ashecon.orgashecon.org/wp-content/uploads/2017/12/S… · XLS file · Web view · 2017-12-07Program Title Abstract Presenting Author Presenting Author Email Address Presenting Author

Program Title Abstract Title

Competition in Medical and Health Insurance Markets

Do Competitors’ Quality Improvements Improve Own Quality? An Empirical Test of Florida Hospitals


Marketplace Competition and Insurers’ Behavior under the Affordable Care Act


Health Care Competition or Regulation: The Unusual Case of Albany Georgia


Weighing the Effects of Vertical Integration versus Market Concentration on Hospital Quality


Does Competition in Home Health Affect Geographic Service Coverage


Did entry of freestanding emergency departments in Texas alleviate high visit volume in hospital-based emergency departments?


The effect of between- and within-molecule competition on drug prices


The effect of between- and within-molecule competition on drug prices


Association between Chemotherapy Provider Competition and Access, Cost and Quality of Care for Medicare Beneficiaries with Cancer


Imperfect Competition among Differentiated Products with Adverse Selection


How Important Is Price Variation Between Insurers?


The Impact of Competition on Investment: Evidence from California Hospitals


Primary Care Access Disparities in Washington State: A Spatial Econometric Analysis


Hospital Market Competition Impact on Hospital Quality – Evidence from Hospital Quality Rating


The Impact of Physician Competition and Market Power on Medical Group Quality and Costs


Association between Physician Practice Prices and Health Care Quality


What Drives Variation in Healthcare Provider Prices?

Abstract

Do Competitors’ Quality Improvements Improve Own Quality? An Empirical Test of Florida Hospitals

For almost two decades, the safety and quality of inpatient care in U.S. hospitals has been of national concern. Efforts to improve these conditions have stemmed mainly from non-market policies and regulation. Economic theory about the relationship between hospital competition and hospital quality is ambiguous: quality may improve, deteriorate or remain unchanged in response to changes in competition. Empirical research on the relationship between hospital market competition and quality has likewise been inconclusive. In order to better understand the mechanisms by which market competition leads to quality improvement within hospitals, we investigate whether an observable improvement in the quality of competitors leads to an increase in investment in a hospital’s own quality. We focus on investment rather than measured quality improvement as the outcome, for two reasons. First, outcomes associated with efforts to improve quality are uncertain; not all quality improvement efforts will be successful, or persist. Secondly, it is not clear what length of time is appropriate to expect measurable improvement to emerge from the quality improvement of competitors. However, the more certain and immediate effects of competition will very likely be through the investments hospitals make to improve their own quality of care. To investigate the relationship between competitor quality and own quality improvement efforts, we use a game-theoretic framework to analyze a panel of all Florida general hospitals from 2004-2015 obtained from Florida’s Center for Health Information and Policy Analysis, a department of the state’s Agency for Health Care Administration. These data include annual hospital inpatient discharge information and financial data. Specifically, we look at the how the quality of competitors, measured by a composite of Agency for Healthcare Research and Quality patient safety (adverse) events, is related to several measures of hospital quality investment, or spending.. We find that, for certain investments, in hospital personnel and wages, there is a beneficial impact from competition within small geographic markets, although the size of the effect is quite small, with estimated elasticities of less than 0.05. Thus, it would seem that the rise of non-market policies from mainly public payers to promote quality (or, to punish poor quality) in U.S. hospitals has been needed to create incentives that are not otherwise provided by market forces.

Introduction: The Affordable Care Act (ACA) organizes private insurers offering individual health plans into insurance exchanges where they are required to offer benefits within standardized categories. To protect patients with pre-existing conditions, the ACA mandated premiums based on modified community rating. However, from 2014 to 2017, premiums increased while many insurers exited exchanges. This paper seeks to explain these phenomena from a market competition perspective. Methods: This paper uses all silver plans from Marketplace Public Use File (PUF) to assess the market concentration and insurer behaviors in Florida, where each county is its own rating area. We conducted tests of independence and constructed two sets of regression models. For the logistic model of whether a rating area has only a single insurer, predictors include: number of issuers and plans lagged one year, the previous year’s median silver premium offered in the county by the dominant issuer (Blue Cross & Blue Shield in Florida, BCBS), as well as the age-adjusted prevalence of chronic disease, rural-urban status, and 2010 population. The model also includes fixed effects for year. The generalized linear model of the individual plan’s average monthly premium started with the same independent variables. Results: The distributions of counties with single insurer were approximately the same from 2015 to 2018 (~ 32% single insurer, ~ 68% not). All urban counties had more than one insurer while 72.2% of the most rural counties had single insurer (p<.001). The marginal effects (ME) of one-year-lag characteristics show an increased probability of single insurer in a rating area given more hospitals, and fewer issuers and plans; but their effects are statistically insignificant. From 2014 to 2018, the number of insurers offering silver plans increased in the first year and dropped after that (12 to 14 to 6), while the number of offered plans decreased 32.2% (1576 to 1069). Meanwhile, premiums increased 77.6% ($434.59 vs. $771.69, p<0.001 on fixed effects for year). Comparing to urban counties, premiums in most rural counties were $15.75 higher (p=.002). Premiums had a predicted $0.82 decrease upon a one-unit increase in the one-year-lag number of plans (p<.001), and a $0.79 increase upon $1 increase in one-year lag median premium of the dominant issuer (p<.001). Population and prevalence of chronic diseases (e.g. cancer and diabetes) had minimal to none impacts in both models. Discussion & Conclusion: The changes in numbers of insurers and plans indicate wax and wane in the market competition. Higher premiums from the dominant insurer did not provide a niche for potential competitors to undercut prices but became a barrier for entry. This is likely associated with the adverse selection issue where higher-risk individuals signed up for insurance in the absence of medical underwriting. The ACA also didn’t address the issue of rural-urban disparity (urban always have access to more than one issuers and on average lower premiums). Further study should investigate approaches for reducing adverse selection, increasing access to insurance and keep market competition at proper equilibrium.

On December 15, 2011, Phoebe Putney Health System acquired the only other hospital in Albany, Georgia—Palmyra Medical Center—despite the Federal Trade Commission’s challenge of the merger as anticompetitive. The acquisition was consummated after the district and appellate courts ruled that Phoebe Putney had antitrust immunity due to its regulation by the local Hospital Authority of Albany-Dougherty County. In February 2013, the Supreme Court reversed these rulings and remanded the case back to the lower courts, after Palmyra Medical Center had become part of Phoebe Putney Memorial Hospital, making a divestiture infeasible. Thus, the acquisition of Palmyra Medical Center by Phoebe Putney provides a natural experiment to study the effects of an otherwise anticompetitive hospital merger subject to local regulation. We found that, after a large price spike in the first post-merger year, the commercial price of inpatient hospital services in Albany, Georgia moderated toward the control group price in subsequent post-merger years. Regarding quality, we found a significant post-merger reduction in inpatient hospital quality relative to controls across many quality metrics. We discuss the implications of these findings for recent initiatives that grant hospitals antitrust immunity in exchange for local regulation.

Provider organizations are increasing in complexity, as hospitals acquire physician practices and physician organizations grow in size. At the same time, hospitals are merging with each other to improve bargaining power with insurers. Greater integration should increase care coordination and limit redundancies, which could improve patient outcomes. However, larger organizations could instead feel less incentive to compete on the basis of quality. We analyze multiple measures of quality from the Medicare Hospital Compare to test whether vertical integration between hospitals and physicians or increases in hospital market concentration influence patient outcomes. We analyze data on 30 measures of hospital quality that were reported to the Center for Medicare and Medicaid Services for the years 2008 to 2015. The measures include hospital readmission rates, process of care measures, and patient satisfaction scores. Identifiers for different types of vertical integration (e.g. no integration, vs. independent practice association, vs. physicians employed by hospitals) are drawn from the American Hospital Association annual survey. We use panel data methods to estimate the effects of within-hospital changes in vertical integration and hospital market concentration on hospital quality, controlling for costly hospital services. We find that vertical integration reduces hospital readmission rates for pneumonia, but less so for other disease conditions. We also find that vertical integration improves quality for a limited set of process and patient satisfaction measures. Yet, increased hospital market concentration is strongly associated with reduced quality across multiple measures, particularly patient satisfaction measures. While better patient experience may not always correlate with higher clinical quality, it is imperative that policies consider the data such as that presented in the CAHPS survey since consumers are now shifting their focus to these and similar reviews of patient experience to select their providers. Our results suggest that regulators should continue to focus scrutiny on proposed hospital mergers. Although vertical integration does not necessarily harm quality, future studies should test whether it is associated with increased hospital prices.

Today’s healthcare setting is characterized by an increased role of post-acute care home health providers. The total number of home health agencies (HHA) participating in Medicare increased by 63% over the past decade. According to Centers for Medicare and Medicaid Services' reports, Medicare fee-for-service (FFS) payments in 2015 totaled $18.1 billion with approximately 3.5 million of Medicare fee-for-service beneficiaries using HHA services. The home healthcare industry, which is part of the post-acute care services, provides patients with home care after hospital discharge, and is needed to improve the recovery process. In this study we look at whether changes in competition among home health agencies affect the quality and quantity of healthcare services provision in the context of geographic service coverage. Existing literature analyzing competition in home health is focused on its effects on price, utilization, and to a lesser extent – quality. To our knowledge, its effects on geographic service coverage was not studied, even though it may have important welfare implications for the population at large. To evaluate the effectiveness of hospital discharge planning in the presence of varying degrees of competition, we develop a spatial model, in which home health agencies bear the travel cost. This allows us to study whether competition increases market coverage, as well as whether competition affects the composition of patients accepted by the home health agency. Since patients with varying health conditions require different levels of service intensity, their value for HHAs may differ as well. This differential value may lead agencies to provide differential coverage by severity dimensions and others. We use Medicare claims data combining the 100% Medicare Provider Analysis and Review (MedPAR) file, which contains claims data for Medicare fee-for-service beneficiaries from the Medicare-certified inpatient hospitals and the Medicare Home Health Agency file, which contains claims data for Medicare home health episodes from 2010 to 2014. We employ advanced spatial econometric techniques to establish a link between changes in competition among providers and its resulting changes in market-level coverage. We also study markets who experienced changes in market concentration over time as well as use within Hospital Referral Region variation across states with and without Certificate of Need legislation for home health to instrument for competition. Empirically documenting which types of patients experience increased coverage with more competition will tell us which type of patients has a higher value to a home health agency.

Background and Objective: Ever since the passage of the Texas Freestanding Emergency Medical Care Facility Licensing Act in 2009, freestanding emergency departments (EDs) have flourished in Texas. While freestanding EDs may provide timely emergency care for patients facing over-crowded hospital emergency rooms, the literature contains only limited case studies of the impact of freestanding EDs on access to emergency care. This study aims to measure the impact of entry of freestanding EDs on the volume of visits to nearby hospital-based emergency departments. Methods: We use American Hospital Association hospital-based ED visit volume in 2010 and 2015 as the dependent variable of interest. Our main explanatory variables are the numbers of freestanding EDs and hospital EDs within certain distance bands, and a dummy variable for whether the hospital built its own satellite EDs in outlying areas. We estimated generalized linear models with Gamma-distributed dependent variables to investigate whether the entry of freestanding EDs helped relieve the burden of ED congestion in nearby hospitals and improve the efficiency of hospital ED services. The analyses control for multiple demographic characteristics and include hospital-level fixed effects. Results: Preliminary results reveal that hospital ED visits are not significantly influenced by the entry of freestanding EDs nearby when taking all the competitors into consideration. The results remain the same with subset tests for hospital EDs in rural areas and urban areas. Setting up a hospital-affiliated freestanding ED slightly increases the overall number of ED visits, although the effect is imprecisely estimated. However, there are significant increases in hospital ED visits in the year 2015 compared to 2010. In conclusion, the entry of freestanding EDs doesn’t help reduce the visit volume in hospital EDs. We are conducting analyses of hospital ED wait times and drop-out rates to address the concerns for ED congestion. We will update this abstract before the conference to incorporate more findings. Conclusion: Texas launched the licensing act for freestanding EDs in 2009, in an effort to relieve hospital emergency congestion and help patients access care in emergency service shortage areas. Our previous study showed that most of the freestanding EDs located in areas with high income and their presences did not reduce average waiting times in hospital EDs. We also noticed many hospitals built hospital-affiliated freestanding EDs to attract more patients, suggesting there are some financial incentives in these markets. We are concerned that the existing policy, instead of relieving the hospital burden, stimulates the demand for emergency services and increases healthcare spending in Texas. We will add data for the years between 2010 and 2015 and refine our measures of freestanding ED entry. We may modify this conclusion after we complete our full analysis.

Previous studies have argued that at least two important types of competition occur in the pharmaceutical industry: within-molecule and between-molecule. Numerous studies have shown that increased within-molecule competition (e.g. due to generic entry following patent expiration) results in substantial reductions in the average price of a molecule. In contrast, there is little or no systematic evidence about the effect of increased between-molecule competition (e.g. due to the entry of new molecules within a class) on the average price of existing molecules. Using data from a number of countries, this research examines the effect of the entry of new molecules on the prices of existing molecules within the same class (i.e. the same therapeutic, pharmacological, or chemical subgroup in the Anatomical Therapeutic Chemical (ATC) classification system), controlling for the number of producers of the molecule. If the entry of new molecules reduces the prices of existing molecules, the net cost to a health care system of the new molecules is lower, perhaps substantially lower, than expenditure on the new molecules. To test the hypothesis that the entry of new molecules reduces the prices of existing molecules within the same class, I estimate the following difference-in-differences model: ln(Pmt) = b ln(N_Molecules_Classmt) + p ln(N_Producers_Moleculemt) + am + dt + emt where

Pmt

= the price (revenue per standard unit) of molecule m in year t

N_Molecules_Classmt

= the number of molecules in molecule m’s (4th, 3rd, or 2nd level) ATC class in year t

N_Producers_Moleculemt

= the number of producers of molecule m in year t

am

= a fixed effect for molecule m

dt = a fixed effect for year t

I estimate models that include lagged values of N_Molecules_Classmt (e.g. N_Molecules_Classm,t-3) as well as the contemporaneous value. Drug price data. For the USA, one source of data on drug prices is Medical Expenditure Panel Survey (MEPS) Prescribed Medicines files for the years 1996-2015. For many countries, data on manufacturer revenue (in USD) and number of standard units, by molecule and year, are obtained from the IMS Health MIDAS database. The molecule’s price (revenue per standard unit) is calculated from these data. Data on between- and within-molecule competition. Data on N_Molecules_Class and N_Producers_Molecule, by molecule and year, are constructed for a number of countries from several different sources. One source that is used for many countries is the IMS Health New Product Focus database, which provides data on all drugs launched in many countries since 1982. We also use country-specific databases on drug registrations, such as the Drugs@FDA database (USA), the Drug Product Database (Canada), Theriaque (France), the SwissMedic Extended Product List (Switzerland), and the Danish Medicines Agency List of Authorized Medicinal Products.

The market structure of oncology care is undergoing dramatic consolidation, yet few studies have examined how changes in market structure impact oncology care for patients receiving physician-administered chemotherapy. Recent research has documented substantial vertical integration in recent years such that the proportion of oncology practices that are owned by hospitals increased from less than 30% to close to 60% between 2004 and 2015. We investigate how changes in market structure impact geographic access to care, Medicare spending on infused chemotherapy, and the quality of care patients receive. Using a 20% sample of Medicare Fee-For-Service claims from 2008 to 2014, we identified a cohort of 142,770 patients receiving infused chemotherapy for cancer and 89,096 new users of chemotherapy. We assessed the relationship between decreases in competition and the following outcomes: the number of chemotherapy-administering physicians within 25, 50, and 75 miles of the patient’s zip code and the distance traveled to receive chemotherapy; Medicare expenditures for infused chemotherapy; and all-cause emergency room visits and chemotherapy-associated hospitalization. Our primary measure of market competition used a modified Herfindahl-Hirschman Index (HHI), similar to the HHI developed by Kessler and McClellan, measured at the core-based statistical area (CBSA). Secondary analysis measured markets at the Hospital Referral Region (HRR). We defined firms based on the tax identifying number listed for the physician administering chemotherapy and their market share as the number of chemotherapy infusions administered by the firm. We used generalized linear models with fixed effects at the market level to examine the relationship between area-level measures of provider competition and each outcome. We log-transformed the HHI which ensures that both small (i.e., a hospital acquiring a community oncologist) and large (i.e., hospitals merging) changes in HHI will be captured. All models adjust for patient level clinical and demographic factors. We find that a one standard deviation increase in logged HHI (i.e., market becoming less competitive) increases the average distance traveled from 100 to 112 miles and decreases the average number of physicians administering chemotherapy within 75 miles from 346 to 312 physicians. When examining spending by Medicare, we find that spending decreases as markets become less competitive at the claim service-line level and the day level but not when considering total spending during six months following treatment initiation. Finally, we do not find any association between the impact of competition on the quality of care that patients receive. Results were consistent when varying the geographic market from the CBSA to the HRR. For patients receiving chemotherapy, competition impacts geographic access to care. However, the association between competition and healthcare spending is not consistent and we do not observe an association between competition and quality of care. Health care administrators should consider how acquisitions and mergers may reduce access to care when assessing the potential consequences of consolidation. While hospitals may improve their financial performance under consolidated delivery systems, this research suggests that patients’ access to care may suffer under less competitive markets.

Standard models of competition between differentiated products in markets with adverse selection assume that the product characteristics are fixed. In the context of insurance markets, this implies firms compete on price while taking as given the terms of the insurance contracts they offer. This assumption confines any welfare conclusions to those based only on the marginal price effect, without anything to say about the degree to which consumers will be insured against risk. In this paper, I will make three main contributions. First, I present a theoretical model of imperfect competition among differentiated insurance contracts in a market with asymmetric. I identify conditions under which a model of symmetric firms has a unique, pure-strategy equilibrium. In a parametric example, I show that both the resource loss of adverse selection and consumer welfare are monotonically increasing in the number of competitors in the market. More general results are forthcoming. Second, I plan to estimate the model using a novel dataset on household health insurance choices made through eHealthInsurance.com, a national online marketplace for purchasing non-group health insurance plans. The data contains age, income, and the truncated zip code of the household. This data can be combined with data collected by the Robert Wood Johnson Foundation on the health insurance plans offered to estimate the demand for health insurance. I will then use the model to estimate marginal cost functions that rationalize the observed data. Finally, I will use the model to make predictions about several policy proposals, including repealing the fine for being uninsured, altering the premium subsidy mechanism, and requiring standardized contracts. This model will be able to make predictions about consumer welfare while accounting for an endogenous change in the space of insurance products available.

Existing research has shown wide variation between hospitals in the prices they negotiate for procedures. Here, we show that variation between insurers is also substantial, even within a hospital. Using the Massachusetts All-Payer Claims Database, we measure negotiated prices for each hospital-insurer pair in the state. We find that differences across payers explain about the same amount of price variation in the data as hospitals. We also show important interactions between insurers and hospitals. These results have implications for insurance markets. In particular, we document the impact of these differences on the value of insurance to individuals. We find that an example high deductible health plan could have an actuarial value ranging from 0.3 to 0.6 depending on the level of prices. We discuss the incentives for insurers to negotiate lower price and how that depends on the sensitivity to both premiums and the insurer's negotiated price level.

Rising numbers of commentators attribute an apparent decline in corporate investment to increased market concentration. To substantiate this view, they primarily cite to changes in industry-level measures. Such metrics are certainly suggestive. However, industrial economists have long recognized the difficulty of drawing strong conclusions from inter-industry -- as opposed to intra-market -- analyses. Moreover, the discussion has largely ignored the fact that economic theory offers ambiguous predictions about the relationship between competition and investment.

This paper begins to fill the gap in the literature by investigating the causal relationship between competition and investment in a context where meaningful variation in both competitive pressure and investment can be cleanly measured. Specifically, I focus on general acute care (GAC) hospitals where there is compelling evidence that competition has declined in many -- though not all -- local markets in recent decades. Moreover, technological progress, depreciation of physical capital, and changing demographics necessitate regular capital investment by hospitals, providing statistical power to the analysis.

Using rich, longitudinal data on the finances and utilization of California hospitals, I employ an empirical approach that incorporates measures of competitive pressure into a hospital investment specification. The richness of the data allows me to account for potential unobserved confounders in a way that is typically impossible in cross-industry analyses. I find economically and statistically significant evidence that more competitive markets foster greater capital investment. My analyses imply that, all else equal, a one standard deviation increase in the level of hospital-specific market concentration leads to a 9 to12 percentage point decline in the investment rate of a hospital. Interestingly, I also find robust evidence that changes in hospital systems' statewide importance, as captured by a measure of health system strength commonly used to assess merger consequences, are associated with even larger magnitude effects. A one standard deviation increase in statewide willingness-to-pay for a hospital system is associated with 13 to 19 percentage point decreases in net asset accumulation at that system's hospitals.

Finding that competition increases investment is consistent with a theoretical literature showing that for many models of competitive interaction, the relative value of waiting to pursue investment projects declines when growth opportunities are contestable. Moreover, it is in keeping with the institutional details of the hospital industry. In many local health care markets, it may be difficult for more than one hospital to cover the fixed costs of becoming a trauma center or providing certain types of very high acuity care. Therefore, competition may spur firms to invest more rapidly than would a monopolist, for the competitors want to establish a strong incumbent position. In addition, the finding that system strength appears negatively correlated with investment may fit with the emerging literature suggesting that cross-market hospital mergers have meaningful effects on economic outcomes.

A significant amount of empirical studies have been dedicated to investigating disparities in primary care access in the United States. Its importance stems from primary care’s positive association with prevention, management of chronic diseases, and cost-savings intervention. The growing body of literature on spatial disparities in health care access has either primarily focused on rural areas or has used counties or states as the level of observation. As a result, little is known about access gaps to primary care within urban settings and the implications of using aggregated data at the county or state level as opposed to smaller area units to inform health policy makers. The objective of this study is to analyze the association between provider-to-population ratio with the socioeconomic, demographic, and insurance characteristics of the population within the Seattle-Tacoma combined statistical area (CSA) at the census tract level. We implement different spatial econometric models to account for spatial autocorrelation. The provider to population ratio for each census tract is a combination of both primary care providers within the census tract and those outside of the tract but within a 25 minute drive radius, in agreement with the guidelines set by the Department of Health and Human Services (DHHS). Data for primary care providers was collected from the National Provider Identification (NPI) database. Primary care providers are defined as individual physicians, nurse practitioners, and physician assistants whose primary taxonomy code is one of the following: General Practice, Family Medicine, Geriatric Medicine, Internal Medicine, or Primary Care. The socioeconomic, demographic, and insurance characteristics data was collected from the American Community Survey 5-Year estimates (2015). As of October 2016, Washington State received a five-year demonstration waiver from the Centers for Medicare & Medicaid Services (CMS) which will allow for the transformation of Washington State’s Medicaid system. Initiative 1 of Medicaid Transformation “is intended to build incentives for providers who are committed to changing how we deliver care” (HCA). Part of this initiative would be to ensure the provision of primary care at the local level. In light of this, it is important to inform policy makers about disparities in primary care access not only in rural areas but also in urban areas.

Recent wave of merger, acquisition and hospital closure restructured the hospital market competition. It is important to examine how hospital market competition affects healthcare service quality. Our study addresses this topic and contributes to existing literature in several ways. First, different from conventional quality measures such as the mortality rate, a comprehensive star rating encompassing 7 quality aspects is used. The method is developed by Center for Medicare and Medicaid Services (CMS) and released in 2016. Second, healthcare quality ratings and variations are examined in different Metropolitan Statistical Areas (MSAs), which aims to evaluate the hospital market competition impact on healthcare services disparity. Controlling for covariates commonly used in existing literature, our preliminary study confirms the finding that hospital market competition is positively related to hospital overall quality rating. In MSAs with more intensive hospital market competition, patients get higher quality of care with less variation. The study is based upon two data sets. Hospital Compare by CMS provides more than 100 healthcare quality measures on patient mortality, safety of care, readmission, patient experience, effectiveness of care, timeliness of care, and efficient use of medical imaging. In 2016, CMS released the overall hospital quality star rating encompassing these aspects for the first time. The overall star rating on hospital quality is used as the dependent variable in various models. Control variables are generated from the Medicare Cost Report data. In addition, metropolitan statistical areas (MSAs) from the United States Census Bureau are used to differentiate geographic areas. The current findings are results of cross-section regressions using hospital quality ratings as the dependent variable and Herfindahl-Hirschman index (HHI) as a major predictor controlling for commonly used hospital characteristics such as rural/urban, not-for-profit/for-profit, and teaching status. In addition, size, case-mix index, payer-mix, and financial health are also included as controls. For MSAs regional level regression, the mean and standard deviation of hospital ratings in an MSA are used as dependent variables, which shed some light on disparities of hospital quality. Controls include demographic covariates such as income, health status, and insurance coverage, etc. We are currently examining a series of further studies on: • Introducing price or charge as an endogenous covariate for hospitals with more commercial payers;• Examine different aspects of healthcare service quality separately as dependent variables;• Replicating the star rating algorithm developed by CMS for historical years and develop panel regression considering the endogeneity of hospital competition HHI;

We first examine the rapid increase in the consolidation of physicians from 2010 to 2016. We find that almost all of it is due to systems purchasing medical groups via vertical integration, and not due to the horizontal consolidation of groups or of systems. Next, we examine what impact this has had on the quality and costs of medical groups, using 2013-2015 Medicare Value Modifier QRUR data (Quality & Resource Use Reports) and IQVIA SK&A physician data. Using a semi-structural equation approach with a random utility model, we estimate the impact of willingness-to-pay (WTP) measures of physician market power on quality and costs, including interactions with ACO affiliation. We compare WTP with other market measures, such as the Dunn-Shapiro and Kessler-McClellan Herfindahl-Hirschman Indices (HHI) and multi-market contact measures. More market power and multi-market contact lead to worse readmission rates for the medical group. Preventable hospitalizations become worse as the Kessler-McClellan HHI increases. However, medical group costs decline with market power and HHI.

Provider consolidation has intensified concerns that providers with market power may be able to charge higher prices without having to deliver better care. Providers, on the other hand, have argued that higher prices cover the costs of delivering higher-quality care, and that integrated practices are better able to coordinate patients' care, thus providing long-run value to patients and health care payers. Due to the limited availability of data linking providers’ prices to measures of their quality of care, little prior research has rigorously examined whether higher provider prices are associated with better care, and most extant studies have focused on hospital, but not physician, prices. In this study, we examined the relationship between physician practice prices for outpatient services and the quality and efficiency of care provided to their patients. Using commercial claims, we classified practices as high-priced or low-priced relative to the average for their geographic area. We compared care quality, utilization, and spending between high-priced and low-priced practices in the same areas using data from the Consumer Assessment of Health Care Providers and Systems survey and linked claims for Medicare beneficiaries. Compared with low-priced practices, high-priced practices were much larger and received 36% higher prices. Patients of high-priced practices reported significantly higher scores on some measures of care coordination and management, but did not differ meaningfully in their overall care ratings, other domains of patient experiences (including physician ratings and access to care), receipt of mammography, vaccinations, or diabetes services, acute care use, or total Medicare spending. Where we did find an association between higher physician prices and higher quality, we found no evidence of continued improvements in quality associated with prices exceeding the average for a geographic area. These findings suggest an overall weak relationship between practices’ prices and their quality and efficiency of care, calling into question claims that high-priced providers deliver substantially higher-value care. Consequently, our findings cast substantial doubt on assertions that consolidation among health care providers -- which contributes to rising health care prices -- provides a net benefit to consumers by substantively improving the quality or value of care.

Researchers have identified significant variation in the prices commercial insurers pay for similar healthcare services. A number of explanations of this variation have been proposed, some focusing on the provider, some focusing on the payer, and some focusing on geographic variation in costs. Learning which of these factors is most important will focus policymakers and researchers looking to evaluate the most important measures to constrain costs. We use the Colorado All-Payer Claims Database to determine whether geography, providers, or payers are the main sources of variation in provider prices. In contrast to previous research, such as that using HCCI or Truven Marketscan data, we have both provider and payer identifiers. This enables us to assess the relative importance of payers and providers in driving costs. Also, by studying Colorado, we are able to evaluate this variation in a state without a single large dominant system and with multiple major metropolitan areas. In line with previous research, we use two main approaches to evaluate the amount and drivers of price variation. First, we isolate clinically homogeneous services and focus on the prices for those services. Second, we construct aggregate price indices after adjusting for treatment acuity. At present, we are waiting on permission from our data-provider to disclose details on these results, but we find qualitatively similar patterns across the two approaches. Having established the relative amount of variation that different sets of fixed effects account for, we are now in the process of assessing how observable characteristics of payers and providers correlate with differences in their average prices.

Abstract

For almost two decades, the safety and quality of inpatient care in U.S. hospitals has been of national concern. Efforts to improve these conditions have stemmed mainly from non-market policies and regulation. Economic theory about the relationship between hospital competition and hospital quality is ambiguous: quality may improve, deteriorate or remain unchanged in response to changes in competition. Empirical research on the relationship between

In order to better understand the mechanisms by which market competition leads to quality improvement within hospitals, we investigate whether an observable improvement in the quality of competitors leads to an increase in investment in a hospital’s own quality. We focus on investment rather than measured quality improvement as the outcome, for two reasons. First, outcomes associated with efforts to improve quality are uncertain; not all quality improvement efforts will be successful, or persist. Secondly, it is not clear what length of time is appropriate to expect measurable improvement to emerge from the quality improvement of competitors. However, the more certain and immediate effects of competition will very

To investigate the relationship between competitor quality and own quality improvement efforts, we use a game-theoretic framework to analyze a panel of all Florida general hospitals from 2004-2015 obtained from Florida’s Center for Health Information and Policy Analysis, a department of the state’s Agency for Health Care Administration. These data include annual hospital inpatient discharge information and financial data. Specifically, we look at the how the quality of competitors, measured by a composite of Agency for Healthcare Research and Quality patient safety (adverse) events, is related to several measures of hospital quality investment, or spending.. We find that, for certain investments, in hospital personnel and wages, there is a beneficial impact from competition within small geographic markets, although the size of the effect is quite small, with estimated elasticities of less than 0.05. Thus, it would seem that the rise of non-market policies from mainly public payers to promote quality (or, to punish poor quality) in U.S. hospitals has been needed to create incentives that are not otherwise provided by market forces.

The Affordable Care Act (ACA) organizes private insurers offering individual health plans into insurance exchanges where they are required to offer benefits within standardized categories. To protect patients with pre-existing conditions, the ACA mandated premiums based on modified community rating. However, from 2014 to 2017, premiums increased while many insurers exited exchanges. This paper seeks to explain these phenomena from a market competition perspective. Methods: This paper uses all silver plans from Marketplace Public Use File (PUF) to assess the market concentration and insurer behaviors in Florida, where each county is its own rating area. We conducted tests of independence and constructed two sets of regression models. For the logistic model of whether a rating area has only a single insurer, predictors include: number of issuers and plans lagged one year, the previous year’s median silver premium offered in the county by the dominant issuer (Blue Cross & Blue Shield in Florida, BCBS), as well as the age-adjusted prevalence of chronic disease, rural-urban status, and 2010 population. The model also includes fixed effects for year. The generalized linear model of the individual plan’s average

The distributions of counties with single insurer were approximately the same from 2015 to 2018 (~ 32% single insurer, ~ 68% not). All urban counties had more than one insurer while 72.2% of the most rural counties had single insurer (p<.001). The marginal effects (ME) of one-year-lag characteristics show an increased probability of single insurer in a rating area given more hospitals, and fewer issuers and plans; but their effects are statistically insignificant. From 2014 to 2018, the number of insurers offering silver plans increased in the first year and dropped after that (12 to 14 to 6), while the number of offered plans decreased 32.2% (1576 to 1069). Meanwhile, premiums increased 77.6% ($434.59 vs. $771.69, p<0.001 on fixed effects for year). Comparing to urban counties, premiums in most rural counties were $15.75 higher (p=.002). Premiums had a predicted $0.82 decrease upon a one-unit increase in the one-year-lag number of plans (p<.001), and a $0.79 increase upon $1 increase in one-year lag median premium of the dominant issuer (p<.001). Population and prevalence of chronic diseases (e.g. cancer and diabetes) had minimal to none impacts in both models. Discussion & Conclusion: The changes in numbers of insurers and plans indicate wax and wane in the market competition. Higher premiums from the dominant insurer did not provide a niche for potential competitors to undercut prices but became a barrier for entry. This is likely associated with the adverse selection issue where higher-risk individuals signed up for insurance in the absence of medical underwriting. The ACA also didn’t address the issue of rural-urban disparity (urban always have access to more than one issuers and on average lower premiums). Further study should investigate approaches for reducing adverse selection, increasing access to insurance and keep market competition at proper equilibrium.

On December 15, 2011, Phoebe Putney Health System acquired the only other hospital in Albany, Georgia—Palmyra Medical Center—despite the Federal Trade Commission’s challenge of the merger as anticompetitive. The acquisition was consummated after the district and appellate courts ruled that Phoebe Putney had antitrust immunity due to its regulation by the local Hospital Authority of Albany-Dougherty County. In February 2013, the Supreme Court reversed these rulings and remanded the case back to the lower courts, after Palmyra Medical Center had become part of Phoebe Putney Memorial Hospital, making a divestiture infeasible. Thus, the acquisition of Palmyra Medical Center by Phoebe Putney provides a natural experiment to study the effects of an otherwise anticompetitive hospital merger subject to local regulation. We found that, after a large price spike in the first post-merger year, the commercial price of inpatient hospital services in Albany, Georgia moderated toward the control group price in subsequent post-merger years. Regarding quality, we found a significant post-merger reduction in inpatient hospital quality relative to controls across many quality metrics. We discuss the implications of these findings for recent initiatives that grant hospitals antitrust immunity in exchange for local regulation.

Provider organizations are increasing in complexity, as hospitals acquire physician practices and physician organizations grow in size. At the same time, hospitals are merging with each other to improve bargaining power with insurers. Greater integration should increase care coordination and limit redundancies, which could improve patient outcomes. However, larger organizations could instead feel less incentive to compete on the basis of quality. We analyze multiple measures of quality from the Medicare Hospital Compare to test whether vertical integration between hospitals and physicians or increases in hospital market concentration influence patient outcomes. We analyze data on 30 measures of hospital quality that were reported to the Center for Medicare and Medicaid Services for the years 2008 to 2015. The measures include hospital readmission rates, process of care measures, and patient satisfaction scores. Identifiers for different types of vertical integration (e.g. no integration, vs. independent practice association, vs. physicians employed by hospitals) are drawn from the American Hospital Association annual survey. We use panel data methods to estimate the effects of within-hospital changes in vertical integration and hospital market concentration on hospital quality, controlling for costly hospital services. We find that vertical integration reduces hospital readmission rates for pneumonia, but less so for other disease conditions. We also find that vertical integration improves quality for a limited set of process and patient satisfaction measures. Yet, increased hospital market concentration is strongly associated with reduced quality across multiple measures, particularly patient satisfaction measures. While better patient experience may not always correlate with higher clinical quality, it is imperative that policies consider the data such as that presented in the CAHPS survey since consumers are now shifting their focus to these and similar reviews of patient experience to select their providers. Our results suggest that regulators should continue to focus scrutiny on proposed hospital mergers. Although vertical integration does not necessarily harm quality, future studies should test whether it is associated with increased hospital prices.

Today’s healthcare setting is characterized by an increased role of post-acute care home health providers. The total number of home health agencies (HHA) participating in Medicare increased by 63% over the past decade. According to Centers for Medicare and Medicaid Services' reports, Medicare fee-for-service (FFS) payments in 2015 totaled $18.1 billion with approximately 3.5 million of Medicare fee-for-service beneficiaries using HHA services. The home healthcare industry, which is part of the post-acute care services, provides patients with home care after hospital discharge, and is needed to improve the recovery process. In this study we look at whether changes in competition among home health agencies affect the quality and quantity of healthcare services provision in the context of geographic service coverage. Existing literature analyzing competition in home health is focused on its effects on price, utilization, and to a lesser extent – quality. To our knowledge, its effects on geographic service coverage was not studied, even though it may have important welfare implications for the population at large. To evaluate the effectiveness of hospital discharge planning in the presence of varying degrees of competition, we develop a spatial model, in which home health agencies bear the travel cost. This allows us to study whether competition increases market coverage, as well as whether competition affects the composition of patients accepted by the home health agency. Since patients with varying health conditions require different levels of service intensity, their value for HHAs may differ as well. This differential value may lead agencies to provide differential coverage by severity dimensions and others. We use Medicare claims data combining the 100% Medicare Provider Analysis and Review (MedPAR) file, which contains claims data for Medicare fee-for-service beneficiaries from the Medicare-certified inpatient hospitals and the Medicare Home Health Agency file, which contains claims data for Medicare home health episodes from 2010 to 2014. We employ advanced spatial econometric techniques to establish a link between changes in competition among providers and its resulting changes in market-level coverage. We also study markets who experienced changes in market concentration over time as well as use within Hospital Referral Region variation across states with and without Certificate of Need legislation for home health to instrument for competition. Empirically documenting which types of patients experience increased coverage with more competition will tell us which type of patients has a higher value to a home health agency.

: Ever since the passage of the Texas Freestanding Emergency Medical Care Facility Licensing Act in 2009, freestanding emergency departments (EDs) have flourished in Texas. While freestanding EDs may provide timely emergency care for patients facing over-crowded hospital emergency rooms, the literature contains only limited case studies of the impact of freestanding EDs on access to emergency care. This study aims to measure the impact of entry of

: We use American Hospital Association hospital-based ED visit volume in 2010 and 2015 as the dependent variable of interest. Our main explanatory variables are the numbers of freestanding EDs and hospital EDs within certain distance bands, and a dummy variable for whether the hospital built its own satellite EDs in outlying areas. We estimated generalized linear models with Gamma-distributed dependent variables to investigate whether the entry of freestanding EDs helped relieve the burden of ED congestion in nearby hospitals and improve the efficiency of hospital ED services. The analyses control for multiple demographic characteristics and include hospital-level fixed effects.

: Preliminary results reveal that hospital ED visits are not significantly influenced by the entry of freestanding EDs nearby when taking all the competitors into consideration. The results remain the same with subset tests for hospital EDs in rural areas and urban areas. Setting up a hospital-affiliated freestanding ED slightly increases the overall number of ED visits, although the effect is imprecisely estimated. However, there are significant increases in hospital ED visits in the year 2015 compared to 2010. In conclusion, the entry of freestanding EDs doesn’t help reduce the visit volume in hospital EDs. We are conducting analyses of hospital ED wait times and drop-out rates to address the concerns for ED congestion. We will update

: Texas launched the licensing act for freestanding EDs in 2009, in an effort to relieve hospital emergency congestion and help patients access care in emergency service shortage areas. Our previous study showed that most of the freestanding EDs located in areas with high income and their presences did not reduce average waiting times in hospital EDs. We also noticed many hospitals built hospital-affiliated freestanding EDs to attract more patients, suggesting there are some financial incentives in these markets. We are concerned that the existing policy, instead of relieving the hospital burden, stimulates the demand for emergency services and increases healthcare spending in Texas. We will add data for the years between 2010 and 2015 and refine our measures of freestanding ED entry. We may modify this conclusion after we complete our full analysis.

Previous studies have argued that at least two important types of competition occur in the pharmaceutical industry: within-molecule and between-molecule. Numerous studies have shown that increased within-molecule competition (e.g. due to generic entry following patent expiration) results in substantial reductions in the average price of a molecule. In contrast, there is little or no systematic evidence about the effect of increased between-molecule competition (e.g. due to the entry of

Using data from a number of countries, this research examines the effect of the entry of new molecules on the prices of existing molecules within the same class (i.e. the same therapeutic, pharmacological, or chemical subgroup in the Anatomical Therapeutic Chemical (ATC) classification system), controlling for the number of producers of the molecule. If the entry of new molecules reduces the prices of existing molecules, the net cost to a health care system of the new molecules is lower,

To test the hypothesis that the entry of new molecules reduces the prices of existing molecules within the same class, I estimate the following difference-in-differences model:

) as well as the contemporaneous value. For the USA, one source of data on drug prices is Medical Expenditure Panel Survey (MEPS) Prescribed Medicines files for the years 1996-2015. For many countries, data on manufacturer revenue (in USD) and number of standard

units, by molecule and year, are obtained from the IMS Health MIDAS database. The molecule’s price (revenue per standard unit) is calculated from these data. Data on N_Molecules_Class and N_Producers_Molecule, by molecule and year, are constructed for a number of countries from several different sources. One source that is used for many countries

database, which provides data on all drugs launched in many countries since 1982. We also use country-specific databases on drug registrations, such as the Drugs@FDA database (USA), the Drug Product Database (Canada), Theriaque (France), the SwissMedic Extended Product List (Switzerland), and the Danish Medicines Agency List of Authorized Medicinal Products.

The market structure of oncology care is undergoing dramatic consolidation, yet few studies have examined how changes in market structure impact oncology care for patients receiving physician-administered chemotherapy. Recent research has documented substantial vertical integration in recent years such that the proportion of oncology practices that are owned by hospitals increased from less than 30% to close to 60% between 2004 and 2015. We investigate how changes in market structure impact geographic access to care, Medicare spending on infused chemotherapy, and the quality of care patients receive. Using a 20% sample of Medicare Fee-For-Service claims from 2008 to 2014, we identified a cohort of 142,770 patients receiving infused chemotherapy for cancer and 89,096 new users of chemotherapy. We assessed the relationship between decreases in competition and the following outcomes: the number of chemotherapy-administering physicians within 25, 50, and 75 miles of the patient’s zip code and the distance traveled to receive chemotherapy; Medicare expenditures for infused chemotherapy; and all-cause emergency room visits and chemotherapy-associated hospitalization. Our primary measure of market competition used a modified Herfindahl-Hirschman Index (HHI), similar to the HHI developed by Kessler and McClellan, measured at the core-based statistical area (CBSA). Secondary analysis measured markets at the Hospital Referral Region (HRR). We defined firms based on the tax identifying number listed for the physician administering chemotherapy and their market share as the number of chemotherapy infusions administered by the firm. We used generalized linear models with fixed effects at the market level to examine the relationship between area-level measures of provider competition and each outcome. We log-transformed the HHI which ensures that both small (i.e., a hospital acquiring a community oncologist) and large (i.e., hospitals merging) changes in HHI will be captured. All models adjust for patient level clinical and demographic factors. We find that a one standard deviation increase in logged HHI (i.e., market becoming less competitive) increases the average distance traveled from 100 to 112 miles and decreases the average number of physicians administering chemotherapy within 75 miles from 346 to 312 physicians. When examining spending by Medicare, we find that spending decreases as markets become less competitive at the claim service-line level and the day level but not when considering total spending during six months following treatment initiation. Finally, we do not find any association between the impact of competition on the quality of care that patients receive. Results were consistent when varying the geographic market from the CBSA to the HRR. For patients receiving chemotherapy, competition impacts geographic access to care. However, the association between competition and healthcare spending is not consistent and we do not observe an association between competition and quality of care. Health care administrators should consider how acquisitions and mergers may reduce access to care when assessing the potential consequences of consolidation. While hospitals may improve their financial performance under consolidated delivery systems, this research suggests that patients’ access to care may suffer under less competitive markets.

Standard models of competition between differentiated products in markets with adverse selection assume that the product characteristics are fixed. In the context of insurance markets, this implies firms compete on price while taking as given the terms of the insurance contracts they offer. This assumption confines any welfare conclusions to those based only on the marginal price effect, without anything to say about the degree to which consumers will be insured against risk. In this paper, I will make three main contributions. First, I present a theoretical model of imperfect competition among differentiated insurance contracts in a market with asymmetric. I identify conditions under which a model of symmetric firms has a unique, pure-strategy equilibrium. In a parametric example, I show that both the resource loss of adverse selection and consumer welfare are monotonically increasing in the number of competitors in the market. More general results are

Second, I plan to estimate the model using a novel dataset on household health insurance choices made through eHealthInsurance.com, a national online marketplace for purchasing non-group health insurance plans. The data contains age, income, and the truncated zip code of the household. This data can be combined with data collected by the Robert Wood Johnson Foundation on the health insurance plans offered to estimate the demand for health insurance. I will then use the model to

Finally, I will use the model to make predictions about several policy proposals, including repealing the fine for being uninsured, altering the premium subsidy mechanism, and requiring standardized contracts. This model will be able to make predictions about consumer welfare while accounting for an endogenous change in the space of insurance products available.


Rising numbers of commentators attribute an apparent decline in corporate investment to increased market concentration. To substantiate this view, they primarily cite to changes in industry-level measures. Such metrics are certainly suggestive. However, industrial economists have long recognized the difficulty of drawing strong conclusions from inter-industry -- as opposed to intra-market -- analyses. Moreover, the discussion has largely ignored the fact that economic theory offers

This paper begins to fill the gap in the literature by investigating the causal relationship between competition and investment in a context where meaningful variation in both competitive pressure and investment can be cleanly measured. Specifically, I focus on general acute care (GAC) hospitals where there is compelling evidence that competition has declined in many -- though not all -- local markets in recent decades. Moreover, technological progress, depreciation of physical capital, and changing demographics necessitate regular capital investment by hospitals, providing statistical power to the analysis.

Using rich, longitudinal data on the finances and utilization of California hospitals, I employ an empirical approach that incorporates measures of competitive pressure into a hospital investment specification. The richness of the data allows me to account for potential unobserved confounders in a way that is typically impossible in cross-industry analyses. I find economically and statistically significant evidence that more competitive markets foster greater capital investment. My analyses imply that, all else equal, a one standard deviation increase in the level of hospital-specific market concentration leads to a 9 to12 percentage point decline in the investment rate of a hospital. Interestingly, I also find robust evidence that changes in hospital systems' statewide importance, as captured by a measure of health system strength commonly used to assess merger consequences, are associated with even larger magnitude effects. A one standard deviation increase in statewide willingness-to-pay for a hospital system is associated with 13 to 19 percentage point decreases in net asset accumulation at that system's hospitals.

Finding that competition increases investment is consistent with a theoretical literature showing that for many models of competitive interaction, the relative value of waiting to pursue investment projects declines when growth opportunities are contestable. Moreover, it is in keeping with the institutional details of the hospital industry. In many local health care markets, it may be difficult for more than one hospital to cover the fixed costs of becoming a trauma center or providing certain types of very high acuity care. Therefore, competition may spur firms to invest more rapidly than would a monopolist, for the competitors want to establish a strong incumbent position. In addition, the finding that system strength appears negatively correlated with investment may fit with the emerging literature suggesting that cross-market hospital mergers have meaningful effects on economic outcomes.

A significant amount of empirical studies have been dedicated to investigating disparities in primary care access in the United States. Its importance stems from primary care’s positive association with prevention, management of chronic diseases,

The growing body of literature on spatial disparities in health care access has either primarily focused on rural areas or has used counties or states as the level of observation. As a result, little is known about access gaps to primary care within urban settings and the implications of using aggregated data at the county or state level as opposed to smaller area units to inform health policy makers. The objective of this study is to analyze the association between provider-to-population ratio with the socioeconomic, demographic, and insurance characteristics of the population within the Seattle-Tacoma combined statistical area (CSA) at the census tract level. We implement different spatial econometric models to account for spatial autocorrelation. The provider to population ratio for each census tract is a combination of both primary care providers within the census tract and those outside of the tract but within a 25 minute drive radius, in agreement with the guidelines set by the Department of Health and Human Services (DHHS). Data for primary care providers was collected from the National Provider Identification (NPI) database. Primary care providers are defined as individual physicians, nurse practitioners, and physician assistants whose primary taxonomy code is one of the following: General Practice, Family Medicine, Geriatric Medicine, Internal Medicine, or Primary Care. The socioeconomic, demographic, and insurance characteristics data was collected

As of October 2016, Washington State received a five-year demonstration waiver from the Centers for Medicare & Medicaid Services (CMS) which will allow for the transformation of Washington State’s Medicaid system. Initiative 1 of Medicaid Transformation “is intended to build incentives for providers who are committed to changing how we deliver care” (HCA). Part of this initiative would be to ensure the provision of primary care at the local level. In light of this, it is important to inform policy makers about disparities in primary care access not only in rural areas but also in urban areas.

Recent wave of merger, acquisition and hospital closure restructured the hospital market competition. It is important to examine how hospital market competition affects healthcare service quality. Our study addresses this topic and contributes to existing literature in several ways. First, different from conventional quality measures such as the mortality rate, a comprehensive star rating encompassing 7 quality aspects is used. The method is developed by Center for Medicare and Medicaid Services (CMS) and released in 2016. Second, healthcare quality ratings and variations are examined in different Metropolitan Statistical Areas (MSAs), which aims to evaluate the hospital market competition impact on healthcare services disparity. Controlling for covariates commonly used in existing literature, our preliminary study confirms the finding that hospital market competition is positively related to hospital overall quality rating. In MSAs with more intensive hospital market

The study is based upon two data sets. Hospital Compare by CMS provides more than 100 healthcare quality measures on patient mortality, safety of care, readmission, patient experience, effectiveness of care, timeliness of care, and efficient use of medical imaging. In 2016, CMS released the overall hospital quality star rating encompassing these aspects for the first time. The overall star rating on hospital quality is used as the dependent variable in various models. Control variables are generated from the Medicare Cost Report data. In addition, metropolitan statistical areas (MSAs) from the United States Census Bureau are used to differentiate geographic areas. The current findings are results of cross-section regressions using hospital quality ratings as the dependent variable and Herfindahl-Hirschman index (HHI) as a major predictor controlling for commonly used hospital characteristics such as rural/urban, not-for-profit/for-profit, and teaching status. In addition, size, case-mix index, payer-mix, and financial health are also included as controls. For MSAs regional level regression, the mean and standard deviation of hospital ratings in an MSA are used as dependent variables, which shed some light on disparities of hospital quality. Controls include demographic covariates such as income, health status, and insurance coverage, etc. We are currently examining a series of further studies on: • Introducing price or charge as an endogenous covariate for hospitals with more commercial payers;

Replicating the star rating algorithm developed by CMS for historical years and develop panel regression considering the endogeneity of hospital competition HHI;

We first examine the rapid increase in the consolidation of physicians from 2010 to 2016. We find that almost all of it is due to systems purchasing medical groups via vertical integration, and not due to the horizontal consolidation of groups or of systems. Next, we examine what impact this has had on the quality and costs of medical groups, using 2013-2015 Medicare Value Modifier QRUR data (Quality & Resource Use Reports) and IQVIA SK&A physician data. Using a semi-structural equation approach with a random utility model, we estimate the impact of willingness-to-pay (WTP) measures of physician market power on quality and costs, including interactions with ACO affiliation. We compare WTP with other market measures, such as the Dunn-Shapiro and Kessler-McClellan Herfindahl-Hirschman Indices (HHI) and multi-market contact measures. More market power and multi-market contact lead to worse readmission rates for the medical group. Preventable hospitalizations become worse as the Kessler-McClellan HHI increases. However, medical group costs decline with market power and HHI.

Provider consolidation has intensified concerns that providers with market power may be able to charge higher prices without having to deliver better care. Providers, on the other hand, have argued that higher prices cover the costs of delivering higher-quality care, and that integrated practices are better able to coordinate patients' care, thus providing long-run value to patients and health care payers. Due to the limited availability of data linking providers’ prices to measures of their quality of care, little prior research has rigorously examined whether higher provider prices are associated with better care, and most extant studies have focused on hospital, but not physician, prices. In this study, we examined the relationship between physician practice prices for outpatient services and the quality and efficiency of care provided to their patients. Using commercial claims, we classified practices as high-priced or low-priced relative to the average for their geographic area. We compared care quality, utilization, and spending between high-priced and low-priced practices in the same areas using data from the Consumer Assessment of Health Care Providers and Systems survey and linked claims for Medicare beneficiaries. Compared with low-priced practices, high-priced practices were much larger and received 36% higher prices. Patients of high-priced practices reported significantly higher scores on some measures of care coordination and management, but did not differ meaningfully in their overall care ratings, other domains of patient experiences (including physician ratings and access to care), receipt of mammography, vaccinations, or diabetes services, acute care use, or total Medicare spending. Where we did find an association between higher physician prices and higher quality, we found no evidence of continued improvements in quality associated with prices exceeding the average for a geographic area. These findings suggest an overall weak relationship between practices’ prices and their quality and efficiency of care, calling into question claims that high-priced providers deliver substantially higher-value care. Consequently, our findings cast substantial doubt on assertions that consolidation among health care providers -- which contributes to rising health care prices -- provides a net benefit to consumers by substantively improving the quality or value of care.

Researchers have identified significant variation in the prices commercial insurers pay for similar healthcare services. A number of explanations of this variation have been proposed, some focusing on the provider, some focusing on the payer, and some focusing on geographic variation in costs. Learning which of these factors is most important will focus policymakers and researchers looking to evaluate the most important measures to constrain costs. We use the Colorado All-Payer Claims Database to determine whether geography, providers, or payers are the main sources of variation in provider prices. In contrast to previous research, such as that using HCCI or Truven Marketscan data, we have

and providers in driving costs. Also, by studying Colorado, we are able to evaluate this variation in a state without a single large dominant system and with

In line with previous research, we use two main approaches to evaluate the amount and drivers of price variation. First, we isolate clinically homogeneous services and focus on the prices for those services. Second, we construct aggregate price indices after adjusting for treatment acuity. At present, we are waiting on permission from our data-provider to disclose details on these results, but we find qualitatively similar patterns across the two approaches. Having established the relative amount of variation that different sets of fixed effects account for, we are now in the process of assessing how observable characteristics of payers and providers correlate with differences in their average prices.

Abstract Presenting Author Presenting Author Email Address

Linda Dynan [email protected]

Yuxian Du [email protected]

Christopher Garmon [email protected]

For almost two decades, the safety and quality of inpatient care in U.S. hospitals has been of national concern. Efforts to improve these conditions have stemmed mainly from non-market policies and regulation. Economic theory about the relationship between hospital competition and hospital quality is ambiguous: quality may improve, deteriorate or remain unchanged in response to changes in competition. Empirical research on the relationship between

In order to better understand the mechanisms by which market competition leads to quality improvement within hospitals, we investigate whether an observable improvement in the quality of competitors leads to an increase in investment in a hospital’s own quality. We focus on investment rather than measured quality improvement as the outcome, for two reasons. First, outcomes associated with efforts to improve quality are uncertain; not all quality improvement efforts will be successful, or persist. Secondly, it is not clear what length of time is appropriate to expect measurable improvement to emerge from the quality improvement of competitors. However, the more certain and immediate effects of competition will very

To investigate the relationship between competitor quality and own quality improvement efforts, we use a game-theoretic framework to analyze a panel of all Florida general hospitals from 2004-2015 obtained from Florida’s Center for Health Information and Policy Analysis, a department of the state’s Agency for Health Care Administration. These data include annual hospital inpatient discharge information and financial data. Specifically, we look at the how the quality of competitors,

We find that, for certain investments, in hospital personnel and wages, there is a beneficial impact from competition within small geographic markets, although the size of the effect is quite small, with estimated elasticities of less than 0.05. Thus, it would seem that the rise of non-market policies from mainly public payers to promote quality (or, to punish poor quality) in U.S. hospitals has been needed to create incentives that are not otherwise provided by market forces.

The Affordable Care Act (ACA) organizes private insurers offering individual health plans into insurance exchanges where they are required to offer benefits within standardized categories. To protect patients with pre-existing conditions, the ACA mandated premiums based on modified community rating. However, from 2014 to 2017, premiums increased while many insurers exited exchanges. This paper seeks to explain these phenomena from a market competition perspective. Methods: This paper uses all silver plans from Marketplace Public Use File (PUF) to assess the market concentration and insurer behaviors in Florida, where each county is its own rating area. We conducted tests of independence and constructed two sets of regression models. For the logistic model of whether a rating area has only a single insurer, predictors include: number of issuers and plans lagged one year, the previous year’s median silver premium offered in the county by the dominant issuer (Blue Cross & Blue Shield in Florida, BCBS), as well as the age-adjusted prevalence of chronic disease, rural-urban status, and 2010 population. The model also includes fixed effects for year. The generalized linear model of the individual plan’s average

The distributions of counties with single insurer were approximately the same from 2015 to 2018 (~ 32% single insurer, ~ 68% not). All urban counties had more than one insurer while 72.2% of the most rural counties had single insurer (p<.001). The marginal effects (ME) of one-year-lag characteristics show an increased probability of single insurer in a rating area given more hospitals, and fewer issuers and plans; but their effects are statistically insignificant. From 2014 to 2018, the number of insurers offering silver plans increased in the first year and dropped after that (12 to 14 to 6), while the number of offered plans decreased 32.2% (1576 to 1069). Meanwhile, premiums increased 77.6% ($434.59 vs. $771.69, p<0.001 on fixed effects for year). Comparing to urban counties, premiums in most rural counties were $15.75 higher (p=.002). Premiums had a predicted $0.82 decrease upon a one-unit increase in the one-year-lag number of plans (p<.001), and a $0.79 increase upon $1 increase in one-year lag median premium of the dominant issuer (p<.001). Population and prevalence of chronic diseases (e.g. cancer and diabetes) had minimal to none impacts in both models. Discussion & Conclusion: The changes in numbers of insurers and plans indicate wax and wane in the market competition. Higher premiums from the dominant insurer did not provide a niche for potential competitors to undercut prices but became a barrier for entry. This is likely associated with the adverse selection issue where higher-risk individuals signed up for insurance in the absence of medical underwriting. The ACA also didn’t address the issue of rural-urban disparity (urban always have access to more than one issuers and on average lower premiums). Further study should investigate approaches for reducing adverse selection, increasing access to insurance and keep market competition at proper equilibrium.

On December 15, 2011, Phoebe Putney Health System acquired the only other hospital in Albany, Georgia—Palmyra Medical Center—despite the Federal Trade Commission’s challenge of the merger as anticompetitive. The acquisition was consummated after the district and appellate courts ruled that Phoebe Putney had antitrust immunity due to its regulation by the local Hospital Authority of Albany-Dougherty County. In February 2013, the Supreme Court reversed these rulings and remanded the case back to the lower courts, after Palmyra Medical Center had become part of Phoebe Putney Memorial Hospital, making a divestiture infeasible. Thus, the acquisition of Palmyra Medical Center by Phoebe Putney provides a natural experiment to study the effects of an otherwise anticompetitive hospital merger subject to local regulation. We found that, after a large price spike in the first post-merger year, the commercial price of inpatient hospital services in Albany, Georgia moderated toward the control group price in subsequent post-merger years. Regarding quality, we found a significant post-merger reduction in inpatient hospital quality relative to controls across many quality metrics. We discuss the implications of

Marah Short [email protected]

Elena Andreyeva [email protected]

Yingying Xu [email protected]

Frank Lichtenberg [email protected]

Provider organizations are increasing in complexity, as hospitals acquire physician practices and physician organizations grow in size. At the same time, hospitals are merging with each other to improve bargaining power with insurers. Greater integration should increase care coordination and limit redundancies, which could improve patient outcomes. However, larger organizations could instead feel less incentive to compete on the basis of quality. We analyze multiple measures of quality from the Medicare Hospital Compare to test whether vertical integration between hospitals and physicians or increases in hospital market concentration influence patient outcomes. We analyze data on 30 measures of hospital quality that were reported to the Center for Medicare and Medicaid Services for the years 2008 to 2015. The measures include hospital readmission rates, process of care measures, and patient satisfaction scores. Identifiers for different types of vertical integration (e.g. no integration, vs. independent practice association, vs. physicians employed by hospitals) are drawn from the American Hospital Association annual survey. We use panel data methods to estimate the

We find that vertical integration reduces hospital readmission rates for pneumonia, but less so for other disease conditions. We also find that vertical integration improves quality for a limited set of process and patient satisfaction measures. Yet, increased hospital market concentration is strongly associated with reduced quality across multiple measures, particularly patient satisfaction measures. While better patient experience may not always correlate with higher clinical quality, it is imperative that policies consider the data such as that presented in the CAHPS survey since consumers are now shifting their focus to these and similar reviews of patient experience to select their providers. Our results suggest that regulators should continue to focus scrutiny on proposed hospital mergers. Although vertical integration does not necessarily harm quality, future studies should test whether it is associated with increased hospital prices.

Today’s healthcare setting is characterized by an increased role of post-acute care home health providers. The total number of home health agencies (HHA) participating in Medicare increased by 63% over the past decade. According to Centers for Medicare and Medicaid Services' reports, Medicare fee-for-service (FFS) payments in 2015 totaled $18.1 billion with approximately 3.5 million of Medicare fee-for-service beneficiaries using HHA services. The home healthcare industry, which is part of the post-acute care services, provides patients with home care after hospital discharge, and is needed to improve the recovery process. In this study we look at whether changes in competition among home health agencies affect the quality and quantity of healthcare services provision in the context of geographic service coverage. Existing literature analyzing competition in home health is focused on its effects on price, utilization, and to a lesser

To evaluate the effectiveness of hospital discharge planning in the presence of varying degrees of competition, we develop a spatial model, in which home health agencies bear the travel cost. This allows us to study whether competition increases market coverage, as well as whether competition affects the composition of patients accepted by the home health agency. Since patients with varying health conditions require different levels of service intensity, their value for HHAs may differ as

We use Medicare claims data combining the 100% Medicare Provider Analysis and Review (MedPAR) file, which contains claims data for Medicare fee-for-service beneficiaries from the Medicare-certified inpatient hospitals and the Medicare Home Health Agency file, which contains claims data for Medicare home health episodes from 2010 to 2014. We employ advanced spatial econometric techniques to establish a link between changes in competition among providers and its resulting changes in market-level coverage. We also study markets who experienced changes in market concentration over time as well as use within Hospital Referral Region variation across states with and without Certificate of Need legislation for home health to instrument for competition. Empirically documenting which types of patients experience increased coverage with more competition will tell us which type of patients has a higher value to a home health agency.

: Ever since the passage of the Texas Freestanding Emergency Medical Care Facility Licensing Act in 2009, freestanding emergency departments (EDs) have flourished in Texas. While freestanding EDs may provide timely emergency care for patients facing over-crowded hospital emergency rooms, the literature contains only limited case studies of the impact of freestanding EDs on access to emergency care. This study aims to measure the impact of entry of

: We use American Hospital Association hospital-based ED visit volume in 2010 and 2015 as the dependent variable of interest. Our main explanatory variables are the numbers of freestanding EDs and hospital EDs within certain distance bands, and a dummy variable for whether the hospital built its own satellite EDs in outlying areas. We estimated generalized linear models with Gamma-distributed dependent variables to investigate whether the entry of freestanding EDs helped relieve the burden of ED congestion in nearby hospitals and improve the efficiency of hospital ED services. The analyses control for multiple demographic characteristics and include hospital-level fixed effects.

: Preliminary results reveal that hospital ED visits are not significantly influenced by the entry of freestanding EDs nearby when taking all the competitors into consideration. The results remain the same with subset tests for hospital EDs in rural areas and urban areas. Setting up a hospital-affiliated freestanding ED slightly increases the overall number of ED visits, although the effect is imprecisely estimated. However, there are significant increases in hospital ED visits in the year 2015 compared to 2010. In conclusion, the entry of freestanding EDs doesn’t help reduce the visit volume in hospital EDs. We are conducting analyses of hospital ED wait times and drop-out rates to address the concerns for ED congestion. We will update

: Texas launched the licensing act for freestanding EDs in 2009, in an effort to relieve hospital emergency congestion and help patients access care in emergency service shortage areas. Our previous study showed that most of the freestanding EDs located in areas with high income and their presences did not reduce average waiting times in hospital EDs. We also noticed many hospitals built hospital-affiliated freestanding EDs to attract more patients, suggesting there are some financial incentives in these markets. We are concerned that the existing policy, instead of relieving the hospital burden, stimulates the demand for emergency services and increases healthcare spending in Texas. We will add data for the years

Previous studies have argued that at least two important types of competition occur in the pharmaceutical industry: within-molecule and between-molecule. Numerous studies have shown that increased within-molecule competition (e.g. due to generic entry following patent expiration) results in substantial reductions in the average price of a molecule. In contrast, there is little or no systematic evidence about the effect of increased between-molecule competition (e.g. due to the entry of

Using data from a number of countries, this research examines the effect of the entry of new molecules on the prices of existing molecules within the same class (i.e. the same therapeutic, pharmacological, or chemical subgroup in the Anatomical Therapeutic Chemical (ATC) classification system), controlling for the number of producers of the molecule. If the entry of new molecules reduces the prices of existing molecules, the net cost to a health care system of the new molecules is lower,


Aaron Winn [email protected]

Conor Ryan [email protected]

For the USA, one source of data on drug prices is Medical Expenditure Panel Survey (MEPS) Prescribed Medicines files for the years 1996-2015. For many countries, data on manufacturer revenue (in USD) and number of standard

Data on N_Molecules_Class and N_Producers_Molecule, by molecule and year, are constructed for a number of countries from several different sources. One source that is used for many countries database, which provides data on all drugs launched in many countries since 1982. We also use country-specific databases on drug registrations, such as the Drugs@FDA database (USA), the Drug Product

The market structure of oncology care is undergoing dramatic consolidation, yet few studies have examined how changes in market structure impact oncology care for patients receiving physician-administered chemotherapy. Recent research has documented substantial vertical integration in recent years such that the proportion of oncology practices that are owned by hospitals increased from less than 30% to close to 60% between 2004 and 2015. We investigate how changes in market

Using a 20% sample of Medicare Fee-For-Service claims from 2008 to 2014, we identified a cohort of 142,770 patients receiving infused chemotherapy for cancer and 89,096 new users of chemotherapy. We assessed the relationship between decreases in competition and the following outcomes: the number of chemotherapy-administering physicians within 25, 50, and 75 miles of the patient’s zip code and the distance traveled to receive chemotherapy; Medicare expenditures for infused

Our primary measure of market competition used a modified Herfindahl-Hirschman Index (HHI), similar to the HHI developed by Kessler and McClellan, measured at the core-based statistical area (CBSA). Secondary analysis measured markets at the Hospital Referral Region (HRR). We defined firms based on the tax identifying number listed for the physician administering chemotherapy and their market share as the number of chemotherapy infusions administered by the firm. We used generalized linear models with fixed effects at the market level to examine the relationship between area-level measures of provider competition and each outcome. We log-transformed the HHI which ensures that both small (i.e., a hospital

We find that a one standard deviation increase in logged HHI (i.e., market becoming less competitive) increases the average distance traveled from 100 to 112 miles and decreases the average number of physicians administering chemotherapy within 75 miles from 346 to 312 physicians. When examining spending by Medicare, we find that spending decreases as markets become less competitive at the claim service-line level and the day level but not when considering total spending during six months following treatment initiation. Finally, we do not find any association between the impact of competition on the quality of care that patients receive. Results were consistent when varying the geographic market from the CBSA to the HRR. For patients receiving chemotherapy, competition impacts geographic access to care. However, the association between competition and healthcare spending is not consistent and we do not observe an association between competition and quality of care. Health care administrators should consider how acquisitions and mergers may reduce access to care when assessing the potential consequences of consolidation. While hospitals may improve their financial performance under consolidated

Standard models of competition between differentiated products in markets with adverse selection assume that the product characteristics are fixed. In the context of insurance markets, this implies firms compete on price while taking as given the terms of the insurance contracts they offer. This assumption confines any welfare conclusions to those based only on the marginal price effect, without anything to say about the degree to which consumers will be insured against risk. In this paper, I will make three main contributions. First, I present a theoretical model of imperfect competition among differentiated insurance contracts in a market with asymmetric. I identify conditions under which a model of symmetric firms has a unique, pure-strategy equilibrium. In a parametric example, I show that both the resource loss of adverse selection and consumer welfare are monotonically increasing in the number of competitors in the market. More general results are

Second, I plan to estimate the model using a novel dataset on household health insurance choices made through eHealthInsurance.com, a national online marketplace for purchasing non-group health insurance plans. The data contains age, income, and the truncated zip code of the household. This data can be combined with data collected by the Robert Wood Johnson Foundation on the health insurance plans offered to estimate the demand for health insurance. I will then use the model to

Finally, I will use the model to make predictions about several policy proposals, including repealing the fine for being uninsured, altering the premium subsidy mechanism, and requiring standardized contracts. This model will be able to make

Stuart Craig [email protected]

Nathan Wilson [email protected]

German Izon [email protected]


Rising numbers of commentators attribute an apparent decline in corporate investment to increased market concentration. To substantiate this view, they primarily cite to changes in industry-level measures. Such metrics are certainly suggestive. However, industrial economists have long recognized the difficulty of drawing strong conclusions from inter-industry -- as opposed to intra-market -- analyses. Moreover, the discussion has largely ignored the fact that economic theory offers

This paper begins to fill the gap in the literature by investigating the causal relationship between competition and investment in a context where meaningful variation in both competitive pressure and investment can be cleanly measured. Specifically, I focus on general acute care (GAC) hospitals where there is compelling evidence that competition has declined in many -- though not all -- local markets in recent decades. Moreover, technological progress, depreciation of physical capital, and

Using rich, longitudinal data on the finances and utilization of California hospitals, I employ an empirical approach that incorporates measures of competitive pressure into a hospital investment specification. The richness of the data allows me to account for potential unobserved confounders in a way that is typically impossible in cross-industry analyses. I find economically and statistically significant evidence that more competitive markets foster greater capital investment. My analyses imply that, all else equal, a one standard deviation increase in the level of hospital-specific market concentration leads to a 9 to12 percentage point decline in the investment rate of a hospital. Interestingly, I also find robust evidence that changes in hospital systems' statewide importance, as captured by a measure of health system strength commonly used to assess merger consequences, are associated with even larger magnitude effects. A one standard deviation increase in statewide

Finding that competition increases investment is consistent with a theoretical literature showing that for many models of competitive interaction, the relative value of waiting to pursue investment projects declines when growth opportunities are contestable. Moreover, it is in keeping with the institutional details of the hospital industry. In many local health care markets, it may be difficult for more than one hospital to cover the fixed costs of becoming a trauma center or providing certain types of very high acuity care. Therefore, competition may spur firms to invest more rapidly than would a monopolist, for the competitors want to establish a strong incumbent position. In addition, the finding that system strength appears negatively

A significant amount of empirical studies have been dedicated to investigating disparities in primary care access in the United States. Its importance stems from primary care’s positive association with prevention, management of chronic diseases,

The growing body of literature on spatial disparities in health care access has either primarily focused on rural areas or has used counties or states as the level of observation. As a result, little is known about access gaps to primary care within urban settings and the implications of using aggregated data at the county or state level as opposed to smaller area units to inform health policy makers. The objective of this study is to analyze the association between provider-to-population ratio with the socioeconomic, demographic, and insurance characteristics of the population within the Seattle-Tacoma combined statistical area (CSA) at the census tract level. We implement different spatial econometric models to account for spatial autocorrelation. The provider to population ratio for each census tract is a combination of both primary care providers within the census tract and those outside of the tract but within a 25 minute drive radius, in agreement with the guidelines set by the Department of Health and Human Services (DHHS). Data for primary care providers was collected from the National Provider Identification (NPI) database. Primary care providers are defined as individual physicians, nurse practitioners, and physician assistants whose primary taxonomy code is one of the following: General Practice, Family Medicine, Geriatric Medicine, Internal Medicine, or Primary Care. The socioeconomic, demographic, and insurance characteristics data was collected

As of October 2016, Washington State received a five-year demonstration waiver from the Centers for Medicare & Medicaid Services (CMS) which will allow for the transformation of Washington State’s Medicaid system. Initiative 1 of Medicaid Transformation “is intended to build incentives for providers who are committed to changing how we deliver care” (HCA). Part of this initiative would be to ensure the provision of primary care at the local level. In light of this, it is important to inform

Bo Shi [email protected]

Bill Encinosa [email protected]

Eric Roberts [email protected]

Nathan Wilson [email protected]

Recent wave of merger, acquisition and hospital closure restructured the hospital market competition. It is important to examine how hospital market competition affects healthcare service quality. Our study addresses this topic and contributes to existing literature in several ways. First, different from conventional quality measures such as the mortality rate, a comprehensive star rating encompassing 7 quality aspects is used. The method is developed by Center for Medicare and Medicaid Services (CMS) and released in 2016. Second, healthcare quality ratings and variations are examined in different Metropolitan Statistical Areas (MSAs), which aims to evaluate the hospital market competition impact on healthcare services disparity. Controlling for covariates commonly used in existing literature, our preliminary study confirms the finding that hospital market competition is positively related to hospital overall quality rating. In MSAs with more intensive hospital market

The study is based upon two data sets. Hospital Compare by CMS provides more than 100 healthcare quality measures on patient mortality, safety of care, readmission, patient experience, effectiveness of care, timeliness of care, and efficient use of medical imaging. In 2016, CMS released the overall hospital quality star rating encompassing these aspects for the first time. The overall star rating on hospital quality is used as the dependent variable in various models. Control variables are

The current findings are results of cross-section regressions using hospital quality ratings as the dependent variable and Herfindahl-Hirschman index (HHI) as a major predictor controlling for commonly used hospital characteristics such as rural/urban, not-for-profit/for-profit, and teaching status. In addition, size, case-mix index, payer-mix, and financial health are also included as controls. For MSAs regional level regression, the mean and standard deviation of hospital ratings in an MSA are used as dependent variables, which shed some light on disparities of hospital quality. Controls include demographic covariates such as income, health status, and insurance coverage, etc.

We first examine the rapid increase in the consolidation of physicians from 2010 to 2016. We find that almost all of it is due to systems purchasing medical groups via vertical integration, and not due to the horizontal consolidation of groups or of systems. Next, we examine what impact this has had on the quality and costs of medical groups, using 2013-2015 Medicare Value Modifier QRUR data (Quality & Resource Use Reports) and IQVIA SK&A physician data. Using a semi-structural equation approach with a random utility model, we estimate the impact of willingness-to-pay (WTP) measures of physician market power on quality and costs, including interactions with ACO affiliation. We compare WTP with other market measures, such as the Dunn-Shapiro and Kessler-McClellan Herfindahl-Hirschman Indices (HHI) and multi-market contact measures. More market power and multi-market contact lead to worse readmission rates for the medical group. Preventable hospitalizations

Provider consolidation has intensified concerns that providers with market power may be able to charge higher prices without having to deliver better care. Providers, on the other hand, have argued that higher prices cover the costs of delivering higher-quality care, and that integrated practices are better able to coordinate patients' care, thus providing long-run value to patients and health care payers. Due to the limited availability of data linking providers’ prices to measures of their quality of care, little prior research has rigorously examined whether higher provider prices are associated with better care, and most extant studies have focused on hospital, but not physician, prices. In this study, we examined the relationship between physician practice prices for outpatient services and the quality and efficiency of care provided to their patients. Using commercial claims, we classified practices as high-priced or low-priced relative to the average for their geographic area. We compared care quality, utilization, and spending between high-priced and low-priced practices in the same areas using data from the Consumer Assessment of Health Care Providers and Systems survey and linked claims for Medicare beneficiaries. Compared with low-priced practices, high-priced practices were much larger and received 36% higher prices. Patients of high-priced practices reported significantly higher scores on some measures of care coordination and management, but did not differ meaningfully in their overall care ratings, other domains of patient experiences (including physician ratings and access to care), receipt of mammography, vaccinations, or diabetes services, acute care use, or total Medicare spending. Where we did find an association between higher physician prices and higher quality, we found no evidence of continued improvements in quality associated with prices exceeding the average for a geographic area. These findings suggest an overall weak relationship between practices’ prices and their quality and efficiency of care, calling into question claims that high-priced providers deliver substantially higher-value care. Consequently, our findings cast substantial doubt on assertions that

Researchers have identified significant variation in the prices commercial insurers pay for similar healthcare services. A number of explanations of this variation have been proposed, some focusing on the provider, some focusing on the payer, and some focusing on geographic variation in costs. Learning which of these factors is most important will focus policymakers and researchers looking to evaluate the most important measures to constrain costs. We use the Colorado All-Payer Claims Database to determine whether geography, providers, or payers are the main sources of variation in provider prices. In contrast to previous research, such as that using HCCI or Truven Marketscan data, we have

providers in driving costs. Also, by studying Colorado, we are able to evaluate this variation in a state without a single large dominant system and with

In line with previous research, we use two main approaches to evaluate the amount and drivers of price variation. First, we isolate clinically homogeneous services and focus on the prices for those services. Second, we construct aggregate price indices after adjusting for treatment acuity. At present, we are waiting on permission from our data-provider to disclose details on these results, but we find qualitatively similar patterns across the two approaches. Having established the relative amount of variation that different sets of fixed effects account for, we are now in the process of assessing how observable characteristics of payers and providers correlate with differences in their average prices.

Presenting Author Affiliation Co-Author(s)

Northern Kentucky University Richard Smith Complete

Texas A&M University Robert Ohsfeldt; Michael Morrisey Complete

University of Missouri Kansas City Laura Kmitch Complete

Submission Complete

Rice University Vivian Ho Complete

The Wharton School, University of Pennsylvania Rachel Werner; Guy David Complete

Complete

Columbia University - Columbia Business School Complete

Rice University, James A. Baker III Institute for Public Policy


Medical College of Wisconsin Complete

University of Minnesota Complete

Justin Trodgon; Stacie Dusetzina; Ethan Basch; G. Mark Holmes; Nancy Keating

University of Pennsylvania Keith Ericson; Amanda Starc Complete

Federal Trade Commission Complete

Eastern Washington University Complete

Morehead State University Complete

Agency for Healthcare Research and Quality Complete

J. McWilliams; Ateev Mehrotra Complete

Federal Trade Commission Ted Rosenbaum; Matthew Panhans Complete

University of Pittsburgh Graduate School of Public Health, Department of Health Policy and Management

Program Title Abstract Title Abstract

Consumer Decision Making in Health Care




A Test of Supply-side Explanations of Geographic Variation In Healthcare Use

The existence of significant regional variation in health care utilization has been well documented over the past 40 years. Yet considerable uncertainty persists about whether this variation is primarily the result of supply-side or demand-side forces. We use a model of physician market power to derive an empirical test of supply-side explanations. Specifically, we examine changes in the use of healthcare by the near-elderly across regions that differ in Medicare spending levels as they transition from being uninsured into Medicare. Estimates indicate that gaining Medicare coverage in an above-median spending region is associated with a 48% increase in the probability of at least one hospital visit and a 26% increase in the probability of having more than five doctor visits relative to similar individuals in below-median spending regions. These estimates suggest that supply-side factors can explain much of the observed geographic variation in Medicare spending.

Measuring income equity in the demand for healthcare with finite mixture models

Guaranteeing equity for the poor is a major challenge for healthcare systems in developed countries. Overall, equity is an ethical issue related to the judgments about healthcare accessibility. At the same time, an economic concept of horizontal equity deals with “an equal treatment for equal need” and “means that persons in equal need of medical care should receive the same treatment, irrespective of whether they happen to be poor or rich” . In practical terms, there is a general agreement about striving for “minimal variation of [healthcare] use with income” and ensuring equity for the poor. According to theoretical predictions, a well-designed social health insurance system may provide an equitable redistribution of medical care between the rich and the poor. However, the actual performance of social health insurance systems with respect to guaranteeing equity for the poor is an ultimately empirical question

The paper exploits panel data finite mixture (latent class) models to measure consumer equity in healthcare access and utilization. The finite mixture approach accounts for unobservable consumer heterogeneity. Additionally, we employ the generalized linear models with latent classes to address a retransformation problem of logged dependent variable. Using the data of the Japan Household Panel Survey (2009-2014), we discover that consumers separate into latent classes in the binary choice models for healthcare use and generalized linear models for outpatient/inpatient healthcare expenditure. The results reveal that healthcare access in Japan is pro-poor for the most sick consumers, while utilization of outpatient care is equitable with respect to disposable income. The novelty of the paper is twofold. Firstly, we examine inpatient and outpatient healthcare access, and analyze expenditure within health insurance, exploiting the longitudinal data of the Japan Household Panel Survey. The unique feature of the survey is the fact that it distinguishes between non-users of healthcare, the users of inpatient and outpatient care, and provides a wide range of consumer characteristics, such as health status, index of psychological distress and life-style variables. Secondly, we measure income inequity with the generalized finite mixture models for healthcare use in the longitudinal context. It may be noted that the applicability of the finite mixture models for analyzing healthcare demand is well established. However, the use of generalized finite mixture models for measuring healthcare expenditure is often limited to experimental literature or cross-sectional estimates. The results of our estimations indicate that consumers separate into two latent classes in the binary choice models for use of any care or inpatient care, as well as in the loglinear and generalized linear models for outpatient and inpatient healthcare expenditure. The classes may be naturally interpreted as most frequent and most seek consumers (“high users”), infrequent and most healthy consumers (“low users”), and consumers with median use and median health status (“median users”).

Income Shocks and Health Care Decision among Single-Mother Families

Using two-year panel data from the Medical Expenditure Panel Survey (MEPS) for 2004 to 2012, we examine how income shocks affect health care spending decisions among single-mother families. This analysis focuses upon total out-of-pocket family health care spending as well as the allocation of such spending to a variety of specific health care services: dental care, vision care, prescription drugs, office-based visits, and emergency department visits. We hypothesize that the family’s response to income shocks is likely to be complex. On the one hand, families experiencing income loss may, by necessity, be required to prioritize their health care spending among specific health care services. On the other hand, loss of income may cause stress and anxiety which can have a negative impact on the health status of family members pressuring to families to maintain or even increase health care spending. Since a substantial proportion of the population does not use particular health care services and since the distribution of spending is positively skewed, we estimate a series of two-part-model health care spending models. These models are specified with probit equation for the likelihood of an expenditure in the first part of the model, and a generalized linear model (GLM) with a log link and gamma or inverse Gaussian variance function in the second part for families with positive out-of-pocket spending. To control for unobserved heterogeneity across families in the sample, we estimate the two-part model using the correlated random effects framework. There are several important finding in this study. First, we find that single-mother families experiencing an income loss tend to reduce their out-of-pocket health care spending. This does not necessarily imply a decline in health services utilization. For instance, a middle-income single-mother family that becomes a low-income family decreases its total out-of-pocket spending by an average of $585 annually but increases a likelihood of any health service use (with or without cost-sharing) increases by an average of 2.7 percentage points. Second, we find that an income loss among low-income single-mother families is associated with a decrease in out-of-pocket prescription drug spending by between $67-135 annually. Third, we find that some families appear to reallocate their health care spending in response to an income shock. For instance, high income single-mother families that become middle income families tend to increase their out-of-pocket prescription drug spending by an average of $82 while decreasing out-of-pocket spending for office-based visits by an average of $109. Fourth, we also find that that income loss among single-mother families is associated with a statistically significant decline in out-of-pocket spending toward emergency department visits. However, these declines are small in size. Finally, we find no statistically significant effects of income loss on dental care out-of-pocket spending among single-mother families.

The Impact of High Deductible Health Insurance Plans on Spending and Enrollee Behavioral Response

I assess the extent to which high deductible health plans (HDHPs) reduce ex post moral hazard. Recently, HDHPs have become commonplace in the employer insurance market; however, the effect of adding an HDHP option into an individual’s offer set remains understudied. This paper answers three questions regarding HDHPs. First, do HDHPs lower total medical spending, and is there a behavioral response or simply a shifting of costs to the individual? I find HDHPs lower spending by 16 percent and reduce utilization as predicted by demand theory. Second, I find reductions in hospital-based medical care spending account for 60 percent of the savings. Finally, contrary to recently published papers, I find evidence of discriminatory cutbacks in service utilization and no evidence that these cutbacks impact health outcomes.





Occupational Licensing and the Causes and Consequences of Patient Sorting: a Machine Learning Approach

The focus of this paper is an important (but understudied) driver of health spending: patients' sorting decisions over type of health care provider from which to obtain care. Occupational licensing restrictions ("scope of practice (SOP) laws") play a major role in these decisions by specifying the scope of treatment performed by various provider types. These restrictions vary across states, and several states have relaxed their laws in recent years with more states planning to do so in the future. Efficient patient sorting will become increasingly important given two recent trends in health care: (1) the rise of non-physician healthcare providers (such as nurse practitioners and physician assistants), and (2) the large predicted physician shortage (created in part by the ACA). As non-physicians (NPs) become more accessible, more patients will be able to choose whether to receive care from an NP instead of a physician (MD), and the efficiency of these choices will become a greater focus of health care policy. The increasing relaxation of SOP licensing laws is one factor contributing to the increased access to NPs. Similarly, in the presence of a physician shortage, many patients will be forced to sort to NPs to receive care. Thus, efficient sorting is an increasingly crucial component of several general health care policy issues. In this paper I study the relationship between patient sorting, SOP laws, and health care costs, disparities, quality, and outcomes. I first document both the types of patients and the types of care that sort to NPs versus MDs. I then take a machine learning approach to study the efficiency of sorting in the context of patient mis-prediction of personal risk and complexity of required treatment. Next, I exploit three specific natural experiments to study the types of care that are on the margin between provider types. The three experiments each affect the access to NPs relative to MDs, but through different channels. They are: the state-specific relaxation of SOP laws, insurer changes in relative copays between provider types, and a large government subsidy for the training of NPs at five different US medical schools (the Graduate Nurse Education Demonstration). I combine these experiments with the machine learning results on mis-prediction to estimate a personalized, claim-level measure of the effect of receiving care from an MD versus an NP. I show the correlation between this effect and the patient's algorithm-generated predicted risk of an adverse outcome. The final portion of the paper is devoted to understanding the broader implications of patient sorting over provider types on explanations of empirical facts in the health economics literature. I estimate the predicted effects of counterfactually altering SOP laws and I show that patient sorting plays an important role in health care costs, disparities, quality, and outcomes.

The Impact of Online Physician Ratings on Patients' Choices

With the rising of Internet, more and more websites are providing review information on health care providers. Differing from the traditional report cards, these reviews and ratings are typically written by patients themselves. Therefore, these reviews are easier for patients to understand and also addresses more of patients concerns. An obvious trend in recent years is that a growing number of patients rely on the information from these review websites to choose health care providers. We exploit the physician ratings from Vitals.com, one of the largest and most comprehensive physician-review websites in US, and inpatient claims data of coronary artery bypass graft (CABG) surgeries in Pennsylvania to examine the impact of online physician ratings on patients’ physician choices. Using a discrete choice model with random coefficients, we find that the probability that patients receive CABG surgery from high-rating surgeons is significantly higher than that from surgeons without ratings, and the probability that patients receive CABG surgery from low-rating surgeons is significantly lower than that from surgeons without ratings.

Nudges and Incentives for HIV Testing: A Field Experiment in Ecuador

Many individuals with HIV/AIDS are not receiving treatment, in part because they are not aware of their status. The CDC and other health agencies recommend that all individuals be routinely tested for HIV/AIDS. Underdetection is particularly concerning in low- and middle-income countries because the transmission of the disease can stretch scarce public health resources. We conduct a randomized controlled field experiment in Ecuador, in a province that carries a disproportionate burden of HIV/AIDS. The overall goal of the study is to compare the effects of different strategies, namely information, a behavioral nudge (soft-commitment), and a $ 10 financial incentive (paid either at the time of testing or when the participant picks up their test results) in inducing voluntary HIV testing. In our study, we test these various strategies on a broad target population recruited in several well-transited locations in a major city in the province. Behavioral nudges and rewards have the potential to induce individual testing by overcoming psychological biases or bridging information gaps, and by overcoming social stigma concerns. Participant recruitment is in progress and is expected to be completed by December, 2017. Outcomes include percentage of participants deciding to get tested, percentage of participants picking up their test results, and percentage of participants being diagnosed with HIV. Preliminary results indicate that: 1. About 15% of subjects provided with "information only" agreed to get tested; 2. The "soft-commitment" opportunity did not have additional effects; 3. The $10 incentive paid at the time of testing increased the fraction of subjects who got tested to 60%; 4. The $10 incentive paid when the participants picked up their test results, instead, did not show any additional effect; 5. Between 1.5% and 2% of individual tested were HIV positive; 6. About 40% of non-incentivized subjects chose to learn their test results, vs. 20% of participants who received the incentive at the time of testing. Our preliminary results indicate that incentives provided at the time of testing can overcome economic or psychological barriers to get tested, although the relatively low proportion of incentivized subjects who chose to pick up their test results suggests that other strategies need to be devised to motivate individuals to learn their HIV status.

“Fearing Ebola? Get a Flu Shot”: Impact of Ebola on influenza vaccine uptake in the US

The start of the 2014-15 influenza season was overshadowed by the fear of Ebola. As the first Ebola patient diagnosed in the US died and two nurses confirmed the infection in October 2014, the anxiety of Ebola was elevated. As part of the massive media coverage on Ebola, many health experts compared Ebola with influenza and pointed out the importance of receiving influenza vaccination in that year. The rationale was that Ebola-related hospitalization and deaths were far less than those caused by influenza and having more people receiving influenza vaccines would reduce false alarms and help the public health system respond to Ebola. Meanwhile, CDC also recommended people being actively monitored for potential Ebola virus exposure to receive influenza vaccines if they had not done so. It is unclear, however, whether the public responded to these messages. This study examines the impact of public awareness around Ebola on influenza vaccine uptake during the 2014-15 influenza season in the US. We use individual-level data from the Behavioral Risk Factor Surveillance System and examine changes in the likelihood of receiving influenza vaccines. We examine the robustness of these findings in a doubly-robust difference-in-difference with propensity score matching framework as well as a synthetic control framework. We examine the impact of three types of treatment: residing in a state with an Ebola case (Texas and New York), residing in a state with Ebola or medically evacuated cases (Texas, New York, Georgia, Maryland, and Nebraska), or having higher level of attention to Ebola as measured by web searches. Our findings provide the first quantification of the spillovers from messaging targeting one rare disease to health behaviors related to a second, more common infection. Future immunization programs that consider phrasing their promoting messages by relating influenza to threatening diseases such as Ebola would be able to predict the impact of this strategy based on our results.




Does the framing of patient cost-sharing incentives matter? The effects of deductibles vs. no-claim refunds

Research question and motivation In light of increasing health care expenditures, patient cost-sharing schemes have emerged as one of the main policy tools to reduce medical spending. In this study we show that health care utilization is affected not only by the economic incentives provided by cost-sharing schemes, but also by the way these economic incentives are presented. Specifically, we compare patients’ responses to a deductible and to a no-claim refund. The economic incentives under a deductible and a no-claim refund are very similar, but they are framed in a different way. Under a deductible policy, individuals pay out-of-pocket for all medical care up to the deductible limit. Under a no-claim refund policy, individuals receive a payment at the end of the year if their health care spending during the year was below the no-claim refund limit. Prospect theory predicts that individuals respond stronger to losses than to gains. If individuals perceive deductible payments as losses and lower no-claim refunds as foregone gains then we might expect that individuals will react stronger to deductibles than to no-claim rebates. Data and methods We make use of the fact that in the Netherlands, both schemes have been in place at different points of time while the patient population and the services covered by health insurance remained comparable. In the years 2006 and 2007 Dutch law has mandated that health insurance contracts included a no-claim refund, and from the year 2008 onward, health insurance contracts had to feature an annual deductible. Our analysis is based on unique claims-level data from a Dutch health insurer for the years 2006-2015 which we aggregate to around 9 million person month observations. In our empirical strategy we exploit variation in cost-sharing incentives within a year. Under both a deductible policy and a no-claim refund the price of healthcare utilization can vary over the course of the year depending on whether or not an individual has exceeded her deductible or no-claim refund limit. We examine how the reaction to prices differs between the years when a no-claim refund policy was in place and the years when a deductible policy was in place. We account for the possible endogeneity of prices with a simulated instrumental variables approach. As instrumental variable for the price at the beginning of the month we use a simulated average price for people with the same risk score decile, age, and gender in a given year. Results and conclusions We find that patients react to comparable incentives twice as strongly when they are implemented as a deductible, which suggests that the framing of incentives can be quantitatively almost as important as the incentive itself. Our preferred explanation is that individuals are loss-averse and respond differently to both schemes because they perceive a deductible payment as a loss and a no-claim refund as a gain. Our results are robust to a number of sensitivity analyses. Specifically, our results cannot be explained by differences in the timing of payments, or by end of year effects.

Using Insights from Behavioral Economics for the Design of Financial Incentives Improving Medication Persistence - An Experimental Analysis

Objectives We develop a financial incentive scheme based on the concept of loss aversion to improve persistence behavior, a primary target of efforts to improve health outcomes for patients with chronic disease. According to the conceptual framework of medical persistence by Djawadi et al. (2014) a combination of loss aversion and mental accounting operations dynamically influences patients’ cost-benefit assessments. In the beginning of the treatment patients take the medicine without experiencing any improvements. Once health state improvements evolve patients comply with medication to compensate the losses of their previous health investments, but gradually discontinue with therapy, as soon as these losses are compensated. Methods We design a conventional economic laboratory experiment which simulates the course of events inherent in medical treatments from an economic perspective. Our experiment consists of two stages. The working stage mimics the beginning of the treatment and induces feelings of losses as subjects have to work on a task but only receive a fraction of their proper income. Entering the investment stage with these losses subjects decide over 12 periods between lottery A and lottery B. These lotteries represent the economic consequences of discontinuing and continuing with therapy. Lottery A with a higher risk of losing money can be chosen without any prior investments whereas for playing Lottery B with higher winning chances subjects have to invest some of their monetary endowment. Once a lottery is lost subjects drop out of the experiment and are not allowed to make any more decisions. We incorporate loss aversion and the timing in our incentive scheme in the following way: as soon as subjects have compensated the losses from the working stage they receive an up-front bonus which is added to their balance account. Subjects are only allowed to keep this bonus if they do not dropout of the experiment before the last period. Results Our persistence measure is based on the lottery choices A and B. We define the persistence rate as the ratio of lottery B over lottery A choices. We find that persistence rates in the incentive treatment and the baseline sample of Djawadi et al. (2014) are almost equally high in early periods, but from period 7 on where subjects compensated their losses, significantly higher persistence rates are observed in the incentive treatment (Log Rank χ² =34.69 ; p<0.0001). We further compare this behavioral pattern with an additional control treatment which does not provide any losses in the working stage and thus serves as an upper bound for high persistence rates. We find that persistence rates in the incentive treatment are significantly higher than in the control treatment (Log Rank χ² = 28.91 ; p<0.0001), indicating that the bonus not only mitigated the steady decline of persistence behavior but rather encouraged subjects to continue steadily with lottery B until the end of the experiment.

On-the-job Treating: Patient Response to a Reduction in Time Cost through Worksite Health

Innovative organizational forms of health care delivery have recently developed that lower the time cost of care. While there is an extensive literature on consumer response to changes in the out-of-pocket price for care, little work has studied the equally-salient dimension of time cost. In this paper, I develop a theoretical model of patient decision-making and predict that when a new provider enters the market and offers services with lower time cost, patients engage in new utilization and/or substitution across providers. I then test these predictions in a unique empirical setting: A large corporation opened a worksite health clinic on its California campus in 2013, but did not feature a clinic on its Texas campus. I utilize novel data, 2011-2015 medical claims for the corporation’s employees. My primary empirical strategy is a difference-in-differences approach, where I compare California employees to the control group of Texas employees. I find that the effects of clinic availability are concentrated among the narrow set of services that can be provided onsite. For primary care in particular, I observe both an increase in utilization and substitution towards onsite care. While services beyond the clinic’s scope of practice are mostly unaffected, California employees reduce their utilization of outpatient care; this spillover effect is offset by a substantial increase in demand for office-based care. Ultimately, new consumption of primary care and other office-based services drives a small increase in spending. For example, at the 70th percentile of the conditional distribution, the estimated increase in monthly total spending is $14.67. My findings suggest that consumer demand is sensitive to changes in time cost, and this has important implications for the potential welfare benefits of providing patients with convenient access to high-value services.



The Effects of Financial Incentives on Intrinsic Motivation for Health Behaviors

Objective: To assess whether financial incentives for health behaviors crowd out individuals’ intrinsic motivation to engage in those behaviors. The use of financial incentives to promote changes in health behaviors is widespread among payers and employers, however there are concerns that if incentives crowd out intrinsic motivation, behavior would fall below even pre-incentive levels following the removal of incentives, hindering any long-run impact of incentives on behavior change. Further, consumers’ health-related decisions are likely impacted by the interaction between incentives and motivation. Few studies have assessed the impact of financial incentives on patients’ intrinsic motivation for health behaviors using direct measures of motivation. We examined this question in the context of five randomized controlled trials of financial incentives for health behavior change. We investigated whether effects varied by incentive type or behavior and assessed whether baseline or changes in motivation were associated with performance in behavior change programs. Setting: We used the Treatment Self-Regulation Questionnaire to measure intrinsic motivation at baseline and at least once following the incentive intervention period in randomized controlled trials of financial incentives for weight loss (two studies), home health monitoring, walking among older adults, and adherence to use of a Positive Airways Pressure device for sleep apnea. In addition to varying health behaviors, these trials utilized different forms of incentives, including conditional payments, regret lotteries, and deposit contracts. Methods: Multivariate regressions with participant-level data and random effects were used to assess the relationships between baseline and change in intrinsic motivation and performance in each study, measured as achieving study goals. Similar analyses were used to examine the effect of incentive eligibility and receipt on changes in intrinsic motivation to test for crowding out.

Sample: 561 participants in five randomized controlled trials of financial incentives for health behavior change. Results: First, we found that an increase in intrinsic motivation during the intervention was associated with increased odds of success in the program, defined as achieving program goals such as a pre-determined weight loss target. Second, we found no evidence of crowding out of intrinsic motivation by incentives; that is, there was no significant association between incentive eligibility or receipt and the odds of a decrease in intrinsic motivation pre- versus post-incentives. The lack of evidence of crowding out was consistent across all five studies. (Further sub-group analyses to examine heterogeneity in our results as well as sensitivity checks are forthcoming.) Conclusions: Financial incentives did not crowd out intrinsic motivation across a range of health behaviors and incentive designs. Improving our understanding in this area is critical in order to understand consumer decision-making in the context of health behaviors as well as to design the most effective incentives, understand in what settings they are most likely to work, and improve the potential for long-run behavior change.

Offering Vouchers to Low-Income Minority Populations Increases Follow-up for Free Glaucoma Services

Purpose: Despite limited availability of free eye care services in Baltimore City, utilization by low-income at-risk minority individuals remains low. This may be driven by a lack of perceived value for these free services. We examine the effect of providing vouchers redeemable for free eye care services on uptake among participants in the ongoing SToP Glaucoma study. Methods: A cluster randomized trial was conducted within the SToP Glaucoma study, an investigation of a community-based screening program which identifies glaucoma suspects and offers them free follow-up appointments at the Wilmer Eye Institute. Appointments are scheduled at the time of screening and reminder calls are made for all patients. Screening events were randomized to standard verbal and written counseling offering the individual a free appointment, or counseling in addition to provision of one of two types of vouchers redeemable for free appointments. Both voucher types included the patient’s name, the appointment date and an expiration date 90 days following the screening. One also included the approximate monetary value of the service ($250). The primary outcome was presenting for follow-up within the voucher eligibility period. A hierarchical mixed-effects logistic model allowing for random effects from the screening event was used to assess the effects of each voucher type on presentation to a follow-up appointment. Data collection is ongoing. Results: Follow-up through November 2017 yielded complete data for 431 glaucoma suspects identified at one of 64 screening events. Overall, 76% of individuals were African American, 65% were female, and the mean age was 69. There were no significant differences in these factors between study arms. Those referred in the traditional manner had a 49% attendance rate, whereas 67% of individuals receiving a voucher without monetary value information and 62% of individuals receiving a voucher with monetary value information presented. For a given screening event, offering vouchers without monetary value information increased the odds of presenting for follow-up by 152% (p =.03) compared to not offering a voucher. Offering vouchers with monetary value information increased the odds of presenting by 112%, but this effect was not significant (p=0.09). Conclusions: Offering vouchers for redemption of free eye care services increases utilization. Voucher provision may increase perceived value for these services, particularly among low-income minority populations. Further investigation into the elements of vouchers that drive this effect is warranted.




Initiation and Sustainable Preventative Healthcare Utilization among Aging Couples: Asymmetric Spousal Influences and Behavioral Learning

Research Objective: This study investigates the life partners’ influence on each other’s preventative health service usage among those over age 50, and further, disentangles the active learning and passive imitation channels among spousal concordance. The study also explores how behavioral traits such as risk aversion and time discounting would mediate the learning channel.

Population Studied: The study uses a biennial and representative dataset, Health & Retirement Study (HRS) 2006-2014. Four study cohorts are included, HRS (born between 1931-1941), WB (war baby, 1942-1947), EBB (early baby boomer, 1948-1953) and MBB (middle baby boomer, 1954-1959). A sample of about 19,362 respondents is yielded for the study.

Study Design: The study focuses on individuals with stable life partners throughout the study period. The outcome variables are whether or not a respondent engaged in (initiation or drop-out) preventative activities over years, i.e. a flu shot, a blood test for cholesterol screening and rigorous physical activities. In order to capture positive and negative partner influences, a life partner’s previous usage pattern is classified into one of four categories: Non-User, those life partners who did not use a certain activity for two consecutive waves; Stopper, those who used in earlier wave but stopped using in later wave; Continuous, those who used in both waves; and Starter, those who did not use in earlier wave but started to use in later wave. Self-reported health status changes and physician verified new chronic conditions for both self and the partner are used as proxies of health outcomes to gauge the learning channel. The main research of interest is the extent to which the interactions between the spousal behaviors and health outcomes would predict preventive activity engagement and changes. Risk-aversion and time-discounting are also used to explore the potential mediation effects.

Preliminary Findings: Couples show concordance in preventive behaviors, and further, start using a preventative activity by a life partner encourages the other couple more than stop using would discourage. The learning channel is supported by the evidence that worse-off self-health and better-off spousal-health exaggerate the encouraging effects, while better-off self-health and worse-off spousal-health exaggerate the discouraging effects. Moreover, those with higher risk-aversion and time-discounting (less patient) are more responsive to learning and behavioral changes.

Conclusions & Implications: Findings in the study suggest that the partners' influences are asymmetric and happen through active learning. The study provides insights on the potential efficacy of family-based promotion strategies to increase preventative activity engagement, as well as the perfect timing of intervention. Moreover, this study also sheds light on patient-level medical and healthcare decision-making in general beyond the preventative stage.

Commit to Change, or Change Your Commitment? Dynamic Demand for Goal Difficulty

In two studies—an online experiment using a real effort task and a field study of weight loss among Weight Watchers members—we investigate how demand for difficult goals changes over time in the presence or absence of goal-contingent incentives. To facilitate motivation and self-control in an effortful task, people may choose aggressive goals. However, because individuals in incentive conditions only earn bonus payments when they achieve their goals, they simultaneously have an economic incentive to select (and potentially modify over time) performance and weight loss targets that can be easily reached. In Study 1, we recruited online workers to complete three trials of a computerized effort task and asked them to set performance goals for each trial. Some workers were offered bonus incentives that were contingent upon goal achievement, thus allowing these participants to ensure incentive collection by choosing an easy goal or to put their incentive at risk by selecting a difficult goal. Other workers were offered equivalently-sized incentives that were independent from goal achievement, whereas yet other workers received no bonus incentives. We found that the provision of goal-contingent incentives reduced demand for difficult goals relative to conditions in which no incentives were offered or in which incentives were not contingent upon goal achievement. That is, people behaved rationally in selecting goals when incentives were made contingent upon goal achievement in a context where participants likely had little intrinsic motivation regarding the task itself (a typing task). These differences in goal selection among the conditions persisted over multiple trials, even as participants selected more difficult goals on later trials overall. In Study 2, we partnered with Weight Watchers to conduct a field experiment and to determine how contingent incentives influence the selection of weight goals over a six-month period. 191 Weight Watchers members were randomly assigned to either a control condition of daily weigh-ins and daily feedback, or to an incentive condition with daily weigh-ins, daily feedback, and a daily financial incentive of $2.80 for meeting a weight loss goal. Participants selected weight loss goals each month. Despite having an economic incentive to choose the most modest goal possible, nearly 90% of incentivized participants picked an initial goal that was more challenging than the minimum needed. However, as people gained experience with goal-setting procedures and their own patterns of weight change over time, those offered monetary incentives were more likely to shift to less aggressive goals. Across both studies, we find substantial initial demand for goals that are more ambitious than strictly required, regardless of incentive provision. In the absence of goal-contingent incentives, this demand for difficult goals remains high over time. However, when considering commitment devices in which monetary gains are foregone if goals are not met, people are responsive to the likelihood of goal achievement and demonstrate dynamic demand over iterated trials.

Do Financial Incentives for Medicaid Beneficiaries Increase Use of Preventive Services and Lower Expenditures? Evidence from Ten States

Medicaid beneficiaries face substantial burden from chronic diseases. Incentive programs have been proposed to connect and engage Medicaid beneficiaries with preventive services, thereby changing behavior and preventing chronic diseases. We evaluated the impact of the Medicaid Incentives for Prevention of Chronic Disease program, which funded Medicaid incentive initiatives in 10 states. The focus of initiatives varied by state and included smoking cessation, diabetes prevention, weight loss, and disease management. The value of incentives also varied widely, ranging from a maximum of $50 in one state to over $1,000 annually in another state. Most states randomly assigned participants to incentive and control arms, with both arms eligible to receive preventive services. We conducted a mixed methods evaluation that included document review, site visits, focus groups, a beneficiary survey, and Medicaid claims analysis. The claims analysis used regression and difference-in-difference methods to test whether incentives increased the use of preventive services and lowered use of other Medicaid services and expenditures. We find that states can implement Medicaid programs successfully, although they had to overcome administrative challenges and recruiting participants was more difficult than anticipated. Participants in the incentive arms attended significantly more diabetes prevention and weight watchers classes, made more smoking quitline calls, and attended more smoking cessation sessions. Participants who received incentives were very satisfied with program access and quality and believed that the incentives helped them meet their health goals. The programs had no significant impact on the use and costs of other Medicaid services during the evaluation. Administrative costs were relatively high. State evaluations suggest that the incentives had mixed effects on short-term health outcomes, with some significant reductions in smoking and generally insignificant effects on diabetes prevention, weight loss, and risk factor management. Future study is needed to determine whether the impact of Medicaid incentive programs on health outcomes can be improved.




Tell me where to get my flu shot - Effects of information in prevention

From a public health perspective, the flu vaccination rate is viewed as too low in many countries. While up until recently, only physicians were allowed to vaccinate in Switzerland, some pharmacies received the license to vaccinate during the course of the year 2016. Public health authorities hope to increase the vaccination rate as there is generally no need for an appointment in a pharmacy. This setting allows us to test whether the vaccination rate can be increased by a letter informing consumers about the access to prevention care as well as whether consumers can be nudged to get their flu shot in a recommended pharmacy. We use a randomized control trial where a large Swiss health insurer sent letters to 22'000 of its customers (between 50 and 75 years of age). The letter included information about the new possibility to get a flu shot in a pharmacy, a pledge to bear the costs and an indication of the nearest pharmacy which had the license to vaccinate. The letter increased immunization rates by 2.7%-points (17.9%) with significant heterogeneity in background variables as well as with respect to the distance to the pharmacy. We find some provider substitution effects from physicians to pharmacists. Yet, the majority of the increase in the vaccination rate is driven by additional vaccinations in the pharmacies. More than half of the customers visited the recommended pharmacy even if a closer one had been available. Currently, we do not find any significant effects of the flu shot on covered health care expenditures nor on the number of doctor visits. Information letters proofed to be an effective tool to increase the use of prevention care as well as to guide consumers to specific health care providers. The new possibility to vaccinate in a pharmacy increased the take-up rate of the vaccination. However, the effectiveness of the flu shot remains currently doubtful.

The Impact of Connecticut’s Paid Sick Leave Law on Preventive Services

Do state laws that require the availability of paid sick leave increase the use of preventive services? Paid sick leave benefits allow workers to maintain job security when they leave for medical reasons. In recent years, paid sick leave laws have received more attention by health policy makers in terms of its potential to improve public health. Currently, seven states and D.C. have laws that require employers to provide paid sick leave. However, empirical evidence on this topic is limited (DeRigne, et al 2017; Peipins et al 2012). Although prior studies examined the association of paid sick leave and preventive service use, these studies did not use rigorous study designs. Because workers with higher demand for healthcare services may select into firms which offer better health benefits including paid sick leave, the estimated association in these previous studies may suffer from selection bias. The goal of this study is to use a quasi-experimental study design to estimate the impact of Connecticut’s 2012 paid sick leave law on the use of preventive services. Connecticut was the first state to require private employers to offer paid sick leave benefits to their employees. Using state and time variation from 2007-2016 Behavioral Risk Factor Surveillance System (BRFSS) data, we compare the use of preventive services in Connecticut and in other New England states before and after the implementation of the 2012 paid sick leave law. For general preventive service outcomes, we examined routine checkups, flu vaccinations, and dental visits in the past year. We also examined the use of Pap tests, clinical breast exams, and mammograms for women. Overall, we found that Connecticut’s 2012 paid sick leave law increased the use of preventive services. Specifically, the rate of routine checkups (1.4%, p<.1), flu shots (2.0%, p<.05), and dental visits (2.7%, p<.01) increased over this period. With respect to cancer screening outcomes for women, we found that the use of Pap tests (9.4%, p<.01), and clinical breast exams (4.4%, p<.01) increased. However, the higher rate of mammograms (1.3%, p=.38) was not statistically significant. Our findings provide rigorous evidence on the positive impact of a state’s paid sick leave law on preventive service outcomes. These empirical results also suggest that a lack of paid sick leave among some private employers may represent a barrier to accessing preventive services. Finally, policymakers can use legislation to support the use of preventive services and improve the health and productivity of workers.

Is Less More? The Effect of Simplifying Plan Information on Medicare Part D Choices

Research Objective: Medicare Part D enrollees frequently make suboptimal plan choices, typically leaving hundreds of dollars on the table every year. This overspending is largely driven by systematic errors in the way enrollees assess the value of plan attributes (e.g., over-valuing premiums, under-valuing expected out-of-pocket [OOP] costs), despite the existence of decision support tools that facilitate accurate valuations through the provision of personalized cost estimates. In particular, CMS’s Plan Finder tool has had limited effects on consumer choices, in part because the complexity and presentation of the information it provides limits the salience of the personalized cost estimates. Simplifying Plan Finder has the potential to improve plan choices. This study uses a survey-based randomized experiment to examine the effect of simplifying the financial information presented on Plan Finder on the way individuals choose Part D plans. Methods: We used the American Life Panel, a nationally representative internet panel, to field an experiment among 1,278 adults age 55+. Participants made simulated Part D plan choices on behalf of a friend with a stated preference for minimizing total drug spending. Respondents were randomized into 4 study arms (1 control, 3 treatment). The control group was shown a plan menu which mimicked the current Plan Finder tool. Total cost estimates were displayed for each plan alongside detailed financial information (premiums, deductibles, copays) and non-financial information (e.g., 5-star quality scores, pharmacy network size). The 3 treatment groups, which varied in the amount of financial information shown by default, were as follows: 1) total cost only; 2) total cost and premium; 3) total cost, premium, and estimated OOP cost. All treatment groups could view full financial information by clicking a link within the plan menu. Plan choices were evaluated using discrete choice analyses and conditional logit estimation. Differences in the decision weights placed on plan attributes across study arms were evaluated to test the effect of Plan Finder simplifications on the trade-offs individuals make when selecting a plan. Results: Simplifying financial information resulted in the selection of lower cost plans in all treatment groups relative to the control group. These improvements were largely driven by increases in the weight placed on OOP costs and reductions in the weighting of deductibles. Respondents in the “total cost, premium, and OOP cost” group had decision weights that were most consistent with cost-minimization: they placed equal weight on premium and OOP costs and no additional weight on cost-sharing attributes beyond their direct impact on anticipated spending. The “total cost and premium” group continued to weight premiums more than OOP costs, while the “total cost only” group was more likely to seek additional plan information which resulted in plan choices more like those of the control group. Implications: Simplifying the financial information on Plan Finder helps individuals adhere to a cost-minimization decision rule. Displaying plan premiums and OOP costs alongside total cost estimates was most effective in encouraging cost-minimization and reducing additional information seeking, suggesting that this format improves individuals’ trust of the information and ability to use it.



Social Learning in Health Insurance Choices: Evidence from Employer-Sponsored Health Plans

Research has documented that consumers often have imperfect information about the health insurance plans from which they are asked to choose; but we know less about the sources of that information. Given the difficulty in obtaining reliable information from independent sources, consumers may draw on their peers for recommendations. This paper investigates the role of social learning in health insurance selection, using longitudinal data from the University of California on plan choices of employees and peers in their department. The data from 2011 to 2016 span a major change in the insurance choice set, which aids in the statistical identification of social learning effects among both incumbent employees as well as new hires. I start by documenting the high similarity in plan choices within peer groups, suggesting the possibility of strong peer effects, and then use a variety of approaches to test for potential confounding from unobserved heterogeneity. I employ a discrete choice conditional logit estimator to formally model plan choice behavior, finding that a 10 percentage point increase in the share of peers who select a particular insurance plan will lead to a 14 percentage point increase in the probability that an individual will choose the same plan. This large effect on plan choice is equivalent to lowering the monthly premium by 18 percent. I then use this model to simulate employer strategies that could exploit social learning to better promote the employer’s insurance objectives. For illustration, I conduct counterfactual analyses of incentives to promote adoption of a new consumer-driven insurance. At the actuarially fair premium in this setting, demand for a consumer-driven plan is low, and social learning further discourages take-up. However, with sufficient premium subsidies, the model projects that the social learning effects will become positive and can be harnessed by employers to more effectively achieve their cost and insurance coverage goals.

Physician retirement, practice closures and discontinuity of primary care – What are the causal impacts on patients?

From the perspective of patients, the closing of a primary-care practice causes a discontinuity of care, which bears consequences for patients with long-standing doctor-patient relationships. First, interruptions in care may lead to inefficient utilization of healthcare services. Second, the literature consistently finds that continuity of care is beneficial for patients' health-related outcomes. Moreover, practice closures decrease the local availability of primary care, which disproportionally affects peripheral areas.

This paper studies closures of primary-care practices in Switzerland from 2005 to 2015 to estimate the causal impacts of discontinuities of primary care on patients' utilization patterns, medical expenditures and health-related outcomes. Employing a difference-in-difference framework, we identify causal effects by comparing changes in outcomes between an affected group of patients (‘treatment group’) and an unaffected group that does not experience changes in primary care provision (‘control group’). Our main findings are twofold. First, when faced with a discontinuity of primary care, patients adjust their utilization pattern by shifting visits away from ambulatory primary care providers (-5%) towards specialized care (+10%) and emergency departments of hospitals (+14%). Secondly, practice closures increase patients’ total health care expenditures by 4.6% and raise the probability of incurring non-zero costs in a given month by 10%. Two policy-relevant implications are that practice closures may lead to an inefficient use of healthcare services and have adverse effects on social health insurance which must cover higher costs. These implications are relevant for health planners and policy makers, health insurers and physicians.

Abstract


Guaranteeing equity for the poor is a major challenge for healthcare systems in developed countries. Overall, equity is an ethical issue related to the judgments about healthcare accessibility. At the same time, an economic concept of horizontal equity deals with “an equal treatment for equal need” and “means that persons in equal need of medical care should receive the same treatment, irrespective of whether they happen to be poor or rich” . In practical terms, there is a general agreement about striving for “minimal variation of [healthcare] use with income” and ensuring equity for the poor. According to theoretical predictions, a well-designed social health insurance system may provide an equitable redistribution of medical care between the rich and the poor. However, the actual performance of social health insurance systems with respect to guaranteeing equity for the poor is an ultimately empirical question

The paper exploits panel data finite mixture (latent class) models to measure consumer equity in healthcare access and utilization. The finite mixture approach accounts for unobservable consumer heterogeneity. Additionally, we employ the generalized linear models with latent classes to address a retransformation problem of logged dependent variable. Using the data of the Japan Household Panel Survey (2009-2014), we discover that consumers separate into latent classes in the binary choice models for healthcare use and generalized linear models for outpatient/inpatient healthcare expenditure. The results reveal that healthcare access in Japan is pro-poor for the most sick consumers, while utilization of outpatient care is equitable with respect to disposable income. The novelty of the paper is twofold. Firstly, we examine inpatient and outpatient healthcare access, and analyze expenditure within health insurance, exploiting the longitudinal data of the Japan Household Panel Survey. The unique feature of the survey is the fact that it distinguishes between non-users of healthcare, the users of inpatient and outpatient care, and provides a wide range of consumer characteristics, such as health status, index of psychological distress and life-style variables. Secondly, we measure income inequity with the generalized finite mixture models for healthcare use in the longitudinal context. It may be noted that the applicability of the finite mixture models for analyzing healthcare demand is well established. However, the use of generalized finite mixture models for measuring healthcare expenditure is often limited to experimental literature or cross-sectional estimates. The results of our estimations indicate that consumers separate into two latent classes in the binary choice models for use of any care or inpatient care, as well as in the loglinear and generalized linear models for outpatient and inpatient healthcare expenditure. The classes may be naturally interpreted as most frequent and most seek consumers (“high users”), infrequent and most healthy consumers (“low users”), and consumers with median use and median health status (“median users”).

Using two-year panel data from the Medical Expenditure Panel Survey (MEPS) for 2004 to 2012, we examine how income shocks affect health care spending decisions among single-mother families. This analysis focuses upon total out-of-pocket family health care spending as well as the allocation of such spending to a variety of specific health care services: dental care, vision care, prescription drugs, office-based visits, and emergency department visits. We hypothesize that the family’s response to income shocks is likely to be complex. On the one hand, families experiencing income loss may, by necessity, be required to prioritize their health care spending among specific health care services. On the other hand, loss of income may cause stress and anxiety which can have a negative impact on the health status of family members pressuring to families to maintain or even increase health care spending. Since a substantial proportion of the population does not use particular health care services and since the distribution of spending is positively skewed, we estimate a series of two-part-model health care spending models. These models are specified with probit equation for the likelihood of an expenditure in the first part of the model, and a generalized linear model (GLM) with a log link and gamma or inverse Gaussian variance function in the second part for families with positive out-of-pocket spending. To control for unobserved heterogeneity across families in the sample, we estimate the two-part model using the correlated random effects framework. There are several important finding in this study. First, we find that single-mother families experiencing an income loss tend to reduce their out-of-pocket health care spending. This does not necessarily imply a decline in health services utilization. For instance, a middle-income single-mother family that becomes a low-income family decreases its total out-of-pocket spending by an average of $585 annually but increases a likelihood of any health service use (with or without cost-sharing) increases by an average of 2.7 percentage points. Second, we find that an income loss among low-income single-mother families is associated with a decrease in out-of-pocket prescription drug spending by between $67-135 annually. Third, we find that some families appear to reallocate their health care spending in response to an income shock. For instance, high income single-mother families that become middle income families tend to increase their out-of-pocket prescription drug spending by an average of $82 while decreasing out-of-pocket spending for office-based visits by an average of $109. Fourth, we also find that that income loss among single-mother families is associated with a statistically significant decline in out-of-pocket spending toward emergency department visits. However, these declines are small in size. Finally, we find no statistically significant effects of income loss on dental care out-of-pocket spending among single-mother families.

I assess the extent to which high deductible health plans (HDHPs) reduce ex post moral hazard. Recently, HDHPs have become commonplace in the employer insurance market; however, the effect of adding an HDHP option into an individual’s offer set remains understudied. This paper answers three questions regarding HDHPs. First, do HDHPs lower total medical spending, and is there a behavioral response or simply a shifting of costs to the individual? I find HDHPs lower spending by 16 percent and reduce utilization as predicted by demand theory. Second, I find reductions in hospital-based medical care spending account for 60 percent of the savings. Finally, contrary to recently published papers, I find evidence of discriminatory cutbacks in service utilization and no evidence that these cutbacks impact health outcomes.

The focus of this paper is an important (but understudied) driver of health spending: patients' sorting decisions over type of health care provider from which to obtain care. Occupational licensing restrictions ("scope of practice (SOP) laws") play a major role in these decisions by specifying the scope of treatment performed by various provider types. These restrictions vary across states, and several states have relaxed their laws in recent years with more states planning to do so in the future. Efficient patient sorting will become increasingly important given two recent trends in health care: (1) the rise of non-physician healthcare providers (such as nurse practitioners and physician assistants), and (2) the large predicted physician shortage (created in part by the ACA). As non-physicians (NPs) become more accessible, more patients will be able to choose whether to receive care from an NP instead of a physician (MD), and the efficiency of these choices will become a greater focus of health care policy. The increasing relaxation of SOP licensing laws is one factor contributing to the increased access to NPs. Similarly, in the presence of a physician shortage, many patients will be forced to sort to NPs to receive care. Thus, efficient sorting is an increasingly crucial component of several general health care policy issues. In this paper I study the relationship between patient sorting, SOP laws, and health care costs, disparities, quality, and outcomes. I first document both the types of patients and the types of care that sort to NPs versus MDs. I then take a machine learning approach to study the efficiency of sorting in the context of patient mis-prediction of personal risk and complexity of required treatment. Next, I exploit three specific natural experiments to study the types of care that are on the margin between provider types. The three experiments each affect the access to NPs relative to MDs, but through different channels. They are: the state-specific relaxation of SOP laws, insurer changes in relative copays between provider types, and a large government subsidy for the training of NPs at five different US medical schools (the Graduate Nurse Education Demonstration). I combine these experiments with the machine learning results on mis-prediction to estimate a personalized, claim-level measure of the effect of receiving care from an MD versus an NP. I show the correlation between this effect and the patient's algorithm-generated predicted risk of an adverse outcome. The final portion of the paper is devoted to understanding the broader implications of patient sorting over provider types on explanations of empirical facts in the health economics literature. I estimate the predicted effects of counterfactually altering SOP laws and I show that patient sorting plays an important role in health care costs, disparities, quality, and outcomes.



The start of the 2014-15 influenza season was overshadowed by the fear of Ebola. As the first Ebola patient diagnosed in the US died and two nurses confirmed the infection in October 2014, the anxiety of Ebola was elevated. As part of the massive media coverage on Ebola, many health experts compared Ebola with influenza and pointed out the importance of receiving influenza vaccination in that year. The rationale was that Ebola-related hospitalization and deaths were far less than those caused by influenza and having more people receiving influenza vaccines would reduce false alarms and help the public health system respond to Ebola. Meanwhile, CDC also recommended people being actively monitored for potential Ebola virus exposure to receive influenza vaccines if they had not done so. It is unclear, however, whether the public responded to these messages. This study examines the impact of public awareness around Ebola on influenza vaccine uptake during the 2014-15 influenza season in the US. We use individual-level data from the Behavioral Risk Factor Surveillance System and examine changes in the likelihood of receiving influenza vaccines. We examine the robustness of these findings in a doubly-robust difference-in-difference with propensity score matching framework as well as a synthetic control framework. We examine the impact of three types of treatment: residing in a state with an Ebola case (Texas and New York), residing in a state with Ebola or medically evacuated cases (Texas, New York, Georgia, Maryland, and Nebraska), or having higher level of attention to Ebola as measured by web searches. Our findings provide the first quantification of the spillovers from messaging targeting one rare disease to health behaviors related to a second, more common infection. Future immunization programs that consider phrasing their promoting messages by relating influenza to threatening diseases such as Ebola would be able to predict the impact of this strategy based on our results.

Research question and motivation In light of increasing health care expenditures, patient cost-sharing schemes have emerged as one of the main policy tools to reduce medical spending. In this study we show that health care utilization is affected not only by the economic incentives provided by cost-sharing schemes, but also by the way these economic incentives are presented. Specifically, we compare patients’ responses to a deductible and to a no-claim refund. The economic incentives under a deductible and a no-claim refund are very similar, but they are framed in a different way. Under a deductible policy, individuals pay out-of-pocket for all medical care up to the deductible limit. Under a no-claim refund policy, individuals receive a payment at the end of the year if their health care spending during the year was below the no-claim refund limit. Prospect theory predicts that individuals respond stronger to losses than to gains. If individuals perceive deductible payments as losses and lower no-claim refunds as foregone gains then we might expect that individuals will react stronger to deductibles than to no-claim rebates.

We make use of the fact that in the Netherlands, both schemes have been in place at different points of time while the patient population and the services covered by health insurance remained comparable. In the years 2006 and 2007 Dutch law has mandated that health insurance contracts included a no-claim refund, and from the year 2008 onward, health insurance contracts had to feature an annual deductible. Our analysis is based on unique claims-level data from a Dutch health insurer for the years 2006-2015 which we aggregate to around 9 million person month observations. In our empirical strategy we exploit variation in cost-sharing incentives within a year. Under both a deductible policy and a no-claim refund the price of healthcare utilization can vary over the course of the year depending on whether or not an individual has exceeded her deductible or no-claim refund limit. We examine how the reaction to prices differs between the years when a no-claim refund policy was in place and the years when a deductible policy was in place. We account for the possible endogeneity of prices with a simulated instrumental variables approach. As instrumental variable for the price at the beginning of the month we use a simulated average price for people with the same risk score decile, age, and gender in a given year. Results and conclusions We find that patients react to comparable incentives twice as strongly when they are implemented as a deductible, which suggests that the framing of incentives can be quantitatively almost as important as the incentive itself. Our preferred explanation is that individuals are loss-averse and respond differently to both schemes because they perceive a deductible payment as a loss and a no-claim refund as a gain. Our results are robust to a number of sensitivity analyses. Specifically, our results cannot be explained by differences in the timing of payments, or by end of year effects.

We develop a financial incentive scheme based on the concept of loss aversion to improve persistence behavior, a primary target of efforts to improve health outcomes for patients with chronic disease. According to the conceptual framework of medical persistence by Djawadi et al. (2014) a combination of loss aversion and mental accounting operations dynamically influences patients’ cost-benefit assessments. In the beginning of the treatment patients take the medicine without experiencing any improvements. Once health state improvements evolve patients comply with medication to compensate the losses of their previous health investments, but gradually discontinue with therapy, as soon as these losses are

We design a conventional economic laboratory experiment which simulates the course of events inherent in medical treatments from an economic perspective. Our experiment consists of two stages. The working stage mimics the beginning of the treatment and induces feelings of losses as subjects have to work on a task but only receive a fraction of their proper income. Entering the investment stage with these losses subjects decide over 12 periods between lottery A and lottery B. These lotteries represent the economic consequences of discontinuing and continuing with therapy. Lottery A with a higher risk of losing money can be chosen without any prior investments whereas for playing Lottery B with higher winning chances subjects have to invest some of their monetary endowment. Once a lottery is lost subjects drop out of the experiment and are not allowed to make any more decisions. We incorporate loss aversion and the timing in our incentive scheme in the following way: as soon as subjects have compensated the losses from the working stage they receive an up-front bonus which is added to their balance account. Subjects are only allowed to keep this bonus if they do not dropout of the experiment before the last period.

Our persistence measure is based on the lottery choices A and B. We define the persistence rate as the ratio of lottery B over lottery A choices. We find that persistence rates in the incentive treatment and the baseline sample of Djawadi et al. (2014) are almost equally high in early periods, but from period 7 on where subjects compensated their losses, significantly higher persistence rates are observed in the incentive treatment (Log Rank χ² =34.69 ; p<0.0001). We further compare this behavioral pattern with an additional control treatment which does not provide any losses in the working stage and thus serves as an upper bound for high persistence rates. We find that persistence rates in the incentive treatment are significantly higher than in the control treatment (Log Rank χ² = 28.91 ; p<0.0001), indicating that the bonus not only mitigated the steady decline of persistence behavior but rather encouraged subjects to continue steadily with lottery B until the end of the

Innovative organizational forms of health care delivery have recently developed that lower the time cost of care. While there is an extensive literature on consumer response to changes in the out-of-pocket price for care, little work has studied the equally-salient dimension of time cost. In this paper, I develop a theoretical model of patient decision-making and predict that when a new provider enters the market and offers services with lower time cost, patients engage in new utilization and/or substitution across providers. I then test these predictions in a unique empirical setting: A large corporation opened a worksite health clinic on its California campus in 2013, but did not feature a clinic on its Texas campus. I utilize novel data, 2011-2015 medical claims for the corporation’s employees. My primary empirical strategy is a difference-in-differences approach, where I compare California employees to the control group of Texas employees. I find that the effects of clinic availability are concentrated among the narrow set of services that can be provided onsite. For primary care in particular, I observe both an increase in utilization and substitution towards onsite care. While services beyond the clinic’s scope of practice are mostly unaffected, California employees reduce their utilization of outpatient care; this spillover effect is offset by a substantial increase in demand for office-based care. Ultimately, new consumption of primary care and other office-based services drives a small increase in spending. For example, at the 70th percentile of the conditional distribution, the estimated increase in monthly total spending is $14.67. My findings suggest that consumer demand is sensitive to changes in time cost, and this has important implications for the potential welfare benefits of providing patients with convenient access to high-value services.

To assess whether financial incentives for health behaviors crowd out individuals’ intrinsic motivation to engage in those behaviors. The use of financial incentives to promote changes in health behaviors is widespread among payers and employers, however there are concerns that if incentives crowd out intrinsic motivation, behavior would fall below even pre-incentive levels following the removal of incentives, hindering any long-run impact of incentives on behavior change. Further, consumers’ health-related decisions are likely impacted by the interaction between incentives and motivation. Few studies have assessed the impact of financial incentives on patients’ intrinsic motivation for health behaviors using direct measures of motivation. We examined this question in the context of five randomized controlled trials of financial incentives for health behavior change. We investigated whether effects varied by incentive type or behavior and assessed whether baseline or changes in motivation were associated with performance in behavior change programs.

We used the Treatment Self-Regulation Questionnaire to measure intrinsic motivation at baseline and at least once following the incentive intervention period in randomized controlled trials of financial incentives for weight loss (two studies), home health monitoring, walking among older adults, and adherence to use of a Positive Airways Pressure device for sleep apnea. In addition to varying health behaviors, these trials utilized different forms of incentives, including conditional payments, regret lotteries, and deposit contracts.

Multivariate regressions with participant-level data and random effects were used to assess the relationships between baseline and change in intrinsic motivation and performance in each study, measured as achieving study goals. Similar analyses were used to examine the effect of incentive eligibility and receipt on changes in intrinsic motivation to test for crowding out.

561 participants in five randomized controlled trials of financial incentives for health behavior change. First, we found that an increase in intrinsic motivation during the intervention was associated with increased odds of success in the program, defined as achieving program goals such as a pre-determined weight loss target. Second, we found

no evidence of crowding out of intrinsic motivation by incentives; that is, there was no significant association between incentive eligibility or receipt and the odds of a decrease in intrinsic motivation pre- versus post-incentives. The lack of evidence of crowding out was consistent across all five studies. (Further sub-group analyses to examine heterogeneity in our results as well as sensitivity checks are forthcoming.)

Financial incentives did not crowd out intrinsic motivation across a range of health behaviors and incentive designs. Improving our understanding in this area is critical in order to understand consumer decision-making in the context of health behaviors as well as to design the most effective incentives, understand in what settings they are most likely to work, and improve the potential for long-run behavior change.

Purpose: Despite limited availability of free eye care services in Baltimore City, utilization by low-income at-risk minority individuals remains low. This may be driven by a lack of perceived value for these free services. We examine the effect of providing vouchers redeemable for free eye care services on uptake among participants in the ongoing SToP Glaucoma study. Methods: A cluster randomized trial was conducted within the SToP Glaucoma study, an investigation of a community-based screening program which identifies glaucoma suspects and offers them free follow-up appointments at the Wilmer Eye Institute. Appointments are scheduled at the time of screening and reminder calls are made for all patients. Screening events were randomized to standard verbal and written counseling offering the individual a free appointment, or counseling in addition to provision of one of two types of vouchers redeemable for free appointments. Both voucher types included the patient’s name, the appointment date and an expiration date 90 days following the screening. One also included the approximate monetary value of the service ($250). The primary outcome was presenting for follow-up within the voucher eligibility period. A hierarchical mixed-effects logistic model allowing for random effects from the screening event was used to assess the effects of each voucher type on presentation to a follow-up appointment. Data collection is ongoing. Results: Follow-up through November 2017 yielded complete data for 431 glaucoma suspects identified at one of 64 screening events. Overall, 76% of individuals were African American, 65% were female, and the mean age was 69. There were no significant differences in these factors between study arms. Those referred in the traditional manner had a 49% attendance rate, whereas 67% of individuals receiving a voucher without monetary value information and 62% of individuals receiving a voucher with monetary value information presented. For a given screening event, offering vouchers without monetary value information increased the odds of presenting for follow-up by 152% (p =.03) compared to not offering a voucher. Offering vouchers with monetary value information increased the odds of presenting by 112%, but this effect was not significant (p=0.09). Conclusions: Offering vouchers for redemption of free eye care services increases utilization. Voucher provision may increase perceived value for these services, particularly among low-income minority populations. Further investigation into the elements of vouchers that drive this effect is warranted.

This study investigates the life partners’ influence on each other’s preventative health service usage among those over age 50, and further, disentangles the active learning and passive imitation channels among spousal concordance. The study also explores how behavioral traits such as risk aversion and time discounting would mediate the learning channel.

The study uses a biennial and representative dataset, Health & Retirement Study (HRS) 2006-2014. Four study cohorts are included, HRS (born between 1931-1941), WB (war baby, 1942-1947), EBB (early baby boomer, 1948-1953) and MBB (middle baby boomer, 1954-1959). A sample of about 19,362 respondents is yielded for the study.

The study focuses on individuals with stable life partners throughout the study period. The outcome variables are whether or not a respondent engaged in (initiation or drop-out) preventative activities over years, i.e. a flu shot, a blood test for cholesterol screening and rigorous physical activities. In order to capture positive and negative partner influences, a life partner’s previous usage pattern is classified into one of four categories: Non-User, those life partners who did not use a certain activity for two consecutive waves; Stopper, those who used in earlier wave but stopped using in later wave; Continuous, those who used in both waves; and Starter, those who did not use in earlier wave but started to use in later wave. Self-reported health status changes and physician verified new chronic conditions for both self and the partner are used as proxies of health outcomes to gauge the learning channel. The main research of interest is the extent to which the interactions between the spousal behaviors and health outcomes would predict preventive activity engagement and changes. Risk-aversion and time-discounting are also used to explore the potential mediation effects.

Couples show concordance in preventive behaviors, and further, start using a preventative activity by a life partner encourages the other couple more than stop using would discourage. The learning channel is supported by the evidence that worse-off self-health and better-off spousal-health exaggerate the encouraging effects, while better-off self-health and worse-off spousal-health exaggerate the discouraging effects. Moreover, those with higher risk-aversion and time-discounting (less patient) are more responsive to learning and behavioral changes.

Conclusions & Implications: Findings in the study suggest that the partners' influences are asymmetric and happen through active learning. The study provides insights on the potential efficacy of family-based promotion strategies to increase preventative activity engagement, as well as the perfect timing of intervention. Moreover, this study also sheds light on patient-level medical and healthcare decision-making in general beyond the preventative stage.


Medicaid beneficiaries face substantial burden from chronic diseases. Incentive programs have been proposed to connect and engage Medicaid beneficiaries with preventive services, thereby changing behavior and preventing chronic diseases. We evaluated the impact of the Medicaid Incentives for Prevention of Chronic Disease program, which funded Medicaid incentive initiatives in 10 states. The focus of initiatives varied by state and included smoking cessation, diabetes prevention, weight loss, and disease management. The value of incentives also varied widely, ranging from a maximum of $50 in one state to over $1,000 annually in another state. Most states randomly assigned participants to incentive and control arms, with both arms eligible to receive preventive services. We conducted a mixed methods evaluation that included document review, site visits, focus groups, a beneficiary survey, and Medicaid claims analysis. The claims analysis used regression and difference-in-difference methods to test whether incentives increased the use of preventive services and lowered use of other Medicaid services and expenditures. We find that states can implement Medicaid programs successfully, although they had to overcome administrative challenges and recruiting participants was more difficult than anticipated. Participants in the incentive arms attended significantly more diabetes prevention and weight watchers classes, made more smoking quitline calls, and attended more smoking cessation sessions. Participants who received incentives were very satisfied with program access and quality and believed that the incentives helped them meet their health goals. The programs had no significant impact on the use and costs of other Medicaid services during the evaluation. Administrative costs were relatively high. State evaluations suggest that the incentives had mixed effects on short-term health outcomes, with some significant reductions in smoking and generally insignificant effects on diabetes prevention, weight loss, and risk factor management. Future study is needed to determine whether the impact of Medicaid incentive programs on health outcomes can be improved.

From a public health perspective, the flu vaccination rate is viewed as too low in many countries. While up until recently, only physicians were allowed to vaccinate in Switzerland, some pharmacies received the license to vaccinate during the course of the year 2016. Public health authorities hope to increase the vaccination rate as there is generally no need for an appointment in a pharmacy. This setting allows us to test whether the vaccination rate can be increased by a letter informing consumers about the access to prevention care as well as whether consumers can be nudged to get their flu shot in a recommended pharmacy. We use a randomized control trial where a large Swiss health insurer sent letters to 22'000 of its customers (between 50 and 75 years of age). The letter included information about the new possibility to get a flu shot in a pharmacy, a pledge to bear the costs and an indication of the nearest pharmacy which had the license to vaccinate. The letter increased immunization rates by 2.7%-points (17.9%) with significant heterogeneity in background variables as well as with respect to the distance to the pharmacy. We find some provider substitution effects from physicians to pharmacists. Yet, the majority of the increase in the vaccination rate is driven by additional vaccinations in the pharmacies. More than half of the customers visited the recommended pharmacy even if a closer one had been available. Currently, we do not find any significant effects of the flu shot on covered health care expenditures nor on the number of doctor visits. Information letters proofed to be an effective tool to increase the use of prevention care as well as to guide consumers to specific health care providers. The new possibility to vaccinate in a pharmacy increased the take-up rate of the vaccination. However, the effectiveness of the flu shot remains currently doubtful.

Do state laws that require the availability of paid sick leave increase the use of preventive services? Paid sick leave benefits allow workers to maintain job security when they leave for medical reasons. In recent years, paid sick leave laws have received more attention by health policy makers in terms of its potential to improve public health. Currently, seven states and D.C. have laws that require employers to provide paid sick leave. However, empirical evidence on this topic is limited (DeRigne, et al 2017; Peipins et al 2012). Although prior studies examined the association of paid sick leave and preventive service use, these studies did not use rigorous study designs. Because workers with higher demand for healthcare services may select into firms which offer better health benefits including paid sick leave, the estimated association in these previous studies may suffer from selection bias. The goal of this study is to use a quasi-experimental study design to estimate the impact of Connecticut’s 2012 paid sick leave law on the use of preventive services. Connecticut was the first state to require private employers to offer paid sick leave benefits to their employees. Using state and time variation from 2007-2016 Behavioral Risk Factor Surveillance System (BRFSS) data, we compare the use of preventive services in Connecticut and in other New England states before and after the implementation of the 2012 paid sick leave law. For general preventive service outcomes, we examined routine checkups, flu vaccinations, and dental visits in the past year. We also examined the use of Pap tests, clinical breast exams, and mammograms for women. Overall, we found that Connecticut’s 2012 paid sick leave law increased the use of preventive services. Specifically, the rate of routine checkups (1.4%, p<.1), flu shots (2.0%, p<.05), and dental visits (2.7%, p<.01) increased over this period. With respect to cancer screening outcomes for women, we found that the use of Pap tests (9.4%, p<.01), and clinical breast exams (4.4%, p<.01) increased. However, the higher rate of mammograms (1.3%, p=.38) was not statistically significant. Our findings provide rigorous evidence on the positive impact of a state’s paid sick leave law on preventive service outcomes. These empirical results also suggest that a lack of paid sick leave among some private employers may represent a barrier to accessing preventive services. Finally, policymakers can use legislation to support the use of preventive services and improve the health and productivity of workers.

: Medicare Part D enrollees frequently make suboptimal plan choices, typically leaving hundreds of dollars on the table every year. This overspending is largely driven by systematic errors in the way enrollees assess the value of plan attributes (e.g., over-valuing premiums, under-valuing expected out-of-pocket [OOP] costs), despite the existence of decision support tools that facilitate accurate valuations through the provision of personalized cost estimates. In particular, CMS’s Plan Finder tool has had limited effects on consumer choices, in part because the complexity and presentation of the information it provides limits the salience of the personalized cost estimates. Simplifying Plan Finder has the potential to improve plan choices. This study uses a survey-based randomized experiment to examine the effect of simplifying the financial information presented on Plan Finder on the way individuals choose Part D plans.

: We used the American Life Panel, a nationally representative internet panel, to field an experiment among 1,278 adults age 55+. Participants made simulated Part D plan choices on behalf of a friend with a stated preference for minimizing total drug spending. Respondents were randomized into 4 study arms (1 control, 3 treatment). The control group was shown a plan menu which mimicked the current Plan Finder tool. Total cost estimates were displayed for each plan alongside detailed financial information (premiums, deductibles, copays) and non-financial information (e.g., 5-star quality scores, pharmacy network size). The 3 treatment groups, which varied in the amount of financial information shown by default, were as follows: 1) total cost only; 2) total cost and premium; 3) total cost, premium, and estimated OOP cost. All treatment groups could view full financial information by clicking a link within the plan menu. Plan choices were evaluated using discrete choice analyses and conditional logit estimation. Differences in the decision weights placed on plan attributes across study arms were evaluated to test the effect of Plan Finder simplifications on the trade-offs individuals make when selecting a plan.

: Simplifying financial information resulted in the selection of lower cost plans in all treatment groups relative to the control group. These improvements were largely driven by increases in the weight placed on OOP costs and reductions in the weighting of deductibles. Respondents in the “total cost, premium, and OOP cost” group had decision weights that were most consistent with cost-minimization: they placed equal weight on premium and OOP costs and no additional weight on cost-sharing attributes beyond their direct impact on anticipated spending. The “total cost and premium” group continued to weight premiums more than OOP costs, while the “total cost only” group was more likely to seek additional plan information which resulted in plan choices more like those of the control group.

: Simplifying the financial information on Plan Finder helps individuals adhere to a cost-minimization decision rule. Displaying plan premiums and OOP costs alongside total cost estimates was most effective in encouraging cost-minimization and reducing additional information seeking, suggesting that this format improves individuals’ trust of the information and ability to use it.

Research has documented that consumers often have imperfect information about the health insurance plans from which they are asked to choose; but we know less about the sources of that information. Given the difficulty in obtaining reliable information from independent sources, consumers may draw on their peers for recommendations. This paper investigates the role of social learning in health insurance selection, using longitudinal data from the University of California on plan choices of employees and peers in their department. The data from 2011 to 2016 span a major change in the insurance choice set, which aids in the statistical identification of social learning effects among both incumbent employees as well as new hires. I start by documenting the high similarity in plan choices within peer groups, suggesting the possibility of strong peer effects, and then use a variety of approaches to test for potential confounding from unobserved heterogeneity. I employ a discrete choice conditional logit estimator to formally model plan choice behavior, finding that a 10 percentage point increase in the share of peers who select a particular insurance plan will lead to a 14 percentage point increase in the probability that an individual will choose the same plan. This large effect on plan choice is equivalent to lowering the monthly premium by 18 percent. I then use this model to simulate employer strategies that could exploit social learning to better promote the employer’s insurance objectives. For illustration, I conduct counterfactual analyses of incentives to promote adoption of a new consumer-driven insurance. At the actuarially fair premium in this setting, demand for a consumer-driven plan is low, and social learning further discourages take-up. However, with sufficient premium subsidies, the model projects that the social learning effects will become positive and can be harnessed by employers to more effectively achieve their cost and insurance

From the perspective of patients, the closing of a primary-care practice causes a discontinuity of care, which bears consequences for patients with long-standing doctor-patient relationships. First, interruptions in care may lead to inefficient utilization of healthcare services. Second, the literature consistently finds that continuity of care is beneficial for patients' health-related outcomes. Moreover, practice closures decrease the local availability of primary care, which disproportionally affects

This paper studies closures of primary-care practices in Switzerland from 2005 to 2015 to estimate the causal impacts of discontinuities of primary care on patients' utilization patterns, medical expenditures and health-related outcomes. Employing a difference-in-difference framework, we identify causal effects by comparing changes in outcomes between an affected group of patients (‘treatment group’) and an unaffected group that does not experience changes in primary care provision (‘control group’). Our main findings are twofold. First, when faced with a discontinuity of primary care, patients adjust their utilization pattern by shifting visits away from ambulatory primary care providers (-5%) towards specialized care (+10%) and emergency departments of hospitals (+14%). Secondly, practice closures increase patients’ total health care expenditures by 4.6% and raise the probability of incurring non-zero costs in a given month by 10%. Two policy-relevant implications are that practice closures may lead to an inefficient use of healthcare services and have adverse effects on social health insurance which must cover higher costs. These implications are relevant for health planners and policy makers, health insurers and

Abstract


Guaranteeing equity for the poor is a major challenge for healthcare systems in developed countries. Overall, equity is an ethical issue related to the judgments about healthcare accessibility. At the same time, an economic concept of horizontal equity deals with “an equal treatment for equal need” and “means that persons in equal need of medical care should receive the same treatment, irrespective of whether they happen to be poor or rich” . In practical terms, there is a general agreement about striving for “minimal variation of [healthcare] use with income” and ensuring equity for the poor. According to theoretical predictions, a well-designed social health insurance system may provide an equitable redistribution of medical care between the rich and the poor. However, the actual performance of social health insurance systems with

The paper exploits panel data finite mixture (latent class) models to measure consumer equity in healthcare access and utilization. The finite mixture approach accounts for unobservable consumer heterogeneity. Additionally, we employ the generalized linear models with latent classes to address a retransformation problem of logged dependent variable. Using the data of the Japan Household Panel Survey (2009-2014), we discover that consumers separate into latent classes in the binary choice models for healthcare use and generalized linear models for outpatient/inpatient healthcare expenditure. The results reveal that healthcare access in Japan is pro-poor for the most sick consumers, while utilization of outpatient care is

The novelty of the paper is twofold. Firstly, we examine inpatient and outpatient healthcare access, and analyze expenditure within health insurance, exploiting the longitudinal data of the Japan Household Panel Survey. The unique feature of the survey is the fact that it distinguishes between non-users of healthcare, the users of inpatient and outpatient care, and provides a wide range of consumer characteristics, such as health status, index of psychological distress and life-style variables. Secondly, we measure income inequity with the generalized finite mixture models for healthcare use in the longitudinal context. It may be noted that the applicability of the finite mixture models for analyzing healthcare demand is well established. However, the use of generalized finite mixture models for measuring healthcare expenditure is often limited to experimental

The results of our estimations indicate that consumers separate into two latent classes in the binary choice models for use of any care or inpatient care, as well as in the loglinear and generalized linear models for outpatient and inpatient healthcare expenditure. The classes may be naturally interpreted as most frequent and most seek consumers (“high users”), infrequent and most healthy consumers (“low users”), and consumers with median use and median health status (“median users”).

Using two-year panel data from the Medical Expenditure Panel Survey (MEPS) for 2004 to 2012, we examine how income shocks affect health care spending decisions among single-mother families. This analysis focuses upon total out-of-pocket family health care spending as well as the allocation of such spending to a variety of specific health care services: dental care, vision care, prescription drugs, office-based visits, and emergency department visits. We hypothesize that the family’s response to income shocks is likely to be complex. On the one hand, families experiencing income loss may, by necessity, be required to prioritize their health care spending among specific health care services. On the other hand, loss of income may cause stress and anxiety which can have a negative impact on the health status of family members pressuring to families to maintain or even increase health care spending. Since a substantial proportion of the population does not use particular health care services and since the distribution of spending is positively skewed, we estimate a series of two-part-model health care spending models. These models are specified with probit equation for the likelihood of an expenditure in the first part of the model, and a generalized linear model (GLM) with a log link and gamma or inverse Gaussian variance function in the second part for families with positive out-of-pocket spending. To control for unobserved heterogeneity across families in the sample, we estimate the two-part model using the correlated random effects framework. There are several important finding in this study. First, we find that single-mother families experiencing an income loss tend to reduce their out-of-pocket health care spending. This does not necessarily imply a decline in health services utilization. For instance, a middle-income single-mother family that becomes a low-income family decreases its total out-of-pocket spending by an average of $585 annually but increases a likelihood of any health service use (with or without cost-sharing) increases

Second, we find that an income loss among low-income single-mother families is associated with a decrease in out-of-pocket prescription drug spending by between $67-135 annually. Third, we find that some families appear to reallocate their health care spending in response to an income shock. For instance, high income single-mother families that become middle income families tend to increase their out-of-pocket prescription drug spending by an average of $82 while decreasing out-

Fourth, we also find that that income loss among single-mother families is associated with a statistically significant decline in out-of-pocket spending toward emergency department visits. However, these declines are small in size. Finally, we find no

moral hazard. Recently, HDHPs have become commonplace in the employer insurance market; however, the effect of adding an HDHP option into an individual’s offer set remains understudied. This paper answers three questions regarding HDHPs. First, do HDHPs lower total medical spending, and is there a behavioral response or simply a shifting of costs to the individual? I find HDHPs lower spending by 16 percent and reduce utilization as predicted by demand theory. Second, I find reductions in hospital-based medical care spending account for 60 percent of the savings. Finally, contrary to recently published papers, I find evidence of discriminatory

The focus of this paper is an important (but understudied) driver of health spending: patients' sorting decisions over type of health care provider from which to obtain care. Occupational licensing restrictions ("scope of practice (SOP) laws") play a major role in these decisions by specifying the scope of treatment performed by various provider types. These restrictions vary across states, and several states have relaxed their laws in recent years with more states planning to do so in the future. Efficient patient sorting will become increasingly important given two recent trends in health care: (1) the rise of non-physician healthcare providers (such as nurse practitioners and physician assistants), and (2) the large predicted physician shortage (created in part by the ACA). As non-physicians (NPs) become more accessible, more patients will be able to choose whether to receive care from an NP instead of a physician (MD), and the efficiency of these choices will become a greater focus of health care policy. The increasing relaxation of SOP licensing laws is one factor contributing to the increased access to NPs. Similarly, in the presence of a physician shortage, many patients will be forced to sort to NPs to receive care. Thus, efficient

In this paper I study the relationship between patient sorting, SOP laws, and health care costs, disparities, quality, and outcomes. I first document both the types of patients and the types of care that sort to NPs versus MDs. I then take a machine learning approach to study the efficiency of sorting in the context of patient mis-prediction of personal risk and complexity of required treatment. Next, I exploit three specific natural experiments to study the types of care that are on the margin between provider types. The three experiments each affect the access to NPs relative to MDs, but through different channels. They are: the state-specific relaxation of SOP laws, insurer changes in relative copays between provider types, and a large government subsidy for the training of NPs at five different US medical schools (the Graduate Nurse Education Demonstration). I combine these experiments with the machine learning results on mis-prediction to estimate a personalized, claim-level measure of the effect of receiving care from an MD versus an NP. I show the correlation between this effect and the patient's algorithm-generated predicted risk of an adverse outcome. The final portion of the paper is devoted to understanding the broader implications of patient sorting over provider types on explanations of empirical facts in the health economics literature. I estimate the predicted effects of counterfactually altering SOP laws and I show that patient sorting plays an important role in health care costs, disparities, quality, and outcomes.



The start of the 2014-15 influenza season was overshadowed by the fear of Ebola. As the first Ebola patient diagnosed in the US died and two nurses confirmed the infection in October 2014, the anxiety of Ebola was elevated. As part of the massive media coverage on Ebola, many health experts compared Ebola with influenza and pointed out the importance of receiving influenza vaccination in that year. The rationale was that Ebola-related hospitalization and deaths were far less than those caused by influenza and having more people receiving influenza vaccines would reduce false alarms and help the public health system respond to Ebola. Meanwhile, CDC also recommended people being actively monitored for potential Ebola virus exposure to receive influenza vaccines if they had not done so. It is unclear, however, whether the public responded to these messages. This study examines the impact of public awareness around Ebola on influenza vaccine uptake during the 2014-15 influenza season in the US. We use individual-level data from the Behavioral Risk Factor Surveillance System and examine changes in the likelihood of receiving influenza vaccines. We examine the robustness of these findings in a doubly-robust difference-in-difference with propensity score matching framework as well as a synthetic control framework. We examine the impact of three types of treatment: residing in a state with an Ebola case (Texas and New York), residing in a state with Ebola or medically evacuated cases (Texas, New York, Georgia, Maryland, and Nebraska), or having higher level of attention to Ebola as

Our findings provide the first quantification of the spillovers from messaging targeting one rare disease to health behaviors related to a second, more common infection. Future immunization programs that consider phrasing their promoting messages by relating influenza to threatening diseases such as Ebola would be able to predict the impact of this strategy based on our results.

In light of increasing health care expenditures, patient cost-sharing schemes have emerged as one of the main policy tools to reduce medical spending. In this study we show that health care utilization is affected not only by the economic incentives provided by cost-sharing schemes, but also by the way these economic incentives are presented. Specifically, we compare patients’ responses to a deductible and to a no-claim refund. The economic incentives under a deductible and a no-claim refund are very similar, but they are framed in a different way. Under a deductible policy, individuals pay out-of-pocket for all medical care up to the deductible limit. Under a no-claim refund policy, individuals receive a payment at the end of the year if their health care spending during the year was below the no-claim refund limit. Prospect theory predicts that individuals respond stronger to losses than to gains. If individuals perceive deductible payments as losses and lower no-claim refunds as foregone gains then we might expect that individuals will react stronger to deductibles than to no-claim rebates.

We make use of the fact that in the Netherlands, both schemes have been in place at different points of time while the patient population and the services covered by health insurance remained comparable. In the years 2006 and 2007 Dutch law has mandated that health insurance contracts included a no-claim refund, and from the year 2008 onward, health insurance contracts had to feature an annual deductible. Our analysis is based on unique claims-level data from a Dutch health insurer for the years 2006-2015 which we aggregate to around 9 million person month observations. In our empirical strategy we exploit variation in cost-sharing incentives within a year. Under both a deductible policy and a no-claim refund the price of healthcare utilization can vary over the course of the year depending on whether or not an individual has exceeded her deductible or no-claim refund limit. We examine how the reaction to prices differs between the years when a no-claim refund policy was in place and the years when a deductible policy was in place. We account for the possible endogeneity of prices with a simulated instrumental variables approach. As instrumental variable for the price at the beginning of the month we use

We find that patients react to comparable incentives twice as strongly when they are implemented as a deductible, which suggests that the framing of incentives can be quantitatively almost as important as the incentive itself. Our preferred explanation is that individuals are loss-averse and respond differently to both schemes because they perceive a deductible payment as a loss and a no-claim refund as a gain. Our results are robust to a number of sensitivity analyses. Specifically, our


We design a conventional economic laboratory experiment which simulates the course of events inherent in medical treatments from an economic perspective. Our experiment consists of two stages. The working stage mimics the beginning of the treatment and induces feelings of losses as subjects have to work on a task but only receive a fraction of their proper income. Entering the investment stage with these losses subjects decide over 12 periods between lottery A and lottery B. These lotteries represent the economic consequences of discontinuing and continuing with therapy. Lottery A with a higher risk of losing money can be chosen without any prior investments whereas for playing Lottery B with higher winning chances subjects have to invest some of their monetary endowment. Once a lottery is lost subjects drop out of the experiment and are not allowed to make any more decisions. We incorporate loss aversion and the timing in our incentive scheme in the following way: as soon as subjects have compensated the losses from the working stage they receive an up-front bonus which is added to their balance account. Subjects are only allowed to keep this bonus if they do not dropout of the experiment


Innovative organizational forms of health care delivery have recently developed that lower the time cost of care. While there is an extensive literature on consumer response to changes in the out-of-pocket price for care, little work has studied the equally-salient dimension of time cost. In this paper, I develop a theoretical model of patient decision-making and predict that when a new provider enters the market and offers services with lower time cost, patients engage in new utilization and/or substitution across providers. I then test these predictions in a unique empirical setting: A large corporation opened a worksite health clinic on its California campus in 2013, but did not feature a clinic on its Texas campus. I utilize novel data, 2011-2015 medical claims for the corporation’s employees. My primary empirical strategy is a difference-in-differences approach, where I compare California employees to the control group of Texas employees. I find that the effects of clinic availability are concentrated among the narrow set of services that can be provided onsite. For primary care in particular, I observe both an increase in utilization and substitution towards onsite care. While services beyond the clinic’s scope of practice are mostly unaffected, California employees reduce their utilization of outpatient care; this spillover effect is offset by a substantial increase in demand for office-based care. Ultimately, new consumption of primary care and other office-based services drives a

percentile of the conditional distribution, the estimated increase in monthly total spending is $14.67. My findings suggest that consumer demand is sensitive to changes in time cost, and this has important implications for the potential welfare benefits of providing patients with convenient access to high-value services.

To assess whether financial incentives for health behaviors crowd out individuals’ intrinsic motivation to engage in those behaviors. The use of financial incentives to promote changes in health behaviors is widespread among payers and employers, however there are concerns that if incentives crowd out intrinsic motivation, behavior would fall below even pre-incentive levels following the removal of incentives, hindering any long-run impact of incentives on behavior change. Further, consumers’ health-related decisions are likely impacted by the interaction between incentives and motivation. Few studies have assessed the impact of financial incentives on patients’ intrinsic motivation for health behaviors using direct measures of motivation. We examined this question in the context of five randomized controlled trials of financial incentives for health behavior change. We investigated whether effects varied by incentive type or behavior and assessed whether

We used the Treatment Self-Regulation Questionnaire to measure intrinsic motivation at baseline and at least once following the incentive intervention period in randomized controlled trials of financial incentives for weight loss (two studies), home health monitoring, walking among older adults, and adherence to use of a Positive Airways Pressure device for sleep apnea. In addition to varying health behaviors, these trials utilized different forms of incentives, including conditional

Multivariate regressions with participant-level data and random effects were used to assess the relationships between baseline and change in intrinsic motivation and performance in each study, measured as achieving study goals. Similar analyses were used to examine the effect of incentive eligibility and receipt on changes in intrinsic motivation to test for crowding out.

First, we found that an increase in intrinsic motivation during the intervention was associated with increased odds of success in the program, defined as achieving program goals such as a pre-determined weight loss target. Second, we found no evidence of crowding out of intrinsic motivation by incentives; that is, there was no significant association between incentive eligibility or receipt and the odds of a decrease in intrinsic motivation pre- versus post-incentives. The lack of evidence of crowding out was consistent across all five studies. (Further sub-group analyses to examine heterogeneity in our results as well as sensitivity checks are forthcoming.)

Financial incentives did not crowd out intrinsic motivation across a range of health behaviors and incentive designs. Improving our understanding in this area is critical in order to understand consumer decision-making in the context of health behaviors as well as to design the most effective incentives, understand in what settings they are most likely to work, and improve the potential for long-run behavior change.

Purpose: Despite limited availability of free eye care services in Baltimore City, utilization by low-income at-risk minority individuals remains low. This may be driven by a lack of perceived value for these free services. We examine the effect of providing vouchers redeemable for free eye care services on uptake among participants in the ongoing SToP Glaucoma study. Methods: A cluster randomized trial was conducted within the SToP Glaucoma study, an investigation of a community-based screening program which identifies glaucoma suspects and offers them free follow-up appointments at the Wilmer Eye Institute. Appointments are scheduled at the time of screening and reminder calls are made for all patients. Screening events were randomized to standard verbal and written counseling offering the individual a free appointment, or counseling in addition to provision of one of two types of vouchers redeemable for free appointments. Both voucher types included the patient’s name, the appointment date and an expiration date 90 days following the screening. One also included the approximate monetary value of the service ($250). The primary outcome was presenting for follow-up within the voucher eligibility period. A hierarchical mixed-effects logistic model allowing for random effects from the screening event was used to

Results: Follow-up through November 2017 yielded complete data for 431 glaucoma suspects identified at one of 64 screening events. Overall, 76% of individuals were African American, 65% were female, and the mean age was 69. There were no significant differences in these factors between study arms. Those referred in the traditional manner had a 49% attendance rate, whereas 67% of individuals receiving a voucher without monetary value information and 62% of individuals receiving a voucher with monetary value information presented. For a given screening event, offering vouchers without monetary value information increased the odds of presenting for follow-up by 152% (p =.03) compared to not offering a voucher. Offering vouchers with monetary value information increased the odds of presenting by 112%, but this effect was not significant (p=0.09). Conclusions: Offering vouchers for redemption of free eye care services increases utilization. Voucher provision may increase perceived value for these services, particularly among low-income minority populations. Further investigation into the

This study investigates the life partners’ influence on each other’s preventative health service usage among those over age 50, and further, disentangles the active learning and passive imitation channels among spousal concordance. The study also

The study uses a biennial and representative dataset, Health & Retirement Study (HRS) 2006-2014. Four study cohorts are included, HRS (born between 1931-1941), WB (war baby, 1942-1947), EBB (early baby boomer, 1948-1953) and MBB (middle

The study focuses on individuals with stable life partners throughout the study period. The outcome variables are whether or not a respondent engaged in (initiation or drop-out) preventative activities over years, i.e. a flu shot, a blood test for cholesterol screening and rigorous physical activities. In order to capture positive and negative partner influences, a life partner’s previous usage pattern is classified into one of four categories: Non-User, those life partners who did not use a certain

Continuous, those who used in both waves; and Starter, those who did not use in earlier wave but started to use in later wave. Self-reported health status changes and physician verified new chronic conditions for both self and the partner are used as proxies of health outcomes to gauge the learning channel. The main research of interest is the extent to which the interactions between the spousal behaviors and health outcomes would predict preventive activity engagement and changes. Risk-aversion and time-discounting are also used to explore the potential mediation effects.

Couples show concordance in preventive behaviors, and further, start using a preventative activity by a life partner encourages the other couple more than stop using would discourage. The learning channel is supported by the evidence that worse-off self-health and better-off spousal-health exaggerate the encouraging effects, while better-off self-health and worse-off spousal-health exaggerate the discouraging effects. Moreover, those with higher risk-aversion and time-discounting (less

Findings in the study suggest that the partners' influences are asymmetric and happen through active learning. The study provides insights on the potential efficacy of family-based promotion strategies to increase preventative activity engagement, as well as the perfect timing of intervention. Moreover, this study also sheds light on patient-level medical and healthcare decision-making in general beyond the preventative stage.


Medicaid beneficiaries face substantial burden from chronic diseases. Incentive programs have been proposed to connect and engage Medicaid beneficiaries with preventive services, thereby changing behavior and preventing chronic diseases. We evaluated the impact of the Medicaid Incentives for Prevention of Chronic Disease program, which funded Medicaid incentive initiatives in 10 states. The focus of initiatives varied by state and included smoking cessation, diabetes prevention, weight loss, and disease management. The value of incentives also varied widely, ranging from a maximum of $50 in one state to over $1,000 annually in another state. Most states randomly assigned participants to incentive and control arms, with both

We conducted a mixed methods evaluation that included document review, site visits, focus groups, a beneficiary survey, and Medicaid claims analysis. The claims analysis used regression and difference-in-difference methods to test whether

We find that states can implement Medicaid programs successfully, although they had to overcome administrative challenges and recruiting participants was more difficult than anticipated. Participants in the incentive arms attended significantly more diabetes prevention and weight watchers classes, made more smoking quitline calls, and attended more smoking cessation sessions. Participants who received incentives were very satisfied with program access and quality and believed that the incentives helped them meet their health goals. The programs had no significant impact on the use and costs of other Medicaid services during the evaluation. Administrative costs were relatively high. State evaluations suggest that the incentives had mixed effects on short-term health outcomes, with some significant reductions in smoking and generally insignificant effects on diabetes prevention, weight loss, and risk factor management. Future study is needed to determine whether the

From a public health perspective, the flu vaccination rate is viewed as too low in many countries. While up until recently, only physicians were allowed to vaccinate in Switzerland, some pharmacies received the license to vaccinate during the course of the year 2016. Public health authorities hope to increase the vaccination rate as there is generally no need for an appointment in a pharmacy. This setting allows us to test whether the vaccination rate can be increased by a letter informing consumers about the access to prevention care as well as whether consumers can be nudged to get their flu shot in a recommended pharmacy. We use a randomized control trial where a large Swiss health insurer sent letters to 22'000 of its customers (between 50 and 75 years of age). The letter included information about the new possibility to get a flu shot in a pharmacy, a pledge to bear the costs and an indication of the nearest pharmacy which had the license to vaccinate. The letter increased immunization rates by 2.7%-points (17.9%) with significant heterogeneity in background variables as well as with respect to the distance to the pharmacy. We find some provider substitution effects from physicians to pharmacists. Yet, the majority of the increase in the vaccination rate is driven by additional vaccinations in the pharmacies. More than half of the customers visited the recommended pharmacy even if a closer one had been available. Currently, we do not find any significant effects of the flu shot on covered health care expenditures nor on the number of doctor visits. Information letters proofed to be an effective tool to increase the use of prevention care as well as to guide consumers to specific health care providers. The new possibility to vaccinate in a pharmacy increased the take-up rate of the vaccination.

Do state laws that require the availability of paid sick leave increase the use of preventive services? Paid sick leave benefits allow workers to maintain job security when they leave for medical reasons. In recent years, paid sick leave laws have received more attention by health policy makers in terms of its potential to improve public health. Currently, seven states and D.C. have laws that require employers to provide paid sick leave. However, empirical evidence on this topic is limited (DeRigne, et al 2017; Peipins et al 2012). Although prior studies examined the association of paid sick leave and preventive service use, these studies did not use rigorous study designs. Because workers with higher demand for healthcare services may select into firms which offer better health benefits including paid sick leave, the estimated association in these previous studies may suffer from selection bias. The goal of this study is to use a quasi-experimental study design to estimate the impact of Connecticut’s 2012 paid sick leave law on the use of preventive services. Connecticut was the first state to require private employers to offer paid sick leave benefits to their employees. Using state and time variation from 2007-2016 Behavioral Risk Factor Surveillance System (BRFSS) data, we compare the use of preventive services in Connecticut and in other New England states before and after the implementation of the 2012 paid sick leave law. For general preventive service outcomes, we examined routine checkups, flu vaccinations, and dental visits in the past year. We also examined the use of Pap tests, clinical breast exams, and

Overall, we found that Connecticut’s 2012 paid sick leave law increased the use of preventive services. Specifically, the rate of routine checkups (1.4%, p<.1), flu shots (2.0%, p<.05), and dental visits (2.7%, p<.01) increased over this period. With respect to cancer screening outcomes for women, we found that the use of Pap tests (9.4%, p<.01), and clinical breast exams (4.4%, p<.01) increased. However, the higher rate of mammograms (1.3%, p=.38) was not statistically significant. Our findings provide rigorous evidence on the positive impact of a state’s paid sick leave law on preventive service outcomes. These empirical results also suggest that a lack of paid sick leave among some private employers may represent a barrier to accessing preventive services. Finally, policymakers can use legislation to support the use of preventive services and improve the health and productivity of workers.



: Simplifying financial information resulted in the selection of lower cost plans in all treatment groups relative to the control group. These improvements were largely driven by increases in the weight placed on OOP costs and reductions in the weighting of deductibles. Respondents in the “total cost, premium, and OOP cost” group had decision weights that were most consistent with cost-minimization: they placed equal weight on premium and OOP costs and no additional weight on cost-sharing attributes beyond their direct impact on anticipated spending. The “total cost and premium” group continued to weight premiums more than OOP costs, while the “total cost only” group was more likely to seek additional plan information

: Simplifying the financial information on Plan Finder helps individuals adhere to a cost-minimization decision rule. Displaying plan premiums and OOP costs alongside total cost estimates was most effective in encouraging cost-minimization and reducing additional information seeking, suggesting that this format improves individuals’ trust of the information and ability to use it.



This paper studies closures of primary-care practices in Switzerland from 2005 to 2015 to estimate the causal impacts of discontinuities of primary care on patients' utilization patterns, medical expenditures and health-related outcomes. Employing a difference-in-difference framework, we identify causal effects by comparing changes in outcomes between an affected group of patients (‘treatment group’) and an unaffected group that does not experience changes in primary care provision (‘control group’). Our main findings are twofold. First, when faced with a discontinuity of primary care, patients adjust their utilization pattern by shifting visits away from ambulatory primary care providers (-5%) towards specialized care (+10%) and emergency departments of hospitals (+14%). Secondly, practice closures increase patients’ total health care expenditures by 4.6% and raise the probability of incurring non-zero costs in a given month by 10%. Two policy-relevant implications are that practice closures may lead to an inefficient use of healthcare services and have adverse effects on social health insurance which must cover higher costs. These implications are relevant for health planners and policy makers, health insurers and

Abstract Presenting Author

Jason Ward

Galina Besstremyannaya

Irina Grafova

Alicia Atwood

Presenting Author Email

Address

The existence of significant regional variation in health care utilization has been well documented over the past 40 years. Yet considerable uncertainty persists about whether this variation is primarily the result of supply-side or demand-side forces. We use a model of physician market power to derive an empirical test of supply-side explanations. Specifically, we examine changes in the use of healthcare by the near-elderly across regions that differ in Medicare spending levels as they transition from being uninsured into Medicare. Estimates indicate that gaining Medicare coverage in an above-median spending region is associated with a 48% increase in the probability of at least one hospital visit and a 26% increase in the probability of having more than five doctor visits relative to similar individuals in below-median spending regions. These estimates suggest that supply-side factors can explain much of the observed geographic variation in Medicare spending. jward28@u

ic.edu

Guaranteeing equity for the poor is a major challenge for healthcare systems in developed countries. Overall, equity is an ethical issue related to the judgments about healthcare accessibility. At the same time, an economic concept of horizontal equity deals with “an equal treatment for equal need” and “means that persons in equal need of medical care should receive the same treatment, irrespective of whether they happen to be poor or rich” . In practical terms, there is a general

According to theoretical predictions, a well-designed social health insurance system may provide an equitable redistribution of medical care between the rich and the poor. However, the actual performance of social health insurance systems with

The paper exploits panel data finite mixture (latent class) models to measure consumer equity in healthcare access and utilization. The finite mixture approach accounts for unobservable consumer heterogeneity. Additionally, we employ the generalized linear models with latent classes to address a retransformation problem of logged dependent variable. Using the data of the Japan Household Panel Survey (2009-2014), we discover that consumers separate into latent classes in the binary choice models for healthcare use and generalized linear models for outpatient/inpatient healthcare expenditure. The results reveal that healthcare access in Japan is pro-poor for the most sick consumers, while utilization of outpatient care is

The novelty of the paper is twofold. Firstly, we examine inpatient and outpatient healthcare access, and analyze expenditure within health insurance, exploiting the longitudinal data of the Japan Household Panel Survey. The unique feature of the survey is the fact that it distinguishes between non-users of healthcare, the users of inpatient and outpatient care, and provides a wide range of consumer characteristics, such as health status, index of psychological distress and life-style variables.

It may be noted that the applicability of the finite mixture models for analyzing healthcare demand is well established. However, the use of generalized finite mixture models for measuring healthcare expenditure is often limited to experimental

The results of our estimations indicate that consumers separate into two latent classes in the binary choice models for use of any care or inpatient care, as well as in the loglinear and generalized linear models for outpatient and inpatient healthcare expenditure. The classes may be naturally interpreted as most frequent and most seek consumers (“high users”), infrequent and most healthy consumers (“low users”), and consumers with median use and median health status (“median users”). gbesstre@y

andex.ru

Using two-year panel data from the Medical Expenditure Panel Survey (MEPS) for 2004 to 2012, we examine how income shocks affect health care spending decisions among single-mother families. This analysis focuses upon total out-of-pocket family health care spending as well as the allocation of such spending to a variety of specific health care services: dental care, vision care, prescription drugs, office-based visits, and emergency department visits. We hypothesize that the family’s response to income shocks is likely to be complex. On the one hand, families experiencing income loss may, by necessity, be required to prioritize their health care spending among specific health care services. On the other hand, loss of income may cause stress and anxiety which can have a negative impact on the health status of family members pressuring to families to maintain or even increase health care spending. Since a substantial proportion of the population does not use particular health care services and since the distribution of spending is positively skewed, we estimate a series of two-part-model health care spending models. These models are specified with probit equation for the likelihood of an expenditure in the first part of the model, and a generalized linear model (GLM) with a log link and gamma or inverse Gaussian variance function in the second part for families with positive out-of-pocket

There are several important finding in this study. First, we find that single-mother families experiencing an income loss tend to reduce their out-of-pocket health care spending. This does not necessarily imply a decline in health services utilization. For instance, a middle-income single-mother family that becomes a low-income family decreases its total out-of-pocket spending by an average of $585 annually but increases a likelihood of any health service use (with or without cost-sharing) increases

Second, we find that an income loss among low-income single-mother families is associated with a decrease in out-of-pocket prescription drug spending by between $67-135 annually. Third, we find that some families appear to reallocate their health care spending in response to an income shock. For instance, high income single-mother families that become middle income families tend to increase their out-of-pocket prescription drug spending by an average of $82 while decreasing out-

Fourth, we also find that that income loss among single-mother families is associated with a statistically significant decline in out-of-pocket spending toward emergency department visits. However, these declines are small in size. Finally, we find no [email protected]

moral hazard. Recently, HDHPs have become commonplace in the employer insurance market; however, the effect of adding an HDHP option into an individual’s offer set remains understudied. This paper answers three questions regarding HDHPs. First, do HDHPs lower total medical spending, and is there a behavioral response or simply a shifting of costs to the individual? I find HDHPs lower spending by 16 percent and reduce utilization as predicted by demand theory. Second, I find reductions in hospital-based medical care spending account for 60 percent of the savings. Finally, contrary to recently published papers, I find evidence of discriminatory

[email protected]

Grant Gannaway

Xuan Li

Miguel Reina Ortiz

Weiwei Chen

of health care provider from which to obtain care. Occupational licensing restrictions ("scope of practice (SOP) laws") play a major role in these decisions by specifying the scope of treatment performed by various provider types. These restrictions vary across states, and several states have relaxed their laws in recent years with more states planning to do so in the future. Efficient patient sorting will become increasingly important given two recent trends in health care: (1) the rise of non-physician healthcare providers (such as nurse practitioners and physician assistants), and (2) the large predicted physician shortage (created in part by the ACA). As non-physicians (NPs) become more accessible, more patients will be able to choose whether to receive care from an NP instead of a physician (MD), and the efficiency of these choices will become a greater focus of health care policy. The increasing relaxation of SOP licensing laws is one factor contributing to the increased access to NPs. Similarly, in the presence of a physician shortage, many patients will be forced to sort to NPs to receive care. Thus, efficient

In this paper I study the relationship between patient sorting, SOP laws, and health care costs, disparities, quality, and outcomes. I first document both the types of patients and the types of care that sort to NPs versus MDs. I then take a machine learning approach to study the efficiency of sorting in the context of patient mis-prediction of personal risk and complexity of required treatment. Next, I exploit three specific natural experiments to study the types of care that are on the margin between provider types. The three experiments each affect the access to NPs relative to MDs, but through different channels. They are: the state-specific relaxation of SOP laws, insurer changes in relative copays between provider types, and a large government subsidy for the training of NPs at five different US medical schools (the Graduate Nurse Education Demonstration). I combine these experiments with the machine learning results on mis-prediction to estimate a personalized, claim-level

The final portion of the paper is devoted to understanding the broader implications of patient sorting over provider types on explanations of empirical facts in the health economics literature. I estimate the predicted effects of counterfactually [email protected]

With the rising of Internet, more and more websites are providing review information on health care providers. Differing from the traditional report cards, these reviews and ratings are typically written by patients themselves. Therefore, these reviews are easier for patients to understand and also addresses more of patients concerns. An obvious trend in recent years is that a growing number of patients rely on the information from these review websites to choose health care providers. We exploit the physician ratings from Vitals.com, one of the largest and most comprehensive physician-review websites in US, and inpatient claims data of coronary artery bypass graft (CABG) surgeries in Pennsylvania to examine the impact of online physician ratings on patients’ physician choices. Using a discrete choice model with random coefficients, we find that the probability that patients receive CABG surgery from high-rating surgeons is significantly higher than that from surgeons without

[email protected]

Many individuals with HIV/AIDS are not receiving treatment, in part because they are not aware of their status. The CDC and other health agencies recommend that all individuals be routinely tested for HIV/AIDS. Underdetection is particularly concerning in low- and middle-income countries because the transmission of the disease can stretch scarce public health resources. We conduct a randomized controlled field experiment in Ecuador, in a province that carries a disproportionate burden of HIV/AIDS. The overall goal of the study is to compare the effects of different strategies, namely information, a behavioral nudge (soft-commitment), and a $ 10 financial incentive (paid either at the time of testing or when the participant picks up their test results) in inducing voluntary HIV testing. In our study, we test these various strategies on a broad target population recruited in several well-transited locations in a major city in the province. Behavioral nudges and rewards have the potential to induce individual testing by overcoming psychological biases or bridging information gaps, and by overcoming social stigma concerns. Participant recruitment is in progress and is expected to be completed by December, 2017. Outcomes include percentage of participants deciding to get tested, percentage of participants picking up their test results, and percentage of participants being diagnosed with HIV. Preliminary results indicate that: 1. About 15% of subjects provided with "information only" agreed to get tested; 2. The "soft-commitment" opportunity did not have additional effects; 3. The $10 incentive paid at the time of testing increased the fraction of subjects who got tested to 60%; 4. The $10 incentive paid when the participants picked up their test results, instead, did not show any additional effect; 5. Between 1.5% and 2% of individual tested were HIV positive; 6. About 40% of non-incentivized subjects chose to learn their test results, vs. 20% of participants who received the incentive at the time of testing. Our preliminary results indicate that incentives provided at the time of testing can overcome economic or psychological barriers to get tested, although the relatively low proportion of

[email protected]

The start of the 2014-15 influenza season was overshadowed by the fear of Ebola. As the first Ebola patient diagnosed in the US died and two nurses confirmed the infection in October 2014, the anxiety of Ebola was elevated. As part of the massive media coverage on Ebola, many health experts compared Ebola with influenza and pointed out the importance of receiving influenza vaccination in that year. The rationale was that Ebola-related hospitalization and deaths were far less than those caused by influenza and having more people receiving influenza vaccines would reduce false alarms and help the public health system respond to Ebola. Meanwhile, CDC also recommended people being actively monitored for potential Ebola virus

This study examines the impact of public awareness around Ebola on influenza vaccine uptake during the 2014-15 influenza season in the US. We use individual-level data from the Behavioral Risk Factor Surveillance System and examine changes in the likelihood of receiving influenza vaccines. We examine the robustness of these findings in a doubly-robust difference-in-difference with propensity score matching framework as well as a synthetic control framework. We examine the impact of three types of treatment: residing in a state with an Ebola case (Texas and New York), residing in a state with Ebola or medically evacuated cases (Texas, New York, Georgia, Maryland, and Nebraska), or having higher level of attention to Ebola as

Our findings provide the first quantification of the spillovers from messaging targeting one rare disease to health behaviors related to a second, more common infection. Future immunization programs that consider phrasing their promoting [email protected]

Martin Salm

Alina Elrich

Julius Chen

In light of increasing health care expenditures, patient cost-sharing schemes have emerged as one of the main policy tools to reduce medical spending. In this study we show that health care utilization is affected not only by the economic incentives provided by cost-sharing schemes, but also by the way these economic incentives are presented. Specifically, we compare patients’ responses to a deductible and to a no-claim refund. The economic incentives under a deductible and a no-claim refund are very similar, but they are framed in a different way. Under a deductible policy, individuals pay out-of-pocket for all medical care up to the deductible limit. Under a no-claim refund policy, individuals receive a payment at the end of the year if their health care spending during the year was below the no-claim refund limit. Prospect theory predicts that individuals respond stronger to losses than to gains. If individuals perceive deductible payments as losses and lower no-claim

We make use of the fact that in the Netherlands, both schemes have been in place at different points of time while the patient population and the services covered by health insurance remained comparable. In the years 2006 and 2007 Dutch law has mandated that health insurance contracts included a no-claim refund, and from the year 2008 onward, health insurance contracts had to feature an annual deductible. Our analysis is based on unique claims-level data from a Dutch health insurer for the years 2006-2015 which we aggregate to around 9 million person month observations. In our empirical strategy we exploit variation in cost-sharing incentives within a year. Under both a deductible policy and a no-claim refund the price of healthcare utilization can vary over the course of the year depending on whether or not an individual has exceeded her deductible or no-claim refund limit. We examine how the reaction to prices differs between the years when a no-claim refund policy was in place and the years when a deductible policy was in place. We account for the possible endogeneity of prices with a simulated instrumental variables approach. As instrumental variable for the price at the beginning of the month we use

We find that patients react to comparable incentives twice as strongly when they are implemented as a deductible, which suggests that the framing of incentives can be quantitatively almost as important as the incentive itself. Our preferred explanation is that individuals are loss-averse and respond differently to both schemes because they perceive a deductible payment as a loss and a no-claim refund as a gain. Our results are robust to a number of sensitivity analyses. Specifically, our

[email protected]


We design a conventional economic laboratory experiment which simulates the course of events inherent in medical treatments from an economic perspective. Our experiment consists of two stages. The working stage mimics the beginning of the treatment and induces feelings of losses as subjects have to work on a task but only receive a fraction of their proper income. Entering the investment stage with these losses subjects decide over 12 periods between lottery A and lottery B. These lotteries represent the economic consequences of discontinuing and continuing with therapy. Lottery A with a higher risk of losing money can be chosen without any prior investments whereas for playing Lottery B with higher winning chances subjects have to invest some of their monetary endowment. Once a lottery is lost subjects drop out of the experiment and are not allowed to make any more decisions. We incorporate loss aversion and the timing in our incentive scheme in the following way: as soon as subjects have compensated the losses from the working stage they receive an up-front bonus which is added to their balance account. Subjects are only allowed to keep this bonus if they do not dropout of the experiment


[email protected]

Innovative organizational forms of health care delivery have recently developed that lower the time cost of care. While there is an extensive literature on consumer response to changes in the out-of-pocket price for care, little work has studied the equally-salient dimension of time cost. In this paper, I develop a theoretical model of patient decision-making and predict that when a new provider enters the market and offers services with lower time cost, patients engage in new utilization and/or substitution across providers. I then test these predictions in a unique empirical setting: A large corporation opened a worksite health clinic on its California campus in 2013, but did not feature a clinic on its Texas campus. I utilize novel data, 2011-2015 medical claims for the corporation’s employees. My primary empirical strategy is a difference-in-differences approach, where I compare California employees to the control group of Texas employees. I find that the effects of clinic availability are concentrated among the narrow set of services that can be provided onsite. For primary care in particular, I observe both an increase in utilization and substitution towards onsite care. While services beyond the clinic’s scope of practice are mostly unaffected, California employees reduce their utilization of outpatient care; this spillover effect is offset by a substantial increase in demand for office-based care. Ultimately, new consumption of primary care and other office-based services drives a

percentile of the conditional distribution, the estimated increase in monthly total spending is $14.67. My findings suggest that consumer demand is sensitive to changes in time cost, and this has [email protected]

Aditi Sen

Seema Kacker

To assess whether financial incentives for health behaviors crowd out individuals’ intrinsic motivation to engage in those behaviors. The use of financial incentives to promote changes in health behaviors is widespread among payers and employers, however there are concerns that if incentives crowd out intrinsic motivation, behavior would fall below even pre-incentive levels following the removal of incentives, hindering any long-run impact of incentives on behavior change. Further, consumers’ health-related decisions are likely impacted by the interaction between incentives and motivation. Few studies have assessed the impact of financial incentives on patients’ intrinsic motivation for health behaviors using direct measures of motivation. We examined this question in the context of five randomized controlled trials of financial incentives for health behavior change. We investigated whether effects varied by incentive type or behavior and assessed whether

We used the Treatment Self-Regulation Questionnaire to measure intrinsic motivation at baseline and at least once following the incentive intervention period in randomized controlled trials of financial incentives for weight loss (two studies), home health monitoring, walking among older adults, and adherence to use of a Positive Airways Pressure device for sleep apnea. In addition to varying health behaviors, these trials utilized different forms of incentives, including conditional

Multivariate regressions with participant-level data and random effects were used to assess the relationships between baseline and change in intrinsic motivation and performance in each study, measured as achieving study goals. Similar

First, we found that an increase in intrinsic motivation during the intervention was associated with increased odds of success in the program, defined as achieving program goals such as a pre-determined weight loss target. Second, we found no evidence of crowding out of intrinsic motivation by incentives; that is, there was no significant association between incentive eligibility or receipt and the odds of a decrease in intrinsic motivation pre- versus post-incentives. The lack of evidence of

Financial incentives did not crowd out intrinsic motivation across a range of health behaviors and incentive designs. Improving our understanding in this area is critical in order to understand consumer decision-making in the context of [email protected]

Purpose: Despite limited availability of free eye care services in Baltimore City, utilization by low-income at-risk minority individuals remains low. This may be driven by a lack of perceived value for these free services. We examine the effect of

Methods: A cluster randomized trial was conducted within the SToP Glaucoma study, an investigation of a community-based screening program which identifies glaucoma suspects and offers them free follow-up appointments at the Wilmer Eye Institute. Appointments are scheduled at the time of screening and reminder calls are made for all patients. Screening events were randomized to standard verbal and written counseling offering the individual a free appointment, or counseling in addition to provision of one of two types of vouchers redeemable for free appointments. Both voucher types included the patient’s name, the appointment date and an expiration date 90 days following the screening. One also included the approximate monetary value of the service ($250). The primary outcome was presenting for follow-up within the voucher eligibility period. A hierarchical mixed-effects logistic model allowing for random effects from the screening event was used to

Results: Follow-up through November 2017 yielded complete data for 431 glaucoma suspects identified at one of 64 screening events. Overall, 76% of individuals were African American, 65% were female, and the mean age was 69. There were no significant differences in these factors between study arms. Those referred in the traditional manner had a 49% attendance rate, whereas 67% of individuals receiving a voucher without monetary value information and 62% of individuals receiving a voucher with monetary value information presented. For a given screening event, offering vouchers without monetary value information increased the odds of presenting for follow-up by 152% (p =.03) compared to not offering a voucher. Offering

Conclusions: Offering vouchers for redemption of free eye care services increases utilization. Voucher provision may increase perceived value for these services, particularly among low-income minority populations. Further investigation into the [email protected]

Ying Cao

Eric VanEpps

Thomas Hoerger [email protected]

This study investigates the life partners’ influence on each other’s preventative health service usage among those over age 50, and further, disentangles the active learning and passive imitation channels among spousal concordance. The study also

The study uses a biennial and representative dataset, Health & Retirement Study (HRS) 2006-2014. Four study cohorts are included, HRS (born between 1931-1941), WB (war baby, 1942-1947), EBB (early baby boomer, 1948-1953) and MBB (middle

The study focuses on individuals with stable life partners throughout the study period. The outcome variables are whether or not a respondent engaged in (initiation or drop-out) preventative activities over years, i.e. a flu shot, a blood test for cholesterol screening and rigorous physical activities. In order to capture positive and negative partner influences, a life partner’s previous usage pattern is classified into one of four categories: Non-User, those life partners who did not use a certain

, those who did not use in earlier wave but started to use in later wave. Self-reported health status changes and physician verified new chronic conditions for both self and the partner are used as proxies of health outcomes to gauge the learning channel. The main research of interest is the extent to which the interactions between the spousal behaviors and health outcomes would predict preventive activity engagement and changes. Risk-aversion and time-discounting are also used to explore the potential mediation effects.

Couples show concordance in preventive behaviors, and further, start using a preventative activity by a life partner encourages the other couple more than stop using would discourage. The learning channel is supported by the evidence that worse-off self-health and better-off spousal-health exaggerate the encouraging effects, while better-off self-health and worse-off spousal-health exaggerate the discouraging effects. Moreover, those with higher risk-aversion and time-discounting (less

Findings in the study suggest that the partners' influences are asymmetric and happen through active learning. The study provides insights on the potential efficacy of family-based promotion strategies to increase preventative activity engagement, [email protected]

In two studies—an online experiment using a real effort task and a field study of weight loss among Weight Watchers members—we investigate how demand for difficult goals changes over time in the presence or absence of goal-contingent incentives. To facilitate motivation and self-control in an effortful task, people may choose aggressive goals. However, because individuals in incentive conditions only earn bonus payments when they achieve their goals, they simultaneously have an

In Study 1, we recruited online workers to complete three trials of a computerized effort task and asked them to set performance goals for each trial. Some workers were offered bonus incentives that were contingent upon goal achievement, thus allowing these participants to ensure incentive collection by choosing an easy goal or to put their incentive at risk by selecting a difficult goal. Other workers were offered equivalently-sized incentives that were independent from goal achievement, whereas yet other workers received no bonus incentives. We found that the provision of goal-contingent incentives reduced demand for difficult goals relative to conditions in which no incentives were offered or in which incentives were not contingent upon goal achievement. That is, people behaved rationally in selecting goals when incentives were made contingent upon goal achievement in a context where participants likely had little intrinsic motivation regarding the task itself (a

In Study 2, we partnered with Weight Watchers to conduct a field experiment and to determine how contingent incentives influence the selection of weight goals over a six-month period. 191 Weight Watchers members were randomly assigned to either a control condition of daily weigh-ins and daily feedback, or to an incentive condition with daily weigh-ins, daily feedback, and a daily financial incentive of $2.80 for meeting a weight loss goal. Participants selected weight loss goals each month. Despite having an economic incentive to choose the most modest goal possible, nearly 90% of incentivized participants picked an initial goal that was more challenging than the minimum needed. However, as people gained experience with

Across both studies, we find substantial initial demand for goals that are more ambitious than strictly required, regardless of incentive provision. In the absence of goal-contingent incentives, this demand for difficult goals remains high over time. However, when considering commitment devices in which monetary gains are foregone if goals are not met, people are responsive to the likelihood of goal achievement and demonstrate dynamic demand over iterated trials.

[email protected]

Medicaid beneficiaries face substantial burden from chronic diseases. Incentive programs have been proposed to connect and engage Medicaid beneficiaries with preventive services, thereby changing behavior and preventing chronic diseases. We evaluated the impact of the Medicaid Incentives for Prevention of Chronic Disease program, which funded Medicaid incentive initiatives in 10 states. The focus of initiatives varied by state and included smoking cessation, diabetes prevention, weight loss, and disease management. The value of incentives also varied widely, ranging from a maximum of $50 in one state to over $1,000 annually in another state. Most states randomly assigned participants to incentive and control arms, with both

We conducted a mixed methods evaluation that included document review, site visits, focus groups, a beneficiary survey, and Medicaid claims analysis. The claims analysis used regression and difference-in-difference methods to test whether

We find that states can implement Medicaid programs successfully, although they had to overcome administrative challenges and recruiting participants was more difficult than anticipated. Participants in the incentive arms attended significantly more diabetes prevention and weight watchers classes, made more smoking quitline calls, and attended more smoking cessation sessions. Participants who received incentives were very satisfied with program access and quality and believed that the incentives helped them meet their health goals. The programs had no significant impact on the use and costs of other Medicaid services during the evaluation. Administrative costs were relatively high. State evaluations suggest that the incentives had mixed effects on short-term health outcomes, with some significant reductions in smoking and generally insignificant effects on diabetes prevention, weight loss, and risk factor management. Future study is needed to determine whether the

Christian Schmid

Chanup Jeung

Brian McGarry

From a public health perspective, the flu vaccination rate is viewed as too low in many countries. While up until recently, only physicians were allowed to vaccinate in Switzerland, some pharmacies received the license to vaccinate during the course of the year 2016. Public health authorities hope to increase the vaccination rate as there is generally no need for an appointment in a pharmacy. This setting allows us to test whether the vaccination rate can be increased by a letter informing

We use a randomized control trial where a large Swiss health insurer sent letters to 22'000 of its customers (between 50 and 75 years of age). The letter included information about the new possibility to get a flu shot in a pharmacy, a pledge to bear the costs and an indication of the nearest pharmacy which had the license to vaccinate. The letter increased immunization rates by 2.7%-points (17.9%) with significant heterogeneity in background variables as well as with respect to the distance to the pharmacy. We find some provider substitution effects from physicians to pharmacists. Yet, the majority of the increase in the vaccination rate is driven by additional vaccinations in the pharmacies. More than half of the customers visited the recommended pharmacy even if a closer one had been available. Currently, we do not find any significant effects of the flu shot on covered health care expenditures nor on the number of doctor visits. Information letters proofed to be an effective tool to increase the use of prevention care as well as to guide consumers to specific health care providers. The new possibility to vaccinate in a pharmacy increased the take-up rate of the vaccination. christian.sc

[email protected]

Do state laws that require the availability of paid sick leave increase the use of preventive services? Paid sick leave benefits allow workers to maintain job security when they leave for medical reasons. In recent years, paid sick leave laws have received more attention by health policy makers in terms of its potential to improve public health. Currently, seven states and D.C. have laws that require employers to provide paid sick leave. However, empirical evidence on this topic is limited (DeRigne, et al 2017; Peipins et al 2012). Although prior studies examined the association of paid sick leave and preventive service use, these studies did not use rigorous study designs. Because workers with higher demand for healthcare services

The goal of this study is to use a quasi-experimental study design to estimate the impact of Connecticut’s 2012 paid sick leave law on the use of preventive services. Connecticut was the first state to require private employers to offer paid sick leave benefits to their employees. Using state and time variation from 2007-2016 Behavioral Risk Factor Surveillance System (BRFSS) data, we compare the use of preventive services in Connecticut and in other New England states before and after the implementation of the 2012 paid sick leave law. For general preventive service outcomes, we examined routine checkups, flu vaccinations, and dental visits in the past year. We also examined the use of Pap tests, clinical breast exams, and

Overall, we found that Connecticut’s 2012 paid sick leave law increased the use of preventive services. Specifically, the rate of routine checkups (1.4%, p<.1), flu shots (2.0%, p<.05), and dental visits (2.7%, p<.01) increased over this period. With respect to cancer screening outcomes for women, we found that the use of Pap tests (9.4%, p<.01), and clinical breast exams (4.4%, p<.01) increased. However, the higher rate of mammograms (1.3%, p=.38) was not statistically significant. Our findings provide rigorous evidence on the positive impact of a state’s paid sick leave law on preventive service outcomes. These empirical results also suggest that a lack of paid sick leave among some private employers may represent a barrier to

[email protected]



: Simplifying financial information resulted in the selection of lower cost plans in all treatment groups relative to the control group. These improvements were largely driven by increases in the weight placed on OOP costs and reductions in the weighting of deductibles. Respondents in the “total cost, premium, and OOP cost” group had decision weights that were most consistent with cost-minimization: they placed equal weight on premium and OOP costs and no additional weight on cost-sharing attributes beyond their direct impact on anticipated spending. The “total cost and premium” group continued to weight premiums more than OOP costs, while the “total cost only” group was more likely to seek additional plan information

: Simplifying the financial information on Plan Finder helps individuals adhere to a cost-minimization decision rule. Displaying plan premiums and OOP costs alongside total cost estimates was most effective in encouraging cost- [email protected]

Chaoran Guo

Tamara Bischof


[email protected]


This paper studies closures of primary-care practices in Switzerland from 2005 to 2015 to estimate the causal impacts of discontinuities of primary care on patients' utilization patterns, medical expenditures and health-related outcomes. Employing a difference-in-difference framework, we identify causal effects by comparing changes in outcomes between an affected group of patients (‘treatment group’) and an unaffected group that does not experience changes in primary care provision (‘control group’). Our main findings are twofold. First, when faced with a discontinuity of primary care, patients adjust their utilization pattern by shifting visits away from ambulatory primary care providers (-5%) towards specialized care (+10%) and emergency departments of hospitals (+14%). Secondly, practice closures increase patients’ total health care expenditures by 4.6% and raise the probability of incurring non-zero costs in a given month by 10%. Two policy-relevant implications are that practice closures may lead to an inefficient use of healthcare services and have adverse effects on social health insurance which must cover higher costs. These implications are relevant for health planners and policy makers, health insurers and tamara.bisc

[email protected]

Complete

Complete

Complete

Complete

Presenting Author

AffiliationCo-

Author(s)Submission Complete

University of Illinois at Chicago

Kevin Callison; Robert Kaestner

New Economic School, CEFIR

Rutgers School of Public Health

Rizie Kumar; Alan Monheit

Economics Department, University of Illinois at Chicago

Complete

Complete

Complete

Complete

University of Chicago

Lehigh University

Mengcen Qian

University of South Florida

Ricardo Izurieta; Phillip Phan; Enrique Teran; Michelle Grunauer; Mario Macis

Florida International University

Charles Stoecker

Complete

Complete

Complete

Tilburg University

Arthur Hayen; Tobias Klein

Paderborn University

René Fahr; Behnud Djawadi

Harvard University

Complete

Complete

Johns Hopkins Bloomberg School of Public Health

Debra Gilbert; David Asch; George Loewenstein; Kevin Volpp

Johns Hopkins University

Prateek Gajwani; Eliseo Guallar; David Friedman; Mario Macis; Natasha Kanwar; Di Zhao

Complete

Complete

Complete

State University of New York at Buffalo

Gregory G. Homish; Ekaterina Noyes

University of Utah

Kevin Volpp; Jingsan Zhu; William Yancy

RTI International

Rebecca Perry; Maria Alva; Melissa Romaire

Complete

Complete

Complete

CSS Institute for Empirical Health Economics

Lukas Kauer

George Mason University

Kyung Min Lee; Gilbert Gimm

Harvard Medical School Health Care Policy

David Grabowski; Nicole Maestas

Complete

Boris Kaiser Complete

UC Berkeley

University of Bern, Department of Economics


Costs and Benefits of Specific Medical Treatments

Intravitreal aflibercept compared to ranibizumab for wet age related macular degeneration in the ‐US: cost-effectiveness analysis


The causal effect of ambulance response time on cardiocirculatory morbidity and mortality


Health literacy and treatment selection: Application to stable angina


High-Dose versus Standard-Dose Influenza Vaccination among Veterans Health Administration Patients: An Instrumental Variable Analysis


Are Medical Prices Still Declining? A Systematic Examination of Quality-Adjusted Price Index Alternatives for Medical Care


Cost-effectiveness of various canine rabies post-elimination vaccination strategies to prevent re-establishment of dog rabies


Comparing Cost Effectiveness of Aripiprazole Augmentation with Other “Next-Step” Depression Treatment Strategies

Abstract

Background Wet age-related macular degeneration (wAMD) is the leading cause of visual impairment and blindness in the United States (US). Intravitreal aflibercept injection (IAI) and ranibizumab (RBZ) are the two most commonly used anti-vascular endothelial growth factor treatments approved by the US Food and Drug Administration for the treatment of wAMD. Randomized clinical trials in patients with wAMD have shown similar efficacy, safety, and tolerability between IAI administered every 8 weeks after 3 initial monthly injections, IAI monthly, and monthly RBZ. Therefore, associated costs and cost-effectiveness comparisons are key factors to determine which treatment represents a more rational investment of scarce health care resources. This assessment may help address the increasing cost of prescription drugs in the US, a source of concern for patients, prescribers, payers, and policy makers. Objective To assess the cost-effectiveness of IAI (2mg every 8 weeks after 3 initial monthly doses [2q8]) vs. RBZ (intravitreal, 0.5mg monthly) and RBZ (intravitreal, 0.5mg as needed [PRN]) for wAMD from a US payer perspective. Methods A Markov cohort model was developed to estimate the lifetime quality-adjusted life years (QALYs) and costs of treating patients with wAMD with IAI 2q8, RBZ dosed monthly, and RBZ dosed PRN. The model considered changes in best-corrected visual acuity (BCVA) in the affected and fellow eye over time, and the impact of blindness on mortality. Efficacy data for IAI 2q8 and RBZ dosed monthly, for the first 52 weeks, were from the VIEW 1 and VIEW 2 studies, and from the CATT trial for RBZ dosed PRN. Patients were assumed to be treated for a maximum of 2 years, with BCVA remaining stable between the first and second year in patients remaining on treatment. Natural progression of BCVA loss was applied in patients discontinuing treatment. Background mortality, from US life tables, was adjusted with published hazard ratios for wAMD and blindness. Utilities were from published literature and based on BCVA in the best-seeing eye. Costs (drug acquisition, monitoring, blindness) were from published literature and estimated in 2016 US dollars. Health outcomes and costs were discounted at 3% per annum. Results Over a lifetime, IAI 2q8 provided equal health benefits vs. RBZ dosed monthly (5.44 QALYs) at a lower total cost ($33,795 versus $48,031) as a result of fewer injections. IAI yielded incremental health benefits vs. RBZ dosed PRN (5.44 versus 5.40 QALYs) at a higher cost ($33,795 versus $33,652), with an incremental cost per QALY gained of $2,715. Results were sensitive to variations in drug acquisition costs and number of injections of both drugs, and the baseline age of the cohort. Conclusion IAI 2q8 can be cost-saving and cost-effective compared with RBZ dosed monthly and RBZ dosed PRN, respectively, for the treatment of wAMD in the US.

I document that a one-minute increase in ambulance response time (RT) for cardiovascular emergency calls increases patient’s morbidity by 2 percentage points and out-of-hospital mortality by 1 percentage point. Patient’s morbidity is recorded at the ambulance arrival on the scene, before performing any medical treatment: in this the compounding effect given specific characteristics of the system in analysis is mitigated, and the result may be generalized. Finally, I find no evidence about the existence of a critical threshold for RT. I perform the analysis using administrative data on emergency calls and ambulance missions performed in the Italian region Liguria in two years. I identify the causal effect of interest by instrumenting RT with rainfall amount at the hourly and municipality level and medical personnel work shift.

Introduction: Low health literacy may act as a barrier to appropriate treatment for stable angina. The treatment alternatives have similar outcomes, but prior research suggests that many patients mistakenly believe more interventional treatment reduces the risk of mortality and myocardial infarction. We hypothesized that low compared to high health literacy would be associated with selection of more interventional treatment. Methods: Secondary analysis of fee-for-service (FFS) Medicare beneficiaries (20% random sample) using Parts A, B and D data. The inclusion criteria were: 1) incident diagnosis of stable angina in 2007-2014), 2) twelve months of Medicare FFS prior to diagnosis, 3) six months of Medicare FFS and Part D after diagnosis, 4) did not receive CABG or PCI prior to diagnosis, and 5) no diagnoses for unstable angina or acute myocardial infarction prior to stable angina diagnosis. The three treatment alternatives are: medication only; percutaneous coronary intervention (PCI), and coronary artery bypass grafting (CABG) surgery. Patients were categorized to PCI or CABG if they received the procedure within twelve months of the first observed diagnosis code for angina, coronary artery disease, or chest pain. Otherwise, patients were categorized to ‘medication only’ if they filled at least one claim for a medication used to treat stable angina and did not receive PCI or CABG. The key independent variable was an area-based health literacy measure at the census block group level derived from a validated predictive model using demographic characteristics. This variable was specified as binary with low vs. high health literacy and quartiles based on the national distribution. The relationship between treatment selection and health literacy was evaluated using multinomial logistic regression, and the results are presented as average marginal effects. Control variables included demographics (age, sex, race/ethnicity), programmatic characteristics (dual eligibility status, receipt of low-income subsidy), Charlson comorbidity index, year of treatment, and state fixed effects. Results: The sample had 17,215 beneficiaries, of which 1,922 and 15,293 were in the low and high health literacy categories, respectively. Patients in the low health literacy category were more likely to be dual-eligible, receive the low-income subsidy, and be a racial or ethnic minority. For the binary specification, patients with low health literacy were more likely to receive medication only (2.8 percentage points [0.1 to 5.5% points]) and less likely to receive CABG (-2.9 percentage points [-4.9 to -0.9 percentage points]). For the quartile specification, the probability of receiving PCI significantly decreased with lower literacy for each quartile, while the probability of medication only significantly increased. The magnitude of the marginal effects for both specifications decreased after controlling for area deprivation separate from literacy. Conclusions: Contrary to the hypothesis, lower health literacy was associated with less use of invasive treatment for stable angina. Other factors such as patient preferences, physician preferences, and symptom severity may explain the use of less invasive treatment alternatives. The results may also reflect access to care since the less intensive treatment of medication only was also relatively higher among groups that have been historically underserved (low-income and racial/ethnic minorities).

Background In randomized trials of adults aged 65 years or older, a high-dose influenza vaccine (HD) was more efficacious in preventing pneumonia admissions compared with the standard-dose vaccine (SD). An observational study in the Veterans Health Administration (VHA) population, however, did not show any additional benefit of HD over SD for this outcome. We reassessed the relative vaccine effectiveness (rVE) of HD vs. SD in the VHA population using the instrumental variable (IV) method to better control for confounding by indication, especially unmeasured confounding. Positive rVE indicates greater efficacy of HD compared to SD. Methods The study included Veterans aged 65 or older who received an influenza vaccine in the 2011-12 influenza season at the VHA. Electronic medical records were used to capture the treatment, outcome (hospitalizations due to pneumonia or influenza and all-cause hospitalizations) and baseline patient characteristics. The IV was the proportion of HD recipients at each VHA facility in the prior season, defined as the number of HD recipients divided by the total number of both HD and SD recipients at each VHA facility. The instrument was defined using the HD proportion from the prior season (2010-11), because it is a strong predictor of receipt of HD in 2011-2012 but less likely to be related to the health or treatment of patients in the 2011-2012 season. Because the dependent variable was a count variable (number of hospitalizations), we performed IV Poisson regression analysis, adjusting for age, sex, race, socioeconomic status as well as baseline comorbidities. We performed tests of the Correlation and the Exclusion Restriction criteria for IV analysis. Finally, we conducted a falsification test, using hospitalization associated with urinary tract infection as the outcome; receipt of HD versus SD vaccine should not affect this outcome. Results Our study population included 678,386 VHA patients who were vaccinated at 1,027 VHA facilities (medical centers and community-based outpatient clinics) in the 2011-2012 season. Of these, 20,880 patients (3%) received HD, while 657,506 patients (97%) received SD. The IV adjusted rVE estimate of HD was 47% (95% CI, 27%–62%) against influenza- or pneumonia-associated hospitalization, and 13% (95% CI, 4%-21%) against all-cause hospitalization. We found that our instrument was strongly predictive of receipt of HD. An F statistic greater than 10 is generally considered sufficient. The F statistic for our sample was 916,371. Every increment of 10% in a facility’s prior HD proportion was associated with doubling the likelihood of a patient being provided with HD. In addition, the instrument was not correlated with other factors that would affect the medical outcomes of patients, such as facility quality, facility type, regions, rurality or the overall health status of patients treated by the facility. The falsification test was confirmative as the IV adjusted rVE estimate of HD was 4% (95% CI, -22%–25%) against hospitalization associated with urinary tract infection. Conclusions HD is more effective than SD in protecting against hospitalizations due to pneumonia or influenza and against any hospitalization in the VHA population during the 2011-12 influenza season.

Health care spending has grown rapidly over the past several decades. However, whether this growth rate may be considered too high depends on the value of the increased length and quality of life due to medical treatment as well as on the increased spending. One way to capture the tradeoff between benefits and costs is with the framework of price measurement. Previous literature has constructed quality-adjusted price indexes for different medical conditions such as acute myocardial infarctions (Cutler et al. 1998) and major depression (Berndt et al. 2002). In general, these studies found that the increased benefits of the technological advances in medical care greatly surpassed the increased spending and therefore that price growth in medical care when measured this way was negative. There is no consensus, however, on what method to use for quality adjustment in medical price indexes. In this paper, we explore the different methods both theoretically and empirically with a view toward exploring how the methods could be implemented in practice. The theoretical model shows that generally the different methods produce the same results when the increase in monetized benefits is approximately equal to the increase in monetized costs. To explore the methods empirically, we construct quality-adjusted price indexes for three acute medical conditions (AMI, heart failure, and pneumonia) for the period 2001-2014 using data for Medicare fee-for-service (FFS) beneficiaries. We measure spending with total spending per patient around the hospitalization, with varying windows and we isolate the contribution of medical care to improvements in outcomes by measuring short-term mortality following the hospitalization. For both our spending measures and outcome measures, we risk-adjust based on the characteristics of the patient. We find spending per patient rose from 2001-2010 and stayed level from 2010 to 2014 while mortality exhibited the reverse pattern. We also find that quality adjustment has a large effect on price growth. As measured by a benchmark cost-of-living index (COLI) where the benefits are measured by the utility value of increases in life expectancy, the average prices for these conditions decline steeply from 2001 through 2010. Following 2010, the declines in quality-adjusted prices level off, at the same time that we observe spending per patient holding flat. In other words, the quality-adjusted prices and the trends in spending per patient appear to move in different directions, highlighting the important role of quality adjustment. When other quality adjustment methods are applied, we find a wide dispersion in the growth rates of different quality-adjusted price indexes. According to our theoretical model, this dispersion results from the increases in benefits exceeding the increases in spending for these conditions; only the benchmark COLI fully captures the value of the increases in benefits to patients.

Human rabies, mostly caused by the canine rabies virus variant, inflicts a heavy burden with approximately 59,000 annual deaths globally. While most countries in the Americas and Europe have eliminated rabies from their dog populations, 122 countries are still dealing with endemic canine rabies. The World Health Organization (WHO) recommends that these countries vaccinate 70% of their dogs each year for seven years in order to end transmission permanently. However, it is unknown what level of vaccination is necessary after these seven years to prevent re-establishing dog rabies from an importation or host-shift event. We consider several different scenarios whereby canine rabies is introduced into a rabies-free population. Dog-to-dog and dog-to-human rabies transmission is defined using an SEIR (susceptible-exposed-infectious-recovered) compartmental model. For each reintroduction scenario, we compare cost-effectiveness of various vaccination strategies including discontinuation (baseline), low-to-high coverage, and vaccination maintained only along the border with the enzootic neighbor. The economic costs and benefits of canine vaccine programs is measured in terms of the number of rabid dogs averted, number of human rabies deaths averted and the average cost per rabies-related human deaths averted over a 20-year horizon. Demonstrating the cost-effectiveness of various vaccination strategies to maintain rabies free status could help countries decide how to implement policies for ongoing vaccination post-elimination.

Background: Atypical antipsychotic drugs are widely prescribed for major depressive disorder as a second-line therapy to augment antidepressant use despite little evidence regarding their comparative cost effectiveness. Atypical antipsychotics such as aripiprazole had much higher prices when first introduced, but recent generic equivalents have substantially narrowed price differentials between them and older antidepressant drugs. We used data from a randomized clinical trial to compare the cost effectiveness of augmenting standard antidepressant therapy with aripiprazole compared to another augmentation agent, bupropion, and to switching to bupropion over a 12-week acute treatment phase. Methods: The cost-effectiveness analysis (CEA) was conducted as part of the Veterans Affairs (VA) augmentation and switching treatments for improving depression outcomes (VAST-D) trial in which 35 participating VA medical centers enrolled 1,522 patients who had failed prior pharmacotherapy. Remission from depression and quality-adjusted life years were estimated from trial data collected at baseline and 12 weeks after randomization. Health care costs from 2015 were obtained from VA administrative data. We compared the cost effectiveness of the 3 strategies by estimating the costs per remission with 12 weeks as the time horizon and the health care sector as the primary perspective. We calculated the incremental cost-effectiveness ratio (ICER) using the difference in costs between each treatment strategy versus the other divided by the differences in remission rates at 12 weeks. We calculated 95% confidence intervals around the ICER’s using bootstrap methods. Results: The mean age of participants enrolled in the trial was 54 years, and participants were predominantly male. The rate of remission at 12 weeks was highest for the aripiprazole augmentation arm (29%), followed by bupropion augmentation (27%), and lowest for bupropion monotherapy (22%). Mean mental health care costs which included the costs of outpatient mental health care visits, inpatient psychiatric stays, and the study drugs did not differ significantly between the groups. The incremental cost effectiveness ratio (ICER) comparing costs per remission was lowest for the bupropion augmentation group relative to the bupropion monotherapy group at -$640/remission. The ICER for the aripiprazole augmentation versus bupropion monotherapy group was $1,074/remission (95% CI =47-5022) with 97.9% of the observations in upper-right quadrant, indicating greater costs and benefits associated with aripiprazole. The ICER for aripiprazole augmentation versus bupropion augmentation was $5,094/remission (95% CI =-34027-32774) with 75.6% of the observations in the upper right quadrant. We did not find any significant differences in mental health care costs, quality-adjusted life years, employment, and other work and social adjustment outcomes between treatment groups during follow up. Conclusion: In treatment of non-responsive depression, augmentation with either aripiprazole or bupropion increased costs as compared to switching to a new antidepressant, but these costs were justified by higher remission indicating that both augmentation approaches were cost-effective compared to switching antidepressants.

Abstract

Wet age-related macular degeneration (wAMD) is the leading cause of visual impairment and blindness in the United States (US). Intravitreal aflibercept injection (IAI) and ranibizumab (RBZ) are the two most commonly used anti-vascular endothelial growth factor treatments approved by the US Food and Drug Administration for the treatment of wAMD. Randomized clinical trials in patients with wAMD have shown similar efficacy, safety, and tolerability between IAI administered every 8 weeks after 3 initial monthly injections, IAI monthly, and monthly RBZ. Therefore, associated costs and cost-effectiveness comparisons are key factors to determine which treatment represents a more rational investment of scarce health care resources. This assessment may help address the increasing cost of prescription drugs in the US, a source of concern for patients, prescribers, payers, and policy makers.

To assess the cost-effectiveness of IAI (2mg every 8 weeks after 3 initial monthly doses [2q8]) vs. RBZ (intravitreal, 0.5mg monthly) and RBZ (intravitreal, 0.5mg as needed [PRN]) for wAMD from a US payer perspective.

A Markov cohort model was developed to estimate the lifetime quality-adjusted life years (QALYs) and costs of treating patients with wAMD with IAI 2q8, RBZ dosed monthly, and RBZ dosed PRN. The model considered changes in best-corrected visual acuity (BCVA) in the affected and fellow eye over time, and the impact of blindness on mortality. Efficacy data for IAI 2q8 and RBZ dosed monthly, for the first 52 weeks, were from the VIEW 1 and VIEW 2 studies, and from the CATT trial for RBZ dosed PRN. Patients were assumed to be treated for a maximum of 2 years, with BCVA remaining stable between the first and second year in patients remaining on treatment. Natural progression of BCVA loss was applied in patients discontinuing treatment. Background mortality, from US life tables, was adjusted with published hazard ratios for wAMD and blindness. Utilities were from published literature and based on BCVA in the best-seeing eye. Costs (drug acquisition, monitoring, blindness) were from published literature and estimated in 2016 US dollars. Health outcomes and costs were discounted at 3% per annum.

Over a lifetime, IAI 2q8 provided equal health benefits vs. RBZ dosed monthly (5.44 QALYs) at a lower total cost ($33,795 versus $48,031) as a result of fewer injections. IAI yielded incremental health benefits vs. RBZ dosed PRN (5.44 versus 5.40 QALYs) at a higher cost ($33,795 versus $33,652), with an incremental cost per QALY gained of $2,715. Results were sensitive to variations in drug acquisition costs and number of injections of both drugs, and the baseline age of the

IAI 2q8 can be cost-saving and cost-effective compared with RBZ dosed monthly and RBZ dosed PRN, respectively, for the treatment of wAMD in the US.

I document that a one-minute increase in ambulance response time (RT) for cardiovascular emergency calls increases patient’s morbidity by 2 percentage points and out-of-hospital mortality by 1 percentage point. Patient’s morbidity is recorded at the ambulance arrival on the scene, before performing any medical treatment: in this the compounding effect given specific characteristics of the system in analysis is mitigated, and the result may be generalized. Finally, I find no evidence about the existence of a critical threshold for RT. I perform the analysis using administrative data on emergency calls and ambulance missions performed in the Italian region Liguria in two years. I identify the causal effect of interest by instrumenting RT with rainfall amount at the hourly and municipality level and medical personnel work shift.

Low health literacy may act as a barrier to appropriate treatment for stable angina. The treatment alternatives have similar outcomes, but prior research suggests that many patients mistakenly believe more interventional treatment reduces the risk of mortality and myocardial infarction. We hypothesized that low compared to high health literacy would be associated with selection of more interventional treatment.

Secondary analysis of fee-for-service (FFS) Medicare beneficiaries (20% random sample) using Parts A, B and D data. The inclusion criteria were: 1) incident diagnosis of stable angina in 2007-2014), 2) twelve months of Medicare FFS prior to diagnosis, 3) six months of Medicare FFS and Part D after diagnosis, 4) did not receive CABG or PCI prior to diagnosis, and 5) no diagnoses for unstable angina or acute myocardial infarction prior to stable angina diagnosis. The three treatment alternatives are: medication only; percutaneous coronary intervention (PCI), and coronary artery bypass grafting (CABG) surgery. Patients were categorized to PCI or CABG if they received the procedure within twelve months of the first observed diagnosis code for angina, coronary artery disease, or chest pain. Otherwise, patients were categorized to ‘medication only’ if they filled at least one claim for a medication used to treat stable angina and did not receive PCI or CABG. The key independent variable was an area-based health literacy measure at the census block group level derived from a validated predictive model using demographic characteristics. This variable was specified as binary with low vs. high health literacy and quartiles based on the national distribution. The relationship between treatment selection and health literacy was evaluated using multinomial logistic regression, and the results are presented as average marginal effects. Control variables included demographics (age, sex, race/ethnicity), programmatic characteristics (dual eligibility status, receipt of low-income subsidy), Charlson comorbidity index, year of treatment, and state fixed effects.

The sample had 17,215 beneficiaries, of which 1,922 and 15,293 were in the low and high health literacy categories, respectively. Patients in the low health literacy category were more likely to be dual-eligible, receive the low-income subsidy, and be a racial or ethnic minority. For the binary specification, patients with low health literacy were more likely to receive medication only (2.8 percentage points [0.1 to 5.5% points]) and less likely to receive CABG (-2.9 percentage points [-4.9 to -0.9 percentage points]). For the quartile specification, the probability of receiving PCI significantly decreased with lower literacy for each quartile, while the probability of medication only significantly increased. The magnitude of the marginal effects for both specifications decreased after controlling for area deprivation separate from literacy.

Contrary to the hypothesis, lower health literacy was associated with less use of invasive treatment for stable angina. Other factors such as patient preferences, physician preferences, and symptom severity may explain the use of less invasive treatment alternatives. The results may also reflect access to care since the less intensive treatment of medication only was also relatively higher among groups that have been historically underserved (low-income and

In randomized trials of adults aged 65 years or older, a high-dose influenza vaccine (HD) was more efficacious in preventing pneumonia admissions compared with the standard-dose vaccine (SD). An observational study in the Veterans Health Administration (VHA) population, however, did not show any additional benefit of HD over SD for this outcome. We reassessed the relative vaccine effectiveness (rVE) of HD vs. SD in the VHA population using the instrumental variable (IV) method to better control for confounding by indication, especially unmeasured confounding. Positive rVE indicates greater efficacy of HD compared to SD.

The study included Veterans aged 65 or older who received an influenza vaccine in the 2011-12 influenza season at the VHA. Electronic medical records were used to capture the treatment, outcome (hospitalizations due to pneumonia or influenza and all-cause hospitalizations) and baseline patient characteristics. The IV was the proportion of HD recipients at each VHA facility in the prior season, defined as the number of HD recipients divided by the total number of both HD and SD recipients at each VHA facility. The instrument was defined using the HD proportion from the prior season (2010-11), because it is a strong predictor of receipt of HD in 2011-2012 but less likely to be related to the health or treatment of patients in the 2011-2012 season. Because the dependent variable was a count variable (number of hospitalizations), we performed IV Poisson regression analysis, adjusting for age, sex, race, socioeconomic status as well as baseline comorbidities. We performed tests of the Correlation and the Exclusion Restriction criteria for IV analysis. Finally, we conducted a falsification test, using hospitalization associated with urinary tract infection as the outcome;

Our study population included 678,386 VHA patients who were vaccinated at 1,027 VHA facilities (medical centers and community-based outpatient clinics) in the 2011-2012 season. Of these, 20,880 patients (3%) received HD, while 657,506 patients (97%) received SD. The IV adjusted rVE estimate of HD was 47% (95% CI, 27%–62%) against influenza- or pneumonia-associated hospitalization, and 13% (95% CI, 4%-21%) against all-cause hospitalization. We found that our instrument was strongly predictive of receipt of HD. An F statistic greater than 10 is generally considered sufficient. The F statistic for our sample was 916,371. Every increment of 10% in a facility’s prior HD proportion was associated with doubling the likelihood of a patient being provided with HD. In addition, the instrument was not correlated with other factors that would affect the medical outcomes of patients, such as facility quality, facility type, regions, rurality or the overall health status of patients treated by the facility. The falsification test was confirmative as the IV adjusted rVE estimate of HD was 4% (95% CI, -22%–25%) against hospitalization associated with urinary tract infection.

HD is more effective than SD in protecting against hospitalizations due to pneumonia or influenza and against any hospitalization in the VHA population during the 2011-12 influenza season.

Health care spending has grown rapidly over the past several decades. However, whether this growth rate may be considered too high depends on the value of the increased length and quality of life due to medical treatment as well as on the increased spending. One way to capture the tradeoff between benefits and costs is with the framework of price measurement. Previous literature has constructed quality-adjusted price indexes for different medical conditions such as acute myocardial infarctions (Cutler et al. 1998) and major depression (Berndt et al. 2002). In general, these studies found that the increased benefits of the technological advances in medical care greatly surpassed the increased spending and therefore that price growth in medical care when measured this way was negative. There is no consensus, however, on what method to use for quality adjustment in medical price indexes. In this paper, we explore the different methods both theoretically and empirically with a view toward exploring how the methods could be implemented in practice. The theoretical model shows that generally the different methods produce the same results when the increase in monetized benefits is approximately equal to the increase in monetized costs. To explore the methods empirically, we construct quality-adjusted price indexes for three acute medical conditions (AMI, heart failure, and pneumonia) for the period 2001-2014 using data for Medicare fee-for-service (FFS) beneficiaries. We measure spending with total spending per patient around the hospitalization, with varying windows and we isolate the contribution of medical care to improvements in outcomes by measuring short-term mortality following the hospitalization. For both our spending measures and outcome measures, we risk-adjust based on the characteristics of the patient. We find spending per patient rose from 2001-2010 and stayed level from 2010 to 2014 while mortality exhibited the reverse pattern. We also find that quality adjustment has a large effect on price growth. As measured by a benchmark cost-of-living index (COLI) where the benefits are measured by the utility value of increases in life expectancy, the average prices for these conditions decline steeply from 2001 through 2010. Following 2010, the declines in quality-adjusted prices level off, at the same time that we observe spending per patient holding flat. In other words, the quality-adjusted prices and the trends in spending per patient appear to move in different directions, highlighting the important role of quality adjustment. When other quality adjustment methods are applied, we find a wide dispersion in the growth rates of different quality-adjusted price indexes. According to our theoretical model, this dispersion results from the increases in benefits exceeding the increases in spending for these conditions; only the benchmark COLI fully captures the value of the increases in benefits to patients.

Human rabies, mostly caused by the canine rabies virus variant, inflicts a heavy burden with approximately 59,000 annual deaths globally. While most countries in the Americas and Europe have eliminated rabies from their dog populations, 122 countries are still dealing with endemic canine rabies. The World Health Organization (WHO) recommends that these countries vaccinate 70% of their dogs each year for seven years in order to end transmission permanently. However, it is unknown what level of vaccination is necessary after these seven years to prevent re-establishing dog rabies from an importation or host-shift event. We consider several different scenarios whereby canine rabies is introduced into a rabies-free population. Dog-to-dog and dog-to-human rabies transmission is defined using an SEIR (susceptible-exposed-infectious-recovered) compartmental model. For each reintroduction scenario, we compare cost-effectiveness of various vaccination strategies including discontinuation (baseline), low-to-high coverage, and vaccination maintained only along the border with the enzootic neighbor. The economic costs and benefits of canine vaccine programs is measured in terms of the number of rabid dogs averted, number of human rabies deaths averted and the average cost per rabies-related human deaths averted over a 20-year horizon. Demonstrating the cost-effectiveness of various vaccination strategies to maintain rabies free status could help countries decide how to implement policies for ongoing vaccination post-elimination.

Atypical antipsychotic drugs are widely prescribed for major depressive disorder as a second-line therapy to augment antidepressant use despite little evidence regarding their comparative cost effectiveness. Atypical antipsychotics such as aripiprazole had much higher prices when first introduced, but recent generic equivalents have substantially narrowed price differentials between them and older antidepressant drugs. We used data from a randomized clinical trial to compare the cost effectiveness of augmenting standard antidepressant therapy with aripiprazole compared to another augmentation agent, bupropion, and to switching to bupropion over a 12-week acute

The cost-effectiveness analysis (CEA) was conducted as part of the Veterans Affairs (VA) augmentation and switching treatments for improving depression outcomes (VAST-D) trial in which 35 participating VA medical centers enrolled 1,522 patients who had failed prior pharmacotherapy. Remission from depression and quality-adjusted life years were estimated from trial data collected at baseline and 12 weeks after randomization. Health care costs from 2015 were obtained from VA administrative data. We compared the cost effectiveness of the 3 strategies by estimating the costs per remission with 12 weeks as the time horizon and the health care sector as the primary perspective. We calculated the incremental cost-effectiveness ratio (ICER) using the difference in costs between each treatment strategy versus the other divided by the differences in remission rates at 12 weeks. We calculated 95% confidence intervals

The mean age of participants enrolled in the trial was 54 years, and participants were predominantly male. The rate of remission at 12 weeks was highest for the aripiprazole augmentation arm (29%), followed by bupropion augmentation (27%), and lowest for bupropion monotherapy (22%). Mean mental health care costs which included the costs of outpatient mental health care visits, inpatient psychiatric stays, and the study drugs did not differ significantly between the groups. The incremental cost effectiveness ratio (ICER) comparing costs per remission was lowest for the bupropion augmentation group relative to the bupropion monotherapy group at -$640/remission. The ICER for the aripiprazole augmentation versus bupropion monotherapy group was $1,074/remission (95% CI =47-5022) with 97.9% of the observations in upper-right quadrant, indicating greater costs and benefits associated with aripiprazole. The ICER for aripiprazole augmentation versus bupropion augmentation was $5,094/remission (95% CI =-34027-32774) with 75.6% of the observations in the upper right quadrant. We did not find any significant differences in mental health care costs, quality-adjusted life years, employment, and other work and social adjustment outcomes between treatment groups during follow up.

In treatment of non-responsive depression, augmentation with either aripiprazole or bupropion increased costs as compared to switching to a new antidepressant, but these costs were justified by higher remission indicating


Andreas Kuznik [email protected]

Elena Lucchese [email protected]

Samuel Savitz [email protected]

Wet age-related macular degeneration (wAMD) is the leading cause of visual impairment and blindness in the United States (US). Intravitreal aflibercept injection (IAI) and ranibizumab (RBZ) are the two most commonly used anti-vascular endothelial growth factor treatments approved by the US Food and Drug Administration for the treatment of wAMD. Randomized clinical trials in patients with wAMD have shown similar efficacy, safety, and tolerability between IAI administered every 8 weeks after 3 initial monthly injections, IAI monthly, and monthly RBZ. Therefore, associated costs and cost-effectiveness comparisons are key factors to determine which treatment represents a more rational investment of scarce health care resources. This assessment may help address the increasing cost of prescription drugs in the US, a source of concern for patients, prescribers, payers, and policy makers.

To assess the cost-effectiveness of IAI (2mg every 8 weeks after 3 initial monthly doses [2q8]) vs. RBZ (intravitreal, 0.5mg monthly) and RBZ (intravitreal, 0.5mg as needed [PRN]) for wAMD from a US payer perspective.

A Markov cohort model was developed to estimate the lifetime quality-adjusted life years (QALYs) and costs of treating patients with wAMD with IAI 2q8, RBZ dosed monthly, and RBZ dosed PRN. The model considered changes in best-corrected visual acuity (BCVA) in the affected and fellow eye over time, and the impact of blindness on mortality. Efficacy data for IAI 2q8 and RBZ dosed monthly, for the first 52 weeks, were from the VIEW 1 and VIEW 2 studies, and from the CATT trial for RBZ dosed PRN. Patients were assumed to be treated for a maximum of 2 years, with BCVA remaining stable between the first and second year in patients remaining on treatment. Natural progression of BCVA loss was applied in patients discontinuing treatment. Background mortality, from US life tables, was adjusted with published hazard ratios for wAMD and blindness. Utilities were from published literature and based on BCVA in the best-seeing eye.

Over a lifetime, IAI 2q8 provided equal health benefits vs. RBZ dosed monthly (5.44 QALYs) at a lower total cost ($33,795 versus $48,031) as a result of fewer injections. IAI yielded incremental health benefits vs. RBZ dosed PRN (5.44 versus 5.40 QALYs) at a higher cost ($33,795 versus $33,652), with an incremental cost per QALY gained of $2,715. Results were sensitive to variations in drug acquisition costs and number of injections of both drugs, and the baseline age of the

I document that a one-minute increase in ambulance response time (RT) for cardiovascular emergency calls increases patient’s morbidity by 2 percentage points and out-of-hospital mortality by 1 percentage point. Patient’s morbidity is recorded at the ambulance arrival on the scene, before performing any medical treatment: in this the compounding effect given specific characteristics of the system in analysis is mitigated, and the result may be generalized. Finally, I find no evidence about the existence of a critical threshold for RT. I perform the analysis using administrative data on emergency calls and ambulance missions performed in the Italian region Liguria in two years. I identify the causal effect of

Low health literacy may act as a barrier to appropriate treatment for stable angina. The treatment alternatives have similar outcomes, but prior research suggests that many patients mistakenly believe more interventional treatment

Secondary analysis of fee-for-service (FFS) Medicare beneficiaries (20% random sample) using Parts A, B and D data. The inclusion criteria were: 1) incident diagnosis of stable angina in 2007-2014), 2) twelve months of Medicare FFS prior to diagnosis, 3) six months of Medicare FFS and Part D after diagnosis, 4) did not receive CABG or PCI prior to diagnosis, and 5) no diagnoses for unstable angina or acute myocardial infarction prior to stable angina diagnosis. The three treatment alternatives are: medication only; percutaneous coronary intervention (PCI), and coronary artery bypass grafting (CABG) surgery. Patients were categorized to PCI or CABG if they received the procedure within twelve months of the first observed diagnosis code for angina, coronary artery disease, or chest pain. Otherwise, patients were categorized to ‘medication only’ if they filled at least one claim for a medication used to treat stable angina and did not receive PCI or CABG. The key independent variable was an area-based health literacy measure at the census block group level derived from a validated predictive model using demographic characteristics. This variable was specified as binary with low vs. high health literacy and quartiles based on the national distribution. The relationship between treatment selection and health literacy was evaluated using multinomial logistic regression, and the results are presented as average marginal effects. Control variables included demographics (age, sex, race/ethnicity), programmatic characteristics (dual eligibility status, receipt of low-income subsidy), Charlson comorbidity index, year of treatment, and state fixed effects.

The sample had 17,215 beneficiaries, of which 1,922 and 15,293 were in the low and high health literacy categories, respectively. Patients in the low health literacy category were more likely to be dual-eligible, receive the low-income subsidy, and be a racial or ethnic minority. For the binary specification, patients with low health literacy were more likely to receive medication only (2.8 percentage points [0.1 to 5.5% points]) and less likely to receive CABG (-2.9 percentage points [-4.9 to -0.9 percentage points]). For the quartile specification, the probability of receiving PCI significantly decreased with lower literacy for each quartile, while the probability of medication only significantly increased.

Contrary to the hypothesis, lower health literacy was associated with less use of invasive treatment for stable angina. Other factors such as patient preferences, physician preferences, and symptom severity may explain the use of less invasive treatment alternatives. The results may also reflect access to care since the less intensive treatment of medication only was also relatively higher among groups that have been historically underserved (low-income and

Yinong Young-Xu [email protected]

Anne Hall [email protected]

Seonghye Jeon [email protected]

In randomized trials of adults aged 65 years or older, a high-dose influenza vaccine (HD) was more efficacious in preventing pneumonia admissions compared with the standard-dose vaccine (SD). An observational study in the Veterans Health Administration (VHA) population, however, did not show any additional benefit of HD over SD for this outcome. We reassessed the relative vaccine effectiveness (rVE) of HD vs. SD in the VHA population using the instrumental

The study included Veterans aged 65 or older who received an influenza vaccine in the 2011-12 influenza season at the VHA. Electronic medical records were used to capture the treatment, outcome (hospitalizations due to pneumonia or influenza and all-cause hospitalizations) and baseline patient characteristics. The IV was the proportion of HD recipients at each VHA facility in the prior season, defined as the number of HD recipients divided by the total number of both HD and SD recipients at each VHA facility. The instrument was defined using the HD proportion from the prior season (2010-11), because it is a strong predictor of receipt of HD in 2011-2012 but less likely to be related to the health or treatment of patients in the 2011-2012 season. Because the dependent variable was a count variable (number of hospitalizations), we performed IV Poisson regression analysis, adjusting for age, sex, race, socioeconomic status as well as baseline comorbidities. We performed tests of the Correlation and the Exclusion Restriction criteria for IV analysis. Finally, we conducted a falsification test, using hospitalization associated with urinary tract infection as the outcome;

Our study population included 678,386 VHA patients who were vaccinated at 1,027 VHA facilities (medical centers and community-based outpatient clinics) in the 2011-2012 season. Of these, 20,880 patients (3%) received HD, while 657,506 patients (97%) received SD. The IV adjusted rVE estimate of HD was 47% (95% CI, 27%–62%) against influenza- or pneumonia-associated hospitalization, and 13% (95% CI, 4%-21%) against all-cause hospitalization. We found that our instrument was strongly predictive of receipt of HD. An F statistic greater than 10 is generally considered sufficient. The F statistic for our sample was 916,371. Every increment of 10% in a facility’s prior HD proportion was associated with doubling the likelihood of a patient being provided with HD. In addition, the instrument was not correlated with other factors that would affect the medical outcomes of patients, such as facility quality, facility type, regions, rurality or the overall health status of patients treated by the facility. The falsification test was confirmative as the IV adjusted rVE estimate of HD was 4% (95% CI, -22%–25%) against hospitalization associated with urinary tract infection.

Health care spending has grown rapidly over the past several decades. However, whether this growth rate may be considered too high depends on the value of the increased length and quality of life due to medical treatment as well as on the increased spending. One way to capture the tradeoff between benefits and costs is with the framework of price measurement. Previous literature has constructed quality-adjusted price indexes for different medical conditions such as acute myocardial infarctions (Cutler et al. 1998) and major depression (Berndt et al. 2002). In general, these studies found that the increased benefits of the technological advances in medical care greatly surpassed the increased spending and therefore that price growth in medical care when measured this way was negative. There is no consensus, however, on what method to use for quality adjustment in medical price indexes. In this paper, we explore the different methods both theoretically and empirically with a view toward exploring how the methods could be implemented in practice. The theoretical model shows that generally the different methods produce the same results when the increase in monetized benefits is approximately equal to the increase in monetized costs. To explore the methods empirically, we construct quality-adjusted price indexes for three acute medical conditions (AMI, heart failure, and pneumonia) for the period 2001-2014 using data for Medicare fee-for-service (FFS) beneficiaries. We measure spending with total spending per patient around the hospitalization, with varying windows and we isolate the contribution of medical care to improvements in outcomes by measuring short-term mortality following the hospitalization. For both our spending measures and outcome measures, we risk-adjust based on the characteristics of the patient. We find spending per patient rose from 2001-2010 and stayed level from 2010 to 2014 while mortality exhibited the reverse pattern. We also find that quality adjustment has a large effect on price growth. As measured by a benchmark cost-of-living index (COLI) where the benefits are measured by the utility value of increases in life expectancy, the average prices for these conditions decline steeply from 2001 through 2010. Following 2010, the declines in quality-adjusted prices level off, at the same time that we observe spending per patient holding flat. In other words, the quality-adjusted prices and the trends in spending per patient appear to move in different directions, highlighting the important role of quality adjustment. When other quality adjustment methods are applied, we find a wide dispersion in the growth rates of different quality-adjusted price indexes. According to our theoretical model, this dispersion results from the increases in benefits exceeding

Human rabies, mostly caused by the canine rabies virus variant, inflicts a heavy burden with approximately 59,000 annual deaths globally. While most countries in the Americas and Europe have eliminated rabies from their dog populations, 122 countries are still dealing with endemic canine rabies. The World Health Organization (WHO) recommends that these countries vaccinate 70% of their dogs each year for seven years in order to end transmission

We consider several different scenarios whereby canine rabies is introduced into a rabies-free population. Dog-to-dog and dog-to-human rabies transmission is defined using an SEIR (susceptible-exposed-infectious-recovered) compartmental model. For each reintroduction scenario, we compare cost-effectiveness of various vaccination strategies including discontinuation (baseline), low-to-high coverage, and vaccination maintained only along the border with the enzootic neighbor. The economic costs and benefits of canine vaccine programs is measured in terms of the number of rabid dogs averted, number of human rabies deaths averted and the average cost per rabies-related human deaths averted over a 20-year horizon. Demonstrating the cost-effectiveness of various vaccination strategies to maintain rabies free status could help countries decide how to implement policies for ongoing vaccination post-elimination.

Jean Yoon [email protected]

Atypical antipsychotic drugs are widely prescribed for major depressive disorder as a second-line therapy to augment antidepressant use despite little evidence regarding their comparative cost effectiveness. Atypical antipsychotics such as aripiprazole had much higher prices when first introduced, but recent generic equivalents have substantially narrowed price differentials between them and older antidepressant drugs. We used data from a randomized clinical trial to compare the cost effectiveness of augmenting standard antidepressant therapy with aripiprazole compared to another augmentation agent, bupropion, and to switching to bupropion over a 12-week acute

The cost-effectiveness analysis (CEA) was conducted as part of the Veterans Affairs (VA) augmentation and switching treatments for improving depression outcomes (VAST-D) trial in which 35 participating VA medical centers enrolled 1,522 patients who had failed prior pharmacotherapy. Remission from depression and quality-adjusted life years were estimated from trial data collected at baseline and 12 weeks after randomization. Health care costs from 2015 were obtained from VA administrative data. We compared the cost effectiveness of the 3 strategies by estimating the costs per remission with 12 weeks as the time horizon and the health care sector as the primary perspective. We calculated the incremental cost-effectiveness ratio (ICER) using the difference in costs between each treatment strategy versus the other divided by the differences in remission rates at 12 weeks. We calculated 95% confidence intervals

The mean age of participants enrolled in the trial was 54 years, and participants were predominantly male. The rate of remission at 12 weeks was highest for the aripiprazole augmentation arm (29%), followed by bupropion augmentation (27%), and lowest for bupropion monotherapy (22%). Mean mental health care costs which included the costs of outpatient mental health care visits, inpatient psychiatric stays, and the study drugs did not differ significantly between the groups. The incremental cost effectiveness ratio (ICER) comparing costs per remission was lowest for the bupropion augmentation group relative to the bupropion monotherapy group at -$640/remission. The ICER for the aripiprazole augmentation versus bupropion monotherapy group was $1,074/remission (95% CI =47-5022) with 97.9% of the observations in upper-right quadrant, indicating greater costs and benefits associated with aripiprazole. The ICER for aripiprazole augmentation versus bupropion augmentation was $5,094/remission (95% CI =-34027-32774) with 75.6% of the observations in the upper right quadrant. We did not find any significant differences in mental health care

In treatment of non-responsive depression, augmentation with either aripiprazole or bupropion increased costs as compared to switching to a new antidepressant, but these costs were justified by higher remission indicating


Regeneron Pharmaceuticals, Inc Complete

Department of Economics Complete

The University of North Carolina at Chapel Hill Complete

Andrea Gibson; Clifford Cele; Hector Toro-Diaz; Tereza Lanitis; Luis Hernandez

Justin Trogdon; Sally Stearns; William Jones; Stacey Bailey; Stacie Dusetzina

Veterans Health Administration Complete

US Treasury Department Abe Dunn Complete

Centers for Disease Control and Prevention Complete

Julia Thornton Snider; Ayman Chit; Jason Lee; Edward Thommes; Salaheddin Mahmud

Martin Meltzer; Emily Kahn; Julie Cleaton; Ryan Wallace

Palo Alto VA Complete


Demand for and Effect of Health Insurance




The Short-Run Effects of Employer-Sponsored Health Insurance on the Labor Market Supply of Ill Workers .

Effects of Copays on Non-Urgent ED Use in a Medicaid Population

Impact of ACA's dependent coverage mandate on health insurance and labor market outcomes among young adults: evidence from regression discontinuity design

Doc, do I Really Need it?: Reductions in Inpatient Services for the Uninsured in Maryland





Does Higher Cost-sharing of Employer-provided Health Insurance Plans Impact Workers’ Compensation Claiming?

Contemporaneous and Long-term Effects of Children’s Public Health Insurance Expansions on SSI

Would Medicare Expansion Crowd Out Prevention? Evidence from the Marginally Mortal

The Trump effect: post-inauguration changes in Marketplace enrollment




The Spillover Effects of Child Health Insurance on Non Beneficiaries within the Household: Evidence from Vietnam

Consumer Choice and Learning in Private Insurance Markets: Evidence from the ACA Marketplaces

How much do demographic, insurance market, and policy factors affect exchange enrollment through healthcare.gov?





Utilization of Health Insurance and Worker's Compensation Cost-Shifting

The Impact of the Minimum Wage on Health Insurance: Evidence from Agricultural Workers

Estimating the Demand for Individual Health Insurance

How much does traditional insurance prevent negative financial outcomes? Evidence from Medicare and the Affordable Care Act



Effects of the Affordable Care Act and Medicaid Expansion on Incident End-Stage Renal Disease Patients

Impact of Health Insurance Coverage and Usual Source of Care on Adult Cancer Patients’ Experiences with Care




The Effect of Health Insurance Coverage on Access to Care for Community Health Center Patients

Employer Mandates and Chronic Disease: Assessing the Impact of the San Francisco Employer Mandate on Cancer Outcomes

Do rural areas have worse access to hospital care? Examining regional variation for people with diabetes


A Claim a Day Keeps the Doctor Away? Premium Refunds and Forward Looking Behaviour in the German Private Health Insurance Market

Abstract

Using NLSY79 data, this paper tests whether the source of health insurance creates incentives for newly-diagnosed workers to remain sufficiently employed to maintain access to health insurance coverage. I compare labor supply responses to new diagnoses of workers dependent on their own employment for health insurance with the responses of workers who are dependent on their spouse's employer for health insurance coverage. I use the latter as the comparison group instead of workers with no health insurance because these workers with coverage through their own employer and workers with coverage through their spouse's employer are more likely to be homogeneous in terms of observable and unobservable attributes. I focus on the following labor supply changes made 0, 6, 12 and 24 months after the diagnosis: changes in hours worked, the probability of remaining employed and the probability of moving from full-time to part-time. I find that workers who depend on their own job for health insurance are 1.5-5.5 percentage points more likely to remain employed and for those employed, are 1.3-5.4 percentage points less likely to reduce their labor hours and are 2.1-6.1 percentage points more likely to remain full-time workers. This paper addresses one of the limitations of the previous studies. The scope of earlier studies was limited to either a single state, a single disease or a particular gender or age group which limits the generalizability of conclusions to other settings. This paper addresses such limitation by using the national NLSY data which reports information on diagnosis of various illnesses. A large percentage of the diagnoses reported in NLSY were hypertension, diabetes, arthritis, and mental health problems, which compared to previous studies is more reflective of the national distribution of chronic health conditions faced by residents in the US. These conditions were not included in any of the previous studies done on this topic. Since most of the diseases in the NLSY are not life-threatening compared diseases investigated in previous studies, this paper contributes analysis for the effects of ESI for workers with a diagnosis of less severe conditions.

Background: Research has shown that cost-sharing affects emergency department (ED) use, with lower enrollee cost-sharing associated with increased use. There has been specific concern about ED use in the Medicaid population, where gaining coverage has been associated with increased ED use, including for conditions better treated in a primary care setting. To promote the use of primary care and encourage appropriate use of the ED, Michigan’s 2014 Medicaid expansion program does not include copays for preventive services but does include cost-sharing for non-urgent ED visits. Enrollees with incomes at less than 100 percent of the federal poverty level incurred $3 copayment for ED use; eligible enrollees with higher incomes incurred an $8 copayment for ED use. Copayments are waived for ED use for urgent conditions and are assessed for those considered non-urgent. It is not known whether copayments assessed on this population will discourage ED visits for non-urgent conditions.

Research question: Does a copay for non-urgent ED use discourage Healthy Michigan Plan enrollees from using the ED for non-urgent visits?

Methods: We use Medicaid administrative claims data for enrollees in Michigan’s Medicaid expansion program who had at least 18 months of continuous enrollment, and who enrolled in the program between its inception in April 2014 and March 2015. We classify ED visits into low-medium-high severity using administrative codes, and separately into copay-eligible/copay-exempt using state algorithms. Using a time series design, we compare ED utilization for urgent visits and non-urgent visits between the first six months of an individual’s enrollment, in which no copays were assessed, with subsequent utilization. We examine use of the ED for low, medium and high severity visits. We also analyze total spending in the ED as well as spending for each type of visit. We control for age, gender, income level and region of the state.

Results: This project is part of a required independent evaluation related to Michigan’s 1115 waiver for the Medicaid expansion demonstration. Per terms of the evaluation agreement, results are reviewed by officials at the Michigan Department of Health and Human Services before external release.

Conclusions: Conclusions will be based on data and available after results are reviewed by the state in early 2018.

This paper identifies the effect of the Affordable Care Act's (ACA’s) dependent coverage mandate on health insurance coverage, health insurance holding and labor market outcomes among young adults, by exploiting an exogenous variability in losing an additional access to health insurance coverage at age 26. To remedy the policy endogeneity problem in the literature, we exploit the discrete jump in health insurance coverage and labor market outcomes at age 26 using a fuzzy regression discontinuity design. Using alternative parametric and non-parametric models, we find that ACA's dependent coverage mandate is associated with about 1, 3-5 and 5-9 percentage points decrease in public insurance coverage, private insurance coverage and coverage from someone living outside RU, respectively, when young adults turn to 26. We also find that ACA' dependent coverage mandate decreases coverage from employment or union by about 0.7-1.5 percentage points. These results are quite robust to different model specifications. We also find that the negative effects of aging out at 26 can be offset by significant increase in employment union, nongroup, other group, and private insurance holdings. ACA's dependent coverage mandate also has spillover effects on labor market outcomes among young adults. Our results imply that it is associated with a decrease in the probability of employed by about 3.9 percentage points and a decrease in the probability of self-employment by about 1.8-2.6 percentage points when young adults turn to 26. We also find an increase in probability of engaging in temporary job when people are just older than 26. We do not find any significant change in hourly wage, weekly hours or job mobility at age 26.Our results imply that “job lock" is not a significant and major problem in labor market of young adults but there might exist some moderate “entrepreneurship lock" among young adults. These results, however, are not robust to different bandwidths and model specifications.

Uninsured individuals receive fewer health care services for at least three reasons: higher prices, responsibility for the entire bill, and potential provider reductions for concern of nonpayment. This study isolates differences in service levels between insured and uninsured individuals where uninsured individuals pay the entire bill without a contribution from an insurance company, but otherwise face the same prices. I capitalize on Maryland's highly regulated health care system, where prices are set by the state, are uniform across all patients, and hospitals are compensated or free care and bad debt, to isolate the difference in quantity demanded by the uninsured. I use a unique feature of the data, multiple readmissions for the same patients who gain or lose insurance between visits, to isolate the reductions in quantity demanded when individuals are faced with paying the full price without an insurance company contribution. While the Oregon studies compare Medicaid individuals and their low-income uninsured counterparts, this paper considers income variation among the uninsured, and quantifies the difference in demand in an environment with uniform prices. A Blinder-Oaxaca decomposition estimates uninsured individuals receive 6% fewer services after accounting for differences in patient, illness, and hospital characteristics than when these same individuals are insured. This difference in service level is larger for patients residing in low income zip codes and smaller in wealthy zip codes. This suggests that income is a substantial constraint for uninsured patients, and as that constraint relaxes, more services are demanded. For illnesses with a high risk of mortality, there was no difference in service provision for insured and uninsured individuals. The difference in service provision is attributable to illnesses with a low risk of mortality. While this paper analyses Maryland, it provides insight into demand for insured individuals with high deductible plans country wide. Prior to meeting their deductible, these insured patients face similar conditions to uninsured patients in Maryland— they have access to negotiated rates but are solely paying the bill.

In the recent decade, cost-sharing of health insurance plans increased substantially. The share of insured workers in plans with a general annual deductible has increased from 55% in 2006 to 81% in 2015, as have the average deductible amounts for covered workers in plans with deductibles from $584 in 2006 to $1,318 in 2015 (as reported by the Kaiser Family Foundation). In this analysis, we investigate the responsiveness of injured workers to changes in cost-sharing burden of their health insurance plans when deciding whether to file a workers’ compensation (WC) claim or not. Previous literature offers abundant evidence that a sizable proportion of injured workers with work-related injuries does not file for WC coverage. These substantial non participation rates suggest that WC filing is not free and associated with the following costs: employer discouragement of filing for WC benefits, stigma and loss of bonuses/overtime pay, reduction in income, upfront costs of the filing process including medical costs with uncertain prospects for reimbursement. However, in the environment of growing deductibles and other out-of-pocket payments, workers may find zero out-of-pocket coverage of medical expenses, provided by WC more appealing and may be more willing to bear the filing costs associated WC claiming. In our analysis, we explore the effect of higher cost-sharing burden on decision to file for WC coverage, while controlling for various injury, worker and employment characteristics as well as firm-specific effects. We were able to isolate the worker’s financial incentives to file for WC from the firm effects, by using variation in the annual remaining cost-sharing burden at the time of the injury across workers employed in the same organization. We find positive and statistically significant effect of higher cost-sharing on decision to file for WC coverage. This analysis relies on workers’ compensation and group health medical data coming from a large commercial national database, Truven MarketScan® for years between 2008 and 2014. It includes individuals employed by mostly large employers and insured or administered by one of approximately 100 group health plans. It also provides a wealth of information on the benefit design of the individual group health plans, including information on in and out of network deductible amounts, co-payments, coinsurance amounts, and out of pocket maximum payments.

This study explores the interplay between two important public programs for vulnerable children: Medicaid and the Supplemental Security Income (SSI) program. Medicaid eligibility for children expanded in the late 1990s and early 2000s, primarily due to the creation of the Children’s Health Insurance Program (CHIP). We employ a generalized difference-in-differences design that takes advantage of the expansion of Medicaid and CHIP within states over time to isolate plausibly causal impacts of public health insurance eligibility expansions on SSI outcomes. The key data sources for the study are the Current Population Survey and Social Security Administration’s Supplemental Security Record files. On average, increases in Medicaid eligibility did not affect contemporaneous youth SSI applications or awards. However, in states where SSI recipients did not automatically receive Medicaid, expansions in public health insurance coverage led to a significant decrease in both SSI applications and awards. These results suggest that the newly available Medicaid/CHIP coverage – noteworthy for the relative ease of its application process compared with SSI – was an attractive potential substitute for SSI, especially among families that may have valued SSI primarily for the associated Medicaid benefit. In the long-term, we find that increased Medicaid eligibility during childhood reduces young adult SSI applications, consistent with recent findings that Medicaid coverage in youth improves adult health and economic outcomes.

This paper uses evidence on how treatment effects vary across individuals to identify the prospective effect of expanding Medicare coverage to earlier ages. First, I document that in 1998, it appears that the formal insurance coverage provided by Medicare has a cross-price moral hazard effect on diabetics' usage of insulin - the proportion of diabetics reporting that they use insulin drops discontinuously by 17.7% when they turn 65 and become eligible for Medicare. This provides new evidence of formal insurance crowding out self-insurance via preventive medicine - the first is a substitute for the second in insuring individuals against the risk of incurring high medical costs in the future. Second, I identify the Marginal Treatment Effect for this subpopulation - in this context, the effect of coverage for an individual just at the margin of surviving to age 65. This requires the use of variation that is, unusually, observed by the econometrician but not the individuals themselves. I argue that in a limited number of cases, there are future medical events that cause parametric shifts of individual-specific survival curves that are not predictable from individuals' point of view. Third, the first stage of estimation requires a new approach to using future information to make inferences about the ex ante probability of events, based on information revealed regarding latent processes. Fourth, I find that, absent Part D, making Medicare coverage for treatment available at earlier ages would crowd out prevention to a lesser extent than the status quo. This is due to self-insurance via prevention being crowded out the most among those with the weakest private expectation that they will survive to get Medicare coverage. At earlier ages, by contrast, there is a higher overall probability of survival - hence eligibility, under an expanded regime - across all individuals. The results suggest that Medicare increased health inequality along some dimensions in the United States before the introduction of coverage for preventive medicine such as insulin with Medicare Part D.

On January 20, 2017, President Donald J. Trump penned his first executive order aiming to “minimiz[e] the economic burden” of the Affordable Care Act, signaling his intention to make good on promises to repeal and replace. The health insurance exchanges (Marketplace) allowed more than 12 million Americans to enroll in guaranteed issue community-rated health plans in 2016 and 2017. These plans are subsidized through advanced premium tax credits for those between 100% and 400% of the federal poverty level. Our analysis uses data from the open enrollment periods for these two years obtained from the federally-facilitated and state-based marketplaces to estimate the enrollment loss attributable to the transition of power and this first executive order of the new administration. We first estimated this effect descriptively with publicly available state-level Marketplace enrollment totals prior to inauguration and at the end of open enrollment using the difference between actual and expected incremental enrollment to estimate state and national enrollment declines in 2017 relative to the 2016 trend. This approach does not rely on the counts of incremental enrollment rather the proportion of overall enrollment that is expected to occur during the final two weeks so as to not bias our estimates due to changes in the size of the Marketplace eligible population from year to year. We estimate an approximately 52% incremental loss in enrollment, corresponding to approximately 365,000 fewer enrollees, during the final two weeks of open enrollment for 2017 compared to 2016 for those states enrolling through healthcare.gov. In contrast, we only observed an 18.6% decrease in enrollment during the same time period in three state-based marketplaces. We also used a difference-in-differences model with Marketplace applications data for 1,476 counties in 37 states obtained from the Centers for Medicare and Medicaid Services (CMS) through a Freedom of Information Act request. We accounted for state and county-level characteristics potentially associated with enrollment volume, including state Medicaid expansion status (CMS), 2010 county population and percentage of county population living in urban areas (Census), number of Marketplace insurance carriers and the Silver gap (difference between benchmark Silver plan and the least expensive Silver plan for a single 40-year old) by county and year (CMS), and the percentage of votes received by President Trump in the 2016 Presidential election in each county (Townhall). We estimate a population-weighted decline of approximately 720 applications per county-week during the final two weeks of the 2017 open enrollment period relative to 2016 (b=–720.5, p<.001), corresponding to a 32% decline in applications submitted. The perceived lack of political support for the law by the incoming administration had an immediate and significant effect on Marketplace enrollment nationwide.

The last three decades have seen a surge in the number of government funded targeted health insurance programs introduced in developing countries. However little is known about the indirect impact of these programs on household members who are not eligible. Using variation introduced by a health insurance program targeted to children below the age of six in Vietnam, I compare changes in expenditures between households who receive insurance to those who do not. I find that beneficiary households increase spending on health and food and reduce expenditures on education. There are two explanations for the changes. The first is an increase in employment hours for adults in the households, especially of women in the households. The second is a reallocation between education and health related expenditures for children who are eligible for insurance at the expense of children who are not.

We examine consumer plan choice and learning in 2015 and 2016 in the Federally-facilitated Marketplaces (FFM) for health insurance established by the Affordable Care Act (ACA). The FFM offers a useful context for studying choice in private insurance markets. Plans offered in the FFM are displayed in an online “exchange” and are required to meet standards regarding coverage, premiums, benefits, and cost-sharing. These requirements as well as the design of the exchange is intended to facilitate consumer shopping and there is evidence that consumers actively shop and switch plans in the FFM at higher rates than in the market for employer-sponsored insurance. Understanding the dynamics of consumer choice in this setting is important given that the Marketplaces are still relatively new markets for consumers and insurers and that the regulations described above were intended to standardize and improve the options for consumers getting insurance through the non-group market. A deeper understanding of consumer decision-making is fundamental to improving the design of insurance markets as well as ensuring that the individual insurance market is attractive to insurers.

A combination of public data on plans available at the county level and associated benefit design features and administrative individual-level panel data on FFM enrollment allows us to identify each enrollee’s choice set and their chosen plan as well as to see how their choices evolved over time. Further, these data allow us to calculate the actual premiums, deductibles, and max OOP levels that consumers faced, taking into account cost-sharing reductions and other subsidies. We characterize enrollees’ choice sets and estimate discrete choice models of individual-level plan choice that account for plan characteristics and interactions with individual characteristics. These choice models allow us to assess the value that consumers place on four benefit design features: premium, deductible, maximum out-of-pocket (OOP), and size of the physician network. We use our estimates to calculate individuals’ willingness-to-pay (WTP) for benefits such as lower deductibles and broader provider networks and to variations in WTP across different populations. Finally, we examine choices by consumer year of entry to assess how WTP changes over time, which may reflect consumer learning.

We find that consumers valued each plan benefit design feature when making their choice and were willing to pay substantial amounts to lower deductibles and max OOP levels and to have access to a broader physician network. Specifically, 2016 Marketplace enrollees were willing to pay $257 annually to reduce their deductible by $1,000, $108 to reduce their max OOP by $1,000, and $151 annually to increase the network penetration of their plan by 25 percentage points. Willingness-to-pay for all features increased with age and was generally higher for women than men; WTP for greater network breadth also increased with income. Further, WTP for bigger networks increased monotonically over time for individuals who enrolled in Marketplace plans for multiple years, suggesting that consumers learned over time. These results are also consistent with the introduction of provider search tools and other network information onto the Marketplace website in 2015 and 2016.

How much do demographic, insurance market, and policy factors affect exchange enrollment through healthcare.gov? The exchange market for health insurance is central to many legislative proposals that the Congressional Budget Office analyzes. We can more accurately predict the budgetary and coverage effects of legislative changes if we better understand how much exchange take-up varies with demographic, insurance market, and policy factors. Estimates from the literature are from survey data or aggregated administrative data on plan selections during open enrollment. We are able to improve on this literature using person-level data on effectuated enrollment through healthcare.gov in 2015 and 2016 from the Center for Medicare and Medicaid Services (CMS). These data allow us to accurately measure exchange take-up and control for variation in the demographic composition of local geographic areas. By reducing measurement error and increasing sample size, our analysis will yield more precise estimators. Our analysis is based on logistic regression models of the take-up of exchange coverage by potential enrollees. Our sample consists of 779 local geographic areas in 2015 and 782 in 2016 stratified by gender, age (20-34, 35-44, 45-54, and 55-64), and income (100/138-250 FPL and 250-400 FPL) resulting in a sample of 24,976 observations. We use CMS’s person level data on effectuated enrollment through healthcare.gov to measure the full-year equivalent number of ‑exchange enrollees. To calculate take-up rates, we divide these totals by estimates of the population of potential enrollees based on the Census Bureau’s Population Estimates Program and the 2011–2015 American Community Survey 5-year public use microdata sample. We define a potential enrollee as a person who is eligible for the premium tax credit, who is not eligible for Medicaid or Medicare, and who does not have employer-sponsored insurance or TRICARE. Additional datasets from the Census Bureau, CMS, the Agency for Healthcare Research and Quality, and the Commonwealth Fund are used to measure important control variables for demographic, insurance market, and policy factors. Our preliminary analysis suggests that many factors affect the odds of exchange enrollment. The odds of enrollment are predicted to be much higher for women than men and to increase with age. The odds of enrollment are predicted to decrease with the share of potential enrollees who are exempt from the individual mandate, who do not speak English very well, who are Hispanic, and who are native to the United States. The size of the non-group market in 2013, the exchange participation of insurers with high market share in the non-group market in 2013, and the number of primary care physicians per capita in a region all increase the predicted odds of enrollment. Potential enrollees with eligibility for high subsidies have much higher predicted odds of enrollment than those with lower subsidies. State restrictions on enrollment assistance substantially lower the predicted odds of enrollment. Finally, several factors that we expected to drive exchange take-up, including reference premiums, insurer competition, and the share of the population living in a rural area did not.

Health insurance contracts are often structured with different forms of cost sharing in order to provide consumption smoothing across different health outcomes while limiting moral hazard. These components of health insurance may influence an individual’s behavior in multiple ways. The research on moral hazard has largely focused on the responsiveness of consumers spending on healthcare to the cost of care. Alternatively, most workers with employer-sponsored health insurance (EHI) also have access to workers’ compensation (WC) for medical coverage of injuries and illnesses which may be related to their work. WC medical coverage has no copayment or deductible associated with coverage, but employees need to prove their injury is work-related and they are often limited in their choice of medical care provider. The costs for the employee associated with using EHI differs across plans in their cost-sharing mechanisms and over the course of a year if a deductible is present. Differences in plans and time of year provide variation in the end-of-year and spot price of medical care and can identify whether or not workers vary their utilization of WC dependent upon the cost of their outside option through EHI. Using MarketScan data containing WC and health insurance claims, the analysis first confirms employee sensitivity to future prices of medical care using a difference-in-difference technique. The differences in plan deductibles and employee enrollment months allows for control of seasonal differences in medical utilization while focusing on variation in the end-of-year price of medical care that comes with differing enrollment months similar to analysis done in Aron-Dine, Einav, Finkelstein, and Cullen (2015). This analysis is then replicating using the WC claims information for the same set of employees. The empirical work finds evidence of cost shifting to WC as employees are less likely to use EHI and more likely to use WC if they enroll in their EHI later in the year and face a higher end-of-year price of medical care using the EHI. Measuring the extent of WC cost-shifting is important for a number of reasons. In addition to providing benefits to employees injured on the job, WC also serves as an incentive for employers who are experience-rated to invest in the safety of their workplaces. Variation in WC claiming due to changing prices of EHI medical care is likely to also impact workplace injuries reported to employers, and therefore measuring the extent of WC cost shifting can better inform occupational safety and health surveillance activities.

The relationship between minimum wages and health insurance coverage among low-wage workers can be complicated. For employers, an increase in the minimum wage that raises the total wage bill may cause changes in health insurance provision; employers may cease to provide health insurance or increase the premiums paid by workers. This would tend to reduce health insurance coverage among low-wage workers. At the same time, for the worker, an increase in the minimum wage may mean that health insurance becomes more affordable, particularly if the cost of the health insurance is split between the worker and the employer, or between the worker and the government. In our research, we quantify the relationship between minimum wages and health insurance coverage among a national sample of agricultural workers, and we examine how that relationship differs between male and female workers. To test the relationship between minimum wages and various types of health insurance coverage among agricultural workers, we use a Probit model in which we control for worker characteristics that are standard in a Mincerian wage equation. Our model also includes state and year dummies, as well as census region by year effects, worker task and crop dummies and legal status. Our sample of full-time crop workers comes from the confidential version of the National Agricultural Workers Survey (NAWS), from the years 2000 through 2014. The NAWS is the only nationally representative survey of demographic, employment, and health insurance characteristics of hired crop workers. Information is obtained directly from farm workers through face-to-face interviews. Our population includes roughly 30,000 crop workers, 82 percent of which are male. This data is unique in its level of detail on the health insurance coverage for crop workers and their families – for each agricultural worker, and each member of the family, the survey reports whether the individual has health insurance and if it was paid for by the worker, the worker’s employer, the spouse, the spouse’s employer, and/or the government. We find that, among male agricultural workers, overall health insurance coverage from all sources combined falls significantly as the minimum wage increases; this effect is largely due to a decline in health insurance paid for by the worker’s employer. Among female agricultural workers, overall health insurance coverage from all sources combined does not change significantly. While we find a decline in health insurance coverage paid for by the female worker’s employer alone, this negative effect is offset by an increase in the incidence of health insurance whose cost is shared between the employer (or the spouse’s employer) and the worker’s family. We also find that an increase in the minimum wage leads to an increase in health insurance coverage for the worker’s children; an effect that is mostly driven by an increase in the government health insurance. While state policies increasing the minimum wage are aimed at improving the well-being of low-wage workers, we find that they may also have some negative consequences on workers’ health insurance coverage, particularly coverage provided by their employers.

We use a novel data set from eHealthInsurance.com to estimate the demand for health insurance in the individual market. We use this data to provide the first estimates of health insurance demand in the 36 states that do not manage state-run health insurance exchanges. We find that median own-price elasticities range between -.59 and -1.55. While uninsurance rates in this market are high---45% on average---we find that diversion towards uninsurance are low. The median diversion ratios range between 10.1% to 18.2%. The low diversion ratios imply that the penalty for uninsurance implemented through the Individual Mandate may not have a large impact on the uninsurance rate. We do a partial equilibrium exercise and find that a repeal of the Individual Mandate would lead to a 1.7 percentage point increase the uninsurance rate in the individual market. This represents solely the immediate demand effect, and does not incorporate any feedback to prices through selection.

We utilize credit report data to study how unpaid medical collections and other financial outcomes are affected by fundamental features of health care. Specifically, we document and discuss the size and age distributions of unpaid medical bills on consumer credit profiles, use Medicare eligibility and the implementation of the Affordable Care Act to identify the causal effect of insurance coverage, and investigate whether geographic variation in healthcare prices is predictive of medical debt. We show that Medicare sharply decreases the dollar value of unpaid medical collections held by consumers. Moreover, as the ACA reduces the extensive margin gain in insurance coverage at age 65, the effect of Medicare on medical debt falls with it. In particular, from 2013 to 2015 the increase in insurance coverage at age 65 falls by 46% and the effect of Medicare on medical collections declines by a similar 53%. However, Medicare has relatively small effects on the number of people who have medical bills sent to collections, instead predominantly reducing the size of bills which go unpaid. In part we reconcile this with the fact that most unpaid bills are relatively small or modestly sized – over half are less than $600. Thus, while Medicare may substantially reduce the size of bills facing the formerly uninsured, that amount may still exceed many consumer’s willingness or ability to pay. In turn, we show that Medicare has muted effects on other outcomes on consumer credit profiles. It is particularly notable that, despite very salient increases in insurance coverage at Medicare eligibility, we observe little change in credit scores at age 65. The patterns we observe are broadly consistent with recent research which argues that the burden of uncompensated care largely falls on health care providers.

Enacted in 2010 and largely implemented in 2014, the Affordable Care Act (ACA) has expanded health insurance coverage to more than 20 million previously uninsured American adults through the Medicaid expansions implemented by 31 states and health insurance exchanges on which individuals can purchase insurance and received subsidies if their incomes are between 100% and 400% of the Federal Poverty Level. While nearly all Americans with end-stage renal disease (ESRD) have insurance after the onset of this condition, those who are uninsured may not have access to effective care before developing ESRD. We assess whether changes in insurance status prior to the onset of ESRD were accompanied by changes in markers of pre-ESRD care in patients receiving their first dialysis treatment pre-enactment of the ACA versus post-enactment of the ACA. We use data from the CMS Medical Evidence Form 2728 which is completed for all patients at incidence of ESRD. This included 511,424 patients from 10/1/11 to 3/31/16. We estimate Difference-in-Differences (D-in-D) models by age/insurance status at ESRD onset [age <65 with no Medicare (treatment group) vs. age ≥ 66 with Medicare (control group)], and by state Medicaid expansion status for age <65 [States expanding Medicaid (treatment group) vs. states not expanding Medicaid (control group)]. Using linear probability models, we determine the ACA’s impact on insurance status and key markers of pre-ESRD care. Models include state and year fixed effects. The assumption of parallel trends in pre-ACA period is supported using plots of monthly data and the results are robust to using narrower age windows. Among patients age <65 not on Medicare, insurance rates increased from 82.4% pre-ACA to 86.7% in 2014, 89.3% in 2015 and 92.1% in 2016. These patients also experienced increases in pre-ESRD nephrology care, use of home dialysis and use of anemia medication; effects on vascular access were equivocal. Patients in states that expanded Medicaid had 3.4 percentage point greater increases in insurance coverage than those in non-expansion states. They also experienced greater increases in pre-ESRD nephrology care and better vascular access, but no change in the use of home dialysis. Overall, these findings demonstrate that several key indicators of pre-ESRD care improved post-ACA implementation, but which indicators improved differed somewhat across the two affected populations: (A) all adults younger than 65 and not on Medicare and (B) adults younger than 65 in Medicaid expansion states. Our findings are consistent with improved access to care that has significant potential to improve clinical outcomes for those with advanced kidney disease in the United States. The magnitudes of the effects are relatively large. Overall, uninsurance fell dramatically during the post-ACA period. About 25-33% of the patients who gained insurance also gained relevant pre-ESRD care, demonstrating the important role of health insurance coverage in access to pre-ESRD care. These are meaningful changes in the context of a clinically and economically vulnerable population. Further research can establish the extent to which the observed improvements in pre-ESRD care affect post-ESRD outcomes such as mortality, hospitalization and access to kidney transplantation.

Background: Lack of access to both health insurance coverage and usual source of care (USC) can be a potential barrier for appropriate use of health care services. For special disease groups, such as adult cancer survivors in the United States, both insurance and USC could be critical for improved access and health outcomes. Yet, no study has evaluated how insurance and USC interact and affect adult cancer survivors’ health care barriers in the U.S.

Objective: We examine whether insurance and USC are associated with adult cancer survivors’ barriers to health care compared with same age and sex individuals without a history of cancer.

Data and Methods: We used the 2012 to 2015 Medical Expenditures Panel Survey (MEPS) data to identify N=4,009 adult cancer survivors, currently aged 18-64 years, who reported ever being diagnosed with cancer. We used one-to-one propensity score matching to identify 4,009 comparison group individuals without a history of cancer. Patients’ barrier was categorized as “1” if they needed a treatment but were unable/delayed in receiving it, “0” otherwise. USC was categorized as “1” if individuals stated that there was a particular doctor’s office, clinic, health center, or other place that they usually went to if sick or needed advice on health, and “0” otherwise. Insurance status was categorized as private, public, or uninsured. Other covariates included demographics (e.g., age, sex, race and ethnicity, marital status etc.) and self-perceived health (physical and mental health status). Chi-squared tests were used to compare proportions within and between groups. Also, logistic regressions were estimated to identify factors associated with patient barriers within and between groups. All analyses were weighted.

Results: Cancer survivors were less likely to be uninsured (8.6% vs. 11.7%, p<0.001) and more likely to have a USC (86.9% vs. 82.1%) than the comparison group. Survivors who had no access to a USC were more likely to be uninsured (21.6% vs. 6.8%) or have public insurance (18.6% vs. 16.4%) than those who had a USC (p<0.001). About 10% of cancer survivors reported needing care but being unable/delayed in receiving treatment vs. 6% of the comparison group (p<0.001). In our pooled adjusted model, survivors reported this barrier more often than the comparison group (OR=1.47, 95% CI: 1.20-1.80) and uninsured (3.40, 95% CI: 2.40-4.83) and publicly insured individuals (1.69, 95% CI: 1.23-2.34) reported this barrier more often than privately insured; however, USC had no significant influence on barrier to care. Further, among cancer survivors, compared with those who had either private insurance or USC, survivors without insurance and USC (OR=2.74, 95% CI: 1.03-7.26) were more likely to report being unable/delayed in receiving treatment.

Conclusion: For adult cancer survivors, insurance coverage is protective of barriers to required care, and also for having access to USC. More emphasis should be placed on providing adequate insurance coverage to adult cancer survivors in the U.S.

Background: Community Health Centers (CHCs) provide access to primary care services to 24 million patients annually, nearly all of whom are low-income and uninsured or publicly insured. While CHCs provide access to preventive services, chronic disease management services, and some behavioral services, for instance, specialty care is often obtained through referral, including that for behavioral health, and prescription drugs may be unaffordable to uninsured patients. Ensuring access to primary care is essential, as supported by $11 billion in ACA funding to expand CHC capacity. However, to the extent that primary care providers refer patients outside of the CHC, recommend follow-up care, or write prescriptions not provided at a discounted price, if patients cannot access these downstream services because of lack of insurance, then increased access to primary care alone will do little to impact health outcomes for these patients. Objective: To estimate the effect of having health insurance coverage on access to necessary medical care, specialty care, behavioral health care, recommended follow-up care, and medications for patients served by community health centers. Methods: We used a nationally representative sample of 5,040 non-elderly adult CHC patients from the 2014 HRSA Health Center Patient Survey, representing 13.9 million patients. We examined 19 patient-reported outcomes related to access to and delayed access to medical care (any), specialty care, behavioral health care, follow-up care after abnormal cancer screenings, any medications, and medications for hypertensive, asthmatic, diabetic, and hyperlipidemic patients. For each outcome, we calculated inverse probability of treatment weights (IPTWs) based on propensity scores to estimate average treatment effects, where patients with insurance were considered treated. Propensity scores included 20 patient-level sociodemographic and clinical covariates. Weights were stabilized to a mean of one and truncated at the 99th percentile. We used logistic regression models with IPTWs to estimate the effect of having health insurance on each outcome. Models used robust variance estimators and directly adjusted for covariates included in the propensity score model; thus, we produced doubly robust estimates. Results: In 2014, having health insurance coverage was associated with better access to most types of care examined. For instance, compared to statistically similar health center patients without insurance, patients with insurance coverage were more likely to have access to necessary medical care (aOR=2.12, 95%CI 1.74-2.58); to see a recommended specialist (aOR=2.73, 95%CI 2.15-3.46); to see a mental health professional if advised (aOR=1.74, 95%CI 1.31-2.32); to receive recommended follow-up care after an abnormal pap (aOR=3.44, 95%CI 1.80-6.54); and to get necessary prescription medications (aOR=2.10, 95%CI 1.75-2.53), particularly for patients with high cholesterol (aOR=2.25, 95%CI 1.48-3.43). Insurance was not associated with access to condition-specific medications for hypertensive, asthmatic, or diabetic patients. Discussion and conclusions: Results highlight the vital role of health insurance in accessing care within the safety-net, particularly for non-primary care services. This is especially important in light of potential reversals to Medicaid expansion, as health centers may not be able to fully compensate for resulting losses in patient insurance coverage. Furthermore, expanding safety-net capacity to provide non-primary care services for uninsured patients remains critical.

Ten years after San Francisco passed an employer mandate, we assess the effect of the law in increasing rates of health insurance and insurance generosity, and the health effects for chronic disease. Specifically, we focus on the immediate and long-term effects on cancer, a condition where early detection and treatment can significantly improve survival and thus access to health insurance is an important determinant of outcomes. We conduct difference-in-difference analyses, comparing San Francisco with three controls groups, the surrounding counties, similar counties in Great California, and another large metropolis, Los Angeles. We use the California Health Interview Survey to establish the increase in insurance rates in San Francisco relative to the control groups. We also test for greater insurance generosity and higher rates of continuous coverage as a result of the mandate. We then use the Surveillance, Epidemiology, and End Results (SEER) Program Database to determine the impact of the employer mandate in improving health effects as a result of the insurance expansion. We find that as a result of the employer mandate, the likelihood of late stage diagnosis for cancer patients was approximately 1% lower in San Francisco, a small, but statistically significant result. Survival outcomes saw much larger changes. The greatest effects occurred for lymphoma and leukemia; implementation of the employer mandate resulted in a 4% increase in one year survival rates and a 2% increase in two-year survival rates. Comparing the pre and post mandate period, San Francisco saw a 7% relative increase in survival rates. The results were similar, but smaller in magnitude for the two most common types of cancer, lung and breast. For lung cancer, the employer mandate resulted in a 3% increase and 1% increase in one and year survival rates, respectively. These numbers were 0.5% and 1% respectively for breast cancer. Breast cancer survival rates are likely attenuated by initially high survival rates; mortality rates are less than 5%, the lowest of all cancers, so there is little room to improve. This study is the first to investigate the effect of newly gained health insurance from an employer mandate on health outcomes. Employer mandates are an important policy mechanism, because they have the largest impact on individuals right above the poverty line, who are not eligible for Medicaid, but often do not have health insurance through their employer or have inadequate insurance. We show that the SF employer mandate resulted in higher rates of insurance coverage which results in the earlier detection of cancer and significantly increased rates of survival, especially among vulnerable populations. This is a positive signal for the long-term health effects in addressing the health burden of chronic conditions of the Affordable Care Act’s employer mandate as that follows the same structure as this mandate.

Background: Regional variations in health care utilization and quality are well documented. Previous studies have identified a number of factors leading to differences in access, including a shortage of healthcare professionals and adequate health insurance. Research examining differences in rural and urban health care utilization suggests that people living in rural areas may utilize health care less often. In 2013, before the Affordable Care Act (ACA), the insured rate in California was 83%. The ACA led to increased health insurance coverage for residents and funding for many healthcare providers, and in 2016 the insured rate increased to 93%. The purpose of this study is to examine regional variation in access to emergency care, hospital admissions, and length of stay for patients with diabetes before the ACA. Objective: This study aims to determine if there are variations in hospital admissions and average length of stay for diabetes for rural, urban, and frontier areas. Methods: Hospital data from the Office of Statewide Health Planning and Development (OSHPD) for the year 2013 were compared across counties using the patient’s zip code to identify the county of residence. The rurality (rural, urban, or frontier) was determined using OSHPD definition. Prevalence estimates of the number of people with diabetes in the county were obtained using the California Health Interview Survey. Data on the population was obtained from Census Bureau estimates. General linear models including Poisson regression models were used to analyze the data. Results: Overall, there were 2,505,042 people diagnosed with diabetes in California, and a total of 54,526 hospital and 72,776 emergency department admissions, with 149 hospital admissions and 199 emergency department admissions per 10,000 people with diabetes. Regional variations in admission rates and length of stay were found to exist for diabetes admissions, both overall and after controlling for patient characteristics including age, gender, insurance type, ethnicity, and comorbidities. In both hospital admissions and emergency department visits, urban counties were found to have lower rates compared to rural counties. When compared with urban areas, people in frontier areas were found to have shorter length of stays by .14 days (p=.026). Conclusion: Differences in rural and urban areas regarding admissions and length of stay were found to exist. This suggests disparities in insurance coverage and utilization exist. A follow up study examining 2016 hospital data would show if the ACA reduced these disparities because of its impact on increasing health insurance coverage.

Zero-claims bonuses, a payback agreement by the insurer contingent on a claim-free calendar year by the insured, is a common feature in the German private health insurancemarket but so far the evidence on its effects have been scarce. We study how such refunds impacts individual utilization and claiming behavior using rich administrative claims data from a large German health insurance company and an insurer policy that unexpectedly increased the refund size of certain plans. We furthermore suggest a novel method to decompose the overall effect on claims into an intensive, extensive and an automatic component. Our findings show that individuals reacted strongly to the changed incentives by reducing their claims on both the extensive and the intensive margin. We argue that this reaction is evidence of forward-looking behavior, but also show that it seems irrational in many cases, since also individuals with predictably high expenditures cut down on their utilization. Since the policy we consider was abolished again, we can also use claims from later years to study persistent effects of the policy. This analysis suggests that the reduced healthcare utilization in the intervention year led to worse health in subsequent years.

Abstract

Using NLSY79 data, this paper tests whether the source of health insurance creates incentives for newly-diagnosed workers to remain sufficiently employed to maintain access to health insurance coverage. I compare labor supply responses to new diagnoses of workers dependent on their own employment for health insurance with the responses of workers who are dependent on their spouse's employer for health insurance coverage. I use the latter as the comparison group instead of workers with no health insurance because these workers with coverage through their own employer and workers with coverage through their spouse's employer are more likely to be homogeneous in terms of observable and unobservable attributes. I focus on the following labor supply changes made 0, 6, 12 and 24 months after the diagnosis: changes in hours worked, the probability of remaining employed and the probability of moving from full-time to part-time. I find that workers who depend on their own job for health insurance are 1.5-5.5 percentage points more likely to remain employed and for those employed, are 1.3-5.4 percentage points less likely to reduce their labor hours and are 2.1-6.1 percentage points more likely to remain full-time workers. This paper addresses one of the limitations of the previous studies. The scope of earlier studies was limited to either a single state, a single disease or a particular gender or age group which limits the generalizability of conclusions to other settings. This paper addresses such limitation by using the national NLSY data which reports information on diagnosis of various illnesses. A large percentage of the diagnoses reported in NLSY were hypertension, diabetes, arthritis, and mental health problems, which compared to previous studies is more reflective of the national distribution of chronic health conditions faced by residents in the US. These conditions were not included in any of the previous studies done on this topic. Since most of the diseases in the NLSY are not life-threatening compared diseases investigated in previous studies, this paper

: Research has shown that cost-sharing affects emergency department (ED) use, with lower enrollee cost-sharing associated with increased use. There has been specific concern about ED use in the Medicaid population, where gaining coverage has been associated with increased ED use, including for conditions better treated in a primary care setting. To promote the use of primary care and encourage appropriate use of the ED, Michigan’s 2014 Medicaid expansion program does not include copays for preventive services but does include cost-sharing for non-urgent ED visits. Enrollees with incomes at less than 100 percent of the federal poverty level incurred $3 copayment for ED use; eligible enrollees with higher incomes incurred an $8 copayment for ED use. Copayments are waived for ED use for urgent conditions and are assessed for those considered non-urgent. It is not known whether copayments assessed on this

: Does a copay for non-urgent ED use discourage Healthy Michigan Plan enrollees from using the ED for non-urgent visits?

: We use Medicaid administrative claims data for enrollees in Michigan’s Medicaid expansion program who had at least 18 months of continuous enrollment, and who enrolled in the program between its inception in April 2014 and March 2015. We classify ED visits into low-medium-high severity using administrative codes, and separately into copay-eligible/copay-exempt using state algorithms. Using a time series design, we compare ED utilization for urgent visits and non-urgent visits between the first six months of an individual’s enrollment, in which no copays were assessed, with subsequent utilization. We examine use of the ED for low, medium and high severity visits. We also analyze total spending in the ED as well as spending for each type of visit. We control for age, gender, income level and region of the state.

This project is part of a required independent evaluation related to Michigan’s 1115 waiver for the Medicaid expansion demonstration. Per terms of the evaluation agreement, results are reviewed by officials at the Michigan

Conclusions will be based on data and available after results are reviewed by the state in early 2018.

This paper identifies the effect of the Affordable Care Act's (ACA’s) dependent coverage mandate on health insurance coverage, health insurance holding and labor market outcomes among young adults, by exploiting an exogenous variability in losing an additional access to health insurance coverage at age 26. To remedy the policy endogeneity problem in the literature, we exploit the discrete jump in health insurance coverage and labor market outcomes at age 26 using a fuzzy regression discontinuity design. Using alternative parametric and non-parametric models, we find that ACA's dependent coverage mandate is associated with about 1, 3-5 and 5-9 percentage points decrease in public insurance coverage, private insurance coverage and coverage from someone living outside RU, respectively, when young adults turn to 26. We also find that ACA' dependent coverage mandate decreases coverage from employment or union by about 0.7-1.5 percentage points. These results are quite robust to different model specifications. We also find that the negative effects of aging out at 26 can be offset by significant increase in employment union, nongroup, other group, and private insurance holdings. ACA's dependent coverage mandate also has spillover effects on labor market outcomes among young adults. Our results imply that it is associated with a decrease in the probability of employed by about 3.9 percentage points and a decrease in the probability of self-employment by about 1.8-2.6 percentage points when young adults turn to 26. We also find an increase in probability of engaging in temporary job when people are just older than 26. We do not find any significant change in hourly wage, weekly hours or job mobility at age 26.Our results imply that “job lock" is not a significant and major problem in labor market of young adults but there might exist some moderate “entrepreneurship lock" among young adults. These results, however, are not robust to different bandwidths and model specifications.

Uninsured individuals receive fewer health care services for at least three reasons: higher prices, responsibility for the entire bill, and potential provider reductions for concern of nonpayment. This study isolates differences in service levels between insured and uninsured individuals where uninsured individuals pay the entire bill without a contribution from an insurance company, but otherwise face the same prices. I capitalize on Maryland's highly regulated health care system, where prices are set by the state, are uniform across all patients, and hospitals are compensated or free care and bad debt, to isolate the difference in quantity demanded by the uninsured. I use a unique feature of the data, multiple readmissions for the same patients who gain or lose insurance between visits, to isolate the reductions in quantity demanded when individuals are faced with paying the full price without an insurance company contribution. While the Oregon studies compare Medicaid individuals and their low-income uninsured counterparts, this paper considers income variation among the uninsured, and quantifies the difference in demand in an environment with uniform prices. A Blinder-Oaxaca decomposition estimates uninsured individuals receive 6% fewer services after accounting for differences in patient, illness, and hospital characteristics than when these same individuals are insured. This difference in service level is larger for patients residing in low income zip codes and smaller in wealthy zip codes. This suggests that income is a substantial constraint for uninsured patients, and as that constraint relaxes, more services are demanded. For illnesses with a high risk of mortality, there was no difference in service provision for insured and uninsured individuals. The difference in service provision is attributable to illnesses with a low risk of mortality. While this paper analyses Maryland, it provides insight into demand for insured individuals with high deductible plans country wide. Prior to meeting their deductible, these insured patients face similar conditions to uninsured patients in Maryland—

In the recent decade, cost-sharing of health insurance plans increased substantially. The share of insured workers in plans with a general annual deductible has increased from 55% in 2006 to 81% in 2015, as have the average deductible amounts for covered workers in plans with deductibles from $584 in 2006 to $1,318 in 2015 (as reported by the Kaiser Family Foundation). In this analysis, we investigate the responsiveness of injured workers to changes in cost-sharing burden of their health insurance plans when deciding whether to file a workers’ compensation (WC) claim or not. Previous literature offers abundant evidence that a sizable proportion of injured workers with work-related injuries does not file for WC coverage. These substantial non participation rates suggest that WC filing is not free and associated with the following costs: employer discouragement of filing for WC benefits, stigma and loss of bonuses/overtime pay, reduction in income, upfront costs of the filing process including medical costs with uncertain prospects for reimbursement. However, in the environment of growing deductibles and other out-of-pocket payments, workers may find zero out-of-pocket coverage of medical expenses, provided by WC more appealing and may be more willing to bear the filing costs associated WC claiming. In our analysis, we explore the effect of higher cost-sharing burden on decision to file for WC coverage, while controlling for various injury, worker and employment characteristics as well as firm-specific effects. We were able to isolate the worker’s financial incentives to file for WC from the firm effects, by using variation in the annual remaining cost-sharing burden at the time of the injury across workers employed in the same organization. We find positive and statistically significant effect of higher cost-sharing on decision to file for WC coverage. This analysis relies on workers’ compensation and group health medical data coming from a large commercial national database, Truven MarketScan® for years between 2008 and 2014. It includes individuals employed by mostly large employers and insured or administered by one of approximately 100 group health plans. It also provides a wealth of information on the benefit design of the individual group health plans, including information on in and out of network

This study explores the interplay between two important public programs for vulnerable children: Medicaid and the Supplemental Security Income (SSI) program. Medicaid eligibility for children expanded in the late 1990s and early 2000s, primarily due to the creation of the Children’s Health Insurance Program (CHIP). We employ a generalized difference-in-differences design that takes advantage of the expansion of Medicaid and CHIP within states over time to isolate plausibly causal impacts of public health insurance eligibility expansions on SSI outcomes. The key data sources for the study are the Current Population Survey and Social Security Administration’s Supplemental Security Record files. On average, increases in Medicaid eligibility did not affect contemporaneous youth SSI applications or awards. However, in states where SSI recipients did not automatically receive Medicaid, expansions in public health insurance coverage led to a significant decrease in both SSI applications and awards. These results suggest that the newly available Medicaid/CHIP coverage – noteworthy for the relative ease of its application process compared with SSI – was an attractive potential substitute for SSI, especially among families that may have valued SSI primarily for the associated Medicaid benefit. In the long-term, we find that increased Medicaid eligibility during childhood reduces young adult SSI applications, consistent with recent findings that Medicaid coverage in youth improves adult health and economic outcomes.


On January 20, 2017, President Donald J. Trump penned his first executive order aiming to “minimiz[e] the economic burden” of the Affordable Care Act, signaling his intention to make good on promises to repeal and replace. The health insurance exchanges (Marketplace) allowed more than 12 million Americans to enroll in guaranteed issue community-rated health plans in 2016 and 2017. These plans are subsidized through advanced premium tax credits for those between 100% and 400% of the federal poverty level. Our analysis uses data from the open enrollment periods for these two years obtained from the federally-facilitated and state-based marketplaces to estimate the enrollment loss attributable to the transition of power and this first executive order of the new administration. We first estimated this effect descriptively with publicly available state-level Marketplace enrollment totals prior to inauguration and at the end of open enrollment using the difference between actual and expected incremental enrollment to estimate state and national enrollment declines in 2017 relative to the 2016 trend. This approach does not rely on the counts of incremental enrollment rather the proportion of overall enrollment that is expected to occur during the final two weeks so as to not bias our estimates due to changes in the size of the Marketplace eligible population from year to year. We estimate an approximately 52% incremental loss in enrollment, corresponding to approximately 365,000 fewer enrollees, during the final two weeks of open enrollment for 2017 compared to 2016 for those states enrolling through healthcare.gov. In contrast, we only observed an 18.6% decrease in enrollment during the same time period in three state-based marketplaces. We also used a difference-in-differences model with Marketplace applications data for 1,476 counties in 37 states obtained from the Centers for Medicare and Medicaid Services (CMS) through a Freedom of Information Act request. We accounted for state and county-level characteristics potentially associated with enrollment volume, including state Medicaid expansion status (CMS), 2010 county population and percentage of county population living in urban areas (Census), number of Marketplace insurance carriers and the Silver gap (difference between benchmark Silver plan and the least expensive Silver plan for a single 40-year old) by county and year (CMS), and the percentage of votes received by President Trump in the 2016 Presidential election in each county (Townhall). We estimate a population-weighted decline of approximately 720 applications per county-week during the final two weeks of the 2017 open enrollment period relative to 2016 (b=–720.5, p<.001), corresponding to a 32% decline in applications submitted. The perceived lack of political support for the law by the incoming administration had an immediate and significant effect on Marketplace enrollment nationwide.


We examine consumer plan choice and learning in 2015 and 2016 in the Federally-facilitated Marketplaces (FFM) for health insurance established by the Affordable Care Act (ACA). The FFM offers a useful context for studying choice in private insurance markets. Plans offered in the FFM are displayed in an online “exchange” and are required to meet standards regarding coverage, premiums, benefits, and cost-sharing. These requirements as well as the design of the exchange is intended to facilitate consumer shopping and there is evidence that consumers actively shop and switch plans in the FFM at higher rates than in the market for employer-sponsored insurance. Understanding the dynamics of consumer choice in this setting is important given that the Marketplaces are still relatively new markets for consumers and insurers and that the regulations described above were intended to standardize and improve the options for consumers getting insurance through the non-group market. A deeper understanding of consumer decision-making is fundamental to improving the design of insurance markets as well as ensuring that the individual insurance market is

A combination of public data on plans available at the county level and associated benefit design features and administrative individual-level panel data on FFM enrollment allows us to identify each enrollee’s choice set and their chosen plan as well as to see how their choices evolved over time. Further, these data allow us to calculate the actual premiums, deductibles, and max OOP levels that consumers faced, taking into account cost-sharing reductions and other subsidies. We characterize enrollees’ choice sets and estimate discrete choice models of individual-level plan choice that account for plan characteristics and interactions with individual characteristics. These choice models allow us to assess the value that consumers place on four benefit design features: premium, deductible, maximum out-of-pocket (OOP), and size of the physician network. We use our estimates to calculate individuals’ willingness-to-pay (WTP) for benefits such as lower deductibles and broader provider networks and to variations in WTP across different populations. Finally, we examine choices by consumer year of entry to assess how WTP changes over time, which may reflect

We find that consumers valued each plan benefit design feature when making their choice and were willing to pay substantial amounts to lower deductibles and max OOP levels and to have access to a broader physician network. Specifically, 2016 Marketplace enrollees were willing to pay $257 annually to reduce their deductible by $1,000, $108 to reduce their max OOP by $1,000, and $151 annually to increase the network penetration of their plan by 25 percentage points. Willingness-to-pay for all features increased with age and was generally higher for women than men; WTP for greater network breadth also increased with income. Further, WTP for bigger networks increased monotonically over time for individuals who enrolled in Marketplace plans for multiple years, suggesting that consumers learned over time. These results are also consistent with the introduction of provider search tools and other network

How much do demographic, insurance market, and policy factors affect exchange enrollment through healthcare.gov? The exchange market for health insurance is central to many legislative proposals that the Congressional Budget Office analyzes. We can more accurately predict the budgetary and coverage effects of legislative changes if we better understand how much exchange take-up varies with demographic, insurance market, and policy factors. Estimates from the literature are from survey data or aggregated administrative data on plan selections during open enrollment. We are able to improve on this literature using person-level data on effectuated enrollment through healthcare.gov in 2015 and 2016 from the Center for Medicare and Medicaid Services (CMS). These data allow us to accurately measure exchange take-up and control for variation in the demographic composition of local geographic areas. By reducing measurement error and increasing sample size, our analysis will yield more precise estimators. Our analysis is based on logistic regression models of the take-up of exchange coverage by potential enrollees. Our sample consists of 779 local geographic areas in 2015 and 782 in 2016 stratified by gender, age (20-34, 35-44, 45-54, and 55-64), and income (100/138-250 FPL and 250-400 FPL) resulting in a sample of 24,976 observations. We use CMS’s person level data on effectuated enrollment through healthcare.gov to measure the full-year equivalent number of ‑exchange enrollees. To calculate take-up rates, we divide these totals by estimates of the population of potential enrollees based on the Census Bureau’s Population Estimates Program and the 2011–2015 American Community Survey 5-year public use microdata sample. We define a potential enrollee as a person who is eligible for the premium tax credit, who is not eligible for Medicaid or Medicare, and who does not have employer-sponsored insurance or TRICARE. Additional datasets from the Census Bureau, CMS, the Agency for Healthcare Research and Quality, and the Commonwealth Fund are used to measure important control variables for demographic, insurance market, and policy factors. Our preliminary analysis suggests that many factors affect the odds of exchange enrollment. The odds of enrollment are predicted to be much higher for women than men and to increase with age. The odds of enrollment are predicted to decrease with the share of potential enrollees who are exempt from the individual mandate, who do not speak English very well, who are Hispanic, and who are native to the United States. The size of the non-group market in 2013, the exchange participation of insurers with high market share in the non-group market in 2013, and the number of primary care physicians per capita in a region all increase the predicted odds of enrollment. Potential enrollees with eligibility for high subsidies have much higher predicted odds of enrollment than those with lower subsidies. State restrictions on enrollment assistance substantially lower the predicted odds of enrollment. Finally, several factors that we expected to drive exchange take-up, including reference premiums, insurer competition, and the share of the population living in a rural area did not.

Health insurance contracts are often structured with different forms of cost sharing in order to provide consumption smoothing across different health outcomes while limiting moral hazard. These components of health insurance may influence an individual’s behavior in multiple ways. The research on moral hazard has largely focused on the responsiveness of consumers spending on healthcare to the cost of care. Alternatively, most workers with employer-sponsored health insurance (EHI) also have access to workers’ compensation (WC) for medical coverage of injuries and illnesses which may be related to their work. WC medical coverage has no copayment or deductible associated with coverage, but employees need to prove their injury is work-related and they are often limited in their choice of medical care provider. The costs for the employee associated with using EHI differs across plans in their cost-sharing mechanisms and over the course of a year if a deductible is present. Differences in plans and time of year provide variation in the end-of-year and spot price of medical care and can identify whether or not workers vary their utilization of WC dependent upon

Using MarketScan data containing WC and health insurance claims, the analysis first confirms employee sensitivity to future prices of medical care using a difference-in-difference technique. The differences in plan deductibles and employee enrollment months allows for control of seasonal differences in medical utilization while focusing on variation in the end-of-year price of medical care that comes with differing enrollment months similar to analysis done in Aron-Dine, Einav, Finkelstein, and Cullen (2015). This analysis is then replicating using the WC claims information for the same set of employees. The empirical work finds evidence of cost shifting to WC as employees are less likely to use EHI and more likely to use WC if they enroll in their EHI later in the year and face a higher end-of-year price of medical care using the EHI. Measuring the extent of WC cost-shifting is important for a number of reasons. In addition to providing benefits to employees injured on the job, WC also serves as an incentive for employers who are experience-rated to invest in the safety of their workplaces. Variation in WC claiming due to changing prices of EHI medical care is likely to also impact workplace injuries reported to employers, and therefore measuring the extent of WC cost shifting can better inform

The relationship between minimum wages and health insurance coverage among low-wage workers can be complicated. For employers, an increase in the minimum wage that raises the total wage bill may cause changes in health insurance provision; employers may cease to provide health insurance or increase the premiums paid by workers. This would tend to reduce health insurance coverage among low-wage workers. At the same time, for the worker, an increase in the minimum wage may mean that health insurance becomes more affordable, particularly if the cost of the health insurance is split between the worker and the employer, or between the worker and the government. In our research, we quantify the relationship between minimum wages and health insurance coverage among a national sample of agricultural workers, and we examine how that relationship differs between male and female workers. To test the relationship between minimum wages and various types of health insurance coverage among agricultural workers, we use a Probit model in which we control for worker characteristics that are standard in a Mincerian wage equation. Our model also includes state and year dummies, as well as census region by year effects, worker task and crop dummies and legal status. Our sample of full-time crop workers comes from the confidential version of the National Agricultural Workers Survey (NAWS), from the years 2000 through 2014. The NAWS is the only nationally representative survey of demographic, employment, and health insurance characteristics of hired crop workers. Information is obtained directly from farm workers through face-to-face interviews. Our population includes roughly 30,000 crop workers, 82 percent of which are male. This data is unique in its level of detail on the health insurance coverage for crop workers and their families – for each agricultural worker, and each member of the family, the survey reports whether the individual has health insurance and if it was paid for by the worker, the worker’s employer, the spouse, the spouse’s

We find that, among male agricultural workers, overall health insurance coverage from all sources combined falls significantly as the minimum wage increases; this effect is largely due to a decline in health insurance paid for by the worker’s employer. Among female agricultural workers, overall health insurance coverage from all sources combined does not change significantly. While we find a decline in health insurance coverage paid for by the female worker’s employer alone, this negative effect is offset by an increase in the incidence of health insurance whose cost is shared between the employer (or the spouse’s employer) and the worker’s family. We also find that an increase in the minimum wage leads to an increase in health insurance coverage for the worker’s children; an effect that is mostly driven by an increase in the government health insurance. While state policies increasing the minimum wage are aimed at improving the well-being of low-wage workers, we find that they may also have some negative consequences on workers’ health insurance coverage, particularly coverage provided by their employers.

We use a novel data set from eHealthInsurance.com to estimate the demand for health insurance in the individual market. We use this data to provide the first estimates of health insurance demand in the 36 states that do not manage state-run health insurance exchanges. We find that median own-price elasticities range between -.59 and -1.55. While uninsurance rates in this market are high---45% on average---we find that diversion towards uninsurance are low. The median diversion ratios range between 10.1% to 18.2%. The low diversion ratios imply that the penalty for uninsurance implemented through the Individual Mandate may not have a large impact on the uninsurance rate. We do a partial equilibrium exercise and find that a repeal of the Individual Mandate would lead to a 1.7 percentage point increase the uninsurance rate in the individual market. This represents solely the immediate demand effect, and does not

We utilize credit report data to study how unpaid medical collections and other financial outcomes are affected by fundamental features of health care. Specifically, we document and discuss the size and age distributions of unpaid medical bills on consumer credit profiles, use Medicare eligibility and the implementation of the Affordable Care Act to identify the causal effect of insurance coverage, and investigate whether geographic variation in healthcare prices is predictive of medical debt. We show that Medicare sharply decreases the dollar value of unpaid medical collections held by consumers. Moreover, as the ACA reduces the extensive margin gain in insurance coverage at age 65, the effect of Medicare on medical debt falls with it. In particular, from 2013 to 2015 the increase in insurance coverage at age 65 falls by 46% and the effect of Medicare on medical collections declines by a similar 53%. However, Medicare has relatively small effects on the number of people who have medical bills sent to collections, instead predominantly reducing the size of bills which go unpaid. In part we reconcile this with the fact that most unpaid bills are relatively small or modestly sized – over half are less than $600. Thus, while Medicare may substantially reduce the size of bills facing the formerly uninsured, that amount may still exceed many consumer’s willingness or ability to pay. In turn, we show that Medicare has muted effects on other outcomes on consumer credit profiles. It is particularly notable that, despite very salient increases in insurance coverage at Medicare eligibility, we observe little change in credit scores at age 65. The patterns we observe are broadly consistent with recent research which argues that the burden of uncompensated care largely falls on health care providers.

Enacted in 2010 and largely implemented in 2014, the Affordable Care Act (ACA) has expanded health insurance coverage to more than 20 million previously uninsured American adults through the Medicaid expansions implemented by 31 states and health insurance exchanges on which individuals can purchase insurance and received subsidies if their incomes are between 100% and 400% of the Federal Poverty Level. While nearly all Americans with end-stage renal disease (ESRD) have insurance after the onset of this condition, those who are uninsured may not have access to effective care before developing ESRD. We assess whether changes in insurance status prior to the onset of ESRD were accompanied by changes in markers of pre-ESRD care in patients receiving their first dialysis treatment pre-enactment of the ACA versus post-enactment of the ACA. We use data from the CMS Medical Evidence Form 2728 which is completed for all patients at incidence of ESRD. This included 511,424 patients from 10/1/11 to 3/31/16. We estimate Difference-in-Differences (D-in-D) models by age/insurance status at ESRD onset [age <65 with no Medicare (treatment group) vs. age ≥ 66 with Medicare (control group)], and by state Medicaid expansion status for age <65 [States expanding Medicaid (treatment group) vs. states not expanding Medicaid (control group)]. Using linear probability models, we determine the ACA’s impact on insurance status and key markers of pre-ESRD care. Models include state and year fixed effects. The assumption of parallel trends in pre-ACA period is supported using plots of monthly data and the results are robust to using narrower age windows. Among patients age <65 not on Medicare, insurance rates increased from 82.4% pre-ACA to 86.7% in 2014, 89.3% in 2015 and 92.1% in 2016. These patients also experienced increases in pre-ESRD nephrology care, use of home dialysis and use of anemia medication; effects on vascular access were equivocal. Patients in states that expanded Medicaid had 3.4 percentage point greater increases in insurance coverage than those in non-expansion states. They also experienced greater increases in pre-ESRD nephrology care and better vascular access, but no change in the use of home dialysis. Overall, these findings demonstrate that several key indicators of pre-ESRD care improved post-ACA implementation, but which indicators improved differed somewhat across the two affected populations: (A) all adults younger than 65 and not on Medicare and (B) adults younger than 65 in Medicaid expansion states. Our findings are consistent with improved access to care that has significant potential to improve clinical outcomes for those with advanced kidney disease in the United States. The magnitudes of the effects are relatively large. Overall, uninsurance fell dramatically during the post-ACA period. About 25-33% of the patients who gained insurance also gained relevant pre-ESRD care, demonstrating the important role of health insurance coverage in access to pre-ESRD care. These are meaningful changes in the context of a clinically and economically vulnerable population. Further research can establish the extent to which the observed improvements in pre-ESRD care affect post-ESRD outcomes such as

Lack of access to both health insurance coverage and usual source of care (USC) can be a potential barrier for appropriate use of health care services. For special disease groups, such as adult cancer survivors in the United States, both insurance and USC could be critical for improved access and health outcomes. Yet, no study has evaluated how insurance and USC interact and affect adult cancer survivors’ health care barriers in the U.S.

We examine whether insurance and USC are associated with adult cancer survivors’ barriers to health care compared with same age and sex individuals without a history of cancer.

We used the 2012 to 2015 Medical Expenditures Panel Survey (MEPS) data to identify N=4,009 adult cancer survivors, currently aged 18-64 years, who reported ever being diagnosed with cancer. We used one-to-one propensity score matching to identify 4,009 comparison group individuals without a history of cancer. Patients’ barrier was categorized as “1” if they needed a treatment but were unable/delayed in receiving it, “0” otherwise. USC was categorized as “1” if individuals stated that there was a particular doctor’s office, clinic, health center, or other place that they usually went to if sick or needed advice on health, and “0” otherwise. Insurance status was categorized as private, public, or uninsured. Other covariates included demographics (e.g., age, sex, race and ethnicity, marital status etc.) and self-perceived health (physical and mental health status). Chi-squared tests were used to compare proportions within and between groups. Also, logistic regressions were estimated to identify factors associated with patient barriers within and between groups. All analyses were weighted.

Cancer survivors were less likely to be uninsured (8.6% vs. 11.7%, p<0.001) and more likely to have a USC (86.9% vs. 82.1%) than the comparison group. Survivors who had no access to a USC were more likely to be uninsured (21.6% vs. 6.8%) or have public insurance (18.6% vs. 16.4%) than those who had a USC (p<0.001). About 10% of cancer survivors reported needing care but being unable/delayed in receiving treatment vs. 6% of the comparison group (p<0.001). In our pooled adjusted model, survivors reported this barrier more often than the comparison group (OR=1.47, 95% CI: 1.20-1.80) and uninsured (3.40, 95% CI: 2.40-4.83) and publicly insured individuals (1.69, 95% CI: 1.23-2.34) reported this barrier more often than privately insured; however, USC had no significant influence on barrier to care. Further, among cancer survivors, compared with those who had either private insurance or USC, survivors without insurance and USC (OR=2.74, 95% CI: 1.03-7.26) were more likely to report being unable/delayed in receiving treatment.

For adult cancer survivors, insurance coverage is protective of barriers to required care, and also for having access to USC. More emphasis should be placed on providing adequate insurance coverage to adult cancer survivors

: Community Health Centers (CHCs) provide access to primary care services to 24 million patients annually, nearly all of whom are low-income and uninsured or publicly insured. While CHCs provide access to preventive services, chronic disease management services, and some behavioral services, for instance, specialty care is often obtained through referral, including that for behavioral health, and prescription drugs may be unaffordable to uninsured patients. Ensuring access to primary care is essential, as supported by $11 billion in ACA funding to expand CHC capacity. However, to the extent that primary care providers refer patients outside of the CHC, recommend follow-up care, or write prescriptions not provided at a discounted price, if patients cannot access these downstream services because of lack of insurance, then increased access to primary care alone will do little to impact health outcomes for these

: To estimate the effect of having health insurance coverage on access to necessary medical care, specialty care, behavioral health care, recommended follow-up care, and medications for patients served by community health

: We used a nationally representative sample of 5,040 non-elderly adult CHC patients from the 2014 HRSA Health Center Patient Survey, representing 13.9 million patients. We examined 19 patient-reported outcomes related to access to and delayed access to medical care (any), specialty care, behavioral health care, follow-up care after abnormal cancer screenings, any medications, and medications for hypertensive, asthmatic, diabetic, and hyperlipidemic patients. For each outcome, we calculated inverse probability of treatment weights (IPTWs) based on propensity scores to estimate average treatment effects, where patients with insurance were considered treated. Propensity scores included 20 patient-level sociodemographic and clinical covariates. Weights were stabilized to a mean of one and truncated at the 99th percentile. We used logistic regression models with IPTWs to estimate the effect of having health insurance on each outcome. Models used robust variance estimators and directly adjusted for covariates included in the propensity score model; thus, we produced doubly robust estimates.

: In 2014, having health insurance coverage was associated with better access to most types of care examined. For instance, compared to statistically similar health center patients without insurance, patients with insurance coverage were more likely to have access to necessary medical care (aOR=2.12, 95%CI 1.74-2.58); to see a recommended specialist (aOR=2.73, 95%CI 2.15-3.46); to see a mental health professional if advised (aOR=1.74, 95%CI 1.31-2.32); to receive recommended follow-up care after an abnormal pap (aOR=3.44, 95%CI 1.80-6.54); and to get necessary prescription medications (aOR=2.10, 95%CI 1.75-2.53), particularly for patients with high cholesterol (aOR=2.25, 95%CI 1.48-3.43). Insurance was not associated with access to condition-specific medications for hypertensive, asthmatic, or diabetic patients.

: Results highlight the vital role of health insurance in accessing care within the safety-net, particularly for non-primary care services. This is especially important in light of potential reversals to Medicaid expansion, as health centers may not be able to fully compensate for resulting losses in patient insurance coverage. Furthermore, expanding safety-net capacity to provide non-primary care services for uninsured patients remains critical.

Ten years after San Francisco passed an employer mandate, we assess the effect of the law in increasing rates of health insurance and insurance generosity, and the health effects for chronic disease. Specifically, we focus on the immediate and long-term effects on cancer, a condition where early detection and treatment can significantly improve survival and thus access to health insurance is an important determinant of outcomes. We conduct difference-in-difference analyses, comparing San Francisco with three controls groups, the surrounding counties, similar counties in Great California, and another large metropolis, Los Angeles. We use the California Health Interview Survey to establish the increase in insurance rates in San Francisco relative to the control groups. We also test for greater insurance generosity and higher rates of continuous coverage as a result of the mandate. We then use the Surveillance, Epidemiology, and End Results (SEER) Program Database to determine the impact of the employer mandate in improving health effects as a result of the insurance expansion. We find that as a result of the employer mandate, the likelihood of late stage diagnosis for cancer patients was approximately 1% lower in San Francisco, a small, but statistically significant result. Survival outcomes saw much larger changes. The greatest effects occurred for lymphoma and leukemia; implementation of the employer mandate resulted in a 4% increase in one year survival rates and a 2% increase in two-year survival rates. Comparing the pre and post mandate period, San Francisco saw a 7% relative increase in survival rates. The results were similar, but smaller in magnitude for the two most common types of cancer, lung and breast. For lung cancer, the employer mandate resulted in a 3% increase and 1% increase in one and year survival rates, respectively. These numbers were 0.5% and 1% respectively for breast cancer. Breast cancer survival rates are likely attenuated by initially high survival rates; mortality rates are

This study is the first to investigate the effect of newly gained health insurance from an employer mandate on health outcomes. Employer mandates are an important policy mechanism, because they have the largest impact on individuals right above the poverty line, who are not eligible for Medicaid, but often do not have health insurance through their employer or have inadequate insurance. We show that the SF employer mandate resulted in higher rates of insurance coverage which results in the earlier detection of cancer and significantly increased rates of survival, especially among vulnerable populations. This is a positive signal for the long-term health effects in addressing the health burden of chronic conditions of the Affordable Care Act’s employer mandate as that follows the same structure as this mandate.

d: Regional variations in health care utilization and quality are well documented. Previous studies have identified a number of factors leading to differences in access, including a shortage of healthcare professionals and adequate health insurance. Research examining differences in rural and urban health care utilization suggests that people living in rural areas may utilize health care less often. In 2013, before the Affordable Care Act (ACA), the insured rate in California was 83%. The ACA led to increased health insurance coverage for residents and funding for many healthcare providers, and in 2016 the insured rate increased to 93%. The purpose of this study is to examine regional variation in access to emergency care, hospital admissions, and length of stay for patients with diabetes before the ACA.

: This study aims to determine if there are variations in hospital admissions and average length of stay for diabetes for rural, urban, and frontier areas. : Hospital data from the Office of Statewide Health Planning and Development (OSHPD) for the year 2013 were compared across counties using the patient’s zip code to identify the county of residence. The rurality (rural, urban, or

frontier) was determined using OSHPD definition. Prevalence estimates of the number of people with diabetes in the county were obtained using the California Health Interview Survey. Data on the population was obtained from Census Bureau estimates. General linear models including Poisson regression models were used to analyze the data.

: Overall, there were 2,505,042 people diagnosed with diabetes in California, and a total of 54,526 hospital and 72,776 emergency department admissions, with 149 hospital admissions and 199 emergency department admissions per 10,000 people with diabetes. Regional variations in admission rates and length of stay were found to exist for diabetes admissions, both overall and after controlling for patient characteristics including age, gender, insurance type, ethnicity, and comorbidities. In both hospital admissions and emergency department visits, urban counties were found to have lower rates compared to rural counties. When compared with urban areas, people in frontier areas were

: Differences in rural and urban areas regarding admissions and length of stay were found to exist. This suggests disparities in insurance coverage and utilization exist. A follow up study examining 2016 hospital data would show if the ACA reduced these disparities because of its impact on increasing health insurance coverage.

Zero-claims bonuses, a payback agreement by the insurer contingent on a claim-free calendar year by the insured, is a common feature in the German private health insurancemarket but so far the evidence on its effects have been scarce. We study how such refunds impacts individual utilization and claiming behavior using rich administrative claims data from a large German health insurance company and an insurer policy that unexpectedly increased the refund size of certain plans. We furthermore suggest a novel method to decompose the overall effect on claims into an intensive, extensive and an automatic component. Our findings show that individuals reacted strongly to the changed incentives by reducing their claims on both the extensive and the intensive margin. We argue that this reaction is evidence of forward-looking behavior, but also show that it seems irrational in many cases, since also individuals with predictably high expenditures cut down on their utilization. Since the policy we consider was abolished again, we can also use claims from later years to study persistent effects of the policy. This analysis suggests that the reduced healthcare utilization in the intervention year led to worse health in subsequent years.


Sheryll Namingit [email protected]

Elizabeth Cliff [email protected]

Linna Xu [email protected]

Amanda Cook [email protected]

Using NLSY79 data, this paper tests whether the source of health insurance creates incentives for newly-diagnosed workers to remain sufficiently employed to maintain access to health insurance coverage. I compare labor supply responses to new diagnoses of workers dependent on their own employment for health insurance with the responses of workers who are dependent on their spouse's employer for health insurance coverage. I use the latter as the comparison group instead of workers with no health insurance because these workers with coverage through their own employer and workers with coverage through their spouse's employer are more likely to be homogeneous in terms of observable and unobservable attributes. I focus on the following labor supply changes made 0, 6, 12 and 24 months after the diagnosis: changes in hours worked, the probability of remaining employed and the probability of moving from full-time to part-time. I find that workers who depend on their own job for health insurance are 1.5-5.5 percentage points more likely to remain employed and for those employed, are 1.3-5.4 percentage points less likely to reduce their labor hours and are 2.1-6.1 percentage points more likely to remain full-time workers. This paper addresses one of the limitations of the previous studies. The scope of earlier studies was limited to either a single state, a single disease or a particular gender or age group which limits the generalizability of conclusions to other settings. This paper addresses such limitation by using the national NLSY data which reports information on diagnosis of various illnesses. A large percentage of the diagnoses reported in NLSY were hypertension, diabetes, arthritis, and mental health problems, which compared to previous studies is more reflective of the national distribution of chronic health conditions faced by residents in the US. These conditions were not included in any of the previous studies done on this topic. Since most of the diseases in the NLSY are not life-threatening compared diseases investigated in previous studies, this paper

: Research has shown that cost-sharing affects emergency department (ED) use, with lower enrollee cost-sharing associated with increased use. There has been specific concern about ED use in the Medicaid population, where gaining coverage has been associated with increased ED use, including for conditions better treated in a primary care setting. To promote the use of primary care and encourage appropriate use of the ED, Michigan’s 2014 Medicaid expansion program does not include copays for preventive services but does include cost-sharing for non-urgent ED visits. Enrollees with incomes at less than 100 percent of the federal poverty level incurred $3 copayment for ED use; eligible enrollees with higher incomes incurred an $8 copayment for ED use. Copayments are waived for ED use for urgent conditions and are assessed for those considered non-urgent. It is not known whether copayments assessed on this

: We use Medicaid administrative claims data for enrollees in Michigan’s Medicaid expansion program who had at least 18 months of continuous enrollment, and who enrolled in the program between its inception in April 2014 and March 2015. We classify ED visits into low-medium-high severity using administrative codes, and separately into copay-eligible/copay-exempt using state algorithms. Using a time series design, we compare ED utilization for urgent visits and non-urgent visits between the first six months of an individual’s enrollment, in which no copays were assessed, with subsequent utilization. We examine use of the ED for low, medium and high severity visits. We also analyze

This project is part of a required independent evaluation related to Michigan’s 1115 waiver for the Medicaid expansion demonstration. Per terms of the evaluation agreement, results are reviewed by officials at the Michigan

This paper identifies the effect of the Affordable Care Act's (ACA’s) dependent coverage mandate on health insurance coverage, health insurance holding and labor market outcomes among young adults, by exploiting an exogenous variability in losing an additional access to health insurance coverage at age 26. To remedy the policy endogeneity problem in the literature, we exploit the discrete jump in health insurance coverage and labor market outcomes at age 26 using a fuzzy regression discontinuity design. Using alternative parametric and non-parametric models, we find that ACA's dependent coverage mandate is associated with about 1, 3-5 and 5-9 percentage points decrease in public insurance coverage, private insurance coverage and coverage from someone living outside RU, respectively, when young adults turn to 26. We also find that ACA' dependent coverage mandate decreases coverage from employment or union by about 0.7-1.5 percentage points. These results are quite robust to different model specifications. We also find that the negative effects of aging out at 26 can be offset by significant increase in employment union, nongroup, other group, and private insurance holdings. ACA's dependent coverage mandate also has spillover effects on labor market outcomes among young adults. Our results imply that it is associated with a decrease in the probability of employed by about 3.9 percentage points and a decrease in the probability of self-employment by about 1.8-2.6 percentage points when young adults turn to 26. We also find an increase in probability of engaging in temporary job when people are just older than 26. We do not find any significant change in hourly wage, weekly hours or job mobility at age 26.Our results imply that “job lock" is not a significant and major problem in labor market of young adults but there might exist some

Uninsured individuals receive fewer health care services for at least three reasons: higher prices, responsibility for the entire bill, and potential provider reductions for concern of nonpayment. This study isolates differences in service levels between insured and uninsured individuals where uninsured individuals pay the entire bill without a contribution from an insurance company, but otherwise face the same prices. I capitalize on Maryland's highly regulated health care system, where prices are set by the state, are uniform across all patients, and hospitals are compensated or free care and bad debt, to isolate the difference in quantity demanded by the uninsured. I use a unique feature of the data, multiple readmissions for the same patients who gain or lose insurance between visits, to isolate the reductions in quantity demanded when individuals are faced with paying the full price without an insurance company contribution. While the Oregon studies compare Medicaid individuals and their low-income uninsured counterparts, this paper considers income variation among the uninsured, and quantifies the difference in demand in an environment with uniform prices. A Blinder-Oaxaca decomposition estimates uninsured individuals receive 6% fewer services after accounting for differences in patient, illness, and hospital characteristics than when these same individuals are insured. This difference in service level is larger for patients residing in low income zip codes and smaller in wealthy zip codes. This suggests that income is a substantial constraint for uninsured patients, and as that constraint relaxes, more services are demanded. For illnesses with a high risk of mortality, there was no difference in service provision for insured and uninsured individuals. The difference in service provision is attributable to illnesses with a low risk of mortality. While this paper analyses Maryland, it provides insight into demand for insured individuals with high deductible plans country wide. Prior to meeting their deductible, these insured patients face similar conditions to uninsured patients in Maryland—

Olesya Fomenko [email protected]

Sean Orzol [email protected]

Daniel Kaliski [email protected]

Paul Shafer [email protected]

In the recent decade, cost-sharing of health insurance plans increased substantially. The share of insured workers in plans with a general annual deductible has increased from 55% in 2006 to 81% in 2015, as have the average deductible amounts for covered workers in plans with deductibles from $584 in 2006 to $1,318 in 2015 (as reported by the Kaiser Family Foundation). In this analysis, we investigate the responsiveness of injured workers to changes in cost-sharing burden of their health insurance plans when deciding whether to file a workers’ compensation (WC) claim or not. Previous literature offers abundant evidence that a sizable proportion of injured workers with work-related injuries does not file for WC coverage. These substantial non participation rates suggest that WC filing is not free and associated with the following costs: employer discouragement of filing for WC benefits, stigma and loss of bonuses/overtime pay, reduction in income, upfront costs of the filing process including medical costs with uncertain prospects for reimbursement. However, in the environment of growing deductibles and other out-of-pocket payments, workers may find zero out-of-pocket coverage of medical expenses, provided by WC more appealing and may be more willing to bear the filing costs associated WC claiming. In our analysis, we explore the effect of higher cost-sharing burden on decision to file for WC coverage, while controlling for various injury, worker and employment characteristics as well as firm-specific effects. We were able to isolate the worker’s financial incentives to file for WC from the firm effects, by using variation in the annual remaining cost-sharing burden at the time of the injury across workers employed in the same organization. We find positive and statistically significant effect of higher cost-sharing on decision to file for WC coverage. This analysis relies on workers’ compensation and group health medical data coming from a large commercial national database, Truven MarketScan® for years between 2008 and 2014. It includes individuals employed by mostly large employers and insured or administered by one of approximately 100 group health plans. It also provides a wealth of information on the benefit design of the individual group health plans, including information on in and out of network

This study explores the interplay between two important public programs for vulnerable children: Medicaid and the Supplemental Security Income (SSI) program. Medicaid eligibility for children expanded in the late 1990s and early 2000s, primarily due to the creation of the Children’s Health Insurance Program (CHIP). We employ a generalized difference-in-differences design that takes advantage of the expansion of Medicaid and CHIP within states over time to isolate plausibly causal impacts of public health insurance eligibility expansions on SSI outcomes. The key data sources for the study are the Current Population Survey and Social Security Administration’s Supplemental Security Record files. On average, increases in Medicaid eligibility did not affect contemporaneous youth SSI applications or awards. However, in states where SSI recipients did not automatically receive Medicaid, expansions in public health insurance coverage led to a significant decrease in both SSI applications and awards. These results suggest that the newly available Medicaid/CHIP coverage – noteworthy for the relative ease of its application process compared with SSI – was an attractive potential substitute for SSI, especially among families that may have valued SSI primarily for the associated Medicaid benefit. In the long-term, we find that increased Medicaid eligibility during childhood reduces young adult SSI


On January 20, 2017, President Donald J. Trump penned his first executive order aiming to “minimiz[e] the economic burden” of the Affordable Care Act, signaling his intention to make good on promises to repeal and replace. The health insurance exchanges (Marketplace) allowed more than 12 million Americans to enroll in guaranteed issue community-rated health plans in 2016 and 2017. These plans are subsidized through advanced premium tax credits for those between 100% and 400% of the federal poverty level. Our analysis uses data from the open enrollment periods for these two years obtained from the federally-facilitated and state-based marketplaces to estimate the enrollment loss attributable to the transition of power and this first executive order of the new administration. We first estimated this effect descriptively with publicly available state-level Marketplace enrollment totals prior to inauguration and at the end of open enrollment using the difference between actual and expected incremental enrollment to estimate state and national enrollment declines in 2017 relative to the 2016 trend. This approach does not rely on the counts of incremental enrollment rather the proportion of overall enrollment that is expected to occur during the final two weeks so as to not bias our estimates due to changes in the size of the Marketplace eligible population from year to year. We estimate an approximately 52% incremental loss in enrollment, corresponding to approximately 365,000 fewer enrollees, during the final two weeks of open enrollment for 2017 compared to 2016 for those states enrolling through healthcare.gov. In contrast, we only observed an 18.6% decrease in enrollment during the same time period in three state-based marketplaces. We also used a difference-in-differences model with Marketplace applications data for 1,476 counties in 37 states obtained from the Centers for Medicare and Medicaid Services (CMS) through a Freedom of Information Act request. We accounted for state and county-level characteristics potentially associated with enrollment volume, including state Medicaid expansion status (CMS), 2010 county population and percentage of county population living in urban areas (Census), number of Marketplace insurance carriers and the Silver gap (difference between benchmark Silver plan and the least expensive Silver plan for a single 40-year old) by county and year (CMS), and the percentage of votes received by President Trump in the 2016 Presidential election in each county (Townhall). We

<.001), corresponding to a 32% decline in

Anaka Aiyar [email protected]

Aditi Sen [email protected]

Ben Hopkins [email protected]


We examine consumer plan choice and learning in 2015 and 2016 in the Federally-facilitated Marketplaces (FFM) for health insurance established by the Affordable Care Act (ACA). The FFM offers a useful context for studying choice in private insurance markets. Plans offered in the FFM are displayed in an online “exchange” and are required to meet standards regarding coverage, premiums, benefits, and cost-sharing. These requirements as well as the design of the exchange is intended to facilitate consumer shopping and there is evidence that consumers actively shop and switch plans in the FFM at higher rates than in the market for employer-sponsored insurance. Understanding the dynamics of consumer choice in this setting is important given that the Marketplaces are still relatively new markets for consumers and insurers and that the regulations described above were intended to standardize and improve the options for consumers getting insurance through the non-group market. A deeper understanding of consumer decision-making is fundamental to improving the design of insurance markets as well as ensuring that the individual insurance market is

A combination of public data on plans available at the county level and associated benefit design features and administrative individual-level panel data on FFM enrollment allows us to identify each enrollee’s choice set and their chosen plan as well as to see how their choices evolved over time. Further, these data allow us to calculate the actual premiums, deductibles, and max OOP levels that consumers faced, taking into account cost-sharing reductions and other subsidies. We characterize enrollees’ choice sets and estimate discrete choice models of individual-level plan choice that account for plan characteristics and interactions with individual characteristics. These choice models allow us to assess the value that consumers place on four benefit design features: premium, deductible, maximum out-of-pocket (OOP), and size of the physician network. We use our estimates to calculate individuals’ willingness-to-pay (WTP) for benefits such as lower deductibles and broader provider networks and to variations in WTP across different populations. Finally, we examine choices by consumer year of entry to assess how WTP changes over time, which may reflect

We find that consumers valued each plan benefit design feature when making their choice and were willing to pay substantial amounts to lower deductibles and max OOP levels and to have access to a broader physician network. Specifically, 2016 Marketplace enrollees were willing to pay $257 annually to reduce their deductible by $1,000, $108 to reduce their max OOP by $1,000, and $151 annually to increase the network penetration of their plan by 25 percentage points. Willingness-to-pay for all features increased with age and was generally higher for women than men; WTP for greater network breadth also increased with income. Further, WTP for bigger networks increased monotonically over time for individuals who enrolled in Marketplace plans for multiple years, suggesting that consumers learned over time. These results are also consistent with the introduction of provider search tools and other network

The exchange market for health insurance is central to many legislative proposals that the Congressional Budget Office analyzes. We can more accurately predict the budgetary and coverage effects of legislative changes if we better understand how much exchange take-up varies with demographic, insurance market, and policy factors. Estimates from the literature are from survey data or aggregated administrative data on plan selections during open enrollment. We are able to improve on this literature using person-level data on effectuated enrollment through healthcare.gov in 2015 and 2016 from the Center for Medicare and Medicaid Services (CMS). These data allow us to accurately measure exchange take-up and control for variation in the demographic composition of local geographic areas. By reducing measurement error and increasing sample size, our analysis will yield more precise estimators. Our analysis is based on logistic regression models of the take-up of exchange coverage by potential enrollees. Our sample consists of 779 local geographic areas in 2015 and 782 in 2016 stratified by gender, age (20-34, 35-44, 45-54, and 55-64), and income (100/138-250 FPL and 250-400 FPL) resulting in a sample of 24,976 observations. We use CMS’s person level data on effectuated enrollment through healthcare.gov to measure the full-year equivalent number of ‑exchange enrollees. To calculate take-up rates, we divide these totals by estimates of the population of potential enrollees based on the Census Bureau’s Population Estimates Program and the 2011–2015 American Community Survey 5-year public use microdata sample. We define a potential enrollee as a person who is eligible for the premium tax credit, who is not eligible for Medicaid or Medicare, and who does not have employer-sponsored insurance or TRICARE. Additional datasets from the Census Bureau, CMS, the Agency for Healthcare Research and Quality, and the Commonwealth Fund are used to measure important control variables for demographic, insurance market, and policy factors. Our preliminary analysis suggests that many factors affect the odds of exchange enrollment. The odds of enrollment are predicted to be much higher for women than men and to increase with age. The odds of enrollment are predicted to decrease with the share of potential enrollees who are exempt from the individual mandate, who do not speak English very well, who are Hispanic, and who are native to the United States. The size of the non-group market in 2013, the exchange participation of insurers with high market share in the non-group market in 2013, and the number of primary care physicians per capita in a region all increase the predicted odds of enrollment. Potential enrollees with eligibility for high subsidies have much higher predicted odds of enrollment than those with lower subsidies. State restrictions on enrollment assistance substantially lower the predicted odds of enrollment. Finally, several factors that we expected

Nicole Nestoriak [email protected]

Amy Kandilov [email protected]

Conor Ryan [email protected]

Benedic Ippolito [email protected]

Health insurance contracts are often structured with different forms of cost sharing in order to provide consumption smoothing across different health outcomes while limiting moral hazard. These components of health insurance may influence an individual’s behavior in multiple ways. The research on moral hazard has largely focused on the responsiveness of consumers spending on healthcare to the cost of care. Alternatively, most workers with employer-sponsored health insurance (EHI) also have access to workers’ compensation (WC) for medical coverage of injuries and illnesses which may be related to their work. WC medical coverage has no copayment or deductible associated with coverage, but employees need to prove their injury is work-related and they are often limited in their choice of medical care provider. The costs for the employee associated with using EHI differs across plans in their cost-sharing mechanisms and over the course of a year if a deductible is present. Differences in plans and time of year provide variation in the end-of-year and spot price of medical care and can identify whether or not workers vary their utilization of WC dependent upon

Using MarketScan data containing WC and health insurance claims, the analysis first confirms employee sensitivity to future prices of medical care using a difference-in-difference technique. The differences in plan deductibles and employee enrollment months allows for control of seasonal differences in medical utilization while focusing on variation in the end-of-year price of medical care that comes with differing enrollment months similar to analysis done in Aron-Dine, Einav, Finkelstein, and Cullen (2015). This analysis is then replicating using the WC claims information for the same set of employees. The empirical work finds evidence of cost shifting to WC as employees are less likely to use EHI and

Measuring the extent of WC cost-shifting is important for a number of reasons. In addition to providing benefits to employees injured on the job, WC also serves as an incentive for employers who are experience-rated to invest in the safety of their workplaces. Variation in WC claiming due to changing prices of EHI medical care is likely to also impact workplace injuries reported to employers, and therefore measuring the extent of WC cost shifting can better inform

The relationship between minimum wages and health insurance coverage among low-wage workers can be complicated. For employers, an increase in the minimum wage that raises the total wage bill may cause changes in health insurance provision; employers may cease to provide health insurance or increase the premiums paid by workers. This would tend to reduce health insurance coverage among low-wage workers. At the same time, for the worker, an increase in the minimum wage may mean that health insurance becomes more affordable, particularly if the cost of the health insurance is split between the worker and the employer, or between the worker and the government. In our research, we quantify the relationship between minimum wages and health insurance coverage among a national sample of agricultural workers, and we examine how that relationship differs between male and female workers. To test the relationship between minimum wages and various types of health insurance coverage among agricultural workers, we use a Probit model in which we control for worker characteristics that are standard in a Mincerian wage equation. Our model also includes state and year dummies, as well as census region by year effects, worker task and crop dummies and legal status. Our sample of full-time crop workers comes from the confidential version of the National Agricultural Workers Survey (NAWS), from the years 2000 through 2014. The NAWS is the only nationally representative survey of demographic, employment, and health insurance characteristics of hired crop workers. Information is obtained directly from farm workers through face-to-face interviews. Our population includes roughly 30,000 crop workers, 82 percent of which are male. This data is unique in its level of detail on the health insurance coverage for crop workers and their families – for each agricultural worker, and each member of the family, the survey reports whether the individual has health insurance and if it was paid for by the worker, the worker’s employer, the spouse, the spouse’s

We find that, among male agricultural workers, overall health insurance coverage from all sources combined falls significantly as the minimum wage increases; this effect is largely due to a decline in health insurance paid for by the worker’s employer. Among female agricultural workers, overall health insurance coverage from all sources combined does not change significantly. While we find a decline in health insurance coverage paid for by the female worker’s employer alone, this negative effect is offset by an increase in the incidence of health insurance whose cost is shared between the employer (or the spouse’s employer) and the worker’s family. We also find that an increase in the minimum wage leads to an increase in health insurance coverage for the worker’s children; an effect that is mostly driven by an increase in the government health insurance. While state policies increasing the minimum wage are aimed at improving

We use a novel data set from eHealthInsurance.com to estimate the demand for health insurance in the individual market. We use this data to provide the first estimates of health insurance demand in the 36 states that do not manage state-run health insurance exchanges. We find that median own-price elasticities range between -.59 and -1.55. While uninsurance rates in this market are high---45% on average---we find that diversion towards uninsurance are low. The median diversion ratios range between 10.1% to 18.2%. The low diversion ratios imply that the penalty for uninsurance implemented through the Individual Mandate may not have a large impact on the uninsurance rate. We do a partial equilibrium exercise and find that a repeal of the Individual Mandate would lead to a 1.7 percentage point increase the uninsurance rate in the individual market. This represents solely the immediate demand effect, and does not

We utilize credit report data to study how unpaid medical collections and other financial outcomes are affected by fundamental features of health care. Specifically, we document and discuss the size and age distributions of unpaid medical bills on consumer credit profiles, use Medicare eligibility and the implementation of the Affordable Care Act to identify the causal effect of insurance coverage, and investigate whether geographic variation in healthcare prices is predictive of medical debt. We show that Medicare sharply decreases the dollar value of unpaid medical collections held by consumers. Moreover, as the ACA reduces the extensive margin gain in insurance coverage at age 65, the effect of Medicare on medical debt falls with it. In particular, from 2013 to 2015 the increase in insurance coverage at age 65 falls by 46% and the effect of Medicare on medical collections declines by a similar 53%. However, Medicare has relatively small effects on the number of people who have medical bills sent to collections, instead predominantly reducing the size of bills which go unpaid. In part we reconcile this with the fact that most unpaid bills are relatively small or modestly sized – over half are less than $600. Thus, while Medicare may substantially reduce the size of bills facing the formerly uninsured, that amount may still exceed many consumer’s willingness or ability to pay. In turn, we show that Medicare has muted effects on other outcomes on consumer credit profiles. It is particularly notable that, despite very salient increases in insurance coverage at Medicare eligibility, we observe little change in credit scores at age 65. The patterns we

Richard Hirth [email protected]

Sapna Kaul [email protected]

Enacted in 2010 and largely implemented in 2014, the Affordable Care Act (ACA) has expanded health insurance coverage to more than 20 million previously uninsured American adults through the Medicaid expansions implemented by 31 states and health insurance exchanges on which individuals can purchase insurance and received subsidies if their incomes are between 100% and 400% of the Federal Poverty Level. While nearly all Americans with end-stage renal disease (ESRD) have insurance after the onset of this condition, those who are uninsured may not have access to effective care before developing ESRD. We assess whether changes in insurance status prior to the onset of ESRD were accompanied by changes in markers of pre-ESRD care in patients receiving their first dialysis treatment pre-enactment of the ACA versus post-enactment of the ACA. We use data from the CMS Medical Evidence Form 2728 which is completed for all patients at incidence of ESRD. This included 511,424 patients from 10/1/11 to 3/31/16. We estimate Difference-in-Differences (D-in-D) models by age/insurance status at ESRD onset [age <65 with no Medicare (treatment group) vs. age ≥ 66 with Medicare (control group)], and by state Medicaid expansion status for age <65 [States expanding Medicaid (treatment group) vs. states not expanding Medicaid (control group)]. Using linear probability models, we determine the ACA’s impact on insurance status and key markers of pre-ESRD care. Models include state and year fixed effects. The assumption of parallel trends in pre-ACA period is supported using plots of monthly data and the results are robust to using narrower age windows. Among patients age <65 not on Medicare, insurance rates increased from 82.4% pre-ACA to 86.7% in 2014, 89.3% in 2015 and 92.1% in 2016. These patients also experienced increases in pre-ESRD nephrology care, use of home dialysis and use of anemia medication; effects on vascular access were equivocal. Patients in states that expanded Medicaid had 3.4 percentage point greater increases in insurance coverage than those in non-expansion states. They also experienced greater increases in pre-ESRD nephrology care and better vascular access, but no change in the use of home dialysis. Overall, these findings demonstrate that several key indicators of pre-ESRD care improved post-ACA implementation, but which indicators improved differed somewhat across the two affected populations: (A) all adults younger than 65 and not on Medicare and (B) adults younger than 65 in Medicaid expansion states. Our findings are consistent with improved access to care that has significant potential to improve clinical outcomes for those with advanced kidney disease in the United States. The magnitudes of the effects are relatively large. Overall, uninsurance fell dramatically during the post-ACA period. About 25-33% of the patients who gained insurance also gained relevant pre-ESRD care, demonstrating the important role of health insurance coverage in access to pre-ESRD care. These are meaningful changes in the context of a clinically and economically vulnerable population. Further research can establish the extent to which the observed improvements in pre-ESRD care affect post-ESRD outcomes such as

Lack of access to both health insurance coverage and usual source of care (USC) can be a potential barrier for appropriate use of health care services. For special disease groups, such as adult cancer survivors in the United States, both insurance and USC could be critical for improved access and health outcomes. Yet, no study has evaluated how insurance and USC interact and affect adult cancer survivors’ health care barriers in the U.S.

We used the 2012 to 2015 Medical Expenditures Panel Survey (MEPS) data to identify N=4,009 adult cancer survivors, currently aged 18-64 years, who reported ever being diagnosed with cancer. We used one-to-one propensity score matching to identify 4,009 comparison group individuals without a history of cancer. Patients’ barrier was categorized as “1” if they needed a treatment but were unable/delayed in receiving it, “0” otherwise. USC was categorized as “1” if individuals stated that there was a particular doctor’s office, clinic, health center, or other place that they usually went to if sick or needed advice on health, and “0” otherwise. Insurance status was categorized as private, public, or uninsured. Other covariates included demographics (e.g., age, sex, race and ethnicity, marital status etc.) and self-perceived health (physical and mental health status). Chi-squared tests were used to compare proportions

Cancer survivors were less likely to be uninsured (8.6% vs. 11.7%, p<0.001) and more likely to have a USC (86.9% vs. 82.1%) than the comparison group. Survivors who had no access to a USC were more likely to be uninsured (21.6% vs. 6.8%) or have public insurance (18.6% vs. 16.4%) than those who had a USC (p<0.001). About 10% of cancer survivors reported needing care but being unable/delayed in receiving treatment vs. 6% of the comparison group (p<0.001). In our pooled adjusted model, survivors reported this barrier more often than the comparison group (OR=1.47, 95% CI: 1.20-1.80) and uninsured (3.40, 95% CI: 2.40-4.83) and publicly insured individuals (1.69, 95% CI: 1.23-2.34) reported this barrier more often than privately insured; however, USC had no significant influence on barrier to care. Further, among cancer survivors, compared with those who had either private insurance or USC, survivors without

For adult cancer survivors, insurance coverage is protective of barriers to required care, and also for having access to USC. More emphasis should be placed on providing adequate insurance coverage to adult cancer survivors

Megan Cole [email protected]

Sonal Parasrampuria [email protected]

Ravi Singh [email protected]

: Community Health Centers (CHCs) provide access to primary care services to 24 million patients annually, nearly all of whom are low-income and uninsured or publicly insured. While CHCs provide access to preventive services, chronic disease management services, and some behavioral services, for instance, specialty care is often obtained through referral, including that for behavioral health, and prescription drugs may be unaffordable to uninsured patients. Ensuring access to primary care is essential, as supported by $11 billion in ACA funding to expand CHC capacity. However, to the extent that primary care providers refer patients outside of the CHC, recommend follow-up care, or write prescriptions not provided at a discounted price, if patients cannot access these downstream services because of lack of insurance, then increased access to primary care alone will do little to impact health outcomes for these

: To estimate the effect of having health insurance coverage on access to necessary medical care, specialty care, behavioral health care, recommended follow-up care, and medications for patients served by community health

: We used a nationally representative sample of 5,040 non-elderly adult CHC patients from the 2014 HRSA Health Center Patient Survey, representing 13.9 million patients. We examined 19 patient-reported outcomes related to access to and delayed access to medical care (any), specialty care, behavioral health care, follow-up care after abnormal cancer screenings, any medications, and medications for hypertensive, asthmatic, diabetic, and hyperlipidemic patients. For each outcome, we calculated inverse probability of treatment weights (IPTWs) based on propensity scores to estimate average treatment effects, where patients with insurance were considered treated. Propensity scores included 20 patient-level sociodemographic and clinical covariates. Weights were stabilized to a mean of one and truncated at the 99th percentile. We used logistic regression models with IPTWs to estimate the effect of having health

: In 2014, having health insurance coverage was associated with better access to most types of care examined. For instance, compared to statistically similar health center patients without insurance, patients with insurance coverage were more likely to have access to necessary medical care (aOR=2.12, 95%CI 1.74-2.58); to see a recommended specialist (aOR=2.73, 95%CI 2.15-3.46); to see a mental health professional if advised (aOR=1.74, 95%CI 1.31-2.32); to receive recommended follow-up care after an abnormal pap (aOR=3.44, 95%CI 1.80-6.54); and to get necessary prescription medications (aOR=2.10, 95%CI 1.75-2.53), particularly for patients with high cholesterol (aOR=2.25, 95%CI

: Results highlight the vital role of health insurance in accessing care within the safety-net, particularly for non-primary care services. This is especially important in light of potential reversals to Medicaid expansion, as health centers may not be able to fully compensate for resulting losses in patient insurance coverage. Furthermore, expanding safety-net capacity to provide non-primary care services for uninsured patients remains critical.

Ten years after San Francisco passed an employer mandate, we assess the effect of the law in increasing rates of health insurance and insurance generosity, and the health effects for chronic disease. Specifically, we focus on the immediate and long-term effects on cancer, a condition where early detection and treatment can significantly improve survival and thus access to health insurance is an important determinant of outcomes. We conduct difference-in-difference analyses, comparing San Francisco with three controls groups, the surrounding counties, similar counties in Great California, and another large metropolis, Los Angeles. We use the California Health Interview Survey to establish the increase in insurance rates in San Francisco relative to the control groups. We also test for greater insurance generosity and higher rates of continuous coverage as a result of the mandate. We then use the

We find that as a result of the employer mandate, the likelihood of late stage diagnosis for cancer patients was approximately 1% lower in San Francisco, a small, but statistically significant result. Survival outcomes saw much larger changes. The greatest effects occurred for lymphoma and leukemia; implementation of the employer mandate resulted in a 4% increase in one year survival rates and a 2% increase in two-year survival rates. Comparing the pre and post mandate period, San Francisco saw a 7% relative increase in survival rates. The results were similar, but smaller in magnitude for the two most common types of cancer, lung and breast. For lung cancer, the employer mandate resulted in a 3% increase and 1% increase in one and year survival rates, respectively. These numbers were 0.5% and 1% respectively for breast cancer. Breast cancer survival rates are likely attenuated by initially high survival rates; mortality rates are

This study is the first to investigate the effect of newly gained health insurance from an employer mandate on health outcomes. Employer mandates are an important policy mechanism, because they have the largest impact on individuals right above the poverty line, who are not eligible for Medicaid, but often do not have health insurance through their employer or have inadequate insurance. We show that the SF employer mandate resulted in higher rates of insurance coverage which results in the earlier detection of cancer and significantly increased rates of survival, especially among vulnerable populations. This is a positive signal for the long-term health effects in addressing the health burden of

d: Regional variations in health care utilization and quality are well documented. Previous studies have identified a number of factors leading to differences in access, including a shortage of healthcare professionals and adequate health insurance. Research examining differences in rural and urban health care utilization suggests that people living in rural areas may utilize health care less often. In 2013, before the Affordable Care Act (ACA), the insured rate in California was 83%. The ACA led to increased health insurance coverage for residents and funding for many healthcare providers, and in 2016 the insured rate increased to 93%. The purpose of this study is to examine regional variation

: Hospital data from the Office of Statewide Health Planning and Development (OSHPD) for the year 2013 were compared across counties using the patient’s zip code to identify the county of residence. The rurality (rural, urban, or frontier) was determined using OSHPD definition. Prevalence estimates of the number of people with diabetes in the county were obtained using the California Health Interview Survey. Data on the population was obtained from Census

: Overall, there were 2,505,042 people diagnosed with diabetes in California, and a total of 54,526 hospital and 72,776 emergency department admissions, with 149 hospital admissions and 199 emergency department admissions per 10,000 people with diabetes. Regional variations in admission rates and length of stay were found to exist for diabetes admissions, both overall and after controlling for patient characteristics including age, gender, insurance type, ethnicity, and comorbidities. In both hospital admissions and emergency department visits, urban counties were found to have lower rates compared to rural counties. When compared with urban areas, people in frontier areas were

: Differences in rural and urban areas regarding admissions and length of stay were found to exist. This suggests disparities in insurance coverage and utilization exist. A follow up study examining 2016 hospital data would show

Martin Karlsson [email protected]

market but so far the evidence on its effects have been scarce. We study how such refunds impacts individual utilization and claiming behavior using rich administrative claims data from a large German health insurance company and an insurer policy that unexpectedly increased the refund size of certain plans. We furthermore suggest a novel method to decompose the overall effect on claims into an intensive, extensive and an automatic component. Our findings show that individuals reacted strongly to the changed incentives by reducing their claims on both the extensive and the intensive margin. We argue that this reaction is evidence of forward-looking behavior, but also show that it seems irrational in many cases, since also individuals with predictably high expenditures cut down on their utilization. Since the policy we consider was abolished again, we can also use claims from later years to study persistent effects of the policy. This


Rollins College Complete

University of Michigan School of Public Health Jeffrey Kullgren; A. Mark Fendrick; Richard Hirth Complete

University at Albany, SUNY Baris Yoruk Complete

Bowling Green State University Complete

Workers Compensation Research Institute Jonathan Gruber Complete

Mathematica Policy Research Nancy Early; Michael Levere; Lindsey Leininger Complete

University of Oxford Complete

University of North Carolina at Chapel Hill David Anderson Complete

Cornell University Complete

Thomas DeLeire Complete

Congressional Budget Office Alexandra Minicozzi; Jessica Banthin Complete


Bureau of Labor Statistics Complete

RTI International Ivan Kandilov Complete

University of Minnesota Stephen Parente; Roger Feldman Complete

American Enterprise Institute Complete

University of Michigan Complete

University of Texas Medical Branch Yong-fang Kuo; Ana Rodriguez; Jaqueline Avila Complete

Diane Steffick; Rajiv Saran; John Ayanian; William Herman; David Hutton; Jillian Schrager; Jeffrey Pearson

Boston University School of Public Health Complete

Lauren Nicholas Complete

University of California, Merced Complete

Johns Hopkins Bloomberg School of Public health

University of Duisburg-Essen Complete


Health Systems, Health Reform and Health Care Financing

The Impact of a Merit-Based Incentive Payment System on Quality of Healthcare: A Framed Field Experiment


(Expected) Value-Based Payment: From Total Cost of Care to Net Present Value of Care


Recent Changes over Time in Disparities in Late Stage Breast Cancer Diagnoses among Younger Women in the US


The impact of Medicaid global budgets on hospital care: Evidence from Oregon


Heterogeneous effect of the hospital financing reform on productivity: panel data quantile regressions with endogeneity


The Persistence of Medicare Advantage Spillovers in the Post-Affordable Care Act Era


The burden of pandemic influenza on secondary care: Costs of H1N1 hospital admissions in England


Effects of Major Joint Replacement Bundled Payments on Medicare and Commercial Post-Discharge Spending


Does Health Information Exchange Improve Patient Outcomes? A Longitudinal Study


The Impact of Federal Public Health Funding on Sexually Transmitted Disease Outcomes


The role of health insurance, race/ethnicity, and income in total medical expenditure for asthma care


Reconciling Medical Expenditure Estimates from the MEPS and the NHEA, 2012


The Impact of Organizational Change on Firm efficiency: Evidence from the Healthcare Sector


Long-run Health and Mortality Effects of Exposure to Universal Healthcare at Birth


Assessing the Effects of the Affordable Care Act’s Marketplaces on Adults with Chronic Conditions


The Impact of Mercy Health Center on Healthcare Utilization and Costs


Payments to Medicare Advantage Plans and Plan Generosity Before and After the Affordable Care Act


Effectiveness of the Community Health Assist Scheme (CHAS) in Reducing Market Failure in Singapore’s Healthcare Sector


Effect of the Community Health Assist Scheme (CHAS) subsidy on income inequality in Singapore


Medicaid Expansion and Changes in the Health Care Workforce


Improving the Allocation of Resources with Modified Data and the Health Plan Payment System


Emergency Department Utilization by Adolescents and Young Adults with Cancer Compared with Children and Non-Geriatric Adults with Cancer: A Pre and Post Affordable Care Act Analysis


Public Transportation and Elderly Access to Healthcare Services – Evidence from a Reform within Israel’s Arab Communities


Waiting Time and Quality of Patient Care: Consequences of a Medical Expense Transaction Reform in China


Examining predictive modeling based approaches to characterizing healthcare fraud


The effect of Medicare Advantage on beneficiaries who are dually enrolled in Medicare and Medicaid.


Two Novel Findings on the Determinants of Medicaid Churn: Perceived Loss of Coverage after Child Birth and Delayed Effects of Changes in Employment


Does Getting Lean Affect Quality? – Hospital Cost Containment Impact on Quality under Value-Based Purchasing Program


Extension of the Extensive Margin for Coronary Revascularization in Medicare Beneficiaries: Implications for Repeat Procedures and Non-Institutionalized Days


The Increasing Progressivity of Healthcare Financing in the United States: 2004 to 2015


The relationship between Medicare Advantage star ratings and beneficiary health outcomes.


Enhanced Access to Primary Care and Preventable Hospitalizations: Evidence from Statewide Medicaid Managed Care in Florida


Program Outcomes Associated with Medicare's Value-Based Payment Modifier: A Regression-Discontinuity Approach

Abstract

We study the impact of a merit-based incentive payment system on provider behavior in the primary care setting using experimental methods that leverage healthcare simulations with patient actors. Our approach allows us to exogenously change a provider’s incentives and to directly measure the consequences of alternative payment systems. Within our sample, we find that merit-based incentive payment systems increase the number of the incentivized outcome measures met, but lower overall quality of care through unintended effects on other indicators of care.

Healthcare in the United States is undergoing a transition from volume to value. Current value-based payment models incentivize providers to efficiently provide high-quality care and generate savings in the short-term. The next wave of models will need to incentivize providers to promote health across the life-course, and reward efforts that generate both short- and long-term savings. Healthcare payment models can take a life-course perspective on value by tying incentives to reductions in long-term actuarial risk, such that an intervention is valuable to the extent it reduces the payer’s predicted future costs. This actuarial risk-based valuation builds in the business case for how increased investment in health promotion can still be cost-neutral to the healthcare system overall. Specifically, as long as the total amount invested does not exceed the expected value of the reduced actuarial risk, the payer can be relatively confident that the intervention will be at least cost-neutral over the long-term. This approach also intrinsically risk adjusts, as larger investments are justified where there is more risk to reduce. In addition, this approach can avert incentivizing perverse outcomes by assigning monetary value to avoiding serious adverse events that may not technically cost a payer very much. Payers can integrate indicators of future reductions in actuarial risk into existing payment models by using “net present value of care” (NPVoC) rather than total cost of care in determining incentives (see Figure 1). NPVoC is the total cost of care (i.e., the lesser than anticipated amount spent on healthcare), plus the expected value of care (i.e., the amount the payer anticipates that the achieved health outcomes will save in later healthcare costs, divided by a discount rate). The expected value of care could differ from payer to payer, based on how long individuals are anticipated to remain with the plan. In instances of high market penetration, it may be five, ten, or even more years. As Medicare, Medicaid and large provider organizations initiate multi-payer arrangements (i.e., multiple health plans agree to similar and mutually beneficial payment terms) the duration of expected value calculations will lengthen, because even when individuals move amongst the plans, they will all reap the collective benefits. At present, payers have limited ability to accurately predict long-term reductions in actuarial risk from short-term clinical outcomes—an undeniably challenging endeavor. To begin testing and improving NPVoC models, substantial precision is unnecessary. Payers can use a conservative estimate of predicted savings associated with a modest magnitude of change on a short-term outcome, minimizing potential overall losses when predicted savings are shared in expected value-based payments. By allowing payers to calculate the expected value of an outcome, NPVoC models can avoid problems from the past where research indicated likely savings that failed to materialize when the payer took the intervention to scale. Precise estimates of future savings for NPVoC models may take years of data collection and integration, but initial models can be tested and iteratively improved to build the foundation.

Recent Changes over Time in Disparities in Late Stage Breast Cancer Diagnoses among Younger Women in the US Srimoyee Bose, Lee R Mobley

About 44% of young women with Breast cancer (BC) are diagnosed with advanced stage, and it is rising at a faster rate than for older women. This study focuses on examining whether there were significant changes on the predictors of late-stage BC diagnosis rate among young women in the US across the two periods i.e. pre (2004-2009) and post (2010-2014) Affordable Care Act implementation. Using the US cancer statistics registry database, we extracted all 139220 BC cases for young women (<=40 years) from 46 states to examine the variation in their late-stage BC diagnosis across the two periods. We used a random intercept logit model with person, county, state and time level covariates after controlling for factors that may moderate the effect of the ACA implementation. Results suggest that young African-American women had higher odds of late-stage diagnosis relative to the whites, and the odds increased over time. Area urbanicity, poverty, unemployment, obesity rate and Medicaid enrollment rate were associated with increased odds of late-stage diagnosis, and the odds were higher in the later period. Also, per-capita healthcare expenses, a residential diversity index, the average age of mothers at first birth, drinking and marital status rates, BC screening rate and private insurance enrollment rate were associated with lower odds of late-stage diagnosis, which increased in the later period. Thus, disparities in late-stage diagnoses of BC among young women increased after 2010, highlighting the increased importance of having better access to medical care and prevention among younger women following implementation of the ACA in 2010. As time passes, further investigation is needed to understand these persistent disparities and whether health care reform will eventually seem to reduce them.

Keywords: disparities, prediction, multilevel, cancer, young women, Affordable Care Act (ACA)

Background: In 2012 Oregon transitioned its Medicaid program to cover 90% of its enrollees in Coordinated Care Organizations (CCOs). CCOs can be seen as a type of Medicaid Accountable Care Organization, but include an administrative layer (similar to a Managed Care Organization) and are at full financial risk through a global budget. As part of its 1115 Medicaid Waiver, Oregon agreed to reduce its historical rate of health care spending growth from 5.4% to 3.4%. The global budget and prespecified rate of growth have analogies to per capita caps and block grants, features that were prominent in several “repeal and replace” bills that were put forth in 2017. We provide an assessment of the Oregon model on inpatient admissions, a high cost service area. Study Design. We estimated changes in inpatient care using a difference-in-differences approach and data from the Healthcare Cost and Utilization Project (HCUP) and monthly and county Medicaid enrollment data from Oregon and comparison states of Washington, Colorado, New Mexico, and Arizona. We used 1-1 nearest neighbor matching to identify similar county and demographic matches across these comparison states. Using 2 years of pre-intervention data (2010-2011) and 2 years of post-intervention data (2013-2014), we assessed the impact of the CCO model on inpatient admissions and length of stay. In subanalyses, we also assessed changes in hospitalization rates for elective vs. emergency admissions; ambulatory-care sensitive admissions, and admissions among individuals from low-income neighborhoods. Finally, since the CCO model provided care for more than one out of every four Oregonians, we tested for spillover effects among the commercially insured. We assessed parallel trends among the treated and comparison groups and found no statistically significant differences across multiple groups, including all admissions, length of stay, and emergency and non-emergency admissions. Results. Oregon’s transition to global budgets was associated with significant reductions in the rate of inpatient admissions of approximately 1 admission per every 1000 enrollees per month, equivalent to a 9% reduction in the admission rate. There were significant reductions in both the emergent and elective rate of admissions, although the changes were larger among elective admissions, and, within that group, particularly among admissions for births and maternity care. The length of stay among Oregon Medicaid enrollees was low prior to the CCO transformation and remained low in the two years following the intervention, suggesting that changes in the extensive margin were not offset by changes in the intensive margin. In subanalyses, we found significant reductions in ambulatory care sensitive admissions and admissions among individuals from low-income neighborhoods. We found no evidence of spillover in the commercial market. Conclusions. Oregon’s global budgets were associated with significant reductions in Medicaid inpatient admissions relative to matched counties in western states. The extent to which the Oregon model can continue to place pressure on inpatient utilization will depend on the longer-term success of infrastructure investments and a portfolio of delivery system changes that are designed to provide care in less intensive settings. These changes may have lessons for Medicaid reforms in other states.

The impact of policy regulation in the health sector depends on hospital technology, which is largely reflected in the productivity of labor, capital and medicines. This paper focuses on acute-care local public hospitals in Japan and examines how technology differences determined the effect of voluntary changeover by hospitals to the prospective payment system. The Japanese hospital sector provides a rare example of a nationwide introduction of the reform with self-selection. We exploit panel data conditional quantile regressions to model a range of technologies for the multi-product output function of hospitals under an endogenous treatment assignment. The analysis reveals technological heterogeneity, and incorrect labor/capital and labor/medicines mix. The impact of prospective payment on output is primarily attributed to labor and is inversely related to productivity. Finally, we contrast the design of prospective payment reforms and labor changes in the U.S. and Japan. The novelty of the present paper is severalfold. To the best of our knowledge, the paper is the first application in health economics of quantile regression models for estimating the longitudinal production function of hospitals. The analysis incorporates multiple outputs, evaluates factor returns and assesses optimality of input mix across quantiles. Secondly, the paper proposes an approach to account for groupwise serial correlation in the estimates of pooled models of conditional quantile regressions under endogeneity. Thirdly, the paper is the first health economics study to assess the heterogeneous treatment effect of a prospective payment system on hospital production. Finally, the paper is the first analysis to model self-selection in participation by hospitals in this financing reform. The paper is unique in using the data on Japanese hospitals from several sources: the longitudinal financial data for municipal and prefectural public hospitals accumulated by the Ministry of Internal Affairs and Communications; databases on hospitals which introduced a prospective payment system and on designated local hospitals, kept by the Ministry of Health, Labor and Welfare; and the database of the Japan Residency Matching Program on teaching hospitals. We discover that a hospital's voluntary decision to introduce prospective payment may be motivated by teaching status, average tenure of doctors and size (floor area) of hospital. The results of our statistical tests show technological heterogeneity at Japanese hospitals. The technology differences are revealed in different productivity and the technical rate of substitution between hospital inputs (labor, capital and medicines). In particular, we discover lower labor returns at high-output hospitals. The findings reveal that the reform has the biggest effect at low and medium-output Japanese hospitals, while it might be less significant for the most productive hospitals. The impact of prospective payment on output is primarily attributed to labor. The technology distinctions may explain the heterogeneous effect of the reform. We believe that considerations of technological heterogeneity and its link with outcomes of the hospital financing reform would offer helpful guidance for policy measures. Moreover, knowledge of the determinants of self-selection could serve as a useful tool for modification of the Japanese prospective payment system.

Spillovers can arise in markets with multiple purchasers relying on shared producers. If producers are constrained in their ability to adjust quality and cost across purchasers, then the influence of a dominant purchaser affects the entire market. Prior studies have found such spillovers in health care, from managed care to non-managed care populations — reducing spending, utilization, and improving outcomes. Similar effects have been identified in the Medicare Advantage market as well, with studies finding declines in utilization and reductions in resource use among the Traditional Medicare population associated with increases in county-level Medicare Advantage penetration. However, no study to date has provided plausibly causal estimates of such spillovers in the post-Affordable Care Act era. Our study does so by exploiting idiosyncratic differences in payments to Medicare Advantage plans that are unrelated to traditional Medicare spending. Further controlling for health status and other potential confounders, we estimate that a one percentage point increase in county-level Medicare Advantage penetration results in a $146 (1.7%) reduction in standardized per enrollee Traditional Medicare spending. We find evidence for reductions in utilization both on the intensive and extensive margins (including reductions in the number of inpatient stays) and across many types of health care services including: home health, inpatient, and imaging, not all of which have been analyzed in prior Medicare Advantage spillover studies. Our results suggest that spillovers from Medicare Advantage to Traditional Medicare have persisted in the post-Affordable Care Act era.

Influenza pandemics can place considerable burden on affected health systems by straining hospitals’ capacities due to surges in the number and cost of inpatient admissions. Substantial variations exist in estimates of hospital admissions and their costs due to the 2009/10 influenza A/H1N1 pandemic. Previous studies assessing the impact of pandemics on hospitalizations have either analyzed the subgroup of laboratory-confirmed H1N1 patients, thus underestimating the pandemic cases, or they included ordinary seasonal influenza in their analysis, thus overestimating the pandemic cases. The objective of our study is to provide robust estimates of the overall and age-specific weekly H1N1 admissions and costs between June 2009 and March 2011 in 170 English hospitals. We use routine hospital administrative records of all patients admitted for influenza-like illnesses. Since our data does not allow us to distinguish seasonal from pandemic influenza cases, we use time series models and pre-pandemic (2004-2008) admission data to establish a counterfactual of expected weekly seasonal influenza admissions over the pandemic period. We calculate the weekly number and cost of H1N1 admissions as the difference between our data and the counterfactual estimates. We find that there were two distinct waves of pandemic admissions. The first wave coincided roughly with the official pandemic period, with 10,348 excess admissions and £20.5 million secondary care costs between June 2009 and March 2010. The second wave occurred after the pandemic had been declared over by the World Health Organization. Although it was much shorter (November 2010 – March 2011), there were more admissions compared to the first wave – 11,775 – costing £24.8 million. Patients aged 0-4 years had the highest H1N1 admission rate, and patients aged 25-44 and 65+ years had the highest costs. Our estimates are over 4 times higher than those formerly reported, suggesting that the pandemic’s burden on secondary care has been previously underestimated. Our findings support improvements in pandemic preparedness and demonstrate the value of surveillance using data on routine hospital admissions as a possible tool. These results can help hospitals manage unexpected surges in admissions and resource use.

Objective: Medicare’s voluntary bundled payments program has been heralded as a potential solution to align incentives and curb rising health care expenditures. Interest in acute care based bundled payments for major lower extremity joint replacement has been particularly high, with multiple cohorts of providers enrolling from October 2013 through October 2015. Early studies of the program show that bundled payments resulted in provider behavior change, leading to decreased spending for inpatient post-acute care. This paper augments previous studies by examining whether spending reductions persist over time, and whether behavior changes by bundled payment providers were specific to the treatment of Medicare patients or were more broadly implemented, spilling over onto commercial patients. Methods: We examined 90-day, post-discharge spending for hospital-based bundled payments for Medicare and commercially insured patients treated for major lower extremity joint replacement from January 2012 through March 2016. We used claims data from the Michigan Value Collaborative, a statewide, 76-hospital consortium. We compared changes in spending by patients admitted to bundled payment providers, compared to patients admitted to non-participating control providers, using difference-in-differences analyses. Our treatment group consisted of 4,687 episodes from five early entrants (January 2014 provider enrollees) and 7,569 episodes from five late entrants (April 2015 provider enrollees). First, we examined dynamic effects of bundled payments on Medicare spending over a period of 27 months for early entrants. Second, we compared changes in Medicare spending between early and late entrants to determine whether there were heterogeneous cohort effects. Third, we assessed whether bundled payments led to spillover effects on spending. Results: In difference-in-differences comparisons against episodes in the baseline period from 2012 through 2013, we found no statistically significant differences in average 90-day post-discharge Medicare spending in early entrant hospitals during the first year of bundled payments, but statistically significant but non-persistent decreases in 2015. Medicare spending in early entrant hospitals in the first quarter of 2015 declined by $887 (SE: $446; p<0.10) more than control episodes in the same periods, and declined by $1,270 (SE: $608; p<0.05) in the subsequent six-month period. We observed a positive but imprecisely estimated $298 increase (SE: $1,176) by late 2015. We found no evidence of heterogeneous effects on Medicare spending between early and late entrants (F-statistic, 0.04; p=0.85), nor consistent difference-in-differences estimates indicative of spillovers between Medicare and commercially insured patients. Conclusions: Participation in bundled payments led to short-lived reductions in post-discharge spending. The effects of bundled payments were mitigated as non-bundled payments hospitals appeared to adopt alternative cost containment efforts and spending reached equally low levels, making it more difficult for bundled payments providers to achieve differential savings in later years. Finally, our findings suggest that behavior changes from bundled payments providers were isolated to Medicare beneficiaries and did not lead to broad-based practice change. Ultimately, the future success of bundled payments will hinge on its ability to incentivize continuous improvement and its role in transforming practice culture among providers.

The 2009 Health Information Technology for Economic and Clinical Health (HITECH) Act devoted $28 billion through its “meaningful use” incentive program to providers that adopted Electronic Health Records (EHRs) and participated in Health Information Exchange (HIE). Underlying these substantial investments is a belief that timely transfer of standardized electronic health information such as laboratory results and clinical summaries across the care continuum can facilitate coordinated patient care and improve health outcomes. Despite the recent growth of HIE and its potential benefits, only a few studies have examined the impact on quality of care in inpatient settings, and, to the best of our knowledge, no US-based studies have reported on HIE impacts on the tradeoff between readmission and a variety of quality measures. We conducted a large scale retrospective study to examine the impact of HIE engagement on individual patients’ outcomes of Acute Myocardial Infarction (AMI) that is directly targeted by the Hospital Readmissions Reduction Program and a variety of other Centers for Medicare and Medicaid Services (CMS) programs, including the Hospital Value-Based Purchasing Program, and the Bundled Payments for Care Improvement (BPCI) Initiative. We linked the Florida State Inpatient Discharge (SID) data, which allows us to track a patient’s longitudinal visits across hospitals, with the American Hospital Association Annual and Information Technology Supplement surveys. Using a difference-in-differences (DID) estimation approach, we compared changes in outcomes of a treatment group of targeted admissions before and after HIE engagement, relative to changes in outcomes of a control group that never participated in HIE. Our main outcome measures are the 30-day, 45-day, and 60-day all-cause readmission rates. To investigate whether the changes in readmission came at the cost of other quality measures, we also analyzed the impacts on length of stay, total charges, total number of procedures, discharges to a nursing facility or home health care, and in-hospital mortality. Our models adjust for patient characteristics and include hospital specific fixed effects to control for unobservable confounding factors. We employed placebo tests to rule out the concern that changes in outcome measures may have already started in time periods prior to the participation of HIE. Overall, we found that HIE engagement did lower 30-day all-cause readmission rates for AMI patients. The decrease in readmissions after HIE engagement primarily came from reduced readmission to a different hospital. In addition, associated with the reduction in readmission were the rises in length of stay, number of procedures, and total charges, but there were no statistically significant changes in transfer, discharge destination or in-hospital mortality. These results suggest that the decrease in readmission was achieved through the increased treatment intensity of inpatient care, but was not due to any strategic transfers or changes in discharge destination. HIE may have played an important role in determining the optimal cost tradeoff between inpatient care and readmission. A back-of-the-envelope calculation reveals that, for AMI condition alone, the HIE participation in Florida hospitals reduced 235 avoidable readmissions and saved $2,964,290 in cost annually.

Federal public health funding is a major component of sexually transmitted disease (STD) prevention efforts in the United States. The Division of STD Prevention (DSTDP) at the Centers for Disease Control and Prevention allocates annual funds to 57 project areas across the country (50 states plus 7 distinct metropolitan areas). This funding helps achieve various division goals, which include reducing the incidence of STDs in the United States and decreasing health disparities in STDs. This paper estimates the impact of changes in funding on both of these outcomes from 1981 to 2016. Variation in project area funding comes from changes in the overall budget of DSTDP, as well as changes in how the division distributes funds among recipients. Between 2013 and 2014, funding became more targeted to areas with historically higher STD burdens. One empirical concern when estimating the impact of funding on incidence is that more funds will be targeted at areas with relatively larger expected increases in STD rates. To account for this, DSTDP rules governing funding allocations are used to develop an instrumental variable that strongly predicts funding levels but is unrelated to contemporaneous changes in reported STD rates. Specifically, DSTDP requires funding reductions to no exceed 5 percent annually and sets limits on the overall levels of gains and reductions. This induces variation in funding allocation that is based on historic funding rates and unrelated to current changes in incidence. Project areas may use funds to increase STD screening efforts, which may lead to more cases being detected even if true incidence remains unchanged. To account for this possibility, the paper focuses on gonorrhea in males, which is less likely to be asymptomatic than in females or for other common bacterial STDs. Because of this, men with gonorrhea generally seek out treatment, so identification of new cases is less dependent on STD screening efforts in their area. Preliminary results find that a one percent increase in funding decreases male gonorrhea rates by 0.96 percent. Further, findings estimate that the change in distribution of funds starting in 2014 led to 7.5 percent fewer reported cases over 2014-2016 than if the funding allocation had remained unchanged. Finally, the impact of funding on racial disparities in gonorrhea outcomes depends on whether an absolute or relative disparity measure is used. Findings suggest that targeting public health funds may lead to improved efficiency, and care must be taken when interpreting changes in health disparity.

Rationale: Asthma is a chronic disease that affects quality of life, productivity at work and school, and healthcare utilization; although controllable, it can even result in death. To control asthma symptoms and prevent severe asthma attacks, the evidence-based guidelines developed by medical professionals should be followed. These recommendations include assessment of asthma severity, prescribing and ensuring adherence to asthma control medications, providing asthma self-management education, and identifying and avoiding ambient and indoor environmental triggers. In this paper, we estimated the effect of health insurance status, race/ethnicity, and income on asthma-related incremental medical expenditure. Methods: The primary source of data was the 2008-2013 household component of the Medical Expenditure Panel Survey. We defined treated asthma as the presence of at least one medical or pharmaceutical encounter or claim associated with asthma. For the main analysis, we applied two-part regression models to estimate asthma-related annual per-person incremental medical expenditure (APIME). Routine outpatient care was defined as scheduled nonemergency physician office visits or hospital outpatient visits. Results: During 2008-2013 the national average of APIME in the United States was $3,266 (in 2015 US dollars); more than 80% of that amount was attributable to prescription medication and routine outpatient care, while roughly 20% was attributable to hospitalizations and emergency room (ER) visits. APIME was significantly lower than the national average for uninsured persons, Blacks, Hispanics, and persons whose income was equal to or above the national poverty level. Conversely, APIME was higher than the national average for insured persons, Whites, Asians, and for those whose income was below the national poverty level. Studies show that use of ERs and hospitalization services, previously a major driver of high total medical expenditures for asthma, occurs more often among uninsured persons, Blacks, and Hispanics. Our study shows that prescription medications and routine outpatient care are comprising an increasingly large proportion of total asthma care expenditure. Uninsured persons may tend to seek care through use of the ER or hospitalization services rather than through routine outpatient care and filling prescription medications, which lowers overall total medical expenditure for asthma care. Our results also show that persons with income below the national poverty level have higher APIME than persons in higher income brackets. These persons are more likely to live in areas with a higher concentration of outdoor and indoor environmental asthma triggers, which are a major cause of asthma attacks. On the other hand, these individuals are more likely to qualify for Medicaid, which may facilitate access to a wider range of routine and urgent medical services, contributing to higher total medical cost for asthma care. Conclusion: Lack of health insurance hinders access to prescription medications and routine outpatient care and, as a result, contributes to higher use of ER visits and hospitalizations for persons with asthma. Persons with income lower than the national poverty level have higher APIME and, despite being potentially eligible for Medicaid, may need additional financial support, such as health insurance reimbursement for environmental interventions, to maintain an indoor environment free of asthma triggers.

In this study, we compare and align health care expenditure estimates from the Medical Expenditure Panel Survey and the National Health Expenditure Accounts. Reconciling MEPS and NHEA estimates serves two important purposes. First, it is an important quality assurance exercise for improving and ensuring the integrity of estimates from both sources. Second, the reconciliation provides a consistent baseline of health expenditure data for policy simulations. MEPS is often used in developing microsimulation models because it contains person-level expenditures. Reliable estimates of national health spending need to be used as a baseline for analyzing the impact of potential policy changes on health care costs. Based on results from our study, analysts can adjust MEPS to be consistent with the NHEA so that the projected costs as well as budgetary and tax implications of any policy change are consistent with national health spending estimates. Previous reconciliations have been used as the baseline by the Department of Health and Human Services, the Congressional Budget Office, RAND and other researchers in simulating the impact of potential policy changes. The NHEA and MEPS both provide comprehensive estimates of health care spending in the U.S. The NHEA is primarily based on aggregate provider revenue data and administrative records of publically administered programs and covers the entire U.S. population and a full range of health care expenditures, including personal health care spending, public health services, research, and investment in structures and equipment. NHEA estimates are produced annually in the U.S. by the Office of the Actuary at the Centers of Medicare and Medicaid Services (CMS). MEPS, on the other hand, provides person-level information on health expenditures from a nationally representative sample of households in the civilian, non-institutionalized population. MEPS is produced by the Agency for Health Care Research and Quality (AHRQ) and the National Center for Health Statistics. The reconciliation is conducted every five years when the quinquennial Economic Census is available as it is the only data source that contains expenditures at the required level of detail so that specific expenditures reported in different service categories in NHEA and MEPS can be aligned. . The previous reconciliations were conducted for 1996, 2002 and 2007 (Selden, Levit and Cohen et al. 2001, Sing, Banthin and Selden et al. 2006 and Bernard, Cowan and Selden et al. 2013). Although each source provides a measure of total national spending on personal health care (PHC), at first glance the estimates appear to diverge significantly. We make adjustments to account for the differences in underlying populations, covered services and other measurement concepts to reconcile the expenditure estimates. Once we adjust the NHEA to make it consistent with MEPS, we compare and discuss potential reasons for the differences for each service category and source of payment. We also discuss how the expenditure estimates have changed since the previous reconciliation in 2007. Identifying service types and sources of payment with larger gaps helps AHRQ and CMS focus future research efforts aimed at improving expenditure estimates from the MEPS and the NHEA.

How organizational changes relate to firm performance has been the focus of research across disciplines. However, despite its importance, empirical evidence is scant. In healthcare most studies focus on the impact of ownership on performance, but evidence is mixed. We argue that this ambiguity is due to two reasons: first, these studies compare performance across firms that operate in heterogeneous rather than homogeneous markets. Second, there is little assessment of how organizational change per-se relates to performance. We, on the other hand, assess the performance effects of organizational change in the context of a homogenous healthcare market by analyzing its impact on hospital costs and the extent to which efficiency gains from such change interact with scale and scope of hospital services. We also explore how such gains vary with planned vs. unplanned hospital activities and hospital heterogeneity, namely: hospital functional diversity and relative performance. Exploiting detailed 2001-2008 panel-data for English hospitals and the introduction of the Foundation Trust policy that triggered major organizational change - we find that hospitals exhibit economies of scale, but not scope; hospitals that underwent organizational change are more efficient than those that did not; and the organizational change facilitates economies of scope but not scale. However, efficiency gains vary importantly with hospital heterogeneity. Our results suggest that, the FT policy enabled cost-efficiencies, especially for worst-performing and less functionally diverse hospitals. This highlights that organizational changes can be instrumental in promoting the long-term sustainability of healthcare systems.

In this paper we investigate to what extent the childhood healthcare environment influences later life health outcomes. We examine a fundamental re-organisation of the healthcare environment in the U.K., which occurred through the introduction of the National Health Service (NHS) in July 1948. Immediate large decreases in infant mortality of 17% ensued, which were focused on the neo-natal period and larger for individuals who prior to the NHS had a lower access to medical services. Data: We combine historic county-level data with the Office of National Statistics Longitudinal Study of linked census records combined with administrative mortality data, and a large new dataset - the UK Biobank - recording health measurements linked to administrative hospital records to assess the long run impact of birth exposure to the NHS on health and mortality 50 to 60 years after its introduction. Method: As the NHS was introduced nationwide on a single date, we employ a Regression Discontinuity Design, where we will allow for preexisting trends in the outcomes to be different either side of the threshold (i.e. the timing of the NHS introduction).We combine this method with geographic variation in access to medical services through the NHS. Findings: Our findings indicate that survival rates are systematically higher among lower class individuals whose maternity care expanded through the NHS, with the magnitude of the effectincreasing monotonically with age and becoming statistically signicant from age 57 onwards. The increase in the benecial impact of the NHS on survival rates in this population group represents a 12% reduction in mortality (and a 1% increase in survival) at age 57. We supplement these findings with analysis of hospital records, which reveal a similar decrease in hospitalisations for cardiovascular disease, one of the major causes of death, for lower class individuals. Our results suggest that the expansion to universal healthcare (and individual exposure to this universal system at birth) leads to a narrowing in the mortality gap between social classes at older ages.

Research Objective: Before the Affordable Care Act (ACA), people with chronic conditions were typically denied coverage or faced high, experience-rated premiums or preexisting condition exclusions in the nongroup market. Expanding access to nongroup coverage for these individuals while keeping premiums affordable was thus a key objective of the ACA. Recent policies threaten to undermine ACA provisions designed to include healthier and sicker individuals in a single risk pool, yet relatively little is known about the medical needs of people with nongroup coverage who would be affected by these policies. For this study, we examined the health status and health care experiences of adults covered by nongroup plans within and outside of the Marketplaces. Study Design: The study draws on 2012-2015 Medical Expenditure Panel Survey (MEPS) data and focuses on adults ages 18 to 64. We analyzed changes over time in nongroup coverage for this age group. We then estimated changes between pre- and post-ACA implementation periods in the treated prevalence of chronic conditions among adults with nongroup coverage, based on diagnosed conditions that were linked to health care provider visits and prescription drug fills. We also compared treatment for chronic conditions by coverage type (Marketplace, other nongroup, employer-sponsored, and public); other measures of interest included disability status, service use, spending, and sources of payment for care. Because open enrollment periods vary by coverage type, the analysis focused on service use and treatment occurring in the last six months of the year among those with continuous coverage during that period. Key Findings: The share of nonelderly adults reporting nongroup coverage more than doubled following ACA implementation, with all enrollment growth occurring through the Marketplaces. Between the pre- and post-ACA implementation periods, there were increases in the shares of nongroup enrollees who were treated for multiple chronic conditions and who were in the top decile of spending for this age group. These changes were driven primarily by the poorer health of adults with Marketplace coverage, many of whom were uninsured prior to ACA implementation. In 2014-2015, nearly 45 percent of Marketplace enrollees were treated for a chronic condition during the reference period, compared with 35 percent of those with non-Marketplace nongroup coverage and 38 percent of those with employer coverage. Relative to other privately insured adults, those with Marketplace coverage were more likely to have been treated for multiple chronic conditions and had higher service use, and most of their spending was covered by private insurers. Policy Implications: The Marketplaces expanded coverage for adults with chronic conditions, but their higher service use has contributed to rising nongroup premiums. Policymakers seeking to address this challenge face a choice between proposals that aim to strengthen the ACA’s risk pooling arrangements and proposals to concentrate the risk of high costs among those with the greatest medical needs. The outcome of these policy decisions will have a significant impact on a vulnerable population of adults who depend on Marketplace coverage to treat their chronic conditions.

Background: Mercy Health Center (MHC) is a faith-based health resource center serving the underserved population across six Georgia counties. The staff and volunteers provide free healthcare to the uninsured including primary care, pharmacy, dental, chronic disease management, and behavioral health counseling. The purpose of this analysis is to explore the impact that MHC has on the healthcare utilization of its patients over time and the associated healthcare costs. Methods: A cohort of 185 adult patients was identified, and 27 months of hospital utilization was recorded (9 months pre-MHC and 18 months post-MHC). MHC utilization during this period was also recorded. After attaching unit costs to MHC visits, emergency department usage, and outpatient services, we analyzed healthcare utilization and costs over time. Simple analyses included non-parametric longitudinal comparisons of utilization by category and of total healthcare costs. Some simple assumptions were made regarding the trajectory of healthcare utilization beyond the 18-month follow-up period. Further, a recurrent event survival analysis of each category of healthcare utilization was conducted using a stratified Cox proportional hazard model and the gap time approach. Results: Emergency department utilization decreased and outpatient services increased after patients gained access to primary care through MHC. The primary healthcare category of healthcare savings was found to be from a reduction in emergency department visits. However, estimated healthcare cost savings were not large enough during follow-up to offset MHC costs. Overall, results were confirmed in the survival analysis as the risk for emergency department decreased significantly following access to MHC. Discussion: Though net cost savings were not realized within the first 18 months at Mercy for this cohort, we would expect cumulative net cost savings to begin to accrue after a patient’s second year at MHC. Our results suggests that investment in a local free clinic can decrease unnecessary emergency department utilization and eventually lead to cost savings for the healthcare system. Limitations: Patient data were gathered from two local hospital databases, as well as from MHC, leading to three specific limitations. First, verification of patient residence in the service area during the full 27-month period was not possible. Second, we did not have data on healthcare utilization outside of the service area or from local private clinics. Finally, our analysis did not consider medication costs, as these data were not available to the research team.

Background One third of Medicare beneficiaries are now enrolled in private plans through Medicare Advantage. After years of growth in federal payments to Medicare Advantage plans, the Affordable Care Act (ACA) slowed or cut such payments. To date, little is known about the impact of these ACA-related payment changes on plan behavior and on benefits provided to beneficiaries. We examined how plans responded to ACA payment reductions relative to their response to pre-ACA payment increases, which could help reveal whether plans are operating above their costs and inform policymakers regarding future payment policy.

Methods We used 2006-2015 data from the Centers for Medicare and Medicaid Services (CMS) to examine the impact of changes in the maximum federal payments to plans (the “benchmark”) on plans’ asking prices (their “bids”) and on benefits received by beneficiaries (the “rebate”) before and after the ACA. This rebate, which equals a portion of the difference between the bid and the benchmark for plans that bid below the benchmark, must be passed on to beneficiaries in the form of lower premiums or additional benefits including reductions in out-of-pocket costs, reductions in drug costs, and increased coverage for vision, dental, and hearing services. We also assessed differences in plan behavior among plans facing larger benchmarks as compared with smaller benchmarks. Analyses used longitudinal models that exploit the variation in benchmark changes before and after the ACA benchmark cuts, adjusted for beneficiary risk, market concentration, fee-for-service Medicare spending, and fixed differences across counties and across years.

Results In real terms, average monthly Medicare Advantage benchmarks grew by $35 before the ACA (2006-2009) and decreased by $81 after the ACA-related benchmark cuts (2012-2015). Before the ACA, for every $1 increase in the benchmark, plans raised their bids by $0.60 (p<0.001) and beneficiaries received $0.30 in rebates (p<0.001) on average. After the ACA, plans lowered their bids by $0.57 on average for every $1 decrease in the benchmark (p=0.03). This symmetrical bid response after the ACA lessened the resulting decline in beneficiary rebates. Moreover, declines in final plan payments and beneficiary rebates were further offset by new bonuses from quality incentives and increases in beneficiary risk scores. Within rebates, after the ACA plans reduced benefits by about twice as much on the margin as they had raised benefits before the ACA for each dollar change in the benchmark. However, plans changed premiums by similar amounts in response to benchmark changes pre- and post-ACA. Plans in more competitive markets were less responsive to benchmark changes than plans in less competitive markets, implying that plans in more competitive markets may be bidding closer to their average costs.

Conclusion In contrast to before the ACA, Medicare Advantage benchmarks decreased after the ACA. Plans responded to these cuts by lowering their bids, suggesting that plans were operating above cost. This plan bid response, combined with additional payments due to quality bonuses and growth in risk scores, helped lessen the decrease in beneficiary rebates, which may explain the continued growth in Medicare Advantage enrollment after the ACA.

This study addresses the research question: How effective has the Community Health Assist Scheme (CHAS) been in reducing market failure in Singapore’s healthcare sector? The CHAS policy, introduced in 2012 in Singapore, aims to improve accessibility and affordability of healthcare by offering subsidies to low and middle-income groups and elderly individuals for general practice consultations and healthcare. The investigation was undertaken by acquiring and analysing primary and secondary research data from 3 main sources, including handwritten survey responses of 334 individuals who were valid CHAS subsidy recipients (CHAS cardholders) from 5 different locations in Singapore, interview responses from two established general practitioner doctors with working knowledge of the scheme, and information from literature available online. Survey responses were analysed to determine how CHAS has affected the affordability and consumption of healthcare, and other benefits or drawbacks for CHAS users. The interview responses were used to explain the benefits of healthcare consumption and provide different perspectives on the impacts of CHAS on the various parties involved. Online sources provided useful information on changes in healthcare consumerism and Singapore’s government policies. The study revealed that CHAS has been largely effective in reducing market failure as the subsidies granted to consumers have improved the consumption of healthcare. This has allowed for the external benefits of healthcare consumption to be realized, thus reducing market failure. However the study also revealed that CHAS cannot be fully effective in reducing market failure as the scope of CHAS prevents healthcare consumption from fully reaching the socially optimal level. Hence, the study concluded that CHAS has been effective to a large extent in reducing market failure in Singapore’s healthcare sector, albeit with some benefits to third parties yet to be realised. There are certain elements of the investigation, which may limit the validity of the conclusion, such as the means used to determine the socially optimal level of healthcare consumption, and the survey sample size. Keywords—Healthcare consumption, Health economics, Market failure, Subsidies.

CHAS is a government subsidy, introduced in 2012 in Singapore, to improve accessibility and affordability of healthcare to low and middle income groups and elderly individuals for general practice consultations and healthcare. As the subsidy provides financial benefit to low and middle-income groups, the aim of this study was to examine the effect of the CHAS subsidy on income inequality. The Lorenz curve and corresponding Gini coefficient values were used to study the effect on income inequality. Curves were plotted to represent income distribution in 3 separate groups: (a) The nation as a whole for incomes up to $19,999 per month per household, using publically available government data. (b) The same data to which is applied theoretical maximum claims by CHAS-eligible households. Information on maximum claims was obtained from the CHAS website. (c) Forty-eight subjects who provided real world data on their income and CHAS subsidy claims. Gini coefficients were calculated for the corresponding Lorenz curves. The paired t-test was used to determine whether differences between curves were statistically significant. Findings from the Lorenz curve and Gini coefficient data showed that in a theoretical situation of maximum CHAS subsidy claims, there would be a clear decrease in income inequality (change in Gini coefficient = 0.036, p<0.0001). For the real-world data, CHAS claims created a small but statistically significant reduction in income inequality (change in Gini coefficient = 0.005, p<0.0001). There was a small but visible shift of the Lorenz curve. This study suggests that the CHAS subsidy has redistributive effect, with the potential to reduce income inequality at a national level. This effect could be potentiated by greater use of the CHAS scheme by eligible patients. A number of caveats were identified in making the conclusions. First, theoretical maximum claims far exceed actual usage in the population surveyed, and may not represent the real world situation. Second, the study does not account for monthly households with income above $19,999, which make up 12% of the national population. The data therefore does not account for the entire population. Third, the real world data sample size is small and a much larger survey would be needed to verify real world findings.

Medicaid Expansion and Changes in the Health Care Workforce

Abstract:

Eligible individuals in states that expanded Medicaid have reported gains in health care access measures, including having a personal physician, affordability of care, and insurance coverage. However, this change in insurance eligibility led to increases in demand for health care, which raised the question of whether the Medicaid expansion would exacerbate clinician shortages. This paper examines several aspects of the health care workforce across states with different expansion status. First, we assess whether expansion and non-expansion states have different baseline levels of workforce capacity. Then, using a difference-in-differences approach, we test whether the Medicaid expansion had any effects on the number of physicians, nurses, physician assistants, and other health care professionals working in these states. Lastly, we utilize a more granular approach by studying whether the expansion had differential effects within states on the health care workforce based on which counties experienced the largest changes in eligibility.

We find that the pre-ACA per-capita health care workforce was greater, on average, in states that expanded Medicaid (e.g. 322 physicians per capita in expansion vs. 261 in non-expansion states in 2013). Further, non-expansion states do not offset their lower baseline physician supply with greater numbers of other health care professionals, as non-expansion states lag behind expansion states in their supply of all health care professionals.

Post 2014 expansion, we find no evidence that the Medicaid expansion had any significant impact on the available health care workforce post-reform across multiple categories of clinicians. County-based analyses are in process but will be available in time for the conference.

Our results have important economic and policy implications. While Medicaid expansion thus far had led to improvements in access to care across multiple studies, the remaining non-expansion states may have different experiences should they expand given significantly smaller health care workforces at baseline. Furthermore, at present we do not find a Medicaid expansion effect on the health care workforce, suggesting there has been little effect of expansion on short-term entry or exit into health professions, or selective migration by clinicians into (or away) from Medicaid expansion states. However, these effects could manifest in the future, given the lengthy educational requirements within health care professions.

In the conventional framework for designing health plan payment models, the regulator chooses variables to be used as risk adjustors, the risk adjustment weights, and other policy parameters, but the data from which estimates are derived are taken as given. This approach implicitly assumes the observed spending patterns are optimal. In this paper we explore an entirely novel approach: using the data itself as a policy tool. We take the risk adjustors, the estimation method, and other plan payment features as given, and change the data used for estimation to achieve a policy objective. We develop a general model for the provision of health care services by health plans. The key insight of our model is that there is a two-way relationship between plan actions and health plan payment: plan actions (outcomes) are a function of health plan payment, and plan payments are a function of the insurer actions the payment system is meant to affect. Importantly, we show when plan payments are calibrated on data generated by plan actions, equilibrium is when plan actions lead to a set of prices (via the algorithm) that induce the current (possibly inefficient) health care system. Using Medicare data we apply these ideas to two areas of misallocation in health care: undercompensation for individuals with mental health diagnoses and disparities in health care spending between high and low income groups. We transfer spending to the group of interest, re-fit the risk adjustment model on the modified outcome, and illustrate the relationship between the transfer amount and targeted measure. We show spending can be transferred between disease groups to eliminate undercompensation with a minimal impact to overall fit of the risk adjustment model, while correcting disparities requires shifting much larger amounts of spending.

Background: While the Affordable Care Act (ACA) in the U.S. has led to significant gains in health insurance coverage and access to care, less is known on how this policy change affects non-geriatric cancer patients’ utilization and outcomes of emergency departments (EDs). Previous studies demonstrate that younger patients and those who lack access to usual source of care, in general, are at an increased risk of using EDs. Similarly, patients with Medicaid or no insurance use EDs more often than those with private insurance. Within cancer patients, adolescents and young adults (AYAs), aged 15-39 years, are more likely to experience insurance- and cost-related barriers to care. Yet, it is unknown how the ACA has affected ED use outcomes for AYAs with cancer when compared with children and older adults with cancer.

Objective: We examined changes in ED use and outcomes for AYA cancer patients compared with those of children and non-geriatric adults with cancer in the U.S. before and after the implementation of the ACA.

Data and Methods: We used the 2013 and 2014 National Emergency Department Sample (NEDS) from the Healthcare Cost and Utilization Project. Our subpopulation consisted of cancer patients (any cancer diagnosis) currently aged 64 years and younger. We compared patients’ demographic (e.g., sex, primary payer, county of residence) and clinical (e.g., number of chronic conditions and procedures, cancer diagnosis) characteristics, and hospital characteristics (trauma designation, teaching status) between years. We also examined the ED outcome (treated and released vs. admitted to the same hospital). Variables were examined for the overall sample and also stratified by age categories (i.e., children 0-14, AYAs 15-39, and adults 40-64). Chi-squared tests compared proportions, and logistic regressions were used to identify factors that affected the ED outcome. All analyses were weighted.

Results: Overall, cancer patients accounted for over 1.8 million ED visits in 2013 and 1.9 million visits in 2014. A significant decrease was observed in self-paid visits from 2013 to 2014 (9% to 6%). Self-paid visits decreased from 16% to 12% (p<0.001) for AYA visits and from 8% to 5% (p<0.001) for adult cancer visits, however, no differences were observed in primary payer of ED visits by children (2% in both years). Within each group, ED visit outcome did not differ between years. Yet, in each year, AYAs were more likely to be released than admitted (69% vs. 31%) compared with children (58% vs. 42%) and older adults with cancer (51% vs. 49%) (p<0.001). In our adjusted models that accounted for other covariates, in both years, self-pay AYA cancer visits were less likely to be admitted compared with children.

Conclusion: While self-pay AYA visits decreased in 2014 following implementation of the ACA, self-pay visits remained significantly higher for AYAs (12%) as compared to children/adults. Our analyses demonstrate that the ED outcome differs depending on health insurance status. This is especially true for AYA cancer patients who were more likely than children/adults to be uninsured and, thereby less likely to be admitted to the same hospital following an ED visit.

Transportation can be a significant barrier to attaining healthcare services, in particular for disadvantaged populations. We exploit a reform that introduced public transportation services to Arab towns in Israel to evaluate the effect of greater access to healthcare services via public transportation among a disadvantaged population on health outcomes. In 2007, the Israeli ministry of transportation (MOT) announced a reform within non-Jewish communities in Israel: the introduction of public transportation (PT) to these communities. Until then, Non-Jewish communities had been significantly deprived of PT infrastructure, with generally no official services. Furthermore, private car ownership rates are relatively low among Arabs, and many women do not have a driving license due to traditional barriers. The new bus network, which gradually developed over the next 7 years, represented a substantial increase in access to healthcare services. Health services in the form of the best doctors or specialty clinics and hospitals are found in Israel outside Arab towns, generally in Jewish cities.

We use very detailed data from the Israeli MOT documenting the frequency of all bus lines in Israel, their routes and bus stops, for 2008-2014 on a bi-annual basis. We matched our measures of bus frequencies for each town and period to a survey of the Arab population in Israel conducted in 2004, 2007, 2010 and 2014. Our analysis focuses on the elderly population - ages 50-70 - based on the health conditions inquired about in the survey - high blood pressure, diabetes, heart problems, high cholesterol, and back problems. The questions in the survey specifically ask about diagnosis and receiving medical treatment for these health conditions.

Our initial results show statistically significant increases in the diagnosis of heart disease, high cholesterol, asthma, back problems, and migraine headaches among the population aged 50-70 when the penetration of buses to these individuals’ communities increases. We do not observe statistically significant changes in response to public transportation penetration for diabetes and high blood pressure. We observe some differential effects based on respondents’ sex. We also observe elderly individuals reporting that they are overall less healthy when public transportation is greater in their community.

A naive interpretation of these results can suggest that public transportation penetration is adversely affecting health outcomes among the elderly population. We believe that given that the survey inquires about diagnosis of these health conditions, a more plausible interpretation of the results is that there are greater diagnosis levels and awareness of health conditions that were existent prior to PT penetration but were unobserved due to lower access to healthcare services. Our results will be further corroborated by deceased records we have obtained showing that deaths do not increase in response to public transportation penetration.

In China and other developing countries, long waiting times during a hospital visit are pervasive due to the rapidly growing demand for health care services in already overcrowded hospitals. However, technological innovations have been introduced to reduce patients’ waiting times in hospitals and to streamline the process of healthcare services. Such innovations have the potential to improve the efficiency and quality of health care as well as increase patient satisfaction. In the past five years, more than 100 Chinese cities have introduced a Resident Card allowing the patient to schedule appointments online and to pay for medical care efficiently. The Resident Card allows patients to schedule an appointment up to a few days before the visit, avoiding more crowded dates and times. The Resident Card also enables automatic e-payment at physicians’ office and through self-service machines; therefore patients no longer need to wait in line to pay medical bills. Government statistics show that patients save about 45 minutes on average during each hospital visit. We use a large medical claims data set from a major hospital in a Chinese city of 9 million people to analyze the impact of the Resident Card on patients. We have records of 4 million outpatient transactions during 2011-2013. This hospital has adopted the Resident Card since 2012. We use the standard difference-in-differences approach to compare changes in patient health-seeking behavior and health outcomes between the treatment groups (i.e. those using the Resident Card) and the control group. The key parallel trends assumptions are satisfied for both health seeking and health outcomes. Linking each patient by their identifier, we find that compared to the control group, those using the card experience a greater reduction in waiting time for hospital visits and an increase in service use. These two impacts seem to result in a decline in the severity of the health problems being treated due to the greater accessibility. Also, the almost fixed supply of physicians in the short term and the increase in demand for health services give doctors more power relative to patients and payers. However, with government established prices, they make profit by prescribing unnecessary medications and more tests. Moreover, the new transaction technology makes patients’ type of insurance more visible to doctors, further intensifying their ability to make profit, especially from card users with more generous insurance coverage. This study highlights the multiple, intended and unintended consequences of this major reform in medical expenses transactions. The Resident Card produces both benefits and costs, and also systematic heterogeneity across populations. Further analysis is required to fully assess the impact of the reform on public welfare.

Background: Healthcare fraud can represent upwards of hundreds of billions of dollars in spending that could be better spent on patient care. There is often not sufficient detail on the underlying methodologies and data samples that lead to fraud estimates, which may be due to different purposes of these reports or the need to obscure the details of fraud detection methods to prevent fraudulent operators from responding to existing methods. Objectives: The objective of this study was to provide a systematic evaluation and synthesis of the methodologies and data samples used in current peer-reviewed studies on characterizing healthcare fraud. Data Sources: The academic databases searched were Academic Search Complete, Business Source Complete, EconLit, Medline (EBSCO), OneSearch, ProQuest Business Collection, ScienceDirect, and Web of Science. Governmental and commercial sources were also used for background research. Synthesis of Methods: This examination was conducted using a systematic review methodology to identify relevant studies and determine their relevance. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement was used to guide the performance of reviewing the literature. Study criteria for eligibility were collected by applying specific search terms: healthcare, health insurance, Medicare, Medicaid, Obamacare, Affordable Care Act, or health services; fraud, cheat, falsification, corruption, or kickback; detect, detection, prevent, prevention, deterrence, audit, or auditing. Results were restricted to scholarly journals, academic journals, working papers, and conference proceedings. Study selection occurred through two independent reviews of each study for inclusion or exclusion. Disagreements between reviewers were resolved through discussion by the entire research team. Results: Our search terms resulted in 450 articles that were potentially appropriate for inclusion in our report. The results of independent reviews ended with twenty-seven studies considered as relevant to include after the application of our inclusion criteria. Variables are identified from the literature to synthesize each method of fraud detection used. Limitations: One limitation of this study is that the strength of the evidence is reliant on the quality and number of studies previously performed on the topic. Another limitation is the quality of studies with regard to their applicability to different types of insurers. Finally, the majority of studies could not provide proof of intent to commit fraud. Conclusions: A limited number of validated methods are used to detect healthcare fraud. The literature on this topic is spread among several academic fields. The majority of available studies utilize public or social health insurance systems such as Medicare or Medicaid in order to study fraud. The main gaps we identified are validation of existing methods and proof of intent to commit fraud in the studies analyzed. Implication of Key Findings: Our insurer agnostic approach examines the availability and effectiveness of healthcare fraud analytic methods across different types of health insurers, posing great value for members of the health sectors.

Over the past several years, there has been a substantial shift of Medicare enrollment from the traditional, government-administered Medicare program (“traditional Medicare”) to the private, subsidized, and mostly managed care plans offered in Medicare Advantage (MA). While previous work has examined the effect of this trend on the efficiency of care and on beneficiary health outcomes overall, my study is one of the first to consider the implications of increasing MA enrollment for a particularly disadvantaged group of beneficiaries known as dual eligibles (i.e., Medicare beneficiaries who also receive full or partial Medicaid benefits). Dual eligibles merit special attention because they are economically vulnerable, often have significant health needs, and face unique barriers in navigating the health system (e.g., as about three-fifths report having cognitive impairments). Further, while the share of Medicare beneficiaries enrolled in MA has risen overall, the increase has been particularly dramatic among dual eligibles, rising from 1% in 2004 to 32% in 2015. This study evaluates the effect of MA penetration on the number and length of hospital stays (including potentially-preventable admissions) and all-cause mortality rates among dual eligibles. I rely on complete Medicare enrollment and MedPAR files and public MA data from 2009 through 2015. These sources provide detailed information about dual enrollment and capture all hospital discharges at the vast majority of acute care PPS hospitals and all beneficiary deaths. The discharge data provide enough granularity to identify potentially-preventable hospitalizations based on AHRQ Quality Indicators.

My primary identification strategy relies on a regression discontinuity design used by a prior study to evaluate outcomes among the overall Medicare population (Afendulis, Chernew, and Kessler 2017). This approach exploits a discontinuous jump in average benchmark payment rates for MA plans – which is subsequently associated with a sharp increase in MA enrollment – in metropolitan statistical areas that exceed a population threshold. Preliminary results suggest that this increase in plan payments is also associated with a jump in MA penetration rates among dual eligibles who receive partial benefits. I use this potentially exogenous source of variation in enrollment to explore the relationship between MA and beneficiary outcomes. Because dual eligibles may also be enrolled in comprehensive or limited Medicaid managed care plans, I also run analyses restricted to the subset of counties where such enrollment is rare or nonexistent among relevant groups of dual eligibles. My findings will help policymakers understand how the increased role of MA plans has affected a vulnerable subset of the Medicare population and will be informative as states continue to delegate Medicaid benefits for dual eligibles to managed care plans.

Research Objective To explore the reasons why people churn out of Medicaid. Losing Medicaid coverage can have a negative effect on people’s health; it affects continuity of care, especially when they become uninsured. Although not all forms of churn are detrimental (e.g., individuals who find a new job can obtain employer insurance), these transitions can affect costs and place an unnecessary administrative burden on both states and enrollees. Study Design We follow a representative sample of 4,453 non-elderly adults who reported being covered by Medicaid in January, 2013 on wave 1 of the 2014 SIPP. Our model focuses on people’s decision to obtain and maintain Medicaid coverage. We understand churn as a change in this status. We also used first differencing to create our set of (mutually exclusive) predictors: becoming employed or unemployed, gaining or losing a job, having a wage increase or decrease, and having a newborn. We included lagged variables of every life event in our model to allow for a 2-month delayed effect. Working with longitudinal data allowed us to identify these life events, and also use first differencing variables and eliminate any bias due to time-invariant unobserved characteristics. Principal Findings We estimate that 5.1% of nonelderly adults who were enrolled in Medicaid in January, 2013 lost this coverage at some point in the following year. Most of these (70%) had at least one month of uninsurance, while the rest shifted to other forms of coverage. We found that becoming employed and gaining a job increased people’s likelihood of churning off Medicaid, though these changes did not affect churn immediately. If someone’s family member became employed in one month, their probability of churning out from Medicaid the same month did not change, but it increased by 1.2 percentage points (pp) the next month and by 0.7 pp in the month after. Another determinant of churn was having a child, which makes individuals more likely to churn by 1.1 pp the same month, 1.2 pp the following month, and the month after. This result is a novel finding, as prior research has not focused on this life event (most research on churn has not focused on actual changes in coverage). Discussion Our results are consistent with the theory that becoming employed or gaining a job would increase family income, which could make some enrollees ineligible. Although this would generally be viewed as a positive development, it could result in the individual becoming uninsured or underinsured, or experiencing discontinuities in care. These effects were only significant after a lag; the effect on churn does not happen within its month of occurrence. This has implications for researchers studying churn because omitting these lags could bias their estimates. Since women are still eligible for enrollment during the months immediately postpartum, our results could indicate that some women believe their coverage ends at birth. This is important, because if women believe they lost Medicaid coverage, it may impact their healthcare decisions as much as actually being uninsured.

To accommodate Medicare and Medicaid budget cuts and reimbursement method innovations, hospitals have made great effort on cost containment and quality improvement. The first goal of the paper is to study the empirical relationship between healthcare service quality and hospital cost containment. Our current finding suggests that cost containment is negatively related to quality improvement. But the relationship diminished after 2013, which is the year when Center for Medicare and Medicaid Services (CMS) implemented Value-Based Purchasing (VBP) program and incorporated quality as important dimension in acute care hospital inpatient services reimbursement. Therefore, our more important goal is to examine whether VBP effectively incentivized hospitals to improve or maintain quality while reducing costs. We use two dataset primarily: Hospital Compare and Medicare Cost Report. Hospital Compare provides hospital quality measures on clinical care, patient experience, safety, and efficiency. A Total Performance Score (TPS) is calculated as a weighted average of these quality aspects. And a linear exchange function translates TPSs into VBP adjustment factor, which determines a hospital’s value-based incentive payments. Top performers are rewarded with bonus and bottom ones get a discount of full reimbursement. CMS started to reimburse acute care hospital’s inpatient services using this method since 2013. Medicare Cost Report data records hospital operation, financial management, and claims comprehensively. To answer questions of the relationship between cost containment and quality change and effectiveness of VBP incentives, we designed three sets of studies: First, to observe empirical relationship between cost containment and quality change, we use VBP-eligible acute care hospital TPS change as the dependent variable and cost containment proxies as the main predictor controlling for covariates commonly used in previous healthcare service quality research. And we examine the sample in two time periods separately: financial crisis years before VBP 2007 – 2012 and VBP years 2013 – 2016. Hospitals were motivated to reduce costs in both periods but for different reasons. And we use the TPS calculation method in 2013 consistently for all years. We found that the negative relationship between cost containment and quality improvement is stronger before 2013. Second, we address whether the VBP program incentivized hospitals to improve quality while reducing costs. And we apply a ‘quasi’ Difference-in-Difference (DD) study. The VBP program is applied to all eligible acute care hospitals simultaneously since 2013, which excluded a clean control group. Therefore, the DD method is compromised by taking hospitals with less than 10% of Medicare revenue as a ‘quasi’ control group and hospitals with Medicare revenue more than 50% as the treatment group. The DD analysis shows marginally significance of the VBP factor. Besides, we plan to further examine whether the VBP calculation method change and the progressive payment adjustment affect results. Finally, we offer a possible explanation. In particular, we examine the relationship between quality change and cost containment for hospitals in different financial standings. For some hospitals, benefit of cost containment may balance or even surpass the Medicare payment reduction resulting from the subsequent lower quality.

Health economics defines the extensive margin for surgical procedures as the rate at which people with similar diagnoses and medical conditions receive various treatments. While much attention has focused on geographic variation in the extensive margin for various procedures, the expanding use of certain procedures in an increasingly aged and frail population may constitute a growing problem given limited health care resources. An important illustration of this problem pertains to percutaneous coronary intervention (PCI) and coronary artery bypass surgery (CABG) procedures. While these treatments can be life-extending in many circumstances, they are often associated with significant morbidity and delayed recovery among elderly or frail patients. In particular, prolonged rehabilitation can be required due to complications or from frailty and poor functional status at the time of presentation. Randomized trials of PCI and CABG typically exclude frail and elderly patients, particularly those with renal dysfunction, and do not report the rate of nursing home discharges and delayed physical recovery. This analysis describes current trends in coronary revascularization and subsequent care. Medicare claims data from 2006-2015 for a 20% sample of Medicare beneficiaries age 65 and older are used to described trends over time in procedure rates for CABG and PCI for all beneficiaries as well as beneficiaries with selected coronary diagnoses (e.g., acute myocardial infarction, acute coronary syndromes, unstable angina) to adjust for changes in underlying disease rates over time. The descriptive trends for inpatient procedures are calculated overall and for sub-groups (e.g., age groups, case mix severity, and receipt of hemodialysis) for fee-for-service as well as Medicare Advantage enrollees. Analyses of trends for total procedures (inpatient and outpatient) and outcomes are conducted only for fee-for-service enrollees given lack of claims for physician and post-acute services for Medicare Advantage enrollees. We use regression analysis of claims for fee-for-service enrollees receiving PCI or CABG to assess key outcomes: repeat revascularization, post-acute care including skilled and non-skilled nursing home days, hospital readmission, hospice use, post-discharge mortality, total Medicare reimbursements, and non-institutionalized days. Preliminary results show that from 2007 to 2015, rates of CABG declined overall; CABG rates increased slightly among persons aged 85 and older or persons but declined slightly among persons with more complex disease (Charlson>3). In contrast, rates of PCI were fairly stable over time but increased among persons aged 85 and older as well as among persons with more complex disease (Charlson>3). Among beneficiaries receiving coronary revascularization, overall rates of discharge to skilled nursing care increased over time while discharges to home without home health care decreased. The outcome analyses that are in process will assess one-year outcomes and resource use trajectories over a longer period (up to seven years) by the sub-groups identified above. Descriptions of current trends in procedure rates and outcomes including resource use enable discussion of the implications of expansion of the extensive margin for coronary revascularization. The estimates will provide valuable information to policy makers as well as healthcare professionals who routinely consider the relative risks and benefits of PCI and CABG.

The United States has a complex system for financing healthcare, combining an array of public and private components. In addition to out-of-pocket amounts paid directly to providers, healthcare financing includes: the payment of premiums directly to insurers, employer contributions for workers’ health plans, and public programs that provide health coverage and draw on state and federal tax revenues. Given the growth of health care spending and the fact that individuals ultimately bear these costs in some form, there is substantial value in developing a thorough understanding of how much individuals and families pay for healthcare, how these payments are distributed, and how the incidence of financing evolved over time. Previous attempts to understand how much individuals pay for health care typically focus on particular types of healthcare payments in isolation, such as out-of-pocket medical costs (Banthin et al, 2008) or premiums for private insurance (Gruber and McKnight, 2003). Little comprehensive analysis of equity in the finance of U.S. healthcare has been conducted, with two notable exceptions being Wagstaff et al. (1999), which examined data from 1987, and Ketsche et al. (2015), reflecting financing in 2004. We develop an updated analysis of healthcare finance equity using the nationally-representative Medical Expenditure Panel Survey (MEPS) combined with a variety of supplementary datasets to benchmark and to enhance our estimates. We align the MEPS distribution of income to Internal Revenue Service data, simulate a full array of federal and state income tax expenditures and state sales taxes, and report sources of financing for healthcare by quintiles of equivalent income. Our preliminary assessments cover the 2004 to 2013 time period, but our final analysis will be extended to 2015 to include the early years of the main Affordable Care Act reforms. Our systematic analysis of all major components of health spending reveals that the financing of healthcare in the United States – notwithstanding a relatively progressive structure for income taxes – is regressive. The bottom quintile of households in the United States paid 11.7 percent of their income in health payments in 2013 compared with 8.1 percent for the top 1 percent of households. However, our estimates also show that the U.S. system for financing healthcare has become more progressive over time with the share of income devoted to health spending for the bottom quintile falling 6.1 percentage points from 17.7% in 2005 to 11.7% in 2013 and the share paid by the top 1 percent of households increasing 1.9 percentage points from 6.2% to 8.1% over that same period. Our preliminary study period encompasses the introduction of several significant changes to healthcare and tax policy, which are consistent with our findings of increased progressivity in the healthcare sector, including the beginning of the Part D prescription drug program in Medicare (2006), the introduction of income-related premiums in Medicare (2007), and the Additional Medicare Tax for higher-income households (2013). Although we do not yet have complete data for 2014 and 2015, the increased insurance coverage in those years likely led to further increases in progressivity.

To help Medicare beneficiaries choose among Medicare Advantage (MA) plan options, the government collects data on several dimensions of care and summarizes this information through a star rating of plan quality, measured on a scale of one to five. Prior research suggests that beneficiaries do indeed rely on star ratings when making enrollment decisions and the government has used star ratings as the basis for providing billions of dollars in bonus payments. Despite their importance, it is unclear whether these aggregated ratings of plans – based in large part on process of care, patient experience, and intermediate outcome measures – ultimately signify differences that lead to improvements in health. Indeed, there remain questions about the direct relationship between many star ratings and health outcomes, whether star rating measures correlate to strong or weak performance on other dimensions of quality, and the extent to which star ratings reflect differences in quality or differences in underlying populations.

This study evaluates the relationship of star ratings with the number and length of hospital stays (including potentially-preventable admissions) and all-cause mortality rates. I rely on complete Medicare enrollment and MedPAR files and public MA data from 2009 through 2015. These data identify whether a beneficiary enrolled in MA and, if so, the star rating of their plan if they received drug coverage (as is currently the case for approximately 90 percent of MA enrollees). They also capture all hospital discharges at the vast majority of acute care PPS hospitals and all beneficiary deaths. The discharge data provide enough granularity to identify potentially-preventable hospitalizations based on AHRQ Quality Indicators.

My primary empirical strategy exploits potentially exogenous changes in enrollment to explore the relationship between star ratings and beneficiary outcomes. To mitigate the role of selective enrollment, I make use of (1) MA plan exits (which force enrollees to switch plans) and (2) the varying circumstances following plan exit that affect whether beneficiaries switch to a higher- or lower-star plan. I operationalize this approach by estimating an event-study model with a varying treatment. The event is an MA plan exiting the market and the treatment is the difference between the star rating of the terminated plan and the enrollment-weighted average star of the remaining plan options. To address the possibility that these factors might correspond to other regional changes over time, I include enrollees in non-exiting plans who reside in the same county as an additional control group. I focus on plan exits between January 2012 and January 2014, which are associated with about 800,000 beneficiary-year observations after applying sample restrictions (e.g., excluding beneficiaries in terminated private fee-for-service plans).

My findings will help policymakers determine how much weight to attach to star ratings in plan regulations and payment rules and will be informative as the government continues to pilot test a quality rating system on the federally-facilitated marketplaces.

Research Objective: Florida implemented mandatory managed care for Medicaid enrollees in April 2014 via the Statewide Medicaid Managed Care (SMMC) program to improve access and coordination of care. This program enhanced access to primary care by increasing the number of primary care providers (PCPs), and after-hour appointment availability. The research objective of this study is to analyze the impact of the enhanced access to primary care on preventable hospitalizations among Medicaid enrollees. Study Design: We estimate a difference-in-difference (DD) model, comparing the change in the number of preventable hospitalizations in Zip Code Tabulated Areas (ZCTAs) with greater improvement in access to primary care (measured by percentage change of enrollee to provider ratio in the area from 2013 to 2015), compared with the changes of that in ZCTAs with less improvement in the access to primary care after the implementation of SMMC. We control for ZCTA specific socio-demographic characteristics, fixed effects for county of residence, year fixed effects, and a county-specific linear trend. The key explanatory variable is an interaction between the indicator for being a ZCTA in the top quartile of improvement in access to primary care, and the indicator for the post period. The main outcomes are numbers of preventable hospitalizations, i.e., whether the hospitalization was for an ambulatory care sensitive condition (ACSC), per 1000 residents in each enrollees’ ZCTA. We adopt the Prevention Quality Indicator (PQIs) developed by Agency for Healthcare Research and Quality (AHRQ) to identify hospitalizations for ACSCs. Population Studied: We compiled the analytic sample from three data sources. Florida inpatient discharge data from 2010 to 2015 provided information on inpatient visit. There were 1,837,294 discharges for Florida residents between the ages of 18 and 64 with a primary payer of Medicaid insurance, and no missing values on covariates used. We stratify the data into cohorts according to ZCTA, and quarter. The final analytic sample includes 19,621 stratified observations at the ZCTA-quarter level. We supplement the analyses with enrollee to PCP ratio in each ZCTA, created from Florida Medicaid provider data repository and 2010–2014 United States Census American Community Survey (ACS). Principal Findings: We find that areas with greater improvement in the access to primary care experienced reductions in the incidence of overall preventable hospitalizations of 7.2 per 100,000 residents (18.9 percent) compared to areas with less improvement. Those areas also saw reductions in the hospitalizations for chronic ACSCs (reduction of 6.9 preventable hospitalizations per 100,000 residents (25.0 percent)) in the post-implementation period relative to other areas. Conclusions: Our results show that areas with greater improvement in the access to primary care have greater reductions in hospitalizations for ACSCs, especially hospitalizations for chronic ACSCs, compared to areas with less improvement. Implications for Policy or Practice: Managed care in Medicaid provides a foundation for enrolling vulnerable populations and guaranteeing them better access to primary care and care coordination, to reduce cost of the program. Our study provides direct evidence that improved access to primary care under Medicaid managed care is associated with reductions in preventable hospitalizations.

Background: In 2017, the Centers for Medicare and Medicaid Services implemented the Merit-Based Incentive Payment System (MIPS), establishing a new payment program for clinicians participating in the fee-for-service Medicare program. As part of a broader push to link provider payments to value, the MIPS is a pay-for-performance model that intends to reward clinicians for improving quality of care and lowering spending by providing practices with bonuses or penalties based on their performance on quality and spending measures. Although the effects of the MIPS will not be known for several years, its basic design is similar to that of its predecessor, the Value-Based Payment Modifier (VM). From 2014 to 2016, the VM was phased in for physician practices meeting specific size thresholds (i.e., number of constituent clinicians), creating abrupt discontinuities in the exposure of practices to pay-for-performance incentives. We harnessed these discontinuities in a quasi-experimental regression discontinuity design to evaluate the VM's effects on performance measures assessed for all practices subject to the program. Study Design: Exploiting the phase-in of VM incentives based on practice size, we used regression discontinuity analysis and Medicare claims in 2014 to estimate differences in practice performance associated with the abrupt exposure of practices with ≥100 clinicians to full VM incentives (bonuses and penalties) and the exposure of practices with ≥10 clinicians to partial incentives (bonuses only). We repeated analyses using 2015 claims to assess the association of a second year of exposure to pay-for-performance incentives. We examined performance on four sets of outcomes: hospital admissions for ambulatory case-sensitive conditions (ACSCs), all-cause 30-day readmissions, mortality, and Medicare (Part A and Part B) spending per beneficiary. We conducted supplementary analyses to assess the robustness of our results to model specification and placebo tests to check whether discontinuities in outcomes at the VM's implementation thresholds (i.e., ≥10 and ≥100 clinicians) exceeded those at arbitrary thresholds of practice size where incentives did not differ. Results: In 2014, there were no significant discontinuities at the ≥10-clinician threshold in the relationship between practice size and admissions for ACSCs (adjusted discontinuity:+0.003 admissions/beneficiary; 95% CI:-0.0003,0.006), proportion of admissions with readmission (+0.1 percentage points; 95% CI:-0.4,0.6), Medicare spending ($234/beneficiary; 95% CI:-$148,$616), or mortality (+0.2 percentage points; 95% CI:-0.1,0.5). Similarly, there were no discontinuities at the ≥100-clinician threshold in admissions for ACSCs (-0.002 admissions/beneficiary; 95% CI:-0.006,0.003), proportion of admissions with readmission (+0.3 percentage points; 95% CI:-0.6,1.2), spending (-$152/beneficiary; 95% CI:-$712,$408), or mortality (-0.1 percentage points; 95% CI:-0.5,0.3). Analyses of the ≥100-clinician threshold using 2015 data revealed no discontinuities associated with a second year of full exposure to the VM. Discontinuities estimated over all practice size thresholds, and using various practice size ranges, revealed no consistent evidence that discontinuities were largest at the ≥10 and ≥100-clinician thresholds. Conclusions: The VM was not associated with significant differences in performance on program measures at thresholds where physicians' incentives differed. Our findings suggest that incentives in the VM were not sufficiently strong to affect practice performance, questioning whether the similarly-designed MIPS will achieve its intended policy goals.

Abstract

We study the impact of a merit-based incentive payment system on provider behavior in the primary care setting using experimental methods that leverage healthcare simulations with patient actors. Our approach allows us to exogenously change a provider’s incentives and to directly measure the consequences of alternative payment systems. Within our sample, we find that merit-based incentive payment systems increase the number of the incentivized outcome measures met, but lower overall quality of care through unintended effects on other indicators of care.

Healthcare in the United States is undergoing a transition from volume to value. Current value-based payment models incentivize providers to efficiently provide high-quality care and generate savings in the short-term. The next wave of models will need to incentivize providers to promote health across the life-course, and reward efforts that generate both short- and long-term savings. Healthcare payment models can take a life-course perspective on value by tying incentives to reductions in long-term actuarial risk, such that an intervention is valuable to the extent it reduces the payer’s predicted future costs. This actuarial risk-based valuation builds in the business case for how increased investment in health promotion can still be cost-neutral to the healthcare system overall. Specifically, as long as the total amount invested does not exceed the expected value of the reduced actuarial risk, the payer can be relatively confident that the intervention will be at least cost-neutral over the long-term. This approach also intrinsically risk adjusts, as larger investments are justified where there is more risk to reduce. In addition, this approach can avert incentivizing perverse outcomes by assigning monetary value to avoiding serious adverse events that may not technically cost a payer very much. Payers can integrate indicators of future reductions in actuarial risk into existing payment models by using “net present value of care” (NPVoC) rather than total cost of care in determining incentives (see Figure 1). NPVoC is the total cost of care (i.e., the lesser than anticipated amount spent on healthcare), plus the expected value of care (i.e., the amount the payer anticipates that the achieved health outcomes will save in later healthcare costs, divided by a discount rate). The expected value of care could differ from payer to payer, based on how long individuals are anticipated to remain with the plan. In instances of high market penetration, it may be five, ten, or even more years. As Medicare, Medicaid and large provider organizations initiate multi-payer arrangements (i.e., multiple health plans agree to similar and mutually beneficial payment terms) the duration of expected value calculations will lengthen, because even when

At present, payers have limited ability to accurately predict long-term reductions in actuarial risk from short-term clinical outcomes—an undeniably challenging endeavor. To begin testing and improving NPVoC models, substantial precision is unnecessary. Payers can use a conservative estimate of predicted savings associated with a modest magnitude of change on a short-term outcome, minimizing potential overall losses when predicted savings are shared in expected value-based payments. By allowing payers to calculate the expected value of an outcome, NPVoC models can avoid problems from the past where research indicated likely savings that failed to materialize when the payer took the intervention to scale. Precise estimates of future savings for NPVoC models may take years of data collection and integration, but initial models can be tested and iteratively improved to build the foundation.

Recent Changes over Time in Disparities in Late Stage Breast Cancer Diagnoses among Younger Women in the US

About 44% of young women with Breast cancer (BC) are diagnosed with advanced stage, and it is rising at a faster rate than for older women. This study focuses on examining whether there were significant changes on the predictors of late-stage BC diagnosis rate among young women in the US across the two periods i.e. pre (2004-2009) and post (2010-2014) Affordable Care Act implementation. Using the US cancer statistics registry database, we extracted all 139220 BC cases for young women (<=40 years) from 46 states to examine the variation in their late-stage BC diagnosis across the two periods. We used a random intercept logit model with person, county, state and time level covariates after controlling for factors that may moderate the effect of the ACA implementation. Results suggest that young African-American women had higher odds of late-stage diagnosis relative to the whites, and the odds increased over time. Area urbanicity, poverty, unemployment, obesity rate and Medicaid enrollment rate were associated with increased odds of late-stage diagnosis, and the odds were higher in the later period. Also, per-capita healthcare expenses, a residential diversity index, the average age of mothers at first birth, drinking and marital status rates, BC screening rate and private insurance enrollment rate were associated with lower odds of late-stage diagnosis, which increased in the later period. Thus, disparities in late-stage diagnoses of BC among young women increased after 2010, highlighting the increased importance of having better access to medical care and prevention among younger women following implementation of the ACA in 2010. As time passes, further investigation is needed to understand these persistent disparities and whether health care reform will eventually seem to

Background: In 2012 Oregon transitioned its Medicaid program to cover 90% of its enrollees in Coordinated Care Organizations (CCOs). CCOs can be seen as a type of Medicaid Accountable Care Organization, but include an administrative layer (similar to a Managed Care Organization) and are at full financial risk through a global budget. As part of its 1115 Medicaid Waiver, Oregon agreed to reduce its historical rate of health care spending growth from 5.4% to 3.4%. The global budget and prespecified rate of growth have analogies to per capita caps and block grants, features that were prominent in several “repeal and replace” bills that were put forth in 2017. We provide an assessment of the Oregon

Study Design. We estimated changes in inpatient care using a difference-in-differences approach and data from the Healthcare Cost and Utilization Project (HCUP) and monthly and county Medicaid enrollment data from Oregon and comparison states of Washington, Colorado, New Mexico, and Arizona. We used 1-1 nearest neighbor matching to identify similar county and demographic matches across these comparison states. Using 2 years of pre-intervention data (2010-2011) and 2 years of post-intervention data (2013-2014), we assessed the impact of the CCO model on inpatient admissions and length of stay. In subanalyses, we also assessed changes in hospitalization rates for elective vs. emergency admissions; ambulatory-care sensitive admissions, and admissions among individuals from low-income neighborhoods. Finally, since the CCO model provided care for more than one out of every four Oregonians, we tested for spillover effects among the commercially insured. We assessed parallel trends among the treated and comparison groups and found no statistically significant differences across multiple groups, including all admissions, length of stay, and

Results. Oregon’s transition to global budgets was associated with significant reductions in the rate of inpatient admissions of approximately 1 admission per every 1000 enrollees per month, equivalent to a 9% reduction in the admission rate. There were significant reductions in both the emergent and elective rate of admissions, although the changes were larger among elective admissions, and, within that group, particularly among admissions for births and maternity care. The length of stay among Oregon Medicaid enrollees was low prior to the CCO transformation and remained low in the two years following the intervention, suggesting that changes in the extensive margin were not offset by changes in the intensive margin. In subanalyses, we found significant reductions in ambulatory care sensitive admissions and admissions among individuals from low-income neighborhoods. We found no evidence of spillover in the commercial

Conclusions. Oregon’s global budgets were associated with significant reductions in Medicaid inpatient admissions relative to matched counties in western states. The extent to which the Oregon model can continue to place pressure on inpatient utilization will depend on the longer-term success of infrastructure investments and a portfolio of delivery system changes that are designed to provide care in less intensive settings. These changes may have lessons for Medicaid

The impact of policy regulation in the health sector depends on hospital technology, which is largely reflected in the productivity of labor, capital and medicines. This paper focuses on acute-care local public hospitals in Japan and examines how technology differences determined the effect of voluntary changeover by hospitals to the prospective payment system. The Japanese hospital sector provides a rare example of a nationwide introduction of the reform with self-selection. We exploit panel data conditional quantile regressions to model a range of technologies for the multi-product output function of hospitals under an endogenous treatment assignment. The analysis reveals technological heterogeneity, and incorrect labor/capital and labor/medicines mix. The impact of prospective payment on output is primarily attributed to labor and is inversely related to productivity. Finally, we contrast the design of prospective

The novelty of the present paper is severalfold. To the best of our knowledge, the paper is the first application in health economics of quantile regression models for estimating the longitudinal production function of hospitals. The analysis incorporates multiple outputs, evaluates factor returns and assesses optimality of input mix across quantiles. Secondly, the paper proposes an approach to account for groupwise serial correlation in the estimates of pooled models of conditional quantile regressions under endogeneity. Thirdly, the paper is the first health economics study to assess the heterogeneous treatment effect of a prospective payment system on hospital production. Finally, the paper is the first

The paper is unique in using the data on Japanese hospitals from several sources: the longitudinal financial data for municipal and prefectural public hospitals accumulated by the Ministry of Internal Affairs and Communications; databases on hospitals which introduced a prospective payment system and on designated local hospitals, kept by the Ministry of Health, Labor and Welfare; and the database of the Japan Residency Matching Program on teaching hospitals. We discover that a hospital's voluntary decision to introduce prospective payment may be motivated by teaching status, average tenure of doctors and size (floor area) of hospital.

The technology differences are revealed in different productivity and the technical rate of substitution between hospital inputs (labor, capital and medicines). In particular, we discover lower labor returns at high-output hospitals. The findings reveal that the reform has the biggest effect at low and medium-output Japanese hospitals, while it might be less significant for the most productive hospitals. The impact of prospective payment on output is primarily attributed to

We believe that considerations of technological heterogeneity and its link with outcomes of the hospital financing reform would offer helpful guidance for policy measures. Moreover, knowledge of the determinants of self-selection could

Spillovers can arise in markets with multiple purchasers relying on shared producers. If producers are constrained in their ability to adjust quality and cost across purchasers, then the influence of a dominant purchaser affects the entire market. Prior studies have found such spillovers in health care, from managed care to non-managed care populations — reducing spending, utilization, and improving outcomes. Similar effects have been identified in the Medicare Advantage market as well, with studies finding declines in utilization and reductions in resource use among the Traditional Medicare population associated with increases in county-level Medicare Advantage penetration. However, no study to date has provided plausibly causal estimates of such spillovers in the post-Affordable Care Act era. Our study does so by exploiting idiosyncratic differences in payments to Medicare Advantage plans that are unrelated to traditional Medicare spending. Further controlling for health status and other potential confounders, we estimate that a one percentage point increase in county-level Medicare Advantage penetration results in a $146 (1.7%) reduction in standardized per enrollee Traditional Medicare spending. We find evidence for reductions in utilization both on the intensive and extensive margins (including reductions in the number of inpatient stays) and across many types of health care services including: home health, inpatient, and imaging, not all of which have been analyzed in prior Medicare Advantage spillover studies. Our results suggest that spillovers from Medicare Advantage to Traditional Medicare have

Influenza pandemics can place considerable burden on affected health systems by straining hospitals’ capacities due to surges in the number and cost of inpatient admissions. Substantial variations exist in estimates of hospital admissions and their costs due to the 2009/10 influenza A/H1N1 pandemic. Previous studies assessing the impact of pandemics on hospitalizations have either analyzed the subgroup of laboratory-confirmed H1N1 patients, thus underestimating the pandemic cases, or they included ordinary seasonal influenza in their analysis, thus overestimating the pandemic cases. The objective of our study is to provide robust estimates of the overall and age-specific weekly H1N1 admissions and costs between June 2009 and March 2011 in 170 English hospitals. We use routine hospital administrative records of all patients admitted for influenza-like illnesses. Since our data does not allow us to distinguish seasonal from pandemic influenza cases, we use time series models and pre-pandemic (2004-2008) admission data to establish a counterfactual of expected weekly seasonal influenza admissions over the pandemic period. We calculate the weekly number and

We find that there were two distinct waves of pandemic admissions. The first wave coincided roughly with the official pandemic period, with 10,348 excess admissions and £20.5 million secondary care costs between June 2009 and March 2010. The second wave occurred after the pandemic had been declared over by the World Health Organization. Although it was much shorter (November 2010 – March 2011), there were more admissions compared to the first wave – 11,775 – costing £24.8 million. Patients aged 0-4 years had the highest H1N1 admission rate, and patients aged 25-44 and 65+ years had the highest costs. Our estimates are over 4 times higher than those formerly reported, suggesting that the pandemic’s burden on secondary care has been previously underestimated. Our findings support improvements in pandemic preparedness and demonstrate the value of surveillance using data on routine hospital admissions as a possible tool. These results can help hospitals manage unexpected surges in admissions and resource use.

Medicare’s voluntary bundled payments program has been heralded as a potential solution to align incentives and curb rising health care expenditures. Interest in acute care based bundled payments for major lower extremity joint replacement has been particularly high, with multiple cohorts of providers enrolling from October 2013 through October 2015. Early studies of the program show that bundled payments resulted in provider behavior change, leading to decreased spending for inpatient post-acute care. This paper augments previous studies by examining whether spending reductions persist over time, and whether behavior changes by bundled payment providers were specific to the treatment of Medicare patients or were more broadly implemented, spilling over onto commercial patients.

We examined 90-day, post-discharge spending for hospital-based bundled payments for Medicare and commercially insured patients treated for major lower extremity joint replacement from January 2012 through March 2016. We used claims data from the Michigan Value Collaborative, a statewide, 76-hospital consortium. We compared changes in spending by patients admitted to bundled payment providers, compared to patients admitted to non-participating control providers, using difference-in-differences analyses. Our treatment group consisted of 4,687 episodes from five early entrants (January 2014 provider enrollees) and 7,569 episodes from five late entrants (April 2015 provider enrollees). First, we examined dynamic effects of bundled payments on Medicare spending over a period of 27 months for early entrants. Second, we compared changes in Medicare spending between early and late entrants to determine whether there were heterogeneous cohort effects. Third, we assessed whether bundled payments led to spillover effects on spending.

In difference-in-differences comparisons against episodes in the baseline period from 2012 through 2013, we found no statistically significant differences in average 90-day post-discharge Medicare spending in early entrant hospitals during the first year of bundled payments, but statistically significant but non-persistent decreases in 2015. Medicare spending in early entrant hospitals in the first quarter of 2015 declined by $887 (SE: $446; p<0.10) more than control episodes in the same periods, and declined by $1,270 (SE: $608; p<0.05) in the subsequent six-month period. We observed a positive but imprecisely estimated $298 increase (SE: $1,176) by late 2015. We found no evidence of heterogeneous effects on Medicare spending between early and late entrants (F-statistic, 0.04; p=0.85), nor consistent difference-in-differences estimates indicative of spillovers between Medicare and commercially insured patients.

Participation in bundled payments led to short-lived reductions in post-discharge spending. The effects of bundled payments were mitigated as non-bundled payments hospitals appeared to adopt alternative cost containment efforts and spending reached equally low levels, making it more difficult for bundled payments providers to achieve differential savings in later years. Finally, our findings suggest that behavior changes from bundled payments providers were isolated to Medicare beneficiaries and did not lead to broad-based practice change. Ultimately, the future success of bundled payments will hinge on its ability to incentivize continuous improvement and its role in transforming practice culture

The 2009 Health Information Technology for Economic and Clinical Health (HITECH) Act devoted $28 billion through its “meaningful use” incentive program to providers that adopted Electronic Health Records (EHRs) and participated in Health Information Exchange (HIE). Underlying these substantial investments is a belief that timely transfer of standardized electronic health information such as laboratory results and clinical summaries across the care continuum can facilitate coordinated patient care and improve health outcomes. Despite the recent growth of HIE and its potential benefits, only a few studies have examined the impact on quality of care in inpatient settings, and, to the best of our knowledge, no US-based studies have reported on HIE impacts on the tradeoff between readmission and a variety of quality measures. We conducted a large scale retrospective study to examine the impact of HIE engagement on individual patients’ outcomes of Acute Myocardial Infarction (AMI) that is directly targeted by the Hospital Readmissions Reduction Program and a variety of other Centers for Medicare and Medicaid Services (CMS) programs, including the Hospital Value-Based Purchasing Program, and the Bundled Payments for Care Improvement (BPCI) Initiative. We linked the Florida State Inpatient Discharge (SID) data, which allows us to track a patient’s longitudinal visits across hospitals, with the American Hospital Association Annual and Information Technology Supplement surveys. Using a difference-in-differences (DID) estimation approach, we compared changes in outcomes of a treatment group of targeted admissions before and after HIE engagement, relative to changes in outcomes of a control group that never participated in HIE. Our main outcome measures are the 30-day, 45-day, and 60-day all-cause readmission rates. To investigate whether the changes in readmission came at the cost of other quality measures, we also analyzed the impacts on length of stay, total charges, total number of procedures, discharges to a nursing facility or home health care, and in-hospital mortality. Our models adjust for patient characteristics and include hospital specific fixed effects to control for unobservable confounding factors. We employed placebo tests to rule out the concern that changes in outcome measures may have already started in time periods prior to the participation of HIE. Overall, we found that HIE engagement did lower 30-day all-cause readmission rates for AMI patients. The decrease in readmissions after HIE engagement primarily came from reduced readmission to a different hospital. In addition, associated with the reduction in readmission were the rises in length of stay, number of procedures, and total charges, but there were no statistically significant changes in transfer, discharge destination or in-hospital mortality. These results suggest that the decrease in readmission was achieved through the increased treatment intensity of inpatient care, but was not due to any strategic transfers or changes in discharge destination. HIE may have played an important role in determining the optimal cost tradeoff between inpatient care and readmission. A back-of-the-envelope calculation reveals that, for AMI condition alone, the HIE participation in Florida hospitals reduced 235 avoidable


: Asthma is a chronic disease that affects quality of life, productivity at work and school, and healthcare utilization; although controllable, it can even result in death. To control asthma symptoms and prevent severe asthma attacks, the evidence-based guidelines developed by medical professionals should be followed. These recommendations include assessment of asthma severity, prescribing and ensuring adherence to asthma control medications, providing asthma self-management education, and identifying and avoiding ambient and indoor environmental triggers. In this paper, we estimated the effect of health insurance status, race/ethnicity, and income on asthma-related incremental

The primary source of data was the 2008-2013 household component of the Medical Expenditure Panel Survey. We defined treated asthma as the presence of at least one medical or pharmaceutical encounter or claim associated with asthma. For the main analysis, we applied two-part regression models to estimate asthma-related annual per-person incremental medical expenditure (APIME). Routine outpatient care was defined as scheduled

During 2008-2013 the national average of APIME in the United States was $3,266 (in 2015 US dollars); more than 80% of that amount was attributable to prescription medication and routine outpatient care, while roughly 20% was attributable to hospitalizations and emergency room (ER) visits. APIME was significantly lower than the national average for uninsured persons, Blacks, Hispanics, and persons whose income was equal to or above the national poverty level. Conversely, APIME was higher than the national average for insured persons, Whites, Asians, and for those whose income was below the national poverty level. Studies show that use of ERs and hospitalization services, previously a major driver of high total medical expenditures for asthma, occurs more often among uninsured persons, Blacks, and Hispanics. Our study shows that prescription medications and routine outpatient care are comprising an increasingly large proportion of total asthma care expenditure. Uninsured persons may tend to seek care through use of the ER or hospitalization services rather than through routine outpatient care and filling prescription medications, which lowers overall

Our results also show that persons with income below the national poverty level have higher APIME than persons in higher income brackets. These persons are more likely to live in areas with a higher concentration of outdoor and indoor environmental asthma triggers, which are a major cause of asthma attacks. On the other hand, these individuals are more likely to qualify for Medicaid, which may facilitate access to a wider range of routine and urgent medical services,

Lack of health insurance hinders access to prescription medications and routine outpatient care and, as a result, contributes to higher use of ER visits and hospitalizations for persons with asthma. Persons with income lower than the national poverty level have higher APIME and, despite being potentially eligible for Medicaid, may need additional financial support, such as health insurance reimbursement for environmental interventions, to maintain an indoor

In this study, we compare and align health care expenditure estimates from the Medical Expenditure Panel Survey and the National Health Expenditure Accounts. Reconciling MEPS and NHEA estimates serves two important purposes. First, it is an important quality assurance exercise for improving and ensuring the integrity of estimates from both sources. Second, the reconciliation provides a consistent baseline of health expenditure data for policy simulations. MEPS is often used in developing microsimulation models because it contains person-level expenditures. Reliable estimates of national health spending need to be used as a baseline for analyzing the impact of potential policy changes on health care costs. Based on results from our study, analysts can adjust MEPS to be consistent with the NHEA so that the projected costs as well as budgetary and tax implications of any policy change are consistent with national health spending estimates. Previous reconciliations have been used as the baseline by the Department of Health and Human Services, the Congressional Budget Office, RAND and other researchers in simulating the impact of potential policy changes. The NHEA and MEPS both provide comprehensive estimates of health care spending in the U.S. The NHEA is primarily based on aggregate provider revenue data and administrative records of publically administered programs and covers the entire U.S. population and a full range of health care expenditures, including personal health care spending, public health services, research, and investment in structures and equipment. NHEA estimates are produced annually in the U.S. by the Office of the Actuary at the Centers of Medicare and Medicaid Services (CMS). MEPS, on the other hand, provides person-level information on health expenditures from a nationally representative sample of households in the civilian, non-institutionalized population. MEPS is produced by the Agency for Health Care Research and Quality (AHRQ) and the National Center for Health Statistics. The reconciliation is conducted every five years when the quinquennial Economic Census is available as it is the only data source that contains expenditures at the required level of detail so that specific expenditures reported in different service categories in NHEA and MEPS can be aligned. . The previous reconciliations were conducted for 1996, 2002 and 2007 (Selden, Levit and Cohen et al. 2001, Sing, Banthin and Selden et al. 2006 and Bernard, Cowan and Selden et al. 2013). Although each source provides a measure of total national spending on personal health care (PHC), at first glance the estimates appear to diverge significantly. We make adjustments to account for the differences in underlying populations, covered services and other measurement concepts to reconcile the expenditure estimates. Once we adjust the NHEA to make it consistent with MEPS, we compare and discuss potential reasons for the differences for each service category and source of payment. We also discuss how the expenditure estimates have changed since the previous reconciliation in 2007. Identifying service types and sources of payment with larger gaps helps AHRQ and CMS focus future research

How organizational changes relate to firm performance has been the focus of research across disciplines. However, despite its importance, empirical evidence is scant. In healthcare most studies focus on the impact of ownership on performance, but evidence is mixed. We argue that this ambiguity is due to two reasons: first, these studies compare performance across firms that operate in heterogeneous rather than homogeneous markets. Second, there is little assessment of how organizational change per-se relates to performance. We, on the other hand, assess the performance effects of organizational change in the context of a homogenous healthcare market by analyzing its impact on hospital costs and the extent to which efficiency gains from such change interact with scale and scope of hospital services. We also explore how such gains vary with planned vs. unplanned hospital activities and hospital heterogeneity, namely: hospital functional diversity and relative performance. Exploiting detailed 2001-2008 panel-data for English hospitals and the introduction of the Foundation Trust policy that triggered major organizational change - we find that hospitals exhibit economies of scale, but not scope; hospitals that underwent organizational change are more efficient than those that did not; and the organizational change facilitates economies of scope but not scale. However, efficiency gains vary importantly with hospital heterogeneity. Our results suggest that, the FT policy enabled cost-efficiencies, especially for worst-performing and less functionally diverse hospitals. This highlights that organizational changes can be

In this paper we investigate to what extent the childhood healthcare environment influences later life health outcomes. We examine a fundamental re-organisation of the healthcare environment in the U.K., which occurred through the introduction of the National Health Service (NHS) in July 1948. Immediate large decreases in infant mortality of 17% ensued, which were focused on the neo-natal period and larger for individuals who prior to the NHS had a lower access to

Data: We combine historic county-level data with the Office of National Statistics Longitudinal Study of linked census records combined with administrative mortality data, and a large new dataset - the UK Biobank - recording health measurements linked to administrative hospital records to assess the long run impact of birth exposure to the NHS on health and mortality 50 to 60 years after its introduction. Method: As the NHS was introduced nationwide on a single date, we employ a Regression Discontinuity Design, where we will allow for preexisting trends in the outcomes to be different either side of the threshold (i.e. the timing of the NHS introduction).We combine this method with geographic variation in access to medical services through the NHS. Findings: Our findings indicate that survival rates are systematically higher among lower class individuals whose maternity care expanded through the NHS, with the magnitude of the effectincreasing monotonically with age and becoming statistically signicant from age 57 onwards. The increase in the benecial impact of the NHS on survival rates in this population group represents a 12% reduction in mortality (and a 1% increase in survival) at age 57. We supplement these findings with analysis of hospital records, which reveal a similar decrease in hospitalisations for cardiovascular disease, one of the major causes of death, for lower class individuals. Our results suggest that the expansion to universal healthcare (and individual exposure to this universal system at birth) leads to a narrowing in the mortality gap between social classes at older ages.

Before the Affordable Care Act (ACA), people with chronic conditions were typically denied coverage or faced high, experience-rated premiums or preexisting condition exclusions in the nongroup market. Expanding access to nongroup coverage for these individuals while keeping premiums affordable was thus a key objective of the ACA. Recent policies threaten to undermine ACA provisions designed to include healthier and sicker individuals in a single risk pool, yet relatively little is known about the medical needs of people with nongroup coverage who would be affected by these policies. For this study, we examined the health status and health care experiences of adults covered

The study draws on 2012-2015 Medical Expenditure Panel Survey (MEPS) data and focuses on adults ages 18 to 64. We analyzed changes over time in nongroup coverage for this age group. We then estimated changes between pre- and post-ACA implementation periods in the treated prevalence of chronic conditions among adults with nongroup coverage, based on diagnosed conditions that were linked to health care provider visits and prescription drug fills. We also compared treatment for chronic conditions by coverage type (Marketplace, other nongroup, employer-sponsored, and public); other measures of interest included disability status, service use, spending, and sources of payment for care. Because open enrollment periods vary by coverage type, the analysis focused on service use and treatment occurring in the last six months of the year among those with continuous coverage during that period.

The share of nonelderly adults reporting nongroup coverage more than doubled following ACA implementation, with all enrollment growth occurring through the Marketplaces. Between the pre- and post-ACA implementation periods, there were increases in the shares of nongroup enrollees who were treated for multiple chronic conditions and who were in the top decile of spending for this age group. These changes were driven primarily by the poorer health of adults with Marketplace coverage, many of whom were uninsured prior to ACA implementation. In 2014-2015, nearly 45 percent of Marketplace enrollees were treated for a chronic condition during the reference period, compared with 35 percent of those with non-Marketplace nongroup coverage and 38 percent of those with employer coverage. Relative to other privately insured adults, those with Marketplace coverage were more likely to have been treated for multiple chronic conditions and had higher service use, and most of their spending was covered by private insurers.

The Marketplaces expanded coverage for adults with chronic conditions, but their higher service use has contributed to rising nongroup premiums. Policymakers seeking to address this challenge face a choice between proposals that aim to strengthen the ACA’s risk pooling arrangements and proposals to concentrate the risk of high costs among those with the greatest medical needs. The outcome of these policy decisions will have a significant impact on a vulnerable population of adults who depend on Marketplace coverage to treat their chronic conditions.

: Mercy Health Center (MHC) is a faith-based health resource center serving the underserved population across six Georgia counties. The staff and volunteers provide free healthcare to the uninsured including primary care, pharmacy, dental, chronic disease management, and behavioral health counseling. The purpose of this analysis is to explore the impact that MHC has on the healthcare utilization of its patients over time and the associated healthcare

: A cohort of 185 adult patients was identified, and 27 months of hospital utilization was recorded (9 months pre-MHC and 18 months post-MHC). MHC utilization during this period was also recorded. After attaching unit costs to MHC visits, emergency department usage, and outpatient services, we analyzed healthcare utilization and costs over time. Simple analyses included non-parametric longitudinal comparisons of utilization by category and of total healthcare costs. Some simple assumptions were made regarding the trajectory of healthcare utilization beyond the 18-month follow-up period. Further, a recurrent event survival analysis of each category of healthcare utilization was conducted

: Emergency department utilization decreased and outpatient services increased after patients gained access to primary care through MHC. The primary healthcare category of healthcare savings was found to be from a reduction in emergency department visits. However, estimated healthcare cost savings were not large enough during follow-up to offset MHC costs. Overall, results were confirmed in the survival analysis as the risk for emergency department

: Though net cost savings were not realized within the first 18 months at Mercy for this cohort, we would expect cumulative net cost savings to begin to accrue after a patient’s second year at MHC. Our results suggests that investment in a local free clinic can decrease unnecessary emergency department utilization and eventually lead to cost savings for the healthcare system.

: Patient data were gathered from two local hospital databases, as well as from MHC, leading to three specific limitations. First, verification of patient residence in the service area during the full 27-month period was not possible. Second, we did not have data on healthcare utilization outside of the service area or from local private clinics. Finally, our analysis did not consider medication costs, as these data were not available to the research team.

One third of Medicare beneficiaries are now enrolled in private plans through Medicare Advantage. After years of growth in federal payments to Medicare Advantage plans, the Affordable Care Act (ACA) slowed or cut such payments. To date, little is known about the impact of these ACA-related payment changes on plan behavior and on benefits provided to beneficiaries. We examined how plans responded to ACA payment reductions relative to their response to pre-ACA payment increases, which could help reveal whether plans are operating above their costs and inform policymakers regarding future payment policy.

We used 2006-2015 data from the Centers for Medicare and Medicaid Services (CMS) to examine the impact of changes in the maximum federal payments to plans (the “benchmark”) on plans’ asking prices (their “bids”) and on benefits received by beneficiaries (the “rebate”) before and after the ACA. This rebate, which equals a portion of the difference between the bid and the benchmark for plans that bid below the benchmark, must be passed on to beneficiaries in the form of lower premiums or additional benefits including reductions in out-of-pocket costs, reductions in drug costs, and increased coverage for vision, dental, and hearing services. We also assessed differences in plan behavior among plans facing larger benchmarks as compared with smaller benchmarks. Analyses used longitudinal models that exploit the variation in benchmark changes before and after the ACA benchmark cuts, adjusted for beneficiary risk, market concentration, fee-for-service Medicare spending, and fixed differences across counties and across years.

In real terms, average monthly Medicare Advantage benchmarks grew by $35 before the ACA (2006-2009) and decreased by $81 after the ACA-related benchmark cuts (2012-2015). Before the ACA, for every $1 increase in the benchmark, <0.001) on average. After the ACA, plans lowered their bids by $0.57 on average for every $1 decrease in the benchmark (p=0.03). This symmetrical bid

response after the ACA lessened the resulting decline in beneficiary rebates. Moreover, declines in final plan payments and beneficiary rebates were further offset by new bonuses from quality incentives and increases in beneficiary risk scores. Within rebates, after the ACA plans reduced benefits by about twice as much on the margin as they had raised benefits before the ACA for each dollar change in the benchmark. However, plans changed premiums by similar amounts in response to benchmark changes pre- and post-ACA. Plans in more competitive markets were less responsive to benchmark changes than plans in less competitive markets, implying that plans in more competitive markets may

In contrast to before the ACA, Medicare Advantage benchmarks decreased after the ACA. Plans responded to these cuts by lowering their bids, suggesting that plans were operating above cost. This plan bid response, combined with additional payments due to quality bonuses and growth in risk scores, helped lessen the decrease in beneficiary rebates, which may explain the continued growth in Medicare Advantage enrollment after the ACA.

How effective has the Community Health Assist Scheme (CHAS) been in reducing market failure in Singapore’s healthcare sector? The CHAS policy, introduced in 2012 in Singapore, aims to improve accessibility and affordability of healthcare by offering subsidies to low and middle-income groups and elderly individuals for general practice consultations and healthcare. The investigation was undertaken by acquiring and analysing primary and secondary research data from 3 main sources, including handwritten survey responses of 334 individuals who were valid CHAS subsidy recipients (CHAS cardholders) from 5 different locations in Singapore, interview responses from two established general practitioner doctors with working knowledge of the scheme, and information from literature available online. Survey responses were analysed to determine how CHAS has affected the affordability and consumption of healthcare, and other benefits or drawbacks for CHAS users. The interview responses were used to explain the benefits of healthcare consumption and provide different perspectives on the impacts of CHAS on the various parties involved. Online sources provided useful information on changes in healthcare consumerism and Singapore’s government policies. The study revealed that CHAS has been largely effective in reducing market failure as the subsidies granted to consumers have improved the consumption of healthcare. This has allowed for the external benefits of healthcare consumption to be realized, thus reducing market failure. However the study also revealed that CHAS cannot be fully effective in reducing market failure as the scope of CHAS prevents healthcare consumption from fully reaching the socially optimal

CHAS has been effective to a large extent in reducing market failure in Singapore’s healthcare sector, albeit with some benefits to third parties yet to be realised. There are certain elements of the investigation, which may limit the validity of the conclusion, such as the means used to determine the socially optimal level of healthcare consumption, and the survey sample size.

CHAS is a government subsidy, introduced in 2012 in Singapore, to improve accessibility and affordability of healthcare to low and middle income groups and elderly individuals for general practice consultations and healthcare. As the examine the effect of the CHAS subsidy on income inequality.

The Lorenz curve and corresponding Gini coefficient values were used to study the effect on income inequality. Curves were plotted to represent income distribution in 3 separate groups: (a) The nation as a whole for incomes up to $19,999 per month per household, using publically available government data. (b) The same data to which is applied theoretical maximum claims by CHAS-eligible households. Information on maximum claims was obtained from the CHAS website. (c) Forty-eight subjects who provided real world data on their income and CHAS subsidy claims. Gini coefficients were calculated for the corresponding Lorenz curves. The paired t-test was used to determine whether differences

Findings from the Lorenz curve and Gini coefficient data showed that in a theoretical situation of maximum CHAS subsidy claims, there would be a clear decrease in income inequality (change in Gini coefficient = 0.036, p<0.0001). For the real-world data, CHAS claims created a small but statistically significant reduction in income inequality (change in Gini coefficient = 0.005, p<0.0001). There was a small but visible shift of the Lorenz curve.

CHAS subsidy has redistributive effect, with the potential to reduce income inequality at a national level. This effect could be potentiated by greater use of the CHAS scheme by eligible patients. A number of caveats were identified in making the conclusions. First, theoretical maximum claims far exceed actual usage in the population surveyed, and may not represent the real world situation. Second, the study does not account for monthly households with income above $19,999, which make up 12% of the national population. The data therefore does not account for the entire population. Third, the real world data sample size is small and a much larger survey

Eligible individuals in states that expanded Medicaid have reported gains in health care access measures, including having a personal physician, affordability of care, and insurance coverage. However, this change in insurance eligibility led to increases in demand for health care, which raised the question of whether the Medicaid expansion would exacerbate clinician shortages. This paper examines several aspects of the health care workforce across states with different expansion status. First, we assess whether expansion and non-expansion states have different baseline levels of workforce capacity. Then, using a difference-in-differences approach, we test whether the Medicaid expansion had any effects on the number of physicians, nurses, physician assistants, and other health care professionals working in these states. Lastly, we utilize a more granular approach by studying whether the expansion had differential effects within states on


Post 2014 expansion, we find no evidence that the Medicaid expansion had any significant impact on the available health care workforce post-reform across multiple categories of clinicians. County-based analyses are in process but will be

Our results have important economic and policy implications. While Medicaid expansion thus far had led to improvements in access to care across multiple studies, the remaining non-expansion states may have different experiences should they expand given significantly smaller health care workforces at baseline. Furthermore, at present we do not find a Medicaid expansion effect on the health care workforce, suggesting there has been little effect of expansion on short-term entry or exit into health professions, or selective migration by clinicians into (or away) from Medicaid expansion states. However, these effects could manifest in the future, given the lengthy educational requirements within

In the conventional framework for designing health plan payment models, the regulator chooses variables to be used as risk adjustors, the risk adjustment weights, and other policy parameters, but the data from which estimates are derived are taken as given. This approach implicitly assumes the observed spending patterns are optimal. In this paper we explore an entirely novel approach: using the data itself as a policy tool. We take the risk adjustors, the estimation method, and other plan payment features as given, and change the data used for estimation to achieve a policy objective. We develop a general model for the provision of health care services by health plans. The key insight of our model is that there is a two-way relationship between plan actions and health plan payment: plan actions (outcomes) are a function of health plan payment, and plan payments are a function of the insurer actions the payment system is meant to affect. Importantly, we show when plan payments are calibrated on data generated by plan actions, equilibrium is when plan actions lead to a set of prices (via the algorithm) that induce the current (possibly inefficient) health care system. Using Medicare data we apply these ideas to two areas of misallocation in health care: undercompensation for individuals with mental health diagnoses and disparities in health care spending between high and low income groups. We transfer spending to the group of interest, re-fit the risk adjustment model on the modified outcome, and illustrate the relationship between the transfer amount and targeted measure. We show spending can be transferred between disease groups to eliminate undercompensation with a minimal impact to overall fit of the risk adjustment model, while correcting disparities requires shifting much larger amounts of spending.

While the Affordable Care Act (ACA) in the U.S. has led to significant gains in health insurance coverage and access to care, less is known on how this policy change affects non-geriatric cancer patients’ utilization and outcomes of emergency departments (EDs). Previous studies demonstrate that younger patients and those who lack access to usual source of care, in general, are at an increased risk of using EDs. Similarly, patients with Medicaid or no insurance use EDs more often than those with private insurance. Within cancer patients, adolescents and young adults (AYAs), aged 15-39 years, are more likely to experience insurance- and cost-related barriers to care. Yet, it is unknown how the ACA has affected ED use outcomes for AYAs with cancer when compared with children and older adults with cancer.

We examined changes in ED use and outcomes for AYA cancer patients compared with those of children and non-geriatric adults with cancer in the U.S. before and after the implementation of the ACA.

We used the 2013 and 2014 National Emergency Department Sample (NEDS) from the Healthcare Cost and Utilization Project. Our subpopulation consisted of cancer patients (any cancer diagnosis) currently aged 64 years and younger. We compared patients’ demographic (e.g., sex, primary payer, county of residence) and clinical (e.g., number of chronic conditions and procedures, cancer diagnosis) characteristics, and hospital characteristics (trauma designation, teaching status) between years. We also examined the ED outcome (treated and released vs. admitted to the same hospital). Variables were examined for the overall sample and also stratified by age categories (i.e., children 0-14, AYAs 15-39, and adults 40-64). Chi-squared tests compared proportions, and logistic regressions were used to identify factors that affected the ED outcome. All analyses were weighted.

Overall, cancer patients accounted for over 1.8 million ED visits in 2013 and 1.9 million visits in 2014. A significant decrease was observed in self-paid visits from 2013 to 2014 (9% to 6%). Self-paid visits decreased from 16% to 12% (p<0.001) for AYA visits and from 8% to 5% (p<0.001) for adult cancer visits, however, no differences were observed in primary payer of ED visits by children (2% in both years). Within each group, ED visit outcome did not differ between years. Yet, in each year, AYAs were more likely to be released than admitted (69% vs. 31%) compared with children (58% vs. 42%) and older adults with cancer (51% vs. 49%) (p<0.001). In our adjusted models that accounted for other covariates, in both years, self-pay AYA cancer visits were less likely to be admitted compared with children.

While self-pay AYA visits decreased in 2014 following implementation of the ACA, self-pay visits remained significantly higher for AYAs (12%) as compared to children/adults. Our analyses demonstrate that the ED outcome differs depending on health insurance status. This is especially true for AYA cancer patients who were more likely than children/adults to be uninsured and, thereby less likely to be admitted to the same hospital following an ED visit.

Transportation can be a significant barrier to attaining healthcare services, in particular for disadvantaged populations. We exploit a reform that introduced public transportation services to Arab towns in Israel to evaluate the effect of greater access to healthcare services via public transportation among a disadvantaged population on health outcomes. In 2007, the Israeli ministry of transportation (MOT) announced a reform within non-Jewish communities in Israel: the introduction of public transportation (PT) to these communities. Until then, Non-Jewish communities had been significantly deprived of PT infrastructure, with generally no official services. Furthermore, private car ownership rates are relatively low among Arabs, and many women do not have a driving license due to traditional barriers. The new bus network, which gradually developed over the next 7 years, represented a substantial increase in access to healthcare services. Health services in the form of the best doctors or specialty clinics and hospitals are found in Israel outside Arab

We use very detailed data from the Israeli MOT documenting the frequency of all bus lines in Israel, their routes and bus stops, for 2008-2014 on a bi-annual basis. We matched our measures of bus frequencies for each town and period to a survey of the Arab population in Israel conducted in 2004, 2007, 2010 and 2014. Our analysis focuses on the elderly population - ages 50-70 - based on the health conditions inquired about in the survey - high blood pressure, diabetes, heart problems, high cholesterol, and back problems. The questions in the survey specifically ask about diagnosis and receiving medical treatment for these health conditions.

Our initial results show statistically significant increases in the diagnosis of heart disease, high cholesterol, asthma, back problems, and migraine headaches among the population aged 50-70 when the penetration of buses to these individuals’ communities increases. We do not observe statistically significant changes in response to public transportation penetration for diabetes and high blood pressure. We observe some differential effects based on respondents’ sex. We also observe elderly individuals reporting that they are overall less healthy when public transportation is greater in their community.

A naive interpretation of these results can suggest that public transportation penetration is adversely affecting health outcomes among the elderly population. We believe that given that the survey inquires about diagnosis of these health conditions, a more plausible interpretation of the results is that there are greater diagnosis levels and awareness of health conditions that were existent prior to PT penetration but were unobserved due to lower access to healthcare services. Our results will be further corroborated by deceased records we have obtained showing that deaths do not increase in response to public transportation penetration.

In China and other developing countries, long waiting times during a hospital visit are pervasive due to the rapidly growing demand for health care services in already overcrowded hospitals. However, technological innovations have been introduced to reduce patients’ waiting times in hospitals and to streamline the process of healthcare services. Such innovations have the potential to improve the efficiency and quality of health care as well as increase patient satisfaction.

allowing the patient to schedule appointments online and to pay for medical care efficiently. The Resident Card allows patients to schedule an Resident Card also enables automatic e-payment at physicians’ office and through self-service machines; therefore patients no longer need to

wait in line to pay medical bills. Government statistics show that patients save about 45 minutes on average during each hospital visit. We use a large medical claims data set from a major hospital in a Chinese city of 9 million people to analyze the impact of the Resident Card on patients. We have records of 4 million outpatient transactions during 2011-2013. This hospital

since 2012. We use the standard difference-in-differences approach to compare changes in patient health-seeking behavior and health outcomes between the treatment groups (i.e. those using the Resident ) and the control group. The key parallel trends assumptions are satisfied for both health seeking and health outcomes.

Linking each patient by their identifier, we find that compared to the control group, those using the card experience a greater reduction in waiting time for hospital visits and an increase in service use. These two impacts seem to result in a decline in the severity of the health problems being treated due to the greater accessibility. Also, the almost fixed supply of physicians in the short term and the increase in demand for health services give doctors more power relative to patients and payers. However, with government established prices, they make profit by prescribing unnecessary medications and more tests. Moreover, the new transaction technology makes patients’ type of insurance more visible to doctors, further intensifying their ability to make profit, especially from card users with more generous insurance coverage. This study highlights the multiple, intended and unintended consequences of this major reform in medical expenses transactions. The Resident Card produces both benefits and costs, and also systematic heterogeneity across populations.

Healthcare fraud can represent upwards of hundreds of billions of dollars in spending that could be better spent on patient care. There is often not sufficient detail on the underlying methodologies and data samples that lead to fraud estimates, which may be due to different purposes of these reports or the need to obscure the details of fraud detection methods to prevent fraudulent operators from responding to existing methods.

The objective of this study was to provide a systematic evaluation and synthesis of the methodologies and data samples used in current peer-reviewed studies on characterizing healthcare fraud. The academic databases searched were Academic Search Complete, Business Source Complete, EconLit, Medline (EBSCO), OneSearch, ProQuest Business Collection, ScienceDirect, and Web of Science. Governmental and

This examination was conducted using a systematic review methodology to identify relevant studies and determine their relevance. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement was used to guide the performance of reviewing the literature. Study criteria for eligibility were collected by applying specific search terms: healthcare, health insurance, Medicare, Medicaid, Obamacare, Affordable Care Act, or health services; fraud, cheat, falsification, corruption, or kickback; detect, detection, prevent, prevention, deterrence, audit, or auditing. Results were restricted to scholarly journals, academic journals, working papers, and conference proceedings. Study selection occurred through two independent reviews of each study for inclusion or exclusion. Disagreements between reviewers were resolved through discussion by the entire research team.

Our search terms resulted in 450 articles that were potentially appropriate for inclusion in our report. The results of independent reviews ended with twenty-seven studies considered as relevant to include after the application of our inclusion criteria. Variables are identified from the literature to synthesize each method of fraud detection used.

One limitation of this study is that the strength of the evidence is reliant on the quality and number of studies previously performed on the topic. Another limitation is the quality of studies with regard to their applicability to different types of insurers. Finally, the majority of studies could not provide proof of intent to commit fraud.

A limited number of validated methods are used to detect healthcare fraud. The literature on this topic is spread among several academic fields. The majority of available studies utilize public or social health insurance systems such as Medicare or Medicaid in order to study fraud. The main gaps we identified are validation of existing methods and proof of intent to commit fraud in the studies analyzed.

Our insurer agnostic approach examines the availability and effectiveness of healthcare fraud analytic methods across different types of health insurers, posing great value for members of the health sectors.

Over the past several years, there has been a substantial shift of Medicare enrollment from the traditional, government-administered Medicare program (“traditional Medicare”) to the private, subsidized, and mostly managed care plans offered in Medicare Advantage (MA). While previous work has examined the effect of this trend on the efficiency of care and on beneficiary health outcomes overall, my study is one of the first to consider the implications of increasing MA enrollment for a particularly disadvantaged group of beneficiaries known as dual eligibles (i.e., Medicare beneficiaries who also receive full or partial Medicaid benefits). Dual eligibles merit special attention because they are economically vulnerable, often have significant health needs, and face unique barriers in navigating the health system (e.g., as about three-fifths report having cognitive impairments). Further, while the share of Medicare beneficiaries enrolled in MA has risen overall, the increase has been particularly dramatic among dual eligibles, rising from 1% in 2004 to 32% in 2015. This study evaluates the effect of MA penetration on the number and length of hospital stays (including potentially-preventable admissions) and all-cause mortality rates among dual eligibles. I rely on complete Medicare enrollment and MedPAR files and public MA data from 2009 through 2015. These sources provide detailed information about dual enrollment and capture all hospital discharges at the vast majority of acute care PPS hospitals and all beneficiary deaths. The discharge data provide enough granularity to identify potentially-preventable hospitalizations based on AHRQ Quality Indicators.

My primary identification strategy relies on a regression discontinuity design used by a prior study to evaluate outcomes among the overall Medicare population (Afendulis, Chernew, and Kessler 2017). This approach exploits a discontinuous jump in average benchmark payment rates for MA plans – which is subsequently associated with a sharp increase in MA enrollment – in metropolitan statistical areas that exceed a population threshold. Preliminary results suggest that this increase in plan payments is also associated with a jump in MA penetration rates among dual eligibles who receive partial benefits. I use this potentially exogenous source of variation in enrollment to explore the relationship between MA and beneficiary outcomes. Because dual eligibles may also be enrolled in comprehensive or limited Medicaid managed care plans, I also run analyses restricted to the subset of counties where such enrollment is

My findings will help policymakers understand how the increased role of MA plans has affected a vulnerable subset of the Medicare population and will be informative as states continue to delegate Medicaid benefits for dual eligibles to

Losing Medicaid coverage can have a negative effect on people’s health; it affects continuity of care, especially when they become uninsured. Although not all forms of churn are detrimental (e.g., individuals who find a new job can obtain employer insurance), these transitions can affect costs and place an unnecessary administrative burden on both states and enrollees. Study Design We follow a representative sample of 4,453 non-elderly adults who reported being covered by Medicaid in January, 2013 on wave 1 of the 2014 SIPP. Our model focuses on people’s decision to obtain and maintain Medicaid coverage. We understand churn as a change in this status. We also used first differencing to create our set of (mutually exclusive) predictors: becoming employed or unemployed, gaining or losing a job, having a wage increase or decrease, and having a newborn. We included lagged variables of every life event in our model to allow for a 2-month delayed effect. Working with longitudinal data allowed us to identify these life events, and also use first differencing variables and eliminate any bias due to time-invariant unobserved characteristics. Principal Findings We estimate that 5.1% of nonelderly adults who were enrolled in Medicaid in January, 2013 lost this coverage at some point in the following year. Most of these (70%) had at least one month of uninsurance, while the rest shifted to other

We found that becoming employed and gaining a job increased people’s likelihood of churning off Medicaid, though these changes did not affect churn immediately. If someone’s family member became employed in one month, their probability of churning out from Medicaid the same month did not change, but it increased by 1.2 percentage points (pp) the next month and by 0.7 pp in the month after. Another determinant of churn was having a child, which makes individuals more likely to churn by 1.1 pp the same month, 1.2 pp the following month, and the month after. This result is a novel finding, as prior research has not focused on this life event (most research on churn has not focused on actual changes in coverage). Discussion Our results are consistent with the theory that becoming employed or gaining a job would increase family income, which could make some enrollees ineligible. Although this would generally be viewed as a positive development, it could result in the individual becoming uninsured or underinsured, or experiencing discontinuities in care. These effects were only significant after a lag; the effect on churn does not happen within its month of occurrence. This has implications

Since women are still eligible for enrollment during the months immediately postpartum, our results could indicate that some women believe their coverage ends at birth. This is important, because if women believe they lost Medicaid

To accommodate Medicare and Medicaid budget cuts and reimbursement method innovations, hospitals have made great effort on cost containment and quality improvement. The first goal of the paper is to study the empirical relationship between healthcare service quality and hospital cost containment. Our current finding suggests that cost containment is negatively related to quality improvement. But the relationship diminished after 2013, which is the year when Center for Medicare and Medicaid Services (CMS) implemented Value-Based Purchasing (VBP) program and incorporated quality as important dimension in acute care hospital inpatient services reimbursement. Therefore, our more important goal is to examine whether VBP effectively incentivized hospitals to improve or maintain quality while reducing costs. We use two dataset primarily: Hospital Compare and Medicare Cost Report. Hospital Compare provides hospital quality measures on clinical care, patient experience, safety, and efficiency. A Total Performance Score (TPS) is calculated as a weighted average of these quality aspects. And a linear exchange function translates TPSs into VBP adjustment factor, which determines a hospital’s value-based incentive payments. Top performers are rewarded with bonus and bottom ones get a discount of full reimbursement. CMS started to reimburse acute care hospital’s inpatient services using this method since 2013. Medicare Cost Report data records hospital operation, financial management, and claims

To answer questions of the relationship between cost containment and quality change and effectiveness of VBP incentives, we designed three sets of studies: First, to observe empirical relationship between cost containment and quality change, we use VBP-eligible acute care hospital TPS change as the dependent variable and cost containment proxies as the main predictor controlling for covariates commonly used in previous healthcare service quality research. And we examine the sample in two time periods separately: financial crisis years before VBP 2007 – 2012 and VBP years 2013 – 2016. Hospitals were motivated to reduce costs in both periods but for different reasons. And we use the TPS calculation method in 2013 consistently for all years. We found that the negative relationship between cost containment and quality improvement is stronger

Second, we address whether the VBP program incentivized hospitals to improve quality while reducing costs. And we apply a ‘quasi’ Difference-in-Difference (DD) study. The VBP program is applied to all eligible acute care hospitals simultaneously since 2013, which excluded a clean control group. Therefore, the DD method is compromised by taking hospitals with less than 10% of Medicare revenue as a ‘quasi’ control group and hospitals with Medicare revenue more than 50% as the treatment group. The DD analysis shows marginally significance of the VBP factor. Besides, we plan to further examine whether the VBP calculation method change and the progressive payment adjustment affect results. Finally, we offer a possible explanation. In particular, we examine the relationship between quality change and cost containment for hospitals in different financial standings. For some hospitals, benefit of cost containment may balance or

Health economics defines the extensive margin for surgical procedures as the rate at which people with similar diagnoses and medical conditions receive various treatments. While much attention has focused on geographic variation in the extensive margin for various procedures, the expanding use of certain procedures in an increasingly aged and frail population may constitute a growing problem given limited health care resources. An important illustration of this problem pertains to percutaneous coronary intervention (PCI) and coronary artery bypass surgery (CABG) procedures. While these treatments can be life-extending in many circumstances, they are often associated with significant morbidity and delayed recovery among elderly or frail patients. In particular, prolonged rehabilitation can be required due to complications or from frailty and poor functional status at the time of presentation. Randomized trials of PCI and CABG typically exclude frail and elderly patients, particularly those with renal dysfunction, and do not report the rate of nursing home discharges and delayed physical recovery. This analysis describes current trends in coronary revascularization and subsequent care. Medicare claims data from 2006-2015 for a 20% sample of Medicare beneficiaries age 65 and older are used to described trends over time in procedure rates for CABG and PCI for all beneficiaries as well as beneficiaries with selected coronary diagnoses (e.g., acute myocardial infarction, acute coronary syndromes, unstable angina) to adjust for changes in underlying disease rates over time. The descriptive trends for inpatient procedures are calculated overall and for sub-groups (e.g., age groups, case mix severity, and receipt of hemodialysis) for fee-for-service as well as Medicare Advantage enrollees. Analyses of trends for total procedures (inpatient and outpatient) and outcomes are conducted only for fee-for-service enrollees given lack of claims for physician and post-acute services for Medicare Advantage enrollees. We use regression analysis of claims for fee-for-service enrollees receiving PCI or CABG to assess key outcomes: repeat revascularization, post-acute care including skilled and non-skilled nursing home days, hospital readmission, hospice use, post-discharge mortality,

Preliminary results show that from 2007 to 2015, rates of CABG declined overall; CABG rates increased slightly among persons aged 85 and older or persons but declined slightly among persons with more complex disease (Charlson>3). In contrast, rates of PCI were fairly stable over time but increased among persons aged 85 and older as well as among persons with more complex disease (Charlson>3). Among beneficiaries receiving coronary revascularization, overall rates of discharge to skilled nursing care increased over time while discharges to home without home health care decreased. The outcome analyses that are in process will assess one-year outcomes and resource use trajectories over a longer

Descriptions of current trends in procedure rates and outcomes including resource use enable discussion of the implications of expansion of the extensive margin for coronary revascularization. The estimates will provide valuable information to policy makers as well as healthcare professionals who routinely consider the relative risks and benefits of PCI and CABG.

The United States has a complex system for financing healthcare, combining an array of public and private components. In addition to out-of-pocket amounts paid directly to providers, healthcare financing includes: the payment of premiums directly to insurers, employer contributions for workers’ health plans, and public programs that provide health coverage and draw on state and federal tax revenues. Given the growth of health care spending and the fact that individuals ultimately bear these costs in some form, there is substantial value in developing a thorough understanding of how much individuals and families pay for healthcare, how these payments are distributed, and how the incidence

Previous attempts to understand how much individuals pay for health care typically focus on particular types of healthcare payments in isolation, such as out-of-pocket medical costs (Banthin et al, 2008) or premiums for private insurance (Gruber and McKnight, 2003). Little comprehensive analysis of equity in the finance of U.S. healthcare has been conducted, with two notable exceptions being Wagstaff et al. (1999), which examined data from 1987, and Ketsche et al.

We develop an updated analysis of healthcare finance equity using the nationally-representative Medical Expenditure Panel Survey (MEPS) combined with a variety of supplementary datasets to benchmark and to enhance our estimates. We align the MEPS distribution of income to Internal Revenue Service data, simulate a full array of federal and state income tax expenditures and state sales taxes, and report sources of financing for healthcare by quintiles of equivalent income. Our preliminary assessments cover the 2004 to 2013 time period, but our final analysis will be extended to 2015 to include the early years of the main Affordable Care Act reforms. Our systematic analysis of all major components of health spending reveals that the financing of healthcare in the United States – notwithstanding a relatively progressive structure for income taxes – is regressive. The bottom quintile of households in the United States paid 11.7 percent of their income in health payments in 2013 compared with 8.1 percent for the top 1 percent of households. However, our estimates also show that the U.S. system for financing healthcare has become more progressive over time with the share of income devoted to health spending for the bottom quintile falling 6.1 percentage points from 17.7% in 2005 to 11.7% in 2013 and the share paid by the top 1 percent of

Our preliminary study period encompasses the introduction of several significant changes to healthcare and tax policy, which are consistent with our findings of increased progressivity in the healthcare sector, including the beginning of the Part D prescription drug program in Medicare (2006), the introduction of income-related premiums in Medicare (2007), and the Additional Medicare Tax for higher-income households (2013). Although we do not yet have complete data for 2014 and 2015, the increased insurance coverage in those years likely led to further increases in progressivity.

To help Medicare beneficiaries choose among Medicare Advantage (MA) plan options, the government collects data on several dimensions of care and summarizes this information through a star rating of plan quality, measured on a scale of one to five. Prior research suggests that beneficiaries do indeed rely on star ratings when making enrollment decisions and the government has used star ratings as the basis for providing billions of dollars in bonus payments. Despite their importance, it is unclear whether these aggregated ratings of plans – based in large part on process of care, patient experience, and intermediate outcome measures – ultimately signify differences that lead to improvements in health. Indeed, there remain questions about the direct relationship between many star ratings and health outcomes, whether star rating measures correlate to strong or weak performance on other dimensions of quality, and the extent

This study evaluates the relationship of star ratings with the number and length of hospital stays (including potentially-preventable admissions) and all-cause mortality rates. I rely on complete Medicare enrollment and MedPAR files and public MA data from 2009 through 2015. These data identify whether a beneficiary enrolled in MA and, if so, the star rating of their plan if they received drug coverage (as is currently the case for approximately 90 percent of MA enrollees). They also capture all hospital discharges at the vast majority of acute care PPS hospitals and all beneficiary deaths. The discharge data provide enough granularity to identify potentially-preventable hospitalizations based on


My findings will help policymakers determine how much weight to attach to star ratings in plan regulations and payment rules and will be informative as the government continues to pilot test a quality rating system on the federally-

: Florida implemented mandatory managed care for Medicaid enrollees in April 2014 via the Statewide Medicaid Managed Care (SMMC) program to improve access and coordination of care. This program enhanced access to primary care by increasing the number of primary care providers (PCPs), and after-hour appointment availability. The research objective of this study is to analyze the impact of the enhanced access to primary care on preventable

: We estimate a difference-in-difference (DD) model, comparing the change in the number of preventable hospitalizations in Zip Code Tabulated Areas (ZCTAs) with greater improvement in access to primary care (measured by percentage change of enrollee to provider ratio in the area from 2013 to 2015), compared with the changes of that in ZCTAs with less improvement in the access to primary care after the implementation of SMMC. We control for ZCTA specific socio-demographic characteristics, fixed effects for county of residence, year fixed effects, and a county-specific linear trend. The key explanatory variable is an interaction between the indicator for being a ZCTA in the top quartile of improvement in access to primary care, and the indicator for the post period. The main outcomes are numbers of preventable hospitalizations, i.e., whether the hospitalization was for an ambulatory care sensitive condition (ACSC), per 1000 residents in each enrollees’ ZCTA. We adopt the Prevention Quality Indicator (PQIs) developed by Agency for Healthcare Research and Quality (AHRQ) to identify hospitalizations for ACSCs.

: We compiled the analytic sample from three data sources. Florida inpatient discharge data from 2010 to 2015 provided information on inpatient visit. There were 1,837,294 discharges for Florida residents between the ages of 18 and 64 with a primary payer of Medicaid insurance, and no missing values on covariates used. We stratify the data into cohorts according to ZCTA, and quarter. The final analytic sample includes 19,621 stratified observations at the ZCTA-quarter level. We supplement the analyses with enrollee to PCP ratio in each ZCTA, created from Florida Medicaid provider data repository and 2010–2014 United States Census American Community Survey (ACS).

: We find that areas with greater improvement in the access to primary care experienced reductions in the incidence of overall preventable hospitalizations of 7.2 per 100,000 residents (18.9 percent) compared to areas with less improvement. Those areas also saw reductions in the hospitalizations for chronic ACSCs (reduction of 6.9 preventable hospitalizations per 100,000 residents (25.0 percent)) in the post-implementation period relative to other

: Our results show that areas with greater improvement in the access to primary care have greater reductions in hospitalizations for ACSCs, especially hospitalizations for chronic ACSCs, compared to areas with less

: Managed care in Medicaid provides a foundation for enrolling vulnerable populations and guaranteeing them better access to primary care and care coordination, to reduce cost of the program. Our study provides direct evidence that improved access to primary care under Medicaid managed care is associated with reductions in preventable hospitalizations.

In 2017, the Centers for Medicare and Medicaid Services implemented the Merit-Based Incentive Payment System (MIPS), establishing a new payment program for clinicians participating in the fee-for-service Medicare program. As part of a broader push to link provider payments to value, the MIPS is a pay-for-performance model that intends to reward clinicians for improving quality of care and lowering spending by providing practices with bonuses or penalties based on their performance on quality and spending measures. Although the effects of the MIPS will not be known for several years, its basic design is similar to that of its predecessor, the Value-Based Payment Modifier (VM). From 2014 to 2016, the VM was phased in for physician practices meeting specific size thresholds (i.e., number of constituent clinicians), creating abrupt discontinuities in the exposure of practices to pay-for-performance incentives. We harnessed these discontinuities in a quasi-experimental regression discontinuity design to evaluate the VM's effects on performance measures assessed for all practices subject to the program.

Exploiting the phase-in of VM incentives based on practice size, we used regression discontinuity analysis and Medicare claims in 2014 to estimate differences in practice performance associated with the abrupt exposure of practices with ≥100 clinicians to full VM incentives (bonuses and penalties) and the exposure of practices with ≥10 clinicians to partial incentives (bonuses only). We repeated analyses using 2015 claims to assess the association of a second year of exposure to pay-for-performance incentives. We examined performance on four sets of outcomes: hospital admissions for ambulatory case-sensitive conditions (ACSCs), all-cause 30-day readmissions, mortality, and Medicare (Part A and Part B) spending per beneficiary. We conducted supplementary analyses to assess the robustness of our results to model specification and placebo tests to check whether discontinuities in outcomes at the VM's implementation thresholds (i.e., ≥10 and ≥100 clinicians) exceeded those at arbitrary thresholds of practice size where incentives did not differ.

In 2014, there were no significant discontinuities at the ≥10-clinician threshold in the relationship between practice size and admissions for ACSCs (adjusted discontinuity:+0.003 admissions/beneficiary; 95% CI:-0.0003,0.006), proportion of admissions with readmission (+0.1 percentage points; 95% CI:-0.4,0.6), Medicare spending ($234/beneficiary; 95% CI:-$148,$616), or mortality (+0.2 percentage points; 95% CI:-0.1,0.5). Similarly, there were no discontinuities at the ≥100-clinician threshold in admissions for ACSCs (-0.002 admissions/beneficiary; 95% CI:-0.006,0.003), proportion of admissions with readmission (+0.3 percentage points; 95% CI:-0.6,1.2), spending (-$152/beneficiary; 95% CI:-$712,$408), or mortality (-0.1 percentage points; 95% CI:-0.5,0.3). Analyses of the ≥100-clinician threshold using 2015 data revealed no discontinuities associated with a second year of full exposure to the VM. Discontinuities estimated over all practice size thresholds, and using various practice size ranges, revealed no consistent evidence that discontinuities were largest at the ≥10 and ≥100-clinician thresholds.

The VM was not associated with significant differences in performance on program measures at thresholds where physicians' incentives differed. Our findings suggest that incentives in the VM were not sufficiently strong to affect practice performance, questioning whether the similarly-designed MIPS will achieve its intended policy goals.


Ellen Green [email protected]

Nathaniel Counts [email protected]

Srimoyee Bose [email protected]

We study the impact of a merit-based incentive payment system on provider behavior in the primary care setting using experimental methods that leverage healthcare simulations with patient actors. Our approach allows us to exogenously change a provider’s incentives and to directly measure the consequences of alternative payment systems. Within our sample, we find that merit-based incentive payment systems increase the number of the incentivized

Healthcare in the United States is undergoing a transition from volume to value. Current value-based payment models incentivize providers to efficiently provide high-quality care and generate savings in the short-term. The next wave of

Healthcare payment models can take a life-course perspective on value by tying incentives to reductions in long-term actuarial risk, such that an intervention is valuable to the extent it reduces the payer’s predicted future costs. This actuarial risk-based valuation builds in the business case for how increased investment in health promotion can still be cost-neutral to the healthcare system overall. Specifically, as long as the total amount invested does not exceed the expected value of the reduced actuarial risk, the payer can be relatively confident that the intervention will be at least cost-neutral over the long-term. This approach also intrinsically risk adjusts, as larger investments are justified where there is more risk to reduce. In addition, this approach can avert incentivizing perverse outcomes by assigning monetary value to avoiding serious adverse events that may not technically cost a payer very much. Payers can integrate indicators of future reductions in actuarial risk into existing payment models by using “net present value of care” (NPVoC) rather than total cost of care in determining incentives (see Figure 1). NPVoC is the total cost of care (i.e., the lesser than anticipated amount spent on healthcare), plus the expected value of care (i.e., the amount the payer anticipates that the achieved health outcomes will save in later healthcare costs, divided by a discount rate). The expected value of care could differ from payer to payer, based on how long individuals are anticipated to remain with the plan. In instances of high market penetration, it may be five, ten, or even more years. As Medicare, Medicaid and large provider organizations initiate multi-payer arrangements (i.e., multiple health plans agree to similar and mutually beneficial payment terms) the duration of expected value calculations will lengthen, because even when

At present, payers have limited ability to accurately predict long-term reductions in actuarial risk from short-term clinical outcomes—an undeniably challenging endeavor. To begin testing and improving NPVoC models, substantial precision is unnecessary. Payers can use a conservative estimate of predicted savings associated with a modest magnitude of change on a short-term outcome, minimizing potential overall losses when predicted savings are shared in expected value-based payments. By allowing payers to calculate the expected value of an outcome, NPVoC models can avoid problems from the past where research indicated likely savings that failed to materialize when the payer took the intervention to scale. Precise estimates of future savings for NPVoC models may take years of data collection and integration, but initial models can be tested and iteratively improved to build the foundation.

About 44% of young women with Breast cancer (BC) are diagnosed with advanced stage, and it is rising at a faster rate than for older women. This study focuses on examining whether there were significant changes on the predictors of

Using the US cancer statistics registry database, we extracted all 139220 BC cases for young women (<=40 years) from 46 states to examine the variation in their late-stage BC diagnosis across the two periods. We used a random intercept logit model with person, county, state and time level covariates after controlling for factors that may moderate the effect of the ACA implementation. Results suggest that young African-American women had higher odds of late-stage diagnosis relative to the whites, and the odds increased over time. Area urbanicity, poverty, unemployment, obesity rate and Medicaid enrollment rate were associated with increased odds of late-stage diagnosis, and the odds were higher in the later period. Also, per-capita healthcare expenses, a residential diversity index, the average age of mothers at first birth, drinking and marital status rates, BC screening rate and private insurance enrollment rate were associated with lower odds of late-stage diagnosis, which increased in the later period. Thus, disparities in late-stage diagnoses of BC among young women increased after 2010, highlighting the increased importance of having better access to medical care and prevention among younger women following implementation of the ACA in 2010. As time passes, further investigation is needed to understand these persistent disparities and whether health care reform will eventually seem to

Kenneth John McConnell [email protected]

Galina Besstremyannaya [email protected]

Yevgeniy Feyman [email protected]

Background: In 2012 Oregon transitioned its Medicaid program to cover 90% of its enrollees in Coordinated Care Organizations (CCOs). CCOs can be seen as a type of Medicaid Accountable Care Organization, but include an administrative layer (similar to a Managed Care Organization) and are at full financial risk through a global budget. As part of its 1115 Medicaid Waiver, Oregon agreed to reduce its historical rate of health care spending growth from 5.4% to 3.4%. The global budget and prespecified rate of growth have analogies to per capita caps and block grants, features that were prominent in several “repeal and replace” bills that were put forth in 2017. We provide an assessment of the Oregon

Study Design. We estimated changes in inpatient care using a difference-in-differences approach and data from the Healthcare Cost and Utilization Project (HCUP) and monthly and county Medicaid enrollment data from Oregon and comparison states of Washington, Colorado, New Mexico, and Arizona. We used 1-1 nearest neighbor matching to identify similar county and demographic matches across these comparison states. Using 2 years of pre-intervention data (2010-2011) and 2 years of post-intervention data (2013-2014), we assessed the impact of the CCO model on inpatient admissions and length of stay. In subanalyses, we also assessed changes in hospitalization rates for elective vs. emergency admissions; ambulatory-care sensitive admissions, and admissions among individuals from low-income neighborhoods. Finally, since the CCO model provided care for more than one out of every four Oregonians, we tested for spillover effects among the commercially insured. We assessed parallel trends among the treated and comparison groups and found no statistically significant differences across multiple groups, including all admissions, length of stay, and

Results. Oregon’s transition to global budgets was associated with significant reductions in the rate of inpatient admissions of approximately 1 admission per every 1000 enrollees per month, equivalent to a 9% reduction in the admission rate. There were significant reductions in both the emergent and elective rate of admissions, although the changes were larger among elective admissions, and, within that group, particularly among admissions for births and maternity care. The length of stay among Oregon Medicaid enrollees was low prior to the CCO transformation and remained low in the two years following the intervention, suggesting that changes in the extensive margin were not offset by changes in the intensive margin. In subanalyses, we found significant reductions in ambulatory care sensitive admissions and admissions among individuals from low-income neighborhoods. We found no evidence of spillover in the commercial

Conclusions. Oregon’s global budgets were associated with significant reductions in Medicaid inpatient admissions relative to matched counties in western states. The extent to which the Oregon model can continue to place pressure on inpatient utilization will depend on the longer-term success of infrastructure investments and a portfolio of delivery system changes that are designed to provide care in less intensive settings. These changes may have lessons for Medicaid

The impact of policy regulation in the health sector depends on hospital technology, which is largely reflected in the productivity of labor, capital and medicines. This paper focuses on acute-care local public hospitals in Japan and examines how technology differences determined the effect of voluntary changeover by hospitals to the prospective payment system. The Japanese hospital sector provides a rare example of a nationwide introduction of the reform with self-selection. We exploit panel data conditional quantile regressions to model a range of technologies for the multi-product output function of hospitals under an endogenous treatment assignment. The analysis reveals technological heterogeneity, and incorrect labor/capital and labor/medicines mix. The impact of prospective payment on output is primarily attributed to labor and is inversely related to productivity. Finally, we contrast the design of prospective

The novelty of the present paper is severalfold. To the best of our knowledge, the paper is the first application in health economics of quantile regression models for estimating the longitudinal production function of hospitals. The analysis incorporates multiple outputs, evaluates factor returns and assesses optimality of input mix across quantiles. Secondly, the paper proposes an approach to account for groupwise serial correlation in the estimates of pooled models of conditional quantile regressions under endogeneity. Thirdly, the paper is the first health economics study to assess the heterogeneous treatment effect of a prospective payment system on hospital production. Finally, the paper is the first

The paper is unique in using the data on Japanese hospitals from several sources: the longitudinal financial data for municipal and prefectural public hospitals accumulated by the Ministry of Internal Affairs and Communications; databases on hospitals which introduced a prospective payment system and on designated local hospitals, kept by the Ministry of Health, Labor and Welfare; and the database of the Japan Residency Matching Program on teaching hospitals. We

The technology differences are revealed in different productivity and the technical rate of substitution between hospital inputs (labor, capital and medicines). In particular, we discover lower labor returns at high-output hospitals. The findings reveal that the reform has the biggest effect at low and medium-output Japanese hospitals, while it might be less significant for the most productive hospitals. The impact of prospective payment on output is primarily attributed to

We believe that considerations of technological heterogeneity and its link with outcomes of the hospital financing reform would offer helpful guidance for policy measures. Moreover, knowledge of the determinants of self-selection could

Spillovers can arise in markets with multiple purchasers relying on shared producers. If producers are constrained in their ability to adjust quality and cost across purchasers, then the influence of a dominant purchaser affects the entire market. Prior studies have found such spillovers in health care, from managed care to non-managed care populations — reducing spending, utilization, and improving outcomes. Similar effects have been identified in the Medicare Advantage market as well, with studies finding declines in utilization and reductions in resource use among the Traditional Medicare population associated with increases in county-level Medicare Advantage penetration. However, no study to date has provided plausibly causal estimates of such spillovers in the post-Affordable Care Act era. Our study does so by exploiting idiosyncratic differences in payments to Medicare Advantage plans that are unrelated to traditional Medicare spending. Further controlling for health status and other potential confounders, we estimate that a one percentage point increase in county-level Medicare Advantage penetration results in a $146 (1.7%) reduction in standardized per enrollee Traditional Medicare spending. We find evidence for reductions in utilization both on the intensive and extensive margins (including reductions in the number of inpatient stays) and across many types of health care services including: home health, inpatient, and imaging, not all of which have been analyzed in prior Medicare Advantage spillover studies. Our results suggest that spillovers from Medicare Advantage to Traditional Medicare have

Krystal Lau [email protected]

Jun Li [email protected]

Min Chen [email protected]

Influenza pandemics can place considerable burden on affected health systems by straining hospitals’ capacities due to surges in the number and cost of inpatient admissions. Substantial variations exist in estimates of hospital admissions and their costs due to the 2009/10 influenza A/H1N1 pandemic. Previous studies assessing the impact of pandemics on hospitalizations have either analyzed the subgroup of laboratory-confirmed H1N1 patients, thus underestimating the pandemic cases, or they included ordinary seasonal influenza in their analysis, thus overestimating the pandemic cases. The objective of our study is to provide robust estimates of the overall and age-specific weekly H1N1 admissions and costs between June 2009 and March 2011 in 170 English hospitals. We use routine hospital administrative records of all patients admitted for influenza-like illnesses. Since our data does not allow us to distinguish seasonal from pandemic influenza cases, we use time series models and pre-pandemic (2004-2008) admission data to establish a counterfactual of expected weekly seasonal influenza admissions over the pandemic period. We calculate the weekly number and

We find that there were two distinct waves of pandemic admissions. The first wave coincided roughly with the official pandemic period, with 10,348 excess admissions and £20.5 million secondary care costs between June 2009 and March 2010. The second wave occurred after the pandemic had been declared over by the World Health Organization. Although it was much shorter (November 2010 – March 2011), there were more admissions compared to the first wave – 11,775 – costing £24.8 million. Patients aged 0-4 years had the highest H1N1 admission rate, and patients aged 25-44 and 65+ years had the highest costs. Our estimates are over 4 times higher than those formerly reported, suggesting that the pandemic’s burden on secondary care has been previously underestimated. Our findings support improvements in pandemic preparedness and demonstrate the value of surveillance using data on routine hospital admissions as a

Medicare’s voluntary bundled payments program has been heralded as a potential solution to align incentives and curb rising health care expenditures. Interest in acute care based bundled payments for major lower extremity joint replacement has been particularly high, with multiple cohorts of providers enrolling from October 2013 through October 2015. Early studies of the program show that bundled payments resulted in provider behavior change, leading to decreased spending for inpatient post-acute care. This paper augments previous studies by examining whether spending reductions persist over time, and whether behavior changes by bundled payment providers were specific to the

We examined 90-day, post-discharge spending for hospital-based bundled payments for Medicare and commercially insured patients treated for major lower extremity joint replacement from January 2012 through March 2016. We used claims data from the Michigan Value Collaborative, a statewide, 76-hospital consortium. We compared changes in spending by patients admitted to bundled payment providers, compared to patients admitted to non-participating control providers, using difference-in-differences analyses. Our treatment group consisted of 4,687 episodes from five early entrants (January 2014 provider enrollees) and 7,569 episodes from five late entrants (April 2015 provider enrollees). First, we examined dynamic effects of bundled payments on Medicare spending over a period of 27 months for early entrants. Second, we compared changes in Medicare spending between early and late entrants to determine whether

In difference-in-differences comparisons against episodes in the baseline period from 2012 through 2013, we found no statistically significant differences in average 90-day post-discharge Medicare spending in early entrant hospitals during the first year of bundled payments, but statistically significant but non-persistent decreases in 2015. Medicare spending in early entrant hospitals in the first quarter of 2015 declined by $887 (SE: $446; p<0.10) more than control episodes in the same periods, and declined by $1,270 (SE: $608; p<0.05) in the subsequent six-month period. We observed a positive but imprecisely estimated $298 increase (SE: $1,176) by late 2015. We found no evidence of heterogeneous effects on Medicare spending between early and late entrants (F-statistic, 0.04; p=0.85), nor consistent difference-in-differences estimates indicative of spillovers between Medicare and commercially insured patients.

Participation in bundled payments led to short-lived reductions in post-discharge spending. The effects of bundled payments were mitigated as non-bundled payments hospitals appeared to adopt alternative cost containment efforts and spending reached equally low levels, making it more difficult for bundled payments providers to achieve differential savings in later years. Finally, our findings suggest that behavior changes from bundled payments providers were isolated to Medicare beneficiaries and did not lead to broad-based practice change. Ultimately, the future success of bundled payments will hinge on its ability to incentivize continuous improvement and its role in transforming practice culture

The 2009 Health Information Technology for Economic and Clinical Health (HITECH) Act devoted $28 billion through its “meaningful use” incentive program to providers that adopted Electronic Health Records (EHRs) and participated in Health Information Exchange (HIE). Underlying these substantial investments is a belief that timely transfer of standardized electronic health information such as laboratory results and clinical summaries across the care continuum can facilitate coordinated patient care and improve health outcomes. Despite the recent growth of HIE and its potential benefits, only a few studies have examined the impact on quality of care in inpatient settings, and, to the best of our

We conducted a large scale retrospective study to examine the impact of HIE engagement on individual patients’ outcomes of Acute Myocardial Infarction (AMI) that is directly targeted by the Hospital Readmissions Reduction Program and a variety of other Centers for Medicare and Medicaid Services (CMS) programs, including the Hospital Value-Based Purchasing Program, and the Bundled Payments for Care Improvement (BPCI) Initiative. We linked the Florida State Inpatient Discharge (SID) data, which allows us to track a patient’s longitudinal visits across hospitals, with the American Hospital Association Annual and Information Technology Supplement surveys. Using a difference-in-differences (DID) estimation approach, we compared changes in outcomes of a treatment group of targeted admissions before and after HIE engagement, relative to changes in outcomes of a control group that never participated in HIE. Our main outcome measures are the 30-day, 45-day, and 60-day all-cause readmission rates. To investigate whether the changes in readmission came at the cost of other quality measures, we also analyzed the impacts on length of stay, total charges, total number of procedures, discharges to a nursing facility or home health care, and in-hospital mortality. Our models adjust for patient characteristics and include hospital specific fixed effects to control for unobservable confounding factors. We employed placebo tests to rule out the concern that changes in outcome measures may have already started in time periods prior to the participation of HIE. Overall, we found that HIE engagement did lower 30-day all-cause readmission rates for AMI patients. The decrease in readmissions after HIE engagement primarily came from reduced readmission to a different hospital. In addition, associated with the reduction in readmission were the rises in length of stay, number of procedures, and total charges, but there were no statistically significant changes in transfer, discharge destination or in-hospital mortality. These results suggest that the decrease in readmission was achieved through the increased treatment intensity of inpatient care, but was not due to any strategic transfers or changes in discharge destination. HIE may have played an important role in determining the optimal cost tradeoff between inpatient care and readmission. A back-of-the-envelope calculation reveals that, for AMI condition alone, the HIE participation in Florida hospitals reduced 235 avoidable

Austin Williams [email protected]

Tursynbek Nurmagambetov [email protected]

Didem Bernard [email protected]


: Asthma is a chronic disease that affects quality of life, productivity at work and school, and healthcare utilization; although controllable, it can even result in death. To control asthma symptoms and prevent severe asthma attacks, the evidence-based guidelines developed by medical professionals should be followed. These recommendations include assessment of asthma severity, prescribing and ensuring adherence to asthma control medications, providing asthma self-management education, and identifying and avoiding ambient and indoor environmental triggers. In this paper, we estimated the effect of health insurance status, race/ethnicity, and income on asthma-related incremental

as the presence of at least one medical or pharmaceutical encounter or claim associated with asthma. For the main analysis, we applied two-part regression models to estimate asthma-related annual per-person incremental medical expenditure (APIME). Routine outpatient care was defined as scheduled

During 2008-2013 the national average of APIME in the United States was $3,266 (in 2015 US dollars); more than 80% of that amount was attributable to prescription medication and routine outpatient care, while roughly 20% was attributable to hospitalizations and emergency room (ER) visits. APIME was significantly lower than the national average for uninsured persons, Blacks, Hispanics, and persons whose income was equal to or above the national poverty level. Conversely, APIME was higher than the national average for insured persons, Whites, Asians, and for those whose income was below the national poverty level. Studies show that use of ERs and hospitalization services, previously a major driver of high total medical expenditures for asthma, occurs more often among uninsured persons, Blacks, and Hispanics. Our study shows that prescription medications and routine outpatient care are comprising an increasingly large proportion of total asthma care expenditure. Uninsured persons may tend to seek care through use of the ER or hospitalization services rather than through routine outpatient care and filling prescription medications, which lowers overall

Our results also show that persons with income below the national poverty level have higher APIME than persons in higher income brackets. These persons are more likely to live in areas with a higher concentration of outdoor and indoor environmental asthma triggers, which are a major cause of asthma attacks. On the other hand, these individuals are more likely to qualify for Medicaid, which may facilitate access to a wider range of routine and urgent medical services,

Lack of health insurance hinders access to prescription medications and routine outpatient care and, as a result, contributes to higher use of ER visits and hospitalizations for persons with asthma. Persons with income lower than the national poverty level have higher APIME and, despite being potentially eligible for Medicaid, may need additional financial support, such as health insurance reimbursement for environmental interventions, to maintain an indoor

In this study, we compare and align health care expenditure estimates from the Medical Expenditure Panel Survey and the National Health Expenditure Accounts. Reconciling MEPS and NHEA estimates serves two important purposes. First, it is an important quality assurance exercise for improving and ensuring the integrity of estimates from both sources. Second, the reconciliation provides a consistent baseline of health expenditure data for policy simulations. MEPS is often used in developing microsimulation models because it contains person-level expenditures. Reliable estimates of national health spending need to be used as a baseline for analyzing the impact of potential policy changes on health care costs. Based on results from our study, analysts can adjust MEPS to be consistent with the NHEA so that the projected costs as well as budgetary and tax implications of any policy change are consistent with national health spending estimates. Previous reconciliations have been used as the baseline by the Department of Health and Human Services, the Congressional Budget Office, RAND and other researchers in simulating the impact of potential policy changes. The NHEA and MEPS both provide comprehensive estimates of health care spending in the U.S. The NHEA is primarily based on aggregate provider revenue data and administrative records of publically administered programs and covers the entire U.S. population and a full range of health care expenditures, including personal health care spending, public health services, research, and investment in structures and equipment. NHEA estimates are produced annually in the U.S. by the Office of the Actuary at the Centers of Medicare and Medicaid Services (CMS). MEPS, on the other hand, provides person-level information on health expenditures from a nationally representative sample of households in the civilian, non-institutionalized population. MEPS is produced by the Agency for Health Care Research and Quality (AHRQ) and the National Center for Health Statistics. The reconciliation is conducted every five years when the quinquennial Economic Census is available as it is the only data source that contains expenditures at the required level of detail so that specific expenditures reported in different service categories in NHEA and MEPS can be aligned. . The previous

Although each source provides a measure of total national spending on personal health care (PHC), at first glance the estimates appear to diverge significantly. We make adjustments to account for the differences in underlying populations, covered services and other measurement concepts to reconcile the expenditure estimates. Once we adjust the NHEA to make it consistent with MEPS, we compare and discuss potential reasons for the differences for each service category and source of payment. We also discuss how the expenditure estimates have changed since the previous reconciliation in 2007. Identifying service types and sources of payment with larger gaps helps AHRQ and CMS focus future research

Mujaheed Shaikh [email protected]

Melanie Luhrmann [email protected]

Michael Karpman [email protected]

How organizational changes relate to firm performance has been the focus of research across disciplines. However, despite its importance, empirical evidence is scant. In healthcare most studies focus on the impact of ownership on performance, but evidence is mixed. We argue that this ambiguity is due to two reasons: first, these studies compare performance across firms that operate in heterogeneous rather than homogeneous markets. Second, there is little assessment of how organizational change per-se relates to performance. We, on the other hand, assess the performance effects of organizational change in the context of a homogenous healthcare market by analyzing its impact on hospital costs and the extent to which efficiency gains from such change interact with scale and scope of hospital services. We also explore how such gains vary with planned vs. unplanned hospital activities and hospital heterogeneity, namely: hospital functional diversity and relative performance. Exploiting detailed 2001-2008 panel-data for English hospitals and the introduction of the Foundation Trust policy that triggered major organizational change - we find that hospitals exhibit economies of scale, but not scope; hospitals that underwent organizational change are more efficient than those that did not; and the organizational change facilitates economies of scope but not scale. However, efficiency gains vary importantly with hospital heterogeneity. Our results suggest that, the FT policy enabled cost-efficiencies, especially for worst-performing and less functionally diverse hospitals. This highlights that organizational changes can be

In this paper we investigate to what extent the childhood healthcare environment influences later life health outcomes. We examine a fundamental re-organisation of the healthcare environment in the U.K., which occurred through the introduction of the National Health Service (NHS) in July 1948. Immediate large decreases in infant mortality of 17% ensued, which were focused on the neo-natal period and larger for individuals who prior to the NHS had a lower access to

Data: We combine historic county-level data with the Office of National Statistics Longitudinal Study of linked census records combined with administrative mortality data, and a large new dataset - the UK Biobank - recording health

Method: As the NHS was introduced nationwide on a single date, we employ a Regression Discontinuity Design, where we will allow for preexisting trends in the outcomes to be different either side of the threshold (i.e. the timing of the

increasing monotonically with age and becoming statistically signicant from age 57 onwards. The increase in the benecial impact of the NHS on survival rates in this population group represents a 12% reduction in mortality (and a 1% increase in survival) at age 57. We supplement these findings with analysis of hospital records, which reveal a similar decrease in hospitalisations for cardiovascular disease, one of the major causes of death, for lower class individuals. Our results suggest that the expansion to universal healthcare (and individual exposure to this universal system at birth) leads to a narrowing in the mortality gap between social classes at older ages.

Before the Affordable Care Act (ACA), people with chronic conditions were typically denied coverage or faced high, experience-rated premiums or preexisting condition exclusions in the nongroup market. Expanding access to nongroup coverage for these individuals while keeping premiums affordable was thus a key objective of the ACA. Recent policies threaten to undermine ACA provisions designed to include healthier and sicker individuals in a single risk pool, yet relatively little is known about the medical needs of people with nongroup coverage who would be affected by these policies. For this study, we examined the health status and health care experiences of adults covered

The study draws on 2012-2015 Medical Expenditure Panel Survey (MEPS) data and focuses on adults ages 18 to 64. We analyzed changes over time in nongroup coverage for this age group. We then estimated changes between pre- and post-ACA implementation periods in the treated prevalence of chronic conditions among adults with nongroup coverage, based on diagnosed conditions that were linked to health care provider visits and prescription drug fills. We also compared treatment for chronic conditions by coverage type (Marketplace, other nongroup, employer-sponsored, and public); other measures of interest included disability status, service use, spending, and sources of payment for care. Because open enrollment periods vary by coverage type, the analysis focused on service use and treatment occurring in the last six months of the year among those with continuous coverage during that period.

The share of nonelderly adults reporting nongroup coverage more than doubled following ACA implementation, with all enrollment growth occurring through the Marketplaces. Between the pre- and post-ACA implementation periods, there were increases in the shares of nongroup enrollees who were treated for multiple chronic conditions and who were in the top decile of spending for this age group. These changes were driven primarily by the poorer health of adults with Marketplace coverage, many of whom were uninsured prior to ACA implementation. In 2014-2015, nearly 45 percent of Marketplace enrollees were treated for a chronic condition during the reference period, compared with 35 percent of those with non-Marketplace nongroup coverage and 38 percent of those with employer coverage. Relative to other privately insured adults, those with Marketplace coverage were more likely to have

The Marketplaces expanded coverage for adults with chronic conditions, but their higher service use has contributed to rising nongroup premiums. Policymakers seeking to address this challenge face a choice between proposals that aim to strengthen the ACA’s risk pooling arrangements and proposals to concentrate the risk of high costs among those with the greatest medical needs. The outcome of these policy decisions will have a significant impact

Phaedra Corso [email protected]

Daria Pelech [email protected]

: Mercy Health Center (MHC) is a faith-based health resource center serving the underserved population across six Georgia counties. The staff and volunteers provide free healthcare to the uninsured including primary care, pharmacy, dental, chronic disease management, and behavioral health counseling. The purpose of this analysis is to explore the impact that MHC has on the healthcare utilization of its patients over time and the associated healthcare

: A cohort of 185 adult patients was identified, and 27 months of hospital utilization was recorded (9 months pre-MHC and 18 months post-MHC). MHC utilization during this period was also recorded. After attaching unit costs to MHC visits, emergency department usage, and outpatient services, we analyzed healthcare utilization and costs over time. Simple analyses included non-parametric longitudinal comparisons of utilization by category and of total healthcare costs. Some simple assumptions were made regarding the trajectory of healthcare utilization beyond the 18-month follow-up period. Further, a recurrent event survival analysis of each category of healthcare utilization was conducted

: Emergency department utilization decreased and outpatient services increased after patients gained access to primary care through MHC. The primary healthcare category of healthcare savings was found to be from a reduction in emergency department visits. However, estimated healthcare cost savings were not large enough during follow-up to offset MHC costs. Overall, results were confirmed in the survival analysis as the risk for emergency department

: Though net cost savings were not realized within the first 18 months at Mercy for this cohort, we would expect cumulative net cost savings to begin to accrue after a patient’s second year at MHC. Our results suggests that

: Patient data were gathered from two local hospital databases, as well as from MHC, leading to three specific limitations. First, verification of patient residence in the service area during the full 27-month period was not possible. Second, we did not have data on healthcare utilization outside of the service area or from local private clinics. Finally, our analysis did not consider medication costs, as these data were not available to the research team.

One third of Medicare beneficiaries are now enrolled in private plans through Medicare Advantage. After years of growth in federal payments to Medicare Advantage plans, the Affordable Care Act (ACA) slowed or cut such payments. To date, little is known about the impact of these ACA-related payment changes on plan behavior and on benefits provided to beneficiaries. We examined how plans responded to ACA payment reductions relative to their response to pre-ACA

We used 2006-2015 data from the Centers for Medicare and Medicaid Services (CMS) to examine the impact of changes in the maximum federal payments to plans (the “benchmark”) on plans’ asking prices (their “bids”) and on benefits received by beneficiaries (the “rebate”) before and after the ACA. This rebate, which equals a portion of the difference between the bid and the benchmark for plans that bid below the benchmark, must be passed on to beneficiaries in the form of lower premiums or additional benefits including reductions in out-of-pocket costs, reductions in drug costs, and increased coverage for vision, dental, and hearing services. We also assessed differences in plan behavior among plans facing larger benchmarks as compared with smaller benchmarks. Analyses used longitudinal models that exploit the variation in benchmark changes before and after the ACA benchmark cuts, adjusted for beneficiary risk, market

In real terms, average monthly Medicare Advantage benchmarks grew by $35 before the ACA (2006-2009) and decreased by $81 after the ACA-related benchmark cuts (2012-2015). Before the ACA, for every $1 increase in the benchmark, <0.001) on average. After the ACA, plans lowered their bids by $0.57 on average for every $1 decrease in the benchmark (p=0.03). This symmetrical bid

response after the ACA lessened the resulting decline in beneficiary rebates. Moreover, declines in final plan payments and beneficiary rebates were further offset by new bonuses from quality incentives and increases in beneficiary risk scores. Within rebates, after the ACA plans reduced benefits by about twice as much on the margin as they had raised benefits before the ACA for each dollar change in the benchmark. However, plans changed premiums by similar amounts in response to benchmark changes pre- and post-ACA. Plans in more competitive markets were less responsive to benchmark changes than plans in less competitive markets, implying that plans in more competitive markets may

In contrast to before the ACA, Medicare Advantage benchmarks decreased after the ACA. Plans responded to these cuts by lowering their bids, suggesting that plans were operating above cost. This plan bid response, combined with additional payments due to quality bonuses and growth in risk scores, helped lessen the decrease in beneficiary rebates, which may explain the continued growth in Medicare Advantage enrollment after the ACA.

Matthew Lau [email protected]

Matthew Lau [email protected]

Adrian Garcia Mosqueira [email protected]

The CHAS policy, introduced in 2012 in Singapore, aims to improve accessibility and affordability of healthcare by offering subsidies to low and middle-income groups and elderly individuals for general practice consultations and healthcare. The investigation was undertaken by acquiring and analysing primary and secondary research data from 3 main sources, including handwritten survey responses of 334 individuals who were valid CHAS subsidy recipients (CHAS cardholders) from 5 different locations in Singapore, interview responses from two established general practitioner doctors with working knowledge of the scheme, and information from literature available online. Survey responses were analysed to determine how CHAS has affected the affordability and consumption of healthcare, and other benefits or drawbacks for CHAS users. The interview responses were used to explain the benefits of healthcare consumption and provide different perspectives on the impacts of CHAS on the various parties involved. Online sources provided useful information on changes in healthcare consumerism and Singapore’s government policies. The study revealed that CHAS has been largely effective in reducing market failure as the subsidies granted to consumers have improved the consumption of healthcare. This has allowed for the external benefits of healthcare consumption to be realized, thus reducing market failure. However the study also revealed that CHAS cannot be fully effective in reducing market failure as the scope of CHAS prevents healthcare consumption from fully reaching the socially optimal

, albeit with some benefits to third parties yet to be realised. There are certain elements of the

CHAS is a government subsidy, introduced in 2012 in Singapore, to improve accessibility and affordability of healthcare to low and middle income groups and elderly individuals for general practice consultations and healthcare. As the

The Lorenz curve and corresponding Gini coefficient values were used to study the effect on income inequality. Curves were plotted to represent income distribution in 3 separate groups: (a) The nation as a whole for incomes up to $19,999 per month per household, using publically available government data. (b) The same data to which is applied theoretical maximum claims by CHAS-eligible households. Information on maximum claims was obtained from the CHAS website. (c) Forty-eight subjects who provided real world data on their income and CHAS subsidy claims. Gini coefficients were calculated for the corresponding Lorenz curves. The paired t-test was used to determine whether differences

Findings from the Lorenz curve and Gini coefficient data showed that in a theoretical situation of maximum CHAS subsidy claims, there would be a clear decrease in income inequality (change in Gini coefficient = 0.036, p<0.0001). For the real-world data, CHAS claims created a small but statistically significant reduction in income inequality (change in Gini coefficient = 0.005, p<0.0001). There was a small but visible shift of the Lorenz curve.

. This effect could be potentiated by greater use of the CHAS scheme by eligible patients. A number of caveats were identified in making the conclusions. First, theoretical maximum claims far exceed actual usage in the population surveyed, and may not represent the real world situation. Second, the study does not account for monthly households with income above $19,999, which make up 12% of the national population. The data therefore does not account for the entire population. Third, the real world data sample size is small and a much larger survey

Eligible individuals in states that expanded Medicaid have reported gains in health care access measures, including having a personal physician, affordability of care, and insurance coverage. However, this change in insurance eligibility led to increases in demand for health care, which raised the question of whether the Medicaid expansion would exacerbate clinician shortages. This paper examines several aspects of the health care workforce across states with different expansion status. First, we assess whether expansion and non-expansion states have different baseline levels of workforce capacity. Then, using a difference-in-differences approach, we test whether the Medicaid expansion had any effects on the number of physicians, nurses, physician assistants, and other health care professionals working in these states. Lastly, we utilize a more granular approach by studying whether the expansion had differential effects within states on


Post 2014 expansion, we find no evidence that the Medicaid expansion had any significant impact on the available health care workforce post-reform across multiple categories of clinicians. County-based analyses are in process but will be

Our results have important economic and policy implications. While Medicaid expansion thus far had led to improvements in access to care across multiple studies, the remaining non-expansion states may have different experiences should they expand given significantly smaller health care workforces at baseline. Furthermore, at present we do not find a Medicaid expansion effect on the health care workforce, suggesting there has been little effect of expansion on short-term entry or exit into health professions, or selective migration by clinicians into (or away) from Medicaid expansion states. However, these effects could manifest in the future, given the lengthy educational requirements within

Savannah Bergquist [email protected]

Sapna Kaul [email protected]

In the conventional framework for designing health plan payment models, the regulator chooses variables to be used as risk adjustors, the risk adjustment weights, and other policy parameters, but the data from which estimates are derived are taken as given. This approach implicitly assumes the observed spending patterns are optimal. In this paper we explore an entirely novel approach: using the data itself as a policy tool. We take the risk adjustors, the estimation

We develop a general model for the provision of health care services by health plans. The key insight of our model is that there is a two-way relationship between plan actions and health plan payment: plan actions (outcomes) are a function of health plan payment, and plan payments are a function of the insurer actions the payment system is meant to affect. Importantly, we show when plan payments are calibrated on data generated by plan actions, equilibrium is

Using Medicare data we apply these ideas to two areas of misallocation in health care: undercompensation for individuals with mental health diagnoses and disparities in health care spending between high and low income groups. We transfer spending to the group of interest, re-fit the risk adjustment model on the modified outcome, and illustrate the relationship between the transfer amount and targeted measure. We show spending can be transferred between disease groups to eliminate undercompensation with a minimal impact to overall fit of the risk adjustment model, while correcting disparities requires shifting much larger amounts of spending.

While the Affordable Care Act (ACA) in the U.S. has led to significant gains in health insurance coverage and access to care, less is known on how this policy change affects non-geriatric cancer patients’ utilization and outcomes of emergency departments (EDs). Previous studies demonstrate that younger patients and those who lack access to usual source of care, in general, are at an increased risk of using EDs. Similarly, patients with Medicaid or no insurance use EDs more often than those with private insurance. Within cancer patients, adolescents and young adults (AYAs), aged 15-39 years, are more likely to experience insurance- and cost-related barriers to care. Yet, it is unknown how the ACA

We examined changes in ED use and outcomes for AYA cancer patients compared with those of children and non-geriatric adults with cancer in the U.S. before and after the implementation of the ACA.

We used the 2013 and 2014 National Emergency Department Sample (NEDS) from the Healthcare Cost and Utilization Project. Our subpopulation consisted of cancer patients (any cancer diagnosis) currently aged 64 years and younger. We compared patients’ demographic (e.g., sex, primary payer, county of residence) and clinical (e.g., number of chronic conditions and procedures, cancer diagnosis) characteristics, and hospital characteristics (trauma designation, teaching status) between years. We also examined the ED outcome (treated and released vs. admitted to the same hospital). Variables were examined for the overall sample and also stratified by age categories (i.e., children 0-

Overall, cancer patients accounted for over 1.8 million ED visits in 2013 and 1.9 million visits in 2014. A significant decrease was observed in self-paid visits from 2013 to 2014 (9% to 6%). Self-paid visits decreased from 16% to 12% (p<0.001) for AYA visits and from 8% to 5% (p<0.001) for adult cancer visits, however, no differences were observed in primary payer of ED visits by children (2% in both years). Within each group, ED visit outcome did not differ between years. Yet, in each year, AYAs were more likely to be released than admitted (69% vs. 31%) compared with children (58% vs. 42%) and older adults with cancer (51% vs. 49%) (p<0.001). In our adjusted models that accounted for other

While self-pay AYA visits decreased in 2014 following implementation of the ACA, self-pay visits remained significantly higher for AYAs (12%) as compared to children/adults. Our analyses demonstrate that the ED outcome differs depending on health insurance status. This is especially true for AYA cancer patients who were more likely than children/adults to be uninsured and, thereby less likely to be admitted to the same hospital following an ED visit.

Shirlee Lichtman-Sadot [email protected]

Xi Chen [email protected]

Transportation can be a significant barrier to attaining healthcare services, in particular for disadvantaged populations. We exploit a reform that introduced public transportation services to Arab towns in Israel to evaluate the effect of

In 2007, the Israeli ministry of transportation (MOT) announced a reform within non-Jewish communities in Israel: the introduction of public transportation (PT) to these communities. Until then, Non-Jewish communities had been significantly deprived of PT infrastructure, with generally no official services. Furthermore, private car ownership rates are relatively low among Arabs, and many women do not have a driving license due to traditional barriers. The new bus network, which gradually developed over the next 7 years, represented a substantial increase in access to healthcare services. Health services in the form of the best doctors or specialty clinics and hospitals are found in Israel outside Arab

We use very detailed data from the Israeli MOT documenting the frequency of all bus lines in Israel, their routes and bus stops, for 2008-2014 on a bi-annual basis. We matched our measures of bus frequencies for each town and period to a survey of the Arab population in Israel conducted in 2004, 2007, 2010 and 2014. Our analysis focuses on the elderly population - ages 50-70 - based on the health conditions inquired about in the survey - high blood pressure, diabetes,

Our initial results show statistically significant increases in the diagnosis of heart disease, high cholesterol, asthma, back problems, and migraine headaches among the population aged 50-70 when the penetration of buses to these individuals’ communities increases. We do not observe statistically significant changes in response to public transportation penetration for diabetes and high blood pressure. We observe some differential effects based on respondents’ sex.

A naive interpretation of these results can suggest that public transportation penetration is adversely affecting health outcomes among the elderly population. We believe that given that the survey inquires about diagnosis of these health conditions, a more plausible interpretation of the results is that there are greater diagnosis levels and awareness of health conditions that were existent prior to PT penetration but were unobserved due to lower access to healthcare

In China and other developing countries, long waiting times during a hospital visit are pervasive due to the rapidly growing demand for health care services in already overcrowded hospitals. However, technological innovations have been introduced to reduce patients’ waiting times in hospitals and to streamline the process of healthcare services. Such innovations have the potential to improve the efficiency and quality of health care as well as increase patient satisfaction.

Resident Card allows patients to schedule an also enables automatic e-payment at physicians’ office and through self-service machines; therefore patients no longer need to

on patients. We have records of 4 million outpatient transactions during 2011-2013. This hospital since 2012. We use the standard difference-in-differences approach to compare changes in patient health-seeking behavior and health outcomes between the treatment groups (i.e. those using the Resident

Linking each patient by their identifier, we find that compared to the control group, those using the card experience a greater reduction in waiting time for hospital visits and an increase in service use. These two impacts seem to result in a decline in the severity of the health problems being treated due to the greater accessibility. Also, the almost fixed supply of physicians in the short term and the increase in demand for health services give doctors more power relative to patients and payers. However, with government established prices, they make profit by prescribing unnecessary medications and more tests. Moreover, the new transaction technology makes patients’ type of insurance more visible to

produces both benefits and costs, and also systematic heterogeneity across populations.

Robert Lieberthal [email protected]

Zachary Levinson [email protected]

Healthcare fraud can represent upwards of hundreds of billions of dollars in spending that could be better spent on patient care. There is often not sufficient detail on the underlying methodologies and data samples that lead to fraud estimates, which may be due to different purposes of these reports or the need to obscure the details of fraud detection methods to prevent fraudulent operators from responding to existing methods.

The objective of this study was to provide a systematic evaluation and synthesis of the methodologies and data samples used in current peer-reviewed studies on characterizing healthcare fraud. The academic databases searched were Academic Search Complete, Business Source Complete, EconLit, Medline (EBSCO), OneSearch, ProQuest Business Collection, ScienceDirect, and Web of Science. Governmental and

This examination was conducted using a systematic review methodology to identify relevant studies and determine their relevance. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement was used to guide the performance of reviewing the literature. Study criteria for eligibility were collected by applying specific search terms: healthcare, health insurance, Medicare, Medicaid, Obamacare, Affordable Care Act, or health services; fraud, cheat, falsification, corruption, or kickback; detect, detection, prevent, prevention, deterrence, audit, or auditing. Results were restricted to scholarly journals, academic journals, working papers, and conference proceedings. Study selection occurred through two independent reviews of each study for inclusion or exclusion. Disagreements between reviewers were resolved through discussion by the entire research team.

Our search terms resulted in 450 articles that were potentially appropriate for inclusion in our report. The results of independent reviews ended with twenty-seven studies considered as relevant to include after the application of

One limitation of this study is that the strength of the evidence is reliant on the quality and number of studies previously performed on the topic. Another limitation is the quality of studies with regard to their applicability to

A limited number of validated methods are used to detect healthcare fraud. The literature on this topic is spread among several academic fields. The majority of available studies utilize public or social health insurance systems

Our insurer agnostic approach examines the availability and effectiveness of healthcare fraud analytic methods across different types of health insurers, posing great value for members of the health sectors.

Over the past several years, there has been a substantial shift of Medicare enrollment from the traditional, government-administered Medicare program (“traditional Medicare”) to the private, subsidized, and mostly managed care plans offered in Medicare Advantage (MA). While previous work has examined the effect of this trend on the efficiency of care and on beneficiary health outcomes overall, my study is one of the first to consider the implications of increasing MA enrollment for a particularly disadvantaged group of beneficiaries known as dual eligibles (i.e., Medicare beneficiaries who also receive full or partial Medicaid benefits). Dual eligibles merit special attention because they are economically vulnerable, often have significant health needs, and face unique barriers in navigating the health system (e.g., as about three-fifths report having cognitive impairments). Further, while the share of Medicare beneficiaries enrolled in MA

This study evaluates the effect of MA penetration on the number and length of hospital stays (including potentially-preventable admissions) and all-cause mortality rates among dual eligibles. I rely on complete Medicare enrollment and MedPAR files and public MA data from 2009 through 2015. These sources provide detailed information about dual enrollment and capture all hospital discharges at the vast majority of acute care PPS hospitals and all beneficiary deaths.

My primary identification strategy relies on a regression discontinuity design used by a prior study to evaluate outcomes among the overall Medicare population (Afendulis, Chernew, and Kessler 2017). This approach exploits a discontinuous jump in average benchmark payment rates for MA plans – which is subsequently associated with a sharp increase in MA enrollment – in metropolitan statistical areas that exceed a population threshold. Preliminary results suggest that this increase in plan payments is also associated with a jump in MA penetration rates among dual eligibles who receive partial benefits. I use this potentially exogenous source of variation in enrollment to explore the relationship between MA and beneficiary outcomes. Because dual eligibles may also be enrolled in comprehensive or limited Medicaid managed care plans, I also run analyses restricted to the subset of counties where such enrollment is

My findings will help policymakers understand how the increased role of MA plans has affected a vulnerable subset of the Medicare population and will be informative as states continue to delegate Medicaid benefits for dual eligibles to

Giovann Alarcon Espinoza [email protected]

Bo Shi [email protected]

Losing Medicaid coverage can have a negative effect on people’s health; it affects continuity of care, especially when they become uninsured. Although not all forms of churn are detrimental (e.g., individuals who find a new job can obtain

Our model focuses on people’s decision to obtain and maintain Medicaid coverage. We understand churn as a change in this status. We also used first differencing to create our set of (mutually exclusive) predictors: becoming employed or unemployed, gaining or losing a job, having a wage increase or decrease, and having a newborn. We included lagged variables of every life event in our model to allow for a 2-month delayed effect. Working with longitudinal data allowed us to identify these life events, and also use first differencing variables and eliminate any bias due to time-invariant unobserved characteristics. Principal Findings We estimate that 5.1% of nonelderly adults who were enrolled in Medicaid in January, 2013 lost this coverage at some point in the following year. Most of these (70%) had at least one month of uninsurance, while the rest shifted to other

We found that becoming employed and gaining a job increased people’s likelihood of churning off Medicaid, though these changes did not affect churn immediately. If someone’s family member became employed in one month, their

Another determinant of churn was having a child, which makes individuals more likely to churn by 1.1 pp the same month, 1.2 pp the following month, and the month after. This result is a novel finding, as prior research has not focused on

Our results are consistent with the theory that becoming employed or gaining a job would increase family income, which could make some enrollees ineligible. Although this would generally be viewed as a positive development, it could result in the individual becoming uninsured or underinsured, or experiencing discontinuities in care. These effects were only significant after a lag; the effect on churn does not happen within its month of occurrence. This has implications

Since women are still eligible for enrollment during the months immediately postpartum, our results could indicate that some women believe their coverage ends at birth. This is important, because if women believe they lost Medicaid

To accommodate Medicare and Medicaid budget cuts and reimbursement method innovations, hospitals have made great effort on cost containment and quality improvement. The first goal of the paper is to study the empirical relationship between healthcare service quality and hospital cost containment. Our current finding suggests that cost containment is negatively related to quality improvement. But the relationship diminished after 2013, which is the year when Center for Medicare and Medicaid Services (CMS) implemented Value-Based Purchasing (VBP) program and incorporated quality as important dimension in acute care hospital inpatient services reimbursement. Therefore, our more

We use two dataset primarily: Hospital Compare and Medicare Cost Report. Hospital Compare provides hospital quality measures on clinical care, patient experience, safety, and efficiency. A Total Performance Score (TPS) is calculated as a weighted average of these quality aspects. And a linear exchange function translates TPSs into VBP adjustment factor, which determines a hospital’s value-based incentive payments. Top performers are rewarded with bonus and bottom ones get a discount of full reimbursement. CMS started to reimburse acute care hospital’s inpatient services using this method since 2013. Medicare Cost Report data records hospital operation, financial management, and claims

First, to observe empirical relationship between cost containment and quality change, we use VBP-eligible acute care hospital TPS change as the dependent variable and cost containment proxies as the main predictor controlling for covariates commonly used in previous healthcare service quality research. And we examine the sample in two time periods separately: financial crisis years before VBP 2007 – 2012 and VBP years 2013 – 2016. Hospitals were motivated to reduce costs in both periods but for different reasons. And we use the TPS calculation method in 2013 consistently for all years. We found that the negative relationship between cost containment and quality improvement is stronger

Second, we address whether the VBP program incentivized hospitals to improve quality while reducing costs. And we apply a ‘quasi’ Difference-in-Difference (DD) study. The VBP program is applied to all eligible acute care hospitals simultaneously since 2013, which excluded a clean control group. Therefore, the DD method is compromised by taking hospitals with less than 10% of Medicare revenue as a ‘quasi’ control group and hospitals with Medicare revenue more than 50% as the treatment group. The DD analysis shows marginally significance of the VBP factor. Besides, we plan to further examine whether the VBP calculation method change and the progressive payment adjustment affect results. Finally, we offer a possible explanation. In particular, we examine the relationship between quality change and cost containment for hospitals in different financial standings. For some hospitals, benefit of cost containment may balance or

Sally Stearns [email protected]

Paul Jacobs [email protected]

Health economics defines the extensive margin for surgical procedures as the rate at which people with similar diagnoses and medical conditions receive various treatments. While much attention has focused on geographic variation in the extensive margin for various procedures, the expanding use of certain procedures in an increasingly aged and frail population may constitute a growing problem given limited health care resources. An important illustration of this problem pertains to percutaneous coronary intervention (PCI) and coronary artery bypass surgery (CABG) procedures. While these treatments can be life-extending in many circumstances, they are often associated with significant morbidity and delayed recovery among elderly or frail patients. In particular, prolonged rehabilitation can be required due to complications or from frailty and poor functional status at the time of presentation. Randomized trials of PCI and CABG typically exclude frail and elderly patients, particularly those with renal dysfunction, and do not report the rate of nursing home discharges and delayed physical recovery. This analysis describes current trends in coronary revascularization and subsequent care. Medicare claims data from 2006-2015 for a 20% sample of Medicare beneficiaries age 65 and older are used to described trends over time in procedure rates for CABG and PCI for all beneficiaries as well as beneficiaries with selected coronary diagnoses (e.g., acute myocardial infarction, acute coronary syndromes, unstable angina) to adjust for changes in underlying disease rates over time. The descriptive trends for inpatient procedures are calculated overall and for sub-groups (e.g., age groups, case mix severity, and receipt of hemodialysis) for fee-for-service as well as Medicare Advantage enrollees. Analyses of trends for total procedures (inpatient and outpatient) and outcomes are conducted only for fee-for-service enrollees given lack of claims for physician and post-acute services for Medicare Advantage enrollees. We use regression analysis of claims for fee-for-service enrollees receiving PCI or CABG to assess key outcomes: repeat revascularization, post-acute care including skilled and non-skilled nursing home days, hospital readmission, hospice use, post-discharge mortality,

Preliminary results show that from 2007 to 2015, rates of CABG declined overall; CABG rates increased slightly among persons aged 85 and older or persons but declined slightly among persons with more complex disease (Charlson>3). In contrast, rates of PCI were fairly stable over time but increased among persons aged 85 and older as well as among persons with more complex disease (Charlson>3). Among beneficiaries receiving coronary revascularization, overall rates of discharge to skilled nursing care increased over time while discharges to home without home health care decreased. The outcome analyses that are in process will assess one-year outcomes and resource use trajectories over a longer

Descriptions of current trends in procedure rates and outcomes including resource use enable discussion of the implications of expansion of the extensive margin for coronary revascularization. The estimates will provide valuable

The United States has a complex system for financing healthcare, combining an array of public and private components. In addition to out-of-pocket amounts paid directly to providers, healthcare financing includes: the payment of premiums directly to insurers, employer contributions for workers’ health plans, and public programs that provide health coverage and draw on state and federal tax revenues. Given the growth of health care spending and the fact that individuals ultimately bear these costs in some form, there is substantial value in developing a thorough understanding of how much individuals and families pay for healthcare, how these payments are distributed, and how the incidence

Previous attempts to understand how much individuals pay for health care typically focus on particular types of healthcare payments in isolation, such as out-of-pocket medical costs (Banthin et al, 2008) or premiums for private insurance (Gruber and McKnight, 2003). Little comprehensive analysis of equity in the finance of U.S. healthcare has been conducted, with two notable exceptions being Wagstaff et al. (1999), which examined data from 1987, and Ketsche et al.

We develop an updated analysis of healthcare finance equity using the nationally-representative Medical Expenditure Panel Survey (MEPS) combined with a variety of supplementary datasets to benchmark and to enhance our estimates. We align the MEPS distribution of income to Internal Revenue Service data, simulate a full array of federal and state income tax expenditures and state sales taxes, and report sources of financing for healthcare by quintiles of equivalent

Our systematic analysis of all major components of health spending reveals that the financing of healthcare in the United States – notwithstanding a relatively progressive structure for income taxes – is regressive. The bottom quintile of households in the United States paid 11.7 percent of their income in health payments in 2013 compared with 8.1 percent for the top 1 percent of households. However, our estimates also show that the U.S. system for financing healthcare has become more progressive over time with the share of income devoted to health spending for the bottom quintile falling 6.1 percentage points from 17.7% in 2005 to 11.7% in 2013 and the share paid by the top 1 percent of

Our preliminary study period encompasses the introduction of several significant changes to healthcare and tax policy, which are consistent with our findings of increased progressivity in the healthcare sector, including the beginning of the Part D prescription drug program in Medicare (2006), the introduction of income-related premiums in Medicare (2007), and the Additional Medicare Tax for higher-income households (2013). Although we do not yet have complete

Zachary Levinson [email protected]

Tianyan Hu [email protected]

To help Medicare beneficiaries choose among Medicare Advantage (MA) plan options, the government collects data on several dimensions of care and summarizes this information through a star rating of plan quality, measured on a scale of one to five. Prior research suggests that beneficiaries do indeed rely on star ratings when making enrollment decisions and the government has used star ratings as the basis for providing billions of dollars in bonus payments. Despite their importance, it is unclear whether these aggregated ratings of plans – based in large part on process of care, patient experience, and intermediate outcome measures – ultimately signify differences that lead to improvements in health. Indeed, there remain questions about the direct relationship between many star ratings and health outcomes, whether star rating measures correlate to strong or weak performance on other dimensions of quality, and the extent

This study evaluates the relationship of star ratings with the number and length of hospital stays (including potentially-preventable admissions) and all-cause mortality rates. I rely on complete Medicare enrollment and MedPAR files and public MA data from 2009 through 2015. These data identify whether a beneficiary enrolled in MA and, if so, the star rating of their plan if they received drug coverage (as is currently the case for approximately 90 percent of MA enrollees). They also capture all hospital discharges at the vast majority of acute care PPS hospitals and all beneficiary deaths. The discharge data provide enough granularity to identify potentially-preventable hospitalizations based on


My findings will help policymakers determine how much weight to attach to star ratings in plan regulations and payment rules and will be informative as the government continues to pilot test a quality rating system on the federally-

: Florida implemented mandatory managed care for Medicaid enrollees in April 2014 via the Statewide Medicaid Managed Care (SMMC) program to improve access and coordination of care. This program enhanced access to primary care by increasing the number of primary care providers (PCPs), and after-hour appointment availability. The research objective of this study is to analyze the impact of the enhanced access to primary care on preventable

: We estimate a difference-in-difference (DD) model, comparing the change in the number of preventable hospitalizations in Zip Code Tabulated Areas (ZCTAs) with greater improvement in access to primary care (measured by percentage change of enrollee to provider ratio in the area from 2013 to 2015), compared with the changes of that in ZCTAs with less improvement in the access to primary care after the implementation of SMMC. We control for ZCTA specific socio-demographic characteristics, fixed effects for county of residence, year fixed effects, and a county-specific linear trend. The key explanatory variable is an interaction between the indicator for being a ZCTA in the top quartile of improvement in access to primary care, and the indicator for the post period. The main outcomes are numbers of preventable hospitalizations, i.e., whether the hospitalization was for an ambulatory care sensitive condition (ACSC), per 1000 residents in each enrollees’ ZCTA. We adopt the Prevention Quality Indicator (PQIs) developed by Agency for Healthcare Research and Quality (AHRQ) to identify hospitalizations for ACSCs.

: We compiled the analytic sample from three data sources. Florida inpatient discharge data from 2010 to 2015 provided information on inpatient visit. There were 1,837,294 discharges for Florida residents between the ages of 18 and 64 with a primary payer of Medicaid insurance, and no missing values on covariates used. We stratify the data into cohorts according to ZCTA, and quarter. The final analytic sample includes 19,621 stratified observations at the ZCTA-quarter level. We supplement the analyses with enrollee to PCP ratio in each ZCTA, created from Florida Medicaid provider data repository and 2010–2014 United States Census American Community Survey (ACS).

: We find that areas with greater improvement in the access to primary care experienced reductions in the incidence of overall preventable hospitalizations of 7.2 per 100,000 residents (18.9 percent) compared to areas with less improvement. Those areas also saw reductions in the hospitalizations for chronic ACSCs (reduction of 6.9 preventable hospitalizations per 100,000 residents (25.0 percent)) in the post-implementation period relative to other

: Our results show that areas with greater improvement in the access to primary care have greater reductions in hospitalizations for ACSCs, especially hospitalizations for chronic ACSCs, compared to areas with less

: Managed care in Medicaid provides a foundation for enrolling vulnerable populations and guaranteeing them better access to primary care and care coordination, to reduce cost of the program. Our


In 2017, the Centers for Medicare and Medicaid Services implemented the Merit-Based Incentive Payment System (MIPS), establishing a new payment program for clinicians participating in the fee-for-service Medicare program. As part of a broader push to link provider payments to value, the MIPS is a pay-for-performance model that intends to reward clinicians for improving quality of care and lowering spending by providing practices with bonuses or penalties based on their performance on quality and spending measures. Although the effects of the MIPS will not be known for several years, its basic design is similar to that of its predecessor, the Value-Based Payment Modifier (VM). From 2014 to 2016, the VM was phased in for physician practices meeting specific size thresholds (i.e., number of constituent clinicians), creating abrupt discontinuities in the exposure of practices to pay-for-performance incentives. We harnessed these discontinuities in a quasi-experimental regression discontinuity design to evaluate the VM's effects on performance measures assessed for all practices subject to the program.

Exploiting the phase-in of VM incentives based on practice size, we used regression discontinuity analysis and Medicare claims in 2014 to estimate differences in practice performance associated with the abrupt exposure of practices with ≥100 clinicians to full VM incentives (bonuses and penalties) and the exposure of practices with ≥10 clinicians to partial incentives (bonuses only). We repeated analyses using 2015 claims to assess the association of a second year of exposure to pay-for-performance incentives. We examined performance on four sets of outcomes: hospital admissions for ambulatory case-sensitive conditions (ACSCs), all-cause 30-day readmissions, mortality, and Medicare (Part A and Part B) spending per beneficiary. We conducted supplementary analyses to assess the robustness of our results to model specification and placebo tests to check whether discontinuities in outcomes at the VM's implementation

In 2014, there were no significant discontinuities at the ≥10-clinician threshold in the relationship between practice size and admissions for ACSCs (adjusted discontinuity:+0.003 admissions/beneficiary; 95% CI:-0.0003,0.006), proportion of admissions with readmission (+0.1 percentage points; 95% CI:-0.4,0.6), Medicare spending ($234/beneficiary; 95% CI:-$148,$616), or mortality (+0.2 percentage points; 95% CI:-0.1,0.5). Similarly, there were no discontinuities at the ≥100-clinician threshold in admissions for ACSCs (-0.002 admissions/beneficiary; 95% CI:-0.006,0.003), proportion of admissions with readmission (+0.3 percentage points; 95% CI:-0.6,1.2), spending (-$152/beneficiary; 95% CI:-$712,$408), or mortality (-0.1 percentage points; 95% CI:-0.5,0.3). Analyses of the ≥100-clinician threshold using 2015 data revealed no discontinuities associated with a second year of full exposure to the VM. Discontinuities estimated

The VM was not associated with significant differences in performance on program measures at thresholds where physicians' incentives differed. Our findings suggest that incentives in the VM were not sufficiently strong to


ASU Complete

Mental Health America Daniel Crowley; Justin Smith Complete

Georgia State University Lee Mobley Complete

Oregon Health & Sci. Univ. Complete

New Economic School, CEFIR Complete

Harvard T.H. Chan School of Public Health Austin Frakt Complete

Marisa Miraldo; Katharina Hauck Complete

University of Michigan Scott Regenbogen; Edward Norton Complete

Florida International University Xuan Tan; Sheng Guo Complete

Centre for Health Economics & Policy Innovation, Imperial College Business School


Centers for Disease Control and Prevention Paul Garbe; Robin Kuwahara Complete

DRM, CFACT, AHRQ CompleteDavid Lassman; Steven Heffler; Cathy Cowan; Thomas Selden

Vienna University of Economics and Business Renata Kosova; Giorgia Marini; Marisa Miraldo Complete

Royal Holloway, University of London Tanya Wilson Complete

The Urban Institute Lea Bart; Sharon K. Long Complete

University of Georgia Rebecca Walcott; Justin Ingels Complete

Congressional Budget Office Zirui Song Complete

Anglo-Chinese School (Independent) Complete

Anglo-Chinese School (Independent) Complete

Harvard University Complete

Harvard University Timothy Layton; Sherri Rose; Thomas McGuire Complete

University of Texas Medical Branch CompleteDaniel Jupiter; Ana Rodriguez; ThuyQuynh Do; Rahul Shah; Heidi Russell; John Livingston

Ben-Gurion University of the Negev Complete

Yale University Complete

Thomas Jefferson University Rachel Wojciechowski; Skyla Smith; Jing Ai Complete


University of Minnesota - SHADAC Brett Fried Complete

Morehead State University Complete

The University of North Carolina at Chapel Hill Kristine Falk; Joseph Rossi; Samuel Savitz Complete

Agency for Healthcare Research and Quality Complete


Florida International University CompleteImelda Moise; Karoline Mortensen; Sandra Decker

J. McWilliams; Alan Zaslavsky Complete



Health, Labor Markets, and the Economy






Income Elasticity of Demand for Tanning Bed Usage: Evidence from Survey Data

Sick and Tell: A Field Experiment Analyzing the Effects of an Illness-Related Employment Gap on the Callback Rate

Early life malnutrition and the evolution of human capital

Labor Market Effects of Medical Innovation: The Case of Breast and Prostate Cancer

Watch for Motorcycles! The Effects of Texting and Handheld Bans on Motorcyclist Fatalities

The Affordable Care Act and Women's Self-Employment





Health, Labor Markets, and the Economy Innovation and Diffusion of Medical Treatment

The Effects of Medicaid Expansion on Labor Market Outcomes: Evidence from Border Counties

Exploring the impact of new medical technology on workforce planning

The Impact of Nurse Practitioner Scope-of-Practice on NP Employment

Tattoos, Employment, and Earnings: Are Multiple, Visible, and Offensive Images Bad for Business?





Measles Vaccination and Human Capital Development in the United States

Estimated Annual and Lifetime Labor Productivity in the United States, 2006-2016

Does the Declining NCDs_DALYs Contribute to Economic Growth?—GMM Estimation on Global Countries Data

The long reach of childhood health problems on education, earnings and health




The Impact of Immigration Enforcement on Health Care Access and Utilization

Health Insurance and the Earnings Stability of Low-Income Households

Work-Family Conflict: A Comparative Analysis of Staff, Managerial and Executive Nurses

Health, Labor Markets, and the Economy Do Minimum Wages Improve Child Health?



The Effect of the Affordable Care Act on Entrepreneurship

Using Predictive Analytics for Early Identification of Short-Term Disability Claimants Who Exhaust Their Benefits





IMPACT OF THE AFFORDABLE CARE ACT’S EMPLOYER MANDATE ON INSURANCE COVERAGE

How did Affordable Care Act Exchanges Affect Individuals’ Willingness to Quit their Jobs?

Gender Homophily in Referral Networks: Consequences for the Medicare Physician Earnings Gap

The Relationship Between Abortion Rates and Macroeconomic Fluctuations




Health, Labor Markets, and the Economy Local Economic Conditions and Quality of Care

Effects of Informal Elderly Care on Labor Supply: Exploitation of Government Intervention on the Supply Side of Elderly Care Market

The Impacts of the Food Stamp Program on Mortality

Infant Health, Cognitive Performance and Earnings: Evidence from Inception of the Welfare State in Sweden

Health, Labor Markets, and the Economy Minimum Wages and Immigrant Health

Abstract

Using data on U.S. adults from the National Health Interview Survey, we estimate the income elasticity of demand for tanning bed usage among users and evaluate the factors that influence non-use of tanning beds. While controlling for individual characteristics, we show that the income elasticity of demand indicates that tanning bed usage is a normal good that is viewed as a necessity for users. For non-users, estimates show, after controlling for individual characteristics, that income is negatively associated with the probability of becoming a tanning bed user. Results suggest that usage of tanning beds grows for users as income increases and makes tanning usage more affordable. The findings imply that policymakers wanting to discourage the usage of tanning beds may have to tax usage considerably to cause a significant reduction in usage among continuing users.

Using a résumé-based correspondence test, we compare the employment consequences of an illness-related employment gap to those of an unexplained employment gap. Previous research shows that employment gaps, in general, have adverse effects on the probability of getting hired. It is not clear, however, if employers view spells of joblessness due to health issues distinctly. To shed light on this, we present a theoretical model in which employers use information on employment gaps as a signal of unobserved productivity and healthcare costs. We investigate the empirical implications of the model by sending three types of fictitious résumés to real job vacancies. One résumé indicates that the applicant is newly unemployed. The other résumés indicate employment gaps which are either unexplained or explained as being related to an illness. To signal an illness-related employment gap, a phrase in the cover letter explained that the employment gap was due to a physical illness followed by a full recovery. An additional signal on medical history was sent via information in the résumé that indicates involvement in a cancer recovery support group. The corresponding cover letters of the résumés with unexplained gaps did not provide any explanation for the gap. For the résumé of newly unemployed applicants, the length of the gap is limited to less than two months. Based on the literature, this is too short a gap to bring about adverse effects. The corresponding cover letter of newly unemployed applicants notes that the applicant left the last job because her family had to move from another state and that she is currently looking for a new job. From March to September, 2016, we sent 3,771 résumés to 1,257 sales, administrative, and accounting assistant jobs. Outcomes are measured in terms of differences in the callback rate of each type of résumé. The results of the experiment show that newly unemployed applicants had the highest callback rate (27.4%). Consistent with previous studies, résumés with an employment gap received lower callback rates, indicating that such gaps negatively affect hiring outcomes. However, résumés with an explained illness-related gap received a higher callback rate than résumés with an unexplained gap (25.6% versus 23.3%). Within the context of our theoretical model, these results suggest that the negative productivity signal of an unexplained gap outweighs undesirable factors associated with poor health history.

Extensive literature in health economics examine the persisting effect of early life adverse shocks on adult outcomes, and sharp exogenous shocks in fetal health are exploited to provide compelling evidence on the fetal origins hypothesis. This paper is motivated by the idea to synthesize the fetal origins literature and the evolution of children’s cognitive and non-cognitive abilities to provide a life-cycle framework to understand the origins of health inequality. In this framework, adult outcomes to the development of cognitive abilities, meanwhile cognitive abilities are jointly determined by environment, investment and initial genes. As argued by previous fetal origins literature, early life deprivation has long-standing and negative shocks on adult health or educational outcomes. We make the argument that even though the effect of early life (in utero) deprivation is disastrous, remediation and resilience induced by later investment can contribute to narrow the gaps between adverse shocks exposed cohort and their reference groups. This paper provides novel evidence from 1958-1961 China great famine on long-standing effect of early life deprivation on adult outcomes and how gaps induced by early life deprivation are narrowed by later life compensating investment. Exploiting unique datasets obtained from China Health and Retirement Longitudinal Survey (CHARLS), 2011 national baseline survey and 2014 life history survey, this paper intends to examine the long-term impacts of fetal malnutrition based on survivors in their 50s who were born during the 1959-1961 China’s great Famine. In addition, this study is interested to incorporate early life adverse shocks, family compensating investment into a life cycle development of human capital Our results support the Fetal Origins Hypothesis (Barker, 1992) that exposure to adverse conditions in early life (such as exposure to infectious diseases or malnutrition status) may causally affect health and mortality at old ages. We find that fetal exposure to malnutrition has large and long- lasting impacts on cognitive abilities, including immediate word recalling ability, delayed word recalling ability and having difficulty with drawing a picture. In addition, our results shows the effect of early life adverse shocks can be narrowed by family compensating investment, thus as guardian’s love and affection, their time and effort devoted to taking care of children, which is consistent with the framework of evolution of human capital put forward by J.J Heckman (2007).

Innovations in cancer treatment have lowered mortality, but little is know about their economic benefits and who benefits from them. In this paper, we assess the effect of improved treatment options over the last three decades on the labor market outcomes of breast and prostate cancer patients. We combine administrative tax return and cancer registry data from Canada with measures of medical innovation (approved drugs, academic publications, and patents) to estimate triple-differences regressions of employment and annual earnings. Our results show that the reductions in these labor market outcomes among cancer patients are partially offset by improved treatment options. Specifically, the decline in employment due to a cancer diagnosis is cut by 50 to 75 percent by the medical innovation that has occurred during the last three decades. We also find that the benefits of medical innovation are limited to cancer patients with high educational attainment. This result provides one explanation for the observed positive correlation between education, health, and income.

Motorcyclists account for a much higher proportion of traffic fatalities relative to the share of motorcycles among all vehicles and vehicle miles driven in the U.S. In this paper, we examine whether state-specific texting/handheld bans significantly influence motorcyclist fatalities. We use longitudinal multivariate analysis of state-specific traffic fatality data in the U.S. (2005-2015) from the Fatality Analysis Reporting System (FARS) merged with state-specific characteristics, texting/handheld device laws, and other traffic policies. We find that states with moderate and strong texting/handheld bans have significantly lower motorcyclist fatality rates even after controlling for numerous other factors and state fixed-effects. This result is driven mainly by multiple-vehicle motorcycle crashes as opposed to single-vehicle crashes. Although research is mixed on the effectiveness of texting/handheld device policies for overall traffic fatalities, our research indicates that motorcyclists may be at elevated risk of distracted driving and thus benefit greatly from these policies.

The Affordable Care Act (ACA) of 2010 improved and expanded availability of non-group health insurance. Previous studies have shown that women in the U.S. workforce value health insurance more than men do. Because prior to the ACA self-employed individuals did not have guaranteed access to affordable health insurance, womens lower rate of self-employment may partly have reflected job lock due to reliance on employer-based group coverage. This paper employs nationally-representative survey data for 2012-2016 and a difference-in-difference modeling approach to demonstrate that in fact, unmarried women have had significantly higher probability of self-employment since the ACA health insurance exchanges opened in 2014, coincident with their relatively higher uptake of private non-group health insurance purchased on state exchanges. This evidence demonstrates additional economic benefits of the ACA legislation, beyond its direct effects on healthcare access and medical costs.

The recent Medicaid expansions under the Affordable Care Act (ACA) have substantially increased access to health insurance among low-income individuals. Despite potential cost savings from better access to preventive care by Medicaid recipients, one common argument against the expansion of public insurance is that it may reduce the incentive for eligible individuals to remain in the workforce. Therefore, a correct understanding of the labor market impacts of expanding public insurance programs is important in evaluating the welfare effects of such policies. We seek to provide new empirical evidence on the effects of the recent Medicaid expansions on labor market outcomes. Unlike most existing studies that conduct the analysis at the state-level (e.g. Leung and Mas, 2016; Kaestner et al, 2015; Frisvold and Jung, 2017), our main analysis is at the county-level, which is a more appropriate approximation for the boundaries of local labor markets. Our identification strategy is based on the comparison of employment and wages in contiguous county-pairs in neighboring states (i.e. border counties) with different Medicaid expansion status. This is similar to the method employed by Dube et al (2010) to study the employment effects of minimum wage laws. Compared to a standard difference-in-differences approach, restricting the analysis to border-county pairs greatly improves the comparability between treatment and control units and controls for spatial heterogeneity, which can confound the relationship between Medicaid expansion status and employment or wages. In particular, our preferred specification uses only within county-pair variation in Medicaid expansion status and allows for arbitrary time effects across county-pairs. The main data source for our study is the 2008-2016 Quarterly Census of Employment and Wages (QCEW) collected by the Bureau of Labor Statistics (BLS). The QCEW is a comprehensive census of all establishments that report to the Unemployment Insurance programs, which contains about 97% of civilian employment nationwide. In contrast to other popular datasets such as the CPS or ACS, the QCEW allows the measurement of changes in employment and wages over time at the county-level. We estimate a set of distributed lag models which allow us to examine the dynamic effects of the Medicaid expansions. All of our models include controls for effective minimum wage, county population, state poverty rate, and state median household income. Consistent with previous studies, we do not find any statistically significant effects on either employment or wages when estimating a conventional county and year fixed effects on the full sample. However, in our border county-pair sample we find a small but statistically significant decrease in employment of between 1.2-1.5 percent one year after the implementation of the Medicaid expansions, but no statistically significant change in wages. Overall, our results suggest that the ACA Medicaid expansions had a modest effect on labor supply at the extensive margin. These findings contribute to the growing literature on the impact of public insurance programs on labor market outcomes.

This paper contributes to the existing literature on the diffusion of medical technologies. We apply panel data techniques to determine the manner in which technology is diffused across the NHS, with a particular emphasis on the impact that technology has on the workforce composition. We first examine the substitution or complementarity effects across different types of new technologies introduced into the NHS. Drawing on the work by Cutler and Huckman, we consider the diffusion of PTCA as it replaces CABG in the treatment of cardiovascular disease in England. We then estimate the degree to which the workforce reacts to the introduction of new technology, through calculating elasticity of supply measures. The data is combined from different sources to analyse these relationships: mainly, the UK Hospital Episodes Statistics (HES) and the NHS Electronic Staff Records (ESR). Analysis is at the provider level and the empirical specification explores the relationship between volume and workforce also controlling for provider and at risk population characteristics. Given the lack of quantitative evidence on the degree of substitution or complementarity across different forms of input in treating surgical cases, such analysis gives indicative estimates of productivity gains attributable to flexible workforce planning and technology uptake.

Keywords: Technology, workforce, substitution, production function

JEL Classification: O33, I12, C41, C33, J2

The costs of primary care have been rising and access to it may become limited because of a possible shortage in primary care physicians. Some state governments have addressed this issue by allowing Advanced Practice Registered Nurses (APRNs) to serve the population without the supervision of physicians. About half of the states permit nurse practitioners (NPs) to practice and/or prescribe drugs without physician supervision or collaboration. NPs in primary care charge lower prices than physicians and provide satisfactory quality of care, supported by existent literature. Moreover, increasing the number of NPs could alleviate access problems from a low supply of physicians. NP scope-of-practice (SOP) regulations have been changing in many states. This paper focuses on the impact of NP SOP regulations on access to primary health care. In particular, it will assess how state NP SOP regulations affect NP employment in the United States.

A recent study reports that having a tattoo does not diminish one’s likelihood of employment conditional on labor force participation or earnings conditional on being employed. Although novel in its design, scope, and contribution, the findings are somewhat limited because neither of the two datasets used by the authors contain information beyond the reporting of having one or more tattoos. To address this important shortcoming in the literature, the present study collected detailed data on number, coverage, and characteristics of tattoos (as well as employment status, labor supply, earnings, human capital measures, and other personal characteristics) among a sample of approximately 2,000 adults. We then estimate whether tattoos are significantly related to labor market outcomes using a much broader range of tattoo features (any, number, visible, offensive). Results show that few of the tattoo measures are significantly related to employment, labor supply, or earnings for either gender. These results have important and timely implications for those entering, or already participating in, the labor market that may have, or are considering, a tattoo(s). The implications also extend to employers as they establish workplace policies pertaining to tattoos among their job applicants and employees.

We develop and estimate a dynamic structural model of demand for a product line whose characteristics evolve over time as a consequence of consumer choices. We provide a new approach to the econometric challenge of estimating demand under uncertain innovation that includes sporadic breakthroughs and frequent, incremental changes. We use our framework to analyze consumer choice and the realized path of innovations over a long time horizon in a maturing product market: HIV drugs. In our model product quality is multidimensional since medications differ by their efficacy and their propensity to cause side effects. We allow for the possibility that new, more effective medicines can sometimes have harsher side effects. Atomistic consumers do not account for the role of aggregate demand on the speed and direction of innovation, leading to possible externalities. Using our estimated model we find that a planner that internalizes the externalities can increase welfare by at least 2% by increasing experimentation. Our results also indicate that providing monetary incentives for trial participation can be welfare improving.

My work evaluates changes in human capital development after the introduction of the measles vaccine in 1963. Prior to the vaccine 90% of people in the United States contracted the measles virus before turning 16. Mortality from measles was low (1 death in 1000 reported cases); however, after recovery from measles an individual is more susceptible to other infectious diseases for a prolonged period of time. Once the vaccine became available measles rates dramatically decreased. Using the unique biological profile of the measles virus I show the introduction of the measles vaccine led to positive short run health and educational outcomes and that these gains translated into higher earnings later in life through higher quality education and better health.

Human capital-based lifetime personal productivity estimates have long been used to measure the expected burden of productivity losses associated with diseases, injuries, or risk factors that lead to disability or premature death and the expected economic benefits from prevention. The most recent published estimates, which are now a decade old, followed convention in assuming that labor productivity would grow indefinitely at 1% per year in real (inflation-adjusted) terms. We present updated estimates of annual productivity for the period 2006-2016 using data from the American Community Survey, the American Time Use Survey, and the Current Population Survey. Productivity is the sum of market productivity calculated as gross annual personal labor earnings adjusted for employer-paid benefits and the imputed value of the non-market time spent producing household, caring, and volunteer services. Hours spent in non-market services were valued using the replacement cost method of the hourly market cost of hiring equivalent services to be performed. The present value of lifetime productivity at various ages was calculated for synthetic cohorts using annual productivity estimates, US life tables, discount rates, and assumptions about future labor productivity growth rates. Mean annual productivity was $57,324 for US adults in 2016, including $36,935 in market and $20,389 in non-marked productivity. Productivity in 2016 remained below pre-crisis (2006-2008) levels after adjustment for inflation, which implies negative real growth in labor productivity in the resident adult population during the study period. The present value of lifetime productivity at birth in 2016 calculated using a 3% real discount rate ranged from $1,193,498, assuming a 0.5% annual real growth in labor productivity in future years, to $1,468,669, assuming 1% annual productivity growth. These findings demonstrate that decisions as to whether to estimate total or just market productivity and what assumption to make about growth in future productivity are influential in estimates of avoidable economic productivity losses from sickness, disability, and premature death.

Relationships between health and economic growth are difficult to assess. Health is multidimensional and measured with errors. It is argued that commonly used health indicators in macroeconomic studies (life expectancy, infant mortality or specific diseases such as malaria or HIV/AIDS) imperfectly represent the global health status of population. The health indicators used in the previous literatures capture only one dimension of the population health. Actually, health is rather a complex notion and includes several dimensions which concern fatal and non-fatal issues of illness. The main thesis of this paper is that macroeconomic effects of the global health status are accurately caught by the Disability-Adjusted Life Year (DALYs) calculated by WHO. DALYs represents the burden of disease and can be thought of as a measurement of the gap between current health status and an ideal health situation. DALYs are commonly used in cost-effectiveness analyses but rarely used in macro economy. With the improvement of data, its impact to economic growth is starting to be noticed. According to the statistics of GBD 2015, between 1990 to 2015, global DALYs rate for all causes was projected to decrease from 48297 to 33440 per 100000, an overall decline of about 31%. For the NCDs_DALYs (Non-Communicable Diseases), although it declined from 20606 to 19990 per 100000, the decline proportion is only 3% and its proportion to all causes increased from 43% to 60%. That means the global disease burden is shifting from communicable disease and early life mortality to NCDs. The research is focused on the effects of NCDs_DALYs on the growth rate of per capita GDP in global countries. This paper use an expanded Solow growth model and a dynamic panel GMM estimator to testify whether DALYs contribute to economic growth. At the same time, physical capital and another important human capital—education are also included in the model. Considering the epidemiological transition and the income gap in different countries and regions, the whole sample of global countries were divided into four sub-samples according to the income level: high income countries, high and upper middle income countries, middle and low income countries, lower middle and low income countries. The empirical study has four important findings: First, NCDs_DALYs has a lagged and negative effect on GDP both in global countries and different income level countries. That means the declining of NCDs_DALYs contributes to the growth rate of per capita GDP. Second, the effect of NCDs_DALYs on economic growth is much more significant in developing countries than in developed countries. The correlation coefficients for lower middle and low income countries, middle and low income countries, and high income countries are separately -1.08***, -0.71*** and -0.04**.Third, education has a positive effect on GDP in high income countries and negative effect in lower middle and low income countries. Their coefficients are 0.33*** and -0.76***. Forth, physical capital always has significantly positive effect on GDP for the whole sample and sub-samples.

According to existing literature, early-life health affects later labor market outcomes such as earnings and work effort. We examine whether this holds for multiple dimensions of health and regardless of a country’s health-care system. We ask whether mental and physical health problems and poor general health by age 15 have similar or different influences on lifetime earnings, on years of schooling and on health problems later in life. Then we ask whether the health-care system the child lived in influenced the estimated effects of early health problems on lifetime earnings. We expect that early health problems reduce earnings and the most generous system is tied to the least negative long-term effects. Our analysis uses individual-level data from the first three waves of SHARE, a multidisciplinary and representative cross-national panel of the European population aged 50-plus. Waves 1 (2004/05) and 2 (2006/07) include information on sociodemographic background characteristics, current health, and socioeconomic status, as well as expectations of retirement age. Most of the data we use are from the third wave, SHARELIFE (2008/09), which is a retrospective survey conducted in 13 European countries as part of the SHARE project. We use our respondent’s country of childhood and a four-way system to characterize the health-care systems they lived in as children based on descriptions in the U.S. Social Security Administration’ Office of Policy (2002). These four groupings are: full coverage; considerable use of co-payments; limited coverage; and Socialist (full coverage but limited care). We find that the health-care system does make a difference in the size of the earnings penalty.

This paper examines the impact of increases in immigration enforcement on health care access and utilization of immigrants. While previous studies have examined the chilling effects of immigration enforcement on the take-up of public health insurance (Watson 2014), as well as the direct impact on self-reported health (Venkataramani et al. 2017), few studies have directly examined how deportations affect immigrant access to and use of health care. Harsher immigration enforcement and increased deportations may affect health of immigrants who remain in the country for multiple reasons. First, stress associated with the fear of being deported can have a negative effect on both mental and physical health outcomes, including anxiety, depression, cardiovascular disease, and high blood pressure. Second, increased deportations may deter undocumented immigrants, as well as their families, from using health care available to them for fear of interacting with authorities. Finally, chilling effects (as documented by Watson 2014) may discourage immigrants from obtaining health care coverage, even when they or their children are eligible, thus leading to lower health care use and worse health outcomes. Taking advantage of variation across counties in ICE apprehensions and deportations, combined with hospital inpatient discharge records, I will focus on the last two channels and examine the effect of deportations on emergency room (ER) admissions and Prevention Quality Indicators (PQIs). Both of these variables are measures of access to outpatient care that can be studied with inpatient data. Many ER admissions result from a lack of access to other channels of health care, such as preventive care or primary care, whiles PQIs reflect conditions for which hospitalizations can be easily avoided with regular outpatient care. Data on immigration enforcement comes from individual records of ICE apprehensions and removals across US counties under the Secure Communities Program for fiscal years 2007-2017. These data will be combined with information on policies that reflect the degree of cooperation between local law enforcement and federal immigration authorities, including 287(g) agreements at the county-level, state-level omnibus immigration enforcement laws, and “sanctuary city” designations across the US. Hospital discharge data comes from the Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases (SID). These databases contain administrative records on inpatient discharges from community hospitals for an unbalanced panel of states across the US from 1996-2011. Given both the current debates about immigration policy and general concern about issues of access to health care in the US, this study will provide valuable evidence on the effects of immigration enforcement as a barrier to health care access and the impact of changes in health care utilization among immigrants.

We evaluate the effect of health insurance on negative earnings shocks using the administrative tax data and survey responses of 4,975 low-income households. We exploit exogenous variation in the cost of private insurance under the Affordable Care Act using a regression discontinuity design. The probability of income loss falls by 22% at the income threshold for receiving insurance subsidies. An otherwise uninsured household that gets subsidized coverage is 25 percentage points less likely to report a job loss. Effects are concentrated among households with past health costs and exist only for “unexpected” forms of earnings variation, suggesting a health-productivity link. Rudimentary calculations based on our RD estimate imply a $256–$476 per year welfare benefit of health insurance in terms of reduced exposure to job loss.

Research Objective: The nursing workforce increasingly faces issues that affect clinical and managerial practice. One such issue is work-family conflict (WFC) and family-work conflict (FWC). Nurses face role strain as they confront the pressures from often competing work-and-family roles. While several studies have explored this issue among staff nurses, none to our knowledge have studied nurse managers. This study assesses WFC/FWC among staff versus executive and managerial nurses. Study design: This is an exploratory, cross-sectional survey. Survey questions included demographics, practice settings and roles, perceptions regarding the work environment, and perceptions of WFC/FWC. The survey instrument was validated in a number of prior studies. Descriptive statistics were conducted. Two separate ANOVAs were run to test the between and within groups scores for staff, managerial and executive nurses on WFC and FWC respectively. Two separate OLS regressions were run on models in which the dependent variables were WFC and FWC scales respectively and the independent variables were demographic, professional and work environment measures, focusing on the three different nursing roles. Population Studied: We randomly sampled registered nurses across the state of Florida. Of the nearly 5,000 email surveys, over 400 were completed. Nurses of varying roles and practice settings participated,. Principal Findings: Descriptively, nurses experienced more work-family conflict than family-work conflict. Regression analyses and ANOVAs indicated that staff nurses experienced less work-family conflict than nursing managers (second most) and nursing executives (highest). None of the nurse roles experienced significant levels of FWC. White nurses, compared to non-white nurses, experienced less WFC and FWC. WFC increased with shift length but FWC was not significantly affected by it. Paid leave for childbirth was associated with lower FWC. Age, gender, marital status and number of children in the home were demographics not significantly related to WFC/FWC. Practice setting, length of employment in current job, professional tenure, and educational level were professional variables not significant in the model. Managerial issues, staffing, nurse/physician collaboration and nursing competence were work environment issues not related to WFC/FWC. Conclusions and Implications for Policy or Practice: This study holds significant implications for practice. Nurse managers and executives showed significantly higher WRC than staff nurses. This may discourage a nurse from taking on leadership roles or lead to leaving them. In an era where nurse managers and leaders are needed, efforts must be taken to decrease WFC/FWC factors. Nonwhite nurses reported higher levels of both WFC and FWC. This may contribute to tension at the workplace and a difficult family life. Recruitment of people of color into nursing and their retention may be adversely affected. Leaders must continue to create platforms for people of all races and ethnicities to voice their work and family needs, and to be supported when doing so. Nurses working shifts over 8 hours had higher WFC levels. Although 12-hour shifts have been popular among staff and management, their use should be reevaluated. Finally, paid leave for childbirth is a program worth supporting as it was a factor in lower FWC.

Early life environmental conditions play a critical role in well-being over the life-course. Most studies of these roles have treated extreme shocks, such as famines, as natural experiments to study the causal impact of early life conditions. However, little attention is paid to the potential of labor market institutions, specifically income protection legislation, in shaping opportunities for investing in early life health and the subsequent impact on early life health outcomes. We study minimum wages around the time of birth and their effects on child stunting in Indonesia up to 5 years after birth. Indonesia is interesting not only because it carries the fifth highest burden of stunted children in the world, but also because minimum wages are an integral part of their social policy debate where worker protests are a regular occurrence. To the extent that minimum wages influence parental wage income, they might affect parental investments in child health. For example, if wages were to increase, families might be more likely to avail themselves of health services and engage in other salutary behaviors which may be particularly effective around the time of birth. However, mothers may also be more likely to spend more time in the labor market at the expense of care-giving activities. The time around birth is widely understood to be a critical period in shaping child nutrition and stunting levels, so that changes in parental economic conditions around the time of births may have particularly large effects on child health and nutrition. Using variation in annual fluctuations in real minimum wages in different provinces of Indonesia, we find that children exposed to increases in minimum wages in the year of birth have higher height-for-age (HAZ) scores in the first five years of their lives. Furthermore, we use data on parental wages to focus on children of parents for whom the minimum wage is most likely to be binding- those whose parents are in the bottom 25th and 50th quantiles of the wage distribution. Parents in upper tails of the wage distribution are considered as part of a placebo sample. Our estimated impacts are evident with difference-in-difference models with province and year-of-birth fixed effects and are robust to inclusion of biological sibling fixed effects, measures of child characteristics (age, gender) and parental characteristics (such as employment status, age and educational attainment, household income and assets) as well as community covariates (provincial GDP and unemployment rates). The effects are prominent particularly among children whose fathers earn in the bottom of the wage distribution, where as no effects are found for fathers earnings in the top part of the wage distribution (placebo). We also use multiple sources of consumer prices indices (CPI) (provincial and national) to explore robustness to different measures of real minimum wages. Our results are consistent with recent work from Indonesia based on “big push” models where increases in minimum wages lead to a movement away from an equilibrium of low wages and low labor demand to an equilibrium with high demand and high wages.

We aim to determine the effect of the main provisions of the Affordable Care Act on entrepreneurship. Previous work has found substantial evidence that some Americans are deterred from self-employment because they are concerned about losing their employer-based health insurance. The Affordable Care Act (ACA) introduced several reforms aimed at improving the market for individual insurance, with one objective of reducing the problem of "entrepreneurship lock." We will assess whether, and the extent to which the ACA has made progress towards this objective, across various relevant margins and sub-populations, based on a quasi-experimental research design applied to data from the American Community Survey and Current Population Survey. Because many of the main ACA provisions took effect in January 2014, it is only now becoming possible to study their effects; previous work on the ACA and entrepreneurship could only investigate the effect of relatively minor early provisions such as the dependent coverage mandate (Bailey 2013). As potential entrepreneurs are deterred because of concerns over their ability to maintain health insurance when they start a company, it is important to understand whether and to what extent the ACA solves this market failure for various groups, as well as the extent to which further reforms would be needed to keep the employer-based health insurance system from distorting the labor market and slowing the formation of new businesses. Market frictions that impede entrepreneurship may have adverse implications for efficient matching between worker skills and work and lead to labor market inefficiencies. Our main analysis focuses on evaluating the net effect of the main ACA provisions as a whole. However, if policymakers aim to expand, scale back, or replace the ACA it will be important to understand potential effects of individual components. Because many of the main provisions, including guaranteed issue, community rating, and the subsidized exchanges began simultaneously on January 1, 2014, it is empirically difficult to separate their effects. Nevertheless, first-step evidence on the partial effects of some of these individual components can be gleaned by exploiting previous state laws mandating guaranteed issue and community rating and by further exploiting individuals from Massachusetts, who were subject to an individual mandate, as an additional control group.

Objective Short-term disability insurance (STDI) pays partial wage replacements to employees temporarily unable to work due to “off-the-job” medical conditions. Most STDI policies replace wages for a fixed period, such as six months. Because wages are replaced only partially, STDI claimants have an incentive to return to work. Those who are unable to return before benefits expire may be at higher risk of job loss and receipt of long-term disability insurance (LTDI) or Social Security Disability Insurance (SSDI) benefits. An STDI claim can be an early identification point of workers with medical conditions who could, with adequate support, remain in the workforce. However, little is known about the factors influencing STDI duration or the transition to LTDI or SSDI benefits. Furthermore, careful timing and targeting of interventions is critical to efficiency; some workers may return to work without intervention, while others may not benefit from it. In this paper, we: (1) compare the performance of alternative models using information in claims data to predict exhaustion of STDI benefits; and (2) assess if waiting for some claims to resolve without intervention can improve the efficiency of targeting individuals for early intervention aimed at helping them remain in the workforce. Data Integrated Benefits Institute (IBI) Health and Productivity Benchmarking Data from 2011 through 2015, including 820,751 closed STDI claims from 8,587 small, medium, and large businesses associated with 9 disability insurance carriers and third-party leave administrators. The data include claim outcomes and claimant, employer, and insurance plan design characteristics. The primary outcome of interest is exhaustion of the STDI benefit. Methods We fit several predictive models to the data, including logistic regression, regularized logistic regression (using an elastic net), and random forests. We randomly divide our sample into training and testing sets, and select a model based on the area under the receiver operating characteristic curve. Individuals are flagged as having a high probability of exhausting their benefits using a predicted probability threshold, which was chosen to balance the tradeoff between sensitivity and specificity. We report the predictive performance of our models when applied to the test set. We perform the analysis for claims with benefit duration of 26 weeks, first using the full sample, then sequentially eliminating claims that resolved within 2, 4, and 6 weeks. Comparing across durations illustrates the potential efficiency gains of waiting to allow some claims to resolve on their own. Results The factors most strongly associated with exhaustion of STDI benefits are age, diagnosis, and employer industry. Waiting to allow some claims to resolve without intervention improves the efficiency of targeting efforts. Modeling based on observable factors helps further narrow the target population, with the machine learning techniques we use expected to outperform logistic regression in predictive performance. Conclusions Depending on the cost structure of the intervention, our approach could represent significant savings through efficient targeting of interventions to those STDI claimants who are most likely to benefit from them. We simulate the cost and benefit of several existing early intervention proposals under our predictive modeling framework.

I evaluate the impact of the Patient Protection Affordable Care Act’s (ACA) employer mandate on private employer sponsored insurance (ESI) coverage, private non-ESI coverage, and un-insurance status among less-skilled (high school dropouts and graduates) workers. The employer mandate requires firms with 100 (50) or more full-time equivalent (FTE) workers to provide affordable health insurance from 2015 (2016). The mandate was intended to increase insurance coverage among 19-64 year olds who participate in the labor market on a full-time basis, cannot qualify for Medicaid in their state, and are traditionally not offered or eligible for insurance. In addition, it prevents crowd-out into the private non-ESI insurance exchange marketplace or Medicaid. The literature notes that previous state and local level employer mandates increased private ESI coverage among previously uninsured workers. However, the ACA’s employer mandate may be ineffective since (1) small firms—traditionally the firms least likely to offer their workers with ESI coverage—are exempted from the mandate, (2) firms can avoid mandate by reducing workers’ hours so that they are not considered as FTE workers, (3) some firms can marginally reduce total FTE labor units and be exempted from the mandate, and (4) some low-income workers can qualify for Medicaid. In order to assess insurance coverage status among less-skilled workers—a group traditionally less likely to have private ESI coverage or afford private non-ESI coverage— based on firm size, I pool annual March supplements of the Current Population Survey from 2006 to 2016. The pooled cross-section data provides income, labor supply, and health insurance information from 2005 to 2015. Using a difference-in-difference model, I do not find evidence that that less-skilled workers were significantly more likely to report private ESI coverage if they worked in mid-sized and larger firms after the ACA was legislated in 2011, or after major components of the ACA was implemented in 2014. However, compared to less-skilled workers in small firms, workers were significantly less likely to report private non-ESI coverage if they worked in mid-sized by 0.03 percentage points (0.7%) and in large firms by 0.02 percentage points (0.6%) from 2013. In addition, while uninsured status wa not significantly different between workers in small firms and mid-sized firms after the legislation, workers in large firms were less likely to be uninsured by 0.04 percentage points (0.10%) between 2011 and 2013 when the ACA legislation was passed but the major components (individual mandate, Medicaid expansions, and employer mandate) were not implemented. Results are robust to parallel trends tests and falsification tests. While the employer mandate may not have increased private-ESI among workers, workers are less likely to have private non-ESI coverage in mid-sized and large firms. Evidence suggests that mandated firms may have avoided penalties by either reducing units of labor along the intensive and extensive margin, and/or ensured low-income, less-skilled workers enrolled in Medicaid. Less-skilled workers in mandate firms. Based on preliminary results, the mandate was unsuccessful in increasing private ESI coverage, decreasing un-insurance, and preventing crowd-out into Medicaid.

Background: In October 2013, the Affordable Care Act (ACA) health insurance exchanges began offering access to subsidized community-rated plans previously unavailable to many non-elderly adults. This shock represents a decrease in the opportunity cost to employment for those with employer-sponsored insurance (ESI). Studies have examined labor participation, but not willingness to voluntarily quit as a result of expanded availability of guaranteed issue coverage. There is similarly little evidence on the reasons people quit or the activities that individuals pursue after quitting. Objective: To test the effect of the launch of ACA health insurance exchanges (late 2013) on individuals’ willingness to quit their current main job. Methods: Repeated panels (2006-2015) from the Medical Expenditure Panel Survey (MEPS) contain information on current employment and quitting in five survey rounds over a two-year period. Linear probability models (LPM) with standard errors clustered at the household level predict the probability of quitting, controlling for temporal and seasonal trends using panel/round fixed effects. Models include controls for demographic and socioeconomic characteristics and are weighted using the provided longitudinal weights. The main effect compares the first open enrollment period (Q4 2013-Q1-2014 [panel 17, round 5]) with the same seasonal period (round 5) in different panels (first difference), then controls for time trends by subtracting out the same comparison from a different round (e.g., panel 17, round 2) (second difference). Results: The sample contains 105,348 individuals, 2,782 of whom quit their current main job over 10 overlapping panels (10 years) of 4 rounds each (round 1 established current job). The unadjusted rate of voluntary quitting in round 5 across all pre-exchange panels was 3.1%. During the first open enrollment (panel 17, round 5), the predicted probability of voluntary quitting was nearly a full percentage point higher than the average of the prior panels [0.93 percentage points (95% CI: 0.01, 1.80)] and consistent for individual years (first difference). Controlling for temporal changes (second difference), using the second and third rounds as referent, the average effects are even stronger [2.01 (0.93, 3.11)] and [1.54 (0.37, 2.70)], respectively. We also observe a stronger effect in earlier panel years (pre-2010), suggesting the period of and following the Great Recession may have dampened willingness to quit. The largest change in reasons for voluntarily quitting a position between the first open enrollment and pre-exchange periods was ‘quit to take care of home/family’ (10.6%) followed by ‘quit to go to school’ (7.58%). Conclusions: These results show support for the job lock hypothesis, indicating that individuals may be willing to quit their jobs when access to other sources of affordable health insurance become available. The magnitude of this result is small, but generalizable to a sizable population, unlike other previous attempts to characterize the effects of the ACA on labor outcomes. Specifically, we describe the pathway where individuals may be willing to quit their current job differentially around the time of the first ACA open enrollment period. Affordable access to health insurance can improve flexibility to move in and out of the labor force.

In this paper, I assess the extent to which gender gaps in earnings may be driven by physicians’ preference for working with specialists of the same gender. Analyzing administrative data on 100 million Medicare patient referrals, I provide robust evidence that physicians refer more to others of their same gender (i.e., referrals exhibit gender homophily). I show that homophily in referrals is predominantly driven by physicians’ decisions, rather than by endogenous sorting of physicians or patients. As 75% of referring physicians are men, my estimates suggest that gender homophily in referrals makes, all else being equal, demand for female physicians 5% lower than demand for male physicians, thus contributing to the persistence of gender inequality. Overall, my results point to the positive externality associated with increased female participation in medicine, and perhaps in other contexts where networking is important.

This is a test

BACKGROUND The goal of this study is to examine the causal effect of informal care on labor supply. Related studies in the United States and Europe analyzing the effect of informal care on labor supply have employed family structure and parental health as instrumental variables. They have not utilized institutional change as a natural experiment in estimating the effect of informal care on labor supply. As Van Houtven, Coe, and Skira (2013) point out, some of the instruments employed in literature are weak or their exogeneity is questionable. In 2000, the Japanese government has also implemented LTCI. In the Japanese care system, there are two important characteristics related to our study. First, there are three types of public nursing homes. Second, the supply of these nursing homes is regulated by the government. Our analysis utilizes the exogenous variation of government intervention on the supply side of the elderly care market to estimate the causal effect of informal care on labor supply.

METHODS We use the instrumental variables method. To the best of our knowledge, no study has thus far utilized exogenous institutional variation as an instrument to estimate the causal effect of informal care on labor supply.

RESULTS Analysis results reveal that the effect of informal elderly care on female labor force participation is negative. By contrast, male labor force participation is not affected by such care, since, in Japan, females spend more time on informal care than males. The increase in nursing home capacity is thus effective for decreasing the female burden of informal care.

CONCLUSIONS The effect of informal care for elderly on labor supply in both males and females is small. Especially, when compared with literature, the effect is smaller than in extant studies. The time spent on informal care in households is the focus on female household members. The government intervention is effective for increasing female labor supply.

The Supplemental Nutrition Assistance Program (SNAP; formerly called the Food Stamp Program or FSP) provided $66.5 billion in nutrition assistance to 44.2 million participants in fiscal year 2016. Prior research has examined the effects of SNAP on many different outcomes, but despite the program’s economic importance and potential health impacts, there are relatively few studies investigating how receiving food stamps affects health outcomes. Even fewer studies have examined SNAP’s health impacts on adult recipients, and these studies' findings are somewhat mixed. Given recent research finding a significantly higher risk of death for food stamp recipients compared to eligible non-recipients, these health effects may be significant. However, no prior study has satisfactorily examined the causal effects of food stamps on adult mortality outcomes. This study examines the effects of food stamps on health, with a focus on adult health. Specifically, I use the county-level rollout of the FSP from 1961 to 1975 as a source of plausibly exogenous variation in access to food stamps. I examine the effects of contemporaneous and multiple-year access to the FSP on various county-year-level mortality rates using fixed effects models. I consider effects on aggregate mortality rates, subgroup rates for sex, race, and age groups, and rates for specific causes of death to examine the different mechanisms through which food stamps might affect health. I find mixed results for the entire sample that indicate small overall effects of access to food stamps on mortality rates. However, among subsamples of poorer counties that are likely to benefit the most from food stamps, I find that implementation of the FSP reduces mortality rates for most groups over time.

We estimate impacts of exposure to a preventive infant health intervention trialled in Sweden in the early 1930s using purposively digitised birth registers linked to school catalogues, census files and tax records to generate longitudinal microdata that track 25,000 individuals through four stages of the life-course, from birth to age 71. This allows us to measure impacts on childhood health and cognitive skills at ages 7 and 10, educational choice during young adulthood, employment, earnings and occupation at age 36-40, and pension income at age 71. Leveraging quasi-random variation in eligibility by birth date and birth parish, we estimate that an additional year of exposure was associated with substantial increases in earnings and (public sector) employment among women, alongside no improvements for men. We also identify intervention effects on primary school test scores for men and women, and on secondary school completion for women only. A large part of the income gain for women can be attributed to secondary schooling and test score improvements, in particular at the top of the distribution. Using recent innovations in mediation analysis, we are able to show that school performance and secondary schooling enrollment are important mechanisms behind the adult gains in earnings. The greater investments of women in education are consistent with their comparative advantage in cognitive tasks, but opportunities are also likely to have played a role. Our sample cohorts were exposed to a massive expansion of the Swedish welfare state, which created unprecedented employment opportunities for women.

Quality of care is linked to improved health outcomes, cost efficiency, and high-value healthcare delivery. Recent policies, such as Medicare's value-based purchasing programs, seek to improve quality of care by tying provider reimbursements to the quality of services provided. However, the impact of factors external to the healthcare system, such as economic fluctuations, on healthcare quality is largely unexplored and recent work in this area has focused on elderly populations in nursing homes (Antwi and Bowblis, 2017; Stevens et al., 2012). The extent to which the broader adult population experiences cyclical changes in healthcare quality remains unclear. Patient experience is considered a component of quality as it captures the quality of health inputs as perceived by the patient. A large patient experience literature, primarily using cross-sectional and hospital-level data, identifies correlates between patient experience measures and individual and provider attributes. Using patient experience ratings from a nationally representative survey, I merge these two literatures by exploring the relationship between local economic conditions and the quality of healthcare services. In contrast to previous studies, I use individual-level, short panel data from the Medical Expenditure Panel Survey from 2002 to 2011 to explore local economic conditions as a potential determinant of improvements in patient ratings of healthcare quality. Specifically, I find evidence that increases in county unemployment rates during the Great Recession are associated with higher patient ratings of having enough time with their healthcare providers in the past year. The relationship is particularly strong for those with chronic conditions, suggesting changes in quality generated by economic fluctuations are larger for high-users of care. I find neither changes in individual's employment status nor insurance coverage mediate the positive correlation. I find some evidence that improvements in health during the Great Recession are a potential pathway. My findings imply that value-based purchasing programs may need to adjust quality targets to account for external factors, such as local economic conditions, when measures of patient satisfaction determine the incentive payment.

Immigrants are a vulnerable population due to their high risk for poor physical, psychological, and social health outcomes (Derose et al., 2007). Compared to similarly poor native-born citizens, low-income immigrants are more likely to lack health insurance. A sizeable segment of the immigrant population in the U.S. work at minimum wage jobs because they have lower educational attainment, limited language skills, and less social capital. Given the size and the rapid growth of immigrants in the U.S. workforce, it is important to examine the impact of minimum wage increases on immigrants’ health and access to care. Conventional economic theory predicts that increases in minimum wages raise hourly earnings and reduce employment. Orrenius and Zavodny (2008) find that hourly earnings for low-skilled adult immigrants increased with minimum wage increases but they find no adverse employment effects. Increases in earnings resulting from higher minimum wages create an income effect (among those who keep their jobs) which could then improve health outcomes. The potential for better health has led some policymakers to call for higher minimum wages specifically to improve health (e.g. Bhatia, 2014). Empirically, there is a growing literature aimed at determining if minimum wage increases positively affect the health of those individuals who retain their jobs and have higher earnings (e.g., Averett et al. 2016, Kronenberg et al. 2015, Lenhart 2015, Reeves et al. 2014, 2016, Strain et al. 2016, and Wehby et al. 2016). This paper examines whether minimum wage increases affect the health of low-skilled immigrants. We do so by using data from the National Health Interview Survey and estimating the following equation: yist=α+δ1MWst+ δ2Zit+ δ2Xst+θs+τt+εist where yist is our health outcome for individual i, residing in state s at year t; MWst is the minimum wage (or the ratio of the minimum wage to the state’s average wage); Zit is a vector of individual controls including age, marital status, language of interview, race/ethnicity, citizenship status, years in the U.S. Xst is a vector of state-specific time-varying economic and policy controls that may be correlated with minimum wages and health including immigrant’s access to health care, welfare and food benefits after the 1996 welfare reform act, the percent of the state’s workforce covered by a collective bargaining agreement, the percent of union membership, the state unemployment rate, state cigarette taxes, the percent of the state’s population below the poverty line, whether the state has an e-verify mandate and whether the state allows for public health insurance for unauthorized children and adults. θ s is the time-invariant state effect; τt is the time-invariant year effect, and εist is an error term. Standard errors will be clustered by state. Preliminary results using the Current Population Survey which has a much more limited set of health outcomes shows that there is likely to be an income effect and an effect of minimum wages on self-rated health for working immigrants. Using the NHIS data we will be able to test a variety of health and health related outcomes.

Abstract

Using data on U.S. adults from the National Health Interview Survey, we estimate the income elasticity of demand for tanning bed usage among users and evaluate the factors that influence non-use of tanning beds. While controlling for individual characteristics, we show that the income elasticity of demand indicates that tanning bed usage is a normal good that is viewed as a necessity for users. For non-users, estimates show, after controlling for individual characteristics, that income is negatively associated with the probability of becoming a tanning bed user. Results suggest that usage of tanning beds grows for users as income increases and makes tanning usage more affordable. The findings imply that policymakers wanting to discourage the usage of tanning beds may have to tax usage considerably to cause a significant reduction in usage among continuing users.

Using a résumé-based correspondence test, we compare the employment consequences of an illness-related employment gap to those of an unexplained employment gap. Previous research shows that employment gaps, in general, have adverse effects on the probability of getting hired. It is not clear, however, if employers view spells of joblessness due to health issues distinctly. To shed light on this, we present a theoretical model in which employers use information on employment gaps as a signal of unobserved productivity and healthcare costs. We investigate the empirical implications of the model by sending three types of fictitious résumés to real job vacancies. One résumé indicates that the applicant is newly unemployed. The other résumés indicate employment gaps which are either unexplained or explained as being related to an illness. To signal an illness-related employment gap, a phrase in the cover letter explained that the employment gap was due to a physical illness followed by a full recovery. An additional signal on medical history was sent via information in the résumé that indicates involvement in a cancer recovery support group. The corresponding cover letters of the résumés with unexplained gaps did not provide any explanation for the gap. For the résumé of newly unemployed applicants, the length of the gap is limited to less than two months. Based on the literature, this is too short a gap to bring about adverse effects. The corresponding cover letter of newly unemployed applicants notes that the applicant left the last job because her family had to move from another state and that she is currently looking for a new job. From March to September, 2016, we sent 3,771 résumés to 1,257 sales, administrative, and accounting assistant jobs. Outcomes are measured in terms of differences in the callback rate of each type of résumé. The results of the experiment show that newly unemployed applicants had the highest callback rate (27.4%). Consistent with previous studies, résumés with an employment gap received lower callback rates, indicating that such gaps negatively affect hiring outcomes. However, résumés with an explained illness-related gap received a higher callback rate than résumés with an unexplained gap (25.6% versus 23.3%). Within the context of our theoretical model, these results suggest that the negative productivity signal of an unexplained gap outweighs undesirable factors associated with poor health history.

Extensive literature in health economics examine the persisting effect of early life adverse shocks on adult outcomes, and sharp exogenous shocks in fetal health are exploited to provide compelling evidence on the fetal origins hypothesis. This paper is motivated by the idea to synthesize the fetal origins literature and the evolution of children’s cognitive and non-cognitive abilities to provide a life-cycle framework to understand the origins of health inequality. In this framework, adult outcomes to the development of cognitive abilities, meanwhile cognitive abilities are jointly determined by environment, investment and initial genes. As argued by previous fetal origins literature, early life deprivation has long-standing and negative shocks on adult health or educational outcomes. We make the argument that even though the effect of early life (in utero) deprivation is disastrous, remediation and resilience induced by later investment can contribute to narrow the gaps between adverse shocks exposed cohort and their reference groups. This paper provides novel evidence from 1958-1961 China great famine on long-standing effect of early life deprivation on adult outcomes and how gaps induced by early life deprivation are narrowed by later life compensating investment. Exploiting unique datasets obtained from China Health and Retirement Longitudinal Survey (CHARLS), 2011 national baseline survey and 2014 life history survey, this paper intends to examine the long-term impacts of fetal malnutrition based on survivors in their 50s who were born during the 1959-1961 China’s great Famine. In addition, this study is interested to incorporate early life adverse shocks, family compensating investment into a life cycle

Our results support the Fetal Origins Hypothesis (Barker, 1992) that exposure to adverse conditions in early life (such as exposure to infectious diseases or malnutrition status) may causally affect health and mortality at old ages. We find that fetal exposure to malnutrition has large and long- lasting impacts on cognitive abilities, including immediate word recalling ability, delayed word recalling ability and having difficulty with drawing a picture. In addition, our results shows the effect of early life adverse shocks can be narrowed by family compensating investment, thus as guardian’s love and affection, their time and effort devoted to taking care of children, which is consistent with the framework of

Innovations in cancer treatment have lowered mortality, but little is know about their economic benefits and who benefits from them. In this paper, we assess the effect of improved treatment options over the last three decades on the labor market outcomes of breast and prostate cancer patients. We combine administrative tax return and cancer registry data from Canada with measures of medical innovation (approved drugs, academic publications, and patents) to estimate triple-differences regressions of employment and annual earnings. Our results show that the reductions in these labor market outcomes among cancer patients are partially offset by improved treatment options. Specifically, the decline in employment due to a cancer diagnosis is cut by 50 to 75 percent by the medical innovation that has occurred during the last three decades. We also find that the benefits of medical innovation are limited to cancer patients with high educational attainment. This result provides one explanation for the observed positive correlation between education, health, and income.

Motorcyclists account for a much higher proportion of traffic fatalities relative to the share of motorcycles among all vehicles and vehicle miles driven in the U.S. In this paper, we examine whether state-specific texting/handheld bans significantly influence motorcyclist fatalities. We use longitudinal multivariate analysis of state-specific traffic fatality data in the U.S. (2005-2015) from the Fatality Analysis Reporting System (FARS) merged with state-specific characteristics, texting/handheld device laws, and other traffic policies. We find that states with moderate and strong texting/handheld bans have significantly lower motorcyclist fatality rates even after controlling for numerous other factors and state fixed-effects. This result is driven mainly by multiple-vehicle motorcycle crashes as opposed to single-vehicle crashes. Although research is mixed on the effectiveness of texting/handheld device policies for overall traffic fatalities, our research indicates that motorcyclists may be at elevated risk of distracted driving and thus benefit greatly from these policies.

The Affordable Care Act (ACA) of 2010 improved and expanded availability of non-group health insurance. Previous studies have shown that women in the U.S. workforce value health insurance more than men do. Because prior to the ACA self-employed individuals did not have guaranteed access to affordable health insurance, womens lower rate of self-employment may partly have reflected job lock due to reliance on employer-based group coverage. This paper employs nationally-representative survey data for 2012-2016 and a difference-in-difference modeling approach to demonstrate that in fact, unmarried women have had significantly higher probability of self-employment since the ACA health insurance exchanges opened in 2014, coincident with their relatively higher uptake of private non-group health insurance purchased on state exchanges. This evidence demonstrates additional economic benefits of the ACA legislation,

The recent Medicaid expansions under the Affordable Care Act (ACA) have substantially increased access to health insurance among low-income individuals. Despite potential cost savings from better access to preventive care by Medicaid recipients, one common argument against the expansion of public insurance is that it may reduce the incentive for eligible individuals to remain in the workforce. Therefore, a correct understanding of the labor market impacts of expanding public insurance programs is important in evaluating the welfare effects of such policies. We seek to provide new empirical evidence on the effects of the recent Medicaid expansions on labor market outcomes. Unlike most existing studies that conduct the analysis at the state-level (e.g. Leung and Mas, 2016; Kaestner et al, 2015; Frisvold and Jung, 2017), our main analysis is at the county-level, which is a more appropriate approximation for the boundaries of local labor markets. Our identification strategy is based on the comparison of employment and wages in contiguous county-pairs in neighboring states (i.e. border counties) with different Medicaid expansion status. This is similar to the method employed by Dube et al (2010) to study the employment effects of minimum wage laws. Compared to a standard difference-in-differences approach, restricting the analysis to border-county pairs greatly improves the comparability between treatment and control units and controls for spatial heterogeneity, which can confound the relationship between Medicaid expansion status and employment or wages. In particular, our preferred specification uses only within county-pair variation in Medicaid expansion status and allows for arbitrary time effects

The main data source for our study is the 2008-2016 Quarterly Census of Employment and Wages (QCEW) collected by the Bureau of Labor Statistics (BLS). The QCEW is a comprehensive census of all establishments that report to the Unemployment Insurance programs, which contains about 97% of civilian employment nationwide. In contrast to other popular datasets such as the CPS or ACS, the QCEW allows the measurement of changes in employment and wages

We estimate a set of distributed lag models which allow us to examine the dynamic effects of the Medicaid expansions. All of our models include controls for effective minimum wage, county population, state poverty rate, and state median household income. Consistent with previous studies, we do not find any statistically significant effects on either employment or wages when estimating a conventional county and year fixed effects on the full sample. However, in our border county-pair sample we find a small but statistically significant decrease in employment of between 1.2-1.5 percent one year after the implementation of the Medicaid expansions, but no statistically significant change in wages. Overall, our results suggest that the ACA Medicaid expansions had a modest effect on labor supply at the extensive margin. These findings contribute to the growing literature on the impact of public insurance programs on labor market

This paper contributes to the existing literature on the diffusion of medical technologies. We apply panel data techniques to determine the manner in which technology is diffused across the NHS, with a particular emphasis on the impact that technology has on the workforce composition. We first examine the substitution or complementarity effects across different types of new technologies introduced into the NHS. Drawing on the work by Cutler and Huckman, we consider the diffusion of PTCA as it replaces CABG in the treatment of cardiovascular disease in England. We then estimate the degree to which the workforce reacts to the introduction of new technology, through calculating elasticity of supply measures. The data is combined from different sources to analyse these relationships: mainly, the UK Hospital Episodes Statistics (HES) and the NHS Electronic Staff Records (ESR). Analysis is at the provider level and the empirical specification explores the relationship between volume and workforce also controlling for provider and at risk population characteristics. Given the lack of quantitative evidence on the degree of substitution or complementarity across different forms of input in treating surgical cases, such analysis gives indicative estimates of productivity gains attributable to flexible workforce planning and technology uptake.


A recent study reports that having a tattoo does not diminish one’s likelihood of employment conditional on labor force participation or earnings conditional on being employed. Although novel in its design, scope, and contribution, the findings are somewhat limited because neither of the two datasets used by the authors contain information beyond the reporting of having one or more tattoos. To address this important shortcoming in the literature, the present study collected detailed data on number, coverage, and characteristics of tattoos (as well as employment status, labor supply, earnings, human capital measures, and other personal characteristics) among a sample of approximately 2,000 adults. We then estimate whether tattoos are significantly related to labor market outcomes using a much broader range of tattoo features (any, number, visible, offensive). Results show that few of the tattoo measures are significantly related to employment, labor supply, or earnings for either gender. These results have important and timely implications for those entering, or already participating in, the labor market that may have, or are considering, a tattoo(s). The implications also extend to employers as they establish workplace policies pertaining to tattoos among their job applicants and employees.


Human capital-based lifetime personal productivity estimates have long been used to measure the expected burden of productivity losses associated with diseases, injuries, or risk factors that lead to disability or premature death and the expected economic benefits from prevention. The most recent published estimates, which are now a decade old, followed convention in assuming that labor productivity would grow indefinitely at 1% per year in real (inflation-adjusted) terms. We present updated estimates of annual productivity for the period 2006-2016 using data from the American Community Survey, the American Time Use Survey, and the Current Population Survey. Productivity is the sum of market productivity calculated as gross annual personal labor earnings adjusted for employer-paid benefits and the imputed value of the non-market time spent producing household, caring, and volunteer services. Hours spent in non-market services were valued using the replacement cost method of the hourly market cost of hiring equivalent services to be performed. The present value of lifetime productivity at various ages was calculated for synthetic cohorts using annual productivity estimates, US life tables, discount rates, and assumptions about future labor productivity growth rates. Mean annual productivity was $57,324 for US adults in 2016, including $36,935 in market and $20,389 in non-marked productivity. Productivity in 2016 remained below pre-crisis (2006-2008) levels after adjustment for inflation, which implies negative real growth in labor productivity in the resident adult population during the study period. The present value of lifetime productivity at birth in 2016 calculated using a 3% real discount rate ranged from $1,193,498, assuming a 0.5% annual real growth in labor productivity in future years, to $1,468,669, assuming 1% annual productivity growth. These findings demonstrate that decisions as to whether to estimate total or just market productivity and what assumption to make about growth in future productivity are influential in estimates of avoidable economic

Relationships between health and economic growth are difficult to assess. Health is multidimensional and measured with errors. It is argued that commonly used health indicators in macroeconomic studies (life expectancy, infant mortality or specific diseases such as malaria or HIV/AIDS) imperfectly represent the global health status of population. The health indicators used in the previous literatures capture only one dimension of the population health. Actually, health is rather a complex notion and includes several dimensions which concern fatal and non-fatal issues of illness. The main thesis of this paper is that macroeconomic effects of the global health status are accurately caught by the Disability-Adjusted Life Year (DALYs) calculated by WHO. DALYs represents the burden of disease and can be thought of as a measurement of the gap between current health status and an ideal health situation. DALYs are commonly used in cost-effectiveness analyses but rarely used in macro economy. With the improvement of data, its impact to economic growth is starting to be noticed. According to the statistics of GBD 2015, between 1990 to 2015, global DALYs rate for all causes was projected to decrease from 48297 to 33440 per 100000, an overall decline of about 31%. For the NCDs_DALYs (Non-Communicable Diseases), although it declined from 20606 to 19990 per 100000, the decline proportion is only 3% and its proportion to all causes increased from 43% to 60%. That means the global disease burden is

The research is focused on the effects of NCDs_DALYs on the growth rate of per capita GDP in global countries. This paper use an expanded Solow growth model and a dynamic panel GMM estimator to testify whether DALYs contribute to economic growth. At the same time, physical capital and another important human capital—education are also included in the model. Considering the epidemiological transition and the income gap in different countries and regions, the whole sample of global countries were divided into four sub-samples according to the income level: high income countries, high and upper middle income countries, middle and low income countries, lower middle and low income

The empirical study has four important findings: First, NCDs_DALYs has a lagged and negative effect on GDP both in global countries and different income level countries. That means the declining of NCDs_DALYs contributes to the growth rate of per capita GDP. Second, the effect of NCDs_DALYs on economic growth is much more significant in developing countries than in developed countries. The correlation coefficients for lower middle and low income countries, middle and low income countries, and high income countries are separately -1.08***, -0.71*** and -0.04**.Third, education has a positive effect on GDP in high income countries and negative effect in lower middle and low income countries. Their coefficients are 0.33*** and -0.76***. Forth, physical capital always has significantly positive effect on GDP for the whole sample and sub-samples.

According to existing literature, early-life health affects later labor market outcomes such as earnings and work effort. We examine whether this holds for multiple dimensions of health and regardless of a country’s health-care system. We ask whether mental and physical health problems and poor general health by age 15 have similar or different influences on lifetime earnings, on years of schooling and on health problems later in life. Then we ask whether the health-care system the child lived in influenced the estimated effects of early health problems on lifetime earnings. We expect that early health problems reduce earnings and the most generous system is tied to the least negative long-term effects. Our analysis uses individual-level data from the first three waves of SHARE, a multidisciplinary and representative cross-national panel of the European population aged 50-plus. Waves 1 (2004/05) and 2 (2006/07) include information on sociodemographic background characteristics, current health, and socioeconomic status, as well as expectations of retirement age. Most of the data we use are from the third wave, SHARELIFE (2008/09), which is a retrospective survey

We use our respondent’s country of childhood and a four-way system to characterize the health-care systems they lived in as children based on descriptions in the U.S. Social Security Administration’ Office of Policy (2002). These four groupings are: full coverage; considerable use of co-payments; limited coverage; and Socialist (full coverage but limited care). We find that the health-care system does make a difference in the size of the earnings penalty.

This paper examines the impact of increases in immigration enforcement on health care access and utilization of immigrants. While previous studies have examined the chilling effects of immigration enforcement on the take-up of public health insurance (Watson 2014), as well as the direct impact on self-reported health (Venkataramani et al. 2017), few studies have directly examined how deportations affect immigrant access to and use of health care. Harsher immigration enforcement and increased deportations may affect health of immigrants who remain in the country for multiple reasons. First, stress associated with the fear of being deported can have a negative effect on both mental and physical health outcomes, including anxiety, depression, cardiovascular disease, and high blood pressure. Second, increased deportations may deter undocumented immigrants, as well as their families, from using health care available to them for fear of interacting with authorities. Finally, chilling effects (as documented by Watson 2014) may discourage immigrants from obtaining health care coverage, even when they or their children are eligible, thus leading to lower health care use and worse health outcomes. Taking advantage of variation across counties in ICE apprehensions and deportations, combined with hospital inpatient discharge records, I will focus on the last two channels and examine the effect of deportations on emergency room (ER) admissions and Prevention Quality Indicators (PQIs). Both of these variables are measures of access to outpatient care that can be studied with inpatient data. Many ER admissions result from a lack of access to other channels of health care, such as preventive care or primary care, whiles PQIs reflect conditions for which hospitalizations can be easily avoided with regular outpatient care. Data on immigration enforcement comes from individual records of ICE apprehensions and removals across US counties under the Secure Communities Program for fiscal years 2007-2017. These data will be combined with information on policies that reflect the degree of cooperation between local law enforcement and federal immigration authorities, including 287(g) agreements at the county-level, state-level omnibus immigration enforcement laws, and “sanctuary city” designations across the US. Hospital discharge data comes from the Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases (SID). These databases contain administrative records on inpatient discharges from community hospitals for an unbalanced panel of states across the US from 1996-2011. Given both the current debates about immigration policy and general concern about issues of access to health care in the US, this study will provide valuable evidence on the effects of immigration enforcement as a barrier to health care access and the impact of changes in health care utilization among immigrants.

We evaluate the effect of health insurance on negative earnings shocks using the administrative tax data and survey responses of 4,975 low-income households. We exploit exogenous variation in the cost of private insurance under the Affordable Care Act using a regression discontinuity design. The probability of income loss falls by 22% at the income threshold for receiving insurance subsidies. An otherwise uninsured household that gets subsidized coverage is 25 percentage points less likely to report a job loss. Effects are concentrated among households with past health costs and exist only for “unexpected” forms of earnings variation, suggesting a health-productivity link. Rudimentary calculations based on our RD estimate imply a $256–$476 per year welfare benefit of health insurance in terms of reduced exposure to job loss.

: The nursing workforce increasingly faces issues that affect clinical and managerial practice. One such issue is work-family conflict (WFC) and family-work conflict (FWC). Nurses face role strain as they confront the pressures from often competing work-and-family roles. While several studies have explored this issue among staff nurses, none to our knowledge have studied nurse managers. This study assesses WFC/FWC among staff versus executive

: This is an exploratory, cross-sectional survey. Survey questions included demographics, practice settings and roles, perceptions regarding the work environment, and perceptions of WFC/FWC. The survey instrument was validated in a number of prior studies. Descriptive statistics were conducted. Two separate ANOVAs were run to test the between and within groups scores for staff, managerial and executive nurses on WFC and FWC respectively. Two separate OLS regressions were run on models in which the dependent variables were WFC and FWC scales respectively and the independent variables were demographic, professional and work environment measures, focusing on the

: We randomly sampled registered nurses across the state of Florida. Of the nearly 5,000 email surveys, over 400 were completed. Nurses of varying roles and practice settings participated,. : Descriptively, nurses experienced more work-family conflict than family-work conflict. Regression analyses and ANOVAs indicated that staff nurses experienced less work-family conflict than nursing managers (second

most) and nursing executives (highest). None of the nurse roles experienced significant levels of FWC. White nurses, compared to non-white nurses, experienced less WFC and FWC. WFC increased with shift length but FWC was not significantly affected by it. Paid leave for childbirth was associated with lower FWC. Age, gender, marital status and number of children in the home were demographics not significantly related to WFC/FWC. Practice setting, length of employment in current job, professional tenure, and educational level were professional variables not significant in the model. Managerial issues, staffing, nurse/physician collaboration and nursing competence were work environment

This study holds significant implications for practice. Nurse managers and executives showed significantly higher WRC than staff nurses. This may discourage a nurse from taking on leadership roles or lead to leaving them. In an era where nurse managers and leaders are needed, efforts must be taken to decrease WFC/FWC factors. Nonwhite nurses reported higher levels of both WFC and FWC. This may contribute to tension at the workplace and a difficult family life. Recruitment of people of color into nursing and their retention may be adversely affected. Leaders must continue to create platforms for people of all races and ethnicities to voice their work and family needs, and to be supported when doing so. Nurses working shifts over 8 hours had higher WFC levels. Although 12-hour shifts have been popular among staff and management, their use should be reevaluated. Finally, paid leave for childbirth is a program worth supporting as it was

Early life environmental conditions play a critical role in well-being over the life-course. Most studies of these roles have treated extreme shocks, such as famines, as natural experiments to study the causal impact of early life conditions. However, little attention is paid to the potential of labor market institutions, specifically income protection legislation, in shaping opportunities for investing in early life health and the subsequent impact on early life health outcomes. We study minimum wages around the time of birth and their effects on child stunting in Indonesia up to 5 years after birth. Indonesia is interesting not only because it carries the fifth highest burden of stunted children in the world, but also because minimum wages are an integral part of their social policy debate where worker protests are a regular occurrence. To the extent that minimum wages influence parental wage income, they might affect parental investments in child health. For example, if wages were to increase, families might be more likely to avail themselves of health services and engage in other salutary behaviors which may be particularly effective around the time of birth. However, mothers may also be more likely to spend more time in the labor market at the expense of care-giving activities. The time around birth is widely understood to be a critical period in shaping child nutrition and stunting levels, so that changes in parental economic conditions around the time of births may have particularly large effects on child health and nutrition. Using variation in annual fluctuations in real minimum wages in different provinces of Indonesia, we find that children exposed to increases in minimum wages in the year of birth have higher height-for-age (HAZ) scores in the first five years of their lives. Furthermore, we use data on parental wages to focus on children of parents for whom the minimum wage is most likely to be binding- those whose parents are in the bottom 25th and 50th quantiles of the wage distribution. Parents in upper tails of the wage distribution are considered as part of a placebo sample. Our estimated impacts are evident with difference-in-difference models with province and year-of-birth fixed effects and are robust to inclusion of biological sibling fixed effects, measures of child characteristics (age, gender) and parental characteristics (such as employment status, age and educational attainment, household income and assets) as well as community covariates (provincial GDP and unemployment rates). The effects are prominent particularly among children whose fathers earn in the bottom of the wage distribution, where as no effects are found for fathers earnings in the top part of the wage distribution (placebo). We also use multiple sources of consumer prices indices (CPI) (provincial and national) to explore robustness to different measures of real minimum wages. Our results are consistent with recent work from Indonesia based on “big push” models where increases in minimum wages lead to a movement away from an equilibrium of low wages and low labor demand to an equilibrium with high demand and high wages.

We aim to determine the effect of the main provisions of the Affordable Care Act on entrepreneurship. Previous work has found substantial evidence that some Americans are deterred from self-employment because they are concerned about losing their employer-based health insurance. The Affordable Care Act (ACA) introduced several reforms aimed at improving the market for individual insurance, with one objective of reducing the problem of "entrepreneurship lock." We will assess whether, and the extent to which the ACA has made progress towards this objective, across various relevant margins and sub-populations, based on a quasi-experimental research design applied to data from the

Because many of the main ACA provisions took effect in January 2014, it is only now becoming possible to study their effects; previous work on the ACA and entrepreneurship could only investigate the effect of relatively minor early provisions such as the dependent coverage mandate (Bailey 2013). As potential entrepreneurs are deterred because of concerns over their ability to maintain health insurance when they start a company, it is important to understand whether and to what extent the ACA solves this market failure for various groups, as well as the extent to which further reforms would be needed to keep the employer-based health insurance system from distorting the labor market and slowing the formation of new businesses. Market frictions that impede entrepreneurship may have adverse implications for efficient matching between worker skills and work and lead to labor market inefficiencies. Our main analysis focuses on evaluating the net effect of the main ACA provisions as a whole. However, if policymakers aim to expand, scale back, or replace the ACA it will be important to understand potential effects of individual components. Because many of the main provisions, including guaranteed issue, community rating, and the subsidized exchanges began simultaneously on January 1, 2014, it is empirically difficult to separate their effects. Nevertheless, first-step evidence on the partial effects of some of these individual components can be gleaned by exploiting previous state laws mandating guaranteed issue and community rating and by further exploiting individuals from

Short-term disability insurance (STDI) pays partial wage replacements to employees temporarily unable to work due to “off-the-job” medical conditions. Most STDI policies replace wages for a fixed period, such as six months. Because wages are replaced only partially, STDI claimants have an incentive to return to work. Those who are unable to return before benefits expire may be at higher risk of job loss and receipt of long-term disability insurance (LTDI) or Social Security Disability Insurance (SSDI) benefits. An STDI claim can be an early identification point of workers with medical conditions who could, with adequate support, remain in the workforce. However, little is known about the factors influencing STDI duration or the transition to LTDI or SSDI benefits. Furthermore, careful timing and targeting of interventions is critical to efficiency; some workers may return to work without intervention, while others may not benefit from it. In this paper, we: (1) compare the performance of alternative models using information in claims data to predict exhaustion of STDI benefits; and (2) assess if waiting for some claims to resolve without intervention can improve the efficiency of targeting individuals for early intervention aimed at helping them remain in the workforce.

Integrated Benefits Institute (IBI) Health and Productivity Benchmarking Data from 2011 through 2015, including 820,751 closed STDI claims from 8,587 small, medium, and large businesses associated with 9 disability insurance carriers and third-party leave administrators. The data include claim outcomes and claimant, employer, and insurance plan design characteristics. The primary outcome of interest is exhaustion of the STDI benefit.

We fit several predictive models to the data, including logistic regression, regularized logistic regression (using an elastic net), and random forests. We randomly divide our sample into training and testing sets, and select a model based on the area under the receiver operating characteristic curve. Individuals are flagged as having a high probability of exhausting their benefits using a predicted probability threshold, which was chosen to balance the tradeoff between sensitivity and specificity. We report the predictive performance of our models when applied to the test set. We perform the analysis for claims with benefit duration of 26 weeks, first using the full sample, then sequentially eliminating claims that resolved within 2, 4, and 6 weeks. Comparing across durations illustrates the potential efficiency gains of waiting to allow some claims to resolve on their own.

The factors most strongly associated with exhaustion of STDI benefits are age, diagnosis, and employer industry. Waiting to allow some claims to resolve without intervention improves the efficiency of targeting efforts. Modeling based on observable factors helps further narrow the target population, with the machine learning techniques we use expected to outperform logistic regression in predictive performance.

Depending on the cost structure of the intervention, our approach could represent significant savings through efficient targeting of interventions to those STDI claimants who are most likely to benefit from them. We simulate the cost and

on private employer sponsored insurance (ESI) coverage, private non-ESI coverage, and un-insurance status among less-skilled (high school dropouts and graduates) workers. The employer mandate requires firms with 100 (50) or more full-time equivalent (FTE) workers to provide affordable health insurance from 2015 (2016). The mandate was intended to increase insurance coverage among 19-64 year olds who participate in the labor market on a full-time basis, cannot qualify for Medicaid in their state, and are traditionally not offered or eligible for insurance. In addition, it prevents crowd-out into the private

The literature notes that previous state and local level employer mandates increased private ESI coverage among previously uninsured workers. However, the ACA’s employer mandate may be ineffective since (1) small firms—traditionally the firms least likely to offer their workers with ESI coverage—are exempted from the mandate, (2) firms can avoid mandate by reducing workers’ hours so that they are not considered as FTE workers, (3) some firms can marginally reduce total FTE labor units and be exempted from the mandate, and (4) some low-income workers can qualify for Medicaid. In order to assess insurance coverage status among less-skilled workers—a group traditionally less likely to have private ESI coverage or afford private non-ESI coverage— based on firm size, I pool annual March supplements of the Current Population Survey from 2006 to 2016. The pooled cross-section data provides income, labor supply, and health insurance information from 2005 to 2015. Using a difference-in-difference model, I do not find evidence that that less-skilled workers were significantly more likely to report private ESI coverage if they worked in mid-sized and larger firms after the ACA was legislated in 2011, or after major components of the ACA was implemented in 2014. However, compared to less-skilled workers in small firms, workers were significantly less likely to report private non-ESI coverage if they worked in mid-sized by 0.03 percentage points (0.7%) and in large firms by 0.02 percentage points (0.6%) from 2013. In addition, while uninsured status wa not significantly different between workers in small firms and mid-sized firms after the legislation, workers in large firms were less likely to be uninsured by 0.04 percentage points (0.10%) between 2011 and 2013 when the ACA legislation was passed but the major components (individual mandate, Medicaid expansions, and employer mandate)

While the employer mandate may not have increased private-ESI among workers, workers are less likely to have private non-ESI coverage in mid-sized and large firms. Evidence suggests that mandated firms may have avoided penalties by either reducing units of labor along the intensive and extensive margin, and/or ensured low-income, less-skilled workers enrolled in Medicaid. Less-skilled workers in mandate firms. Based on preliminary results, the mandate was unsuccessful in increasing private ESI coverage, decreasing un-insurance, and preventing crowd-out into Medicaid.

: In October 2013, the Affordable Care Act (ACA) health insurance exchanges began offering access to subsidized community-rated plans previously unavailable to many non-elderly adults. This shock represents a decrease in the opportunity cost to employment for those with employer-sponsored insurance (ESI). Studies have examined labor participation, but not willingness to voluntarily quit as a result of expanded availability of guaranteed issue coverage. There is similarly little evidence on the reasons people quit or the activities that individuals pursue after quitting.

: To test the effect of the launch of ACA health insurance exchanges (late 2013) on individuals’ willingness to quit their current main job. : Repeated panels (2006-2015) from the Medical Expenditure Panel Survey (MEPS) contain information on current employment and quitting in five survey rounds over a two-year period. Linear probability models (LPM) with

standard errors clustered at the household level predict the probability of quitting, controlling for temporal and seasonal trends using panel/round fixed effects. Models include controls for demographic and socioeconomic characteristics and are weighted using the provided longitudinal weights. The main effect compares the first open enrollment period (Q4 2013-Q1-2014 [panel 17, round 5]) with the same seasonal period (round 5) in different panels (first difference), then controls for time trends by subtracting out the same comparison from a different round (e.g., panel 17, round 2) (second difference).

: The sample contains 105,348 individuals, 2,782 of whom quit their current main job over 10 overlapping panels (10 years) of 4 rounds each (round 1 established current job). The unadjusted rate of voluntary quitting in round 5 across all pre-exchange panels was 3.1%. During the first open enrollment (panel 17, round 5), the predicted probability of voluntary quitting was nearly a full percentage point higher than the average of the prior panels [0.93 percentage points (95% CI: 0.01, 1.80)] and consistent for individual years (first difference). Controlling for temporal changes (second difference), using the second and third rounds as referent, the average effects are even stronger [2.01 (0.93, 3.11)] and [1.54 (0.37, 2.70)], respectively. We also observe a stronger effect in earlier panel years (pre-2010), suggesting the period of and following the Great Recession may have dampened willingness to quit. The largest change in reasons for voluntarily quitting a position between the first open enrollment and pre-exchange periods was ‘quit to take care of home/family’ (10.6%) followed by ‘quit to go to school’ (7.58%).

: These results show support for the job lock hypothesis, indicating that individuals may be willing to quit their jobs when access to other sources of affordable health insurance become available. The magnitude of this result is small, but generalizable to a sizable population, unlike other previous attempts to characterize the effects of the ACA on labor outcomes. Specifically, we describe the pathway where individuals may be willing to quit their current job differentially around the time of the first ACA open enrollment period. Affordable access to health insurance can improve flexibility to move in and out of the labor force.


The goal of this study is to examine the causal effect of informal care on labor supply. Related studies in the United States and Europe analyzing the effect of informal care on labor supply have employed family structure and parental health as instrumental variables. They have not utilized institutional change as a natural experiment in estimating the effect of informal care on labor supply. As Van Houtven, Coe, and Skira (2013) point out, some of the instruments

In 2000, the Japanese government has also implemented LTCI. In the Japanese care system, there are two important characteristics related to our study. First, there are three types of public nursing homes. Second, the supply of these nursing homes is regulated by the government. Our analysis utilizes the exogenous variation of government intervention on the supply side of the elderly care market to estimate the causal effect of informal care on labor supply.

We use the instrumental variables method. To the best of our knowledge, no study has thus far utilized exogenous institutional variation as an instrument to estimate the causal effect of informal care on labor supply.

Analysis results reveal that the effect of informal elderly care on female labor force participation is negative. By contrast, male labor force participation is not affected by such care, since, in Japan, females spend more time on informal care than males. The increase in nursing home capacity is thus effective for decreasing the female burden of informal care.

The effect of informal care for elderly on labor supply in both males and females is small. Especially, when compared with literature, the effect is smaller than in extant studies. The time spent on informal care in households is the focus on female household members. The government intervention is effective for increasing female labor supply.

The Supplemental Nutrition Assistance Program (SNAP; formerly called the Food Stamp Program or FSP) provided $66.5 billion in nutrition assistance to 44.2 million participants in fiscal year 2016. Prior research has examined the effects of SNAP on many different outcomes, but despite the program’s economic importance and potential health impacts, there are relatively few studies investigating how receiving food stamps affects health outcomes. Even fewer studies have examined SNAP’s health impacts on adult recipients, and these studies' findings are somewhat mixed. Given recent research finding a significantly higher risk of death for food stamp recipients compared to eligible non-recipients, these health effects may be significant. However, no prior study has satisfactorily examined the causal effects of food stamps on adult mortality outcomes. This study examines the effects of food stamps on health, with a focus on adult health. Specifically, I use the county-level rollout of the FSP from 1961 to 1975 as a source of plausibly exogenous variation in access to food stamps. I examine the effects of contemporaneous and multiple-year access to the FSP on various county-year-level mortality rates using fixed effects models. I consider effects on aggregate mortality rates, subgroup rates for sex, race, and age groups, and rates for specific causes of death to examine the different mechanisms through which food stamps might affect health. I find mixed results for the entire sample that indicate small overall effects of access to food stamps on mortality rates. However, among subsamples of poorer counties that are likely to benefit the most from food stamps, I find that implementation of the FSP reduces mortality rates for most groups over time.

We estimate impacts of exposure to a preventive infant health intervention trialled in Sweden in the early 1930s using purposively digitised birth registers linked to school catalogues, census files and tax records to generate longitudinal microdata that track 25,000 individuals through four stages of the life-course, from birth to age 71. This allows us to measure impacts on childhood health and cognitive skills at ages 7 and 10, educational choice during young adulthood, employment, earnings and occupation at age 36-40, and pension income at age 71. Leveraging quasi-random variation in eligibility by birth date and birth parish, we estimate that an additional year of exposure was associated with substantial increases in earnings and (public sector) employment among women, alongside no improvements for men. We also identify intervention effects on primary school test scores for men and women, and on secondary school completion for women only. A large part of the income gain for women can be attributed to secondary schooling and test score improvements, in particular at the top of the distribution. Using recent innovations in mediation analysis, we are able to show that school performance and secondary schooling enrollment are important mechanisms behind the adult gains in earnings. The greater investments of women in education are consistent with their comparative advantage in cognitive tasks, but opportunities are also likely to have played a role. Our sample cohorts were exposed to a massive expansion of the Swedish welfare state, which created unprecedented employment opportunities for

Quality of care is linked to improved health outcomes, cost efficiency, and high-value healthcare delivery. Recent policies, such as Medicare's value-based purchasing programs, seek to improve quality of care by tying provider reimbursements to the quality of services provided. However, the impact of factors external to the healthcare system, such as economic fluctuations, on healthcare quality is largely unexplored and recent work in this area has focused on elderly populations in nursing homes (Antwi and Bowblis, 2017; Stevens et al., 2012). The extent to which the broader adult population experiences cyclical changes in healthcare quality remains unclear. Patient experience is considered a component of quality as it captures the quality of health inputs as perceived by the patient. A large patient experience literature, primarily using cross-sectional and hospital-level data, identifies correlates between patient experience measures and individual and provider attributes. Using patient experience ratings from a nationally representative survey, I merge these two literatures by exploring the relationship between local economic conditions and the quality of

In contrast to previous studies, I use individual-level, short panel data from the Medical Expenditure Panel Survey from 2002 to 2011 to explore local economic conditions as a potential determinant of improvements in patient ratings of healthcare quality. Specifically, I find evidence that increases in county unemployment rates during the Great Recession are associated with higher patient ratings of having enough time with their healthcare providers in the past year. The relationship is particularly strong for those with chronic conditions, suggesting changes in quality generated by economic fluctuations are larger for high-users of care. I find neither changes in individual's employment status nor insurance coverage mediate the positive correlation. I find some evidence that improvements in health during the Great Recession are a potential pathway. My findings imply that value-based purchasing programs may need to adjust quality targets to account for external factors, such as local economic conditions, when measures of patient satisfaction determine the incentive payment.

Immigrants are a vulnerable population due to their high risk for poor physical, psychological, and social health outcomes (Derose et al., 2007). Compared to similarly poor native-born citizens, low-income immigrants are more likely to lack health insurance. A sizeable segment of the immigrant population in the U.S. work at minimum wage jobs because they have lower educational attainment, limited language skills, and less social capital. Given the size and the rapid growth of immigrants in the U.S. workforce, it is important to examine the impact of minimum wage increases on immigrants’ health and access to care. Conventional economic theory predicts that increases in minimum wages raise hourly earnings and reduce employment. Orrenius and Zavodny (2008) find that hourly earnings for low-skilled adult immigrants increased with minimum

Increases in earnings resulting from higher minimum wages create an income effect (among those who keep their jobs) which could then improve health outcomes. The potential for better health has led some policymakers to call for higher minimum wages specifically to improve health (e.g. Bhatia, 2014). Empirically, there is a growing literature aimed at determining if minimum wage increases positively affect the health of those individuals who retain their jobs and have higher earnings (e.g., Averett et al. 2016, Kronenberg et al. 2015, Lenhart 2015, Reeves et al. 2014, 2016, Strain et al. 2016, and Wehby et al. 2016). This paper examines whether minimum wage increases affect the health of low-skilled immigrants. We do so by using data from the National Health Interview Survey and estimating the following equation:

is the minimum wage (or the ratio of the minimum wage to the state’s average wage); Z it is a vector of individual controls including age, marital status, is a vector of state-specific time-varying economic and policy controls that may be correlated with minimum wages and health including immigrant’s access to

health care, welfare and food benefits after the 1996 welfare reform act, the percent of the state’s workforce covered by a collective bargaining agreement, the percent of union membership, the state unemployment rate, state cigarette taxes, the percent of the state’s population below the poverty line, whether the state has an e-verify mandate and whether the state allows for public health insurance for unauthorized children and adults. θ s is the time-invariant state

is an error term. Standard errors will be clustered by state. Preliminary results using the Current Population Survey which has a much more limited set of health outcomes shows that there is likely to be an income effect and an effect of minimum wages on self-rated health for working immigrants.


Neil Meredith [email protected]

Sheryll Namingit [email protected]

Wei Luo [email protected]

R. Vincent Pohl [email protected]

Gulcin Gumus [email protected]

Meg Blume-Kohout [email protected]

Using data on U.S. adults from the National Health Interview Survey, we estimate the income elasticity of demand for tanning bed usage among users and evaluate the factors that influence non-use of tanning beds. While controlling for individual characteristics, we show that the income elasticity of demand indicates that tanning bed usage is a normal good that is viewed as a necessity for users. For non-users, estimates show, after controlling for individual characteristics, that income is negatively associated with the probability of becoming a tanning bed user. Results suggest that usage of tanning beds grows for users as income increases and makes tanning usage more affordable. The

Using a résumé-based correspondence test, we compare the employment consequences of an illness-related employment gap to those of an unexplained employment gap. Previous research shows that employment gaps, in general, have adverse effects on the probability of getting hired. It is not clear, however, if employers view spells of joblessness due to health issues distinctly. To shed light on this, we present a theoretical model in which employers use information on employment gaps as a signal of unobserved productivity and healthcare costs. We investigate the empirical implications of the model by sending three types of fictitious résumés to real job vacancies. One résumé indicates that the applicant is newly unemployed. The other résumés indicate employment gaps which are either unexplained or explained as being related to an illness. To signal an illness-related employment gap, a phrase in the cover letter explained that the employment gap was due to a physical illness followed by a full recovery. An additional signal on medical history was sent via information in the résumé that indicates involvement in a cancer recovery support group. The corresponding cover letters of the résumés with unexplained gaps did not provide any explanation for the gap. For the résumé of newly unemployed applicants, the length of the gap is limited to less than two months. Based on the literature, this is too short a gap to bring about adverse effects. The corresponding cover letter of newly unemployed applicants notes that the applicant left the last job because her family had to move from another state and that she is currently looking for a new job. From March to September, 2016, we sent 3,771 résumés to 1,257 sales, administrative, and accounting assistant jobs. Outcomes are measured in terms of differences in the callback rate of each type of résumé. The results of the experiment show that newly unemployed applicants had the highest callback rate (27.4%). Consistent with previous studies, résumés with an employment gap received lower callback rates, indicating that such gaps negatively affect hiring outcomes. However, résumés with an explained illness-related gap received a higher callback rate than résumés with an unexplained gap (25.6% versus 23.3%). Within the context of our theoretical model, these results suggest that the

Extensive literature in health economics examine the persisting effect of early life adverse shocks on adult outcomes, and sharp exogenous shocks in fetal health are exploited to provide compelling evidence on the fetal origins hypothesis. This paper is motivated by the idea to synthesize the fetal origins literature and the evolution of children’s cognitive and non-cognitive abilities to provide a life-cycle framework to understand the origins of health inequality. In this framework, adult outcomes to the development of cognitive abilities, meanwhile cognitive abilities are jointly determined by environment, investment and initial genes. As argued by previous fetal origins literature, early life deprivation has long-standing and negative shocks on adult health or educational outcomes. We make the argument that even though the effect of early life (in utero) deprivation is disastrous, remediation and resilience induced by later investment

This paper provides novel evidence from 1958-1961 China great famine on long-standing effect of early life deprivation on adult outcomes and how gaps induced by early life deprivation are narrowed by later life compensating investment. Exploiting unique datasets obtained from China Health and Retirement Longitudinal Survey (CHARLS), 2011 national baseline survey and 2014 life history survey, this paper intends to examine the long-term impacts of fetal malnutrition based on survivors in their 50s who were born during the 1959-1961 China’s great Famine. In addition, this study is interested to incorporate early life adverse shocks, family compensating investment into a life cycle

Our results support the Fetal Origins Hypothesis (Barker, 1992) that exposure to adverse conditions in early life (such as exposure to infectious diseases or malnutrition status) may causally affect health and mortality at old ages. We find that fetal exposure to malnutrition has large and long- lasting impacts on cognitive abilities, including immediate word recalling ability, delayed word recalling ability and having difficulty with drawing a picture. In addition, our results shows the effect of early life adverse shocks can be narrowed by family compensating investment, thus as guardian’s love and affection, their time and effort devoted to taking care of children, which is consistent with the framework of

Innovations in cancer treatment have lowered mortality, but little is know about their economic benefits and who benefits from them. In this paper, we assess the effect of improved treatment options over the last three decades on the labor market outcomes of breast and prostate cancer patients. We combine administrative tax return and cancer registry data from Canada with measures of medical innovation (approved drugs, academic publications, and patents) to estimate triple-differences regressions of employment and annual earnings. Our results show that the reductions in these labor market outcomes among cancer patients are partially offset by improved treatment options. Specifically, the decline in employment due to a cancer diagnosis is cut by 50 to 75 percent by the medical innovation that has occurred during the last three decades. We also find that the benefits of medical innovation are limited to cancer patients with

Motorcyclists account for a much higher proportion of traffic fatalities relative to the share of motorcycles among all vehicles and vehicle miles driven in the U.S. In this paper, we examine whether state-specific texting/handheld bans significantly influence motorcyclist fatalities. We use longitudinal multivariate analysis of state-specific traffic fatality data in the U.S. (2005-2015) from the Fatality Analysis Reporting System (FARS) merged with state-specific characteristics, texting/handheld device laws, and other traffic policies. We find that states with moderate and strong texting/handheld bans have significantly lower motorcyclist fatality rates even after controlling for numerous other factors and state fixed-effects. This result is driven mainly by multiple-vehicle motorcycle crashes as opposed to single-vehicle crashes. Although research is mixed on the effectiveness of texting/handheld device policies for overall traffic fatalities, our

The Affordable Care Act (ACA) of 2010 improved and expanded availability of non-group health insurance. Previous studies have shown that women in the U.S. workforce value health insurance more than men do. Because prior to the ACA self-employed individuals did not have guaranteed access to affordable health insurance, womens lower rate of self-employment may partly have reflected job lock due to reliance on employer-based group coverage. This paper employs nationally-representative survey data for 2012-2016 and a difference-in-difference modeling approach to demonstrate that in fact, unmarried women have had significantly higher probability of self-employment since the ACA health insurance exchanges opened in 2014, coincident with their relatively higher uptake of private non-group health insurance purchased on state exchanges. This evidence demonstrates additional economic benefits of the ACA legislation,

Lizhong Peng [email protected]

Laia Maynou [email protected]

Aziza Arifkhanova [email protected]

Michael T. French [email protected]

Andres Hincapie [email protected]

The recent Medicaid expansions under the Affordable Care Act (ACA) have substantially increased access to health insurance among low-income individuals. Despite potential cost savings from better access to preventive care by Medicaid recipients, one common argument against the expansion of public insurance is that it may reduce the incentive for eligible individuals to remain in the workforce. Therefore, a correct understanding of the labor market impacts of

We seek to provide new empirical evidence on the effects of the recent Medicaid expansions on labor market outcomes. Unlike most existing studies that conduct the analysis at the state-level (e.g. Leung and Mas, 2016; Kaestner et al, 2015; Frisvold and Jung, 2017), our main analysis is at the county-level, which is a more appropriate approximation for the boundaries of local labor markets. Our identification strategy is based on the comparison of employment and wages in contiguous county-pairs in neighboring states (i.e. border counties) with different Medicaid expansion status. This is similar to the method employed by Dube et al (2010) to study the employment effects of minimum wage laws. Compared to a standard difference-in-differences approach, restricting the analysis to border-county pairs greatly improves the comparability between treatment and control units and controls for spatial heterogeneity, which can confound the relationship between Medicaid expansion status and employment or wages. In particular, our preferred specification uses only within county-pair variation in Medicaid expansion status and allows for arbitrary time effects

The main data source for our study is the 2008-2016 Quarterly Census of Employment and Wages (QCEW) collected by the Bureau of Labor Statistics (BLS). The QCEW is a comprehensive census of all establishments that report to the Unemployment Insurance programs, which contains about 97% of civilian employment nationwide. In contrast to other popular datasets such as the CPS or ACS, the QCEW allows the measurement of changes in employment and wages

We estimate a set of distributed lag models which allow us to examine the dynamic effects of the Medicaid expansions. All of our models include controls for effective minimum wage, county population, state poverty rate, and state median household income. Consistent with previous studies, we do not find any statistically significant effects on either employment or wages when estimating a conventional county and year fixed effects on the full sample. However, in our border county-pair sample we find a small but statistically significant decrease in employment of between 1.2-1.5 percent one year after the implementation of the Medicaid expansions, but no statistically significant change in wages. Overall, our results suggest that the ACA Medicaid expansions had a modest effect on labor supply at the extensive margin. These findings contribute to the growing literature on the impact of public insurance programs on labor market

This paper contributes to the existing literature on the diffusion of medical technologies. We apply panel data techniques to determine the manner in which technology is diffused across the NHS, with a particular emphasis on the impact that technology has on the workforce composition. We first examine the substitution or complementarity effects across different types of new technologies introduced into the NHS. Drawing on the work by Cutler and Huckman, we consider the diffusion of PTCA as it replaces CABG in the treatment of cardiovascular disease in England. We then estimate the degree to which the workforce reacts to the introduction of new technology, through calculating elasticity of supply measures. The data is combined from different sources to analyse these relationships: mainly, the UK Hospital Episodes Statistics (HES) and the NHS Electronic Staff Records (ESR). Analysis is at the provider level and the empirical specification explores the relationship between volume and workforce also controlling for provider and at risk population characteristics. Given the lack of quantitative evidence on the degree of substitution or complementarity across


A recent study reports that having a tattoo does not diminish one’s likelihood of employment conditional on labor force participation or earnings conditional on being employed. Although novel in its design, scope, and contribution, the findings are somewhat limited because neither of the two datasets used by the authors contain information beyond the reporting of having one or more tattoos. To address this important shortcoming in the literature, the present study collected detailed data on number, coverage, and characteristics of tattoos (as well as employment status, labor supply, earnings, human capital measures, and other personal characteristics) among a sample of approximately 2,000 adults. We then estimate whether tattoos are significantly related to labor market outcomes using a much broader range of tattoo features (any, number, visible, offensive). Results show that few of the tattoo measures are significantly related to employment, labor supply, or earnings for either gender. These results have important and timely implications for those entering, or already participating in, the labor market that may have, or are considering, a tattoo(s). The implications


Alicia Atwood [email protected]

Scott Grosse [email protected]

Yingxi Zhang [email protected]

Barbara Wolfe [email protected]

Human capital-based lifetime personal productivity estimates have long been used to measure the expected burden of productivity losses associated with diseases, injuries, or risk factors that lead to disability or premature death and the expected economic benefits from prevention. The most recent published estimates, which are now a decade old, followed convention in assuming that labor productivity would grow indefinitely at 1% per year in real (inflation-adjusted) terms. We present updated estimates of annual productivity for the period 2006-2016 using data from the American Community Survey, the American Time Use Survey, and the Current Population Survey. Productivity is the sum of market productivity calculated as gross annual personal labor earnings adjusted for employer-paid benefits and the imputed value of the non-market time spent producing household, caring, and volunteer services. Hours spent in non-market services were valued using the replacement cost method of the hourly market cost of hiring equivalent services to be performed. The present value of lifetime productivity at various ages was calculated for synthetic cohorts using annual productivity estimates, US life tables, discount rates, and assumptions about future labor productivity growth rates. Mean annual productivity was $57,324 for US adults in 2016, including $36,935 in market and $20,389 in non-marked productivity. Productivity in 2016 remained below pre-crisis (2006-2008) levels after adjustment for inflation, which implies negative real growth in labor productivity in the resident adult population during the study period. The present value of lifetime productivity at birth in 2016 calculated using a 3% real discount rate ranged from $1,193,498, assuming a 0.5% annual real growth in labor productivity in future years, to $1,468,669, assuming 1% annual productivity growth. These findings demonstrate that decisions as to whether to estimate total or just market productivity and what assumption to make about growth in future productivity are influential in estimates of avoidable economic

Relationships between health and economic growth are difficult to assess. Health is multidimensional and measured with errors. It is argued that commonly used health indicators in macroeconomic studies (life expectancy, infant mortality or specific diseases such as malaria or HIV/AIDS) imperfectly represent the global health status of population. The health indicators used in the previous literatures capture only one dimension of the population health. Actually, health is

The main thesis of this paper is that macroeconomic effects of the global health status are accurately caught by the Disability-Adjusted Life Year (DALYs) calculated by WHO. DALYs represents the burden of disease and can be thought of as a measurement of the gap between current health status and an ideal health situation. DALYs are commonly used in cost-effectiveness analyses but rarely used in macro economy. With the improvement of data, its impact to economic growth is starting to be noticed. According to the statistics of GBD 2015, between 1990 to 2015, global DALYs rate for all causes was projected to decrease from 48297 to 33440 per 100000, an overall decline of about 31%. For the NCDs_DALYs (Non-Communicable Diseases), although it declined from 20606 to 19990 per 100000, the decline proportion is only 3% and its proportion to all causes increased from 43% to 60%. That means the global disease burden is

The research is focused on the effects of NCDs_DALYs on the growth rate of per capita GDP in global countries. This paper use an expanded Solow growth model and a dynamic panel GMM estimator to testify whether DALYs contribute to economic growth. At the same time, physical capital and another important human capital—education are also included in the model. Considering the epidemiological transition and the income gap in different countries and regions, the whole sample of global countries were divided into four sub-samples according to the income level: high income countries, high and upper middle income countries, middle and low income countries, lower middle and low income

The empirical study has four important findings: First, NCDs_DALYs has a lagged and negative effect on GDP both in global countries and different income level countries. That means the declining of NCDs_DALYs contributes to the growth rate of per capita GDP. Second, the effect of NCDs_DALYs on economic growth is much more significant in developing countries than in developed countries. The correlation coefficients for lower middle and low income countries, middle and low income countries, and high income countries are separately -1.08***, -0.71*** and -0.04**.Third, education has a positive effect on GDP in high income countries and negative effect in lower middle and low income countries.

According to existing literature, early-life health affects later labor market outcomes such as earnings and work effort. We examine whether this holds for multiple dimensions of health and regardless of a country’s health-care system. We ask whether mental and physical health problems and poor general health by age 15 have similar or different influences on lifetime earnings, on years of schooling and on health problems later in life. Then we ask whether the health-care system the child lived in influenced the estimated effects of early health problems on lifetime earnings. We expect that early health problems reduce earnings and the most generous system is tied to the least negative long-term effects. Our analysis uses individual-level data from the first three waves of SHARE, a multidisciplinary and representative cross-national panel of the European population aged 50-plus. Waves 1 (2004/05) and 2 (2006/07) include information on sociodemographic background characteristics, current health, and socioeconomic status, as well as expectations of retirement age. Most of the data we use are from the third wave, SHARELIFE (2008/09), which is a retrospective survey

We use our respondent’s country of childhood and a four-way system to characterize the health-care systems they lived in as children based on descriptions in the U.S. Social Security Administration’ Office of Policy (2002). These four groupings are: full coverage; considerable use of co-payments; limited coverage; and Socialist (full coverage but limited care). We find that the health-care system does make a difference in the size of the earnings penalty.

Annie Hines [email protected]

Emily Gallagher [email protected]

Amanda Raffenaud [email protected]

This paper examines the impact of increases in immigration enforcement on health care access and utilization of immigrants. While previous studies have examined the chilling effects of immigration enforcement on the take-up of public health insurance (Watson 2014), as well as the direct impact on self-reported health (Venkataramani et al. 2017), few studies have directly examined how deportations affect immigrant access to and use of health care. Harsher immigration enforcement and increased deportations may affect health of immigrants who remain in the country for multiple reasons. First, stress associated with the fear of being deported can have a negative effect on both mental and physical health outcomes, including anxiety, depression, cardiovascular disease, and high blood pressure. Second, increased deportations may deter undocumented immigrants, as well as their families, from using health care available to them for fear of interacting with authorities. Finally, chilling effects (as documented by Watson 2014) may discourage immigrants from obtaining health care coverage, even when they or their children are eligible, thus leading to lower health care use and worse health outcomes. Taking advantage of variation across counties in ICE apprehensions and deportations, combined with hospital inpatient discharge records, I will focus on the last two channels and examine the effect of deportations on emergency room (ER) admissions and Prevention Quality Indicators (PQIs). Both of these variables are measures of access to outpatient care that can be studied with inpatient data. Many ER admissions result from a lack of access to other channels of health care, such as preventive care or primary care, whiles PQIs reflect conditions for which hospitalizations can be easily avoided with regular outpatient care. Data on immigration enforcement comes from individual records of ICE apprehensions and removals across US counties under the Secure Communities Program for fiscal years 2007-2017. These data will be combined with information on policies that reflect the degree of cooperation between local law enforcement and federal immigration authorities, including 287(g) agreements at the county-level, state-level omnibus immigration enforcement laws, and “sanctuary city” designations across the US. Hospital discharge data comes from the Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases (SID). These databases contain administrative records on inpatient discharges from community hospitals for an unbalanced panel of states across the US from 1996-2011. Given both the current debates about immigration policy and general concern about issues of access to health care in the US, this study will provide valuable evidence on the effects of

We evaluate the effect of health insurance on negative earnings shocks using the administrative tax data and survey responses of 4,975 low-income households. We exploit exogenous variation in the cost of private insurance under the Affordable Care Act using a regression discontinuity design. The probability of income loss falls by 22% at the income threshold for receiving insurance subsidies. An otherwise uninsured household that gets subsidized coverage is 25 percentage points less likely to report a job loss. Effects are concentrated among households with past health costs and exist only for “unexpected” forms of earnings variation, suggesting a health-productivity link. Rudimentary calculations

: The nursing workforce increasingly faces issues that affect clinical and managerial practice. One such issue is work-family conflict (WFC) and family-work conflict (FWC). Nurses face role strain as they confront the pressures from often competing work-and-family roles. While several studies have explored this issue among staff nurses, none to our knowledge have studied nurse managers. This study assesses WFC/FWC among staff versus executive

: This is an exploratory, cross-sectional survey. Survey questions included demographics, practice settings and roles, perceptions regarding the work environment, and perceptions of WFC/FWC. The survey instrument was validated in a number of prior studies. Descriptive statistics were conducted. Two separate ANOVAs were run to test the between and within groups scores for staff, managerial and executive nurses on WFC and FWC respectively. Two separate OLS regressions were run on models in which the dependent variables were WFC and FWC scales respectively and the independent variables were demographic, professional and work environment measures, focusing on the

: We randomly sampled registered nurses across the state of Florida. Of the nearly 5,000 email surveys, over 400 were completed. Nurses of varying roles and practice settings participated,. : Descriptively, nurses experienced more work-family conflict than family-work conflict. Regression analyses and ANOVAs indicated that staff nurses experienced less work-family conflict than nursing managers (second

most) and nursing executives (highest). None of the nurse roles experienced significant levels of FWC. White nurses, compared to non-white nurses, experienced less WFC and FWC. WFC increased with shift length but FWC was not significantly affected by it. Paid leave for childbirth was associated with lower FWC. Age, gender, marital status and number of children in the home were demographics not significantly related to WFC/FWC. Practice setting, length of employment in current job, professional tenure, and educational level were professional variables not significant in the model. Managerial issues, staffing, nurse/physician collaboration and nursing competence were work environment

This study holds significant implications for practice. Nurse managers and executives showed significantly higher WRC than staff nurses. This may discourage a nurse from taking on

Nonwhite nurses reported higher levels of both WFC and FWC. This may contribute to tension at the workplace and a difficult family life. Recruitment of people of color into nursing and their retention may be adversely affected. Leaders

Nurses working shifts over 8 hours had higher WFC levels. Although 12-hour shifts have been popular among staff and management, their use should be reevaluated. Finally, paid leave for childbirth is a program worth supporting as it was

Farhan Majid [email protected]

James Bailey [email protected]

Kara Contreary [email protected]

Early life environmental conditions play a critical role in well-being over the life-course. Most studies of these roles have treated extreme shocks, such as famines, as natural experiments to study the causal impact of early life conditions. However, little attention is paid to the potential of labor market institutions, specifically income protection legislation, in shaping opportunities for investing in early life health and the subsequent impact on early life health outcomes. We study minimum wages around the time of birth and their effects on child stunting in Indonesia up to 5 years after birth. Indonesia is interesting not only because it carries the fifth highest burden of stunted children in the world, but

To the extent that minimum wages influence parental wage income, they might affect parental investments in child health. For example, if wages were to increase, families might be more likely to avail themselves of health services and engage in other salutary behaviors which may be particularly effective around the time of birth. However, mothers may also be more likely to spend more time in the labor market at the expense of care-giving activities. The time around birth is widely understood to be a critical period in shaping child nutrition and stunting levels, so that changes in parental economic conditions around the time of births may have particularly large effects on child health and nutrition. Using variation in annual fluctuations in real minimum wages in different provinces of Indonesia, we find that children exposed to increases in minimum wages in the year of birth have higher height-for-age (HAZ) scores in the first five years of their lives. Furthermore, we use data on parental wages to focus on children of parents for whom the minimum wage is most likely to be binding- those whose parents are in the bottom 25th and 50th quantiles of the wage distribution. Parents in upper tails of the wage distribution are considered as part of a placebo sample. Our estimated impacts are evident with difference-in-difference models with province and year-of-birth fixed effects and are robust to inclusion of biological sibling fixed effects, measures of child characteristics (age, gender) and parental characteristics (such as employment status, age and educational attainment, household income and assets) as well as community covariates (provincial GDP and unemployment rates). The effects are prominent particularly among children whose fathers earn in the bottom of the wage distribution, where as no effects are found for fathers earnings in the top part of the wage distribution (placebo). We also use multiple sources of consumer prices indices (CPI) (provincial and national) to explore robustness to different measures of real minimum wages. Our results are consistent with recent work from Indonesia based on “big push” models where increases in minimum wages lead to a movement away from an equilibrium of low wages and low labor demand to an equilibrium with high demand and high wages.

We aim to determine the effect of the main provisions of the Affordable Care Act on entrepreneurship. Previous work has found substantial evidence that some Americans are deterred from self-employment because they are concerned about losing their employer-based health insurance. The Affordable Care Act (ACA) introduced several reforms aimed at improving the market for individual insurance, with one objective of reducing the problem of "entrepreneurship lock." We will assess whether, and the extent to which the ACA has made progress towards this objective, across various relevant margins and sub-populations, based on a quasi-experimental research design applied to data from the

Because many of the main ACA provisions took effect in January 2014, it is only now becoming possible to study their effects; previous work on the ACA and entrepreneurship could only investigate the effect of relatively minor early provisions such as the dependent coverage mandate (Bailey 2013). As potential entrepreneurs are deterred because of concerns over their ability to maintain health insurance when they start a company, it is important to understand whether and to what extent the ACA solves this market failure for various groups, as well as the extent to which further reforms would be needed to keep the employer-based health insurance system from distorting the labor market and slowing the formation of new businesses. Market frictions that impede entrepreneurship may have adverse implications for efficient matching between worker skills and work and lead to labor market inefficiencies. Our main analysis focuses on evaluating the net effect of the main ACA provisions as a whole. However, if policymakers aim to expand, scale back, or replace the ACA it will be important to understand potential effects of individual components. Because many of the main provisions, including guaranteed issue, community rating, and the subsidized exchanges began simultaneously on January 1, 2014, it is empirically difficult to separate their effects. Nevertheless, first-step evidence on the partial effects of some of these individual components can be gleaned by exploiting previous state laws mandating guaranteed issue and community rating and by further exploiting individuals from

Short-term disability insurance (STDI) pays partial wage replacements to employees temporarily unable to work due to “off-the-job” medical conditions. Most STDI policies replace wages for a fixed period, such as six months. Because wages are replaced only partially, STDI claimants have an incentive to return to work. Those who are unable to return before benefits expire may be at higher risk of job loss and receipt of long-term disability insurance (LTDI) or Social Security Disability Insurance (SSDI) benefits. An STDI claim can be an early identification point of workers with medical conditions who could, with adequate support, remain in the workforce. However, little is known about the factors influencing STDI duration or the transition to LTDI or SSDI benefits. Furthermore, careful timing and targeting of interventions is critical to efficiency; some workers may return to work without intervention, while others may not benefit from it. In this paper, we: (1) compare the performance of alternative models using information in claims data to predict exhaustion of STDI benefits; and (2) assess if waiting for some claims to resolve without intervention can improve the

Integrated Benefits Institute (IBI) Health and Productivity Benchmarking Data from 2011 through 2015, including 820,751 closed STDI claims from 8,587 small, medium, and large businesses associated with 9 disability insurance carriers and third-party leave administrators. The data include claim outcomes and claimant, employer, and insurance plan design characteristics. The primary outcome of interest is exhaustion of the STDI benefit.

We fit several predictive models to the data, including logistic regression, regularized logistic regression (using an elastic net), and random forests. We randomly divide our sample into training and testing sets, and select a model based on the area under the receiver operating characteristic curve. Individuals are flagged as having a high probability of exhausting their benefits using a predicted probability threshold, which was chosen to balance the tradeoff between sensitivity and specificity. We report the predictive performance of our models when applied to the test set. We perform the analysis for claims with benefit duration of 26 weeks, first using the full sample, then sequentially eliminating

The factors most strongly associated with exhaustion of STDI benefits are age, diagnosis, and employer industry. Waiting to allow some claims to resolve without intervention improves the efficiency of targeting efforts. Modeling based on

Depending on the cost structure of the intervention, our approach could represent significant savings through efficient targeting of interventions to those STDI claimants who are most likely to benefit from them. We simulate the cost and

Angshuman Gooptu [email protected]

Paul Shafer [email protected]

Dan Zeltzer [email protected]

Troy Quast [email protected]

on private employer sponsored insurance (ESI) coverage, private non-ESI coverage, and un-insurance status among less-skilled (high school dropouts and graduates) workers. The employer mandate requires firms with 100 (50) or more full-time equivalent (FTE) workers to provide affordable health insurance from 2015 (2016). The mandate was intended to increase insurance coverage among 19-64 year olds who participate in the labor market on a full-time basis, cannot qualify for Medicaid in their state, and are traditionally not offered or eligible for insurance. In addition, it prevents crowd-out into the private

The literature notes that previous state and local level employer mandates increased private ESI coverage among previously uninsured workers. However, the ACA’s employer mandate may be ineffective since (1) small firms—traditionally the firms least likely to offer their workers with ESI coverage—are exempted from the mandate, (2) firms can avoid mandate by reducing workers’ hours so that they are not considered as FTE workers, (3) some firms can marginally reduce

In order to assess insurance coverage status among less-skilled workers—a group traditionally less likely to have private ESI coverage or afford private non-ESI coverage— based on firm size, I pool annual March supplements of the Current

Using a difference-in-difference model, I do not find evidence that that less-skilled workers were significantly more likely to report private ESI coverage if they worked in mid-sized and larger firms after the ACA was legislated in 2011, or after major components of the ACA was implemented in 2014. However, compared to less-skilled workers in small firms, workers were significantly less likely to report private non-ESI coverage if they worked in mid-sized by 0.03 percentage points (0.7%) and in large firms by 0.02 percentage points (0.6%) from 2013. In addition, while uninsured status wa not significantly different between workers in small firms and mid-sized firms after the legislation, workers in large firms were less likely to be uninsured by 0.04 percentage points (0.10%) between 2011 and 2013 when the ACA legislation was passed but the major components (individual mandate, Medicaid expansions, and employer mandate)

While the employer mandate may not have increased private-ESI among workers, workers are less likely to have private non-ESI coverage in mid-sized and large firms. Evidence suggests that mandated firms may have avoided penalties by either reducing units of labor along the intensive and extensive margin, and/or ensured low-income, less-skilled workers enrolled in Medicaid. Less-skilled workers in mandate firms. Based on preliminary results, the mandate was

: In October 2013, the Affordable Care Act (ACA) health insurance exchanges began offering access to subsidized community-rated plans previously unavailable to many non-elderly adults. This shock represents a decrease in the opportunity cost to employment for those with employer-sponsored insurance (ESI). Studies have examined labor participation, but not willingness to voluntarily quit as a result of expanded availability of guaranteed issue coverage.

: Repeated panels (2006-2015) from the Medical Expenditure Panel Survey (MEPS) contain information on current employment and quitting in five survey rounds over a two-year period. Linear probability models (LPM) with standard errors clustered at the household level predict the probability of quitting, controlling for temporal and seasonal trends using panel/round fixed effects. Models include controls for demographic and socioeconomic characteristics and are weighted using the provided longitudinal weights. The main effect compares the first open enrollment period (Q4 2013-Q1-2014 [panel 17, round 5]) with the same seasonal period (round 5) in different panels (first difference),

: The sample contains 105,348 individuals, 2,782 of whom quit their current main job over 10 overlapping panels (10 years) of 4 rounds each (round 1 established current job). The unadjusted rate of voluntary quitting in round 5 across all pre-exchange panels was 3.1%. During the first open enrollment (panel 17, round 5), the predicted probability of voluntary quitting was nearly a full percentage point higher than the average of the prior panels [0.93 percentage points (95% CI: 0.01, 1.80)] and consistent for individual years (first difference). Controlling for temporal changes (second difference), using the second and third rounds as referent, the average effects are even stronger [2.01 (0.93, 3.11)] and [1.54 (0.37, 2.70)], respectively. We also observe a stronger effect in earlier panel years (pre-2010), suggesting the period of and following the Great Recession may have dampened willingness to quit. The largest change in reasons for

: These results show support for the job lock hypothesis, indicating that individuals may be willing to quit their jobs when access to other sources of affordable health insurance become available. The magnitude of this result is small, but generalizable to a sizable population, unlike other previous attempts to characterize the effects of the ACA on labor outcomes. Specifically, we describe the pathway where individuals may be willing to quit their current job


Yoshinori Nishimura [email protected]

Jordan Jones [email protected]

Martin Karlsson [email protected]

Kimberly Groover [email protected]

The goal of this study is to examine the causal effect of informal care on labor supply. Related studies in the United States and Europe analyzing the effect of informal care on labor supply have employed family structure and parental health as instrumental variables. They have not utilized institutional change as a natural experiment in estimating the effect of informal care on labor supply. As Van Houtven, Coe, and Skira (2013) point out, some of the instruments

In 2000, the Japanese government has also implemented LTCI. In the Japanese care system, there are two important characteristics related to our study. First, there are three types of public nursing homes. Second, the supply of these nursing homes is regulated by the government. Our analysis utilizes the exogenous variation of government intervention on the supply side of the elderly care market to estimate the causal effect of informal care on labor supply.

We use the instrumental variables method. To the best of our knowledge, no study has thus far utilized exogenous institutional variation as an instrument to estimate the causal effect of informal care on labor supply.

Analysis results reveal that the effect of informal elderly care on female labor force participation is negative. By contrast, male labor force participation is not affected by such care, since, in Japan, females spend more time on informal care

The effect of informal care for elderly on labor supply in both males and females is small. Especially, when compared with literature, the effect is smaller than in extant studies. The time spent on informal care in households is the focus on

The Supplemental Nutrition Assistance Program (SNAP; formerly called the Food Stamp Program or FSP) provided $66.5 billion in nutrition assistance to 44.2 million participants in fiscal year 2016. Prior research has examined the effects of SNAP on many different outcomes, but despite the program’s economic importance and potential health impacts, there are relatively few studies investigating how receiving food stamps affects health outcomes. Even fewer studies have examined SNAP’s health impacts on adult recipients, and these studies' findings are somewhat mixed. Given recent research finding a significantly higher risk of death for food stamp recipients compared to eligible non-recipients, these

This study examines the effects of food stamps on health, with a focus on adult health. Specifically, I use the county-level rollout of the FSP from 1961 to 1975 as a source of plausibly exogenous variation in access to food stamps. I examine the effects of contemporaneous and multiple-year access to the FSP on various county-year-level mortality rates using fixed effects models. I consider effects on aggregate mortality rates, subgroup rates for sex, race, and age groups, and rates for specific causes of death to examine the different mechanisms through which food stamps might affect health. I find mixed results for the entire sample that indicate small overall effects of access to food stamps on mortality rates.

We estimate impacts of exposure to a preventive infant health intervention trialled in Sweden in the early 1930s using purposively digitised birth registers linked to school catalogues, census files and tax records to generate longitudinal microdata that track 25,000 individuals through four stages of the life-course, from birth to age 71. This allows us to measure impacts on childhood health and cognitive skills at ages 7 and 10, educational choice during young adulthood, employment, earnings and occupation at age 36-40, and pension income at age 71. Leveraging quasi-random variation in eligibility by birth date and birth parish, we estimate that an additional year of exposure was associated with substantial increases in earnings and (public sector) employment among women, alongside no improvements for men. We also identify intervention effects on primary school test scores for men and women, and on secondary school completion for women only. A large part of the income gain for women can be attributed to secondary schooling and test score improvements, in particular at the top of the distribution. Using recent innovations in mediation analysis, we are able to show that school performance and secondary schooling enrollment are important mechanisms behind the adult gains in earnings. The greater investments of women in education are consistent with their comparative advantage in cognitive tasks, but opportunities are also likely to have played a role. Our sample cohorts were exposed to a massive expansion of the Swedish welfare state, which created unprecedented employment opportunities for

Quality of care is linked to improved health outcomes, cost efficiency, and high-value healthcare delivery. Recent policies, such as Medicare's value-based purchasing programs, seek to improve quality of care by tying provider reimbursements to the quality of services provided. However, the impact of factors external to the healthcare system, such as economic fluctuations, on healthcare quality is largely unexplored and recent work in this area has focused on elderly populations in nursing homes (Antwi and Bowblis, 2017; Stevens et al., 2012). The extent to which the broader adult population experiences cyclical changes in healthcare quality remains unclear. Patient experience is considered a component of quality as it captures the quality of health inputs as perceived by the patient. A large patient experience literature, primarily using cross-sectional and hospital-level data, identifies correlates between patient experience measures and individual and provider attributes. Using patient experience ratings from a nationally representative survey, I merge these two literatures by exploring the relationship between local economic conditions and the quality of

In contrast to previous studies, I use individual-level, short panel data from the Medical Expenditure Panel Survey from 2002 to 2011 to explore local economic conditions as a potential determinant of improvements in patient ratings of healthcare quality. Specifically, I find evidence that increases in county unemployment rates during the Great Recession are associated with higher patient ratings of having enough time with their healthcare providers in the past year. The relationship is particularly strong for those with chronic conditions, suggesting changes in quality generated by economic fluctuations are larger for high-users of care. I find neither changes in individual's employment status nor insurance coverage mediate the positive correlation. I find some evidence that improvements in health during the Great Recession are a potential pathway. My findings imply that value-based purchasing programs may need to adjust quality targets

Susan Averett [email protected]

Immigrants are a vulnerable population due to their high risk for poor physical, psychological, and social health outcomes (Derose et al., 2007). Compared to similarly poor native-born citizens, low-income immigrants are more likely to lack health insurance. A sizeable segment of the immigrant population in the U.S. work at minimum wage jobs because they have lower educational attainment, limited language skills, and less social capital. Given the size and the rapid growth

Conventional economic theory predicts that increases in minimum wages raise hourly earnings and reduce employment. Orrenius and Zavodny (2008) find that hourly earnings for low-skilled adult immigrants increased with minimum

Increases in earnings resulting from higher minimum wages create an income effect (among those who keep their jobs) which could then improve health outcomes. The potential for better health has led some policymakers to call for higher minimum wages specifically to improve health (e.g. Bhatia, 2014). Empirically, there is a growing literature aimed at determining if minimum wage increases positively affect the health of those individuals who retain their jobs and

This paper examines whether minimum wage increases affect the health of low-skilled immigrants. We do so by using data from the National Health Interview Survey and estimating the following equation:

is a vector of individual controls including age, marital status, is a vector of state-specific time-varying economic and policy controls that may be correlated with minimum wages and health including immigrant’s access to

health care, welfare and food benefits after the 1996 welfare reform act, the percent of the state’s workforce covered by a collective bargaining agreement, the percent of union membership, the state unemployment rate, state cigarette taxes, the percent of the state’s population below the poverty line, whether the state has an e-verify mandate and whether the state allows for public health insurance for unauthorized children and adults. θ s is the time-invariant state

Preliminary results using the Current Population Survey which has a much more limited set of health outcomes shows that there is likely to be an income effect and an effect of minimum wages on self-rated health for working immigrants.


West Texas A&M University Anne Macy Complete

Rollins College Complete

Complete

University of Georgia Sung-Hee Jeon Complete

Florida Atlantic University Michael T. French Complete

Colgate University Complete

The Hong Kong University of Science and Technology

University of West Georgia Chad Meyerhoefer; Xiaohui Guo Complete

London School of Economics (LSE), Health Victoria Serra-Sastre; Alistair McGuire Complete

CDC; Pardee RAND Graduate School Complete

University of Miami Complete

UNC Complete

Complete

CDC Jamison Pike; Kurt Krueger Complete

Chinese Academy of Social Science Complete

University of Wisconsin, Madison Manuel Flores Complete

Economics Department, University of Illinois at Chicago

UC Davis Complete

Washington University in St. Louis Complete

Adventist University of Health Sciences Lynn Unruh Complete

Rice University Jere Behrman; Zoe Pham Complete

Providence College Dhaval Dave Complete

Mathematica Policy Research CompleteBrian Gifford; Jonathan Gellar; Yonatan Ben-Shalom

Walgreens Co. Complete

University of North Carolina at Chapel Hill Alex Gertner; Jason Rotter Complete

Tel Aviv University Complete

University of South Florida Fidel Gonzalez Complete

Chiba Institute of Technology Masato Oikawa Complete

Georgia State University Charles Courtemanche; James Marton Complete

University of Duisburg-Essen Complete

University of Georgia Complete

Lafayette College Yang Wang; Julie Smith Complete


Hospital Reimbursement and Behavior



Does Hospital Investment in Patient Safety Improve Safety? Evidence from a Panel Study of Florida Hospitals

The Impact of the 340B Drug Pricing Program on Critical Access Hospitals

Rent seeking in spine surgery: A review of 1,018,171 procedures to determine the impact of Certificate of Need programs on the site of surgical care.




Service-level Selection: Strategic Risk Selection in Medicare Advantage in Response to Risk Adjustment

How Do Hospitals Respond to Payment Incentives?

Physicians' altruism in incentives contracts: Medicare's quality race



The implications of high bed occupancy rates on hospital behaviour and quality of care in England

Management styles in the public sector - Evidence from the English NHS






Do hospitals respond to changing incentive structures? Evidence from Medicare's 2007 DRG methodology change.

Estimating the Causal Relationship between Hospital Costs and Quality Measures

“The impact of health insurance on heart attack outcomes.”

"Effects of Insurance Status on Emergency Room Care and Outcomes"

How do Hospitals Set Their Charity Care Policies? Evidence from the IRS 990



Association of the Hospital Value-Based Purchasing Program with Condition-Specific Mortality: Experience from the First Five years of Medicare’s Pay-for-Performance Program

The Impacts of CMS Public Reports of Hospital Charge Data


Changes in Health Care Use Associated with the Introduction of Hospital Global Budgets in Maryland

Abstract

Does Hospital Investment in Patient Safety Improve Safety? Evidence from a Panel Study of Florida Hospitals

For almost two decades, and with renewed intensity since the passage of the Affordable Care Act in 2010, the safety and quality of inpatient care in U.S. hospitals has been of national concern. Of particular interest to hospitals is how responsive is measured quality improvement in outcomes to investment in quality improvement. Indeed, the business case for safety and quality improvement is based on the proposition that given proper financial incentives (via more informed and selective purchasers), health care providers will naturally improve safety and quality by making the appropriate investments. But key to this proposition is that investments in quality improvement are cost effective. In this paper, we take a new and extensive look at the business case for hospital safety while, to the extent possible, avoiding some of the analytical pitfalls of earlier work such as endogenous explanatory variables and simultaneity. We use a unique panel of all general hospitals in the state of Florida between 2004 and 2015 obtained from Florida’s Center for Health Information and Policy Analysis, a department of the state’s Agency for Health Care Administration. These data include annual hospital inpatient discharge information and financial data. We estimate the effect of quality investments on patient safety measured by a composite of Agency for Healthcare Research and Quality patient safety (adverse) events. Specifically, we estimate the elasticity of patient safety with respect to several measures of quality investment using quantile regression analysis, hospital fixed effects and dynamic-panel models to address endogeneity. We find that rates of adverse events are quite inelastic with respect to hospital investment. Although we find that some kinds of hospital investment have favorable effects on safety, these effects are very small, with estimated elasticities of less than 0.1. In addition, we find evidence that policies, particularly the 2008 Centers for Medicare and Medicaid Services reimbursement-withholding policy, have stronger, but still relatively small, effects. Thus, there is little evidence to support the widely asserted “business case” for hospital investment in safety and quality.

In 2010, the Affordable Care Act extended the 340B Drug Pricing Program to allow most rural hospitals to acquire outpatient drugs from manufacturers at discounted prices. Under 340B, hospitals can dispense discounted outpatient drugs to almost any patients regardless of insurance coverage. For many urban hospitals, the program enables them to extract additional profit margin from 340B discounts, as Medicare (yet) and private insurers do not condition payment rates on whether a hospital is acquiring drugs through 340B or not. But for the majority of rural hospitals, which belong to a special classification called Critical Access Hospitals (CAH), the effect of 340B is ambiguous. CAHs receive payment of 101 percent of reasonable costs from Medicare for most inpatient and outpatient services. On the one hand, 340B likely relaxes the liquidity constraints facing CAHs, allowing them to maintain a better cash flow. On the other hand, with cost reimbursement, lower drug costs introduced by 340B lead to lower Medicare payments and less profit margin per drug used. To the extent that CAHs attract more patients by making otherwise expensive drugs more accessible, they can potentially compensate for the loss in revenue. However, CAHs serve communities that tend to have older, less well-off and dwindling populations, which could limit their ability to expand drug-intensive programs. Although close to 80 percent of CAHs have joined the program since the ACA, the net effect of 340B on CAHs and the communities they serve remains an empirical question. In this ongoing research, I quantify the effects of 340B on CAHs by examining a wide array of outcomes ranging from Medicare outpatient drug utilization to hospital financial performance. I assemble data spanning 2007 – 2013 from Medicare claims, Medicare cost reports, the American Hospital Association Annual Survey of Hospitals, and state hospital financial reports. For patient-level outcomes, my main identification strategy exploits geographic variation in outpatient market share of 340B-eligible CAHs in 2009 - measured in terms of the fraction of outpatient visits captured by these hospitals in an area - as a proxy for new exposure to 340B following the ACA. I use an event-study style difference-in-difference design that flexibly estimates the coefficient of interest for each time interval to trace out the relationship throughout the entire study period. For hospital-level financial outcomes, I use a conventional event study analysis focusing on CAHs ever participating in 340B since the expansion. I also plan to study CAH provision of uncompensated care and other service offerings under 340B.

Purpose: This study analyzed 1,018,171 U.S. spine procedures performed during 2009 – 2015, to determine whether the surgery site - hospital inpatient v/s lower-cost alternatives, e.g., ambulatory surgery centers (ASC) and hospital outpatient centers (HOPD) - is associated with the existence of state Certificate of Need (CON) regulation. Background: Rising costs and performance frequency place spine surgery in the highest category of hospital expenditures and make it a research interest to payers, hospitals, and the US government. Spine surgery performed in lower-cost settings (e.g., ASCs) is known to be safe and effective, and can reduce cost by as much as 65%. Private-practice spine surgeons have taken advantage of health technology improvements by partnering with private insurers to move common spine procedures, e.g., spinal decompressions or cervical fusions (ACDFs), from inpatient to ASC setting. However, the CON regulation is imposing in 25 U.S. states. CON regulation requires a non-trivial fee and approval from a state level board before an ASC can operate and compete with incumbent institutions. These institutions usually retain lobbyists to influence the board approval. Methods/Data: Using claims data from the MarketScan® databases, our study tests the hypothesis that the 25 states where the CON program remains in force have fewer spine procedures performed in the lower-cost settings. We analyzed all qualified and reimbursed spine procedures performed between 2009 and 2015 to determine whether the frequency of common spine procedures performed in lower cost settings is associated with the existence of CON regulation. A variable identifying the site of service is available in the data set and the procedures types were identified by MS-DRG or CPT® codes. Importantly, we only used MS-DRG codes without complications/comorbidities or major complications/comorbidities and we controlled for patient diagnosis that could impact the site of service decision using ICD-9 codes. We estimated a multinomial logit model with state clustered errors. Procedure data on males and females were analyzed together and separately to identify any gender differences in health seeking behavior. Results: Results indicate that the presence of a CON program led to significant decreases in the frequency of ACDFs and spinal decompressions performed in both lower-cost settings. Surprisingly, the CON regulation effects analyzed separately for male and female sub-samples differ significantly and this suggests gender differences in care seeking. Estimated risk ratios showed that patients in states with a CON program are 40% less likely to have an ACDF in an ASC and 37% less likely to have a decompression in an ASC. The effect was more pronounced in HOPDs where patients were 46% less likely to have a spinal decompression and 93% less likely to have an ACDF. In states with a CON regulation, males and females underwent significantly fewer procedures in the lower-cost settings and there are differences in the odds ratios of the separately estimated gender regression models. Conclusion/Discussion: State CON regulation is associated with fewer spine procedures performed in lower-cost settings. Study findings have policy implications for both value-based purchasing and the design of alternative payment models in spine surgeries.

The Centers for Medicare and Medicaid Services (CMS) has phased in the Hierarchical Condition Categories (HCC) risk adjustment model during 2004-2006 to more accurately estimate capitated payments to Medicare Advantage (MA) plans to reflect each beneficiary’s health status. However, it is debatable whether the CMS-HCC model has led to strategic evolutions of risk selection. We examine the competing claims and analyze the risk selection behavior of MA plans in response to the CMS-HCC model. We find that the CMS-HCC model reduced the phenomenon that MA plans avoid high-cost beneficiaries in traditional Medicare plans, whereas it led to increased disenrollment of high-cost beneficiaries, conditional on illness severity, from MA plans. We explain this phenomenon in relation to service-level selection. First, we show that MA plans have incentives to effectuate risk selection via service-level selection, by lowering coverage levels for services that are more likely to be used by beneficiaries who could be unprofitable under the CMS-HCC model. Then, we empirically test our theoretical prediction that compared to the pre-implementation period (2001-2003), MA plans have raised copayments disproportionately more for services needed by unprofitable beneficiaries than for other services in the post-implementation period (2007-2009). The disproportionate changes in copayments led to voluntary disenrollment of beneficiaries with need for these services, who tend to incur higher expenditures than their risk-adjusted payments. We also find evidence supporting our hypothesis that those who were less satisfied with out-of-pocket costs were more likely to disenroll from MA plans. Such strategic behavior led to MA plans to save $5.2 billion in 2007-2009 by simply transferring the costs to the federal government, thereby placing significant financial burdens on the federal government. Our results provide key policy implications for CMS in moving towards a better risk adjustment model that accounts for the enrollees’ predicted risk scores while generating economic incentives for MA plans that discourage service-level selection.

How Do Hospitals Respond to Payment Incentives? Given the accelerating rise in U.S. public health care expenditures, Medicare has launched various pilot reimbursement models to reform health care delivery by holding health care providers financially accountable. Under Medicare’s currently dominant payment method, fee-for-service (FFS), each health care provider involved in an episode of care is reimbursed for services provided to a patient. This volume-based payment model incentivizes profit-maximizing health care providers to overuse healthcare resources regardless of the necessity or quality of care. In addition, the inefficient FFS system creates a fragmented health care delivery system, where health care providers have no incentive to coordinate during an episode of treating a health condition. The growing need for alternative reimbursement models is more recognized for certain health conditions including total joint replacement (TJR). There are numerous entities involved in TJR, which are not coordinated under the current fee-for-service payment method. This paper investigates the impact of the Bundled Payments for Care Improvement Model 2 (BPCI Model 2) initiative, one of the most recent and comprehensive alternative payment methods, on reducing health care cost, and reforming the coordination and quality of the care delivered to total joint replacement patients. In BPCI Model 2, participating hospitals are financially accountable for Medicare spending on hospitalization, post-acute care and all related services, including readmission, up to 90 days after hospital discharge. By using hospital discharge-level data from State Inpatient Databases, developed by Agency for Healthcare Research and Quality, I investigate the mechanism used by financially accountable hospitals to reduce health care utilization and aggregate expenditures over a cycle of treating TJR. I estimate a difference-in-difference model where I compare inpatient length of stay, number of procedures, hospital total charges, and disposition pattern of TJR patients of BPCI-participating hospitals to outcomes of non-participating ones before and after the bundled payment was implemented. In order to study the impact of BPCI Model 2 on the quality of care, I estimate the effect of this payment reform on probability of re-hospitalization within 30 or 90 days after hospital discharge. Moreover, I stratify the sample by payer type to investigate the broader effect of BPCI Model 2 and whether participating hospitals discriminate against TJR patients with other insurance types.

My empirical analyses suggest despite a significant 15 percentage point decrease in probability of discharging patients to institutional post-acute care facilities, and one third of day reduction in hospital length of stay, the quality of care is improved and there is a significant decrease in probability of re-hospitalization. I find no cost-shifting effect on non-Medicare patients or any significant evidence that BPCI-participating hospitals use different practice patterns among TJR patients across payer types. In addition, I find no supporting evidence that the orthopedic surgeons performing surgery in BPCI-participating hospitals sort healthier patients to those entities and divert more intensive patients to non-participating ones. The results suggest BPCI Model 2 leads to less post-acute care utilization and higher quality.

JEL Classification— H5, I1, L1 Keywords— Medicare, Bundled Payment, Health Care

Public contracting with hospitals under asymmetric information about their technology provides a classic example of an agency problem, where government as a principal can achieve social optimum in terms of product's quantity and agent's efforts through nonlinear prices. Incentives contracts are targeted at increasing aggregate performance, which is observed in the experimental and empirical literature. However, mean effect hinders heterogeneity in the responses of agents who differ in their abilities. In particular, both theoretical literature and natural experiments point to deteriorating performance of the front-runners. Theoretical explanations include motivation crowding out owing to intrinsic behavior or slacking efforts in tournaments, particularly in the dynamic context. Altruistic agents, however, would be interested in a social value of their performance per se. There is limited theoretical literature on altruism in public good games and piece-rate incentives contracts, but little is known about the influence of altruism on the outcomes of tournaments. The paper analyzes the impact of physicians' altruism and motivation on the outcomes of rank-order tournaments in healthcare, where a fixed price contract on quantity is supplemented with a relative performance contract on quality. Our theoretical model forecasts crowding out of most altruistic types owing to the effect of the participation constraint. In an empirical application to the Medicare's nationwide natural experiment with a relative performance contract on quality for acute inpatient care since 2013, we observe the proof of the model's predictions. Namely, the quality dimensions, which are linked to patient's benefit, demonstrate higher deterioration among top-performing hospitals than other incentivized dimensions. We use the framework of tournaments in healthcare, where a fixed price contract on quantity is supplemented with a relative performance contract on quality. Our model shows quality convergence, however, altruism may lead to quality decrease among subgroups of the high-performing agents. In testing the model's hypotheses, we focus on Medicare's hospitals with top performance and show that quality dimensions, which are not linked to patient's benefit, demonstrate lower quality decrease than other dimensions. In other words, altruism may become a reason for motivation crowding out on the healthcare market. The novelty of our empirical approach is severalfold. Firstly, we use dynamic panel data estimations to account for ``habit-formation''. The analysis excludes ``regression-to-the-mean'' effect by modelling the time-dependent long-term mean as a function of hospital characteristics. Secondly, while previous studies exploited the data for prototypes of value-based purchasing and concentrated on composite measures, we use longitudinal data sets on each quality measure of all acute-care Medicare's hospitals before and after the reform (fiscal years 2004--2016). The data are supplemented with patient case-mix, ownership, share of Medicare population and various hospital control variables, coming from: Medicare's Impact Files, Final Rules, Provider of Service Data, and Provider Utilization and Payment Data.

The English National Health Service has experienced significant increases in the percentage of hospital beds that are occupied at any point in time, from an average of 84.5% in 2006/07 to 91.4% in 2016/17. Moreover, it is not uncommon for some hospitals to reach 100% bed utilisation at certain times of the day. This is the consequence of a developing mismatch between supply and demand side factors, including a consistent decrease in hospital bed stock at a time of significant rises in the number of hospital admissions. Across OECD countries, only hospitals in Canada, Ireland, Israel and Norway are experiencing higher average bed occupancy rates. Clinical leadership has expressed concern about rises in bed occupancy rates and how they might affect the ability of hospital teams to deliver high quality care. There are particular concerns that high bed occupancy rates might lead to an increased likelihood of adverse events, challenges in securing the resources needed to diagnose and treat patients, and problems with planning for the discharge of larger numbers of patients. Despite the salience of the issue to resource-stretched hospitals, very little research has examined the implications of high bed occupancy rates on hospital behaviour and the quality of care. This is the first study to assess daily changes in bed occupancy rates and their relationship with risk-adjusted, patient discharges and their subsequent 30-day readmission rate, from April 2014 to February 2016. Hospital Episode Statistics data was used to identify all patients treated in English hospitals across the observation period, with the daily hospital-level bed occupancy rate calculated as the ratio of registered inpatients at midnight and the recorded stock of hospital beds. We modelled panel data models for each hospital and day across the observation period, using ordinary least squares estimators with hospital fixed effects to relate changes in daily bed occupancy rates to the two outcome variables. Sensitivity analysis was conducted for a range of selected patient subgroups, including patients of different ages, with different comorbidities and from varying socioeconomic backgrounds. We find that an increase in bed occupancy rate by 1% was associated with a 0.49% (p<0.001) rise in the discharge rate, and a 0.011% (p<0.001) increase in the 30-day readmission rate for discharged patients. These associations became more pronounced once bed occupancy reaches the highest tertile of the bed occupancy distribution, around 95%, with each extra 1% rise in bed occupancy associated with a 0.04% (p<0.001) rise in the readmission rate. Older patients (e.g. patients aged 81 to 90 years) and those with a greater number of comorbidities (e.g. 5 comorbidities) were less likely to be discharged at times of high bed occupancy, but if they were discharged, they had an increased risk of readmission. In this study, we find that when bed occupancy rates are high, hospitals discharge a greater proportion of their patients. However, increased bed occupancy was not associated with a substantial increase in 30-day readmission rates. It may be that hospitals are successfully prioritising early discharge among the least vulnerable patients.

This paper investigates the effect of top-managers on the performance of public sector organisations using CEOs of English public hospitals as a case study. In many countries, the public sector is important not only in financing but also in delivering services such as education and healthcare. In the search for greater productivity of public sector organisations, a popular political reform model is to introduce more autonomy and market discipline in the system. An important element of these reforms tends to be a greater emphasis on the role of top managers, accompanied by the use of manager-specific compensation policies, and performance-related pay and dismissals. This approach is underpinned by the belief that top managers are central to the performance of the organisation. The role of top managers for firm performance has been extensively studied in private sector organizations, starting with the seminal paper by Bertrand and Schoar (2003). The evidence on the effectiveness of top management in the public sector, however, is still relatively scarce. In particular, no evidence exists about the effectiveness of individual managers for hospital performance. We address this gap in the literature by examining the role of top managers in English public hospitals. In the late 1980s, the English government embarked on a long-standing reform programme which replaced the consensus management characterizing the National Health Service (NHS) in the 1970s with a decentralized model. In this new model, CEOs could assume a largely undisputed responsibility for the management and performance of individual public hospitals, and individual hospital boards could select and reward individual CEOs in a fully decentralized fashion. These shifts, and the frequent managerial rotations of the same CEOs across NHS hospitals that arose as a consequence of the new policies, provide an ideal setting to study whether individual managers are indeed associated with systematic differences in hospital performance. An additional appealing factor of the NHS system is the richness of the hospital performance data that we can use in the analysis, spanning a wide set of financial and clinical final outcomes, as well as intermediate outputs and operational variables. We begin by documenting systematic differences in CEO pay: moving from the 25th percentile to the 75th percentile of the CEO effects in pay distribution represents a 12% increase in pay relative to mean CEO pay. These differences in managerial pay provide prima-facie evidence of the existence of a perceived difference in ability across different managers. Next, we estimate whether individual managers are associated with actual differences in hospital performance and a variety of intermediate operational outcomes employing three different approaches. First, the fixed effects approach pioneered by Bertrand and Schoar (2003). Second, the alternative two-step method proposed by Bertrand and Schoar (2003) and third, a non-parametric approach that resembles a difference-in-difference matching estimator. Overall, we find little evidence of individual CEOs having an impact on hospital performance. Our results also raise concerns about policy approaches that rely on the contribution of transient "turnaround" top managers to improve the performance of individual hospitals.

In 2007, the Centers for Medicare and Medicaid (CMS) significantly restructured the diagnosis related group (DRG) system by moving from the CMS DRG to the Medicaid Severity (MS) DRG. The MS DRGs expanded the number of illness categories, explicitly recognizing the variation of complications present within certain conditions, and allowing for differential reimbursement depending on the severity of the case. Each DRG has a weights which reflect the average service intensity for patient with that illness. Payments are proportional to weights. After the implementation of the MS-DRG system, hospitals may choose from up to three new DRGs depending upon the severity of the diagnosis reported including either complication/co-morbidity (CC) or no complication/co-morbidity (non-CC) or major complication/comorbidity (MCC). Each of these new DRGs has a different reimbursement weight with the more complicated cases receiving higher weights and more reimbursement, the less complicated cases receive a lower weight and lower reimbursement. We explore whether hospitals, given a diagnosis, systematically code patients into the higher severity MS DRG to increase their reimbursements. In the economics literature, this is known as upcoding (e.g. Dafny 2005, Silverman and Skinner, 2004). Using the National Inpatient Survey (NIS) data from 2005-2008 and MS DRG weights from CMS, we first compare the average DRG weights using the MS DRGs to the counterfactual calculation of the average DRG weights using the old CMS DRGs. Our preliminary results indicate that patients have an increase in average DRG weight of .048 to .071. This indicates that patients are recorded as 4-6 percent sicker, under the MS-DRG methodology. However, this is only suggestive of upcoding because the distribution of severity of illness within a CMS DRG may have changed such that patients needed more services and the higher weights were appropriate. We use a look back approach applying the counterfactual calculation to data for the three years before the policy change to determine the extent to which aggregate weights persist over time. For a more formal test of upcoding, we estimate the equation below for those DRGs where we observe three categories post 2007. Weight refers to the weight associated with the assigned DRG for patient i in hospital h. Spread refers to the difference in the weights between different severity levels in MS-DRG group and γh are hospital fixed effects (Dafney, 2005 and Barros and Braun 2017 employ a similar method). 2008Weightigh=β0+ β1Spread2-1, g+ β2Spread3-2, g + β32007Weightigh + γh +εigh (1) Table 1 (not shown) presents estimates of equation 1 stratified by hospital type. The positive and significant coefficients on the spread variables are indicative of upcoding. Another concern is technological change, which may increase the service intensity of the illness, and increase the weight of the DRG. However, this would only be a threat to identification if the technological change were newly implemented in 2008 and were within the MS DRGs which had seen expansion in the policy change. To address this, we will examine DRG groups with high frequency (common ailments) and investigate technological change in these groups.

This paper evaluates the relationship between hospital cost and quality. The Department of Health and Human Services collects quality information from hospitals and produces a number of quality measures that are publicly available through its Hospital Compare project. The purpose of providing these quality measures is to increase transparency and accountability in the healthcare system. Another area where these quality measures could be useful is in quality adjusting hospital prices in the Consumer Price Index (CPI) and the Producer Price Index (PPI). Currently, the PPI makes some use of these measures, but the methodology used may be problematic as the relationship between the quality measures and hospital costs is unknown. Once the relationship between the quality measures and costs are better understood, it will be possible to revise the existing PPI methods. This paper estimates the causal relationship between the quality measures and hospital costs. The causal relationship is identified using the instrumental variable technique of Doyle, Graves, Gruber, and Kleiner (JPE 2015) and Doyle, Graves, and Gruber (NBER working paper 2017). They develop an instrument for hospital selection based on plausibly exogenous assignment to different ambulance companies (which have different preferences for hospitals). Doyle, Graves, Gruber, and Kleiner (JPE 2015) look at whether hospitals that have higher Medicare reimbursements have better outcomes. Doyle, Graves, and Gruber (NBER working paper 2017) look at the relationship between the Hospital Compare quality measures and outcome measures. This paper uses Medicare claims data for the years 2011-2015. Data on hospital admissions is contained in the Inpatient file, and ambulance billing data are contained in the Outpatient and Carrier files. The claims data are linked to hospital specific cost (hospital cost reports) and quality data (Hospital Compare). Preliminary results suggest a relationship between cost and quality but is sensitive to the cost and quality measures used. A complicating factor in the analysis is that CMS began phasing in the use of quality measures in hospital reimbursements during this period.

This paper investigates whether health insurance status has an impact on survival and admission probabilities for heart attack patients presenting through the emergency department. The paper uses a methodology from Currie, MacLeod and Van Parys (2016) to address endogeneity and evaluate whether the differences across different types of health insurance. Using data from the 2006-2010 HCUP NEDS, I show patterns in the probability of death, hospital admission and treatment patterns patients with a primary diagnosis of acute myocardial infarction before, during, and after the Great Recession.

Health economics researchers have long known insurance status has an impact on the use of health care system (Hadley 2003); however, measuring the causal effect of insurance status on treatment decisions and outcomes has been a difficult problem for economists to tackle. This paper builds upon earlier work by Doyle (2005) in examining the effects of insurance status on injury patients in the emergency room. Doyle (2005) uses a Wisconsin sample of patients reporting with injuries resulting from motor vehicle accidents. The present examination improves on Doyle’s study by broadening the dataset, controlling for overall health status and considering multiple types of health shocks. I use a nationwide dataset of emergency department visits to compare the differences between the effects of insurance status on treatment for patients with injuries from motor vehicle accidents and patients with other types of injuries while controlling for overall health status. This paper reaches three primary conclusions: (1) Patients paying out of pocket experience worse health outcomes; (2) Patients paying out of pocket receive fewer health services in response to motor vehicle accident injuries and other types of injuries; and, (3) Despite no observed differences between patients paying out of pocket and patients receiving charity care, charity care patients generally received more healthcare services and had better health outcomes than patients with private insurance.

Hospitals play an important role in the social insurance system by providing free and discounted care to the uninsured. In 2012, hospitals provided over $46 billion in uncompensated care -- almost 30% of Medicaid inpatient and outpatient spending that year. Despite the large economic cost of uncompensated care to hospitals, relatively little is known about how hospitals determine the amount of uncompensated care they provide. In this paper, I look at how nonprofit hospitals adjust their charity care policies in response to increases in public funding, state regulations, and changes in the local market for medical care. To do so, I use new data from the IRS 990 Schedule H on the universe of non-profit hospital charity care policies between 2010 and 2015. I find significant evidence that hospitals raise their charity care policies in response to changes in hospital income, but limited evidence that hospitals respond to changes in local need. Furthermore, I find that state changes in the regulation of non-profit hospitals community benefit activities have ambiguous on charity care provision, and are most effective when the attorney general has regulatory power.

Medicare’s Hospital Value Based Purchasing Program (HVBP), implemented since 2013, aims to improve inpatient care quality using hospital financial incentives. In 2014, performance on 30-day mortality for acute myocardial infarction (AMI), heart failure (HF), and pneumonia were added to the multiple other quality measures used in determining the size of hospital penalties or bonuses for Inpatient Prospective Payment System (IPPS) hospitals - the hospitals targeted by HVBP. Prior studies used control hospitals that were systematically different from hospitals in which HVBP was implemented, raising concerns that these differences may confound the estimation of changes resulting from the HVBP. We stratified IPPS hospitals based on the extent of their reliance on Medicare patients and using publicly reported data on hospital performance from 2009-2016 examined the association between HVBP incentives and changes in 30-day mortality by comparing pre- to post-HVBP changes in hospital 30-day mortality among high vs. low Medicare share hospitals. We examined 30-day mortality changes for three admission cohorts – acute myocardial infarction (AMI), heart failure (HF) and pneumonia – that were all introduced into the HVBP in 2014. We obtained mortality data from the Centers for Medicare and Medicaid Services’ (CMS’) Hospital Compare, information on hospital type from the CMS Final Impact Rule, and hospital characteristics data from the American Hospital Association Annual Survey. Our anlytic sample comprised 1,915 eligible IPPS hospitals from 2009-2016 (1,659 (256) high (low) Medicare share hospitals). We evaluated the association of the HVBP with the mortality outcomes using a difference-in-differences approach, whereby pre- vs. post-HVBP changes in the outcome in high Medicare share hospitals were contrasted with corresponding changes in low Medicare share hospitals. Specifically, we used a linear (hospital-level) random effects regression model and adjusted for hospital characteristics and year effects to account for secular trends in the mortality outcomes. Given potential changes in 30-day readmission rate for AMI, HF, and pneumonia associated with the Hospital Readmission Reduction Program (HRRP) - another CMS incentive program that was introduced alongside the HVBP in 2010 - there may have been unintended spillover effects on 30-day mortality for the aforementioned conditions. As a sensitivity analysis, we re-estimated variants of our main models that included 30-day risk-adjusted readmission rate for the corresponding admission cohort as an additional covariate. We found that introduction of the HVBP was associated with a relative increase in 30-day AMI mortality (0.27%, 95% confidence interval (CI) [0.01%, 0.53%]) and 30-day mortality for pneumonia admissions (0.29%, 95% CI [0.10%, 0.48%]), but no change in 30-day mortality for HF admissions in high Medicare share hospitals vs. low Medicare share hospitals. Additionally, we did not find evidence of spillover effects from changes in readmission rates due to the HRRP (i.e., higher mortality rates for targeted conditions). The lack of improvement in patient mortality through five years of the HVBP program suggests that careful re-evaluation of the program is warranted, especially in the light of previous work in behavioral economics which show that financial incentives can be detrimental for motivation and can lead to worse hospital performance.

Background In May 2013, the Centers for Medicare and Medicaid Services (CMS) began publishing charges for the 100 most frequently billed diagnosis-related groups (DRGs) for inpatients treated in approximately 3,400 U.S. hospitals. The goal was to increase transparency and bring greater market forces to bear on the steep growth in prices for hospital services. While charges are not the same as payments, the data release has several advantages over claims data. The reports are readily available online, national in scope, and the subject of media attention. Charges generally are the starting point for hospital-insurer price negotiations. High charges also have a direct impact on uninsured individuals, who are exposed to full charges, as well as patients covered out-of-network or under workers’ compensation, who generally pay a portion of full charges. Research Questions This paper examines whether the CMS price transparency initiative is performing as a policy lever in reducing the growth of hospitals prices and explores which stakeholders are the most receptive audiences. More specifically, we address supply side responsiveness by examining trends in hospital inpatient charges and demand side responsiveness by examining trends in volume and market share of inpatient services. Methods Using quasi-experimental designs, we estimated econometric models using charge and utilization data obtained from CMS and from the Florida and New York AHRQ State Inpatient Databases for the years 2011-2015. To examine the impact of the CMS public reports on hospital charges, we estimated difference-in differences models that compared changes in charges for the 100 reported DRGs with changes in charges for the unreported DRGs, controlling for DRG weight, geographic location, and patient factors. To examine the impact of the public reports on consumer choice, we explored shifts in volume and market shares. We selected two reported DRGs based on price sensitivity and high volume: lower extremity total joint replacement and percutaneous coronary intervention (PCI – excluding cases admitted through emergency departments). We estimated the impact of the release of CMS public reports in May 2013 on volume and market share of these conditions using quarterly observations on individual hospitals and controlling for hospital characteristics. Findings Results indicate that in Florida, inflation adjusted charges increased between 2013 and 2015, but were 4% lower in the 100 reported DRGs compared to other DRGs. We did not find differences in either volume or market share for total joint replacement or PCI in high-charge hospitals relative to low-charge hospitals. It appears that in Florida, hospitals were more responsive to CMS public reporting than consumers were. Estimation of models using New York data are in progress.

Background: In 2014, the State of Maryland placed the majority of its hospitals under all-payer global budgets for inpatient, hospital outpatient, and emergency department care. Maryland’s payment reform has garnered considerable attention from policymakers because of its unique approach to reorienting hospitals from a traditional fee-for-service payment model to a model that rewards hospitals for managing population health—including the provision of care outside of hospital settings. Despite reports from CMS and Maryland of the program’s early success in controlling hospital spending, the program's impacts on hospital and primary care use remain largely unknown. Under the structure of Maryland’s program, hospitals can lower spending by reducing hospital utilization and enhancing primary care—as policymakers had intended—or by reducing their prices to meet their budgets. Design, Setting, and Participants: Using a quasi-experimental difference-in-differences design, we compared changes in hospital and primary care use among fee-for-service Medicare beneficiaries in Maryland vs. 27 matched out-of-state control counties from before (2009-13) to after (2014-15) Maryland's introduction of hospital global budgets. We conducted our analyses under two sets of assumptions. First, we assumed pre-intervention differences between Maryland and the control counties would have remained constant past 2014 had Maryland not implemented global budgets (the standard identifying assumption of difference-in-difference analyses). Second, we assumed differences in pre-intervention trends would have continued without the state’s payment change (differential trend assumption). In supplementary analyses, we compared our main difference-in-differences estimates to those generated from a series of placebo tests in which we iteratively re-assigned states outside of Maryland to the intervention group; using estimates generated in these tests, we assessed whether changes in Maryland consistently exceeded changes detected in unaffected (placebo) states. Results: Assuming parallel trends, we estimated a differential change in Maryland of -0.47 annual hospital stays/100 beneficiaries (95% CI: -1.65,0.72; P=0.43) from the pre-intervention period (2009-13) to 2015, but assuming differential trends, we estimated a differential change in Maryland of -1.24 stays/100 beneficiaries (95% CI: -2.46,-0.02; P=0.047). Assuming parallel trends, we found a significant increase in primary care visits (+10.6 annual visits/100 beneficiaries; 4.6,16.6; P=0.001), but assuming differential trends, we found no change (-0.8 visits/100 beneficiaries; 95% CI: -10.6,9.0; P=0.87). Comparing estimates with both trend assumptions, we found no consistent changes in emergency department visits, return hospital stays, HOPD use, or post-hospitalization primary care visits associated with Maryland’s program. Differential changes in Maryland were within the distribution of changes detected in supplementary placebo analyses, suggesting that hospital and primary care utilization did not change differentially in Maryland from changes that we would have expected to see in the absence global budgets. Conclusions and Relevance: We did not find consistent evidence that Maryland’s hospital global budget program was associated with anticipated reductions in hospital use or increases in primary care visits among fee-for-service Medicare beneficiaries after two years. Given the challenge of changing practice patterns over the short-term, evaluations over longer periods and in younger populations—for whom care outside the hospital may be more appropriate—should be pursued.

Abstract

For almost two decades, and with renewed intensity since the passage of the Affordable Care Act in 2010, the safety and quality of inpatient care in U.S. hospitals has been of national concern. Of particular interest to hospitals is how responsive is measured quality improvement in outcomes to investment in quality improvement. Indeed, the business case for safety and quality improvement is based on the proposition that given proper financial incentives (via more informed and selective purchasers), health care providers will naturally improve safety and quality by making the appropriate investments. But key to this proposition is that investments in quality improvement are cost effective. In this paper, we take a new and extensive look at the business case for hospital safety while, to the extent possible, avoiding some of the analytical pitfalls of earlier work such as endogenous explanatory variables and simultaneity. We use a unique panel of all general hospitals in the state of Florida between 2004 and 2015 obtained from Florida’s Center for Health Information and Policy Analysis, a department of the state’s Agency for Health Care Administration. These data include annual hospital inpatient discharge information and financial data. We estimate the effect of quality investments on patient safety measured by a composite of Agency for Healthcare Research and Quality patient safety (adverse) events. Specifically, we estimate the elasticity of patient safety with respect to several measures of quality investment using quantile regression analysis, hospital fixed effects and dynamic-panel models to address endogeneity. We find that rates of adverse events are quite inelastic with respect to hospital investment. Although we find that some kinds of hospital investment have favorable effects on safety, these effects are very small, with estimated elasticities of less than 0.1. In addition, we find evidence that policies, particularly the 2008 Centers for Medicare and Medicaid Services reimbursement-withholding policy, have stronger, but still relatively small, effects. Thus, there is little evidence

In 2010, the Affordable Care Act extended the 340B Drug Pricing Program to allow most rural hospitals to acquire outpatient drugs from manufacturers at discounted prices. Under 340B, hospitals can dispense discounted outpatient drugs to almost any patients regardless of insurance coverage. For many urban hospitals, the program enables them to extract additional profit margin from 340B discounts, as Medicare (yet) and private insurers do not condition payment rates on whether a hospital is acquiring drugs through 340B or not. But for the majority of rural hospitals, which belong to a special classification called Critical Access Hospitals (CAH), the effect of 340B is ambiguous. CAHs receive payment of 101 percent of reasonable costs from Medicare for most inpatient and outpatient services. On the one hand, 340B likely relaxes the liquidity constraints facing CAHs, allowing them to maintain a better cash flow. On the other hand, with cost reimbursement, lower drug costs introduced by 340B lead to lower Medicare payments and less profit margin per drug used. To the extent that CAHs attract more patients by making otherwise expensive drugs more accessible, they can potentially compensate for the loss in revenue. However, CAHs serve communities that tend to have older, less well-off and dwindling populations, which could limit their ability to expand drug-intensive programs. Although close to 80 percent of CAHs have joined the program since the ACA, the net effect of 340B on CAHs and the communities they serve remains an empirical question. In this ongoing research, I quantify the effects of 340B on CAHs by examining a wide array of outcomes ranging from Medicare outpatient drug utilization to hospital financial performance. I assemble data spanning 2007 – 2013 from Medicare claims, Medicare cost reports, the American Hospital Association Annual Survey of Hospitals, and state hospital financial reports. For patient-level outcomes, my main identification strategy exploits geographic variation in outpatient market share of 340B-eligible CAHs in 2009 - measured in terms of the fraction of outpatient visits captured by these hospitals in an area - as a proxy for new exposure to 340B following the ACA. I use an event-study style difference-in-difference design that flexibly estimates the coefficient of interest for each time interval to trace out the relationship throughout the entire study period. For hospital-level financial outcomes, I use a conventional event study analysis focusing on CAHs ever participating in 340B since the expansion. I also plan to study CAH provision of uncompensated care and other service offerings under 340B.

This study analyzed 1,018,171 U.S. spine procedures performed during 2009 – 2015, to determine whether the surgery site - hospital inpatient v/s lower-cost alternatives, e.g., ambulatory surgery centers (ASC) and hospital outpatient centers (HOPD) - is associated with the existence of state Certificate of Need (CON) regulation.

Rising costs and performance frequency place spine surgery in the highest category of hospital expenditures and make it a research interest to payers, hospitals, and the US government. Spine surgery performed in lower-cost settings (e.g., ASCs) is known to be safe and effective, and can reduce cost by as much as 65%. Private-practice spine surgeons have taken advantage of health technology improvements by partnering with private insurers to move common spine procedures, e.g., spinal decompressions or cervical fusions (ACDFs), from inpatient to ASC setting. However, the CON regulation is imposing in 25 U.S. states. CON regulation requires a non-trivial fee and approval from a state level board before an ASC can operate and compete with incumbent institutions. These institutions usually retain lobbyists to influence the board approval.

Using claims data from the MarketScan® databases, our study tests the hypothesis that the 25 states where the CON program remains in force have fewer spine procedures performed in the lower-cost settings. We analyzed all qualified and reimbursed spine procedures performed between 2009 and 2015 to determine whether the frequency of common spine procedures performed in lower cost settings is associated with the existence of CON regulation. A variable identifying the site of service is available in the data set and the procedures types were identified by MS-DRG or CPT® codes. Importantly, we only used MS-DRG codes without complications/comorbidities or major complications/comorbidities and we controlled for patient diagnosis that could impact the site of service decision using ICD-9 codes. We estimated a multinomial logit model with state clustered errors. Procedure data on males and females were analyzed together and separately to identify any gender differences in health seeking behavior.

Results indicate that the presence of a CON program led to significant decreases in the frequency of ACDFs and spinal decompressions performed in both lower-cost settings. Surprisingly, the CON regulation effects analyzed separately for male and female sub-samples differ significantly and this suggests gender differences in care seeking. Estimated risk ratios showed that patients in states with a CON program are 40% less likely to have an ACDF in an ASC and 37% less likely to have a decompression in an ASC. The effect was more pronounced in HOPDs where patients were 46% less likely to have a spinal decompression and 93% less likely to have an ACDF. In states with a CON regulation, males and females underwent significantly fewer procedures in the lower-cost settings and there are differences in the odds ratios of the separately estimated gender regression models.

State CON regulation is associated with fewer spine procedures performed in lower-cost settings. Study findings have policy implications for both value-based purchasing and the design of alternative payment

The Centers for Medicare and Medicaid Services (CMS) has phased in the Hierarchical Condition Categories (HCC) risk adjustment model during 2004-2006 to more accurately estimate capitated payments to Medicare Advantage (MA) plans to reflect each beneficiary’s health status. However, it is debatable whether the CMS-HCC model has led to strategic evolutions of risk selection. We examine the competing claims and analyze the risk selection behavior of MA plans in response to the CMS-HCC model. We find that the CMS-HCC model reduced the phenomenon that MA plans avoid high-cost beneficiaries in traditional Medicare plans, whereas it led to increased disenrollment of high-cost beneficiaries, conditional on illness severity, from MA plans. We explain this phenomenon in relation to service-level selection. First, we show that MA plans have incentives to effectuate risk selection via service-level selection, by lowering coverage levels for services that are more likely to be used by beneficiaries who could be unprofitable under the CMS-HCC model. Then, we empirically test our theoretical prediction that compared to the pre-implementation period (2001-2003), MA plans have raised copayments disproportionately more for services needed by unprofitable beneficiaries than for other services in the post-implementation period (2007-2009). The disproportionate changes in copayments led to voluntary disenrollment of beneficiaries with need for these services, who tend to incur higher expenditures than their risk-adjusted payments. We also find evidence supporting our hypothesis that those who were less satisfied with out-of-pocket costs were more likely to disenroll from MA plans. Such strategic behavior led to MA plans to save $5.2 billion in 2007-2009 by simply transferring the costs to the federal government, thereby placing significant financial burdens on the federal government. Our results provide key policy implications for CMS in moving towards a better risk adjustment model that accounts for the enrollees’ predicted risk scores while generating economic incentives for MA plans

Given the accelerating rise in U.S. public health care expenditures, Medicare has launched various pilot reimbursement models to reform health care delivery by holding health care providers financially accountable. Under Medicare’s currently dominant payment method, fee-for-service (FFS), each health care provider involved in an episode of care is reimbursed for services provided to a patient. This volume-based payment model incentivizes profit-maximizing health care providers to overuse healthcare resources regardless of the necessity or quality of care. In addition, the inefficient FFS system creates a fragmented health care delivery system, where health care providers have no incentive to coordinate during an episode of treating a health condition. The growing need for alternative reimbursement models is more recognized for certain health conditions including total joint replacement (TJR). There are numerous entities

This paper investigates the impact of the Bundled Payments for Care Improvement Model 2 (BPCI Model 2) initiative, one of the most recent and comprehensive alternative payment methods, on reducing health care cost, and reforming the coordination and quality of the care delivered to total joint replacement patients. In BPCI Model 2, participating hospitals are financially accountable for Medicare spending on hospitalization, post-acute care and all related services,

By using hospital discharge-level data from State Inpatient Databases, developed by Agency for Healthcare Research and Quality, I investigate the mechanism used by financially accountable hospitals to reduce health care utilization and aggregate expenditures over a cycle of treating TJR. I estimate a difference-in-difference model where I compare inpatient length of stay, number of procedures, hospital total charges, and disposition pattern of TJR patients of BPCI-participating hospitals to outcomes of non-participating ones before and after the bundled payment was implemented. In order to study the impact of BPCI Model 2 on the quality of care, I estimate the effect of this payment reform on probability of re-hospitalization within 30 or 90 days after hospital discharge. Moreover, I stratify the sample by payer type to investigate the broader effect of BPCI Model 2 and whether participating hospitals discriminate against TJR

My empirical analyses suggest despite a significant 15 percentage point decrease in probability of discharging patients to institutional post-acute care facilities, and one third of day reduction in hospital length of stay, the quality of care is improved and there is a significant decrease in probability of re-hospitalization. I find no cost-shifting effect on non-Medicare patients or any significant evidence that BPCI-participating hospitals use different practice patterns among TJR patients across payer types. In addition, I find no supporting evidence that the orthopedic surgeons performing surgery in BPCI-participating hospitals sort healthier patients to those entities and divert more intensive patients to non-participating ones. The results suggest BPCI Model 2 leads to less post-acute care utilization and higher quality.

Public contracting with hospitals under asymmetric information about their technology provides a classic example of an agency problem, where government as a principal can achieve social optimum in terms of product's quantity and agent's efforts through nonlinear prices. Incentives contracts are targeted at increasing aggregate performance, which is observed in the experimental and empirical literature. However, mean effect hinders heterogeneity in the responses of agents who differ in their abilities. In particular, both theoretical literature and natural experiments point to deteriorating performance of the front-runners. Theoretical explanations include motivation crowding out owing to intrinsic behavior or slacking efforts in tournaments, particularly in the dynamic context. Altruistic agents, however, would be interested in a social value of their performance per se. There is limited theoretical literature on altruism in public good games and piece-rate incentives contracts, but little is known about the influence of altruism on the outcomes of tournaments. The paper analyzes the impact of physicians' altruism and motivation on the outcomes of rank-order tournaments in healthcare, where a fixed price contract on quantity is supplemented with a relative performance contract on quality. Our theoretical model forecasts crowding out of most altruistic types owing to the effect of the participation constraint. In an empirical application to the Medicare's nationwide natural experiment with a relative performance contract on quality for acute inpatient care since 2013, we observe the proof of the model's predictions. Namely, the quality dimensions, which are linked to patient's benefit, demonstrate higher deterioration among top-performing hospitals than

We use the framework of tournaments in healthcare, where a fixed price contract on quantity is supplemented with a relative performance contract on quality. Our model shows quality convergence, however, altruism may lead to quality decrease among subgroups of the high-performing agents. In testing the model's hypotheses, we focus on Medicare's hospitals with top performance and show that quality dimensions, which are not linked to patient's benefit, demonstrate lower quality decrease than other dimensions. In other words, altruism may become a reason for motivation crowding out on the healthcare market. The novelty of our empirical approach is severalfold. Firstly, we use dynamic panel data estimations to account for ``habit-formation''. The analysis excludes ``regression-to-the-mean'' effect by modelling the time-dependent long-term mean as a function of hospital characteristics. Secondly, while previous studies exploited the data for prototypes of value-based purchasing and concentrated on composite measures, we use longitudinal data sets on each quality measure of all acute-care Medicare's hospitals before and after the reform (fiscal years 2004--2016). The data are supplemented with patient case-mix, ownership, share of Medicare population and various hospital control variables, coming from: Medicare's Impact Files, Final Rules, Provider of Service Data, and Provider Utilization and Payment Data.

The English National Health Service has experienced significant increases in the percentage of hospital beds that are occupied at any point in time, from an average of 84.5% in 2006/07 to 91.4% in 2016/17. Moreover, it is not uncommon for some hospitals to reach 100% bed utilisation at certain times of the day. This is the consequence of a developing mismatch between supply and demand side factors, including a consistent decrease in hospital bed stock at a time of significant rises in the number of hospital admissions. Across OECD countries, only hospitals in Canada, Ireland, Israel and Norway are experiencing higher average bed occupancy rates. Clinical leadership has expressed concern about rises in bed occupancy rates and how they might affect the ability of hospital teams to deliver high quality care. There are particular concerns that high bed occupancy rates might lead to an increased likelihood of adverse events, challenges in securing the resources needed to diagnose and treat patients, and problems with planning for the discharge of larger numbers of patients. Despite the salience of the issue to resource-stretched hospitals, very little research has examined the implications of high bed occupancy rates on hospital behaviour and the quality of care. This is the first study to assess daily changes in bed occupancy rates and their relationship with risk-adjusted, patient discharges and their subsequent 30-day readmission rate, from April 2014 to February 2016. Hospital Episode Statistics data was used to identify all patients treated in English hospitals across the observation period, with the daily hospital-level bed occupancy rate calculated as the ratio of registered inpatients at midnight and the recorded stock of hospital beds. We modelled panel data models for each hospital and day across the observation period, using ordinary least squares estimators with hospital fixed effects to relate changes in daily bed occupancy rates to the two outcome variables. Sensitivity analysis was conducted for a range of selected patient subgroups, including patients of different ages, with different comorbidities and from varying socioeconomic backgrounds. We find that an increase in bed occupancy rate by 1% was associated with a 0.49% (p<0.001) rise in the discharge rate, and a 0.011% (p<0.001) increase in the 30-day readmission rate for discharged patients. These associations became more pronounced once bed occupancy reaches the highest tertile of the bed occupancy distribution, around 95%, with each extra 1% rise in bed occupancy associated with a 0.04% (p<0.001) rise in the readmission rate. Older patients (e.g. patients aged 81 to 90 years) and those with a greater number of comorbidities (e.g. 5 comorbidities) were less likely to be discharged at times of high bed occupancy, but if they were discharged, they had an increased risk of readmission. In this study, we find that when bed occupancy rates are high, hospitals discharge a greater proportion of their patients. However, increased bed occupancy was not associated with a substantial increase in 30-day readmission rates. It may

This paper investigates the effect of top-managers on the performance of public sector organisations using CEOs of English public hospitals as a case study. In many countries, the public sector is important not only in financing but also in delivering services such as education and healthcare. In the search for greater productivity of public sector organisations, a popular political reform model is to introduce more autonomy and market discipline in the system. An important element of these reforms tends to be a greater emphasis on the role of top managers, accompanied by the use of manager-specific compensation policies, and performance-related pay and dismissals. This approach is underpinned by the

The role of top managers for firm performance has been extensively studied in private sector organizations, starting with the seminal paper by Bertrand and Schoar (2003). The evidence on the effectiveness of top management in the public sector, however, is still relatively scarce. In particular, no evidence exists about the effectiveness of individual managers for hospital performance. We address this gap in the literature by examining the role of top managers in English public hospitals. In the late 1980s, the English government embarked on a long-standing reform programme which replaced the consensus management characterizing the National Health Service (NHS) in the 1970s with a decentralized model. In this new model, CEOs could assume a largely undisputed responsibility for the management and performance of individual public hospitals, and individual hospital boards could select and reward individual CEOs in a fully decentralized fashion. These shifts, and the frequent managerial rotations of the same CEOs across NHS hospitals that arose as a consequence of the new policies, provide an ideal setting to study whether individual managers are indeed associated with systematic differences in hospital performance. An additional appealing factor of the NHS system is the richness of the hospital performance data that we can use in the analysis, spanning a wide set of financial and clinical final outcomes, as well as intermediate outputs and operational variables.

percentile to the 75th percentile of the CEO effects in pay distribution represents a 12% increase in pay relative to mean CEO pay. These differences in evidence of the existence of a perceived difference in ability across different managers. Next, we estimate whether individual managers are associated with actual differences in hospital performance

and a variety of intermediate operational outcomes employing three different approaches. First, the fixed effects approach pioneered by Bertrand and Schoar (2003). Second, the alternative two-step method proposed by Bertrand and Schoar (2003) and third, a non-parametric approach that resembles a difference-in-difference matching estimator. Overall, we find little evidence of individual CEOs having an impact on hospital performance. Our results also raise concerns about policy approaches that rely on the contribution of transient "turnaround" top managers to improve the performance of individual hospitals.

In 2007, the Centers for Medicare and Medicaid (CMS) significantly restructured the diagnosis related group (DRG) system by moving from the CMS DRG to the Medicaid Severity (MS) DRG. The MS DRGs expanded the number of illness categories, explicitly recognizing the variation of complications present within certain conditions, and allowing for differential reimbursement depending on the severity of the case. Each DRG has a weights which reflect the average service intensity for patient with that illness. Payments are proportional to weights. After the implementation of the MS-DRG system, hospitals may choose from up to three new DRGs depending upon the severity of the diagnosis reported including either complication/co-morbidity (CC) or no complication/co-morbidity (non-CC) or major complication/comorbidity (MCC). Each of these new DRGs has a different reimbursement weight with the more complicated cases receiving higher weights and more reimbursement, the less complicated cases receive a lower weight and lower reimbursement. We explore whether hospitals, given a diagnosis, systematically code patients into the higher severity MS DRG to increase their reimbursements. In the economics literature, this is known as upcoding (e.g. Dafny 2005, Silverman and

Using the National Inpatient Survey (NIS) data from 2005-2008 and MS DRG weights from CMS, we first compare the average DRG weights using the MS DRGs to the counterfactual calculation of the average DRG weights using the old CMS DRGs. Our preliminary results indicate that patients have an increase in average DRG weight of .048 to .071. This indicates that patients are recorded as 4-6 percent sicker, under the MS-DRG methodology. However, this is only suggestive of upcoding because the distribution of severity of illness within a CMS DRG may have changed such that patients needed more services and the higher weights were appropriate. We use a look back approach applying the counterfactual calculation to data for the three years before the policy change to determine the extent to which aggregate weights persist over time. For a more formal test of upcoding, we estimate the equation below for those DRGs where we observe three categories post 2007. Weight refers to the weight associated with the assigned DRG for patient i in hospital h. Spread refers to

are hospital fixed effects (Dafney, 2005 and Barros and Braun 2017 employ a similar method).

Table 1 (not shown) presents estimates of equation 1 stratified by hospital type. The positive and significant coefficients on the spread variables are indicative of upcoding. Another concern is technological change, which may increase the service intensity of the illness, and increase the weight of the DRG. However, this would only be a threat to identification if the technological change were newly implemented in 2008 and were within the MS DRGs which had seen expansion in the policy change. To address this, we will examine DRG groups with high frequency (common ailments) and investigate technological change in these

This paper evaluates the relationship between hospital cost and quality. The Department of Health and Human Services collects quality information from hospitals and produces a number of quality measures that are publicly available through its Hospital Compare project. The purpose of providing these quality measures is to increase transparency and accountability in the healthcare system. Another area where these quality measures could be useful is in quality adjusting hospital prices in the Consumer Price Index (CPI) and the Producer Price Index (PPI). Currently, the PPI makes some use of these measures, but the methodology used may be problematic as the relationship between the quality measures and hospital costs is unknown. Once the relationship between the quality measures and costs are better understood, it will be possible to revise the existing PPI methods. This paper estimates the causal relationship between the quality measures and hospital costs. The causal relationship is identified using the instrumental variable technique of Doyle, Graves, Gruber, and Kleiner (JPE 2015) and Doyle, Graves, and Gruber (NBER working paper 2017). They develop an instrument for hospital selection based on plausibly exogenous assignment to different ambulance companies (which have different preferences for hospitals). Doyle, Graves, Gruber, and Kleiner (JPE 2015) look at whether hospitals that have higher Medicare reimbursements have better outcomes. Doyle, Graves, and Gruber (NBER working paper 2017) look at the relationship between the Hospital

This paper uses Medicare claims data for the years 2011-2015. Data on hospital admissions is contained in the Inpatient file, and ambulance billing data are contained in the Outpatient and Carrier files. The claims data are linked to hospital specific cost (hospital cost reports) and quality data (Hospital Compare). Preliminary results suggest a relationship between cost and quality but is sensitive to the cost and quality measures used. A complicating factor in the analysis is that CMS began phasing in the use of quality measures in hospital reimbursements during this period.

This paper investigates whether health insurance status has an impact on survival and admission probabilities for heart attack patients presenting through the emergency department. The paper uses a methodology from Currie, MacLeod and Van Parys (2016) to address endogeneity and evaluate whether the differences across different types of health insurance. Using data from the 2006-2010 HCUP NEDS, I show patterns in the probability of death, hospital admission and treatment patterns patients with a primary diagnosis of acute myocardial infarction before, during, and after the Great Recession.

Health economics researchers have long known insurance status has an impact on the use of health care system (Hadley 2003); however, measuring the causal effect of insurance status on treatment decisions and outcomes has been a difficult problem for economists to tackle. This paper builds upon earlier work by Doyle (2005) in examining the effects of insurance status on injury patients in the emergency room. Doyle (2005) uses a Wisconsin sample of patients reporting with injuries resulting from motor vehicle accidents. The present examination improves on Doyle’s study by broadening the dataset, controlling for overall health status and considering multiple types of health shocks. I use a nationwide dataset of emergency department visits to compare the differences between the effects of insurance status on treatment for patients with injuries from motor vehicle accidents and patients with other types of injuries while controlling for overall health status. This paper reaches three primary conclusions: (1) Patients paying out of pocket experience worse health outcomes; (2) Patients paying out of pocket receive fewer health services in response to motor vehicle accident injuries and other types of injuries; and, (3) Despite no observed differences between patients paying out of pocket and patients receiving charity care, charity care patients generally received more healthcare services and

Hospitals play an important role in the social insurance system by providing free and discounted care to the uninsured. In 2012, hospitals provided over $46 billion in uncompensated care -- almost 30% of Medicaid inpatient and outpatient spending that year. Despite the large economic cost of uncompensated care to hospitals, relatively little is known about how hospitals determine the amount of uncompensated care they provide. In this paper, I look at how nonprofit hospitals adjust their charity care policies in response to increases in public funding, state regulations, and changes in the local market for medical care. To do so, I use new data from the IRS 990

I find significant evidence that hospitals raise their charity care policies in response to changes in hospital income, but limited evidence that hospitals respond to changes in local need. Furthermore, I find that state changes in the regulation of non-profit hospitals community benefit activities have ambiguous on charity care provision, and are most effective when the attorney general has regulatory power.


In May 2013, the Centers for Medicare and Medicaid Services (CMS) began publishing charges for the 100 most frequently billed diagnosis-related groups (DRGs) for inpatients treated in approximately 3,400 U.S. hospitals. The goal was to increase transparency and bring greater market forces to bear on the steep growth in prices for hospital services. While charges are not the same as payments, the data release has several advantages over claims data. The reports are readily available online, national in scope, and the subject of media attention. Charges generally are the starting point for hospital-insurer price negotiations. High charges also have a direct impact on uninsured individuals, who are exposed to full charges, as well as patients covered out-of-network or under workers’ compensation, who generally pay a portion of full charges.

This paper examines whether the CMS price transparency initiative is performing as a policy lever in reducing the growth of hospitals prices and explores which stakeholders are the most receptive audiences. More specifically, we address supply side responsiveness by examining trends in hospital inpatient charges and demand side responsiveness by examining trends in volume and market share of inpatient services.

Using quasi-experimental designs, we estimated econometric models using charge and utilization data obtained from CMS and from the Florida and New York AHRQ State Inpatient Databases for the years 2011-2015. To examine the impact of the CMS public reports on hospital charges, we estimated difference-in differences models that compared changes in charges for the 100 reported DRGs with changes in charges for the unreported DRGs, controlling for DRG weight, geographic location, and patient factors. To examine the impact of the public reports on consumer choice, we explored shifts in volume and market shares. We selected two reported DRGs based on price sensitivity and high volume: lower extremity total joint replacement and percutaneous coronary intervention (PCI – excluding cases admitted through emergency departments). We estimated the impact of the release of CMS public reports in May 2013 on volume and market share of these conditions using quarterly observations on individual hospitals and controlling for hospital characteristics.

Results indicate that in Florida, inflation adjusted charges increased between 2013 and 2015, but were 4% lower in the 100 reported DRGs compared to other DRGs. We did not find differences in either volume or market share for total joint replacement or PCI in high-charge hospitals relative to low-charge hospitals. It appears that in Florida, hospitals were more responsive to CMS public reporting than consumers were. Estimation of models using New York data are in

In 2014, the State of Maryland placed the majority of its hospitals under all-payer global budgets for inpatient, hospital outpatient, and emergency department care. Maryland’s payment reform has garnered considerable attention from policymakers because of its unique approach to reorienting hospitals from a traditional fee-for-service payment model to a model that rewards hospitals for managing population health—including the provision of care outside of hospital settings. Despite reports from CMS and Maryland of the program’s early success in controlling hospital spending, the program's impacts on hospital and primary care use remain largely unknown. Under the structure of Maryland’s program, hospitals can lower spending by reducing hospital utilization and enhancing primary care—as policymakers had intended—or by reducing their prices to meet their budgets.

Using a quasi-experimental difference-in-differences design, we compared changes in hospital and primary care use among fee-for-service Medicare beneficiaries in Maryland vs. 27 matched out-of-state control counties from before (2009-13) to after (2014-15) Maryland's introduction of hospital global budgets. We conducted our analyses under two sets of assumptions. First, we assumed pre-intervention differences between Maryland and the control counties would have remained constant past 2014 had Maryland not implemented global budgets (the standard identifying assumption of difference-in-difference analyses). Second, we assumed differences in pre-intervention trends would have continued without the state’s payment change (differential trend assumption). In supplementary analyses, we compared our main difference-in-differences estimates to those generated from a series of placebo tests in which we iteratively re-assigned states outside of Maryland to the intervention group; using estimates generated in these tests, we assessed whether changes in Maryland consistently exceeded changes detected in

Assuming parallel trends, we estimated a differential change in Maryland of -0.47 annual hospital stays/100 beneficiaries (95% CI: -1.65,0.72; P=0.43) from the pre-intervention period (2009-13) to 2015, but assuming differential trends, we estimated a differential change in Maryland of -1.24 stays/100 beneficiaries (95% CI: -2.46,-0.02; P=0.047). Assuming parallel trends, we found a significant increase in primary care visits (+10.6 annual visits/100 beneficiaries; 4.6,16.6; P=0.001), but assuming differential trends, we found no change (-0.8 visits/100 beneficiaries; 95% CI: -10.6,9.0; P=0.87). Comparing estimates with both trend assumptions, we found no consistent changes in emergency department visits, return hospital stays, HOPD use, or post-hospitalization primary care visits associated with Maryland’s program. Differential changes in Maryland were within the distribution of changes detected in supplementary placebo analyses, suggesting that hospital and primary care utilization did not change differentially in Maryland from changes that we would have expected to see in the absence global budgets.

We did not find consistent evidence that Maryland’s hospital global budget program was associated with anticipated reductions in hospital use or increases in primary care visits among fee-for-service Medicare beneficiaries after two years. Given the challenge of changing practice patterns over the short-term, evaluations over longer periods and in younger populations—for whom care outside the hospital may be more appropriate—should be


Linda Dynan [email protected]

Dan Han [email protected]

Nicholas Benson [email protected]

For almost two decades, and with renewed intensity since the passage of the Affordable Care Act in 2010, the safety and quality of inpatient care in U.S. hospitals has been of national concern. Of particular interest to hospitals is how responsive is measured quality improvement in outcomes to investment in quality improvement. Indeed, the business case for safety and quality improvement is based on the proposition that given proper financial incentives (via more informed and selective purchasers), health care providers will naturally improve safety and quality by making the appropriate investments. But key to this proposition is that investments in quality improvement are cost effective. In this paper, we take a new and extensive look at the business case for hospital safety while, to the extent possible, avoiding some of the analytical pitfalls of earlier work such as endogenous explanatory variables and simultaneity. We use a unique panel of all general hospitals in the state of Florida between 2004 and 2015 obtained from Florida’s Center for Health Information and Policy Analysis, a department of the state’s Agency for Health Care Administration. These data include annual hospital inpatient discharge information and financial data. We estimate the effect of quality investments on patient safety measured by a composite of Agency for Healthcare Research and Quality patient safety (adverse) events. Specifically, we estimate the elasticity of patient safety with respect to several measures of quality investment using quantile regression analysis, hospital fixed effects and dynamic-panel models to address endogeneity. We find that rates of adverse events are quite inelastic with respect to hospital investment. Although we find that some kinds of hospital investment have favorable effects on safety, these effects are very small, with estimated elasticities of less than 0.1. In addition, we find evidence that policies, particularly the 2008 Centers for Medicare and Medicaid Services reimbursement-withholding policy, have stronger, but still relatively small, effects. Thus, there is little evidence

In 2010, the Affordable Care Act extended the 340B Drug Pricing Program to allow most rural hospitals to acquire outpatient drugs from manufacturers at discounted prices. Under 340B, hospitals can dispense discounted outpatient drugs to almost any patients regardless of insurance coverage. For many urban hospitals, the program enables them to extract additional profit margin from 340B discounts, as Medicare (yet) and private insurers do not condition payment rates on whether a hospital is acquiring drugs through 340B or not. But for the majority of rural hospitals, which belong to a special classification called Critical Access Hospitals (CAH), the effect of 340B is ambiguous. CAHs receive payment of 101 percent of reasonable costs from Medicare for most inpatient and outpatient services. On the one hand, 340B likely relaxes the liquidity constraints facing CAHs, allowing them to maintain a better cash flow. On the other hand, with cost reimbursement, lower drug costs introduced by 340B lead to lower Medicare payments and less profit margin per drug used. To the extent that CAHs attract more patients by making otherwise expensive drugs more accessible, they can potentially compensate for the loss in revenue. However, CAHs serve communities that tend to have older, less well-off and dwindling populations, which could limit their ability to expand drug-intensive

In this ongoing research, I quantify the effects of 340B on CAHs by examining a wide array of outcomes ranging from Medicare outpatient drug utilization to hospital financial performance. I assemble data spanning 2007 – 2013 from Medicare claims, Medicare cost reports, the American Hospital Association Annual Survey of Hospitals, and state hospital financial reports. For patient-level outcomes, my main identification strategy exploits geographic variation in outpatient market share of 340B-eligible CAHs in 2009 - measured in terms of the fraction of outpatient visits captured by these hospitals in an area - as a proxy for new exposure to 340B following the ACA. I use an event-study style difference-in-difference design that flexibly estimates the coefficient of interest for each time interval to trace out the relationship throughout the entire study period. For hospital-level financial outcomes, I use a conventional event study

This study analyzed 1,018,171 U.S. spine procedures performed during 2009 – 2015, to determine whether the surgery site - hospital inpatient v/s lower-cost alternatives, e.g., ambulatory surgery centers (ASC) and hospital

Rising costs and performance frequency place spine surgery in the highest category of hospital expenditures and make it a research interest to payers, hospitals, and the US government. Spine surgery performed in lower-cost settings (e.g., ASCs) is known to be safe and effective, and can reduce cost by as much as 65%. Private-practice spine surgeons have taken advantage of health technology improvements by partnering with private insurers to move common spine procedures, e.g., spinal decompressions or cervical fusions (ACDFs), from inpatient to ASC setting. However, the CON regulation is imposing in 25 U.S. states. CON regulation requires a non-trivial fee and approval from a state level

Using claims data from the MarketScan® databases, our study tests the hypothesis that the 25 states where the CON program remains in force have fewer spine procedures performed in the lower-cost settings. We analyzed all qualified and reimbursed spine procedures performed between 2009 and 2015 to determine whether the frequency of common spine procedures performed in lower cost settings is associated with the existence of CON regulation. A variable identifying the site of service is available in the data set and the procedures types were identified by MS-DRG or CPT® codes. Importantly, we only used MS-DRG codes without complications/comorbidities or major complications/comorbidities and we controlled for patient diagnosis that could impact the site of service decision using ICD-9 codes. We estimated a multinomial logit model with state clustered errors. Procedure data on males and

Results indicate that the presence of a CON program led to significant decreases in the frequency of ACDFs and spinal decompressions performed in both lower-cost settings. Surprisingly, the CON regulation effects analyzed separately for male and female sub-samples differ significantly and this suggests gender differences in care seeking. Estimated risk ratios showed that patients in states with a CON program are 40% less likely to have an ACDF in an ASC and 37% less likely to have a decompression in an ASC. The effect was more pronounced in HOPDs where patients were 46% less likely to have a spinal decompression and 93% less likely to have an ACDF. In states with a CON regulation, males

State CON regulation is associated with fewer spine procedures performed in lower-cost settings. Study findings have policy implications for both value-based purchasing and the design of alternative payment

Sungchul Park [email protected]

Hoda Nouri Khajavi [email protected]

Galina Besstremyannaya [email protected]

The Centers for Medicare and Medicaid Services (CMS) has phased in the Hierarchical Condition Categories (HCC) risk adjustment model during 2004-2006 to more accurately estimate capitated payments to Medicare Advantage (MA) plans to reflect each beneficiary’s health status. However, it is debatable whether the CMS-HCC model has led to strategic evolutions of risk selection. We examine the competing claims and analyze the risk selection behavior of MA plans in response to the CMS-HCC model. We find that the CMS-HCC model reduced the phenomenon that MA plans avoid high-cost beneficiaries in traditional Medicare plans, whereas it led to increased disenrollment of high-cost beneficiaries, conditional on illness severity, from MA plans. We explain this phenomenon in relation to service-level selection. First, we show that MA plans have incentives to effectuate risk selection via service-level selection, by lowering coverage levels for services that are more likely to be used by beneficiaries who could be unprofitable under the CMS-HCC model. Then, we empirically test our theoretical prediction that compared to the pre-implementation period (2001-2003), MA plans have raised copayments disproportionately more for services needed by unprofitable beneficiaries than for other services in the post-implementation period (2007-2009). The disproportionate changes in copayments led to voluntary disenrollment of beneficiaries with need for these services, who tend to incur higher expenditures than their risk-adjusted payments. We also find evidence supporting our hypothesis that those who were less satisfied with out-of-pocket costs were more likely to disenroll from MA plans. Such strategic behavior led to MA plans to save $5.2 billion in 2007-2009 by simply transferring the costs to the federal government, thereby placing significant financial burdens on the federal government. Our results provide key policy implications for CMS in moving towards a better risk adjustment model that accounts for the enrollees’ predicted risk scores while generating economic incentives for MA plans

Given the accelerating rise in U.S. public health care expenditures, Medicare has launched various pilot reimbursement models to reform health care delivery by holding health care providers financially accountable. Under Medicare’s currently dominant payment method, fee-for-service (FFS), each health care provider involved in an episode of care is reimbursed for services provided to a patient. This volume-based payment model incentivizes profit-maximizing health care providers to overuse healthcare resources regardless of the necessity or quality of care. In addition, the inefficient FFS system creates a fragmented health care delivery system, where health care providers have no incentive to coordinate during an episode of treating a health condition. The growing need for alternative reimbursement models is more recognized for certain health conditions including total joint replacement (TJR). There are numerous entities

This paper investigates the impact of the Bundled Payments for Care Improvement Model 2 (BPCI Model 2) initiative, one of the most recent and comprehensive alternative payment methods, on reducing health care cost, and reforming the coordination and quality of the care delivered to total joint replacement patients. In BPCI Model 2, participating hospitals are financially accountable for Medicare spending on hospitalization, post-acute care and all related services,

By using hospital discharge-level data from State Inpatient Databases, developed by Agency for Healthcare Research and Quality, I investigate the mechanism used by financially accountable hospitals to reduce health care utilization and aggregate expenditures over a cycle of treating TJR. I estimate a difference-in-difference model where I compare inpatient length of stay, number of procedures, hospital total charges, and disposition pattern of TJR patients of BPCI-participating hospitals to outcomes of non-participating ones before and after the bundled payment was implemented. In order to study the impact of BPCI Model 2 on the quality of care, I estimate the effect of this payment reform on probability of re-hospitalization within 30 or 90 days after hospital discharge. Moreover, I stratify the sample by payer type to investigate the broader effect of BPCI Model 2 and whether participating hospitals discriminate against TJR

My empirical analyses suggest despite a significant 15 percentage point decrease in probability of discharging patients to institutional post-acute care facilities, and one third of day reduction in hospital length of stay, the quality of care is improved and there is a significant decrease in probability of re-hospitalization. I find no cost-shifting effect on non-Medicare patients or any significant evidence that BPCI-participating hospitals use different practice patterns among TJR patients across payer types. In addition, I find no supporting evidence that the orthopedic surgeons performing surgery in BPCI-participating hospitals sort healthier patients to those entities and divert more intensive patients to non-

Public contracting with hospitals under asymmetric information about their technology provides a classic example of an agency problem, where government as a principal can achieve social optimum in terms of product's quantity and agent's efforts through nonlinear prices. Incentives contracts are targeted at increasing aggregate performance, which is observed in the experimental and empirical literature. However, mean effect hinders heterogeneity in the responses of agents who differ in their abilities. In particular, both theoretical literature and natural experiments point to deteriorating performance of the front-runners. Theoretical explanations include motivation crowding out owing to intrinsic behavior or slacking efforts in tournaments, particularly in the dynamic context. Altruistic agents, however, would be interested in a social value of their performance per se. There is limited theoretical literature on altruism in public good

The paper analyzes the impact of physicians' altruism and motivation on the outcomes of rank-order tournaments in healthcare, where a fixed price contract on quantity is supplemented with a relative performance contract on quality. Our theoretical model forecasts crowding out of most altruistic types owing to the effect of the participation constraint. In an empirical application to the Medicare's nationwide natural experiment with a relative performance contract on quality for acute inpatient care since 2013, we observe the proof of the model's predictions. Namely, the quality dimensions, which are linked to patient's benefit, demonstrate higher deterioration among top-performing hospitals than

We use the framework of tournaments in healthcare, where a fixed price contract on quantity is supplemented with a relative performance contract on quality. Our model shows quality convergence, however, altruism may lead to quality decrease among subgroups of the high-performing agents. In testing the model's hypotheses, we focus on Medicare's hospitals with top performance and show that quality dimensions, which are not linked to patient's benefit,

The novelty of our empirical approach is severalfold. Firstly, we use dynamic panel data estimations to account for ``habit-formation''. The analysis excludes ``regression-to-the-mean'' effect by modelling the time-dependent long-term mean as a function of hospital characteristics. Secondly, while previous studies exploited the data for prototypes of value-based purchasing and concentrated on composite measures, we use longitudinal data sets on each quality measure of all acute-care Medicare's hospitals before and after the reform (fiscal years 2004--2016). The data are supplemented with patient case-mix, ownership, share of Medicare population and various hospital control variables, coming from:

Rocco Friebel [email protected]

Katharina Janke [email protected]

The English National Health Service has experienced significant increases in the percentage of hospital beds that are occupied at any point in time, from an average of 84.5% in 2006/07 to 91.4% in 2016/17. Moreover, it is not uncommon for some hospitals to reach 100% bed utilisation at certain times of the day. This is the consequence of a developing mismatch between supply and demand side factors, including a consistent decrease in hospital bed stock at a time of

Clinical leadership has expressed concern about rises in bed occupancy rates and how they might affect the ability of hospital teams to deliver high quality care. There are particular concerns that high bed occupancy rates might lead to an increased likelihood of adverse events, challenges in securing the resources needed to diagnose and treat patients, and problems with planning for the discharge of larger numbers of patients. Despite the salience of the issue to resource-

This is the first study to assess daily changes in bed occupancy rates and their relationship with risk-adjusted, patient discharges and their subsequent 30-day readmission rate, from April 2014 to February 2016. Hospital Episode Statistics data was used to identify all patients treated in English hospitals across the observation period, with the daily hospital-level bed occupancy rate calculated as the ratio of registered inpatients at midnight and the recorded stock of hospital beds. We modelled panel data models for each hospital and day across the observation period, using ordinary least squares estimators with hospital fixed effects to relate changes in daily bed occupancy rates to the two outcome variables.

We find that an increase in bed occupancy rate by 1% was associated with a 0.49% (p<0.001) rise in the discharge rate, and a 0.011% (p<0.001) increase in the 30-day readmission rate for discharged patients. These associations became more pronounced once bed occupancy reaches the highest tertile of the bed occupancy distribution, around 95%, with each extra 1% rise in bed occupancy associated with a 0.04% (p<0.001) rise in the readmission rate. Older patients (e.g. patients aged 81 to 90 years) and those with a greater number of comorbidities (e.g. 5 comorbidities) were less likely to be discharged at times of high bed occupancy, but if they were discharged, they had an increased risk of readmission. In this study, we find that when bed occupancy rates are high, hospitals discharge a greater proportion of their patients. However, increased bed occupancy was not associated with a substantial increase in 30-day readmission rates. It may

This paper investigates the effect of top-managers on the performance of public sector organisations using CEOs of English public hospitals as a case study. In many countries, the public sector is important not only in financing but also in delivering services such as education and healthcare. In the search for greater productivity of public sector organisations, a popular political reform model is to introduce more autonomy and market discipline in the system. An important element of these reforms tends to be a greater emphasis on the role of top managers, accompanied by the use of manager-specific compensation policies, and performance-related pay and dismissals. This approach is underpinned by the

The role of top managers for firm performance has been extensively studied in private sector organizations, starting with the seminal paper by Bertrand and Schoar (2003). The evidence on the effectiveness of top management in the

We address this gap in the literature by examining the role of top managers in English public hospitals. In the late 1980s, the English government embarked on a long-standing reform programme which replaced the consensus management characterizing the National Health Service (NHS) in the 1970s with a decentralized model. In this new model, CEOs could assume a largely undisputed responsibility for the management and performance of individual public hospitals, and individual hospital boards could select and reward individual CEOs in a fully decentralized fashion. These shifts, and the frequent managerial rotations of the same CEOs across NHS hospitals that arose as a consequence of the new policies, provide an ideal setting to study whether individual managers are indeed associated with systematic differences in hospital performance. An additional appealing factor of the NHS system is the richness of the hospital

percentile of the CEO effects in pay distribution represents a 12% increase in pay relative to mean CEO pay. These differences in evidence of the existence of a perceived difference in ability across different managers. Next, we estimate whether individual managers are associated with actual differences in hospital performance

and a variety of intermediate operational outcomes employing three different approaches. First, the fixed effects approach pioneered by Bertrand and Schoar (2003). Second, the alternative two-step method proposed by Bertrand and Schoar (2003) and third, a non-parametric approach that resembles a difference-in-difference matching estimator. Overall, we find little evidence of individual CEOs having an impact on hospital performance. Our results also raise concerns

Amanda Cook [email protected]

Brett Matsumoto [email protected]

Rachel Childers [email protected]

Rachel Childers [email protected]

Rebecca Sachs [email protected]

In 2007, the Centers for Medicare and Medicaid (CMS) significantly restructured the diagnosis related group (DRG) system by moving from the CMS DRG to the Medicaid Severity (MS) DRG. The MS DRGs expanded the number of illness categories, explicitly recognizing the variation of complications present within certain conditions, and allowing for differential reimbursement depending on the severity of the case. Each DRG has a weights which reflect the average service intensity for patient with that illness. Payments are proportional to weights. After the implementation of the MS-DRG system, hospitals may choose from up to three new DRGs depending upon the severity of the diagnosis reported including either complication/co-morbidity (CC) or no complication/co-morbidity (non-CC) or major complication/comorbidity (MCC). Each of these new DRGs has a different reimbursement weight with the more complicated cases

We explore whether hospitals, given a diagnosis, systematically code patients into the higher severity MS DRG to increase their reimbursements. In the economics literature, this is known as upcoding (e.g. Dafny 2005, Silverman and

Using the National Inpatient Survey (NIS) data from 2005-2008 and MS DRG weights from CMS, we first compare the average DRG weights using the MS DRGs to the counterfactual calculation of the average DRG weights using the old CMS DRGs. Our preliminary results indicate that patients have an increase in average DRG weight of .048 to .071. This indicates that patients are recorded as 4-6 percent sicker, under the MS-DRG methodology. However, this is only suggestive of upcoding because the distribution of severity of illness within a CMS DRG may have changed such that patients needed more services and the higher weights were appropriate. We use a look back

For a more formal test of upcoding, we estimate the equation below for those DRGs where we observe three categories post 2007. Weight refers to the weight associated with the assigned DRG for patient i in hospital h. Spread refers to

Another concern is technological change, which may increase the service intensity of the illness, and increase the weight of the DRG. However, this would only be a threat to identification if the technological change were newly implemented in 2008 and were within the MS DRGs which had seen expansion in the policy change. To address this, we will examine DRG groups with high frequency (common ailments) and investigate technological change in these

This paper evaluates the relationship between hospital cost and quality. The Department of Health and Human Services collects quality information from hospitals and produces a number of quality measures that are publicly available through its Hospital Compare project. The purpose of providing these quality measures is to increase transparency and accountability in the healthcare system. Another area where these quality measures could be useful is in quality adjusting hospital prices in the Consumer Price Index (CPI) and the Producer Price Index (PPI). Currently, the PPI makes some use of these measures, but the methodology used may be problematic as the relationship between the quality

This paper estimates the causal relationship between the quality measures and hospital costs. The causal relationship is identified using the instrumental variable technique of Doyle, Graves, Gruber, and Kleiner (JPE 2015) and Doyle, Graves, and Gruber (NBER working paper 2017). They develop an instrument for hospital selection based on plausibly exogenous assignment to different ambulance companies (which have different preferences for hospitals). Doyle, Graves, Gruber, and Kleiner (JPE 2015) look at whether hospitals that have higher Medicare reimbursements have better outcomes. Doyle, Graves, and Gruber (NBER working paper 2017) look at the relationship between the Hospital

This paper uses Medicare claims data for the years 2011-2015. Data on hospital admissions is contained in the Inpatient file, and ambulance billing data are contained in the Outpatient and Carrier files. The claims data are linked to hospital specific cost (hospital cost reports) and quality data (Hospital Compare). Preliminary results suggest a relationship between cost and quality but is sensitive to the cost and quality measures used. A complicating factor in the

This paper investigates whether health insurance status has an impact on survival and admission probabilities for heart attack patients presenting through the emergency department. The paper uses a methodology from Currie, MacLeod and Van Parys (2016) to address endogeneity and evaluate whether the differences across different types of health insurance. Using data from the 2006-2010 HCUP NEDS, I show patterns in the probability of death, hospital admission and

Health economics researchers have long known insurance status has an impact on the use of health care system (Hadley 2003); however, measuring the causal effect of insurance status on treatment decisions and outcomes has been a difficult problem for economists to tackle. This paper builds upon earlier work by Doyle (2005) in examining the effects of insurance status on injury patients in the emergency room. Doyle (2005) uses a Wisconsin sample of patients reporting with injuries resulting from motor vehicle accidents. The present examination improves on Doyle’s study by broadening the dataset, controlling for overall health status and considering multiple types of health shocks. I use a nationwide dataset of emergency department visits to compare the differences between the effects of insurance status on treatment for patients with injuries from motor vehicle accidents and patients with other types of injuries while controlling for overall health status. This paper reaches three primary conclusions: (1) Patients paying out of pocket experience worse health outcomes; (2) Patients paying out of pocket receive fewer health services in response to motor vehicle accident injuries and other types of injuries; and, (3) Despite no observed differences between patients paying out of pocket and patients receiving charity care, charity care patients generally received more healthcare services and

Hospitals play an important role in the social insurance system by providing free and discounted care to the uninsured. In 2012, hospitals provided over $46 billion in uncompensated care -- almost 30% of Medicaid inpatient and outpatient spending that year. Despite the large economic cost of uncompensated care to hospitals, relatively little is known about how hospitals determine the amount of uncompensated care they provide. In this paper, I look at how nonprofit hospitals adjust their charity care policies in response to increases in public funding, state regulations, and changes in the local market for medical care. To do so, I use new data from the IRS 990

I find significant evidence that hospitals raise their charity care policies in response to changes in hospital income, but limited evidence that hospitals respond to changes in local need. Furthermore, I find that state changes in the regulation

Souvik Banerjee [email protected]

Kathleen Carey [email protected]


In May 2013, the Centers for Medicare and Medicaid Services (CMS) began publishing charges for the 100 most frequently billed diagnosis-related groups (DRGs) for inpatients treated in approximately 3,400 U.S. hospitals. The goal was to increase transparency and bring greater market forces to bear on the steep growth in prices for hospital services. While charges are not the same as payments, the data release has several advantages over claims data. The reports are readily available online, national in scope, and the subject of media attention. Charges generally are the starting point for hospital-insurer price negotiations. High charges also have a direct impact on uninsured individuals, who are

This paper examines whether the CMS price transparency initiative is performing as a policy lever in reducing the growth of hospitals prices and explores which stakeholders are the most receptive audiences. More specifically, we address

Using quasi-experimental designs, we estimated econometric models using charge and utilization data obtained from CMS and from the Florida and New York AHRQ State Inpatient Databases for the years 2011-2015. To examine the impact of the CMS public reports on hospital charges, we estimated difference-in differences models that compared changes in charges for the 100 reported DRGs with changes in charges for the unreported DRGs, controlling for DRG weight, geographic location, and patient factors. To examine the impact of the public reports on consumer choice, we explored shifts in volume and market shares. We selected two reported DRGs based on price sensitivity and high volume: lower extremity total joint replacement and percutaneous coronary intervention (PCI – excluding cases admitted through emergency departments). We estimated the impact of the release of CMS public reports in May 2013 on

Results indicate that in Florida, inflation adjusted charges increased between 2013 and 2015, but were 4% lower in the 100 reported DRGs compared to other DRGs. We did not find differences in either volume or market share for total joint replacement or PCI in high-charge hospitals relative to low-charge hospitals. It appears that in Florida, hospitals were more responsive to CMS public reporting than consumers were. Estimation of models using New York data are in


In 2014, the State of Maryland placed the majority of its hospitals under all-payer global budgets for inpatient, hospital outpatient, and emergency department care. Maryland’s payment reform has garnered considerable attention from policymakers because of its unique approach to reorienting hospitals from a traditional fee-for-service payment model to a model that rewards hospitals for managing population health—including the provision of care outside of hospital settings. Despite reports from CMS and Maryland of the program’s early success in controlling hospital spending, the program's impacts on hospital and primary care use remain largely unknown. Under the structure of Maryland’s program, hospitals can lower spending by reducing hospital utilization and enhancing primary care—as policymakers had intended—or by reducing their prices to meet their budgets.

Using a quasi-experimental difference-in-differences design, we compared changes in hospital and primary care use among fee-for-service Medicare beneficiaries in Maryland vs. 27 matched out-of-state control counties from before (2009-13) to after (2014-15) Maryland's introduction of hospital global budgets. We conducted our analyses under two sets of assumptions. First, we assumed pre-intervention differences between Maryland and the control counties would have remained constant past 2014 had Maryland not implemented global budgets (the standard identifying assumption of difference-in-difference analyses). Second, we assumed differences in pre-intervention trends would have continued without the state’s payment change (differential trend assumption). In supplementary analyses, we compared our main difference-in-differences estimates to those generated from a series of placebo tests in which we iteratively re-assigned states outside of Maryland to the intervention group; using estimates generated in these tests, we assessed whether changes in Maryland consistently exceeded changes detected in

Assuming parallel trends, we estimated a differential change in Maryland of -0.47 annual hospital stays/100 beneficiaries (95% CI: -1.65,0.72; P=0.43) from the pre-intervention period (2009-13) to 2015, but assuming differential trends, we estimated a differential change in Maryland of -1.24 stays/100 beneficiaries (95% CI: -2.46,-0.02; P=0.047). Assuming parallel trends, we found a significant increase in primary care visits (+10.6 annual visits/100 beneficiaries; 4.6,16.6; P=0.001), but assuming differential trends, we found no change (-0.8 visits/100 beneficiaries; 95% CI: -10.6,9.0; P=0.87). Comparing estimates with both trend assumptions, we found no consistent changes in emergency department visits, return hospital stays, HOPD use, or post-hospitalization primary care visits associated with Maryland’s program. Differential changes in Maryland were within the distribution of changes detected in supplementary placebo analyses, suggesting that hospital and primary care utilization did not change differentially in Maryland from changes that we would have expected to see in the absence global budgets.

We did not find consistent evidence that Maryland’s hospital global budget program was associated with anticipated reductions in hospital use or increases in primary care visits among fee-for-service Medicare beneficiaries after two years. Given the challenge of changing practice patterns over the short-term, evaluations over longer periods and in younger populations—for whom care outside the hospital may be more appropriate—should be


Northern Kentucky University Richard Smith

RAND Corporation

University of Memphis Albert Okunade; E. George, PhD; Cyril Chang

University of Washington Fahad Khalil; Norma Coe; Anirban Basu

Graduate Center, City University of New York

New Economic School, CEFIR

The Health Foundation

Lancaster University Carol Propper; Raffaella Sadun

Bowling Green University Susan Averett

Bureau of Labor Statistics

Presbyterian College

Presbyterian College

Harvard Kennedy School

Boston University School of Medicine

Boston University School of Public Health Avi Dor

Michael Paasche-Orlow; Danny McCormick; Amresh Hanchate; Meng-Yun Lin


Ateev Mehrotra; Michael Chernew; Laura Hatfield; J. McWilliams

Complete

Complete

Complete

Complete

Complete

Complete

Complete

Complete

Complete

Complete

Complete

Complete

Complete

Complete

Complete

Complete


Long Term Care, Aging & Demography


Long Term Care, Aging & Demography Medicare Advantage and Nursing Facility Quality

Health, Longevity, and Welfare Inequality of the Elderly

Not all cognitive skills are the same: The role of fluid and crystallized intelligence on health preventive behavior among the elderly



Competition and Pricing Behavior in Long-Term Care Markets: Evidence from the Market for Assistance in Daily Housekeeping Activities

High-cost inpatient admissions among elderly patients with cancer




The effect of co-payments in long term care on the distribution of income and risk.

A distance too far? The implications of failing to account for spatial dependencies when using distances as an instrument

Is Holding Political Office a Blessing or a Curse? Evidence from Lifespans of Candidates in Gubernatorial Races 1946-2000




Does post-acute care improve patient outcomes? A comparison of skilled nursing facilities and home health care

Substitution between health and social care: Evidence from England

Nursing home chains and practice patterns: evidence under the Medicare Prospective Payment System



An evaluation of a new hospice delivery model for Medicare beneficiaries

Ownership, price mark ups and demand elasticities in the nursing home market




Relative Impacts of Informal and Formal care on Health Outcomes for People with Dementia

Inequality of Health in Old Age: The Role of Early Childhood Circumstances

Family-Provided Old-Age Support and Health Shocks: Evidence from Senior Chinese Households



Public Spending on Acute and Long-Term Care for Alzheimer’s and Related Dementias

Late-life Disability, Homeownership, Wealth and Mortality

Abstract

We provide a framework to understand the distribution of individual well-being and its change over time with an application to the U.S. elderly population. We use data from the Health and Retirement Study to estimate life-cycle dynamics and simulate paths for consumption, leisure, health, and mortality starting at age sixty for every individual. We use an expected utility framework and the simulated profiles to construct measures of welfare distribution. Our analysis suggests substantial variation in welfare across individuals driven foremost by inequality in health and mortality followed by consumption. Not accounting for the effect of health on lifetime utility results in over-predicting the relative welfare for those in the bottom end of the distribution and under-predicting for those at the top. Elderly welfare inequality has increased over time due to growing gaps in consumption, health, and mortality. Cross-sectional income and consumption under-estimate aggregate welfare inequality and are only modestly correlated with elderly welfare at the individual level. Cross-sectional health is a better indicator of individual well-being rank than income or consumption. Our findings can be summarized as follows: 1. There is substantial variation in the ex-ante welfare of individuals at age sixty. The Gini coefficient for consumption-equivalent welfare in our benchmark cohort is 0.66. Those at the ninetieth percentile of the distribution have 23 times higher welfare than those at the tenth percentile. 2.Health differences are crucial for understanding the overall distribution of elderly welfare. Excluding the utility cost of poor health and morbidities lowers the welfare Gini coefficient by 23%. This is driven by a positive correlation between health, consumption, and mortality. 3. The largest drivers of welfare inequality are health and mortality gaps followed by gaps in consumption. Differences in leisure play a comparatively minor role. 4. Welfare inequality among the elderly has increased over time due to growing gaps in consumption, health, and mortality. Compared to the cohort of individuals reaching age sixty between 1992-2001, the welfare Gini rose 9% for those reaching sixty between 2002-07 and 22% for those reaching between 2008-14. 5. Ignoring dynamic uncertainty and the persistence in outcomes over the life-cycle greatly underestimates welfare inequality. The Gini of age sixty flow utility is only 70% of that based on our dynamic welfare measure. A key implication of our results is that cross-sectional distributions of income/consumption underestimate aggregate welfare inequality. This occurs for two primary reasons. First, cross-sectional measures ignore dynamic uncertainty and the persistence of inequality over life. Second, there is a positive correlation between health and consumption. However, even in cases where economic outcomes provide a reasonable approximation to aggregate welfare inequality, our results suggest they may still provide a poor ranking of individual well-being. For example, the rank correlation between consumption and welfare is a relatively modest 0.56 for our benchmark cohort. Moreover, we find cross-sectional health utility at age sixty to be a better predictor of remaining lifetime welfare rank, despite the fact that it drastically underestimates aggregate welfare inequality.

Previous studies have shown a positive association between cognitive skills and preventive health behavior. Cognitive skills have two dierent components: crystallized intelligence (the ability to use learned knowledge), and fluid intelligence (the ability to solve new problems). The role that each component plays in connection with health behavior has not yet been fully explored. During a person's elderly years, these two components decline at different rates. As a consequence, an elderly individual might rely more heavily on one of these components so as to compensate for the decline of the other, allowing him or her to maintain functioning capacity and procure healthy behaviors. We develop a theoretical model to describe the production of health of an elderly person who uses consumption goods and the two cognition components as inputs for the production of health behaviors. We test the implications of our model using the English Longitudinal Study of Aging (ELSA) for the years 2004, 2008, and 2012. We conduct the empirical analysis by first using the entire sample elderly population aged 50 and older. We then look at a subsample of these individuals, focusing on those who have diabetes and hypertension. We construct a measure of preventive behavior using individuals' self-reported behavior, as well as biomarker information. We find that memory, a proxy for crystallized intelligence, has a larger and significant impact on the production of health behavior than numeracy, a proxy for fluid intelligence. Similar results occur for seniors with diabetes and hypertension. Insights from this analysis provide evidence that can be used to improve clinical treatment and better target health interventions directed at improving the health of the elderly.

Background: The percent of nursing home patients covered by MA nearly doubled from 2000 to 2013. There is evidence that highly integrated nursing home-hospital linkages have yielded better outcomes for patients, but little is known if Medicare Advantage plans establish special relationships with high quality facilities. The objective of this study is to examine the relationship between Medicare Advantage enrollment and the quality of nursing facilities at the county level.

Methods: Data sources include 2013-2015 Medicare Advantage payment and enrollment files and Nursing Home Compare Star-Rating files (as a measure of quality) from the Center for Medicare and Medicaid Services. I first run ordinary least squares linear regression testing for the association between Medicare Advantage penetration and facility quality at the county-level. Because where managed care organizations operate is endogenous to patient risk profiles, the quality of facilities as measured by clinical outcomes (overall star-rating) may be biased by favorable selection of healthier Medicare Advantage patients. I then instrument MA penetration with county-level payment rates—an exogenous policy shock that influences Medicare Advantage enrollment to test for the association on quality. I also test the association on structural quality measure (staffing ratings) that may be less affected by patient risk profiles. I control for the number of facilities for each county in all models. Robust standard errors are clustered at the county level.

Results: 2,847 counties were included in the sample from 2013-2015. The average MA penetration was 32.8% (SD=15.2%) and the average MA payment was $757 (sd=$64) in the sample. The average number of nursing facilities in a county was 5.4 (sd=9.7). The average proportion of overall 5-star (high quality) facilities in a county was 22% (sd=29%), overall 1-star (low quality) was 13% (SD=24%), staffing 5-star was 10% (sd=21%), and staffing 1-star was 12% (sd=25%). OLS regressions indicate that a one-percentage increase in MA penetration is associated with increase in the proportion of overall low quality facilities (0.04%, p<.1) and low quality staffing facilities (0.19%, p<.01), and a decrease in the proportion of high quality staffing facilities (-0.15%, p<.01). In contrast, IV regressions suggest that a one-percentage increase in MA penetration is associated with decrease in the proportion of overall low quality facilities (-0.86%, p<.01) and low quality staffing facilities (-1.3%, p<.01). First-stage regressions suggest strong instrument (F=45.11, p<.001).

Conclusion: Results suggest that MA plans may be operating in counties where there are also lower proportion of low quality SNFs. Further research is needed to study if MA plans concentrate their enrollees in specific (i.e. high quality) facilities to provide care. It is important to identify possible inequities in nursing home use and care as a consequence of where MA plans operate.

This paper aims to expand the limited knowledge about how competitive forces shape the pricing behavior on markets for long-term care services, using a unique data set on the Dutch market for assistance in daily housekeeping activities (ADHA). This type of long term care services has been decentralized to municipalities as of 2007. As many municipalities chose not to regulate prices in their procurement process, we have market-determined price data for more than 150 local markets (of just over 400 Dutch municipalities in total) for the period 2009-2013. Our identification strategy relies on the exogenous variation in the market shares in the first month after decentralization. This period can be seen as a mere transition phase, as existing users retained the right to receive ADHA from the very same provider as before the decentralization. Market shares in January ’07 are thus exogenous to any market developments in the years thereafter. Moreover, the January ’07 market share is a strong instrument as it turns out to be substantially correlated with market shares in later years, despite the relatively high entry rates during this period. Focusing on the most common form of ADHA (‘basic ADHA’), we obtain that a provider’s market share has a positive and significant effect on the price received. The average market leader in a municipality – possessing a 65 percent market share – secures a price that is 2.1 percent above the price of an atomistic provider. Translating this result to the overall market level tells us that the average price on the market decreases by 0.5 percent as the market becomes 10 percent less concentrated. When viewed in light of other studies on market power in care markets, this effect is considered small to moderate. Our study thus shows that high degrees of market concentration do not hinder competitive pressures on suppliers’ pricing behavior per se. A plausible explanation for this outcome is that incumbents face a serious threat of entry by newcomers, an idea that is backed by the observation that new providers enter relatively easy into the local markets. This also indicates that in order to stimulate competition on health care and long-term markets, lowering entry barriers might be more important than bringing the level of market concentration below some threshold value. Zooming in on our main outcome, we show that the small but significant effect of market size on price is merely driven by the pricing behavior of for-profit providers: while for this type of provider the average market leader receives a prices that exceeds the price of an atomistic supplier by 2.7 percent, there is no systematic relation between size and price for non-profit organizations. From this, we cannot conclude that the entry of for-profit suppliers has led to higher prices, as for-profits may be more efficient than non-profits. Still, the result indicates that the extent to which market concentration influences price formation may very well depend on the ownership composition at the supply side.

Background: Inpatient costs account for one-third of overall health care spending in the United States. Furthermore, the majority of health care costs are driven by patients who are older and have chronic conditions such as cancer. In order to improve value in care, it is thus important to examine high-cost elderly cancer patients and identify opportunities to effectively manage their health care needs. Objective: Examine characteristics of high-cost elderly cancer patients and compare with lower cost elderly cancer patients in hospital settings. Methods: We used the 2014 National Inpatient Sample data, which is an all-payer sample of hospital discharges and inpatient stays in the U.S. We identified 574,367 inpatient visits for individuals aged 65 years and older who had a cancer diagnosis. High-cost visits were defined as visits with costs at or above the 90th percentile (n= 57,437). Lower-cost visits were those with cost below the 90th percentile (n= 516,930). We examined patients’ sociodemographic characteristics, including gender, age at visit, race and ethnicity, primary payer, and median household income at the zip code level. In addition, we examined clinical characteristics, which included 29 Elixhauser comorbidities that were not related to the primary diagnosis combined into a count (0, 1-2, 3-4, ≥5), type of procedure (major or minor), number of procedures, and chemotherapy use while hospitalized. Hospital ownership (public, private), hospital bed size (small/medium, large), location (rural and urban non-teaching, urban teaching), and hospital region (Northeast, Midwest, South, West) were also evaluated. Overall median cost and median cost stratified by cost groups were described with box-plots. Differences in median costs between groups were examined using the Mann–Whitney test. Logistic regression was estimated to identify characteristics associated with being in the high-cost vs. low-cost group. Results: The overall median cost of hospital visits for elderly cancer patients was $9,162 (IQR: 5,347-15,947). The median cost in the high-cost group was $38,194 (IQR: 31,405-51,802), nearly five times the median cost of the low-cost group ($8,257, IQR: 5,032-13,335). Those in the high-cost group were more likely than their low-cost counterparts to have 5 or more comorbidities (38.4% vs.26.2%, p<0.001). They were also significantly more likely to receive major procedures (67.1% vs. 24.3%, p<0.001) as well as a greater number of procedures. In our adjusted model, older patients, compared with younger, and females, compared with males, were more likely to be in the high-cost group. Compared to those with no comorbidities, those with 5 or more comorbidities were 5 times more likely to be in the high-cost group (OR=5.84, 95%CI: 5.37-6.36). Those with 2 or more procedures were 15 times more likely to be in the high-cost group than those with no procedures (OR=15.06, 95%CI: 13.63-16.63). Those who received chemotherapy were also more likely to be in the high-cost group (OR=4.09, 95%CI: 3.78-4.41). Conclusion: High-cost elderly cancer patients who were admitted to hospitals in the U.S. had more comorbidities and received more intensive care than their lower cost counterparts. These features are important for determining strategies to effectively manage elderly cancer patients in inpatient hospital settings.

Motivation: In the Netherlands, a country with one of the most extensive and expensive public long term care systems, concerns about rising costs have led the government to increase the level of co-copayments. An interesting feature of these co-payments is that they are income- and wealth-dependent. This dependency enables the fine-tuning of the financial impact of co-payments across income- and wealth groups, but it also distorts saving and annuitization decisions of individuals. We use a unique dataset to estimate synthetic lifecycle paths of long term care spending over the lifecycle. We analyze how different types of co-payments affect consumption and saving, and how they redistribute income across income groups. We also consider the effects of the co-payment schemes on the optimal annuitization share of pension wealth. The financial wealth of Dutch elderly consists largely of annuitized pension wealth. In the case of co-payments, full annuitization might not be optimal, as people want to hold some of their wealth in cash to finance their health care costs. Moreover, differences in the co-payment rates for (pension) income and financial wealth might distort the annuitization decision. Methods: Modeling long term care expenditures over the lifecycle is challenging because of their very uneven distribution and limited availability of long panel data. We use a semi-parametric nearest-neighbor approach to estimate lifecycle paths of long term care spending. We use extensive administrative data that includes information on long term care spending, household status, income, and wealth for the entire Dutch population. The resulting lifecycle paths have a similar distribution, in terms of income, initial wealth, and long term care costs as the Dutch population. The estimated paths are inputs in a stochastic lifecycle decision model for retirees. This model determines optimal consumption and saving behavior of elderly for different levels of initial wealth and pensions, taking into account their financial risk. We use the model to analyze the effects of different forms of income and wealth dependent co-payments on average consumption and welfare (certainty equivalent consumption) across income groups. Findings: We find that, compared to a fully premium financed system, a fixed co-payment of 25% of care costs decreases average lifetime consumption of the elderly with the lowest financial means by 11 %, and welfare (certainty equivalent consumption) by 16 %. For the elderly with the highest financial means, welfare loss is only 2 %. An income dependent co-payment, raising the same amount of revenues as the fixed one, leads to a much smaller loss in consumption for the poor (3 %), while increasing the loss for the richest to 3.5 %. An income- and wealth-dependent co-payment reduces welfare loss for the poorest even further. For the richest, welfare loss is equal to the loss in the income-dependent case: holding financial assets becomes less attractive, leading to an increase in average lifetime consumption, while at the same time decreasing protection against care costs, leading to a welfare loss. Co-payments decrease the optimal annuitization rate. The annuitziation rate is lowest when co-payments only depend on (pension) income.

Background: Methodology: Economists often use distances as instruments to examine the causal impact of an event on an outcome. For example, distance to the nearest hospital has been used to instrument hospital utilisation. However, often little care is given to the spatial dependencies within the data, in terms of the outcome, the endogenous variable and the instrument. Application: Policy makers are promoting participation in community assets as a way to improve quality of life and reduce demand on healthcare services. These community assets are groups and facilities that facilitate community cohesion, produce social capital and reduce loneliness. In an earlier study, we showed that participants had better health, but this relied on correlation analysis, and we concluded that more causal investigation is required. Aim: To examine the effects of failing to account for spatial dependencies by considering a particular empirical application; the relationship between community asset participation and health outcomes. Data: We collected a bespoke dataset containing information on individuals aged 65 years and older with a chronic condition (N=3,470). We estimated the impact of community asset participation on three outcome measures: health-related quality-of-life (EuroQol-5D-5L); the costs of three types of health care utilisation; and the net-benefits of participation using a range of threshold values for a Quality-Adjusted Life Year (QALY). Respondents were asked to report participation in up to 20 different types of community assets in the past six months. We obtained the geo-location of all community assets (as defined by the Localism Act, 2011) and created a range of potential instruments based on the number of assets within given distances of each individual’s place of residence and the minimum distance to the nearest asset. Methods: To account for the potentially endogenous nature of community asset participation on these outcomes, we used both simple OLS and two-stage models, where the first stage used distance to nearest asset as an instrument. We used the Bayesian Information Criterion to select the preferred instrument specification. We repeated the analyses with and without accounting for the spatial dependencies within the data. Results: We found that participation in community assets significantly increased HRQoL and led to a positive net-benefit. The standard OLS and spatial OLS results were very similar in magnitude, but there was evidence of endogeneity hence the two-stage results were preferred. We further showed in the two-stage models that failing to account for the spatial dependence of individuals led to a markedly higher point estimates which were less precisely estimated. The coefficient on the spatial parameter was highly significant. Discussion: We found that not accounting for spatial dependencies within data can have quite large effects on the point estimate. We recommend that, where possible, the spatial nature of the data is accounted for, or is at least tested for. In the empirical application, we showed that participation in community assets is associated with substantially higher health-related quality-of-life but is not associated with lower healthcare costs. The social value of developing community assets is potentially substantial.

In this paper, we empirically investigate whether winning a political office influences health of the candidates. Our proxy for health is the candidate’s lifespan. We postulate that there are two separate mechanisms through which winning an election affects lifespan of an individual: (i) Wealth Effect: winning an election increases wealth, which in turn increases their longevity (ii) Stress Effect: decrease in longevity due to the stress of holding a political office. We construct our data set using information from candidates who ran in the gubernatorial elections in the US between 1946 and 2000. Specifically, we obtain birth and death dates of winners and runners-up in addition to their other personal attributes, such as their level of education and their experience as a politician. We identify the impact of winning an election by comparing the lifespans of the winners versus the runners-up in close elections. In these close races where the winner’s margin of victory is small, winning the election is arguably random. That is, the winner could have easily lost (and the runner-up could win) if only a small share of voters did not cast a vote for the winner. Our results indicate that the observable characteristics of the winners versus the runners-up are similar on average in elections in which winners’ margins of victory are small. This finding provides support for the randomness of the treatment. We find that winners of the gubernatorial races live about 4 years longer than the runners-up. Our back-of-the-envelope calculations suggest that the increase in wealth due to winning an election increases lifespan by about 7 years. The stress effect accounts for a 3 year reduction in lifespan.

The use of post-acute care has grown substantially over the past few decades. Nearly 40% of Medicare beneficiaries receive post-acute care after a hospital discharge, and most of those go either to a skilled nursing facility (SNF) or home with care from a home health agency (HHA). In 2015, Medicare spent over $60 billion on post-acute care, of which $48 billion went to SNF and HHA. Despite the proliferation of post-acute care, it is uncertain whether post-acute care benefits patients or whether the choice of specific post-acute care setting matters (i.e. choosing SNF versus HHA). Indeed, the use of post-acute care varies significantly across the country, suggesting substantial uncertainty about its value to patients. In this paper, we investigate the impact of discharge to SNF versus HHA on patient outcomes. We use data from 2010-2014 on all Medicare fee-for-service beneficiaries who are discharged from the hospital and receive post-acute care in either SNF or HHA. We estimate the effect of post-acute care setting on the following patient-level outcomes: death within 30 days of hospital discharge, readmission within 30 days of hospital discharge, successful discharge to the community, and improvement in functional status during the post-acute care episode. To address the endogeneity of treatment choice, we use an instrumental variables approach, using as an instrument the differential distance between the beneficiary’s home ZIP code and the closest HHA and the closest SNF. The instrument passes standard tests of first-stage strength. In all regressions, we include measures of patient case mix, diagnosis related groups, year fixed effects, and hospital fixed effects. Using ordinary least squares regression, we find substantial differences in patient outcomes by discharge setting. Compared to patients discharged to SNF, patients discharged to home health have lower readmission and death rates (by 2.5 and 4.7 percentage points respectively), are much more likely to be successfully discharged to the community (by 25 percentage points) and experience improvement in functional status while in post-acute care (by 54.3 percentage points). These findings are consistent with selection; healthier patients are more likely to be discharged with home health. In the instrumental variable specifications which account for this selection bias, these results change. Patients discharged to home health are still more likely to be successfully discharged to the community, although the effect size is about half (13 percentage points). However, patients discharged to home health are also more likely to be readmitted to the hospital (by 5.5 percentage points). They are no more likely to die. The difference in functional status improvement favored home health but was not statistically different from zero. These preliminary results suggest there are important tradeoffs between home health and SNF care for patients needing post-acute care. While current policies may incentivize the use of lower-intensity settings (such as home health care) for patients needing post-acute care, lower intensity settings may have adverse outcomes that need to be taken into account and balanced against the lower cost of using home health.

Many developed countries face growing demographic pressures on their health and long-term care budgets. Improving the efficiency of these services has therefore become an important policy priority across the world. This paper examines the impact of reductions to public spending on adult social (long-term) care on the use of public hospitals in England in order to quantify the extent of substitution between the two types of publicly funded care. It exploits large reductions in public spending on adult social care in England between 2009 and 2015, a period when social care spending fell by 37% as part of widespread government austerity measures, but where public spending on health care was protected. The institutional features of the public health and social care system in England mean that local governments retain the responsibility to fund and organise public services for their local population. As a result, there was considerable geographical variation in the cuts to adult social care spending over the period. We exploit this variation to identify the impact of public funding for social care for individuals aged 65 and above on their use of public hospitals. We control for permanent differences in the use of hospitals across local authorities with the inclusion of area fixed effects, in addition to a rich set of local area characteristics to control for time-varying needs for hospital services. Our results indicate a small but statistically significant, negative impact of public social care spending on a number of measures of public hospital use. These include visits to emergency departments and admissions for inpatient care. The estimates imply an additional spend of $23 million on emergency treatment in 2015-16 relative to 2009-10 as a result of cuts to social care spending. When examining the effects by age, the estimates indicate that the magnitudes of the effects are largest for older individuals. However, the estimates indicate no statistically significant relationship between social care spending and delayed exits from the medical system, either as measured by official delayed discharge statistics or by the average length of stay recorded in administrative patient records. These results have important implications for policy. They suggest that the recent cuts to social care spending have led to a modest increase in public hospital use among the older population. This means that social care spending has a small fiscal externality on health spending. As a result, attempts to reduce overall public spending on health and social care have been less effective than appears when looking only at the reduction in social care spending. However, the results suggest that the cuts have not been the major driving force behind recent increases in emergency department attendances or the number of delayed discharges from the medical system.

In 1998, Medicare changed its payment method for post-acute care provided by skilled nursing facilities (SNFs) from a cost-based system to a per diem Prospective Payment System (PPS), with a goal to control increasingly high Medicare SNF expenditures. Under the PPS, the per diem rates are primarily based on therapy minutes, creating an incentive for SNFs to provide high-intensity therapy services. It is well documented that SNFs have exploited this reimbursement method by over-providing these therapy services to Medicare beneficiaries, regardless of their actual clinical needs. Accounting for more than 50% of the SNFs in the United States, corporate chains can affect affiliated facilities’ practice patterns. This paper aims to examine whether the selection of therapy treatment levels for SNF residents varies by chain ownership. Using a Difference-in-Differences model to compare independent and chain-acquired SNFs during the period from 2003 to 2009, I find that chain acquisition of independent SNFs are associated with about 2.92 percentage points increase in the proportion of residents with the highest therapy treatment level. A more dynamic model suggests that the chain effects on this aggressive billing practice may last for several years after the acquisition. In addition, I find that the main effects are mostly due to acquisitions by large chains and for-profit chains. However, the increase in treatment intensity among chain-affiliated SNFs does not lead to shorter length of stay. Overall, the findings suggest that government agencies should consider chain affiliation when monitoring Medicare reimbursements among nursing homes.

Background: Created by the Centers for Medicare and Medicaid Innovation (CMMI), the Medicare Care Choices Model (MCCM) extends the benefits of hospice services to Medicare beneficiaries wanting to concurrently receive curative treatment. The MCCM aims to increase access, test a new payment model and improve quality of life and patient/family satisfaction. 141 hospice providers were selected by an expert panel in hospice and model implementation on the basis of being diverse in geographic area and having demonstrated experience of coordination and shared decision-making with beneficiaries and their families. Although hospices were positively selected into the program, CMMI randomly assigned half of the providers to begin in January 2016 (Phase 1) and the other half in January 2018 (Phase 2) and will continue until 2020. Objective: To examine the impact of the MCCM on quality of hospice care. Data: Demographics of providers in 2014 and 2015 from Hospice Utilization and Payment Public Use File, 2016 quality ratings from Hospice Compare, 2010-2017 deficiencies from Centers for Medicare and Medicaid Services Quality, Certification and Oversight Reports, and identification of providers participating in the MCCM from CMMI. Methods: This study employs two methods to estimate a causal effect of the MCCM on two measures of quality: quality ratings and deficiencies. First, we compare the Phase 1 hospice providers’ baseline characteristics (volume, geographic location, accreditation, patient characteristics) to the Phase 2 characteristics in order to establish that the randomization was sufficient. Next, we estimate the impact of the MCCM on seven Hospice Compare quality ratings using a generalized linear model (GLM) with a Gamma distribution and log link function, due to the skewed distribution of the ratings. Finally, we employ a difference-in-differences approach, which compares the deficiencies from inspection reports of Phase 1 providers (treatment) after the MCCM began to deficiencies before the MCCM was started and to Phase 2 providers (control). Results: The Phase 1 hospices were comparable in baseline characteristics to Phase 2 hospices. Compared to all hospices in the United States, they were larger and more non-profit. The GLM model found no statistically significant differences between Phase 1 and Phase 2 providers, with coefficient magnitudes close to zero for all seven quality ratings. Graphically examining the deficiencies per hospice provider surveyed from 2010-2016, the parallel trends assumption holds between Phase 1 and Phase 2 providers. Phase 1 providers had on average 1.07 deficiencies in the pre-period and 1.17 in the post-period for an increase of 0.10 from the pre- to post-period. Phase 2 providers had on average 0.82 deficiencies in the pre-period and 0.92 in the post-period, for an increase in 0.12. The difference-in-differences model found no statistically significant difference in deficiencies between Phase 1 and Phase 2 providers, pre- and post-2016. Conclusion: After one year of implementation, there is no impact of the MCCM on quality of hospice. Further information on the number of program enrollees and a longer time period of follow-up are necessary to estimating the impact of the MCCM on Medicare beneficiaries’ quality of end-of-life care.

Background: Economic theory provides a useful framework for analyzing nursing home (NH) ownership, competition and signals about quality to consumers. NHs serve both private pay and Medicaid consumers, but Medicaid rates are set by the state so Medicaid demand is perfectly elastic. However, NHs can set private pay rates over Medicaid rates depending on the market structure. For-profit and corporate chain-owned NHs aim to minimize costs and set prices in order to maximize profits, often signaling low quality. While nonprofit and independent NHs have other objectives, such as altruism, which signal high quality. Despite the conflicting profit motives for nonprofit chain NHs, about 13% of the market, no research has examined competition relative to independent nonprofit and for-profit NHs. Objective: This study will examine competition in Connecticut’s NH market by estimating the association of ownership on price mark ups and private pay residual demand elasticity. Data: Data include private pay rates from Cost of Long-Term Care in Connecticut reports, Medicaid rates from Connecticut Department of Social Services, facility characteristics from Online Survey and Certification Reporting System and quality ratings from Nursing Home Compare. Methods: To assess market power, we calculate the percent private pay mark up over Medicaid rates. We estimate the association between ownership and price mark up at the NH-level using county-year fixed effects with robust standard errors, controlling for quality. Then we calculate residual demand elasticity using the inverse private pay mark up. Results: The sample includes a balanced panel of 153 Connecticut NHs with at least 5% private pay residents from 2013-2015. 75% were for-profit, of which 60% belonged to a chain. 25% were nonprofit, of which 20% of belonged to a chain. On average, price mark ups were 90% ($200). Analysis revealed no difference in price markup between nonprofit and for-profit NHs. Chain ownership was associated with a 14 percentage point increase (0.6 s.d.) in price mark up compared to NHs that are independent, (p<0.05), controlling for quality and profit status. We estimated a residual demand elasticity (absolute value) of 1.13. Conclusions/Implications: NHs that are part of a chain are setting private prices higher above Medicaid rates than independent NHs, regardless of profit status and quality ratings. Previous estimates of residual demand elasticity in the NH market in the 1990s ranged from 1.7 to 3.85. Our elasticity estimate of 1.13 suggests that competition in the NH market has decreased over time. This decrease in competition could be due to the growth of chain owned NHs. While our single state analysis limits generalizability, we utilized a novel dataset with NH-level private pay prices linked with survey data across multiple years. Policy makers should advocate for more price and ownership transparency to better understand price signals about chain owned NHs and direct them toward higher quality NHs.

The need for long-term care services and supports (LTSS) has been growing disproportionately among individuals with Alzheimer’s disease and related dementias (ADRD) due to increasing prevalence of the disease and lack of effective treatment therapies. Informal care (from friends and family members) is the primary source of LTSS for people with ADRD, but individuals with dementia also need formal LTSS (home/community based or institutional settings) as the disease progresses. While the use of different types of LTSS solely focused on costs, relative benefits of informal and formal LTSS may have differential impacts on health of care recipients. The goal of the current study is to investigate the causal effects of informal versus formal care on health and health outcomes of people with ADRD. The results from this study will address a major policy challenge the country is facing in regard to balance the use of different types of LTSS so that individuals with ADRD receive care in home like setting with best possible health outcomes. Data came from the Health and Retirement Study (HRS) (2000-2014) and the subsample of the HRS, the Aging, Demographic and Memory Study (ADAMS). Dementia diagnosis was based on the modified version of the Telephone Interview of Cognitive Status (TICS) in the HRS and detailed neurological and clinical tests in the ADAMS sample. Separate analysis was performed for both samples to account for the sensitivity of dementia diagnoses. Measures: Health and health outcomes include measures of physical, mental, emotional health, and healthcare utilization. Physical health includes functional disability (changes in activities of daily living, instrumental activities of daily living), mobility (difficulties in walking, getting across room, flights of stairs), self-rated health and mortality; mental health by depression (CESD- score). Diener’s measure of life-satisfaction captures emotional health.

Informal care is considered as an indicator variable if care recipients receive it from family members and formal care is measured by the use of home health or nursing home care. As providing informal care is likely to be endogenous, family level instruments (number of daughters, sons, siblings and number of step-children) for informal care were used in the empirical estimation. Preliminary results: The study sample includes 10,716 unique respondents from the HRS and about 856 respondents ADAMS sample members. There are significant differences in physical and mental health outcomes between those diagnosed with dementia in ADAMS sample versus those categorized as demented in the HRS. About 14% of respondents in the ADAMS sample used informal care compared to 7% of respondents in the HRS sample. Preliminary results from the two-stage residual inclusion method suggest that informal care was significantly associated with lower physical and mental health outcomes but higher emotional health in both samples, after controlling for individual level characteristics including chronic health. Ongoing analysis is investigating whether the receipt of informal care varies with and without the presence of formal care to demonstrate whether the presence of one form of LTSS affect the effectiveness of the other and whether the effect of informal care varies by disease severity.

Objectives: To estimate the extent to which childhood circumstances contribute to health inequality in old age and to evaluate the importance of health versus non-health circumstances.

Method: Building on the framework of Inequality of Opportunity (IOP, a.k.a. health inequality due to circumstances), we link the China Health and Retirement Longitudinal Study (CHARLS, HRS-sister study) with the newly released life history survey to quantify health inequality due to childhood circumstances for which they have little control. We evaluate comprehensive dimensions of health. Our analytic sample includes about 5,000 elderly Chinese between age 60 and 79. Results: Using the Shapley value decomposition approach, we first show that childhood circumstances may explain around 40 percent of health inequality in old age across multiple health outcomes. Second, while both health circumstances (especially health and access to health care in childhood) and non-health circumstances (especially family socioeconomic status and regional and urban/rural status during childhood) contribute significantly to health inequality, the latter tends to be more sizable. Discussion: Our findings support the value of a life course approach in identifying the key determinants of health in old age. Distinguishing sources of health inequality and rectifying inequality due to early circumstances should be the basis of policy promoting health equity.

Keywords: Life course; Inequality of Opportunity; Mortality; Physical health; Mental health; Cognitive ability

What role does family-provided old-age support play in alleviating the adverse effects of health shocks on the well-being of senior households? We investigate this question in the context of China, where access to public old-age pension and health insurance is relatively poor despite recent efforts to expand coverage. In the absence of formal insurance mechanisms, Chinese seniors rely heavily on children and other relatives to meet their material and instrumental needs. The question we seek to answer empirically is whether this support responds to changes in elderly household health status. Our empirical analysis draws on data from the 2011, 2013, and 2015 waves of the China Health and Retirement Longitudinal Study (CHARLS). CHARLS offers information on a broad array of individual and household economic and demographic characteristics, including detailed information on family networks and financial and time transfers to and from children and other relatives. To quantify how family support responds to elderly health status we utilize an indicator that captures the difficulty level of performing activities of daily living (ADL). The longitudinal feature of the data allows us to explore within-household variation in difficulties with ADL and, therefore, hold constant any time-invariant unobservable household traits correlated with health. Our models are also supplemented with village-year fixed effects to hold constant time-varying community-level characteristics that are potentially correlated with health outcomes, such as the supply of health care facilities, access to publicly-provided health insurance and other public safety nets, and local economic shocks. We first document that increasing difficulties with ADL significantly lowers labor supply and labor income for senior Chinese households. When health status changes so that individuals who could perform a certain ADL, say walking 100 meters, can no longer execute it (that is, ADL score increases by 3), the likelihood of working is 2.1 percentage points lower; hours worked also decreases by 7.5%. Turning to the main question of the paper, we find that financial assistance from daughters and other relatives increases when households experience increasing difficulties with ADL, 9.4% and 8.8% respectively. Sons do not provide more old-age support in response to health shocks. To measure the impact of health shocks on well-being, we run the same models on various measures of household expenditure. While we find that household per capita expenses with medical costs go up, expenditures with non-medical goods and services do not change when households experience a health shock, suggesting that the financial transfers function as a consumption smoothing mechanism for elderly households. We further investigate these findings by performing our analysis on households whose first reported job was in agriculture and non-agriculture separately. Our health shock measure, ADL, measures physical difficulties and thus we expect to find different results depending on the type of work performed. Overall, we find that households originally working in agriculture drive the results described above, suggesting that a decrease in physical ability is more damaging to households relying on physical labor and those households require (and receive) more assistance from relatives.

An estimated four to five million older adults in the United States are living with Alzheimer’s and related dementias (ADRD) (Langa, 2017), a chronic, progressive condition characterized by cognitive decline of sufficient severity to interfere with a person’s ability to carry out daily activities (Alzheimer’s Association, 2017). Despite recent evidence that prevalence rates of ADRD are declining (Langa, 2017), population forecasts indicate significant growth in the absolute number of adults in the U.S. living with dementia (Prince, 2013). Understanding the magnitude of the medical care and long-term care costs attributable to dementia is important for public and private decision makers, but estimating these costs has been difficult. First, identifying people with ADRD can be difficult in secondary data, since diagnosis can be at different stages of the disease progression or lacking altogether. Second, one must isolate the costs attributable to ADRD among a population that has several co-occurring chronic and acute conditions. Third, our fragmented health system means that many players are responsible for different types of cost; Medicare, Medicaid, and the family all play sizable roles in funding care for individuals with ADRD. We estimate the public spending on ADRD using newly available data from the Health and Retirement Survey matched to Medicare and Medicaid claims data. We identify a retrospective cohort of older adults with ADRD, and perform sensitivity analysis around the definition of dementia onset. We examine Medicare and Medicaid expenditures for the 12 months prior and up to 60 months following a diagnosis of ADRD. In order to isolate the costs attributable to ADRD, we select a comparison group of HRS participants matching on sex, birth year, and HRS entry year. To calculate the marginal effect of ADRD on Medicare expenditures, we use the estimator described by Basu and Manning (2010) for estimating costs under censoring. We estimate costs using a two-part model; the first part estimates the probability of any costs during each month using a logit model, while the second part estimates the magnitude of costs when costs are greater than zero using a generalized linear model with gamma family and power link of 0.95. This estimation is done separately on two samples: (1) months prior to death or censoring, and again (2) for months in which death occurs. An accelerated failure time model based on the lognormal distribution for time is used to estimate each subject’s survival function after accounting for censoring. We use the method of recycled predictions in order to estimate the marginal effects from each of the models. We estimate the costs attributable to ADRD, paying special attention to who bears these costs. We also examine how the total costs and the burden of costs has shifted over time and over the course of the disease.

This paper uses data from the Health and Retirement Study to investigate when the older homeowners suffering from late-life disability exit from homeownership and how this exit influences their total wealth and mortality. There has been a growing emphasis on “aging in place.” Data show that elderly rarely downsize their houses or move unless a drastic event such as an illness or death of a spouse occurs. Housing equity is the most important asset in the portfolios of large fraction of older Americans. Housing directly provides utility, and there are transaction costs associated with the purchasing and selling a house. The expanded life-cycle models have shown that bequest motives, as well as health and medical risks, are the driving forces of the puzzling phenomena. Aging in place in poor health, on the other hand, might require expensive home-based care. Obtaining care at home might cause reduced quality of care leading to increased mortality. If the current housing lacks basic accessibility features, it would also prevent disabled older adults from living safely in their home. Within the scope of benefit-cost framework, we propose that elderly with declining health and functional capacity should exit homeownership (i.e., move to a nursing home, to a retirement facility or move in with a relative) when the expected cost of “aging in place” outweighs the expected benefits. Measurement of health status at older ages is complex. No single indicator fully captures all aspects of health. We focus on functional disability, which reflects restrictions in carrying out specific activities. We measure late-life disability using limitations with six activities of daily life (ADLs) including walking across the room, bathing, dressing, eating, getting in/out bed, and toileting and five instrumental activities of daily life (IADLs) including using telephone, managing money, taking medication, shopping for groceries, and preparing hot meal. Around 11 million Americans report difficulty with performing one or more ADLs or IADLs, and about half of this population is over the age 65. Difficulties with ADLs and IADLs increase with age, and the loss of functional capacity leads to rise in morbidity and mortality. Findings from fixed effects models show that older homeowners are less likely to move unless they experience severe difficulties with ADLs, measured as difficulties with five or six ADLs. On the other hand, elderly needing help with two or more IADLs are more likely to move. When older households with diminished functional capacity move, they are less likely continue to be homeowners, and experience sharp drops in housing and total wealth. We did not find any increases in financial assets and non-housing wealth following the move and exit from homeownership. The decline in total wealth and increase in out-of-pocket health care expenditures generate lower bequest intentions for those who exit from homeownership. Upon leaving homeownership, there are some gains in mortality for those having difficulties with IADLs, but not for those having difficulties with ADLs. Our findings have significant implications for intergenerational wealth transfers, housing market and aging policy.

Abstract

We provide a framework to understand the distribution of individual well-being and its change over time with an application to the U.S. elderly population. We use data from the Health and Retirement Study to estimate life-cycle dynamics and simulate paths for consumption, leisure, health, and mortality starting at age sixty for every individual. We use an expected utility framework and the simulated profiles to construct measures of welfare distribution. Our analysis suggests substantial variation in welfare across individuals driven foremost by inequality in health and mortality followed by consumption. Not accounting for the effect of health on lifetime utility results in over-predicting the relative welfare for those in the bottom end of the distribution and under-predicting for those at the top. Elderly welfare inequality has increased over time due to growing gaps in consumption, health, and mortality. Cross-sectional income and consumption under-estimate aggregate welfare inequality and are only modestly correlated with elderly welfare at the individual level. Cross-sectional health is a better indicator of individual well-being rank than income or consumption.

1. There is substantial variation in the ex-ante welfare of individuals at age sixty. The Gini coefficient for consumption-equivalent welfare in our benchmark cohort is 0.66. Those at the ninetieth percentile of the distribution have 23 times

2.Health differences are crucial for understanding the overall distribution of elderly welfare. Excluding the utility cost of poor health and morbidities lowers the welfare Gini coefficient by 23%. This is driven by a positive correlation

3. The largest drivers of welfare inequality are health and mortality gaps followed by gaps in consumption. Differences in leisure play a comparatively minor role. 4. Welfare inequality among the elderly has increased over time due to growing gaps in consumption, health, and mortality. Compared to the cohort of individuals reaching age sixty between 1992-2001, the welfare Gini rose 9% for those

5. Ignoring dynamic uncertainty and the persistence in outcomes over the life-cycle greatly underestimates welfare inequality. The Gini of age sixty flow utility is only 70% of that based on our dynamic welfare measure. A key implication of our results is that cross-sectional distributions of income/consumption underestimate aggregate welfare inequality. This occurs for two primary reasons. First, cross-sectional measures ignore dynamic uncertainty and the persistence of inequality over life. Second, there is a positive correlation between health and consumption. However, even in cases where economic outcomes provide a reasonable approximation to aggregate welfare inequality, our results suggest they may still provide a poor ranking of individual well-being. For example, the rank correlation between consumption and welfare is a relatively modest 0.56 for our benchmark cohort. Moreover, we find cross-sectional health utility at age sixty to be a better predictor of remaining lifetime welfare rank, despite the fact that it drastically underestimates aggregate welfare inequality.

Previous studies have shown a positive association between cognitive skills and preventive health behavior. Cognitive skills have two dierent components: crystallized intelligence (the ability to use learned knowledge), and fluid intelligence (the ability to solve new problems). The role that each component plays in connection with health behavior has not yet been fully explored. During a person's elderly years, these two components decline at different rates. As a consequence, an elderly individual might rely more heavily on one of these components so as to compensate for the decline of the other, allowing him or her to maintain functioning capacity and procure healthy behaviors. We develop a theoretical model to describe the production of health of an elderly person who uses consumption goods and the two cognition components as inputs for the production of health behaviors. We test the implications of our model using the English Longitudinal Study of Aging (ELSA) for the years 2004, 2008, and 2012. We conduct the empirical analysis by first using the entire sample elderly population aged 50 and older. We then look at a subsample of these individuals, focusing on those who have diabetes and hypertension. We construct a measure of preventive behavior using individuals' self-reported behavior, as well as biomarker information. We find that memory, a proxy for crystallized intelligence, has a larger and significant impact on the production of health behavior than numeracy, a proxy for fluid intelligence. Similar results occur for seniors with diabetes and hypertension. Insights from this analysis provide evidence that can be used to improve clinical treatment and better target health interventions directed at improving the health of the elderly.

The percent of nursing home patients covered by MA nearly doubled from 2000 to 2013. There is evidence that highly integrated nursing home-hospital linkages have yielded better outcomes for patients, but little is known if Medicare Advantage plans establish special relationships with high quality facilities. The objective of this study is to examine the relationship between Medicare Advantage enrollment and the quality of nursing facilities at the county level.

Data sources include 2013-2015 Medicare Advantage payment and enrollment files and Nursing Home Compare Star-Rating files (as a measure of quality) from the Center for Medicare and Medicaid Services. I first run ordinary least squares linear regression testing for the association between Medicare Advantage penetration and facility quality at the county-level. Because where managed care organizations operate is endogenous to patient risk profiles, the quality of facilities as measured by clinical outcomes (overall star-rating) may be biased by favorable selection of healthier Medicare Advantage patients. I then instrument MA penetration with county-level payment rates—an exogenous policy shock that influences Medicare Advantage enrollment to test for the association on quality. I also test the association on structural quality measure (staffing ratings) that may be less affected by patient risk profiles. I control for the number

2,847 counties were included in the sample from 2013-2015. The average MA penetration was 32.8% (SD=15.2%) and the average MA payment was $757 (sd=$64) in the sample. The average number of nursing facilities in a county was 5.4 (sd=9.7). The average proportion of overall 5-star (high quality) facilities in a county was 22% (sd=29%), overall 1-star (low quality) was 13% (SD=24%), staffing 5-star was 10% (sd=21%), and staffing 1-star was 12% (sd=25%). OLS regressions indicate that a one-percentage increase in MA penetration is associated with increase in the proportion of overall low quality facilities (0.04%, p<.1) and low quality staffing facilities (0.19%, p<.01), and a decrease in the proportion of high quality staffing facilities (-0.15%, p<.01). In contrast, IV regressions suggest that a one-percentage increase in MA penetration is associated with decrease in the proportion of overall low quality facilities (-0.86%, p<.01) and low quality staffing facilities (-1.3%, p<.01). First-stage regressions suggest strong instrument (F=45.11, p<.001).

Results suggest that MA plans may be operating in counties where there are also lower proportion of low quality SNFs. Further research is needed to study if MA plans concentrate their enrollees in specific (i.e. high quality) facilities to provide care. It is important to identify possible inequities in nursing home use and care as a consequence of where MA plans operate.

This paper aims to expand the limited knowledge about how competitive forces shape the pricing behavior on markets for long-term care services, using a unique data set on the Dutch market for assistance in daily housekeeping activities (ADHA). This type of long term care services has been decentralized to municipalities as of 2007. As many municipalities chose not to regulate prices in their procurement process, we have market-determined price data for more than 150

Our identification strategy relies on the exogenous variation in the market shares in the first month after decentralization. This period can be seen as a mere transition phase, as existing users retained the right to receive ADHA from the very same provider as before the decentralization. Market shares in January ’07 are thus exogenous to any market developments in the years thereafter. Moreover, the January ’07 market share is a strong instrument as it turns out to be substantially correlated with market shares in later years, despite the relatively high entry rates during this period. Focusing on the most common form of ADHA (‘basic ADHA’), we obtain that a provider’s market share has a positive and significant effect on the price received. The average market leader in a municipality – possessing a 65 percent market share – secures a price that is 2.1 percent above the price of an atomistic provider. Translating this result to the overall market level tells us that the average price on the market decreases by 0.5 percent as the market becomes 10 percent less concentrated. When viewed in light of other studies on market power in care markets, this effect is considered small to moderate. Our study thus shows that high degrees of market concentration do not hinder competitive pressures on suppliers’ pricing behavior per se. A plausible explanation for this outcome is that incumbents face a serious threat of entry by newcomers, an idea that is backed by the observation that new providers enter relatively easy into the local markets. This also indicates that in order to stimulate competition on health care and long-term markets, lowering entry barriers might be more important than bringing the level of market concentration below some threshold value. Zooming in on our main outcome, we show that the small but significant effect of market size on price is merely driven by the pricing behavior of for-profit providers: while for this type of provider the average market leader receives a prices that exceeds the price of an atomistic supplier by 2.7 percent, there is no systematic relation between size and price for non-profit organizations. From this, we cannot conclude that the entry of for-profit suppliers has led to higher prices, as for-profits may be more efficient than non-profits. Still, the result indicates that the extent to which market concentration influences price formation may very well depend on the ownership composition at the supply side.

Inpatient costs account for one-third of overall health care spending in the United States. Furthermore, the majority of health care costs are driven by patients who are older and have chronic conditions such as cancer. In order to improve value in care, it is thus important to examine high-cost elderly cancer patients and identify opportunities to effectively manage their health care needs.

Examine characteristics of high-cost elderly cancer patients and compare with lower cost elderly cancer patients in hospital settings. We used the 2014 National Inpatient Sample data, which is an all-payer sample of hospital discharges and inpatient stays in the U.S. We identified 574,367 inpatient visits for individuals aged 65 years and older who had a cancer

diagnosis. High-cost visits were defined as visits with costs at or above the 90th percentile (n= 57,437). Lower-cost visits were those with cost below the 90th percentile (n= 516,930). We examined patients’ sociodemographic characteristics, including gender, age at visit, race and ethnicity, primary payer, and median household income at the zip code level. In addition, we examined clinical characteristics, which included 29 Elixhauser comorbidities that were not related to the primary diagnosis combined into a count (0, 1-2, 3-4, ≥5), type of procedure (major or minor), number of procedures, and chemotherapy use while hospitalized. Hospital ownership (public, private), hospital bed size (small/medium, large), location (rural and urban non-teaching, urban teaching), and hospital region (Northeast, Midwest, South, West) were also evaluated. Overall median cost and median cost stratified by cost groups were described with box-plots. Differences in median costs between groups were examined using the Mann–Whitney test. Logistic regression was estimated to identify characteristics associated with being in the high-cost vs. low-cost group.

The overall median cost of hospital visits for elderly cancer patients was $9,162 (IQR: 5,347-15,947). The median cost in the high-cost group was $38,194 (IQR: 31,405-51,802), nearly five times the median cost of the low-cost group ($8,257, IQR: 5,032-13,335). Those in the high-cost group were more likely than their low-cost counterparts to have 5 or more comorbidities (38.4% vs.26.2%, p<0.001). They were also significantly more likely to receive major procedures (67.1% vs. 24.3%, p<0.001) as well as a greater number of procedures. In our adjusted model, older patients, compared with younger, and females, compared with males, were more likely to be in the high-cost group. Compared to those with no comorbidities, those with 5 or more comorbidities were 5 times more likely to be in the high-cost group (OR=5.84, 95%CI: 5.37-6.36). Those with 2 or more procedures were 15 times more likely to be in the high-cost group than those with no procedures (OR=15.06, 95%CI: 13.63-16.63). Those who received chemotherapy were also more likely to be in the high-cost group (OR=4.09, 95%CI: 3.78-4.41).

High-cost elderly cancer patients who were admitted to hospitals in the U.S. had more comorbidities and received more intensive care than their lower cost counterparts. These features are important for determining

In the Netherlands, a country with one of the most extensive and expensive public long term care systems, concerns about rising costs have led the government to increase the level of co-copayments. An interesting feature of these co-payments is that they are income- and wealth-dependent. This dependency enables the fine-tuning of the financial impact of co-payments across income- and wealth groups, but it also distorts saving and annuitization decisions of

We use a unique dataset to estimate synthetic lifecycle paths of long term care spending over the lifecycle. We analyze how different types of co-payments affect consumption and saving, and how they redistribute income across income groups. We also consider the effects of the co-payment schemes on the optimal annuitization share of pension wealth. The financial wealth of Dutch elderly consists largely of annuitized pension wealth. In the case of co-payments, full annuitization might not be optimal, as people want to hold some of their wealth in cash to finance their health care costs. Moreover, differences in the co-payment rates for (pension) income and financial wealth might distort the

Modeling long term care expenditures over the lifecycle is challenging because of their very uneven distribution and limited availability of long panel data. We use a semi-parametric nearest-neighbor approach to estimate lifecycle paths of long term care spending. We use extensive administrative data that includes information on long term care spending, household status, income, and wealth for the entire Dutch population. The resulting lifecycle paths have a similar

The estimated paths are inputs in a stochastic lifecycle decision model for retirees. This model determines optimal consumption and saving behavior of elderly for different levels of initial wealth and pensions, taking into account their financial risk. We use the model to analyze the effects of different forms of income and wealth dependent co-payments on average consumption and welfare (certainty equivalent consumption) across income groups.

We find that, compared to a fully premium financed system, a fixed co-payment of 25% of care costs decreases average lifetime consumption of the elderly with the lowest financial means by 11 %, and welfare (certainty equivalent consumption) by 16 %. For the elderly with the highest financial means, welfare loss is only 2 %. An income dependent co-payment, raising the same amount of revenues as the fixed one, leads to a much smaller loss in consumption for the poor (3 %), while increasing the loss for the richest to 3.5 %. An income- and wealth-dependent co-payment reduces welfare loss for the poorest even further. For the richest, welfare loss is equal to the loss in the income-dependent case: holding financial assets becomes less attractive, leading to an increase in average lifetime consumption, while at the same time decreasing protection against care costs, leading to a welfare loss. Co-payments decrease the optimal annuitization rate. The annuitziation rate is lowest when co-payments only depend on (pension) income.

Economists often use distances as instruments to examine the causal impact of an event on an outcome. For example, distance to the nearest hospital has been used to instrument hospital utilisation. However, often little care is given to the spatial dependencies within the data, in terms of the outcome, the endogenous variable and the instrument.

Policy makers are promoting participation in community assets as a way to improve quality of life and reduce demand on healthcare services. These community assets are groups and facilities that facilitate community cohesion, produce social capital and reduce loneliness. In an earlier study, we showed that participants had better health, but this relied on correlation analysis, and we concluded that more causal investigation is required.

To examine the effects of failing to account for spatial dependencies by considering a particular empirical application; the relationship between community asset participation and health outcomes. We collected a bespoke dataset containing information on individuals aged 65 years and older with a chronic condition (N=3,470). We estimated the impact of community asset participation on three outcome measures: health-

related quality-of-life (EuroQol-5D-5L); the costs of three types of health care utilisation; and the net-benefits of participation using a range of threshold values for a Quality-Adjusted Life Year (QALY). Respondents were asked to report participation in up to 20 different types of community assets in the past six months. We obtained the geo-location of all community assets (as defined by the Localism Act, 2011) and created a range of potential instruments based on the number of assets within given distances of each individual’s place of residence and the minimum distance to the nearest asset.

To account for the potentially endogenous nature of community asset participation on these outcomes, we used both simple OLS and two-stage models, where the first stage used distance to nearest asset as an instrument. We used the Bayesian Information Criterion to select the preferred instrument specification. We repeated the analyses with and without accounting for the spatial dependencies within the data.

We found that participation in community assets significantly increased HRQoL and led to a positive net-benefit. The standard OLS and spatial OLS results were very similar in magnitude, but there was evidence of endogeneity hence the two-stage results were preferred. We further showed in the two-stage models that failing to account for the spatial dependence of individuals led to a markedly higher point estimates which were less precisely estimated. The

We found that not accounting for spatial dependencies within data can have quite large effects on the point estimate. We recommend that, where possible, the spatial nature of the data is accounted for, or is at least tested for. In the empirical application, we showed that participation in community assets is associated with substantially higher health-related quality-of-life but is not associated with lower healthcare costs. The social value of developing

In this paper, we empirically investigate whether winning a political office influences health of the candidates. Our proxy for health is the candidate’s lifespan. We postulate that there are two separate mechanisms through which winning an election affects lifespan of an individual: (i) Wealth Effect: winning an election increases wealth, which in turn increases their longevity (ii) Stress Effect: decrease in longevity due to the stress of holding a political office. We construct our data set using information from candidates who ran in the gubernatorial elections in the US between 1946 and 2000. Specifically, we obtain birth and death dates of winners and runners-up in addition to their other personal attributes, such as their level of education and their experience as a politician. We identify the impact of winning an election by comparing the lifespans of the winners versus the runners-up in close elections. In these close races where the winner’s margin of victory is small, winning the election is arguably random. That is, the winner could have easily lost (and the runner-up could win) if only a small share of voters did not cast a vote for the winner. Our results indicate that the observable characteristics of the winners versus the runners-up are similar on average in elections in which winners’ margins of victory are small. This finding provides support for the randomness of the treatment. We find that winners of the gubernatorial races live about 4 years longer than the runners-up. Our back-of-the-envelope calculations suggest that the increase in wealth due to winning an election increases lifespan by about 7

The use of post-acute care has grown substantially over the past few decades. Nearly 40% of Medicare beneficiaries receive post-acute care after a hospital discharge, and most of those go either to a skilled nursing facility (SNF) or home with care from a home health agency (HHA). In 2015, Medicare spent over $60 billion on post-acute care, of which $48 billion went to SNF and HHA. Despite the proliferation of post-acute care, it is uncertain whether post-acute care benefits patients or whether the choice of specific post-acute care setting matters (i.e. choosing SNF versus HHA). Indeed, the use of post-acute care varies significantly across the country, suggesting substantial uncertainty about its value to patients. In this paper, we investigate the impact of discharge to SNF versus HHA on patient outcomes. We use data from 2010-2014 on all Medicare fee-for-service beneficiaries who are discharged from the hospital and receive post-acute care in either SNF or HHA. We estimate the effect of post-acute care setting on the following patient-level outcomes: death within 30 days of hospital discharge, readmission within 30 days of hospital discharge, successful discharge to the community, and improvement in functional status during the post-acute care episode. To address the endogeneity of treatment choice, we use an instrumental variables approach, using as an instrument the differential distance between the beneficiary’s home ZIP code and the closest HHA and the closest SNF. The instrument passes standard tests of first-stage strength. In all regressions, we include measures of patient case mix, diagnosis related groups, year fixed effects, and hospital fixed effects. Using ordinary least squares regression, we find substantial differences in patient outcomes by discharge setting. Compared to patients discharged to SNF, patients discharged to home health have lower readmission and death rates (by 2.5 and 4.7 percentage points respectively), are much more likely to be successfully discharged to the community (by 25 percentage points) and experience improvement in functional status while in post-acute care (by 54.3 percentage points). These findings are consistent with selection; healthier patients are more likely to be discharged with home health. In the instrumental variable specifications which account for this selection bias, these results change. Patients discharged to home health are still more likely to be successfully discharged to the community, although the effect size is about half (13 percentage points). However, patients discharged to home health are also more likely to be readmitted to the hospital (by 5.5 percentage points). They are no more likely to die. The difference in functional status improvement favored home health but was not statistically different from zero. These preliminary results suggest there are important tradeoffs between home health and SNF care for patients needing post-acute care. While current policies may incentivize the use of lower-intensity settings (such as home health care) for patients needing post-acute care, lower intensity settings may have adverse outcomes that need to be taken into account and balanced against the lower cost of using home health.

Many developed countries face growing demographic pressures on their health and long-term care budgets. Improving the efficiency of these services has therefore become an important policy priority across the world. This paper examines the impact of reductions to public spending on adult social (long-term) care on the use of public hospitals in England in order to quantify the extent of substitution between the two types of publicly funded care. It exploits large reductions in public spending on adult social care in England between 2009 and 2015, a period when social care spending fell by 37% as part of widespread government austerity measures, but where public spending on health care was

The institutional features of the public health and social care system in England mean that local governments retain the responsibility to fund and organise public services for their local population. As a result, there was considerable geographical variation in the cuts to adult social care spending over the period. We exploit this variation to identify the impact of public funding for social care for individuals aged 65 and above on their use of public hospitals. We control for permanent differences in the use of hospitals across local authorities with the inclusion of area fixed effects, in addition to a rich set of local area characteristics to control for time-varying needs for hospital services. Our results indicate a small but statistically significant, negative impact of public social care spending on a number of measures of public hospital use. These include visits to emergency departments and admissions for inpatient care. The estimates imply an additional spend of $23 million on emergency treatment in 2015-16 relative to 2009-10 as a result of cuts to social care spending. When examining the effects by age, the estimates indicate that the magnitudes of the effects are largest for older individuals. However, the estimates indicate no statistically significant relationship between social care spending and delayed exits from the medical system, either as measured by official delayed discharge statistics or by the average

These results have important implications for policy. They suggest that the recent cuts to social care spending have led to a modest increase in public hospital use among the older population. This means that social care spending has a small fiscal externality on health spending. As a result, attempts to reduce overall public spending on health and social care have been less effective than appears when looking only at the reduction in social care spending. However, the results suggest that the cuts have not been the major driving force behind recent increases in emergency department attendances or the number of delayed discharges from the medical system.

In 1998, Medicare changed its payment method for post-acute care provided by skilled nursing facilities (SNFs) from a cost-based system to a per diem Prospective Payment System (PPS), with a goal to control increasingly high Medicare SNF expenditures. Under the PPS, the per diem rates are primarily based on therapy minutes, creating an incentive for SNFs to provide high-intensity therapy services. It is well documented that SNFs have exploited this reimbursement method by over-providing these therapy services to Medicare beneficiaries, regardless of their actual clinical needs. Accounting for more than 50% of the SNFs in the United States, corporate chains can affect affiliated facilities’ practice patterns. This paper aims to examine whether the selection of therapy treatment levels for SNF residents varies by chain ownership. Using a Difference-in-Differences model to compare independent and chain-acquired SNFs during the period from 2003 to 2009, I find that chain acquisition of independent SNFs are associated with about 2.92 percentage points increase in the proportion of residents with the highest therapy treatment level. A more dynamic model suggests that the chain effects on this aggressive billing practice may last for several years after the acquisition. In addition, I find that the main effects are mostly due to acquisitions by large chains and for-profit chains. However, the increase in treatment intensity among chain-affiliated SNFs does not lead to shorter length of stay. Overall, the findings suggest that government agencies should consider chain affiliation when monitoring Medicare reimbursements

: Created by the Centers for Medicare and Medicaid Innovation (CMMI), the Medicare Care Choices Model (MCCM) extends the benefits of hospice services to Medicare beneficiaries wanting to concurrently receive curative treatment. The MCCM aims to increase access, test a new payment model and improve quality of life and patient/family satisfaction. 141 hospice providers were selected by an expert panel in hospice and model implementation on the basis of being diverse in geographic area and having demonstrated experience of coordination and shared decision-making with beneficiaries and their families. Although hospices were positively selected into the program, CMMI randomly assigned half of the providers to begin in January 2016 (Phase 1) and the other half in January 2018 (Phase 2) and will continue until 2020.

: Demographics of providers in 2014 and 2015 from Hospice Utilization and Payment Public Use File, 2016 quality ratings from Hospice Compare, 2010-2017 deficiencies from Centers for Medicare and Medicaid Services Quality, Certification and Oversight Reports, and identification of providers participating in the MCCM from CMMI.

: This study employs two methods to estimate a causal effect of the MCCM on two measures of quality: quality ratings and deficiencies. First, we compare the Phase 1 hospice providers’ baseline characteristics (volume, geographic location, accreditation, patient characteristics) to the Phase 2 characteristics in order to establish that the randomization was sufficient. Next, we estimate the impact of the MCCM on seven Hospice Compare quality ratings using a generalized linear model (GLM) with a Gamma distribution and log link function, due to the skewed distribution of the ratings. Finally, we employ a difference-in-differences approach, which compares the deficiencies from inspection reports of Phase 1 providers (treatment) after the MCCM began to deficiencies before the MCCM was started and to Phase 2 providers (control).

: The Phase 1 hospices were comparable in baseline characteristics to Phase 2 hospices. Compared to all hospices in the United States, they were larger and more non-profit. The GLM model found no statistically significant differences between Phase 1 and Phase 2 providers, with coefficient magnitudes close to zero for all seven quality ratings. Graphically examining the deficiencies per hospice provider surveyed from 2010-2016, the parallel trends assumption holds between Phase 1 and Phase 2 providers. Phase 1 providers had on average 1.07 deficiencies in the pre-period and 1.17 in the post-period for an increase of 0.10 from the pre- to post-period. Phase 2 providers had on average 0.82 deficiencies in the pre-period and 0.92 in the post-period, for an increase in 0.12. The difference-in-differences model found no statistically significant difference in deficiencies between Phase 1 and Phase 2 providers, pre-

: After one year of implementation, there is no impact of the MCCM on quality of hospice. Further information on the number of program enrollees and a longer time period of follow-up are necessary to estimating the impact

: Economic theory provides a useful framework for analyzing nursing home (NH) ownership, competition and signals about quality to consumers. NHs serve both private pay and Medicaid consumers, but Medicaid rates are set by the state so Medicaid demand is perfectly elastic. However, NHs can set private pay rates over Medicaid rates depending on the market structure. For-profit and corporate chain-owned NHs aim to minimize costs and set prices in order to maximize profits, often signaling low quality. While nonprofit and independent NHs have other objectives, such as altruism, which signal high quality. Despite the conflicting profit motives for nonprofit chain NHs, about 13% of the market, no research has examined competition relative to independent nonprofit and for-profit NHs.

This study will examine competition in Connecticut’s NH market by estimating the association of ownership on price mark ups and private pay residual demand elasticity. Data include private pay rates from Cost of Long-Term Care in Connecticut reports, Medicaid rates from Connecticut Department of Social Services, facility characteristics from Online Survey and Certification Reporting System and

To assess market power, we calculate the percent private pay mark up over Medicaid rates. We estimate the association between ownership and price mark up at the NH-level using county-year fixed effects with robust standard errors, controlling for quality. Then we calculate residual demand elasticity using the inverse private pay mark up.

The sample includes a balanced panel of 153 Connecticut NHs with at least 5% private pay residents from 2013-2015. 75% were for-profit, of which 60% belonged to a chain. 25% were nonprofit, of which 20% of belonged to a chain. On average, price mark ups were 90% ($200). Analysis revealed no difference in price markup between nonprofit and for-profit NHs. Chain ownership was associated with a 14 percentage point increase (0.6 s.d.) in price mark up compared to NHs that are independent, (p<0.05), controlling for quality and profit status. We estimated a residual demand elasticity (absolute value) of 1.13.

NHs that are part of a chain are setting private prices higher above Medicaid rates than independent NHs, regardless of profit status and quality ratings. Previous estimates of residual demand elasticity in the NH market in the 1990s ranged from 1.7 to 3.85. Our elasticity estimate of 1.13 suggests that competition in the NH market has decreased over time. This decrease in competition could be due to the growth of chain owned NHs. While our single state analysis limits generalizability, we utilized a novel dataset with NH-level private pay prices linked with survey data across multiple years. Policy makers should advocate for more price and ownership transparency to better

The need for long-term care services and supports (LTSS) has been growing disproportionately among individuals with Alzheimer’s disease and related dementias (ADRD) due to increasing prevalence of the disease and lack of effective treatment therapies. Informal care (from friends and family members) is the primary source of LTSS for people with ADRD, but individuals with dementia also need formal LTSS (home/community based or institutional settings) as the disease progresses. While the use of different types of LTSS solely focused on costs, relative benefits of informal and formal LTSS may have differential impacts on health of care recipients. The goal of the current study is to investigate the causal effects of informal versus formal care on health and health outcomes of people with ADRD. The results from this study will address a major policy challenge the country is facing in regard to balance the use of different types of LTSS so that individuals with ADRD receive care in home like setting with best possible health outcomes. Data came from the Health and Retirement Study (HRS) (2000-2014) and the subsample of the HRS, the Aging, Demographic and Memory Study (ADAMS). Dementia diagnosis was based on the modified version of the Telephone Interview of Cognitive Status (TICS) in the HRS and detailed neurological and clinical tests in the ADAMS sample. Separate analysis was performed for both samples to account for the sensitivity of dementia diagnoses.

Health and health outcomes include measures of physical, mental, emotional health, and healthcare utilization. Physical health includes functional disability (changes in activities of daily living, instrumental activities of daily living), mobility (difficulties in walking, getting across room, flights of stairs), self-rated health and mortality; mental health by depression (CESD- score). Diener’s measure of life-satisfaction captures emotional health.

Informal care is considered as an indicator variable if care recipients receive it from family members and formal care is measured by the use of home health or nursing home care. As providing informal care is likely to be endogenous, family level instruments (number of daughters, sons, siblings and number of step-children) for informal care were used in the empirical estimation.

The study sample includes 10,716 unique respondents from the HRS and about 856 respondents ADAMS sample members. There are significant differences in physical and mental health outcomes between those diagnosed with dementia in ADAMS sample versus those categorized as demented in the HRS. About 14% of respondents in the ADAMS sample used informal care compared to 7% of respondents in the HRS sample. Preliminary results from the two-stage residual inclusion method suggest that informal care was significantly associated with lower physical and mental health outcomes but higher emotional health in both samples, after controlling for individual level characteristics including chronic health. Ongoing analysis is investigating whether the receipt of informal care varies with and without the presence of formal care to demonstrate whether the presence of one form of LTSS affect the effectiveness of the other and whether

To estimate the extent to which childhood circumstances contribute to health inequality in old age and to evaluate the importance of health versus non-health circumstances.

Building on the framework of Inequality of Opportunity (IOP, a.k.a. health inequality due to circumstances), we link the China Health and Retirement Longitudinal Study (CHARLS, HRS-sister study) with the newly released life history survey to quantify health inequality due to childhood circumstances for which they have little control. We evaluate comprehensive dimensions of health. Our analytic sample includes about 5,000 elderly Chinese between age 60

Using the Shapley value decomposition approach, we first show that childhood circumstances may explain around 40 percent of health inequality in old age across multiple health outcomes. Second, while both health circumstances (especially health and access to health care in childhood) and non-health circumstances (especially family socioeconomic status and regional and urban/rural status during childhood) contribute significantly to health

Our findings support the value of a life course approach in identifying the key determinants of health in old age. Distinguishing sources of health inequality and rectifying inequality due to early circumstances should be the basis

Life course; Inequality of Opportunity; Mortality; Physical health; Mental health; Cognitive ability

What role does family-provided old-age support play in alleviating the adverse effects of health shocks on the well-being of senior households? We investigate this question in the context of China, where access to public old-age pension and health insurance is relatively poor despite recent efforts to expand coverage. In the absence of formal insurance mechanisms, Chinese seniors rely heavily on children and other relatives to meet their material and instrumental needs. The question we seek to answer empirically is whether this support responds to changes in elderly household health status. Our empirical analysis draws on data from the 2011, 2013, and 2015 waves of the China Health and Retirement Longitudinal Study (CHARLS). CHARLS offers information on a broad array of individual and household economic and demographic characteristics, including detailed information on family networks and financial and time transfers to and from children and other relatives. To quantify how family support responds to elderly health status we utilize an indicator that captures the difficulty level of performing activities of daily living (ADL). The longitudinal feature of the data allows us to explore within-household variation in difficulties with ADL and, therefore, hold constant any time-invariant unobservable household traits correlated with health. Our models are also supplemented with village-year fixed effects to hold constant time-varying community-level characteristics that are potentially correlated with health outcomes, such as the supply of health care facilities, access to publicly-provided health insurance and other public safety nets, and local economic shocks. We first document that increasing difficulties with ADL significantly lowers labor supply and labor income for senior Chinese households. When health status changes so that individuals who could perform a certain ADL, say walking 100 meters, can no longer execute it (that is, ADL score increases by 3), the likelihood of working is 2.1 percentage points lower; hours worked also decreases by 7.5%. Turning to the main question of the paper, we find that financial assistance from daughters and other relatives increases when households experience increasing difficulties with ADL, 9.4% and 8.8% respectively. Sons do not provide more old-age support in response to health shocks. To measure the impact of health shocks on well-being, we run the same models on various measures of household expenditure. While we find that household per capita expenses with medical costs go up, expenditures with non-medical goods and services do not change when households experience a health shock, suggesting that the financial transfers function as a consumption smoothing mechanism for elderly households. We further investigate these findings by performing our analysis on households whose first reported job was in agriculture and non-agriculture separately. Our health shock measure, ADL, measures physical difficulties and thus we expect to find different results depending on the type of work performed. Overall, we find that households originally working in agriculture drive the results described above, suggesting that a decrease in physical ability is more damaging to households relying on physical labor and those households require (and receive) more assistance from relatives.

An estimated four to five million older adults in the United States are living with Alzheimer’s and related dementias (ADRD) (Langa, 2017), a chronic, progressive condition characterized by cognitive decline of sufficient severity to interfere with a person’s ability to carry out daily activities (Alzheimer’s Association, 2017). Despite recent evidence that prevalence rates of ADRD are declining (Langa, 2017), population forecasts indicate significant growth in the absolute number

Understanding the magnitude of the medical care and long-term care costs attributable to dementia is important for public and private decision makers, but estimating these costs has been difficult. First, identifying people with ADRD can be difficult in secondary data, since diagnosis can be at different stages of the disease progression or lacking altogether. Second, one must isolate the costs attributable to ADRD among a population that has several co-occurring chronic and acute conditions. Third, our fragmented health system means that many players are responsible for different types of cost; Medicare, Medicaid, and the family all play sizable roles in funding care for individuals with ADRD. We estimate the public spending on ADRD using newly available data from the Health and Retirement Survey matched to Medicare and Medicaid claims data. We identify a retrospective cohort of older adults with ADRD, and perform sensitivity analysis around the definition of dementia onset. We examine Medicare and Medicaid expenditures for the 12 months prior and up to 60 months following a diagnosis of ADRD. In order to isolate the costs attributable to ADRD, we select a comparison group of HRS participants matching on sex, birth year, and HRS entry year. To calculate the marginal effect of ADRD on Medicare expenditures, we use the estimator described by Basu and Manning (2010) for estimating costs under censoring. We estimate costs using a two-part model; the first part estimates the probability of any costs during each month using a logit model, while the second part estimates the magnitude of costs when costs are greater than zero using a generalized linear model with gamma family and power link of 0.95. This estimation is done separately on two samples: (1) months prior to death or censoring, and again (2) for months in which death occurs. An accelerated failure time model based on the lognormal distribution for time is used to estimate each subject’s survival function after accounting for censoring. We use the method of recycled predictions in order to estimate the

We estimate the costs attributable to ADRD, paying special attention to who bears these costs. We also examine how the total costs and the burden of costs has shifted over time and over the course of the disease.

This paper uses data from the Health and Retirement Study to investigate when the older homeowners suffering from late-life disability exit from homeownership and how this exit influences their total wealth and mortality. There has been a growing emphasis on “aging in place.” Data show that elderly rarely downsize their houses or move unless a drastic event such as an illness or death of a spouse occurs. Housing equity is the most important asset in the portfolios of large fraction of older Americans. Housing directly provides utility, and there are transaction costs associated with the purchasing and selling a house. The expanded life-cycle models have shown that bequest motives, as well as health and medical risks, are the driving forces of the puzzling phenomena. Aging in place in poor health, on the other hand, might require expensive home-based care. Obtaining care at home might cause reduced quality of care leading to increased mortality. If the current housing lacks basic accessibility features, it would also prevent disabled older adults from living safely in their home. Within the scope of benefit-cost framework, we propose that elderly with declining health and functional capacity should exit homeownership (i.e., move to a nursing home, to a retirement facility or move in with a relative) when the expected cost of “aging in place” outweighs the expected benefits. Measurement of health status at older ages is complex. No single indicator fully captures all aspects of health. We focus on functional disability, which reflects restrictions in carrying out specific activities. We measure late-life disability using limitations with six activities of daily life (ADLs) including walking across the room, bathing, dressing, eating, getting in/out bed, and toileting and five instrumental activities of daily life (IADLs) including using telephone, managing money, taking medication, shopping for groceries, and preparing hot meal. Around 11 million Americans report difficulty with performing one or more ADLs or IADLs, and about half of this population is over the age 65. Difficulties with ADLs and IADLs increase with age, and the loss of functional capacity leads to rise in morbidity and mortality. Findings from fixed effects models show that older homeowners are less likely to move unless they experience severe difficulties with ADLs, measured as difficulties with five or six ADLs. On the other hand, elderly needing help with two or more IADLs are more likely to move. When older households with diminished functional capacity move, they are less likely continue to be homeowners, and experience sharp drops in housing and total wealth. We did not find any increases in financial assets and non-housing wealth following the move and exit from homeownership. The decline in total wealth and increase in out-of-pocket health care expenditures generate lower bequest intentions for those who exit from homeownership. Upon leaving homeownership, there are some gains in mortality for those having difficulties with IADLs, but not for those having difficulties with ADLs. Our findings have significant implications for


Ray Miller [email protected]

Antonio Trujillo [email protected]

Shannon Wu [email protected]

We provide a framework to understand the distribution of individual well-being and its change over time with an application to the U.S. elderly population. We use data from the Health and Retirement Study to estimate life-cycle dynamics and simulate paths for consumption, leisure, health, and mortality starting at age sixty for every individual. We use an expected utility framework and the simulated profiles to construct measures of welfare distribution. Our analysis suggests substantial variation in welfare across individuals driven foremost by inequality in health and mortality followed by consumption. Not accounting for the effect of health on lifetime utility results in over-predicting the relative welfare for those in the bottom end of the distribution and under-predicting for those at the top. Elderly welfare inequality has increased over time due to growing gaps in consumption, health, and mortality. Cross-sectional income and consumption under-estimate aggregate welfare inequality and are only modestly correlated with elderly welfare at the individual level. Cross-sectional health is a better indicator of individual well-being rank than income or consumption.

1. There is substantial variation in the ex-ante welfare of individuals at age sixty. The Gini coefficient for consumption-equivalent welfare in our benchmark cohort is 0.66. Those at the ninetieth percentile of the distribution have 23 times

2.Health differences are crucial for understanding the overall distribution of elderly welfare. Excluding the utility cost of poor health and morbidities lowers the welfare Gini coefficient by 23%. This is driven by a positive correlation

4. Welfare inequality among the elderly has increased over time due to growing gaps in consumption, health, and mortality. Compared to the cohort of individuals reaching age sixty between 1992-2001, the welfare Gini rose 9% for those

5. Ignoring dynamic uncertainty and the persistence in outcomes over the life-cycle greatly underestimates welfare inequality. The Gini of age sixty flow utility is only 70% of that based on our dynamic welfare measure. A key implication of our results is that cross-sectional distributions of income/consumption underestimate aggregate welfare inequality. This occurs for two primary reasons. First, cross-sectional measures ignore dynamic uncertainty and the persistence of inequality over life. Second, there is a positive correlation between health and consumption. However, even in cases where economic outcomes provide a reasonable approximation to aggregate welfare inequality, our results suggest they may still provide a poor ranking of individual well-being. For example, the rank correlation between consumption and welfare is a relatively modest 0.56 for our benchmark cohort. Moreover, we find cross-sectional

Previous studies have shown a positive association between cognitive skills and preventive health behavior. Cognitive skills have two dierent components: crystallized intelligence (the ability to use learned knowledge), and fluid intelligence (the ability to solve new problems). The role that each component plays in connection with health behavior has not yet been fully explored. During a person's elderly years, these two components decline at different rates. As a consequence, an elderly individual might rely more heavily on one of these components so as to compensate for the decline of the other, allowing him or her to maintain functioning capacity and procure healthy behaviors. We develop a theoretical model to describe the production of health of an elderly person who uses consumption goods and the two cognition components as inputs for the production of health behaviors. We test the implications of our model using the English Longitudinal Study of Aging (ELSA) for the years 2004, 2008, and 2012. We conduct the empirical analysis by first using the entire sample elderly population aged 50 and older. We then look at a subsample of these individuals, focusing on those who have diabetes and hypertension. We construct a measure of preventive behavior using individuals' self-reported behavior, as well as biomarker information. We find that memory, a proxy for crystallized intelligence, has a larger and significant impact on the production of health behavior than numeracy, a proxy for fluid intelligence. Similar results occur for seniors with diabetes and hypertension. Insights from this analysis provide evidence that can be

The percent of nursing home patients covered by MA nearly doubled from 2000 to 2013. There is evidence that highly integrated nursing home-hospital linkages have yielded better outcomes for patients, but little is known if Medicare Advantage plans establish special relationships with high quality facilities. The objective of this study is to examine the relationship between Medicare Advantage enrollment and the quality of nursing facilities at the county level.

Data sources include 2013-2015 Medicare Advantage payment and enrollment files and Nursing Home Compare Star-Rating files (as a measure of quality) from the Center for Medicare and Medicaid Services. I first run ordinary least squares linear regression testing for the association between Medicare Advantage penetration and facility quality at the county-level. Because where managed care organizations operate is endogenous to patient risk profiles, the quality of facilities as measured by clinical outcomes (overall star-rating) may be biased by favorable selection of healthier Medicare Advantage patients. I then instrument MA penetration with county-level payment rates—an exogenous policy shock that influences Medicare Advantage enrollment to test for the association on quality. I also test the association on structural quality measure (staffing ratings) that may be less affected by patient risk profiles. I control for the number

2,847 counties were included in the sample from 2013-2015. The average MA penetration was 32.8% (SD=15.2%) and the average MA payment was $757 (sd=$64) in the sample. The average number of nursing facilities in a county was 5.4 (sd=9.7). The average proportion of overall 5-star (high quality) facilities in a county was 22% (sd=29%), overall 1-star (low quality) was 13% (SD=24%), staffing 5-star was 10% (sd=21%), and staffing 1-star was 12% (sd=25%). OLS regressions indicate that a one-percentage increase in MA penetration is associated with increase in the proportion of overall low quality facilities (0.04%, p<.1) and low quality staffing facilities (0.19%, p<.01), and a decrease in the proportion of high quality staffing facilities (-0.15%, p<.01). In contrast, IV regressions suggest that a one-percentage increase in MA penetration is associated with decrease in the proportion of overall low quality facilities (-0.86%, p<.01) and low quality

Results suggest that MA plans may be operating in counties where there are also lower proportion of low quality SNFs. Further research is needed to study if MA plans concentrate their enrollees in specific (i.e. high quality) facilities to

Remco van Eijkel [email protected]

Jaqueline Avila [email protected]

This paper aims to expand the limited knowledge about how competitive forces shape the pricing behavior on markets for long-term care services, using a unique data set on the Dutch market for assistance in daily housekeeping activities (ADHA). This type of long term care services has been decentralized to municipalities as of 2007. As many municipalities chose not to regulate prices in their procurement process, we have market-determined price data for more than 150

Our identification strategy relies on the exogenous variation in the market shares in the first month after decentralization. This period can be seen as a mere transition phase, as existing users retained the right to receive ADHA from the very same provider as before the decentralization. Market shares in January ’07 are thus exogenous to any market developments in the years thereafter. Moreover, the January ’07 market share is a strong instrument as it turns out to be

Focusing on the most common form of ADHA (‘basic ADHA’), we obtain that a provider’s market share has a positive and significant effect on the price received. The average market leader in a municipality – possessing a 65 percent market share – secures a price that is 2.1 percent above the price of an atomistic provider. Translating this result to the overall market level tells us that the average price on the market decreases by 0.5 percent as the market becomes 10 percent

Our study thus shows that high degrees of market concentration do not hinder competitive pressures on suppliers’ pricing behavior per se. A plausible explanation for this outcome is that incumbents face a serious threat of entry by newcomers, an idea that is backed by the observation that new providers enter relatively easy into the local markets. This also indicates that in order to stimulate competition on health care and long-term markets, lowering entry barriers

Zooming in on our main outcome, we show that the small but significant effect of market size on price is merely driven by the pricing behavior of for-profit providers: while for this type of provider the average market leader receives a prices that exceeds the price of an atomistic supplier by 2.7 percent, there is no systematic relation between size and price for non-profit organizations. From this, we cannot conclude that the entry of for-profit suppliers has led to higher prices, as for-profits may be more efficient than non-profits. Still, the result indicates that the extent to which market concentration influences price formation may very well depend on the ownership composition at the supply side.

Inpatient costs account for one-third of overall health care spending in the United States. Furthermore, the majority of health care costs are driven by patients who are older and have chronic conditions such as cancer. In

We used the 2014 National Inpatient Sample data, which is an all-payer sample of hospital discharges and inpatient stays in the U.S. We identified 574,367 inpatient visits for individuals aged 65 years and older who had a cancer diagnosis. High-cost visits were defined as visits with costs at or above the 90th percentile (n= 57,437). Lower-cost visits were those with cost below the 90th percentile (n= 516,930). We examined patients’ sociodemographic characteristics, including gender, age at visit, race and ethnicity, primary payer, and median household income at the zip code level. In addition, we examined clinical characteristics, which included 29 Elixhauser comorbidities that were not related to the primary diagnosis combined into a count (0, 1-2, 3-4, ≥5), type of procedure (major or minor), number of procedures, and chemotherapy use while hospitalized. Hospital ownership (public, private), hospital bed size (small/medium, large), location (rural and urban non-teaching, urban teaching), and hospital region (Northeast, Midwest, South, West) were also evaluated. Overall median cost and median cost stratified by cost groups were described with box-plots. Differences in median costs between groups were examined using the Mann–Whitney test. Logistic regression was estimated to identify characteristics associated with being in the high-cost vs. low-cost group.

The overall median cost of hospital visits for elderly cancer patients was $9,162 (IQR: 5,347-15,947). The median cost in the high-cost group was $38,194 (IQR: 31,405-51,802), nearly five times the median cost of the low-cost group ($8,257, IQR: 5,032-13,335). Those in the high-cost group were more likely than their low-cost counterparts to have 5 or more comorbidities (38.4% vs.26.2%, p<0.001). They were also significantly more likely to receive major procedures (67.1% vs. 24.3%, p<0.001) as well as a greater number of procedures. In our adjusted model, older patients, compared with younger, and females, compared with males, were more likely to be in the high-cost group. Compared to those with no comorbidities, those with 5 or more comorbidities were 5 times more likely to be in the high-cost group (OR=5.84, 95%CI: 5.37-6.36). Those with 2 or more procedures were 15 times more likely to be in the high-cost group than those with no procedures (OR=15.06, 95%CI: 13.63-16.63). Those who received chemotherapy were also more likely to be in the high-cost group (OR=4.09, 95%CI: 3.78-4.41).

High-cost elderly cancer patients who were admitted to hospitals in the U.S. had more comorbidities and received more intensive care than their lower cost counterparts. These features are important for determining

Bram Wouterse [email protected]

Luke Munford [email protected]

Duha Tore Altindag [email protected]

In the Netherlands, a country with one of the most extensive and expensive public long term care systems, concerns about rising costs have led the government to increase the level of co-copayments. An interesting feature of these co-payments is that they are income- and wealth-dependent. This dependency enables the fine-tuning of the financial impact of co-payments across income- and wealth groups, but it also distorts saving and annuitization decisions of

We use a unique dataset to estimate synthetic lifecycle paths of long term care spending over the lifecycle. We analyze how different types of co-payments affect consumption and saving, and how they redistribute income across income groups. We also consider the effects of the co-payment schemes on the optimal annuitization share of pension wealth. The financial wealth of Dutch elderly consists largely of annuitized pension wealth. In the case of co-payments, full annuitization might not be optimal, as people want to hold some of their wealth in cash to finance their health care costs. Moreover, differences in the co-payment rates for (pension) income and financial wealth might distort the

Modeling long term care expenditures over the lifecycle is challenging because of their very uneven distribution and limited availability of long panel data. We use a semi-parametric nearest-neighbor approach to estimate lifecycle paths of long term care spending. We use extensive administrative data that includes information on long term care spending, household status, income, and wealth for the entire Dutch population. The resulting lifecycle paths have a similar

The estimated paths are inputs in a stochastic lifecycle decision model for retirees. This model determines optimal consumption and saving behavior of elderly for different levels of initial wealth and pensions, taking into account their financial risk. We use the model to analyze the effects of different forms of income and wealth dependent co-payments on average consumption and welfare (certainty equivalent consumption) across income groups.

We find that, compared to a fully premium financed system, a fixed co-payment of 25% of care costs decreases average lifetime consumption of the elderly with the lowest financial means by 11 %, and welfare (certainty equivalent consumption) by 16 %. For the elderly with the highest financial means, welfare loss is only 2 %. An income dependent co-payment, raising the same amount of revenues as the fixed one, leads to a much smaller loss in consumption for the poor (3 %), while increasing the loss for the richest to 3.5 %. An income- and wealth-dependent co-payment reduces welfare loss for the poorest even further. For the richest, welfare loss is equal to the loss in the income-dependent case: holding financial assets becomes less attractive, leading to an increase in average lifetime consumption, while at the same time decreasing protection against care costs, leading to a welfare loss. Co-payments decrease the optimal

Economists often use distances as instruments to examine the causal impact of an event on an outcome. For example, distance to the nearest hospital has been used to instrument hospital utilisation. However, often little

Policy makers are promoting participation in community assets as a way to improve quality of life and reduce demand on healthcare services. These community assets are groups and facilities that facilitate community cohesion, produce social capital and reduce loneliness. In an earlier study, we showed that participants had better health, but this relied on correlation analysis, and we concluded that more causal investigation is required.

To examine the effects of failing to account for spatial dependencies by considering a particular empirical application; the relationship between community asset participation and health outcomes. We collected a bespoke dataset containing information on individuals aged 65 years and older with a chronic condition (N=3,470). We estimated the impact of community asset participation on three outcome measures: health-

related quality-of-life (EuroQol-5D-5L); the costs of three types of health care utilisation; and the net-benefits of participation using a range of threshold values for a Quality-Adjusted Life Year (QALY). Respondents were asked to report participation in up to 20 different types of community assets in the past six months. We obtained the geo-location of all community assets (as defined by the Localism Act, 2011) and created a range of potential instruments based on the

To account for the potentially endogenous nature of community asset participation on these outcomes, we used both simple OLS and two-stage models, where the first stage used distance to nearest asset as an instrument. We

We found that participation in community assets significantly increased HRQoL and led to a positive net-benefit. The standard OLS and spatial OLS results were very similar in magnitude, but there was evidence of endogeneity hence the two-stage results were preferred. We further showed in the two-stage models that failing to account for the spatial dependence of individuals led to a markedly higher point estimates which were less precisely estimated. The

We found that not accounting for spatial dependencies within data can have quite large effects on the point estimate. We recommend that, where possible, the spatial nature of the data is accounted for, or is at least tested for. In the empirical application, we showed that participation in community assets is associated with substantially higher health-related quality-of-life but is not associated with lower healthcare costs. The social value of developing

In this paper, we empirically investigate whether winning a political office influences health of the candidates. Our proxy for health is the candidate’s lifespan. We postulate that there are two separate mechanisms through which winning an election affects lifespan of an individual: (i) Wealth Effect: winning an election increases wealth, which in turn increases their longevity (ii) Stress Effect: decrease in longevity due to the stress of holding a political office. We construct our data set using information from candidates who ran in the gubernatorial elections in the US between 1946 and 2000. Specifically, we obtain birth and death dates of winners and runners-up in addition to their other personal attributes, such as their level of education and their experience as a politician. We identify the impact of winning an election by comparing the lifespans of the winners versus the runners-up in close elections. In these close races where the winner’s margin of victory is small, winning the election is arguably random. That is, the winner could have easily lost (and the runner-up could win) if only a small share of voters did not cast a vote for the winner. Our results indicate that the observable characteristics of the winners versus the runners-up are similar on average in elections in which winners’ margins of victory are small. This finding provides support for the randomness of the treatment. We find that winners of the gubernatorial races live about 4 years longer than the runners-up. Our back-of-the-envelope calculations suggest that the increase in wealth due to winning an election increases lifespan by about 7

Rachel Werner [email protected]

George Stoye [email protected]

Qing Zheng [email protected]

The use of post-acute care has grown substantially over the past few decades. Nearly 40% of Medicare beneficiaries receive post-acute care after a hospital discharge, and most of those go either to a skilled nursing facility (SNF) or home with care from a home health agency (HHA). In 2015, Medicare spent over $60 billion on post-acute care, of which $48 billion went to SNF and HHA. Despite the proliferation of post-acute care, it is uncertain whether post-acute care benefits patients or whether the choice of specific post-acute care setting matters (i.e. choosing SNF versus HHA). Indeed, the use of post-acute care varies significantly across the country, suggesting substantial uncertainty about its value

We use data from 2010-2014 on all Medicare fee-for-service beneficiaries who are discharged from the hospital and receive post-acute care in either SNF or HHA. We estimate the effect of post-acute care setting on the following patient-level outcomes: death within 30 days of hospital discharge, readmission within 30 days of hospital discharge, successful discharge to the community, and improvement in functional status during the post-acute care episode. To address the endogeneity of treatment choice, we use an instrumental variables approach, using as an instrument the differential distance between the beneficiary’s home ZIP code and the closest HHA and the closest SNF. The instrument passes

Using ordinary least squares regression, we find substantial differences in patient outcomes by discharge setting. Compared to patients discharged to SNF, patients discharged to home health have lower readmission and death rates (by 2.5 and 4.7 percentage points respectively), are much more likely to be successfully discharged to the community (by 25 percentage points) and experience improvement in functional status while in post-acute care (by 54.3 percentage points). These findings are consistent with selection; healthier patients are more likely to be discharged with home health. In the instrumental variable specifications which account for this selection bias, these results change. Patients discharged to home health are still more likely to be successfully discharged to the community, although the effect size is about half (13 percentage points). However, patients discharged to home health are also more likely to be readmitted to the hospital (by 5.5 percentage points). They are no more likely to die. The difference in functional status improvement favored home health but was not statistically different from zero. These preliminary results suggest there are important tradeoffs between home health and SNF care for patients needing post-acute care. While current policies may incentivize the use of lower-intensity settings (such as home health care)

Many developed countries face growing demographic pressures on their health and long-term care budgets. Improving the efficiency of these services has therefore become an important policy priority across the world. This paper examines the impact of reductions to public spending on adult social (long-term) care on the use of public hospitals in England in order to quantify the extent of substitution between the two types of publicly funded care. It exploits large reductions in public spending on adult social care in England between 2009 and 2015, a period when social care spending fell by 37% as part of widespread government austerity measures, but where public spending on health care was

The institutional features of the public health and social care system in England mean that local governments retain the responsibility to fund and organise public services for their local population. As a result, there was considerable geographical variation in the cuts to adult social care spending over the period. We exploit this variation to identify the impact of public funding for social care for individuals aged 65 and above on their use of public hospitals. We control for permanent differences in the use of hospitals across local authorities with the inclusion of area fixed effects, in addition to a rich set of local area characteristics to control for time-varying needs for hospital services. Our results indicate a small but statistically significant, negative impact of public social care spending on a number of measures of public hospital use. These include visits to emergency departments and admissions for inpatient care. The estimates imply an additional spend of $23 million on emergency treatment in 2015-16 relative to 2009-10 as a result of cuts to social care spending. When examining the effects by age, the estimates indicate that the magnitudes of the effects are largest for older individuals. However, the estimates indicate no statistically significant relationship between social care spending and delayed exits from the medical system, either as measured by official delayed discharge statistics or by the average

These results have important implications for policy. They suggest that the recent cuts to social care spending have led to a modest increase in public hospital use among the older population. This means that social care spending has a small fiscal externality on health spending. As a result, attempts to reduce overall public spending on health and social care have been less effective than appears when looking only at the reduction in social care spending. However, the results suggest that the cuts have not been the major driving force behind recent increases in emergency department attendances or the number of delayed discharges from the medical system.

In 1998, Medicare changed its payment method for post-acute care provided by skilled nursing facilities (SNFs) from a cost-based system to a per diem Prospective Payment System (PPS), with a goal to control increasingly high Medicare SNF expenditures. Under the PPS, the per diem rates are primarily based on therapy minutes, creating an incentive for SNFs to provide high-intensity therapy services. It is well documented that SNFs have exploited this reimbursement method by over-providing these therapy services to Medicare beneficiaries, regardless of their actual clinical needs. Accounting for more than 50% of the SNFs in the United States, corporate chains can affect affiliated facilities’ practice patterns. This paper aims to examine whether the selection of therapy treatment levels for SNF residents varies by chain ownership. Using a Difference-in-Differences model to compare independent and chain-acquired SNFs during the period from 2003 to 2009, I find that chain acquisition of independent SNFs are associated with about 2.92 percentage points increase in the proportion of residents with the highest therapy treatment level. A more dynamic model suggests that the chain effects on this aggressive billing practice may last for several years after the acquisition. In addition, I find that the main effects are mostly due to acquisitions by large chains and for-profit chains. However, the increase in treatment intensity among chain-affiliated SNFs does not lead to shorter length of stay. Overall, the findings suggest that government agencies should consider chain affiliation when monitoring Medicare reimbursements

Lacey Loomer [email protected]

Lacey Loomer [email protected]

: Created by the Centers for Medicare and Medicaid Innovation (CMMI), the Medicare Care Choices Model (MCCM) extends the benefits of hospice services to Medicare beneficiaries wanting to concurrently receive curative treatment. The MCCM aims to increase access, test a new payment model and improve quality of life and patient/family satisfaction. 141 hospice providers were selected by an expert panel in hospice and model implementation on the basis of being diverse in geographic area and having demonstrated experience of coordination and shared decision-making with beneficiaries and their families. Although hospices were positively selected into the program, CMMI randomly

: Demographics of providers in 2014 and 2015 from Hospice Utilization and Payment Public Use File, 2016 quality ratings from Hospice Compare, 2010-2017 deficiencies from Centers for Medicare and Medicaid Services Quality,

: This study employs two methods to estimate a causal effect of the MCCM on two measures of quality: quality ratings and deficiencies. First, we compare the Phase 1 hospice providers’ baseline characteristics (volume, geographic location, accreditation, patient characteristics) to the Phase 2 characteristics in order to establish that the randomization was sufficient. Next, we estimate the impact of the MCCM on seven Hospice Compare quality ratings using a generalized linear model (GLM) with a Gamma distribution and log link function, due to the skewed distribution of the ratings. Finally, we employ a difference-in-differences approach, which compares the deficiencies from

: The Phase 1 hospices were comparable in baseline characteristics to Phase 2 hospices. Compared to all hospices in the United States, they were larger and more non-profit. The GLM model found no statistically significant differences between Phase 1 and Phase 2 providers, with coefficient magnitudes close to zero for all seven quality ratings. Graphically examining the deficiencies per hospice provider surveyed from 2010-2016, the parallel trends assumption holds between Phase 1 and Phase 2 providers. Phase 1 providers had on average 1.07 deficiencies in the pre-period and 1.17 in the post-period for an increase of 0.10 from the pre- to post-period. Phase 2 providers had on average 0.82 deficiencies in the pre-period and 0.92 in the post-period, for an increase in 0.12. The difference-in-differences model found no statistically significant difference in deficiencies between Phase 1 and Phase 2 providers, pre-

: After one year of implementation, there is no impact of the MCCM on quality of hospice. Further information on the number of program enrollees and a longer time period of follow-up are necessary to estimating the impact

: Economic theory provides a useful framework for analyzing nursing home (NH) ownership, competition and signals about quality to consumers. NHs serve both private pay and Medicaid consumers, but Medicaid rates are set by the state so Medicaid demand is perfectly elastic. However, NHs can set private pay rates over Medicaid rates depending on the market structure. For-profit and corporate chain-owned NHs aim to minimize costs and set prices in order to maximize profits, often signaling low quality. While nonprofit and independent NHs have other objectives, such as altruism, which signal high quality. Despite the conflicting profit motives for nonprofit chain NHs, about 13% of the

Data include private pay rates from Cost of Long-Term Care in Connecticut reports, Medicaid rates from Connecticut Department of Social Services, facility characteristics from Online Survey and Certification Reporting System and

To assess market power, we calculate the percent private pay mark up over Medicaid rates. We estimate the association between ownership and price mark up at the NH-level using county-year fixed effects with robust standard

The sample includes a balanced panel of 153 Connecticut NHs with at least 5% private pay residents from 2013-2015. 75% were for-profit, of which 60% belonged to a chain. 25% were nonprofit, of which 20% of belonged to a chain. On average, price mark ups were 90% ($200). Analysis revealed no difference in price markup between nonprofit and for-profit NHs. Chain ownership was associated with a 14 percentage point increase (0.6 s.d.) in price mark up

NHs that are part of a chain are setting private prices higher above Medicaid rates than independent NHs, regardless of profit status and quality ratings. Previous estimates of residual demand elasticity in the NH market in the 1990s ranged from 1.7 to 3.85. Our elasticity estimate of 1.13 suggests that competition in the NH market has decreased over time. This decrease in competition could be due to the growth of chain owned NHs. While our single state analysis limits generalizability, we utilized a novel dataset with NH-level private pay prices linked with survey data across multiple years. Policy makers should advocate for more price and ownership transparency to better

Rashmita Basu [email protected]


Jaqueline M Oliveira [email protected]

The need for long-term care services and supports (LTSS) has been growing disproportionately among individuals with Alzheimer’s disease and related dementias (ADRD) due to increasing prevalence of the disease and lack of effective treatment therapies. Informal care (from friends and family members) is the primary source of LTSS for people with ADRD, but individuals with dementia also need formal LTSS (home/community based or institutional settings) as the disease progresses. While the use of different types of LTSS solely focused on costs, relative benefits of informal and formal LTSS may have differential impacts on health of care recipients. The goal of the current study is to investigate the causal effects of informal versus formal care on health and health outcomes of people with ADRD. The results from this study will address a major policy challenge the country is facing in regard to balance the use of different types of LTSS

Data came from the Health and Retirement Study (HRS) (2000-2014) and the subsample of the HRS, the Aging, Demographic and Memory Study (ADAMS). Dementia diagnosis was based on the modified version of the Telephone Interview of Cognitive Status (TICS) in the HRS and detailed neurological and clinical tests in the ADAMS sample. Separate analysis was performed for both samples to account for the sensitivity of dementia diagnoses.

Health and health outcomes include measures of physical, mental, emotional health, and healthcare utilization. Physical health includes functional disability (changes in activities of daily living, instrumental activities of daily living), mobility (difficulties in walking, getting across room, flights of stairs), self-rated health and mortality; mental health by depression (CESD- score). Diener’s measure of life-satisfaction captures emotional health.

Informal care is considered as an indicator variable if care recipients receive it from family members and formal care is measured by the use of home health or nursing home care. As providing informal care is likely to be endogenous,

The study sample includes 10,716 unique respondents from the HRS and about 856 respondents ADAMS sample members. There are significant differences in physical and mental health outcomes between those diagnosed with dementia in ADAMS sample versus those categorized as demented in the HRS. About 14% of respondents in the ADAMS sample used informal care compared to 7% of respondents in the HRS sample. Preliminary results from the two-stage residual inclusion method suggest that informal care was significantly associated with lower physical and mental health outcomes but higher emotional health in both samples, after controlling for individual level characteristics including chronic health. Ongoing analysis is investigating whether the receipt of informal care varies with and without the presence of formal care to demonstrate whether the presence of one form of LTSS affect the effectiveness of the other and whether

Building on the framework of Inequality of Opportunity (IOP, a.k.a. health inequality due to circumstances), we link the China Health and Retirement Longitudinal Study (CHARLS, HRS-sister study) with the newly released life history survey to quantify health inequality due to childhood circumstances for which they have little control. We evaluate comprehensive dimensions of health. Our analytic sample includes about 5,000 elderly Chinese between age 60

Using the Shapley value decomposition approach, we first show that childhood circumstances may explain around 40 percent of health inequality in old age across multiple health outcomes. Second, while both health circumstances (especially health and access to health care in childhood) and non-health circumstances (especially family socioeconomic status and regional and urban/rural status during childhood) contribute significantly to health

Our findings support the value of a life course approach in identifying the key determinants of health in old age. Distinguishing sources of health inequality and rectifying inequality due to early circumstances should be the basis

What role does family-provided old-age support play in alleviating the adverse effects of health shocks on the well-being of senior households? We investigate this question in the context of China, where access to public old-age pension and health insurance is relatively poor despite recent efforts to expand coverage. In the absence of formal insurance mechanisms, Chinese seniors rely heavily on children and other relatives to meet their material and instrumental needs.

Our empirical analysis draws on data from the 2011, 2013, and 2015 waves of the China Health and Retirement Longitudinal Study (CHARLS). CHARLS offers information on a broad array of individual and household economic and demographic characteristics, including detailed information on family networks and financial and time transfers to and from children and other relatives. To quantify how family support responds to elderly health status we utilize an indicator that captures the difficulty level of performing activities of daily living (ADL). The longitudinal feature of the data allows us to explore within-household variation in difficulties with ADL and, therefore, hold constant any time-invariant unobservable household traits correlated with health. Our models are also supplemented with village-year fixed effects to hold constant time-varying community-level characteristics that are potentially correlated with health

We first document that increasing difficulties with ADL significantly lowers labor supply and labor income for senior Chinese households. When health status changes so that individuals who could perform a certain ADL, say walking 100 meters, can no longer execute it (that is, ADL score increases by 3), the likelihood of working is 2.1 percentage points lower; hours worked also decreases by 7.5%. Turning to the main question of the paper, we find that financial assistance from daughters and other relatives increases when households experience increasing difficulties with ADL, 9.4% and 8.8% respectively. Sons do not provide more old-age support in response to health shocks. To measure the impact of health shocks on well-being, we run the same models on various measures of household expenditure. While we find that household per capita expenses with medical costs go up, expenditures with non-medical goods and services do not

We further investigate these findings by performing our analysis on households whose first reported job was in agriculture and non-agriculture separately. Our health shock measure, ADL, measures physical difficulties and thus we expect to find different results depending on the type of work performed. Overall, we find that households originally working in agriculture drive the results described above, suggesting that a decrease in physical ability is more damaging to

Norma Coe [email protected]

Tansel Yilmazer [email protected]

An estimated four to five million older adults in the United States are living with Alzheimer’s and related dementias (ADRD) (Langa, 2017), a chronic, progressive condition characterized by cognitive decline of sufficient severity to interfere with a person’s ability to carry out daily activities (Alzheimer’s Association, 2017). Despite recent evidence that prevalence rates of ADRD are declining (Langa, 2017), population forecasts indicate significant growth in the absolute number

Understanding the magnitude of the medical care and long-term care costs attributable to dementia is important for public and private decision makers, but estimating these costs has been difficult. First, identifying people with ADRD can be difficult in secondary data, since diagnosis can be at different stages of the disease progression or lacking altogether. Second, one must isolate the costs attributable to ADRD among a population that has several co-occurring chronic and acute conditions. Third, our fragmented health system means that many players are responsible for different types of cost; Medicare, Medicaid, and the family all play sizable roles in funding care for individuals with ADRD. We estimate the public spending on ADRD using newly available data from the Health and Retirement Survey matched to Medicare and Medicaid claims data. We identify a retrospective cohort of older adults with ADRD, and perform sensitivity analysis around the definition of dementia onset. We examine Medicare and Medicaid expenditures for the 12 months prior and up to 60 months following a diagnosis of ADRD. In order to isolate the costs attributable to ADRD, we select a comparison group of HRS participants matching on sex, birth year, and HRS entry year. To calculate the marginal effect of ADRD on Medicare expenditures, we use the estimator described by Basu and Manning (2010) for estimating costs under censoring. We estimate costs using a two-part model; the first part estimates the probability of any costs during each month using a logit model, while the second part estimates the magnitude of costs when costs are greater than zero using a generalized linear model with gamma family and power link of 0.95. This estimation is done separately on two samples: (1) months prior to death or censoring, and again (2) for months in which death occurs. An accelerated failure time model based on the lognormal distribution for time is used to estimate each subject’s survival function after accounting for censoring. We use the method of recycled predictions in order to estimate the

We estimate the costs attributable to ADRD, paying special attention to who bears these costs. We also examine how the total costs and the burden of costs has shifted over time and over the course of the disease.

This paper uses data from the Health and Retirement Study to investigate when the older homeowners suffering from late-life disability exit from homeownership and how this exit influences their total wealth and mortality. There has been a growing emphasis on “aging in place.” Data show that elderly rarely downsize their houses or move unless a drastic event such as an illness or death of a spouse occurs. Housing equity is the most important asset in the portfolios of large fraction of older Americans. Housing directly provides utility, and there are transaction costs associated with the purchasing and selling a house. The expanded life-cycle models have shown that bequest motives, as well as health and medical risks, are the driving forces of the puzzling phenomena. Aging in place in poor health, on the other hand, might require expensive home-based care. Obtaining care at home might cause reduced quality of care leading to increased mortality. If the current housing lacks basic accessibility features, it would also prevent disabled older adults from living safely in their home. Within the scope of benefit-cost framework, we propose that elderly with declining health and functional capacity should exit homeownership (i.e., move to a nursing home, to a retirement facility or move in with a relative) when the expected cost of “aging in place” outweighs the expected benefits. Measurement of health status at older ages is complex. No single indicator fully captures all aspects of health. We focus on functional disability, which reflects restrictions in carrying out specific activities. We measure late-life disability using limitations with six activities of daily life (ADLs) including walking across the room, bathing, dressing, eating, getting in/out bed, and toileting and five instrumental activities of daily life (IADLs) including using telephone, managing money, taking medication, shopping for groceries, and preparing hot meal. Around 11 million Americans report difficulty with performing one or more ADLs or IADLs, and about half of this population is over the age 65. Difficulties with

Findings from fixed effects models show that older homeowners are less likely to move unless they experience severe difficulties with ADLs, measured as difficulties with five or six ADLs. On the other hand, elderly needing help with two or more IADLs are more likely to move. When older households with diminished functional capacity move, they are less likely continue to be homeowners, and experience sharp drops in housing and total wealth. We did not find any increases in financial assets and non-housing wealth following the move and exit from homeownership. The decline in total wealth and increase in out-of-pocket health care expenditures generate lower bequest intentions for those who exit from homeownership. Upon leaving homeownership, there are some gains in mortality for those having difficulties with IADLs, but not for those having difficulties with ADLs. Our findings have significant implications for


Neha Bairoliya Complete

Johns Hopkins Emmanuel Garcia Morales Complete

Complete

Harvard Center for Population and Development Studies


Mark Kattenberg; Ab van der Torre Complete

University of Texas Medical Branch Complete

CPB Netherlands Bureau for Economic Policy Analysis

Daniel Jupiter; Sapna Kaul; Claire de Oliveira; Mariana Chavez-MacGregor

Albert Wong; Arjen Hussem Complete

Matt Sutton; Andrew Jones Complete

Auburn University Jie Zhang; Diana Alessandrini Complete

Erasmus School of Health Policy & Management, Erasmus University Rotterdam

Manchester Centre for Health Economics, University of Manchester

University of Pennsylvania Norma Coe; R. Tamara Konetzka Complete

Institute for Fiscal Studies Ben Zaranko; Rowena Crawford Complete


Brown University Complete

Brown University Complete

Texas Tech University Health Sciences Center Complete

Yale University Thomas Gill; Binjian Yan Complete

Rhodes College Complete

University of Pennsylvania Lindsay White Complete

The Ohio State University Patryk Babiarz Complete


Maternal and Child Health




Do Twin Studies Underestimate the Cost of Low Birth Weight?

The Effect of Criminal Justice System on Children of the Incarcerated

Estimating the Effects of Free School Meal Provisions on Child Health: Evidence from the Community Eligibility Provision

Cesarean Section and Children’s Health: Evidence using a Quasi-Experimental Design




Transition Home Plus Program Reduces Medicaid Spending and Health Care Utilization for High-Risk Infants

Utilization with High Out-of-Pocket Costs: Evidence from In-Vitro-Fertilization Treatment

Exploring National Trends of Patient and Family Centered Care among US Children




Parental Investments in Response to Early Life Endowment

IMPACT OF THE QUALITY OF FAMILY PLANNING SERVICES ON MODERN CONTRACEPTIVE PREVALENCE

Long Commute to Work during Pregnancy and Infant Health at Birth

Maternal and Child Health Inequality, Relative Income and Newborn Health




Willingness to pay for child health screening: Evidence from lead poisoning prevention in Illinois

The Effects on Outcomes and Cost Savings from the Provision of Free Long-Acting Reversible Contraceptives

Parental education and non-cognitive skills in children: evidence from a UK schooling reform




The Impact of Access to Prenatal Care on the Benefits of Next Generation: Using the CHIP Unborn Child Option

Investing in the Womb: Identifying Gender Discrimination through the Lens of Prenatal Ultrasounds

Smog and Suicidal Behaviors: Evidence from School-Age Children

Abstract

Low birth weight infants are at a higher risk of infant mortality, although there is debate as to how much of this relationship is causal. Low birth weight may be correlated with infant health even in the absence of a strong causal effect if both low birth weight and other measures of infant health are determined by a third factor. To overcome these challenges, many studies use twin designs that include birth-set fixed effects. Twin studies have high internal validity, but at the potential cost of external validity since it is not clear that twin studies generalize to the singleton population. Twins may be smaller because their optimal birth weight is less than that of a singleton. A 2-kg singleton would be classified as low birth weight and a marginal increase in birth weight may improve infant health, whereas a 2-kg twin would be close to average and may be at her optimal birth weight. Little is known about how the effects of birth weight vary by plurality, which severely limits how informative twin studies can be for public policy. To overcome these challenges, this paper proposes a novel approach to evaluate whether twin studies provide under-, over-, or unbiased estimates of the effects of birth weight on singletons by comparing causal estimates of low birth weight for twins, triplets, and quadruplets. Many of the differences that exist between singletons and twins also exist between twins and triplets, as well as between triplets and quadruplets. For example, the average birth weight of singletons, twins, triplets, and quadruplets are 3.4 kg, 2.4 kg, 1.7 kg, and 1.3 kg, respectively. This study uses data from the universe of multiple births linked to infant death records in the United States from 1995-2000 to estimate how birth weight causally affects infant health. I find that low birth weights increase infant mortality and decrease APGAR scores for twins, triplets, and quadruplets. Although triplets and quadruplets are on average smaller, they experience smaller reductions in infant health when they are LBW. These results suggest that twin studies underestimate the effect of LBW for the general population. In addition to the fixed-effects model, I present cross-sectional estimates of the effect of birth weight on infant health for singletons and twins using a rich set of controls. These estimates are consistent with births from higher pluralities experiencing smaller reductions in health when they are low birth weight. I then estimate the implied infant mortality reduction from increasing all birth weights to at least 2.2 kg. Estimates from twins, triplets, and quadruplets suggest that increasing birth weights would reduce the infant mortality rate by 1, 0.65, and 0.3 deaths per 1000 live births. These estimates suggest that the true decrease in infant mortality for singletons would be approximately 1.3 deaths per 1000 live births, implying that the twin estimates underestimate the statistical value of lives saved from increasing birth weights by at least $8 billion per year.

The effect of criminal justice system involvement on the children of the incarcerated is not well understood. There have been no evaluations examining whether pre-trial confinement worsens child health. We study how specific decisions by the courts and other criminal justice agencies affect child wellbeing during follow-up. We use North Carolina (NC) statewide administrative criminal court and incarceration data for the years 2005-2016 and link parents to their biological children through birth records. Our study uses instrumental variables (e.g., prosecutor’s prosecution rate and judge’s conviction rate) to examine whether holding other factors constant, harsher penalties for criminal offenses imposed by the courts (and also more restrictive pretrial confinement terms) affect children’s health and/or Medicaid spending. We will present preliminary findings.

Over 45 million children participate in the National School Lunch Program (NSLP) and School Breakfast Program (SBP) each year. For many, the free meals they receive in school are their only nutritionally sufficient ones of the day, and for the most at-risk children, it may be their single reliable source of food. A growing literature across the fields of economics, public health, and medicine has linked child nutrition deficiencies and food insecurity to negative health, education, and behavioral outcomes. Alternatively, very little work has been done to estimate the effect that food provision programs like the NSLP and SBP have on the health of disadvantaged children. Utilizing variation in the NSLP and SBP following the passage of 2010’s Healthy Hunger-Free Kids Act (HHKA), this study estimates the effect of providing children with free meals in school on multiple outcomes related to child health across different socioeconomic groups. While the HHKA broadly affected school nutrition across multiple dimensions, one of its main additions was the Community Eligibility Provision (CEP) which allows qualifying schools and districts to offer their students free lunch and breakfast regardless of each child’s personal eligibility. CEP eligibility is determined at both the school and district level through an Identified Student Percentage (ISP) which is the proportion of students who are predicted to be eligible for the NSLP and SBP. Alternatively, students attending non-CEP schools must have their family qualify and apply for the programs directly. At this time, no existing studies have estimated the CEP’s effect on child health. This study uses data regarding eligibility and participation in the CEP program at the school and district levels in combination with the restricted use Early Childhood Longitudinal Study, Kindergarten class of 2010-2011 (ECLS-K:2011) which contains data on child health, education, school characteristics, and family behaviors. The restricted use ECLS-K:2011 if well suited for the study, as it can be linked to schools directly. The study uses CEP eligibility at the school-level as an instrument for a child’s participation in the NSLP and SBP. Outcome variables of interest include parent-reported child health, body composition, prevalence of different chronic and acute illnesses, mental health, and behavioral development. The results of this study provide valuable insight for researchers and policymakers concerned with the effects of child-focused nutrition assistance programs on various health outcomes. Currently, there is little causally interpretable evidence regarding the effect of increased food provisions on child health, and no evidence exploiting variation from the CEP. Disparities in nutrition and food security among children both between and within various socioeconomic groups may produce lifelong gaps in health, education, and economic outcomes. This study evaluates the degree to which the two most common school based nutrition assistance programs correct for these disparities and identifies the subgroups for which the programs produce the best and worst results. With these findings, policymakers can alter existing policies and design supplemental programs which target children who are still at the greatest levels of risk.

The prevalence of inflammatory child health conditions, such as asthma, eczema and food allergy, as well as neurodevelopmental conditions such as attention-deficit disorder and autism, and their associated costs have increased rapidly over the last 20 years, becoming public health and policy concerns. While environmental factors likely underpin these increases, recent studies rely only on associational methods. Cesarean section as a method of obstetric delivery is an understudied environmental factor that increased dramatically in the period of interest, and has been linked to child health outcomes via multiple biological mechanisms. We combine 22 years of birth cohort data from the National Survey of Children’s Health with cesarean section rates generated for subgroups based on state, sex, race, and Hispanic-origin. Then, we employ a quasi-experimental fixed effects design to estimate the effects of cesarean section on rates of asthma, eczema, food allergy, attention-deficit disorder and autism. We find that cesarean section significantly predicts food allergy and autism, with qualitatively impactful implications for each.

Research Objective: To evaluate the effects of a transition home intervention on total Medicaid spending, emergency department (ED) visits, and unplanned readmissions for preterm infants born at ≤ 36.6 weeks gestational age and high-risk full-term infants. Study Design: Rhode Island Medicaid claims data was used to study the 321 infants cared for in the neonatal intensive care unit (NICU) for ≥ 5 days, who were enrolled in the transition home plus (THP) program. The THP program incorporated support services both pre- and post-NICU discharge provided by social workers and family resource specialists (FRS) working with the medical team between October 2012 and October 2014. The THP infants were compared with 365 high-risk infants born and admitted to the NICU in 2011 prior to the full launch of the THP program. Intervention and comparison group outcomes were compared in the eight 3-month quarters after the infant’s birth. Propensity score weights were applied in regression models to balance demographic characteristics between groups. Regression analyses with quarterly fixed effects were run to determine the impact of THP on total Medicaid spending, ED visits, and unplanned readmissions. Population Studied: The THP cohort included prospectively enrolled Rhode Island resident high-risk infants who were born early (< 32 weeks), moderate (32-33 weeks), late preterm (34-36.6 weeks), or full-term (> 36.6 weeks) in terms of gestational age. All the THP infants were on Medicaid (fee-for-service or managed care), and were hospitalized for ≥ 5 days in an 80-bed single room Level 3-4 NICU between October 1, 2012 and September 30, 2014. The comparison group included infants on Medicaid hospitalized in the same NICU for ≥ 5 days in the year prior to the study period. Principal Findings: Infants in the intervention group had significantly lower total Medicaid spending, fewer ED visits, and fewer readmissions than the comparison group. The Medicaid spending savings for the intervention group were $4,591 per infant per quarter (90% CI: −$8,397, −$785) in our study period. The average quarterly difference estimate for ED visits is a decrease of 334 visits (90% CI: −389, −279) per 1,000 patients relative to the comparison group for the first eight quarters after birth, weighted by the number of intervention patients in the quarter. The intervention group is 7.6 percentage points (90% CI: −12.3, −2.9) less likely to have an unplanned readmission during the first eight quarters after birth. Sensitivity analyses looking at Medicaid claims post discharge, which excludes all the NICU-related health care expenses and utilization, showed the results remained largely the same as the main analyses. Conclusions: Transition home support services for high-risk infants provided by social workers and family resource specialists working with the medical team can reduce Medicaid spending and health care utilization. Implications for Policy, Delivery or Practice: Expansion of the medical team to include social workers and FRS could offer a cost-effective approach for clinicians and policy makers to consider in addressing the psychosocial needs of families caring for preterm and high-risk full-term infants.

Does insurance coverage of medical treatments with high out-of-pocket costs affects patients’ utilization. We exploit a policy intervention that mandates coverage for In-Vitro- Fertilization (IVF) –an expensive infertility treatment with low success rates in one cycle of treatment– in private health insurance in the US. Mandated coverage varies from one cycle of treatment in some states to unlimited cycles in some others. Patients’ might increase their chances of conceiving an infant by more aggressive treatments, resulting in risky and costly multiple births. We provide the first estimate of the effects on adverse outcome of aggressive treatments from number of IVF cycles covered in mandated health insurances. We use a Generalized Synthetic Control framework to estimate causal effects. Our estimated effects varies from 0.31 percentage points decrease in share of multiple births in states with only one covered cycle to more than 35 percentage points increase in states with unlimited coverage. Our estimates of effects of mandated IVF coverage on adoption –the main alternative for IVF patients with low chances of success– furthermore shows that adoption rates in states with more covered cycles is lower. These findings suggests that high out-of-pocket costs has strong behavioural responses from patients. In states with more coverage, more patients with low chance of success –who would prefer aggressive treatments– use the treatment. These patients otherwise would have adopted a child. Our findings have important implications for designing policy interventions to increase accessibility of expensive and technologically advance medical treatments while simultaneously decreasing utilization costs.

Background: Patient and family centered care (PFCC) is the hallmark of high quality pediatric care. We explored national trends in the receipt of high quality patient-communication and patient empowerment through behavioral health counseling among the general US child population. Methods: We used data from the Medical Expenditure Panel Survey (pooled cross-section) from 2010-2014. We employed two measures of PFCC: 1) a composite measure of high quality patient-physician communication (n=34,629) and 2) patient empowerment through behavioral health counseling about healthy eating (n=36,527) and exercise (n= 38,318). We used multiple logistic regressions to estimate the variation of receiving PFCC by social determinants of health (SDOH) and time trend. Results: Rates of receiving behavioral health counseling about healthy eating (53-60%) and exercise (37-42%) were lower than the rate of receiving high quality physician-patient communication (92-93%). Parents were significantly more likely to report receiving high quality physician-patient communication in 2014 than in 2010 (OR 1.37, CI 1.08-1.67); however no association was found for empowerment through behavioral health counseling. Low income and parental educational attainment, and lack of insurance were associated with a lower odds of receiving behavioral health counseling. Conclusion: Results showed significant variation of physician-patient communication and empowerment by SDOH. Evidence also suggested that providers need to take the next step and begin empowering parents and their children to self-care.

A large literature documents the effects of early life health shocks on long run outcomes including education, health and earnings. However, the role of parents in reinforcing initial health inequities or mitigating them is not well understood. Furthermore, most studies do not study investments over a life course and look at a single dimension of response such as education or health and ignore non-human capital transfers. We fill these gaps in the literature by studying parental responses over multiple stages of the life course and for multiple dimensions, including non-human capital transfers. We utilize the overlap of Ramadan fasting month with pregnancies for Muslims as a natural experiment for studying parental responses to fetal under nutrition. Ramadan is a holy month in the Islamic calendar. Practicing Muslims are required to observe a strict fast from sunrise to sunset for a month. Since about seventy five percent of all pregnancies overlap with Ramadan, it is estimated more than 1.2 billion Muslims were potentially exposed to their mother’s fasting in utero. Our work builds on some recent studies which show the negative effects of Ramadan exposure in utero on birth weight, cognition and labor market outcomes later in life. We study Muslims in Indonesia, the largest Muslim majority country in the world. Nationally representative data from the 2007 wave of the Indonesian Family Life Survey (IFLS 4) is used to study parental investment response on human capital with detailed expenditures on health and education and on non-human capital transfers (e.g. dowries) over a life cycle. Results show that for children younger than five years, parents are less likely to get their exposed (during Ramadan in utero) children vaccinated and invest less in their diet. However, we find an opposite pattern for older children, with parents investing more in vitamins or supplements but we do not find any differences between exposed and non-exposed children on educational investments. However, our results show that compared to unexposed children, mothers of exposed children have lower expectations about future educational attainment of their children. In contrast to most papers in the larger literature, we also look at non-human capital responses to fetal health shocks and find gender differences in parental response. Females get fewer dowries and bring fewer assets into marriage, an important marker for future bargaining power. For male adults, however, parents are more likely to provide monetary and non-monetary help attempting to mitigate inequities. We do not find any differential responses on prenatal care. Overall our results suggests, parents make complimentary investments in response to initial health endowments particularly for children younger than five years and mostly in the form of investing in their children’s health care but also in terms of dowries for their daughters. These results have important policy implications: if its not feasible to nudge pregnant women from unhealthy cultural practises during pregnancy, one can instead focus on health care and physical capital transfers early and later in life which may help mitigate the initial health inequities.

INTRODUCTION: This study evaluates the association between family planing (FP) service quality and modern contraceptive prevalence (MCP) in Ethiopia. Reducing unmet need for FP in low and middle income countries (LMICs) can reduce child and maternal mortality and disabilities through improved birth spacing, education, women’s empowerment, and reduction of unwanted pregnancies, poverty, and hunger. Evidence regarding the effect of FP service quality (FPSQ) on MCP has been slow due to the challenge of obtaining measures of FPSQ. The Bruce-Jain framework provides one of the earliest comprehensive literature reviews and theoretical frameworks describing the dimensions of FPSQ. Our study uses a potentially generalizable method of data reduction by using an expert panel to score potential variables linked to FP service quality and to prepare a smaller set of FPSQ indices that are in keeping with the Bruce-Jain framework. METHODS: We used Performance, Monitoring and Accountability data from 2015 to assemble a cross-section dataset for Ethiopia with enumeration area (EA) level data from the service delivery points (SDP) and woman of reproductive age household (HH) surveys. We used multivariate ordinary least square regression models with Huber/White robust standard errors clustered at the region level. The dependent variable was mean MCP and the independent variables were six different FPSQ Bruce scores at the EA level. The six different FPSQ Bruce scores were tested combined and individually in the models. Control variables included socioeconomic status (SES), marriage, husband cohabitation, and education status, age groups, and urban. EA level MCP and controls were calculated by averaging across individual female respondents within an EA. RESULTS: Among the 215 EA-observations in Ethiopia, 44% of the women were between 15 and 24 years old, 58% were married, out of those married 92% cohabited with their husband, 29% had a high school education, and 27% (EA range: 0%-71%) were a current user of a modern contraceptive method. Regression results show that for every unit increase in the EA mean FPSQ Bruce 1 score (on choice of method) and Bruce 4 score (on provider-patient interpersonal relations), the mean MCP increased by 21.4% (p<0.01) and by 18.0% (p<0.01) respectively. Bruce scores 2, 3, 5, 6 did not show statistically significant association with EA level MCP. Results were robust to multiple specification tests. CONCLUSION: Multivariate statistical analysis using PMA data from Ethiopia showed that two out of six of the Bruce domains of FPSQ are linked to modern contraceptive prevalence. These results show that efforts to strengthen FP service quality have the potential to significantly increase contraceptive use in LMICs. Likewise, this results may help child and maternal health policymakers advocate for increased investment in interventions that improve FPSQ in LMICs. Future extensions to our analysis will exploit the panel nature of data with three waves and multiple countries to adjust for fixed effects.

In the United States, according to the most recent census data, approximately 2.2 million workers travel at least 50 miles each way between their homes and workplaces, and about 1.7 million workers spend 90 minutes or more commuting in each direction. These long commutes can be physically and mentally demanding, particularly for pregnant women. In this regard, we conduct the first empirical study to examine the health impact of long commute to work during pregnancy on fetuses and infants at birth, using unique data that contain information on not only a woman's home address but also her employer's address during her pregnancy. Our study is also the first to examine the health impact of the chronic strain induced by long commute on fetuses and infants at birth, adding new evidence to the literature. We find that among long-distance commuters, increasing the maternal travel distance during pregnancy by 10 miles is associated with increases in low birth weight and intrauterine growth restriction by 1.0 and 0.6 percentage points, or 25 and 46 percent compared with their means, respectively. In addition, we provide evidence on two possible mechanisms underlying the adverse health outcomes associated with long commute: elevated stress levels of pregnant women who are long-distance commuters, and under-utilization of prenatal care. We examine the presence of long commute induced maternal stress during pregnancy by showing that the likelihood of using c-sections increases among male babies, but not among female babies, born to women who travel long distance to work during pregnancy. Such an increase is consistent with the finding in the medical literature that male fetuses are more sensitive to stressors in utero than female fetuses, resulting in higher likelihood of delivery complications that require c-sections. With regard to prenatal care, we find that among long-distance commuters an increase of 10 miles in maternal travel distance during pregnancy could reduce the number of prenatal visits by 2.53 percent, decrease the probability of the mother’s completing her first prenatal visit within the first trimester by 2.3 percentage points, and increase the probability of the mother's completing her first prenatal visit within the third trimester or having no prenatal visit at all by 1.4 percentage points, all indicating an under-utilization of prenatal care. In addition, our study is the first to calculate the travel distance according to existing public roads instead of using the commonly computed geodetic distance, which represents the length of the shortest curve between two points on earth. Using geodetic distance can incur a greater attenuation bias, potentially leading to a claim of no adverse effect of long commute when the adverse effect could be detected by a less noisily measured commuting distance. Our study has important implications for public policy proposals that consider expanding maternity leave to cover the prenatal period, particularly in the context of the United States. Having the needed time off during the prenatal period can be crucial for pregnant women who are long commuters to alleviate stress and adequately utilize prenatal care.

A major concern over rising inequality is its potential to reduce intergenerational mobility, leading to even greater inequality in the next generation. We estimate the impact of rising inequality over the period 1970-2010 on offspring health at birth, a measure of human capital that has been shown to be highly correlated with future education, IQ and income. We define inequality both at the aggregate level and at the individual level: as a group-level measure (the Gini coefficient for each state or county), and as individual level measures of relative income (relative deprivation, rank, and relative income distance). We document a strong negative relationship between the Gini and newborn health in the cross section, but find that including a modest set of controls, or limiting variation to changes in inequality over time within an area, or instrumenting for inequality eliminates the relationship between the Gini and newborn health completely. However, this null result likely reflects heterogeneity in the effect of rising inequality. When we estimate the impact of relative income on newborn health, we find negative and significant effects of having relatively less income than one’s neighbors on birth weight, even after controlling for area fixed effects and instrumenting for differences in the income distribution.

Childhood lead exposure carries high life-long private and public costs, including reduced IQ and educational attainment and an increased risk of criminal activity. A blood lead screening test is a secondary prevention measure that identifies exposure. A robust body of literature documents disparities in utilization of preventative care, such as immunization coverage, across socioeconomic and racial groups in the US. This literature suggests that several barriers might decrease uptake of preventative care, including high perceived costs and low perceived benefits. On the cost side, lack of information, scheduling challenges, and transportation costs appear to contribute to vaccine delay among low-socioeconomic families. On the benefit side, disease outbreaks are associated with increased immunization rates, likely due to increased salience of disease risk. This paper considers costs and benefits from screening in a unified setting in order to estimate parents’ willingness to pay (WTP) for blood lead level screening and how that WTP varies with latent exposure risk, lead hazard salience, and parents’ characteristics. This paper uses blood lead screening data from 5.4 million tests performed in Illinois from 1997-2016. By linking these test data with birth and death records for almost five million children born in Illinois between 1991 and 2016, I will be able to estimate a model for screening demand as a function of five factors. First, I observe family characteristics in the birth records, but since around 30% of mothers with two children only screen one of the two children, family background might not fully explain disparities in screening. Second, I estimate exposure risk from housing age and pollution sources based on birth address. Third, I estimate provider quality based on observed adherence to screening guidelines and screening rates in providers’ catchment areas. Using distance to providers as a proxy for cost of screening, a procedure common in health access literature, I trace out WTP for screening. Fourth, I study how access to remediation funding impacts demand for screening. Last, I exploit the long sample period to assess how changes in the salience of lead hazards due to “lead scares” affect the WTP estimates. While federal guidelines mandate that all children on Medicaid are screened for lead poisoning at ages one and two, guidelines for non-Medicaid children vary by state and local health department. Illinois requires screening for all children living in high risk zip codes; risk is assessed by housing age and population characteristics. Screening rates in high risk zip codes have been stable at 80% for recent birth cohorts, well above the 56% screening rate for zip codes that do not require universal screening. Yet, it seems important to understand why 20% of young children living in high risk areas are not screened. This is especially relevant as, in the wake of Flint’s lead disaster, some states, including Illinois, are considering switching to a universal screening system like those in Massachusetts and Rhode Island. This analysis aims to sheds light on the relative role of different factors in determining demand for screening.

More than half of pregnancies among women 15-24 in the US are unintended. Long-acting reversible contraceptives (LARC) are extremely effective contraceptive methods with failure rates of less than one percent. LARC use, however, has been hindered by their high cost and lack of acceptance. In early 2009, the Colorado Family Planning Initiative (CFPI) sought to address these access barriers by providing free LARCs and provider training at Title X–funded clinics, which serve a low-income population. The objective of our study was to estimate the causal impact of the CFPI on unintended pregnancies and its impact on Medicaid savings over five years. We used a propensity score weighted difference-in-difference research design to account for unobserved contemporaneous trends in fertility rates using a group of non-Colorado counties as controls. The primary outcome was birth rate per 1000 for women ages 15-24. We obtained birth rates for all US counties from the Centers for Disease Control. We supplemented these data with information on county demographics, educational attainment, income, and unemployment rates from the Area Health Resource Files. The sample period included four years of pre-CFPI implementation and five years of post-CFPI implementation. We first selected controls counties by restricting comparison counties to those within thresholds based on key demographic characteristics, and the then estimated propensity scores with the remaining counties. Based on our model estimates and estimates of the number of women actually affected by the initiative, we calculated Medicaid cost savings. These estimates accounted for mother and child spending, also considering eligibility, attrition, and reimbursement rates over five years post-CFPI. Our final sample consisted of 128 Colorado counties and 204 comparison counties. Baseline characteristics of Colorado and comparison counties were similar and the standardized differences were within the range of acceptable balance, with the exception of percent black females ages 15-19 and percent black female, which were slightly lower in Colorado counties. Our results show that fertility rates decline in both Colorado and comparison counties in the post-period, but Colorado experienced a larger and statistically significant decline of 8.15 births per thousand or about 2900 births averted. Thus, roughly half of the observed decline in fertility rates can be attributed to the CFPI. Our analysis also shows that Medicaid saved approximately $40 million over five years. The CFPI significantly reduced unintended pregnancies in Colorado over and above observed declines in fertility rates across the country during the same time period. Removing LARC cost barriers for low-income women has the potential to reduce the large number of unintended pregnancies that still remain and are increasingly concentrated among poor and low income women. In addition, the initiative reduced state and federal spending on medical care, potentially outweighing the cost of the program.

The importance of non-cognitive skills in shaping later-life outcomes has long been recognised in the psychology literature. However, for economists this is a relatively new and underexplored phenomenon. As the body of evidence continues to grow and the consensus surrounding their importance begins to permeate fields, the need to better understand the determinants of non-cognitive skills becomes ever more apparent. This paper investigates the impact of parental education on a child’s non-cognitive skills. We use data from the National Child Development Study (NCDS) to create measures of the child’s Conscientiousness and Neuroticism. These measures are created using responses, given by the child and their teacher, to questions related to the child’s behaviour. We explore exogenous variation in parental education, induced by the 1947 schooling reform, to identify the causal effect. This reform was announced under the 1944 Butler Act, resulting in an increase in the minimum school leaving age from 14 to 15 in the UK in April 1947. We found that this reform increased the average years of schooling by 3.9 months and 4.7 months for fathers and mothers in our cohort respectively. We utilise an Instrumental Variable framework, using dummy variables to indicate whether the parents were impacted by the reform as instruments for parental education. We restrict the study to those parents born just either side of the reform, and obtain the local average treatment effect. By comparing observations that lie closely on either side of our threshold, we are able to eliminate certain trend effects. Findings based on data from the NCDS suggest that increasing the school leaving age by 1 year has no significant effect on the child’s Conscientiousness or Neuroticism. These results suggest that although parental education may play an important role in shaping a child’s capabilities, it has no impact on a child’s non-cognitive skills.

This paper studies the effect of access to prenatal care on children's health, exploiting the public health policy, so-called the CHIP unborn child option, that enables pregnant noncitizens to get prenatal care regardless of legal immigration status. Using state-level variation in whether to opt in and the timing of policy adoption, I find that eligible female noncitizens increase public health insurance coverage rate and subsequently increase the number of a doctor's office visits in 12 months previous to the survey. For children's outcome, the key result is a decline in the presence of chronic health conditions at ages 4-6, which may result in long-lasting health problems during their entire life. Also, the school attendance rate increases among children who were eligible in utero, implying that CHIP unborn child option enhances children's health and increases their pre-school or kindergarten attendance rate. This paper finds that the guaranteed availability of prenatal care can induce substantial benefits that persist in children's health.

In utero is a critical period of human development during which parents act on children’s behalf in health investments. These investments may have a profound impact on the life trajectory of a child. We investigate whether parents in China who choose to carry the pregnancy to term allocate resources differently between their sons and daughters over the course of pregnancy after the sex of the child is disclosed to parents. Using unique and large-scale hospital electronic records of prenatal ultrasound scans and birth outcomes as well as a longitudinal survey of parents’ health behavior during pregnancy, we estimate how parental health behaviors and prenatal health investments change after parents gain access to gender information from post-20 gestational week ultrasound scans. In addition to the state-of-the-art difference-in-differences model, we employ a novel fetus fixed effect model to identify shifts in prenatal investments when information on child gender is disclosed. We document sex-selective prenatal investments as an early channel through which parents practice discriminatory behavior. We show that parents favorably shift certain parental health investments when pregnant with a boy. Specifically, the chance of exposure to passive smoking decreases while more mothers take nutrient supplements when parents expect boys compared to girls after receiving a post-20th gestational week ultrasound scan. Preferential prenatal treatment of males is greater for areas with stronger son preference. A set of key placebo tests using pre-pregnancy and early pregnancy behaviors reassure us that our identified effects are likely causal. Our findings have implications for eliminating gender discrimination and improving maternal and child health in the earliest stage of life. These findings also call for utilizing the window of opportunity during pregnancy to more effectively promote smoking cessation.

Previous studies evaluating the welfare cost of air pollution have not paid much attention to its potential effect on suicidal behaviors. This paper attempts to fill the gap by estimating the effect of air pollution on suicidal ideation and suicide attempts among the school-age children. We use changes in the local wind direction as instruments for air pollutant concentrations to address endogeneity and measurement errors. By matching a unique youth risk behavior survey in China with rich air quality and weather conditions according to the exact date and location of each interview, we find that fine particulate matter (PM2.5) increases the rate of both suicidal ideation and suicide attempts. Heterogeneity analysis reveals that boys who have bad school performance are more affected in terms of suicidal ideation.

Abstract

Low birth weight infants are at a higher risk of infant mortality, although there is debate as to how much of this relationship is causal. Low birth weight may be correlated with infant health even in the absence of a strong causal effect if both low birth weight and other measures of infant health are determined by a third factor. To overcome these challenges, many studies use twin designs that include birth-set fixed effects. Twin studies have high internal validity, but at the potential cost of external validity since it is not clear that twin studies generalize to the singleton population. Twins may be smaller because their optimal birth weight is less than that of a singleton. A 2-kg singleton would be classified as low birth weight and a marginal increase in birth weight may improve infant health, whereas a 2-kg twin would be close to average and may be at her optimal birth weight. Little is known about how the effects of birth weight vary by plurality, which severely limits how informative twin studies can be for public policy. To overcome these challenges, this paper proposes a novel approach to evaluate whether twin studies provide under-, over-, or unbiased estimates of the effects of birth weight on singletons by comparing causal estimates of low birth weight for twins, triplets, and quadruplets. Many of the differences that exist between singletons and twins also exist between twins and triplets, as well as between triplets and quadruplets. For example, the average birth weight of singletons, twins, triplets, and quadruplets are 3.4 kg, 2.4 kg, 1.7 kg, and 1.3 kg, respectively. This study uses data from the universe of multiple births linked to infant death records in the United States from 1995-2000 to estimate how

I find that low birth weights increase infant mortality and decrease APGAR scores for twins, triplets, and quadruplets. Although triplets and quadruplets are on average smaller, they experience smaller reductions in infant health when they are LBW. These results suggest that twin studies underestimate the effect of LBW for the general population. In addition to the fixed-effects model, I present cross-sectional estimates of the effect of birth weight on infant health for singletons and twins using a rich set of controls. These estimates are consistent with births from higher pluralities experiencing smaller reductions in health when they are low birth weight. I then estimate the implied infant mortality reduction from increasing all birth weights to at least 2.2 kg. Estimates from twins, triplets, and quadruplets suggest that increasing birth weights would reduce the infant mortality rate by 1, 0.65, and 0.3 deaths per 1000 live births. These estimates suggest that the true decrease in infant mortality for singletons would be approximately 1.3 deaths per 1000 live births, implying that the twin estimates underestimate the statistical

The effect of criminal justice system involvement on the children of the incarcerated is not well understood. There have been no evaluations examining whether pre-trial confinement worsens child health. We study how specific decisions by the courts and other criminal justice agencies affect child wellbeing during follow-up. We use North Carolina (NC) statewide administrative criminal court and incarceration data for the years 2005-2016 and link parents to their biological children through birth records. Our study uses instrumental variables (e.g., prosecutor’s prosecution rate and judge’s conviction rate) to examine whether holding other factors constant, harsher penalties for criminal offenses imposed by the courts (and also more restrictive pretrial confinement terms) affect children’s health and/or Medicaid spending. We will present preliminary findings.

Over 45 million children participate in the National School Lunch Program (NSLP) and School Breakfast Program (SBP) each year. For many, the free meals they receive in school are their only nutritionally sufficient ones of the day, and for the most at-risk children, it may be their single reliable source of food. A growing literature across the fields of economics, public health, and medicine has linked child nutrition deficiencies and food insecurity to negative health, education, and behavioral outcomes. Alternatively, very little work has been done to estimate the effect that food provision programs like the NSLP and SBP have on the health of disadvantaged children. Utilizing variation in the NSLP and SBP following the passage of 2010’s Healthy Hunger-Free Kids Act (HHKA), this study estimates the effect of providing children with free meals in school on multiple outcomes related to child health across different socioeconomic groups. While the HHKA broadly affected school nutrition across multiple dimensions, one of its main additions was the Community Eligibility Provision (CEP) which allows qualifying schools and districts to offer their students free lunch and breakfast regardless of each child’s personal eligibility. CEP eligibility is determined at both the school and district level through an Identified Student Percentage (ISP) which is the proportion of students who are predicted to be eligible for the NSLP and SBP. Alternatively, students attending non-CEP schools must have their family qualify and apply for the programs directly. At this time, no existing studies have estimated the CEP’s effect on child health. This study uses data regarding eligibility and participation in the CEP program at the school and district levels in combination with the restricted use Early Childhood Longitudinal Study, Kindergarten class of 2010-2011 (ECLS-K:2011) which contains data on child health, education, school characteristics, and family behaviors. The restricted use ECLS-K:2011 if well suited for the study, as it can be linked to schools directly. The study uses CEP eligibility at the school-level as an instrument for a child’s participation in the NSLP and SBP. Outcome variables of interest include parent-reported child health, body composition, prevalence of different chronic and acute illnesses, mental health, and

The results of this study provide valuable insight for researchers and policymakers concerned with the effects of child-focused nutrition assistance programs on various health outcomes. Currently, there is little causally interpretable evidence regarding the effect of increased food provisions on child health, and no evidence exploiting variation from the CEP. Disparities in nutrition and food security among children both between and within various socioeconomic groups may produce lifelong gaps in health, education, and economic outcomes. This study evaluates the degree to which the two most common school based nutrition assistance programs correct for these disparities and identifies the subgroups for which the programs produce the best and worst results. With these findings, policymakers can alter existing policies and design supplemental programs which target children who are still at the greatest levels of risk.


: To evaluate the effects of a transition home intervention on total Medicaid spending, emergency department (ED) visits, and unplanned readmissions for preterm infants born at ≤ 36.6 weeks gestational age and high-

Rhode Island Medicaid claims data was used to study the 321 infants cared for in the neonatal intensive care unit (NICU) for ≥ 5 days, who were enrolled in the transition home plus (THP) program. The THP program incorporated support services both pre- and post-NICU discharge provided by social workers and family resource specialists (FRS) working with the medical team between October 2012 and October 2014. The THP infants were compared with 365 high-risk infants born and admitted to the NICU in 2011 prior to the full launch of the THP program. Intervention and comparison group outcomes were compared in the eight 3-month quarters after the infant’s birth. Propensity score weights were applied in regression models to balance demographic characteristics between groups. Regression analyses with quarterly fixed effects were run to determine the impact of THP on total Medicaid spending, ED visits, and

The THP cohort included prospectively enrolled Rhode Island resident high-risk infants who were born early (< 32 weeks), moderate (32-33 weeks), late preterm (34-36.6 weeks), or full-term (> 36.6 weeks) in terms of gestational age. All the THP infants were on Medicaid (fee-for-service or managed care), and were hospitalized for ≥ 5 days in an 80-bed single room Level 3-4 NICU between October 1, 2012 and September 30, 2014. The comparison group included infants on Medicaid hospitalized in the same NICU for ≥ 5 days in the year prior to the study period.

Infants in the intervention group had significantly lower total Medicaid spending, fewer ED visits, and fewer readmissions than the comparison group. The Medicaid spending savings for the intervention group were $4,591 per infant per quarter (90% CI: −$8,397, −$785) in our study period. The average quarterly difference estimate for ED visits is a decrease of 334 visits (90% CI: −389, −279) per 1,000 patients relative to the comparison group for the first eight quarters after birth, weighted by the number of intervention patients in the quarter. The intervention group is 7.6 percentage points (90% CI: −12.3, −2.9) less likely to have an unplanned readmission during the first eight quarters after birth. Sensitivity analyses looking at Medicaid claims post discharge, which excludes all the NICU-related health care expenses and utilization, showed the results remained largely the same as the main analyses.

Transition home support services for high-risk infants provided by social workers and family resource specialists working with the medical team can reduce Medicaid spending and health care utilization. Expansion of the medical team to include social workers and FRS could offer a cost-effective approach for clinicians and policy makers to consider in addressing the psychosocial needs of

Does insurance coverage of medical treatments with high out-of-pocket costs affects patients’ utilization. We exploit a policy intervention that mandates coverage for In-Vitro- Fertilization (IVF) –an expensive infertility treatment with low success rates in one cycle of treatment– in private health insurance in the US. Mandated coverage varies from one cycle of treatment in some states to unlimited cycles in some others. Patients’ might increase their chances of conceiving an infant by more aggressive treatments, resulting in risky and costly multiple births. We provide the first estimate of the effects on adverse outcome of aggressive treatments from number of IVF cycles covered in mandated health insurances. We use a Generalized Synthetic Control framework to estimate causal effects. Our estimated effects varies from 0.31 percentage points decrease in share of multiple births in states with only one covered cycle to more than 35 percentage points increase in states with unlimited coverage. Our estimates of effects of mandated IVF coverage on adoption –the main alternative for IVF patients with low chances of success– furthermore shows that adoption rates in states with more covered cycles is lower. These findings suggests that high out-of-pocket costs has strong behavioural responses from patients. In states with more coverage, more patients with low chance of success –who would prefer aggressive treatments– use the treatment. These patients otherwise would have adopted a child. Our findings have important implications for designing policy interventions to increase accessibility of expensive and technologically advance

Patient and family centered care (PFCC) is the hallmark of high quality pediatric care. We explored national trends in the receipt of high quality patient-communication and patient empowerment through behavioral health

We used data from the Medical Expenditure Panel Survey (pooled cross-section) from 2010-2014. We employed two measures of PFCC: 1) a composite measure of high quality patient-physician communication (n=34,629) and 2) patient empowerment through behavioral health counseling about healthy eating (n=36,527) and exercise (n= 38,318). We used multiple logistic regressions to estimate the variation of receiving PFCC by social determinants of health

Rates of receiving behavioral health counseling about healthy eating (53-60%) and exercise (37-42%) were lower than the rate of receiving high quality physician-patient communication (92-93%). Parents were significantly more likely to report receiving high quality physician-patient communication in 2014 than in 2010 (OR 1.37, CI 1.08-1.67); however no association was found for empowerment through behavioral health counseling. Low income and parental educational attainment, and lack of insurance were associated with a lower odds of receiving behavioral health counseling.

Results showed significant variation of physician-patient communication and empowerment by SDOH. Evidence also suggested that providers need to take the next step and begin empowering parents and their children to self-

A large literature documents the effects of early life health shocks on long run outcomes including education, health and earnings. However, the role of parents in reinforcing initial health inequities or mitigating them is not well understood. Furthermore, most studies do not study investments over a life course and look at a single dimension of response such as education or health and ignore non-human capital transfers. We fill these gaps in the literature by studying parental responses over multiple stages of the life course and for multiple dimensions, including non-human capital transfers. We utilize the overlap of Ramadan fasting month with pregnancies for Muslims as a natural experiment for studying parental responses to fetal under nutrition. Ramadan is a holy month in the Islamic calendar. Practicing Muslims are required to observe a strict fast from sunrise to sunset for a month. Since about seventy five percent of all pregnancies overlap with Ramadan, it is estimated more than 1.2 billion Muslims were potentially exposed to their mother’s fasting in utero. Our work builds on some recent studies which show the negative effects of Ramadan exposure in utero on birth weight, cognition and labor market outcomes later in life. We study Muslims in Indonesia, the largest Muslim majority country in the world. Nationally representative data from the 2007 wave of the Indonesian Family Life Survey (IFLS 4) is used to study parental investment response on human capital with detailed expenditures on health and education and on non-human capital transfers (e.g. dowries) over a life cycle. Results show that for children younger than five years, parents are less likely to get their exposed (during Ramadan in utero) children vaccinated and invest less in their diet. However, we find an opposite pattern for older children, with parents investing more in vitamins or supplements but we do not find any differences between exposed and non-exposed children on educational investments. However, our results show that compared to unexposed children, mothers of exposed children have lower expectations about future educational attainment of their children. In contrast to most papers in the larger literature, we also look at non-human capital responses to fetal health shocks and find gender differences in parental response. Females get fewer dowries and bring fewer assets into marriage, an important marker for future bargaining power. For male adults, however, parents are more likely to provide monetary and non-monetary help attempting to mitigate inequities. We do not find any differential responses on prenatal care. Overall our results suggests, parents make complimentary investments in response to initial health endowments particularly for children younger than five years and mostly in the form of investing in their children’s health care but also in terms of dowries for their daughters. These results have important policy implications: if its not feasible to nudge pregnant women from unhealthy cultural practises during pregnancy, one can instead focus on health care and physical capital

This study evaluates the association between family planing (FP) service quality and modern contraceptive prevalence (MCP) in Ethiopia. Reducing unmet need for FP in low and middle income countries (LMICs) can reduce child and maternal mortality and disabilities through improved birth spacing, education, women’s empowerment, and reduction of unwanted pregnancies, poverty, and hunger. Evidence regarding the effect of FP service quality (FPSQ) on MCP has been slow due to the challenge of obtaining measures of FPSQ. The Bruce-Jain framework provides one of the earliest comprehensive literature reviews and theoretical frameworks describing the dimensions of FPSQ. Our study uses a potentially generalizable method of data reduction by using an expert panel to score potential variables linked to FP service quality and to prepare a smaller set of FPSQ indices that are in keeping with the Bruce-Jain

: We used Performance, Monitoring and Accountability data from 2015 to assemble a cross-section dataset for Ethiopia with enumeration area (EA) level data from the service delivery points (SDP) and woman of reproductive age household (HH) surveys. We used multivariate ordinary least square regression models with Huber/White robust standard errors clustered at the region level. The dependent variable was mean MCP and the independent variables were six different FPSQ Bruce scores at the EA level. The six different FPSQ Bruce scores were tested combined and individually in the models. Control variables included socioeconomic status (SES), marriage, husband cohabitation, and education status, age groups, and urban. EA level MCP and controls were calculated by averaging across individual female respondents within an EA. RESULTS: Among the 215 EA-observations in Ethiopia, 44% of the women were between 15 and 24 years old, 58% were married, out of those married 92% cohabited with their husband, 29% had a high school education, and 27% (EA range: 0%-71%) were a current user of a modern contraceptive method. Regression results show that for every unit increase in the EA mean FPSQ Bruce 1 score (on choice of method) and Bruce 4 score (on provider-patient interpersonal relations), the mean MCP increased by 21.4% (p<0.01) and by 18.0% (p<0.01) respectively. Bruce scores 2, 3, 5, 6 did not show statistically significant association with EA level MCP. Results were robust to multiple specification tests. CONCLUSION: Multivariate statistical analysis using PMA data from Ethiopia showed that two out of six of the Bruce domains of FPSQ are linked to modern contraceptive prevalence. These results show that efforts to strengthen FP service quality have the potential to significantly increase contraceptive use in LMICs. Likewise, this results may help child and maternal health policymakers advocate for increased investment in interventions that improve FPSQ in LMICs. Future extensions to our analysis will exploit the panel nature of data with

In the United States, according to the most recent census data, approximately 2.2 million workers travel at least 50 miles each way between their homes and workplaces, and about 1.7 million workers spend 90 minutes or more commuting in each direction. These long commutes can be physically and mentally demanding, particularly for pregnant women. In this regard, we conduct the first empirical study to examine the health impact of long commute to work during pregnancy on fetuses and infants at birth, using unique data that contain information on not only a woman's home address but also her employer's address during her pregnancy. Our study is also the first to examine the health impact of the chronic strain induced by long commute on fetuses and infants at birth, adding new evidence to the literature. We find that among long-distance commuters, increasing the maternal travel distance during pregnancy by 10 miles is associated with increases in low birth weight and intrauterine growth restriction by 1.0 and 0.6 percentage points, or 25 and 46 percent compared with their means, respectively. In addition, we provide evidence on two possible mechanisms underlying the adverse health outcomes associated with long commute: elevated stress levels of pregnant women

We examine the presence of long commute induced maternal stress during pregnancy by showing that the likelihood of using c-sections increases among male babies, but not among female babies, born to women who travel long distance to work during pregnancy. Such an increase is consistent with the finding in the medical literature that male fetuses are more sensitive to stressors in utero than female fetuses, resulting in higher likelihood of delivery complications that require c-sections. With regard to prenatal care, we find that among long-distance commuters an increase of 10 miles in maternal travel distance during pregnancy could reduce the number of prenatal visits by 2.53 percent, decrease the probability of the mother’s completing her first prenatal visit within the first trimester by 2.3 percentage points, and increase the probability of the mother's completing her first prenatal visit within the third trimester or having no prenatal

In addition, our study is the first to calculate the travel distance according to existing public roads instead of using the commonly computed geodetic distance, which represents the length of the shortest curve between two points on earth. Using geodetic distance can incur a greater attenuation bias, potentially leading to a claim of no adverse effect of long commute when the adverse effect could be detected by a less noisily measured commuting distance. Our study has important implications for public policy proposals that consider expanding maternity leave to cover the prenatal period, particularly in the context of the United States. Having the needed time off during the prenatal period can be crucial for pregnant women who are long commuters to alleviate stress and adequately utilize prenatal care.

A major concern over rising inequality is its potential to reduce intergenerational mobility, leading to even greater inequality in the next generation. We estimate the impact of rising inequality over the period 1970-2010 on offspring health at birth, a measure of human capital that has been shown to be highly correlated with future education, IQ and income. We define inequality both at the aggregate level and at the individual level: as a group-level measure (the Gini coefficient for each state or county), and as individual level measures of relative income (relative deprivation, rank, and relative income distance). We document a strong negative relationship between the Gini and newborn health in the cross section, but find that including a modest set of controls, or limiting variation to changes in inequality over time within an area, or instrumenting for inequality eliminates the relationship between the Gini and newborn health completely. However, this null result likely reflects heterogeneity in the effect of rising inequality. When we estimate the impact of relative income on newborn health, we find negative and significant effects of having relatively less income than one’s neighbors on birth weight, even after controlling for area fixed effects and instrumenting for differences in the income distribution.

Childhood lead exposure carries high life-long private and public costs, including reduced IQ and educational attainment and an increased risk of criminal activity. A blood lead screening test is a secondary prevention measure that identifies exposure. A robust body of literature documents disparities in utilization of preventative care, such as immunization coverage, across socioeconomic and racial groups in the US. This literature suggests that several barriers might decrease uptake of preventative care, including high perceived costs and low perceived benefits. On the cost side, lack of information, scheduling challenges, and transportation costs appear to contribute to vaccine delay among low-socioeconomic families. On the benefit side, disease outbreaks are associated with increased immunization rates, likely due to increased salience of disease risk. This paper considers costs and benefits from screening in a unified setting in order to estimate parents’ willingness to pay (WTP) for blood lead level screening and how that WTP varies with latent exposure risk, lead hazard salience, and parents’ characteristics. This paper uses blood lead screening data from 5.4 million tests performed in Illinois from 1997-2016. By linking these test data with birth and death records for almost five million children born in Illinois between 1991 and 2016, I will be able to estimate a model for screening demand as a function of five factors. First, I observe family characteristics in the birth records, but since around 30% of mothers with two children only screen one of the two children, family background might not fully explain disparities in screening. Second, I estimate exposure risk from housing age and pollution sources based on birth address. Third, I estimate provider quality based on observed adherence to screening guidelines and screening rates in providers’ catchment areas. Using distance to providers as a proxy for cost of screening, a procedure common in health access literature, I trace out WTP for screening. Fourth, I study how access to remediation funding impacts demand for screening. Last, I exploit the long sample period to assess how changes in the salience of lead hazards due to “lead scares” affect the WTP estimates. While federal guidelines mandate that all children on Medicaid are screened for lead poisoning at ages one and two, guidelines for non-Medicaid children vary by state and local health department. Illinois requires screening for all children living in high risk zip codes; risk is assessed by housing age and population characteristics. Screening rates in high risk zip codes have been stable at 80% for recent birth cohorts, well above the 56% screening rate for zip codes that do not require universal screening. Yet, it seems important to understand why 20% of young children living in high risk areas are not screened. This is especially relevant as, in the wake of Flint’s lead disaster, some states, including Illinois, are considering switching to a universal screening system like those in Massachusetts and Rhode Island. This analysis aims to sheds light on the relative role of different factors in determining demand for screening.

More than half of pregnancies among women 15-24 in the US are unintended. Long-acting reversible contraceptives (LARC) are extremely effective contraceptive methods with failure rates of less than one percent. LARC use, however, has been hindered by their high cost and lack of acceptance. In early 2009, the Colorado Family Planning Initiative (CFPI) sought to address these access barriers by providing free LARCs and provider training at Title X–funded clinics, which serve a low-income population. The objective of our study was to estimate the causal impact of the CFPI on unintended pregnancies and its impact on Medicaid savings over five years. We used a propensity score weighted difference-in-difference research design to account for unobserved contemporaneous trends in fertility rates using a group of non-Colorado counties as controls. The primary outcome was birth rate per 1000 for women ages 15-24. We obtained birth rates for all US counties from the Centers for Disease Control. We supplemented these data with information on county demographics, educational attainment, income, and unemployment rates from the Area Health Resource Files. The sample period included four years of pre-CFPI implementation and five years of post-CFPI implementation. We first selected controls counties by restricting comparison counties to those within thresholds based on key demographic characteristics, and the then estimated propensity scores with the remaining counties. Based on our model estimates and estimates of the number of women actually affected by the initiative, we calculated Medicaid cost savings. These estimates accounted for mother and child spending, also considering eligibility, attrition, and reimbursement rates over five years post-CFPI. Our final sample consisted of 128 Colorado counties and 204 comparison counties. Baseline characteristics of Colorado and comparison counties were similar and the standardized differences were within the range of acceptable balance, with the exception of percent black females ages 15-19 and percent black female, which were slightly lower in Colorado counties. Our results show that fertility rates decline in both Colorado and comparison counties in the post-period, but Colorado experienced a larger and statistically significant decline of 8.15 births per thousand or about 2900 births averted. Thus, roughly half of the observed decline in fertility rates can be attributed to the CFPI. Our analysis also

The CFPI significantly reduced unintended pregnancies in Colorado over and above observed declines in fertility rates across the country during the same time period. Removing LARC cost barriers for low-income women has the potential to reduce the large number of unintended pregnancies that still remain and are increasingly concentrated among poor and low income women. In addition, the initiative reduced state and federal spending on medical care, potentially

The importance of non-cognitive skills in shaping later-life outcomes has long been recognised in the psychology literature. However, for economists this is a relatively new and underexplored phenomenon. As the body of evidence continues to grow and the consensus surrounding their importance begins to permeate fields, the need to better understand the determinants of non-cognitive skills becomes ever more apparent. This paper investigates the impact of parental education on a child’s non-cognitive skills. We use data from the National Child Development Study (NCDS) to create measures of the child’s Conscientiousness and Neuroticism. These measures are created using responses, given by the child and their teacher, to questions related to the child’s behaviour. We explore exogenous variation in parental education, induced by the 1947 schooling reform, to identify the causal effect. This reform was announced under the 1944 Butler Act, resulting in an increase in the minimum school leaving age from 14 to 15 in the UK in April 1947. We found that this reform increased the average years of schooling by 3.9 months and 4.7 months for fathers and mothers in our cohort respectively. We utilise an Instrumental Variable framework, using dummy variables to indicate whether the parents were impacted by the reform as instruments for parental education. We restrict the study to those parents born just either side of the reform, and obtain the local average treatment effect. By comparing observations that lie closely on either side of our threshold, we are able to eliminate certain trend effects. Findings based on data from the NCDS suggest that increasing the school leaving age by 1 year has no significant effect on the child’s Conscientiousness or Neuroticism. These results suggest that although parental education may play an important role in shaping a child’s

This paper studies the effect of access to prenatal care on children's health, exploiting the public health policy, so-called the CHIP unborn child option, that enables pregnant noncitizens to get prenatal care regardless of legal immigration status. Using state-level variation in whether to opt in and the timing of policy adoption, I find that eligible female noncitizens increase public health insurance coverage rate and subsequently increase the number of a doctor's office visits in 12 months previous to the survey. For children's outcome, the key result is a decline in the presence of chronic health conditions at ages 4-6, which may result in long-lasting health problems during their entire life. Also, the school attendance rate increases among children who were eligible in utero, implying that CHIP unborn child option enhances children's health and increases their pre-school or kindergarten attendance rate. This paper finds that the guaranteed

is a critical period of human development during which parents act on children’s behalf in health investments. These investments may have a profound impact on the life trajectory of a child. We investigate whether parents in China who choose to carry the pregnancy to term allocate resources differently between their sons and daughters over the course of pregnancy after the sex of the child is disclosed to parents. Using unique and large-scale hospital electronic records of prenatal ultrasound scans and birth outcomes as well as a longitudinal survey of parents’ health behavior during pregnancy, we estimate how parental health behaviors and prenatal health investments change after parents gain access to gender information from post-20 gestational week ultrasound scans. In addition to the state-of-the-art difference-in-differences model, we employ a novel fetus fixed effect model to identify shifts in prenatal investments when information on child gender is disclosed. We document sex-selective prenatal investments as an early channel through which parents practice discriminatory behavior. We show that parents favorably shift certain parental health investments when pregnant with a boy. Specifically, the chance of exposure to passive smoking decreases while more mothers take nutrient supplements when parents expect boys compared to girls after receiving a post-

gestational week ultrasound scan. Preferential prenatal treatment of males is greater for areas with stronger son preference. A set of key placebo tests using pre-pregnancy and early pregnancy behaviors reassure us that our identified effects are likely causal. Our findings have implications for eliminating gender discrimination and improving maternal and child health in the earliest stage of life. These findings also call for utilizing the window of opportunity during

Previous studies evaluating the welfare cost of air pollution have not paid much attention to its potential effect on suicidal behaviors. This paper attempts to fill the gap by estimating the effect of air pollution on suicidal ideation and suicide attempts among the school-age children. We use changes in the local wind direction as instruments for air pollutant concentrations to address endogeneity and measurement errors. By matching a unique youth risk behavior survey in China with rich air quality and weather conditions according to the exact date and location of each interview, we find that fine particulate matter (PM2.5) increases the rate of both suicidal ideation and suicide attempts. Heterogeneity analysis reveals that boys who have bad school performance are more affected in terms of suicidal ideation.


Martin Saavedra [email protected]

Frank A. Sloan [email protected]

George Davis [email protected]

Jason Fletcher [email protected]

Low birth weight infants are at a higher risk of infant mortality, although there is debate as to how much of this relationship is causal. Low birth weight may be correlated with infant health even in the absence of a strong causal effect if

Twin studies have high internal validity, but at the potential cost of external validity since it is not clear that twin studies generalize to the singleton population. Twins may be smaller because their optimal birth weight is less than that of a singleton. A 2-kg singleton would be classified as low birth weight and a marginal increase in birth weight may improve infant health, whereas a 2-kg twin would be close to average and may be at her optimal birth weight. Little is known

To overcome these challenges, this paper proposes a novel approach to evaluate whether twin studies provide under-, over-, or unbiased estimates of the effects of birth weight on singletons by comparing causal estimates of low birth weight for twins, triplets, and quadruplets. Many of the differences that exist between singletons and twins also exist between twins and triplets, as well as between triplets and quadruplets. For example, the average birth weight of singletons, twins, triplets, and quadruplets are 3.4 kg, 2.4 kg, 1.7 kg, and 1.3 kg, respectively. This study uses data from the universe of multiple births linked to infant death records in the United States from 1995-2000 to estimate how

I find that low birth weights increase infant mortality and decrease APGAR scores for twins, triplets, and quadruplets. Although triplets and quadruplets are on average smaller, they experience smaller reductions in infant health when they are LBW. These results suggest that twin studies underestimate the effect of LBW for the general population. In addition to the fixed-effects model, I present cross-sectional estimates of the effect of birth weight on infant health for

I then estimate the implied infant mortality reduction from increasing all birth weights to at least 2.2 kg. Estimates from twins, triplets, and quadruplets suggest that increasing birth weights would reduce the infant mortality rate by 1, 0.65, and 0.3 deaths per 1000 live births. These estimates suggest that the true decrease in infant mortality for singletons would be approximately 1.3 deaths per 1000 live births, implying that the twin estimates underestimate the statistical

The effect of criminal justice system involvement on the children of the incarcerated is not well understood. There have been no evaluations examining whether pre-trial confinement worsens child health. We study how specific decisions by the courts and other criminal justice agencies affect child wellbeing during follow-up. We use North Carolina (NC) statewide administrative criminal court and incarceration data for the years 2005-2016 and link parents to their biological children through birth records. Our study uses instrumental variables (e.g., prosecutor’s prosecution rate and judge’s conviction rate) to examine whether holding other factors constant, harsher penalties for criminal offenses imposed by

Over 45 million children participate in the National School Lunch Program (NSLP) and School Breakfast Program (SBP) each year. For many, the free meals they receive in school are their only nutritionally sufficient ones of the day, and for the most at-risk children, it may be their single reliable source of food. A growing literature across the fields of economics, public health, and medicine has linked child nutrition deficiencies and food insecurity to negative health, education, and behavioral outcomes. Alternatively, very little work has been done to estimate the effect that food provision programs like the NSLP and SBP have on the health of disadvantaged children. Utilizing variation in the NSLP and SBP following the passage of 2010’s Healthy Hunger-Free Kids Act (HHKA), this study estimates the effect of providing children with free meals in school on multiple outcomes related to child health across different socioeconomic groups. While the HHKA broadly affected school nutrition across multiple dimensions, one of its main additions was the Community Eligibility Provision (CEP) which allows qualifying schools and districts to offer their students free lunch and breakfast regardless of each child’s personal eligibility. CEP eligibility is determined at both the school and district level through an Identified Student Percentage (ISP) which is the proportion of students who are predicted to be eligible for the NSLP and SBP. Alternatively, students attending non-CEP schools must have their family qualify and apply for the programs directly. At this time, no existing studies have estimated the CEP’s effect on child health. This study uses data regarding eligibility and participation in the CEP program at the school and district levels in combination with the restricted use Early Childhood Longitudinal Study, Kindergarten class of 2010-2011 (ECLS-K:2011) which contains data on child health, education, school characteristics, and family behaviors. The restricted use ECLS-K:2011 if well suited for the study, as it can be linked to schools directly. The study uses CEP eligibility at the school-level as an instrument for a child’s participation in the NSLP and SBP. Outcome variables of interest include parent-reported child health, body composition, prevalence of different chronic and acute illnesses, mental health, and

The results of this study provide valuable insight for researchers and policymakers concerned with the effects of child-focused nutrition assistance programs on various health outcomes. Currently, there is little causally interpretable evidence regarding the effect of increased food provisions on child health, and no evidence exploiting variation from the CEP. Disparities in nutrition and food security among children both between and within various socioeconomic groups may produce lifelong gaps in health, education, and economic outcomes. This study evaluates the degree to which the two most common school based nutrition assistance programs correct for these disparities and identifies the subgroups for which the programs produce the best and worst results. With these findings, policymakers can alter existing policies and design supplemental programs which target children who are still at the greatest levels of risk.


Yiyan Liu [email protected]

Arezou Zaresani [email protected]

Andrew Anderson [email protected]

: To evaluate the effects of a transition home intervention on total Medicaid spending, emergency department (ED) visits, and unplanned readmissions for preterm infants born at ≤ 36.6 weeks gestational age and high-

Rhode Island Medicaid claims data was used to study the 321 infants cared for in the neonatal intensive care unit (NICU) for ≥ 5 days, who were enrolled in the transition home plus (THP) program. The THP program incorporated support services both pre- and post-NICU discharge provided by social workers and family resource specialists (FRS) working with the medical team between October 2012 and October 2014. The THP infants were compared with 365 high-risk infants born and admitted to the NICU in 2011 prior to the full launch of the THP program. Intervention and comparison group outcomes were compared in the eight 3-month quarters after the infant’s birth. Propensity score weights were applied in regression models to balance demographic characteristics between groups. Regression analyses with quarterly fixed effects were run to determine the impact of THP on total Medicaid spending, ED visits, and

The THP cohort included prospectively enrolled Rhode Island resident high-risk infants who were born early (< 32 weeks), moderate (32-33 weeks), late preterm (34-36.6 weeks), or full-term (> 36.6 weeks) in terms of gestational age. All the THP infants were on Medicaid (fee-for-service or managed care), and were hospitalized for ≥ 5 days in an 80-bed single room Level 3-4 NICU between October 1, 2012 and September 30, 2014. The comparison group

Infants in the intervention group had significantly lower total Medicaid spending, fewer ED visits, and fewer readmissions than the comparison group. The Medicaid spending savings for the intervention group were $4,591 per infant per quarter (90% CI: −$8,397, −$785) in our study period. The average quarterly difference estimate for ED visits is a decrease of 334 visits (90% CI: −389, −279) per 1,000 patients relative to the comparison group for the first eight quarters after birth, weighted by the number of intervention patients in the quarter. The intervention group is 7.6 percentage points (90% CI: −12.3, −2.9) less likely to have an unplanned readmission during the first eight quarters after birth. Sensitivity analyses looking at Medicaid claims post discharge, which excludes all the NICU-related health care expenses and utilization, showed the results remained largely the same as the main analyses.

Transition home support services for high-risk infants provided by social workers and family resource specialists working with the medical team can reduce Medicaid spending and health care utilization. Expansion of the medical team to include social workers and FRS could offer a cost-effective approach for clinicians and policy makers to consider in addressing the psychosocial needs of

Does insurance coverage of medical treatments with high out-of-pocket costs affects patients’ utilization. We exploit a policy intervention that mandates coverage for In-Vitro- Fertilization (IVF) –an expensive infertility treatment with low success rates in one cycle of treatment– in private health insurance in the US. Mandated coverage varies from one cycle of treatment in some states to unlimited cycles in some others. Patients’ might increase their chances of conceiving an infant by more aggressive treatments, resulting in risky and costly multiple births. We provide the first estimate of the effects on adverse outcome of aggressive treatments from number of IVF cycles covered in mandated health insurances. We use a Generalized Synthetic Control framework to estimate causal effects. Our estimated effects varies from 0.31 percentage points decrease in share of multiple births in states with only one covered cycle to more than 35 percentage points increase in states with unlimited coverage. Our estimates of effects of mandated IVF coverage on adoption –the main alternative for IVF patients with low chances of success– furthermore shows that adoption rates in states with more covered cycles is lower. These findings suggests that high out-of-pocket costs has strong behavioural responses from patients. In states with more coverage, more patients with low chance of success –who would prefer aggressive treatments– use the treatment. These patients otherwise would have adopted a child. Our findings have important implications for designing policy interventions to increase accessibility of expensive and technologically advance

Patient and family centered care (PFCC) is the hallmark of high quality pediatric care. We explored national trends in the receipt of high quality patient-communication and patient empowerment through behavioral health

We used data from the Medical Expenditure Panel Survey (pooled cross-section) from 2010-2014. We employed two measures of PFCC: 1) a composite measure of high quality patient-physician communication (n=34,629) and 2) patient empowerment through behavioral health counseling about healthy eating (n=36,527) and exercise (n= 38,318). We used multiple logistic regressions to estimate the variation of receiving PFCC by social determinants of health

Rates of receiving behavioral health counseling about healthy eating (53-60%) and exercise (37-42%) were lower than the rate of receiving high quality physician-patient communication (92-93%). Parents were significantly more likely to report receiving high quality physician-patient communication in 2014 than in 2010 (OR 1.37, CI 1.08-1.67); however no association was found for empowerment through behavioral health counseling. Low income and parental

Results showed significant variation of physician-patient communication and empowerment by SDOH. Evidence also suggested that providers need to take the next step and begin empowering parents and their children to self-

Rakesh Banerjee [email protected]

Y. Natalia Alfonso [email protected]

Muzhe Yang [email protected]

A large literature documents the effects of early life health shocks on long run outcomes including education, health and earnings. However, the role of parents in reinforcing initial health inequities or mitigating them is not well understood. Furthermore, most studies do not study investments over a life course and look at a single dimension of response such as education or health and ignore non-human capital transfers. We fill these gaps in the literature by

We utilize the overlap of Ramadan fasting month with pregnancies for Muslims as a natural experiment for studying parental responses to fetal under nutrition. Ramadan is a holy month in the Islamic calendar. Practicing Muslims are required to observe a strict fast from sunrise to sunset for a month. Since about seventy five percent of all pregnancies overlap with Ramadan, it is estimated more than 1.2 billion Muslims were potentially exposed to their mother’s fasting

We study Muslims in Indonesia, the largest Muslim majority country in the world. Nationally representative data from the 2007 wave of the Indonesian Family Life Survey (IFLS 4) is used to study parental investment response on human capital with detailed expenditures on health and education and on non-human capital transfers (e.g. dowries) over a life cycle. Results show that for children younger than five years, parents are less likely to get their exposed (during Ramadan in utero) children vaccinated and invest less in their diet. However, we find an opposite pattern for older children, with parents investing more in vitamins or supplements but we do not find any differences between exposed and non-exposed children on educational investments. However, our results show that compared to unexposed children, mothers of exposed children have lower expectations about future educational attainment of their children. In contrast to most papers in the larger literature, we also look at non-human capital responses to fetal health shocks and find gender differences in parental response. Females get fewer dowries and bring fewer assets into marriage, an important marker for future bargaining power. For male adults, however, parents are more likely to provide monetary and non-monetary help attempting to mitigate inequities. We do not find any differential responses on prenatal care. Overall our results suggests, parents make complimentary investments in response to initial health endowments particularly for children younger than five years and mostly in the form of investing in their children’s health care but also in terms of dowries for their daughters. These results have important policy implications: if its not feasible to nudge pregnant women from unhealthy cultural practises during pregnancy, one can instead focus on health care and physical capital

This study evaluates the association between family planing (FP) service quality and modern contraceptive prevalence (MCP) in Ethiopia. Reducing unmet need for FP in low and middle income countries (LMICs) can reduce child and maternal mortality and disabilities through improved birth spacing, education, women’s empowerment, and reduction of unwanted pregnancies, poverty, and hunger. Evidence regarding the effect of FP service quality (FPSQ) on MCP has been slow due to the challenge of obtaining measures of FPSQ. The Bruce-Jain framework provides one of the earliest comprehensive literature reviews and theoretical frameworks describing the dimensions of FPSQ. Our study uses a potentially generalizable method of data reduction by using an expert panel to score potential variables linked to FP service quality and to prepare a smaller set of FPSQ indices that are in keeping with the Bruce-Jain

: We used Performance, Monitoring and Accountability data from 2015 to assemble a cross-section dataset for Ethiopia with enumeration area (EA) level data from the service delivery points (SDP) and woman of reproductive age household (HH) surveys. We used multivariate ordinary least square regression models with Huber/White robust standard errors clustered at the region level. The dependent variable was mean MCP and the independent variables were six different FPSQ Bruce scores at the EA level. The six different FPSQ Bruce scores were tested combined and individually in the models. Control variables included socioeconomic status (SES), marriage, husband

Among the 215 EA-observations in Ethiopia, 44% of the women were between 15 and 24 years old, 58% were married, out of those married 92% cohabited with their husband, 29% had a high school education, and 27% (EA range: 0%-71%) were a current user of a modern contraceptive method. Regression results show that for every unit increase in the EA mean FPSQ Bruce 1 score (on choice of method) and Bruce 4 score (on provider-patient interpersonal relations), the mean MCP increased by 21.4% (p<0.01) and by

Multivariate statistical analysis using PMA data from Ethiopia showed that two out of six of the Bruce domains of FPSQ are linked to modern contraceptive prevalence. These results show that efforts to strengthen FP service quality have the potential to significantly increase contraceptive use in LMICs. Likewise, this results may help child and maternal health policymakers advocate for increased investment in interventions that improve FPSQ in LMICs. Future extensions to our analysis will exploit the panel nature of data with

In the United States, according to the most recent census data, approximately 2.2 million workers travel at least 50 miles each way between their homes and workplaces, and about 1.7 million workers spend 90 minutes or more commuting in each direction. These long commutes can be physically and mentally demanding, particularly for pregnant women. In this regard, we conduct the first empirical study to examine the health impact of long commute to work during pregnancy on fetuses and infants at birth, using unique data that contain information on not only a woman's home address but also her employer's address during her pregnancy. Our study is also the first to examine the health

We find that among long-distance commuters, increasing the maternal travel distance during pregnancy by 10 miles is associated with increases in low birth weight and intrauterine growth restriction by 1.0 and 0.6 percentage points, or 25 and 46 percent compared with their means, respectively. In addition, we provide evidence on two possible mechanisms underlying the adverse health outcomes associated with long commute: elevated stress levels of pregnant women

We examine the presence of long commute induced maternal stress during pregnancy by showing that the likelihood of using c-sections increases among male babies, but not among female babies, born to women who travel long distance than female fetuses, resulting in higher likelihood of delivery complications that

require c-sections. With regard to prenatal care, we find that among long-distance commuters an increase of 10 miles in maternal travel distance during pregnancy could reduce the number of prenatal visits by 2.53 percent, decrease the probability of the mother’s completing her first prenatal visit within the first trimester by 2.3 percentage points, and increase the probability of the mother's completing her first prenatal visit within the third trimester or having no prenatal

In addition, our study is the first to calculate the travel distance according to existing public roads instead of using the commonly computed geodetic distance, which represents the length of the shortest curve between two points on earth. Using geodetic distance can incur a greater attenuation bias, potentially leading to a claim of no adverse effect of long commute when the adverse effect could be detected by a less noisily measured commuting distance. Our study has important implications for public policy proposals that consider expanding maternity leave to cover the prenatal period, particularly in the context of the United States. Having the needed time off during the prenatal period

Florencia Borrescio Higa [email protected]

Ludovica Gazze [email protected]

Marcelo Perraillon [email protected]

Rose Atkins [email protected]

A major concern over rising inequality is its potential to reduce intergenerational mobility, leading to even greater inequality in the next generation. We estimate the impact of rising inequality over the period 1970-2010 on offspring health at birth, a measure of human capital that has been shown to be highly correlated with future education, IQ and income. We define inequality both at the aggregate level and at the individual level: as a group-level measure (the Gini coefficient for each state or county), and as individual level measures of relative income (relative deprivation, rank, and relative income distance). We document a strong negative relationship between the Gini and newborn health in the cross section, but find that including a modest set of controls, or limiting variation to changes in inequality over time within an area, or instrumenting for inequality eliminates the relationship between the Gini and newborn health completely. However, this null result likely reflects heterogeneity in the effect of rising inequality. When we estimate the impact of relative income on newborn health, we find negative and significant effects of having relatively less income

Childhood lead exposure carries high life-long private and public costs, including reduced IQ and educational attainment and an increased risk of criminal activity. A blood lead screening test is a secondary prevention measure that identifies exposure. A robust body of literature documents disparities in utilization of preventative care, such as immunization coverage, across socioeconomic and racial groups in the US. This literature suggests that several barriers might decrease uptake of preventative care, including high perceived costs and low perceived benefits. On the cost side, lack of information, scheduling challenges, and transportation costs appear to contribute to vaccine delay among low-socioeconomic families. On the benefit side, disease outbreaks are associated with increased immunization rates, likely due to increased salience of disease risk. This paper considers costs and benefits from screening in a unified setting in

This paper uses blood lead screening data from 5.4 million tests performed in Illinois from 1997-2016. By linking these test data with birth and death records for almost five million children born in Illinois between 1991 and 2016, I will be able to estimate a model for screening demand as a function of five factors. First, I observe family characteristics in the birth records, but since around 30% of mothers with two children only screen one of the two children, family background might not fully explain disparities in screening. Second, I estimate exposure risk from housing age and pollution sources based on birth address. Third, I estimate provider quality based on observed adherence to screening guidelines and screening rates in providers’ catchment areas. Using distance to providers as a proxy for cost of screening, a procedure common in health access literature, I trace out WTP for screening. Fourth, I study how access to

While federal guidelines mandate that all children on Medicaid are screened for lead poisoning at ages one and two, guidelines for non-Medicaid children vary by state and local health department. Illinois requires screening for all children living in high risk zip codes; risk is assessed by housing age and population characteristics. Screening rates in high risk zip codes have been stable at 80% for recent birth cohorts, well above the 56% screening rate for zip codes that do not require universal screening. Yet, it seems important to understand why 20% of young children living in high risk areas are not screened. This is especially relevant as, in the wake of Flint’s lead disaster, some states, including Illinois, are considering switching to a universal screening system like those in Massachusetts and Rhode Island. This analysis aims to sheds light on the relative role of different factors in determining demand for screening.

More than half of pregnancies among women 15-24 in the US are unintended. Long-acting reversible contraceptives (LARC) are extremely effective contraceptive methods with failure rates of less than one percent. LARC use, however, has been hindered by their high cost and lack of acceptance. In early 2009, the Colorado Family Planning Initiative (CFPI) sought to address these access barriers by providing free LARCs and provider training at Title X–funded clinics, which

We used a propensity score weighted difference-in-difference research design to account for unobserved contemporaneous trends in fertility rates using a group of non-Colorado counties as controls. The primary outcome was birth rate per 1000 for women ages 15-24. We obtained birth rates for all US counties from the Centers for Disease Control. We supplemented these data with information on county demographics, educational attainment, income, and unemployment rates from the Area Health Resource Files. The sample period included four years of pre-CFPI implementation and five years of post-CFPI implementation. We first selected controls counties by restricting comparison counties to those within thresholds based on key demographic characteristics, and the then estimated propensity scores with the remaining counties. Based on our model estimates and estimates of the number of women actually affected by the initiative, we calculated Medicaid cost savings. These estimates accounted for mother and child spending, also considering eligibility, attrition, and reimbursement rates over five years post-CFPI. Our final sample consisted of 128 Colorado counties and 204 comparison counties. Baseline characteristics of Colorado and comparison counties were similar and the standardized differences were within the range of acceptable balance, with the exception of percent black females ages 15-19 and percent black female, which were slightly lower in Colorado counties. Our results show that fertility rates decline in both Colorado and comparison counties in the post-period, but Colorado experienced a larger and statistically significant decline of 8.15 births per thousand or about 2900 births averted. Thus, roughly half of the observed decline in fertility rates can be attributed to the CFPI. Our analysis also

The CFPI significantly reduced unintended pregnancies in Colorado over and above observed declines in fertility rates across the country during the same time period. Removing LARC cost barriers for low-income women has the potential to reduce the large number of unintended pregnancies that still remain and are increasingly concentrated among poor and low income women. In addition, the initiative reduced state and federal spending on medical care, potentially

The importance of non-cognitive skills in shaping later-life outcomes has long been recognised in the psychology literature. However, for economists this is a relatively new and underexplored phenomenon. As the body of evidence continues to grow and the consensus surrounding their importance begins to permeate fields, the need to better understand the determinants of non-cognitive skills becomes ever more apparent. This paper investigates the impact of parental education on a child’s non-cognitive skills. We use data from the National Child Development Study (NCDS) to create measures of the child’s Conscientiousness and Neuroticism. These measures are created using responses, given by the child and their teacher, to questions related to the child’s behaviour. We explore exogenous variation in parental education, induced by the 1947 schooling reform, to identify the causal effect. This reform was announced under the 1944 Butler Act, resulting in an increase in the minimum school leaving age from 14 to 15 in the UK in April 1947. We found that this reform increased the average years of schooling by 3.9 months and 4.7 months for fathers and mothers in our cohort respectively. We utilise an Instrumental Variable framework, using dummy variables to indicate whether the parents were impacted by the reform as instruments for parental education. We restrict the study to those parents born just either side of the reform, and obtain the local average treatment effect. By comparing observations that lie closely on either side of our threshold, we are able to eliminate certain trend effects. Findings based on data from the NCDS suggest that increasing the school leaving age by 1 year has no significant effect on the child’s Conscientiousness or Neuroticism. These results suggest that although parental education may play an important role in shaping a child’s

Grace Hwang [email protected]



This paper studies the effect of access to prenatal care on children's health, exploiting the public health policy, so-called the CHIP unborn child option, that enables pregnant noncitizens to get prenatal care regardless of legal immigration status. Using state-level variation in whether to opt in and the timing of policy adoption, I find that eligible female noncitizens increase public health insurance coverage rate and subsequently increase the number of a doctor's office visits in 12 months previous to the survey. For children's outcome, the key result is a decline in the presence of chronic health conditions at ages 4-6, which may result in long-lasting health problems during their entire life. Also, the school attendance rate increases among children who were eligible in utero, implying that CHIP unborn child option enhances children's health and increases their pre-school or kindergarten attendance rate. This paper finds that the guaranteed

is a critical period of human development during which parents act on children’s behalf in health investments. These investments may have a profound impact on the life trajectory of a child. We investigate whether parents in China who choose to carry the pregnancy to term allocate resources differently between their sons and daughters over the course of pregnancy after the sex of the child is disclosed to parents. Using unique and large-scale hospital electronic records of prenatal ultrasound scans and birth outcomes as well as a longitudinal survey of parents’ health behavior during pregnancy, we estimate how parental health behaviors and prenatal health investments change after parents gain access to gender information from post-20 gestational week ultrasound scans. In addition to the state-of-the-art difference-in-differences model, we employ a novel fetus fixed effect model to identify shifts in prenatal investments when information on child gender is disclosed. We document sex-selective prenatal investments as an early channel through which parents practice discriminatory behavior. We show that parents favorably shift certain parental health investments when pregnant with a boy. Specifically, the chance of exposure to passive smoking decreases while more mothers take nutrient supplements when parents expect boys compared to girls after receiving a post-

gestational week ultrasound scan. Preferential prenatal treatment of males is greater for areas with stronger son preference. A set of key placebo tests using pre-pregnancy and early pregnancy behaviors reassure us that our identified effects are likely causal. Our findings have implications for eliminating gender discrimination and improving maternal and child health in the earliest stage of life. These findings also call for utilizing the window of opportunity during

Previous studies evaluating the welfare cost of air pollution have not paid much attention to its potential effect on suicidal behaviors. This paper attempts to fill the gap by estimating the effect of air pollution on suicidal ideation and suicide attempts among the school-age children. We use changes in the local wind direction as instruments for air pollutant concentrations to address endogeneity and measurement errors. By matching a unique youth risk behavior survey in China with rich air quality and weather conditions according to the exact date and location of each interview, we find that fine particulate matter (PM2.5) increases the rate of both suicidal ideation and suicide attempts. Heterogeneity


Oberlin College Complete

Duke University Elizabeth Gifford; Lindsey Kozecke Complete

Georgia State University Complete

University of Wisconsin-Madison Jessica Polos Complete

RTI International Complete

University of Melbourne Lucie Schmidt; Lindsay Tedds Complete

University of Maryland Complete

LaShawn Glasgow; Richard Tucker; Elisabeth McGowan; Betty Vohr; Marianne Kluckman

Rice University Farhan Majid Complete

Complete

Lehigh University Yang Wang Complete

Johns Hopkins University Bloomberg School of Public Health

David Bishai; Portia Pan; Mingxin Chen; Suzanne Bell; Sasmita Matta

Universidad Adolfo Ibanez Hernan Winkler; Anna Aizer Complete

Complete

Richard Lindrooth Complete

The University of Manchester Complete

University of Chicago Energy and Environment Lab

University of Colorado Anschutz Medical Campus

The Ohio State University Complete

Yale University Neha Anand Complete

Yale University Xin Zhang Complete


Mental Health

Mental Health

Mental Health

Mental Health

Optical Illusion: Adolescent Body Perception and its Impact on BMI

Life is Precious: Reducing Suicidal Behavior in Latina Adolescents in New York City

How Do Economic Shocks Affect Family Mental Health Spending?

Investigating the impact of cognitive impairment on health care utilisation for older people

Mental Health

Mental Health

The Great Recession and Mental Health: the effect of income loss on depression scores of young mothers

Gender, Body Image and Mental Health Risks of Bullying in U.S. High Schools: A Study of State Anti-Bullying Laws from Survey Data

Mental Health

Mental Health

The Effect of Medicaid Churning on Healthcare Utilization among Adults with Mental Health Disorders

Does your parents’ education affect your well-being? Evidence from an educational reform in England.

Mental Health

Mental Health

Preventing suicidal ideation and intentional self-inflicted injuries among people with substance abuse disorders -- Roles of local health departments in Maryland

Evaluating and Improving Post-Hospitalization Mental Health Follow-Up Care

Mental Health

Mental Health

State religious freedom laws and mental distress among sexual minority adults

Hospitalizations, length of stay, and associated costs among suicide-attempting adolescents: A longitudinal study from California

Mental HealthMental Illness and College Educational Outcomes: Evidence from State Parity Laws

Abstract

Psychology has taught us that self-concepts develop during adolescence—a time when youth are also completing their physical development. This study examines these two simultaneous development processes and how they relate. Using 15-years of panel data on adolescents, I evaluate the relationship between body perception and body weight. I determine how closely perception aligns with BMI and identify those environmental, regional and household characteristics associated with mis/perception. Finally, I identify how over or under estimation of body weight relates to subsequent growth and how perception defines the BMI growth trajectory. Results show that misperception is more common among adolescents at high BMI levels. Blacks have a higher probability of accurately assessing their bodies, while Hispanics have a lower probability. Overestimating body size leads to lower subsequent growth while underestimating is associated with higher growth. Results suggest that although self-views and physical features develop simultaneously, they do not always align.

Background: Latina adolescents have high rates of suicidal ideation and attempts (CDC, 2015). Life is Precious (LIP) serves Latina adolescents at risk for suicide in an after-school, clubhouse model that supplements ongoing outpatient mental health treatment with a set of culturally-competent services, including family support services, supported education services, and creative arts therapy. LIP operates three NYC locations, serving Latina girls aged 12-18 who have been identified as at-risk for suicidal behavior. Methods: Since program inception in 2008, data has been obtained for 279 participants. Psychological assessment results Suicidal Ideation Questionnaire (SIQ), Reynolds Adolescent Depression Scale (RADS2), and Trauma Symptom Checklist for Children (TSCC) are analyzed using linear mixed effects models. Results: There have been no completed suicides among program participants since the program’s inception in 2008, in this high-risk population. In the past year, there has been one reported suicide attempt (of 124 participants served over the course of the year at the three program sites). Suicidal ideation, as measured by SIQ, decreases by about 1 point per year of enrollment (p<0.01). Participants who report a history of sexual abuse have a decrease of about three points (p<0.01) and participants who report a history of alcohol use also have a decrease of about three points (p<0.02). Participants with a history of tobacco use have a decrease of about five points (p<0.01). Depressive symptoms also decrease: about 2.6 points per year as measured by the RADS2 and about one point per year as measured by the TSCC (p<0.01 for all). Other symptoms measured by TSCC also decrease: anger, anxiety and post-traumatic stress symptoms all decrease by about half a point per year (p<0.01 for all). There is no control group at this time, and assessments only measure participants while they are in the program. Conclusions: The program is showing low rates of suicide attempts and statistically significant decreases in suicidal ideation, depressive symptoms, and other symptoms during program participation. While changes to ideation and symptoms are small, they are statistically significant and are seen over time and in multiple assessment measures (SIQ, RADS2, TSCC). An expansion of this evaluation, which will compare outcomes to comparison samples who are receiving outpatient mental health treatment without the additional LIP clubhouse services, has been funded and is beginning.

A negative economic shocks may impose two opposing influences on mental health spending: pressure to decrease spending due to liquidity constraints, and pressure to increase spending due to worsening mental health status. Using two-year panel data from the Medical Expenditure Panel Survey for the period 2004 to 2012, we examine the effect of an economic shock on mental health spending by families with children. Specifically, we focus on the effect of changes in family income, employment status, and health insurance status over the two-year observation period within each MEPS panel. Since a substantial proportion of the population does not use mental health care and the distribution of spending is positively skewed we estimate a series of two-part model with probit equation in the first part and generalized linear model (GLM) with a log link and gamma variance function in the second part. To control for unobserved heterogeneity across families in the sample we apply the two-part model using the correlated random effects (CRE) framework. The results indicate that family mental health spending strongly responds to employment shocks. Employment gains are associated with a decline in both the likelihood of mental health services utilization and in the amount being spent toward mental health services among those who use the services. For instance, gaining employment in two-parent families where a parent was not employed during entire year is associated, on average, with an expected decline of $172 in family mental health spending. Among single-mother families, gaining employment and continuing to be employed throughout the two-year observation period is associated with an expected decline of $307 in family mental health spending. The effect of an employment gain leading to a lower mental health services use is consistent with the hypothesis that gaining employment may improve mental health and thereby lead to a decreased demand for mental health services. We find that employment losses are associated with an increase in both the likelihood of mental health services utilization and in the amount being spent on such services among those who use services. For instance, losing employment after a recent employment gain increases the likelihood of mental health services utilization by about five percentage points. We also find that losing a job and remaining jobless for the remainder of the two-year observation period leads to an increase of $204 in expected family mental health among two-parent families. The sign of the employment loss effect may reflect pressure to increase mental health spending due to the possibility of worsening mental health status when a family member loses employment. We also find that income and health insurance loss is associated with a decline in mental health spending. For instance, mothers in two-parent families where a family member lost health insurance coverage spend less on mental health care. Similarly, negative income shocks in single-mother families decrease the amount of ambulatory mental health spending and the amount of mental health care spending incurred by mothers.

Evaluation of health care utilization is prominent in the economics literature, generally concentrating on the impact of certain health conditions on resource use, using individual level self-reported data. However, the important additional impact of cognitive impairment has not yet been explored. We now propose a model that assesses the impact of word recall, a standard validated measure of cognitive status, on health care utilisation. We develop an approach that simultaneously takes into account the count nature of the data, recall bias and reporting behaviours, and the role of cognitive state in all of these processes. Our model also introduces an individual specific random parameter. Using the Survey of Health, Ageing and Retirement in Europe data, we demonstrate that without modelling for unobserved heterogeneity via captivity effects and random parameters, the effect of cognitive status is largely underestimated. A striking finding is that the effect of cognitive status is relatively constant over age.

Introduction: Studies of the effects of economic shocks on health have shown different conclusions about the nature of the recessionary-health relationship. This lack of a consensus may be due to the variations in size and type of recession. While unemployment was seen in most high income countries during the Great Recession, economic downturns also reduce real income. Furthermore, in Ireland and the United States the financial crisis also triggered mortgage distress. Actual or threatened loss of a home can lead to depression and poor mental health.

Methods: I use data from the first three waves of the infant cohort Growing Up in Ireland study to examine the effect of income loss on the mental health of young mothers. The timing of the three waves encompasses the period of the financial crisis. I take a fixed effects approach to examine changes in total depression score (CES-D), and I also examine changes for the subgroups of mental health that make up the total score. This allows me to differentiate between feelings of failure and fear for the future, and symptoms consistent with sadness and the ‘blues’. I examine the effect of income loss on these measures mental health, and I then examine if this effect is due to the tenancy status of the respondent, or to their income status in the pre-recessionary period.

Results: A balanced panel of 6821 households is used in this study, and the individual of interest is the biological mother of the study child. The results of the fixed effects model show that income loss is associated with an increase in depression (-.245**). When the tenancy status of the individual is taken into account income loss is associated with depression for owners who have a mortgage (-.317***). This effect of income loss is not seen for those who rent from local authority, private owners, parents, or occupy their homes free of rent. When the income quintile of the individual is included, it is seen that those who were in the wealthiest income quintile in wave 1 have an increase in depression associated with income loss (-.253*). This is not seen for the other income quintiles. There is also an association between being in mortgage or rent arrears and an increase in depression (.159**). When the total depression score is broken down into symptoms of depression, income loss is associated with symptoms of being depressed (-.043*), failure (-.031**) and fear (-.04*) but not loneliness, crying, or being sad. However, income loss for those who pay a mortgage was associated with the ‘blues’ (-.036*), feeling depressed (-.05) failure (-.032**) fear (-.053**), restless sleep (-.068**) and being sad (-.044*).

Conclusion: The results indicate that the Great Recession had an effect on the mental health of young mothers. This was due to income loss rather than unemployment, despite the fact that unemployment is commonly used as an indicator of recessionary effect. When the effect is examined based on the tenancy of the household, income loss for mortgage owners was associated with depression.

As of today, all 50 states and District of Columbia have enacted Anti-Bullying Laws (ABLs)to protect students from being bullied. Yet there is still limited evidence on whether such policiescan effectively reduce the frequency and severity of bullying on school grounds, especially forthose students with high-risk of being bullied. Moreover, several previous studies have indicatedthat students with negative self-perceived body image are more likely to be victims of bullyingactivities. In this paper, we investigate the effectiveness of ABLs on U.S high school studentsfrom Youth Risk Behavior Surveillance System (YRBSS) and School Health Profiles (SHP) underdifference-in-difference model framework. The school-level estimates from SHP suggest that ABLsincrease the probability of school staff devoting their effort towards healthy anti-bullying climateby 1.9 percentage points and receiving more professional developments by 5.9 percentage points.Yet, the student-level analysis on YRBSS supports the statement that state ABLs have limitedeffect on high school students' mental health and suicide ideation across different gender and body-image classes. But results also imply that ABLs with certain components (Training/Prevention,Communication and Prohibited Behaviors) can signicantly reduce female students' mental healthissues (1.9-6.6 percentage points) and suicide ideation of male students with negative body image(10.9-11.4 percentage points) compared to other state ABLs. Falsication test using BehavioralRisk Factor Surveillance System (BRFSS) on which the anti-bullying laws should not have animpact, verfies the robustness of results form YRBSS analysis.

Objectives: Recent policy proposals have considered cutting billions in federal funding for Medicaid and restricting the program’s eligibility criteria, potentially leading to coverage loss for beneficiaries with mental health disorders. Little is known about the dynamic relationship between Medicaid coverage loss (or “churning”) and mental health care utilization. This analysis examined how outpatient healthcare and mental health service use changes when patients lose Medicaid coverage and become uninsured among a nationwide cohort of adults with mental health disorders, and how these changes progress over time while the individuals remain uninsured. Methods: Our sample included 4,822 persons (102,261 person-months) ages 18-64 with mental health disorders from the 2001-2014 Medical Expenditure Panel Survey. We examined patterns of healthcare service use pre-churning (while covered by Medicaid) and post-churning (while uninsured) among individuals who lost Medicaid coverage, and compared them with service use in a comparison group with continuous Medicaid enrollment. We identified the comparison group using a propensity score matching method. In our models, a simple “post” indicator was first used to estimate the average differences in healthcare service use post-churning, and a count variable measuring total months since churning—with linear and non-linear terms—was subsequently used to estimate changes over time. We estimated logit models to estimate changes in the likelihoods of any outpatient visit and any mental health related outpatient visit per-person-per-month. We also used two-part models to estimate total healthcare costs and out-of-pocket costs per-person-per-month. Results: Becoming uninsured after losing Medicaid coverage was associated with a 53% reduction (marginal effects [ME] = -15.09 percentage points, 95% CI: [-17.23, -12.95]) in the likelihood of receiving any outpatient services, and a 43% reduction (ME = -2.76 percentage points, 95% CI: [-4.22, -1.31]) in the likelihood of receiving any mental health related outpatient services in a month. Healthcare service use declined the most in the month immediately post-churning. Healthcare use trends declined further and flattened over the next half-year while the patients remained uninsured.

Moreover, the reductions in healthcare use post-churning was translated into a 53% reduction (ME = -$153.73; 95% CI: [-224.24, -83.23]) in total healthcare costs per-person-per-month. Further, we observed a 66% increase (ME = $4.85; 95% CI: [2.48, 7.22]) in patients’ out-of-pocket costs per-person-per-month. When examining how these trends changed over the six months following loss of Medicaid coverage, total costs continued to decline and out-of-pocket costs continued to rise, but these curves flattened over time. Conclusions Our analysis raises important considerations about the implications of Medicaid churning for low-income patients with mental health problems. Findings from this analysis suggest that loss of insurance coverage immediately after dis-enrolling from Medicaid disrupted the receipt of outpatient services among adult beneficiaries with mental health disorders, and this disruption persisted for an extended period of time post-churning. As states consider the future of their Medicaid programs in a dynamic policy landscape, it will be crucial to implement strategies to mitigate Medicaid churning and ensure adequate receipt of mental health related outpatient services for vulnerable populations.

A 2007 United Nations Children’s Fund and an OECD report comparing child well-being across 21 industrialized nations, placed the UK at the bottom of the rankings for measures of children socio-emotional well-being (SEW). In addition to being important for health, SEW relates to the “Big 5” personality traits and non-cognitive skills which are key predictors for later life employment opportunities. This paper evaluates the effect of parental education on children socio-emotional skills. I use a rich cohort dataset, the Avon Longitudinal Study of Parents and Children (ALSPAC) containing measures of SEW such as the Strengths and Difficulty Questionnaire (SDQ) from 4 to 13 years old. I exploit the fact that a proportion of the parents in ALSPAC were impacted by the most recent raising of the minimum school leaving age (RoSLA) in England which occurred in 1972. Using a Regression Discontinuity Design, the structure of the ALSPAC data allows me to identify the causal impact of the policy separately from the effect of parents age at the time of the child’s birth. Theoretical work on the skills production function by Conti and Heckman, Cunha and Heckman, and others suggest that an individual’s skills are multi-dimensional and comprise socio-emotional skills in addition to cognitive skills. There is a dynamic technology of skill formation function that depends, amongst other inputs, on parental investments. As early life SEW improves both SEW throughout the lifecourse and impacts the productivity of future investments, this model suggests that parental education may be more impactful than other late-life interventions. This paper relates to two strands of the economics literature. The first strand examines the determinants of non-cognitive abilities and socio-emotional skills. The majority of these studies have focused on the role of parental income using measures of permanent income or exogenous intervention-related household income shocks. They find that income has a cumulative positive effect on non-cognitive skills which increases as the cohort ages. Likely mechanisms are improvements in parent-child relationships. The second strand of the literature investigates the effect of parental education on children’s outcomes such as education, health and cognitive skills finding effects that are visible at older ages. I contribute to both strands of the empirical literature. Compared to the non-cognitive skills literature, I consider a previously overlooked component of the non-cognitive skills production function, parental education, and exploit a natural experiment. My paper is the first to investigate the intergenerational transmission of socio-emotional skills in a large cohort study in the UK. I consider the mechanisms through which parental education might affect children socio-emotional skills including assortative mating, earnings and better information on parenting skills. I also consider whether the effect varies across the parental educational distribution and depending on the gender of children. Very preliminary results indicate that parental education improve children socio-emotional skills. This effect is stronger for the children of parents who would have not stayed in school longer. There is no difference between children’s gender, but one likely mechanism is parenting skills and time spent reading to children.

Background Suicide is the leading cause of death among people with substance use disorders (SUDs). From 2006 to 2013, the rate of emergency department (ED) adult visits involving SUDs increased by 34% and the rate of ED visits related to suicidal ideation more than doubled from 173 to 376 visits per 100,000 population. Together, 42% of ED suicidal ideation-related visits were associated with SUDs in 2013. Though, numerous studies indicate ED-initiated interventions have been successful in reducing adult suicidal ideation and intentional self-inflicted injury, none are wide-spread; and while other studies recommend interventions outside the ED, none examine the role of local health departments (LHDs) in relation to suicide prevention. Objectives The objective of this study is to examine whether LHDs’ active roles of health promotion are associated with reductions in the rates of suicidal ideation and intentional self-inflicted injury in the ED settings. Methods Using data sets linked from multiple sources, including 2012-2013 State ED Databases for the State of Maryland, the National Association of County and City Health Officials Profiles Survey, the Area Health Resource File, and U.S. Census data, we employed multi-level/hierarchical logistic models to examine whether LHDs’ active provisions of preventive care and primary care services and health policy advocacy (such as affordable housing, mental health, education, etc.) were associated with the reduction of having suicidal ideation and intentional self-inflicted injuries. Analyses were conducted among individuals with substance use disorders (SUDs) aged 18 and above. Results People with SUDs who committed suicidal ideation and intentional self-inflicted injuries were more likely to be White, had more chronic conditions, and were more likely to live in rural areas. The rate of suicidal ideation and intentional self-inflicted injury was 0.78% overall, but the rate increased to 7.26% of among patients with SUDs. Levels of LHDs’ activities vary by services. Approximately 7.23% of patients with SUDs resided in counties with LHDs’ active advocacy for affordable housing, and 92.07% patients with SUDs resided in counties with LHDS’ active education on tobacco, alcohol, or other drugs. After adjusting for individual-, hospital-, LHD-, and county-level characteristics, multilevel logistic regressions showed that LHDs’ health promotions on affordable housing (OR=0.65 , p<0.05) and education on tobacco, alcohol, or other drugs (OR=0.63, p<0.001) were significantly associated with the reduction of suicidal ideation and intentional self-inflicted injury in the ED settings for people with SUDs. Associations of LHDs’ provision of behavioral health preventive care and health services were not significant. Conclusions Suicide is the product of multiple causes, its prevention requires broad, interdisciplinary approaches. Though results demonstrate a benefit of health education related to affordable housing, tobacco, alcohol, and other drugs, there is a need to further examine the impact of LHD activity on mental health and the health system. Future studies should drill down into the SUDs variable to determine if these results vary by alcohol or drug use; and should explore the policy differences of jurisdictional tobacco control and housing policies on suicidal ideation and intentional self-inflicted injury.

Follow-up care after a hospitalization for mental illness within 7 (or 30) days is a widely used performance measure for mental health care, but little empirical evidence supports its use. Coordination of mental health care is difficult, and no empirical studies examine whether improving follow-up after hospitalizations decreases health care spending or utilization. To examine associations between mental health provider follow-up after a hospital visit for mental illness and utilization and cost outcomes using health insurance enrollment and claims data from all payer claims databases in four states: Massachusetts, New Hampshire, Maryland, and Utah. The three outcome measures are total healthcare spending during the 180-day follow-up period after a hospital visit for mental illness, whether the patient is readmitted; and whether the patient has an ED visit post discharge. The independent measure of interest is whether or not the patient had a follow-up visit within 7 days after discharge. I expect to find follow-up with a mental health provider within 7 days is associated with reduced readmissions and reduced ED visits, with minimal impacts on healthcare spending. I expect to find heterogeneity in these effects based on patient demographic and clinical characteristics. Practice improvements enabled by this research — for example, better targeting individuals with an increased risk of poor ambulatory follow-up care, identifying hospital characteristics important for improved follow-up, and the use of decision support mechanisms to improve cross-sector care delivery — have substantial implications for mental health outcomes and health costs. As private insurance and Medicaid dominate the financing of mental health care, this study helps fill a critical hole in existing research by strengthening the empirical evidence base for the calculation and use of mental health performance measures.

Background Between 2015 and 2017, 12 states passed religious freedom laws affecting sexual minority rights. Sexual minorities have a disproportionate burden of poor mental health that may be affected by religious freedom laws. We investigated whether 2015 state religious freedom laws were associated with changes in sexual minority mental health between 2014 and 2016. We used 2014-2016 Behavioral Risk Factor Surveillance System (BRFSS) data representative of adults aged 18-65 years in each state. We used data from three states that implemented religious freedom laws in 2015 and that included questions on sexual orientation identity in the state BRFSS: North Carolina, Utah, and Michigan. We compared these states to three control states that collected data on sexual orientation identity and that had populations most closely matching the racial/ethnic characteristics and income distributions of treatment states: Virginia, Idaho, and Pennsylvania, respectively. The main outcome of interest was severe mental distress, defined as poor mental health in 14 or more of the past 30 days. We conducted a difference-in-differences (DD) analysis, comparing changes in severe mental distress in religious freedom states to changes in severe mental distress in control states. We estimated a linear regression model based on evidence that logistic regression estimates can be biased in the presence of fixed effects. The main exposure of interest was living in a state that implemented a religious freedom law in the prior calendar year and identifying as a sexual minority, modeled as an interaction term. In addition to including binary terms for sexual minority identity and for states implementing religious freedom laws in the prior year, we controlled for individual-level state, year, sex, race, ethnicity, age group, educational attainment, income, employment, and marital status. Controlling for each state had the effect of controlling for all time-invariant state characteristics. Because DD clustered standard error estimates can be biased downward, we conducted permutation tests to estimate statistical significance. We conducted several sensitivity analyses, including estimating a logistic regression model.

Results Prior to religious freedom policies, 12.7% of adults and 20.4% of sexual minority adults reported severe mental distress. In states that passed religious freedom policies in 2015, severe mental distress among sexual minorities increased from 21.9% in 2014 to 32.8% in 2016. In control states, severe mental distress among sexual minorities increased from 19.2% in 2014 to 26.3% in 2016. Based on the difference-in-differences analysis, religious freedom laws were associated with an 8.8 percentage point (95% confidence interval: 3.4 to 14.2 percentage points, permutation-adjusted p-value<0.01) increase, a 43% relative increase, in the proportion of sexual minority adults experiencing severe mental distress. Religious freedom laws were not statistically significantly associated with changes in severe mental distress among all adults. Sensitivity analyses were consistent with the main results.

Conclusion State religious freedom policies were associated with a 43% increase in the proportion of sexual minority adults experiencing severe mental distress. Lawmakers and courts considering religious freedom policies should consider the relationship between these policies and increases in severe mental distress among sexual minority adults.

Aims: Research from both U.S. and international contexts reports that suicide-attempting youths are at high risk for subsequent serious physical injury and disease (e.g., Goldman-Mellor et al. 2014). Such research suggests that these youths – whose numbers are increasing – may require substantial amounts of healthcare over their lifetime, along with incurring associated costs. This study assesses adolescent suicide attempters’ rate of inpatient visits, average length of inpatient stay, and total hospitalization costs over several years of follow-up, using statewide data from California. Methods: The California Office of Statewide Health Planning and Development (OSHPD) provided anonymized individual-level patient encounter data from all California-licensed hospital facilities from 2006-2015. For this study, the dataset consisted of ED and inpatient records for all adolescent patients aged 10 to 19 years with a valid unique identifier (encrypted social security number) and a California residential zip code in 2010 (n = 522,056). Unique identifiers were used to link multiple hospital visits per patient over time, including encounters at any hospital in the state prior (2006-2009) and subsequent (2010-2015) to the adolescent’s index visit in 2010. Suicide attempt cases comprised adolescent ED patients who presented in 2010 with a primary ICD-9-CM self-injury code of E950-959. Controls comprised all adolescent ED patients who presented in 2010 for any reason other than self-injury. For each patient, we assessed three outcomes (subsequent to the patient’s index ED visit) for the period 2010-2015: (1) Total number of inpatient visits, (2) Average length of inpatient stay, and (3) Total hospitalization charges. Charges included those for services rendered during the inpatient stay, including daily hospital services, ancillary services and any patient care services, and were based on the hospital's reported rates. Associations between exposure status (suicide attempter vs. control) and outcome variables were examined using negative binomial regression with robust standard errors. Included as controls in all multivariate analyses were measures of adolescents’ race, age, insurance status, rurality of residential zip code, and history of prior ED visits for self-injury, mental health problems, substance use problems, assault, unintentional injury, as well as total numbers of visits. Results: A total of 5,488 California adolescents made an ED visit for nonfatal suicide attempt in 2010 (mean age=16.6 years (SD=2.0); 64.0% female). Suicide-attempting adolescents had an average of 2.33 inpatient visits (SD=4.21) during the follow-up period, compared to non-attempters’ average of 0.68 inpatient visits (SD=1.81). Average length of inpatient stay was 4.07 days (SD=3.52) among suicide attempters, compared to 3.22 days (SD=3.97) among non-attempters. Total hospitalization charges accrued by suicide attempters during follow-up were greater: An average of $51,115 per patient, compared to $23,686 per patient among non-attempters. Negative binomial regression analyses that controlled for age, gender, race/ethnicity, insurance status, and history of ED visits confirmed that these differences were statistically significant, with suicide attempters experiencing 122% more visits, 5% longer average length of stays, and 48% higher total charges for the five-year follow-up period.

Mental illness is prevalent in the U.S.: 18.3% of American adults experience some form of mental illness. Mental illness treatments are effective; however, majority of individuals with mental illness do not receive any treatment. In 2016, less than half of adults with mental illness received any treatment. Inability to pay for treatment and lack of insurance coverage for mental illness treatment are key barriers to receiving treatment. Mental illness treatment is costly for an uninsured patient: mental healthcare provider reimbursement rates can range from $67 to $144 per visit. Historically, insurance coverage for mental healthcare has been less generous than general healthcare coverage. In an attempt to address discriminatory treatment of mental healthcare coverage, numerous U.S. states have implemented laws that compel private insurers to cover mental healthcare services at ‘parity’ with general healthcare services. Previous research has established that state mental illness parity laws improve access to mental healthcare and, in turn, reduce mental illness. Moreover, most mental illnesses develop during adolescence and early adulthood. I extend this literature in two important ways. First, I study the effect of the state mental illness parity law implementation on mental illness among college-age individuals. Second, I examine the effect of state mental illness parity laws on human capital accumulation. Considering spill-overs to these educational outcomes is important as previous research shows that mental illness impedes college performance. Hence, reduced mental illness through state parity laws could have positive spill-over effects to educational outcomes that have not yet been documented. I use differences-in-differences models to uncover the causal effects of state mental illness parity laws on mental illness and educational outcomes. I leverage plausibly exogenous variation in insurance coverage for mental healthcare using changes in state laws over the period 1998 to 2008. First, to study parity law effects on mental illness I utilize administrative data on completed suicides from National Vital Statistics System and survey data on reported mental illness from Behavioral Risk Factor System. Second, I use longitudinal data from the National Longitudinal Survey of Youth 1997 Cohort to study the effects of the mental health parity law on two important educational outcomes: drop out decisions and grade point average (GPA). Three main findings emerge from my analysis. First, I document that the passage of a mental health parity law leads to reductions in state-level suicide rate for the college-aged population, and reductions in the number of poor mental health days for the student population. Second, I find no evidence that the passage of a mental health parity law influences the propensity to drop out of college. Third, I show that state-level mental health parity laws have a significant positive effect on college GPA. The findings from this study can provide insights into the impact of the Affordable Care Act, which expands access to valuable mental healthcare services to millions of Americans. More broadly, these findings document important spill-over effects from public health policy to educational outcomes, and suggest that considering policies in isolation may underestimate their value to society.

Abstract

Psychology has taught us that self-concepts develop during adolescence—a time when youth are also completing their physical development. This study examines these two simultaneous development processes and how they relate. Using 15-years of panel data on adolescents, I evaluate the relationship between body perception and body weight. I determine how closely perception aligns with BMI and identify those environmental, regional and household characteristics associated with mis/perception. Finally, I identify how over or under estimation of body weight relates to subsequent growth and how perception defines the BMI growth trajectory. Results show that misperception is more common among adolescents at high BMI levels. Blacks have a higher probability of accurately assessing their bodies, while Hispanics have a lower probability. Overestimating body size leads to lower subsequent growth while underestimating is associated with higher growth. Results suggest that although self-views and physical features develop simultaneously, they do not always align.

: Latina adolescents have high rates of suicidal ideation and attempts (CDC, 2015). Life is Precious (LIP) serves Latina adolescents at risk for suicide in an after-school, clubhouse model that supplements ongoing outpatient mental health treatment with a set of culturally-competent services, including family support services, supported education services, and creative arts therapy. LIP operates three NYC locations, serving Latina girls aged 12-18 who have

: Since program inception in 2008, data has been obtained for 279 participants. Psychological assessment results Suicidal Ideation Questionnaire (SIQ), Reynolds Adolescent Depression Scale (RADS2), and Trauma Symptom Checklist

: There have been no completed suicides among program participants since the program’s inception in 2008, in this high-risk population. In the past year, there has been one reported suicide attempt (of 124 participants served over the course of the year at the three program sites). Suicidal ideation, as measured by SIQ, decreases by about 1 point per year of enrollment (p<0.01). Participants who report a history of sexual abuse have a decrease of about three points (p<0.01) and participants who report a history of alcohol use also have a decrease of about three points (p<0.02). Participants with a history of tobacco use have a decrease of about five points (p<0.01). Depressive symptoms also decrease: about 2.6 points per year as measured by the RADS2 and about one point per year as measured by the TSCC (p<0.01 for all). Other symptoms measured by TSCC also decrease: anger, anxiety and post-traumatic stress symptoms all decrease by about half a point per year (p<0.01 for all). There is no control group at this time, and assessments only measure participants while they are in the program.

: The program is showing low rates of suicide attempts and statistically significant decreases in suicidal ideation, depressive symptoms, and other symptoms during program participation. While changes to ideation and symptoms are small, they are statistically significant and are seen over time and in multiple assessment measures (SIQ, RADS2, TSCC). An expansion of this evaluation, which will compare outcomes to comparison samples who are receiving outpatient mental health treatment without the additional LIP clubhouse services, has been funded and is beginning.

A negative economic shocks may impose two opposing influences on mental health spending: pressure to decrease spending due to liquidity constraints, and pressure to increase spending due to worsening mental health status. Using two-year panel data from the Medical Expenditure Panel Survey for the period 2004 to 2012, we examine the effect of an economic shock on mental health spending by families with children. Specifically, we focus on the effect of changes in family income, employment status, and health insurance status over the two-year observation period within each MEPS panel. Since a substantial proportion of the population does not use mental health care and the distribution of spending is positively skewed we estimate a series of two-part model with probit equation in the first part and generalized linear model (GLM) with a log link and gamma variance function in the second part. To control for unobserved heterogeneity across families in the sample we apply the two-part model using the correlated random effects (CRE) framework.

Employment gains are associated with a decline in both the likelihood of mental health services utilization and in the amount being spent toward mental health services among those who use the services. For instance, gaining employment in two-parent families where a parent was not employed during entire year is associated, on average, with an expected decline of $172 in family mental health spending. Among single-mother families, gaining employment and continuing to be employed throughout the two-year observation period is associated with an expected decline of $307 in family mental health spending. The effect of an employment gain leading to a lower mental health services use is consistent with the hypothesis that gaining employment may improve mental health and thereby lead to a decreased demand for mental health

of mental health services utilization and in the amount being spent on such services among those who use services. For instance, losing employment after a recent employment gain increases the likelihood of mental health services utilization by about five percentage points. We also find that losing a job and remaining jobless for the remainder of the two-year observation period leads to an increase of $204 in expected family mental health among two-parent families. The sign of the employment loss effect may reflect pressure to increase mental health spending due to the possibility of worsening mental health status

We also find that income and health insurance loss is associated with a decline in mental health spending. For instance, mothers in two-parent families where a family member lost health insurance coverage spend less on mental health care. Similarly, negative income shocks in single-mother families decrease the amount of ambulatory mental health spending and the amount of mental health care spending incurred by mothers.

Evaluation of health care utilization is prominent in the economics literature, generally concentrating on the impact of certain health conditions on resource use, using individual level self-reported data. However, the important additional impact of cognitive impairment has not yet been explored. We now propose a model that assesses the impact of word recall, a standard validated measure of cognitive status, on health care utilisation. We develop an approach that simultaneously takes into account the count nature of the data, recall bias and reporting behaviours, and the role of cognitive state in all of these processes. Our model also introduces an individual specific random parameter. Using the Survey of Health, Ageing and Retirement in Europe data, we demonstrate that without modelling for unobserved heterogeneity via captivity effects and random parameters, the effect of cognitive status is largely underestimated. A striking

Studies of the effects of economic shocks on health have shown different conclusions about the nature of the recessionary-health relationship. This lack of a consensus may be due to the variations in size and type of recession. While unemployment was seen in most high income countries during the Great Recession, economic downturns also reduce real income. Furthermore, in Ireland and the United States the financial crisis also triggered mortgage distress. Actual or

I use data from the first three waves of the infant cohort Growing Up in Ireland study to examine the effect of income loss on the mental health of young mothers. The timing of the three waves encompasses the period of the financial crisis. I take a fixed effects approach to examine changes in total depression score (CES-D), and I also examine changes for the subgroups of mental health that make up the total score. This allows me to differentiate between feelings of failure and fear for the future, and symptoms consistent with sadness and the ‘blues’. I examine the effect of income loss on these measures mental health, and I then examine if this effect is due to the tenancy status of the respondent, or

A balanced panel of 6821 households is used in this study, and the individual of interest is the biological mother of the study child. The results of the fixed effects model show that income loss is associated with an increase in depression (-.245**). When the tenancy status of the individual is taken into account income loss is associated with depression for owners who have a mortgage (-.317***). This effect of income loss is not seen for those who rent from local authority, private owners, parents, or occupy their homes free of rent. When the income quintile of the individual is included, it is seen that those who were in the wealthiest income quintile in wave 1 have an increase in depression associated with income loss (-.253*). This is not seen for the other income quintiles. There is also an association between being in mortgage or rent arrears and an increase in depression (.159**). When the total depression score is broken down into symptoms of depression, income loss is associated with symptoms of being depressed (-.043*), failure (-.031**) and fear (-.04*) but not loneliness, crying, or being sad. However, income loss for those who pay a mortgage was associated with the ‘blues’ (-.036*), feeling depressed (-.05) failure (-.032**) fear (-.053**), restless sleep (-.068**) and being sad (-.044*).

The results indicate that the Great Recession had an effect on the mental health of young mothers. This was due to income loss rather than unemployment, despite the fact that unemployment is commonly used as an indicator of recessionary effect. When the effect is examined based on the tenancy of the household, income loss for mortgage owners was associated with depression.

effect on high school students' mental health and suicide ideation across different gender and body-

Recent policy proposals have considered cutting billions in federal funding for Medicaid and restricting the program’s eligibility criteria, potentially leading to coverage loss for beneficiaries with mental health disorders. Little is known about the dynamic relationship between Medicaid coverage loss (or “churning”) and mental health care utilization. This analysis examined how outpatient healthcare and mental health service use changes when patients lose Medicaid coverage and become uninsured among a nationwide cohort of adults with mental health disorders, and how these changes progress over time while the individuals remain uninsured.

Our sample included 4,822 persons (102,261 person-months) ages 18-64 with mental health disorders from the 2001-2014 Medical Expenditure Panel Survey. We examined patterns of healthcare service use pre-churning (while covered by Medicaid) and post-churning (while uninsured) among individuals who lost Medicaid coverage, and compared them with service use in a comparison group with continuous Medicaid enrollment. We identified the comparison group using a propensity score matching method. In our models, a simple “post” indicator was first used to estimate the average differences in healthcare service use post-churning, and a count variable measuring total months since churning—with linear and non-linear terms—was subsequently used to estimate changes over time. We estimated logit models to estimate changes in the likelihoods of any outpatient visit and any mental health related outpatient visit per-person-per-month. We also used two-part models to estimate total healthcare costs and out-of-pocket costs per-person-per-month.

Becoming uninsured after losing Medicaid coverage was associated with a 53% reduction (marginal effects [ME] = -15.09 percentage points, 95% CI: [-17.23, -12.95]) in the likelihood of receiving any outpatient services, and a 43% reduction (ME = -2.76 percentage points, 95% CI: [-4.22, -1.31]) in the likelihood of receiving any mental health related outpatient services in a month. Healthcare service use declined the most in the month immediately post-churning. Healthcare use trends declined further and flattened over the next half-year while the patients remained uninsured.

Moreover, the reductions in healthcare use post-churning was translated into a 53% reduction (ME = -$153.73; 95% CI: [-224.24, -83.23]) in total healthcare costs per-person-per-month. Further, we observed a 66% increase (ME = $4.85; 95% CI: [2.48, 7.22]) in patients’ out-of-pocket costs per-person-per-month. When examining how these trends changed over the six months following loss of Medicaid coverage, total costs continued to decline and out-of-pocket costs

Our analysis raises important considerations about the implications of Medicaid churning for low-income patients with mental health problems. Findings from this analysis suggest that loss of insurance coverage immediately after dis-enrolling from Medicaid disrupted the receipt of outpatient services among adult beneficiaries with mental health disorders, and this disruption persisted for an extended period of time post-churning. As states consider the future of their Medicaid programs in a dynamic policy landscape, it will be crucial to implement strategies to mitigate Medicaid churning and ensure adequate receipt of mental health related outpatient services for vulnerable populations.

A 2007 United Nations Children’s Fund and an OECD report comparing child well-being across 21 industrialized nations, placed the UK at the bottom of the rankings for measures of children socio-emotional well-being (SEW). In addition to being important for health, SEW relates to the “Big 5” personality traits and non-cognitive skills which are key predictors for later life employment opportunities. This paper evaluates the effect of parental education on children socio-emotional skills. I use a rich cohort dataset, the Avon Longitudinal Study of Parents and Children (ALSPAC) containing measures of SEW such as the Strengths and Difficulty Questionnaire (SDQ) from 4 to 13 years old. I exploit the fact that a proportion of the parents in ALSPAC were impacted by the most recent raising of the minimum school leaving age (RoSLA) in England which occurred in 1972. Using a Regression Discontinuity Design, the structure of the ALSPAC data allows me to identify the causal impact of the policy separately from the effect of parents age at the time of the child’s birth. Theoretical work on the skills production function by Conti and Heckman, Cunha and Heckman, and others suggest that an individual’s skills are multi-dimensional and comprise socio-emotional skills in addition to cognitive skills. There is a dynamic technology of skill formation function that depends, amongst other inputs, on parental investments. As early life SEW improves both SEW throughout the lifecourse and impacts the productivity of future investments, this model

This paper relates to two strands of the economics literature. The first strand examines the determinants of non-cognitive abilities and socio-emotional skills. The majority of these studies have focused on the role of parental income using measures of permanent income or exogenous intervention-related household income shocks. They find that income has a cumulative positive effect on non-cognitive skills which increases as the cohort ages. Likely mechanisms are improvements in parent-child relationships. The second strand of the literature investigates the effect of parental education on children’s outcomes such as education, health and cognitive skills finding effects that are visible at older ages. I contribute to both strands of the empirical literature. Compared to the non-cognitive skills literature, I consider a previously overlooked component of the non-cognitive skills production function, parental education, and exploit a natural experiment. My paper is the first to investigate the intergenerational transmission of socio-emotional skills in a large cohort study in the UK. I consider the mechanisms through which parental education might affect children socio-emotional skills including assortative mating, earnings and better information on parenting skills. I also consider whether the effect varies across the parental educational distribution and depending on the gender of children. Very preliminary results indicate that parental education improve children socio-emotional skills. This effect is stronger for the children of parents who would have not stayed in school longer. There is no difference between children’s

Suicide is the leading cause of death among people with substance use disorders (SUDs). From 2006 to 2013, the rate of emergency department (ED) adult visits involving SUDs increased by 34% and the rate of ED visits related to suicidal ideation more than doubled from 173 to 376 visits per 100,000 population. Together, 42% of ED suicidal ideation-related visits were associated with SUDs in 2013. Though, numerous studies indicate ED-initiated interventions have been successful in reducing adult suicidal ideation and intentional self-inflicted injury, none are wide-spread; and while other studies recommend interventions outside the ED, none examine the role of local health departments (LHDs) in

The objective of this study is to examine whether LHDs’ active roles of health promotion are associated with reductions in the rates of suicidal ideation and intentional self-inflicted injury in the ED settings.

Using data sets linked from multiple sources, including 2012-2013 State ED Databases for the State of Maryland, the National Association of County and City Health Officials Profiles Survey, the Area Health Resource File, and U.S. Census data, we employed multi-level/hierarchical logistic models to examine whether LHDs’ active provisions of preventive care and primary care services and health policy advocacy (such as affordable housing, mental health, education, etc.) were associated with the reduction of having suicidal ideation and intentional self-inflicted injuries. Analyses were conducted among individuals with substance use disorders (SUDs) aged 18 and above.

People with SUDs who committed suicidal ideation and intentional self-inflicted injuries were more likely to be White, had more chronic conditions, and were more likely to live in rural areas. The rate of suicidal ideation and intentional self-inflicted injury was 0.78% overall, but the rate increased to 7.26% of among patients with SUDs. Levels of LHDs’ activities vary by services. Approximately 7.23% of patients with SUDs resided in counties with LHDs’ active advocacy for affordable housing, and 92.07% patients with SUDs resided in counties with LHDS’ active education on tobacco, alcohol, or other drugs. After adjusting for individual-, hospital-, LHD-, and county-level characteristics, multilevel logistic regressions showed that LHDs’ health promotions on affordable housing (OR=0.65 , p<0.05) and education on tobacco, alcohol, or other drugs (OR=0.63, p<0.001) were significantly associated with the reduction of suicidal ideation and intentional self-inflicted injury in the ED settings for people with SUDs. Associations of LHDs’ provision of behavioral health preventive care and health services were not significant.

Suicide is the product of multiple causes, its prevention requires broad, interdisciplinary approaches. Though results demonstrate a benefit of health education related to affordable housing, tobacco, alcohol, and other drugs, there is a need to further examine the impact of LHD activity on mental health and the health system. Future studies should drill down into the SUDs variable to determine if these results vary by alcohol or drug use; and should explore the policy differences of jurisdictional tobacco control and housing policies on suicidal ideation and intentional self-inflicted injury.

Follow-up care after a hospitalization for mental illness within 7 (or 30) days is a widely used performance measure for mental health care, but little empirical evidence supports its use. Coordination of mental health care is difficult, and no empirical studies examine whether improving follow-up after hospitalizations decreases health care spending or utilization. To examine associations between mental health provider follow-up after a hospital visit for mental illness and utilization and cost outcomes using health insurance enrollment and claims data from all payer claims databases in four states: Massachusetts, New Hampshire, Maryland, and Utah. The three outcome measures are total healthcare spending during the 180-day follow-up period after a hospital visit for mental illness, whether the patient is readmitted; and whether the patient has an ED visit post discharge. The independent measure of interest is whether or not the

I expect to find follow-up with a mental health provider within 7 days is associated with reduced readmissions and reduced ED visits, with minimal impacts on healthcare spending. I expect to find heterogeneity in these effects based on patient demographic and clinical characteristics. Practice improvements enabled by this research — for example, better targeting individuals with an increased risk of poor ambulatory follow-up care, identifying hospital characteristics important for improved follow-up, and the use of decision support mechanisms to improve cross-sector care delivery — have substantial implications for mental health outcomes and health costs. As private insurance and Medicaid dominate the financing of mental health care, this study helps fill a critical hole in existing research by strengthening the empirical evidence base for the calculation and use of mental health performance measures.

Between 2015 and 2017, 12 states passed religious freedom laws affecting sexual minority rights. Sexual minorities have a disproportionate burden of poor mental health that may be affected by religious freedom laws. We investigated whether 2015 state religious freedom laws were associated with changes in sexual minority mental health between 2014 and 2016. We used 2014-2016 Behavioral Risk Factor Surveillance System (BRFSS) data representative of adults aged 18-65 years in each state. We used data from three states that implemented religious freedom laws in 2015 and that included questions on sexual orientation identity in the state BRFSS: North Carolina, Utah, and Michigan. We compared these states to three control states that collected data on sexual orientation identity and that had populations most closely matching the racial/ethnic characteristics and income distributions of treatment states: Virginia, Idaho, and Pennsylvania, respectively. The main outcome of interest was severe mental distress, defined as poor mental health in 14 or more of the past 30 days. We conducted a difference-in-differences (DD) analysis, comparing changes in severe mental distress in religious freedom states to changes in severe mental distress in control states. We estimated a linear regression model based on evidence that logistic regression estimates can be biased in the presence of fixed effects. The main exposure of interest was living in a state that implemented a religious freedom law in the prior calendar year and identifying as a sexual minority, modeled as an interaction term. In addition to including binary terms for sexual minority identity and for states implementing religious freedom laws in the prior year, we controlled for individual-level state, year, sex, race, ethnicity, age group, educational attainment, income, employment, and marital status. Controlling for each state had the effect of controlling for all time-invariant state characteristics. Because DD clustered standard error estimates can be biased downward, we conducted permutation tests to estimate statistical significance. We conducted several sensitivity analyses, including estimating a logistic regression model.

Prior to religious freedom policies, 12.7% of adults and 20.4% of sexual minority adults reported severe mental distress. In states that passed religious freedom policies in 2015, severe mental distress among sexual minorities increased from 21.9% in 2014 to 32.8% in 2016. In control states, severe mental distress among sexual minorities increased from 19.2% in 2014 to 26.3% in 2016. Based on the difference-in-differences analysis, religious freedom laws were associated with an 8.8 percentage point (95% confidence interval: 3.4 to 14.2 percentage points, permutation-adjusted p-value<0.01) increase, a 43% relative increase, in the proportion of sexual minority adults experiencing severe mental distress. Religious freedom laws were not statistically significantly associated with changes in severe mental distress among all adults. Sensitivity analyses were consistent with the main results.

State religious freedom policies were associated with a 43% increase in the proportion of sexual minority adults experiencing severe mental distress. Lawmakers and courts considering religious freedom policies should consider the relationship between these policies and increases in severe mental distress among sexual minority adults.

Research from both U.S. and international contexts reports that suicide-attempting youths are at high risk for subsequent serious physical injury and disease (e.g., Goldman-Mellor et al. 2014). Such research suggests that these youths – whose numbers are increasing – may require substantial amounts of healthcare over their lifetime, along with incurring associated costs. This study assesses adolescent suicide attempters’ rate of inpatient visits, average length of inpatient stay, and total hospitalization costs over several years of follow-up, using statewide data from California.

The California Office of Statewide Health Planning and Development (OSHPD) provided anonymized individual-level patient encounter data from all California-licensed hospital facilities from 2006-2015. For this study, the dataset consisted of ED and inpatient records for all adolescent patients aged 10 to 19 years with a valid unique identifier (encrypted social security number) and a California residential zip code in 2010 (n = 522,056). Unique identifiers were used to link multiple hospital visits per patient over time, including encounters at any hospital in the state prior (2006-2009) and subsequent (2010-2015) to the adolescent’s index visit in 2010. Suicide attempt cases comprised adolescent ED patients who presented in 2010 with a primary ICD-9-CM self-injury code of E950-959. Controls comprised all adolescent ED patients who presented in 2010 for any reason other than self-injury. For each patient, we assessed three outcomes (subsequent to the patient’s index ED visit) for the period 2010-2015: (1) Total number of inpatient visits, (2) Average length of inpatient stay, and (3) Total hospitalization charges. Charges included those for services rendered during the inpatient stay, including daily hospital services, ancillary services and any patient care services, and were based on the hospital's reported rates. Associations between exposure status (suicide attempter vs. control) and outcome variables were examined using negative binomial regression with robust standard errors. Included as controls in all multivariate analyses were measures of adolescents’ race, age, insurance status, rurality of residential zip code, and history of prior ED visits for self-injury, mental health problems, substance use problems, assault, unintentional injury, as well as total numbers of visits.

A total of 5,488 California adolescents made an ED visit for nonfatal suicide attempt in 2010 (mean age=16.6 years (SD=2.0); 64.0% female). Suicide-attempting adolescents had an average of 2.33 inpatient visits (SD=4.21) during the follow-up period, compared to non-attempters’ average of 0.68 inpatient visits (SD=1.81). Average length of inpatient stay was 4.07 days (SD=3.52) among suicide attempters, compared to 3.22 days (SD=3.97) among non-attempters. Total hospitalization charges accrued by suicide attempters during follow-up were greater: An average of $51,115 per patient, compared to $23,686 per patient among non-attempters. Negative binomial regression analyses that controlled for age, gender, race/ethnicity, insurance status, and history of ED visits confirmed that these differences were statistically significant, with suicide attempters experiencing 122% more visits, 5% longer average length of stays, and 48%

Mental illness is prevalent in the U.S.: 18.3% of American adults experience some form of mental illness. Mental illness treatments are effective; however, majority of individuals with mental illness do not receive any treatment. In 2016, less than half of adults with mental illness received any treatment. Inability to pay for treatment and lack of insurance coverage for mental illness treatment are key barriers to receiving treatment. Mental illness treatment is costly for an uninsured patient: mental healthcare provider reimbursement rates can range from $67 to $144 per visit. Historically, insurance coverage for mental healthcare has been less generous than general healthcare coverage. In an attempt to address discriminatory treatment of mental healthcare coverage, numerous U.S. states have implemented laws that compel private insurers to cover mental healthcare services at ‘parity’ with general healthcare services. Previous research has established that state mental illness parity laws improve access to mental healthcare and, in turn, reduce mental illness. Moreover, most mental illnesses develop during adolescence and early adulthood. I extend this literature in two important ways. First, I study the effect of the state mental illness parity law implementation on mental illness among college-age individuals. Second, I examine the effect of state mental illness parity laws on human capital accumulation. Considering spill-overs to these educational outcomes is important as previous research shows that mental illness impedes college performance. Hence, reduced mental illness through state parity laws could have positive spill-over effects to educational outcomes that have not yet been documented. I use differences-in-differences models to uncover the causal effects of state mental illness parity laws on mental illness and educational outcomes. I leverage plausibly exogenous variation in insurance coverage for mental healthcare using changes in state laws over the period 1998 to 2008. First, to study parity law effects on mental illness I utilize administrative data on completed suicides from National Vital Statistics System and survey data on reported mental illness from Behavioral Risk Factor System. Second, I use longitudinal data from the National Longitudinal Survey of Youth 1997 Cohort to study the effects of the mental health parity law on two important educational outcomes: drop out decisions

Three main findings emerge from my analysis. First, I document that the passage of a mental health parity law leads to reductions in state-level suicide rate for the college-aged population, and reductions in the number of poor mental health days for the student population. Second, I find no evidence that the passage of a mental health parity law influences the propensity to drop out of college. Third, I show that state-level mental health parity laws have a significant

The findings from this study can provide insights into the impact of the Affordable Care Act, which expands access to valuable mental healthcare services to millions of Americans. More broadly, these findings document important spill-over effects from public health policy to educational outcomes, and suggest that considering policies in isolation may underestimate their value to society.


Molly Jacobs [email protected]

Jennifer Humensky [email protected]

Irina Grafova [email protected]

Brenda Gannon [email protected]

Psychology has taught us that self-concepts develop during adolescence—a time when youth are also completing their physical development. This study examines these two simultaneous development processes and how they relate. Using 15-years of panel data on adolescents, I evaluate the relationship between body perception and body weight. I determine how closely perception aligns with BMI and identify those environmental, regional and household characteristics associated with mis/perception. Finally, I identify how over or under estimation of body weight relates to subsequent growth and how perception defines the BMI growth trajectory. Results show that misperception is more common among adolescents at high BMI levels. Blacks have a higher probability of accurately assessing their bodies, while Hispanics have a lower probability. Overestimating body size leads to lower subsequent growth while underestimating is

: Latina adolescents have high rates of suicidal ideation and attempts (CDC, 2015). Life is Precious (LIP) serves Latina adolescents at risk for suicide in an after-school, clubhouse model that supplements ongoing outpatient mental health treatment with a set of culturally-competent services, including family support services, supported education services, and creative arts therapy. LIP operates three NYC locations, serving Latina girls aged 12-18 who have

Reynolds Adolescent Depression Scale (RADS2), and Trauma Symptom Checklist

: There have been no completed suicides among program participants since the program’s inception in 2008, in this high-risk population. In the past year, there has been one reported suicide attempt (of 124 participants served over the course of the year at the three program sites). Suicidal ideation, as measured by SIQ, decreases by about 1 point per year of enrollment (p<0.01). Participants who report a history of sexual abuse have a decrease of about three points (p<0.01) and participants who report a history of alcohol use also have a decrease of about three points (p<0.02). Participants with a history of tobacco use have a decrease of about five points (p<0.01). Depressive symptoms also decrease: about 2.6 points per year as measured by the RADS2 and about one point per year as measured by the TSCC (p<0.01 for all). Other symptoms measured by TSCC also decrease: anger, anxiety and post-traumatic stress symptoms

: The program is showing low rates of suicide attempts and statistically significant decreases in suicidal ideation, depressive symptoms, and other symptoms during program participation. While changes to ideation and symptoms are small, they are statistically significant and are seen over time and in multiple assessment measures (SIQ, RADS2, TSCC). An expansion of this evaluation, which will compare outcomes to comparison samples who are receiving

A negative economic shocks may impose two opposing influences on mental health spending: pressure to decrease spending due to liquidity constraints, and pressure to increase spending due to worsening mental health status. Using two-year panel data from the Medical Expenditure Panel Survey for the period 2004 to 2012, we examine the effect of an economic shock on mental health spending by families with children. Specifically, we focus on the effect of changes

Since a substantial proportion of the population does not use mental health care and the distribution of spending is positively skewed we estimate a series of two-part model with probit equation in the first part and generalized linear model (GLM) with a log link and gamma variance function in the second part. To control for unobserved heterogeneity across families in the sample we apply the two-part model using the correlated random effects (CRE) framework.

of mental health services utilization and in the amount being spent toward mental health services among those who use the services. For instance, gaining employment in two-parent families where a parent was not employed during entire year is associated, on average, with an expected decline of $172 in family mental health spending. Among single-mother families, gaining employment and continuing to be employed throughout the two-year observation period is associated with an expected decline of $307 in family mental health spending. The effect of an employment gain leading to a lower mental health services use is consistent with the hypothesis that gaining employment may improve mental health and thereby lead to a decreased demand for mental health

being spent on such services among those who use services. For instance, losing employment after a recent employment gain increases the likelihood of mental health services utilization by about five percentage points. We also find that losing a job and remaining jobless for the remainder of the two-year observation period leads to an increase of $204 in expected family mental health among two-parent families. The sign of the employment loss effect may reflect pressure to increase mental health spending due to the possibility of worsening mental health status

We also find that income and health insurance loss is associated with a decline in mental health spending. For instance, mothers in two-parent families where a family member lost health insurance coverage spend less on mental health care. Similarly, negative income shocks in single-mother families decrease the amount of ambulatory mental health spending and the amount of mental health care spending incurred by mothers.

Evaluation of health care utilization is prominent in the economics literature, generally concentrating on the impact of certain health conditions on resource use, using individual level self-reported data. However, the important additional impact of cognitive impairment has not yet been explored. We now propose a model that assesses the impact of word recall, a standard validated measure of cognitive status, on health care utilisation. We develop an approach that simultaneously takes into account the count nature of the data, recall bias and reporting behaviours, and the role of cognitive state in all of these processes. Our model also introduces an individual specific random parameter. Using the Survey of Health, Ageing and Retirement in Europe data, we demonstrate that without modelling for unobserved heterogeneity via captivity effects and random parameters, the effect of cognitive status is largely underestimated. A striking

Fiona Kiernan [email protected]

Qi Sun [email protected]

Studies of the effects of economic shocks on health have shown different conclusions about the nature of the recessionary-health relationship. This lack of a consensus may be due to the variations in size and type of recession. While unemployment was seen in most high income countries during the Great Recession, economic downturns also reduce real income. Furthermore, in Ireland and the United States the financial crisis also triggered mortgage distress. Actual or

I use data from the first three waves of the infant cohort Growing Up in Ireland study to examine the effect of income loss on the mental health of young mothers. The timing of the three waves encompasses the period of the financial crisis. I take a fixed effects approach to examine changes in total depression score (CES-D), and I also examine changes for the subgroups of mental health that make up the total score. This allows me to differentiate between feelings of failure and fear for the future, and symptoms consistent with sadness and the ‘blues’. I examine the effect of income loss on these measures mental health, and I then examine if this effect is due to the tenancy status of the respondent, or

The results of the fixed effects model show that income loss is associated with an increase in depression (-.245**). When the tenancy status of the individual is taken into account income loss is associated with depression for owners who have a mortgage (-.317***). This effect of income loss is not seen for those who rent from local authority, private owners, parents, or occupy their homes free of rent. When the income quintile of the individual is included, it is seen that those who were in the wealthiest income quintile in wave 1 have an increase in depression associated with income loss (-.253*). This is not seen for the other income quintiles. There is also an association between being in mortgage or rent arrears and an increase in depression (.159**). When the total depression score is broken down into symptoms of depression, income loss is associated with symptoms of being depressed (-.043*), failure (-.031**) and fear (-.04*) but not loneliness, crying, or being sad. However, income loss for those who pay a mortgage was associated with the ‘blues’ (-.036*), feeling depressed (-.05) failure (-.032**) fear (-.053**), restless sleep (-.068**) and being sad (-.044*).

The results indicate that the Great Recession had an effect on the mental health of young mothers. This was due to income loss rather than unemployment, despite the fact that unemployment is commonly used as an indicator of

Xu Ji [email protected]

Eleonora Fichera [email protected]

Recent policy proposals have considered cutting billions in federal funding for Medicaid and restricting the program’s eligibility criteria, potentially leading to coverage loss for beneficiaries with mental health disorders. Little is known about the dynamic relationship between Medicaid coverage loss (or “churning”) and mental health care utilization. This analysis examined how outpatient healthcare and mental health service use changes when patients lose Medicaid

Our sample included 4,822 persons (102,261 person-months) ages 18-64 with mental health disorders from the 2001-2014 Medical Expenditure Panel Survey. We examined patterns of healthcare service use pre-churning (while covered by Medicaid) and post-churning (while uninsured) among individuals who lost Medicaid coverage, and compared them with service use in a comparison group with continuous Medicaid enrollment. We identified the comparison group using a propensity score matching method. In our models, a simple “post” indicator was first used to estimate the average differences in healthcare service use post-churning, and a count variable measuring total months since churning—with linear and non-linear terms—was subsequently used to estimate changes over time. We estimated logit models to estimate changes in the likelihoods of any outpatient visit and any mental health related outpatient visit per-person-

Becoming uninsured after losing Medicaid coverage was associated with a 53% reduction (marginal effects [ME] = -15.09 percentage points, 95% CI: [-17.23, -12.95]) in the likelihood of receiving any outpatient services, and a 43% reduction (ME = -2.76 percentage points, 95% CI: [-4.22, -1.31]) in the likelihood of receiving any mental health related outpatient services in a month. Healthcare service use declined the most in the month immediately post-churning.

Moreover, the reductions in healthcare use post-churning was translated into a 53% reduction (ME = -$153.73; 95% CI: [-224.24, -83.23]) in total healthcare costs per-person-per-month. Further, we observed a 66% increase (ME = $4.85; 95% CI: [2.48, 7.22]) in patients’ out-of-pocket costs per-person-per-month. When examining how these trends changed over the six months following loss of Medicaid coverage, total costs continued to decline and out-of-pocket costs

Our analysis raises important considerations about the implications of Medicaid churning for low-income patients with mental health problems. Findings from this analysis suggest that loss of insurance coverage immediately after dis-enrolling from Medicaid disrupted the receipt of outpatient services among adult beneficiaries with mental health disorders, and this disruption persisted for an extended period of time post-churning. As states consider the future of their Medicaid programs in a dynamic policy landscape, it will be crucial to implement strategies to mitigate Medicaid churning and ensure adequate receipt of mental health related outpatient services for vulnerable populations.

A 2007 United Nations Children’s Fund and an OECD report comparing child well-being across 21 industrialized nations, placed the UK at the bottom of the rankings for measures of children socio-emotional well-being (SEW). In addition to

This paper evaluates the effect of parental education on children socio-emotional skills. I use a rich cohort dataset, the Avon Longitudinal Study of Parents and Children (ALSPAC) containing measures of SEW such as the Strengths and Difficulty Questionnaire (SDQ) from 4 to 13 years old. I exploit the fact that a proportion of the parents in ALSPAC were impacted by the most recent raising of the minimum school leaving age (RoSLA) in England which occurred in 1972. Using a Regression Discontinuity Design, the structure of the ALSPAC data allows me to identify the causal impact of the policy separately from the effect of parents age at the time of the child’s birth. Theoretical work on the skills production function by Conti and Heckman, Cunha and Heckman, and others suggest that an individual’s skills are multi-dimensional and comprise socio-emotional skills in addition to cognitive skills. There is a dynamic technology of skill formation function that depends, amongst other inputs, on parental investments. As early life SEW improves both SEW throughout the lifecourse and impacts the productivity of future investments, this model

This paper relates to two strands of the economics literature. The first strand examines the determinants of non-cognitive abilities and socio-emotional skills. The majority of these studies have focused on the role of parental income using measures of permanent income or exogenous intervention-related household income shocks. They find that income has a cumulative positive effect on non-cognitive skills which increases as the cohort ages. Likely mechanisms are improvements in parent-child relationships. The second strand of the literature investigates the effect of parental education on children’s outcomes such as education, health and cognitive skills finding effects that are visible at older ages. I contribute to both strands of the empirical literature. Compared to the non-cognitive skills literature, I consider a previously overlooked component of the non-cognitive skills production function, parental education, and exploit a natural experiment. My paper is the first to investigate the intergenerational transmission of socio-emotional skills in a large cohort study in the UK. I consider the mechanisms through which parental education might affect children socio-emotional skills including assortative mating, earnings and better information on parenting skills. I also consider whether the effect varies across the parental educational distribution and depending on the gender of children. Very preliminary results indicate that parental education improve children socio-emotional skills. This effect is stronger for the children of parents who would have not stayed in school longer. There is no difference between children’s

Deanna Barath [email protected]

Kimberley Geissler [email protected]

Suicide is the leading cause of death among people with substance use disorders (SUDs). From 2006 to 2013, the rate of emergency department (ED) adult visits involving SUDs increased by 34% and the rate of ED visits related to suicidal ideation more than doubled from 173 to 376 visits per 100,000 population. Together, 42% of ED suicidal ideation-related visits were associated with SUDs in 2013. Though, numerous studies indicate ED-initiated interventions have been successful in reducing adult suicidal ideation and intentional self-inflicted injury, none are wide-spread; and while other studies recommend interventions outside the ED, none examine the role of local health departments (LHDs) in

The objective of this study is to examine whether LHDs’ active roles of health promotion are associated with reductions in the rates of suicidal ideation and intentional self-inflicted injury in the ED settings.

Using data sets linked from multiple sources, including 2012-2013 State ED Databases for the State of Maryland, the National Association of County and City Health Officials Profiles Survey, the Area Health Resource File, and U.S. Census data, we employed multi-level/hierarchical logistic models to examine whether LHDs’ active provisions of preventive care and primary care services and health policy advocacy (such as affordable housing, mental health, education, etc.) were associated with the reduction of having suicidal ideation and intentional self-inflicted injuries. Analyses were conducted among individuals with substance use disorders (SUDs) aged 18 and above.

People with SUDs who committed suicidal ideation and intentional self-inflicted injuries were more likely to be White, had more chronic conditions, and were more likely to live in rural areas. The rate of suicidal ideation and intentional self-inflicted injury was 0.78% overall, but the rate increased to 7.26% of among patients with SUDs. Levels of LHDs’ activities vary by services. Approximately 7.23% of patients with SUDs resided in counties with LHDs’ active advocacy for affordable housing, and 92.07% patients with SUDs resided in counties with LHDS’ active education on tobacco, alcohol, or other drugs. After adjusting for individual-, hospital-, LHD-, and county-level characteristics, multilevel logistic regressions showed that LHDs’ health promotions on affordable housing (OR=0.65 , p<0.05) and education on tobacco, alcohol, or other drugs (OR=0.63, p<0.001) were significantly associated with the reduction of suicidal ideation and

Suicide is the product of multiple causes, its prevention requires broad, interdisciplinary approaches. Though results demonstrate a benefit of health education related to affordable housing, tobacco, alcohol, and other drugs, there is a need to further examine the impact of LHD activity on mental health and the health system. Future studies should drill down into the SUDs variable to determine if these results vary by alcohol or drug use; and should explore the policy

Follow-up care after a hospitalization for mental illness within 7 (or 30) days is a widely used performance measure for mental health care, but little empirical evidence supports its use. Coordination of mental health care is difficult, and no empirical studies examine whether improving follow-up after hospitalizations decreases health care spending or utilization. To examine associations between mental health provider follow-up after a hospital visit for mental illness and utilization and cost outcomes using health insurance enrollment and claims data from all payer claims databases in four states: Massachusetts, New Hampshire, Maryland, and Utah. The three outcome measures are total healthcare spending during the 180-day follow-up period after a hospital visit for mental illness, whether the patient is readmitted; and whether the patient has an ED visit post discharge. The independent measure of interest is whether or not the

I expect to find follow-up with a mental health provider within 7 days is associated with reduced readmissions and reduced ED visits, with minimal impacts on healthcare spending. I expect to find heterogeneity in these effects based on patient demographic and clinical characteristics. Practice improvements enabled by this research — for example, better targeting individuals with an increased risk of poor ambulatory follow-up care, identifying hospital characteristics important for improved follow-up, and the use of decision support mechanisms to improve cross-sector care delivery — have substantial implications for mental health outcomes and health costs. As private insurance and Medicaid dominate the financing of mental health care, this study helps fill a critical hole in existing research by strengthening the empirical evidence base for the calculation and use of mental health performance measures.

Julia Raifman [email protected]

Gilda Zarate-Gonzalez [email protected]

Between 2015 and 2017, 12 states passed religious freedom laws affecting sexual minority rights. Sexual minorities have a disproportionate burden of poor mental health that may be affected by religious freedom laws. We investigated

We used 2014-2016 Behavioral Risk Factor Surveillance System (BRFSS) data representative of adults aged 18-65 years in each state. We used data from three states that implemented religious freedom laws in 2015 and that included questions on sexual orientation identity in the state BRFSS: North Carolina, Utah, and Michigan. We compared these states to three control states that collected data on sexual orientation identity and that had populations most closely matching the racial/ethnic characteristics and income distributions of treatment states: Virginia, Idaho, and Pennsylvania, respectively. The main outcome of interest was severe mental distress, defined as poor mental health in 14 or more of the past 30 days. We conducted a difference-in-differences (DD) analysis, comparing changes in severe mental distress in religious freedom states to changes in severe mental distress in control states. We estimated a linear regression model based on evidence that logistic regression estimates can be biased in the presence of fixed effects. The main exposure of interest was living in a state that implemented a religious freedom law in the prior calendar year and identifying as a sexual minority, modeled as an interaction term. In addition to including binary terms for sexual minority identity and for states implementing religious freedom laws in the prior year, we controlled for individual-level state, year, sex, race, ethnicity, age group, educational attainment, income, employment, and marital status. Controlling for each state had the effect of controlling for all time-invariant state characteristics. Because DD clustered standard error estimates can be biased downward, we conducted permutation tests to estimate statistical significance. We conducted several sensitivity analyses, including estimating a logistic regression model.

Prior to religious freedom policies, 12.7% of adults and 20.4% of sexual minority adults reported severe mental distress. In states that passed religious freedom policies in 2015, severe mental distress among sexual minorities increased from 21.9% in 2014 to 32.8% in 2016. In control states, severe mental distress among sexual minorities increased from 19.2% in 2014 to 26.3% in 2016. Based on the difference-in-differences analysis, religious freedom laws were associated with an 8.8 percentage point (95% confidence interval: 3.4 to 14.2 percentage points, permutation-adjusted p-value<0.01) increase, a 43% relative increase, in the proportion of sexual minority adults experiencing severe mental distress. Religious freedom laws were not statistically significantly associated with changes in severe mental distress among all adults. Sensitivity analyses were consistent with the main results.

State religious freedom policies were associated with a 43% increase in the proportion of sexual minority adults experiencing severe mental distress. Lawmakers and courts considering religious freedom policies should consider the

Research from both U.S. and international contexts reports that suicide-attempting youths are at high risk for subsequent serious physical injury and disease (e.g., Goldman-Mellor et al. 2014). Such research suggests that these youths – whose numbers are increasing – may require substantial amounts of healthcare over their lifetime, along with incurring associated costs. This study assesses adolescent suicide attempters’ rate of inpatient visits, average length of

The California Office of Statewide Health Planning and Development (OSHPD) provided anonymized individual-level patient encounter data from all California-licensed hospital facilities from 2006-2015. For this study, the dataset consisted of ED and inpatient records for all adolescent patients aged 10 to 19 years with a valid unique identifier (encrypted social security number) and a California residential zip code in 2010 (n = 522,056). Unique identifiers were used to link multiple hospital visits per patient over time, including encounters at any hospital in the state prior (2006-2009) and subsequent (2010-2015) to the adolescent’s index visit in 2010. Suicide attempt cases comprised adolescent ED patients who presented in 2010 with a primary ICD-9-CM self-injury code of E950-959. Controls comprised all adolescent ED patients who presented in 2010 for any reason other than self-injury. For each patient, we assessed three outcomes (subsequent to the patient’s index ED visit) for the period 2010-2015: (1) Total number of inpatient visits, (2) Average length of inpatient stay, and (3) Total hospitalization charges. Charges included those for services rendered during the inpatient stay, including daily hospital services, ancillary services and any patient care services, and were based on the hospital's reported rates. Associations between exposure status (suicide attempter vs. control) and outcome variables were examined using negative binomial regression with robust standard errors. Included as controls in all multivariate analyses were measures of adolescents’ race, age, insurance status, rurality of residential zip code, and history

A total of 5,488 California adolescents made an ED visit for nonfatal suicide attempt in 2010 (mean age=16.6 years (SD=2.0); 64.0% female). Suicide-attempting adolescents had an average of 2.33 inpatient visits (SD=4.21) during the follow-up period, compared to non-attempters’ average of 0.68 inpatient visits (SD=1.81). Average length of inpatient stay was 4.07 days (SD=3.52) among suicide attempters, compared to 3.22 days (SD=3.97) among non-attempters. Total hospitalization charges accrued by suicide attempters during follow-up were greater: An average of $51,115 per patient, compared to $23,686 per patient among non-attempters. Negative binomial regression analyses that controlled for age, gender, race/ethnicity, insurance status, and history of ED visits confirmed that these differences were statistically significant, with suicide attempters experiencing 122% more visits, 5% longer average length of stays, and 48%

Keisha Solomon [email protected]

Mental illness is prevalent in the U.S.: 18.3% of American adults experience some form of mental illness. Mental illness treatments are effective; however, majority of individuals with mental illness do not receive any treatment. In 2016, less than half of adults with mental illness received any treatment. Inability to pay for treatment and lack of insurance coverage for mental illness treatment are key barriers to receiving treatment. Mental illness treatment is costly for an

Historically, insurance coverage for mental healthcare has been less generous than general healthcare coverage. In an attempt to address discriminatory treatment of mental healthcare coverage, numerous U.S. states have implemented laws that compel private insurers to cover mental healthcare services at ‘parity’ with general healthcare services. Previous research has established that state mental illness parity laws improve access to mental healthcare and, in turn, reduce mental illness. Moreover, most mental illnesses develop during adolescence and early adulthood. I extend this literature in two important ways. First, I study the effect of the state mental illness parity law implementation on mental illness among college-age individuals. Second, I examine the effect of state mental illness parity laws on human capital accumulation. Considering spill-overs to these educational outcomes is important as previous research shows that mental illness impedes college performance. Hence, reduced mental illness through state parity laws could have positive spill-over effects to educational outcomes that have not yet been documented. I use differences-in-differences models to uncover the causal effects of state mental illness parity laws on mental illness and educational outcomes. I leverage plausibly exogenous variation in insurance coverage for mental healthcare using changes in state laws over the period 1998 to 2008. First, to study parity law effects on mental illness I utilize administrative data on completed suicides from National Vital Statistics System and survey data on reported mental illness from Behavioral Risk Factor System. Second, I use longitudinal data from the National Longitudinal Survey of Youth 1997 Cohort to study the effects of the mental health parity law on two important educational outcomes: drop out decisions

Three main findings emerge from my analysis. First, I document that the passage of a mental health parity law leads to reductions in state-level suicide rate for the college-aged population, and reductions in the number of poor mental health days for the student population. Second, I find no evidence that the passage of a mental health parity law influences the propensity to drop out of college. Third, I show that state-level mental health parity laws have a significant

The findings from this study can provide insights into the impact of the Affordable Care Act, which expands access to valuable mental healthcare services to millions of Americans. More broadly, these findings document important spill-


East Carolina University Complete

Columbia University Dept of Psychiatry/NYSPI Complete

Rutgers School of Public Health Rizie Kumar; Alan Monheit Complete

University of Queensland Mark Harris; Leandro Magnusson Complete

Complete

University at Albany Baris Yoruk Complete

Department of Economics, University College Dublin

Janet Cummings; Benjamin Druss; Adam Wilk Complete

Department of Economics, University of Bath Complete

Emory University Department of Health Policy and Management

University of Maryland Jie Chen Complete

University of Massachusetts-Amherst Complete

Boston University School of Public Health Ellen Moscoe Complete

University of California Merced CompletePaul Brown; Jonathan Boyajian; Kevin Kwan; Dwena Phillips; Sidra Goldman-Mellor

Temple University Complete


Obesity

Obesity

Obesity

Does Higher Education Impact Weight?: Evidence From a Conditional Cash Transfer

Which Factors and Nutritional Ingredients Influence College Students’ Snack Choices ? Evidence from Discrete Choice Experiments

Trends and Costs Associated with the Concerning Rise of Type 2 Diabetes and Comorbidities in Young People

Obesity Impact of the SSB Tax in Boulder on Prices

Obesity

Obesity

Addicted to low calorie sweeteners? A rational addiction model for sweetened beverages in Chile

The Effect of School Lunch on Early Teenagers’ Body Weight

Obesity

Obesity

Obesity

Adolescent BMI: The Importance of Intrinsic and Extrinsic Factors

The Influence of Childhood Nutrition Assistance Program Participation and Food Security on Young Adult Obesity

The Impact of the Philadelphia Beverage Tax on Prices, Household Purchases, and Child and Adult Consumption

ObesityThe effects of soda taxes on consumption and health outcomes in the short and long run

Abstract

Education is an important correlate of health and risky behaviors. However, determining the causal effect of education on health behaviors and health outcomes is difficult due to the endogeneity of education; health and education could be related through multiple pathways, including the potential effect of health on educational attainment. While many studies have examined variation in secondary schooling based on mandated school attendance for teenagers, fewer studies have examined post-secondary education. To test the causal effect of higher education on weight and general health, I exploit a shock in a cash transfer that was conditional on enrollment in college as an instrument for education. First I confirm this cash transfer is a strong and valid instrument for education; the instrument has meaningful effects on educational attainment and appears to be only related to health through education. Second, I find this change in education has important effects on health, specifically weight.

The instrument is based on Social Security benefits for children. Minor children of retired or disabled Social Security beneficiaries as well as children with deceased parents are eligible for their own Social Security benefit. Between the mid-1960s and the early-1980s, college-aged child recipients could continue to receive this benefit conditional on college enrollment. Dynarski (2003) showed that the termination of these benefits to college students caused a sharp decline in college-going among eligible young adults as compared to ineligible young adults. First, I confirm Dynarski’s finding that this program had a large and meaningful effect on education by utilizing (1) a larger dataset (National Health Interview Survey), (2) a more accurate measure of eligibility based on administrative records, and (3) a similar difference-in-differences strategy. I then use the exogenous variation in educational attainment caused by changes in Social Security benefits to determine the causal effect of education on health.

In preliminary analyses, I find that financial aid increased educational attainment for beneficiaries, including increasing the likelihood of attending any college and graduating from college. Interestingly, this effect is concentrated in women, with little to no effect on men. This heterogeneous effect by gender on education allows me to use men in a falsification test for the effect of education on health. Consistent with the first stage, I find meaningful benefits on women’s body mass index and general self-described health, but I find no effects for men.

No recent study has examined whether college students consider the healthiness of a snack compared to other snack attributes, such as taste, convenience, or price. Furthermore, there is limited research on the relative importance of which snack nutritional ingredients influence college students’ snack choices. This study examined college students’ awareness of the healthiness of a snack relative to other snack attributes, and considered the ranking of nutritional ingredients that influence their snack choices to ultimately investigate whether nutrition education could foster healthier snack selections. To examine this, two Discrete Choice Experiments (DCEs) were constructed using a unique approach of Block Fractional Factorial Designs. The first DCE examined four three-level snack attributes (healthiness, taste, convenience and price). The second DCE measured six two-level nutritional ingredients (sugar, salt, calories, fat, “all natural,” and fiber). Nested logit models were used to analyze the DCEs and relative importance scores were calculated. A total of 1,624 undergraduate students from a large diverse public institution from the Orange County area in Southern California completed the DCEs. We found healthiness (55%) and sugar (24%) had the highest relative importance score of snack attributes and nutritional ingredients, respectively. College students tended to prefer quick and cheap snacks, but higher prices on healthy snacks did not affect their decision significantly. In addition, high-sugar snacks were less favored if college students had nutrition education. Interventions to increasing the availability of healthier snacks in vending machines on college campuses and promoting nutrition education could improve students’ snack choices, thus reducing obesity prevalence in college students.

Background: For more than two decades, an epidemic of obesity in the United States has been linked to rising rates of type 2 diabetes. Both conditions have been increasing among the nation’s youth. The increase in type 2 diabetes among young people portends many decades of treatment of the disease and its complications for a larger population than previously known, with all the attendant cost in healthcare resources. This study will analyze the cost to our healthcare system of the rise of type 2 diabetes in today’s youth.

Methods: We analyzed data from our FAIR Health database of billions of privately billed healthcare claims to identify trends and patterns from 2011 to 2015 in obesity, type 2 diabetes and other obesity-related conditions in the nation’s privately insured pediatric population (0-22 years of age). We will perform further data analytics to study claims through 2017 and their associated healthcare costs. Results: As obesity rates have increased in children, adolescents and young adults, our data show that type 2 diabetes also has increased in that population. The percent of claim lines with an obesity diagnosis increased annually in all age groups, from infants and toddlers to adults. In the age group 19 to 22 years, the increase in obesity claim lines was 154 percent. During the same period, the percent of claim lines with a type 2 diabetes diagnosis more than doubled in the pediatric population, increasing 109 percent. In most pediatric age groups, claim lines with an obesity diagnosis were more common for females than males, but claim lines with a type 2 diabetes diagnosis were more common for males than females. Claim lines for two obesity-related conditions, obstructive sleep apnea and hypertension, rose in the pediatric population by 161 percent and 67 percent, respectively. Using both charge and allowed amount data, we will compare the healthcare costs of pediatric patients diagnosed with type 2 diabetes to those of all pediatric patients. As an indicator of the future healthcare costs of this young generation diagnosed with type 2 diabetes as it matures into adulthood, we also will compare the healthcare costs of adult patients with type 2 diabetes to those of all adult patients. Conclusions: Type 2 diabetes associated with obesity in young people has been growing significantly; this study will document its economic impact on the healthcare system. That impact in the form of increasing healthcare costs will likely grow as this generation matures. By understanding the scope of this problem, payors, providers, government officials and policy makers can be better prepared to address and mitigate this public health issue of growing concern.

In an attempt to reduce the consumption of sugar-sweetened beverages (SSBs), stem the rise in obesity, and generate government revenue, taxes on SSBs have become increasingly common across the country. Boulder, Colorado, implemented the largest SSB tax to date on July 1, 2017, increasing the price of SSBs by two cents per ounce. To understand how the tax influcenes consumers, this paper uses multiple data sources to estimate the impact of the SSB tax in Boulder on prices of SSBs and prices of non-SSBs. The first data source is comprised of hand collected prices of beverage items from retailers and restaurants in April, June, August, and October of 2017. The second source is a collection of online restaurant menus from Grubhub.com and Orderup.com, which were collected weekly from March to October of 2017. The final data source is comprised of beverage prices from retailers listed on Instacart.com and Walmart.com, which were collected weekly from November, 2016 to October, 2017. Price data from each source contains observations both within Boulder and directly outside of Boulder. We compare the changes in the price per ounce of SSBs and non-SSBs over time in Boulder compared to the changes during the same period in Boulder County (outside the city) and Fort Collins. The multiple periods of data before implementation allow us to determine if Boulder County and Fort Collins are appropriate comparison groups to Boulder in that the parallel trends assumption is satisfied. Analysis demonstrates that the trends in prices over time prior to July 1 are similar in Boulder and in the rest of Boulder County and Fort Collins. Preliminary estimates indicate that, on average, 50 percent of the SSB tax was passed on to consumers. These pass through results are heterogeneous across firm type, beverage type, and beverage size, varying in magnitude between 30 to 80 percent pass-through. We also find that the price of non-SSB soft drinks significantly increase in Boulder restaurants after the implementation of the tax. Further, with the data dating back to the beginning of November, 2016, prior to the election when voters voted in favor of the tax, we are able to determine if there is a change in price in response to the election as well as the implementation.

Neuroscience studies and observational analyses in the context of a modern average diet, suggest that low-calorie sweeteners (LCS) are linked with overweight and metabolic changes which enhance sweet preference (Murray, Tulloch, Criscitelli, & Avena, 2016; Swithers, 2013; Sylvetsky et al., 2017; Yang, 2010). Random controlled trials (RCT) and studies which match on dietary patterns suggest the opposite—no effect on food preferences or cardiometabolic outcomes (de Koning et al., 2012; Duffey, Gordon-Larsen, Steffen, Jacobs, & Popkin, 2010; Piernas, Tate, Wang, & Popkin, 2013; Tate et al., 2012). As consequence, systematic reviews reflect lack of consensus on the issue of addiction (Rogers et al., 2016) and continue to promote consumption of LCS beverages as a strategy for weight management (U.S. DHHS and USDA, 2015). Overall, understanding LCS addiction relative to sugar is key from a public health perspective. If all sweetened beverages (SB) are similarly addictive, all of them should be regulated to reduce sweetness dependence, and thus, overweight and obesity prevalence. This study provides evidence on the addiction to LCS and sugar in sweetened beverages in Chile. In particular, we will use a random-coefficient discrete choice model, extending from Richards, Patterson, and Tegene (2007) to account for the addiction to products’ nutrients under the rational addiction framework, originally introduced by Becker and Murphy (1988). We test the relative addiction to LCS versus sugar in SB, allowing for several types of beverages and household characteristics. Data source comes from Kantar WorldPanel, which include weekly purchases of packaged foods and beverages from a representative sample of Chilean households (≈2,000) from the 2013-16. Purchase data is connected to nutrition facts panel data collected in several retail venues between 2015 and 2016, to calculate household purchases of specific nutrients, namely sugar and LCS. Chile is ideal as a case study for a rational addiction model of sweetened beverages. First, in contrast to the remarkable success reducing tobacco and risky alcohol consumption, obesity has increased sharply in the last decade, while diet and physical activity does not seem to improve (Ministerio de Salud, 2017). In 2017, three of every four adults are overweight, while obesity prevalence peaked at 34.4%. Additionally, Chile reports the largest consumption of sweetened beverages (Popkin & Hawkes, 2015), with a particularly large market participation of LCS beverages (Caro, Taillie, Ng, & Popkin, 2017). Finally, the last modification to the Chilean SB tax reduced the relative price of LCS beverages compared to sugar-sweetened beverages, potentially inducing larger addictive behavior. Results of this study are critical for current food policy research and practice. First, we can address to which magnitude LCS are addictive in relation to sugar-sweetened beverages at the population level. Secondly, we can observe heterogeneity in addictive behavior based on household characteristics. Last, we can observe how tax policies affect households’ behavior in the context of rational addiction. Presence of LCS beverage addiction suggests that incentives to decrease all SB purchases is key to reduce the obesity epidemic.

Abstract We examine causal effects of Japanese school lunch program on 13- to 15-year-old children’s weight. We use individual level data drawn from the 1975-1994 National Nutrition Survey, a nationally-representative household survey with measured height and weight, and examine the body mass index (BMI), obesity status, and underweight status as outcome variables. Japanese compulsory education consists of six years of elementary school for 6- to 12-year-olds and three years of junior high school for 12- to 15-year-olds, where the majority attend municipal schools. Each municipality decides whether to provide school lunch at municipal elementary and junior high schools, and at a school where school lunch is provided, all students must eat school lunch in principle. While almost all municipalities provide school lunch for elementary schools, there is a large variation in municipal school lunch provision for junior high schools. We use a modified DID framework and compare differences between junior high school students and older (9- to 12-year-old) elementary school students between municipalities with and without school lunch for junior high schools, controlling for district-specific effects to account for unobservable heterogeneity in area characteristics. We conduct subsample analysis by socioeconomic status (SES) to explore treatment heterogeneity, which has not been fully considered in the literature. School lunch alters children’s food intake at lunch, which might increase or decrease their weight, depending on the relative contents of school lunch and counterfactual lunch they would have eaten in the absence of school lunch. This implies possible treatment heterogeneity. Heterogeneous treatment effect might also account for the large variation in previous estimates of the effect of school meal on weight. Studying Japan offers several merits. First, and most importantly, the lack of individual choice in school lunch participation at municipal schools allows little room for self-selection. This enables us to estimate population effect of actual participation in school lunch program and assess treatment heterogeneity by children’s socioeconomic background, both of which are infeasible in previous quasi-experimental studies. Second, strict nutritional standards embedded in Japanese school lunch program minimize program heterogeneity, and also foster understanding the effects of recent introduction of stricter than conventional nutritional requirements in UK and US. Third, there are no other large-scale food provision programs, unlike in US with various food assistance programs. Fourth, the absence of means testing minimizes concerns for stigma-induced reporting bias in school lunch participation. We find no evidence that school lunch affects body weight in full sample analysis. In subsamples of children with low socioeconomic background, however, we find significant negative effect of school lunch on BMI and obesity. We find no evidence that school lunch affects underweight prevalence. These findings are robust to propensity score trimming. Additionally, municipal school lunch provision for junior high schools is not significantly associated with growth or weight problems of younger children, inconsistent with endogeneity concerns. We find some evidence that the weight reduction effect for low SES children persists several years after they graduate from junior high school and stop having school lunch.

Objectives: Research shows that overweight adolescents become overweight adults, but do the same factors contribute to weight in adolescence as adulthood? Are extrinsic factors more important than intrinsic characteristics? This study identifies the correlation between BMI and various intrinsic and extrinsic covariates and evaluates their relative importance in BMI determination. Furthermore, it separates the sample into adolescents (12-20) and adults (21+) and compares the primary determinants of BMI between the two groups. Methods: Using 15 years of panel data, multi-level change models including both fixed and random effects assess the impact of extrinsic—environmental, biological, geographic and household—and intrinsic—sexual activity, substance use, desire to lose weight, etc.—characteristics on BMI. Separate adolescent and adult samples test for differences in relative BMI determinants. Results: Results suggest that race and age are the most significant determinants of BMI at all ages. However, other determinants differ between adolescents are adults. In the early years of adolescents, intrinsic factors are highly deterministic, while extrinsic factors are insignificant. Intrinsic determinates of significance include age of first sexual encounter, tobacco experimentation, perspective on general health and desire to lose/stay the same weight play a significant role for adolescents. As youth age into adults, intrinsic factors decrease in importance, while extrinsic covariates become more deterministic. Conclusion: While biological/genetic attributes, are the largest determinants of BMI at every age, intrinsic factors appear to be more significant than extrinsic characteristics during adolescents. As individuals age, intrinsic determinants decrease in importance as extrinsic characteristics increase in significance. Thus, the weight determinants differ between adolescents and adults suggesting different methods of policy intervention be used to improve the health of both groups.

Purpose: In the United States, 13% of the population is food insecure, and one in seven Americans receive food assistance from the Supplemental Nutrition Assistance Program (SNAP). Half of all children born in the US every year receive supplemental nutrition support from the Special Supplemental Nutrition Program for Women, Infants and Children (WIC). SNAP and WIC improve short-term food security, but their effect on long-term food security is unknown. Nutritional knowledge is an important predictor of diet quality and healthy weight, however less is known about the relationship between parental nutritional knowledge, food security, and the risk of obesity in children as they transition to adulthood. The goal of this study is to examine the relationship between nutrition assistance program participation and parental nutritional knowledge during childhood with obesity and food insecurity risk in young adulthood. Methods: Data for this study come from the 1997-2015 waves of the Panel Study of Income Dynamics (PSID) including the Child Development Supplement (CDS), and Transition into Adulthood Supplement (TAS). Begun in 1968, the PSID is a nationally representative, multigenerational household panel survey. The study sample includes a balanced panel of 2,796 families whose children were age 0-12 years in 1997 during the first wave of the CDS and age 17-31 in 2015. We created a ‘parental nutritional knowledge’ index based on five nutrition knowledge questions. To examine the relationship between SNAP and WIC receipt in childhood (in 1996-2000) and parental nutritional knowledge (in 1999) with weight status and food security outcomes in 2015, we estimated multinomial and logistic multivariable, fixed-effect regression models controlling for individual- and family-level covariates. Results: Preliminary results show that individuals who are food insecure during childhood have a higher risk of obesity in young adulthood (26% vs. 21%), while those who are persistently food insecure have an even larger disparity compared to those who are never food insecure (33% vs. 18%). In adjusted models, among low-income individuals with persistent food insecurity, food stamps participation during childhood reduces risk of obesity from 36% to 34%. Among low-income, food insecure children, SNAP participation increased the chances of becoming food secure in young adulthood from 50% to 54%. Parental nutritional knowledge analyses will also be presented. Conclusions: Food insecurity during childhood has long-term implications for health, however, participation in nutritional assistance programs may mitigate that relationship and lower the risk of obesity in young adulthood. This material was supported by a grant from the University of Kentucky Poverty Research Center.

Several cities in the U.S. have implemented taxes on sugar-sweetened beverages in an attempt to improve public health as well as raise revenue. On January 1, 2017, Philadelphia became the largest U.S. city to implement a beverage tax, which is 1.5 cents per ounce and applies to both caloric and non-caloric but artificially sweetened beverages. In this paper, we estimate the impact of the tax on retail prices, marketing, purchases, and child and adult consumption of taxed and potential substitute beverages. The study includes an innovative nested data collection approach in which we collected: (1) prices and marketing from retailers, (2) purchase information from customers exiting the retailers, and (3) a follow-up household survey of those customers regarding household and child beverage purchases and consumption. We collected the information from a representative group of retailers and their consumers in Philadelphia and a matched group of retailers in surrounding counties that serve as a control group. We collected the information in the months prior to the implementation of the tax and again a year later. In addition, we examine detailed receipt data on beverage purchases in the six months prior to and after implementation of the tax for a sample of households in Philadelphia as well as for two control groups: residents in surrounding counties and a national matched comparison group. We use a difference-in-differences identification strategy to estimate the impact of the tax on prices, marketing, purchases, and consumption of taxed beverages. We also estimate the impacts on untaxed beverages to assess potential substitution by consumers. In addition, we examine differential impacts by proximity to the city border to determine whether spillover effects might dampen the impact on prices close to the border and whether residents increase travel across city borders to purchase beverages after the tax. Furthermore, we examine whether taxes have different impacts on prices and marketing by store type, brand, and beverage container volume, and whether the impacts on purchases and consumption vary by household income, age of the child, race, and ethnicity. Taken together, the findings illuminate how the tax in Philadelphia affected prices and how the changes in prices affected purchases and consumption.

Sugar-sweetened beverages (SSBs) have little nutritive value and add a nontrivial number of calories to diets. Frequent SSB consumption is associated with the incidence of obesity, diabetes, and other health conditions related to the cardiovascular system. Because excessive SSB consumption has a negative health impact, the purchase price of SSBs may not capture the full cost to society. To help internalize this cost, most states apply sales taxes to SSBs and some U.S jurisdictions also levy SSB excise taxes. Numerous studies have examined the effect of SSB taxes on consumption and obesity and have arrived at mixed conclusions. However, most prior studies have estimated the contemporaneous impact of SSB taxes. A focus on the impact of contemporaneous SSB taxes may be insufficient to capture the entire long-run effect since stages of the relationship—the effect of SSB taxes on SSB consumption and the effect of any resulting change in SSB consumption on health—could occur gradually over time. There are at least two reasons why there may be a lag in the (full) consumption response to a SSB cost change. First, it may take time for some consumers to notice a change in SSB costs or to fully appreciate a cost change relative to their food budgets. Second, consumers may be slow to adjust their intake of SSBs because of the habit-forming nature of SSB consumption. In addition, the short- and long-run impact of SSB taxes may differ because tax shifting from suppliers to consumers may vary over time. Using 1999-2014 restricted-use National Health and Nutrition Examination Survey (NHANES) data at the state level, we contribute to the literature by analyzing the short- and long-run effect of SSB taxes on caloric intake, body weight outcomes, blood sugar levels, blood cholesterol levels, blood pressure, as well as a metabolic syndrome index. A noteworthy outcome of interest is blood sugar levels since diabetes as an outcome has been studied relatively little in the literature despite it often being cited as a main reason for implementing a SSB tax. Our study will provide an up-to-date analysis of the health impacts of SSB taxes among children, adolescents, and adults, and will shed light on whether and how the sign and size of any health effects of SSB taxes depends on the time horizon of the analysis. The results of our study will help to inform future tax policy-making decisions by showing the importance of the long-run benefits of SSB taxation.

Abstract

Education is an important correlate of health and risky behaviors. However, determining the causal effect of education on health behaviors and health outcomes is difficult due to the endogeneity of education; health and education could be related through multiple pathways, including the potential effect of health on educational attainment. While many studies have examined variation in secondary schooling based on mandated school attendance for teenagers, fewer studies have examined post-secondary education. To test the causal effect of higher education on weight and general health, I exploit a shock in a cash transfer that was conditional on enrollment in college as an instrument for education. First I confirm this cash transfer is a strong and valid instrument for education; the instrument has meaningful effects on educational attainment and appears to be only related to health through education. Second, I find this change in

The instrument is based on Social Security benefits for children. Minor children of retired or disabled Social Security beneficiaries as well as children with deceased parents are eligible for their own Social Security benefit. Between the mid-1960s and the early-1980s, college-aged child recipients could continue to receive this benefit conditional on college enrollment. Dynarski (2003) showed that the termination of these benefits to college students caused a sharp decline in college-going among eligible young adults as compared to ineligible young adults. First, I confirm Dynarski’s finding that this program had a large and meaningful effect on education by utilizing (1) a larger dataset (National Health Interview Survey), (2) a more accurate measure of eligibility based on administrative records, and (3) a similar difference-in-differences strategy. I then use the exogenous variation in educational attainment caused by changes in Social Security

In preliminary analyses, I find that financial aid increased educational attainment for beneficiaries, including increasing the likelihood of attending any college and graduating from college. Interestingly, this effect is concentrated in women, with little to no effect on men. This heterogeneous effect by gender on education allows me to use men in a falsification test for the effect of education on health. Consistent with the first stage, I find meaningful benefits on women’s body

No recent study has examined whether college students consider the healthiness of a snack compared to other snack attributes, such as taste, convenience, or price. Furthermore, there is limited research on the relative importance of which snack nutritional ingredients influence college students’ snack choices. This study examined college students’ awareness of the healthiness of a snack relative to other snack attributes, and considered the ranking of nutritional ingredients that influence their snack choices to ultimately investigate whether nutrition education could foster healthier snack selections. To examine this, two Discrete Choice Experiments (DCEs) were constructed using a unique approach of Block Fractional Factorial Designs. The first DCE examined four three-level snack attributes (healthiness, taste, convenience and price). The second DCE measured six two-level nutritional ingredients (sugar, salt, calories, fat, “all natural,” and fiber). Nested logit models were used to analyze the DCEs and relative importance scores were calculated. A total of 1,624 undergraduate students from a large diverse public institution from the Orange County area in Southern California completed the DCEs. We found healthiness (55%) and sugar (24%) had the highest relative importance score of snack attributes and nutritional ingredients, respectively. College students tended to prefer quick and cheap snacks, but higher prices on healthy snacks did not affect their decision significantly. In addition, high-sugar snacks were less favored if college students had nutrition education. Interventions to increasing the availability of healthier snacks in vending machines on college campuses and promoting nutrition education could improve students’ snack choices, thus reducing obesity prevalence in college students.

For more than two decades, an epidemic of obesity in the United States has been linked to rising rates of type 2 diabetes. Both conditions have been increasing among the nation’s youth. The increase in type 2 diabetes among young people portends many decades of treatment of the disease and its complications for a larger population than previously known, with all the attendant cost in healthcare resources. This study will analyze the cost to our

We analyzed data from our FAIR Health database of billions of privately billed healthcare claims to identify trends and patterns from 2011 to 2015 in obesity, type 2 diabetes and other obesity-related conditions in the nation’s privately insured pediatric population (0-22 years of age). We will perform further data analytics to study claims through 2017 and their associated healthcare costs.

As obesity rates have increased in children, adolescents and young adults, our data show that type 2 diabetes also has increased in that population. The percent of claim lines with an obesity diagnosis increased annually in all age groups, from infants and toddlers to adults. In the age group 19 to 22 years, the increase in obesity claim lines was 154 percent. During the same period, the percent of claim lines with a type 2 diabetes diagnosis more than doubled in the pediatric population, increasing 109 percent. In most pediatric age groups, claim lines with an obesity diagnosis were more common for females than males, but claim lines with a type 2 diabetes diagnosis were more common for males than females. Claim lines for two obesity-related conditions, obstructive sleep apnea and hypertension, rose in the pediatric population by 161 percent and 67 percent, respectively. Using both charge and allowed amount data, we will compare the healthcare costs of pediatric patients diagnosed with type 2 diabetes to those of all pediatric patients. As an indicator of the future healthcare costs of this young generation diagnosed with type 2 diabetes as it matures into adulthood, we also will compare the healthcare costs of adult patients with type 2 diabetes to those of all adult patients.

Type 2 diabetes associated with obesity in young people has been growing significantly; this study will document its economic impact on the healthcare system. That impact in the form of increasing healthcare costs will likely grow as this generation matures. By understanding the scope of this problem, payors, providers, government officials and policy makers can be better prepared to address and mitigate this public health issue of growing concern.

In an attempt to reduce the consumption of sugar-sweetened beverages (SSBs), stem the rise in obesity, and generate government revenue, taxes on SSBs have become increasingly common across the country. Boulder, Colorado, implemented the largest SSB tax to date on July 1, 2017, increasing the price of SSBs by two cents per ounce. To understand how the tax influcenes consumers, this paper uses multiple data sources to estimate the impact of the SSB tax in

The first data source is comprised of hand collected prices of beverage items from retailers and restaurants in April, June, August, and October of 2017. The second source is a collection of online restaurant menus from Grubhub.com and Orderup.com, which were collected weekly from March to October of 2017. The final data source is comprised of beverage prices from retailers listed on Instacart.com and Walmart.com, which were collected weekly from November, 2016 to October, 2017. Price data from each source contains observations both within Boulder and directly outside of Boulder. We compare the changes in the price per ounce of SSBs and non-SSBs over time in Boulder compared to the changes during the same period in Boulder County (outside the city) and Fort Collins. The multiple periods of data before implementation allow us to determine if Boulder County and Fort Collins are appropriate comparison groups to Boulder in that the parallel trends assumption is satisfied. Analysis demonstrates that the trends in prices over time prior to July 1 are similar in Boulder and in the rest of Boulder County and Fort Collins. Preliminary estimates indicate that, on average, 50 percent of the SSB tax was passed on to consumers. These pass through results are heterogeneous across firm type, beverage type, and beverage size, varying in magnitude between 30 to 80 percent pass-through. We also find that the price of non-SSB soft drinks significantly increase in Boulder restaurants after the implementation of the tax. Further, with the data dating back to the beginning of November, 2016, prior to the election when voters voted in favor of the tax, we are able to determine if there is a change in price in response to the election as well as the implementation.

Neuroscience studies and observational analyses in the context of a modern average diet, suggest that low-calorie sweeteners (LCS) are linked with overweight and metabolic changes which enhance sweet preference (Murray, Tulloch, Criscitelli, & Avena, 2016; Swithers, 2013; Sylvetsky et al., 2017; Yang, 2010). Random controlled trials (RCT) and studies which match on dietary patterns suggest the opposite—no effect on food preferences or cardiometabolic outcomes (de Koning et al., 2012; Duffey, Gordon-Larsen, Steffen, Jacobs, & Popkin, 2010; Piernas, Tate, Wang, & Popkin, 2013; Tate et al., 2012). As consequence, systematic reviews reflect lack of consensus on the issue of addiction (Rogers et al., 2016) and continue to promote consumption of LCS beverages as a strategy for weight management (U.S. DHHS and USDA, 2015). Overall, understanding LCS addiction relative to sugar is key from a public health perspective. If all sweetened beverages (SB) are similarly addictive, all of them should be regulated to reduce sweetness dependence, and thus, overweight and obesity prevalence. This study provides evidence on the addiction to LCS and sugar in sweetened beverages in Chile. In particular, we will use a random-coefficient discrete choice model, extending from Richards, Patterson, and Tegene (2007) to account for the addiction to products’ nutrients under the rational addiction framework, originally introduced by Becker and Murphy (1988). We test the relative addiction to LCS versus sugar in SB, allowing for several types of beverages and household characteristics. Data source comes from Kantar WorldPanel, which include weekly purchases of packaged foods and beverages from a representative sample of Chilean households (≈2,000) from the 2013-16. Purchase data is connected to nutrition facts panel data collected in several retail venues between 2015 and 2016, to calculate household purchases of specific nutrients, namely sugar and LCS. Chile is ideal as a case study for a rational addiction model of sweetened beverages. First, in contrast to the remarkable success reducing tobacco and risky alcohol consumption, obesity has increased sharply in the last decade, while diet and physical activity does not seem to improve (Ministerio de Salud, 2017). In 2017, three of every four adults are overweight, while obesity prevalence peaked at 34.4%. Additionally, Chile reports the largest consumption of sweetened beverages (Popkin & Hawkes, 2015), with a particularly large market participation of LCS beverages (Caro, Taillie, Ng, & Popkin, 2017). Finally, the last modification to the Chilean SB tax reduced the relative price of LCS beverages compared

Results of this study are critical for current food policy research and practice. First, we can address to which magnitude LCS are addictive in relation to sugar-sweetened beverages at the population level. Secondly, we can observe heterogeneity in addictive behavior based on household characteristics. Last, we can observe how tax policies affect households’ behavior in the context of rational addiction. Presence of LCS beverage addiction suggests that incentives to

We examine causal effects of Japanese school lunch program on 13- to 15-year-old children’s weight. We use individual level data drawn from the 1975-1994 National Nutrition Survey, a nationally-representative household survey with measured height and weight, and examine the body mass index (BMI), obesity status, and underweight status as outcome variables. Japanese compulsory education consists of six years of elementary school for 6- to 12-year-olds and three years of junior high school for 12- to 15-year-olds, where the majority attend municipal schools. Each municipality decides whether to provide school lunch at municipal elementary and junior high schools, and at a school where school lunch is provided, all students must eat school lunch in principle. While almost all municipalities provide school lunch for elementary schools, there is a large variation in municipal school lunch provision for junior high schools. We use a modified DID framework and compare differences between junior high school students and older (9- to 12-year-old) elementary school students between municipalities with and without school lunch for junior high schools, controlling for district-specific effects to account for unobservable heterogeneity in area characteristics. We conduct subsample analysis by socioeconomic status (SES) to explore treatment heterogeneity, which has not been fully considered in the literature. School lunch alters children’s food intake at lunch, which might increase or decrease their weight, depending on the relative contents of school lunch and counterfactual lunch they would have eaten in the absence of school lunch. This implies possible treatment heterogeneity. Heterogeneous treatment effect might also

Studying Japan offers several merits. First, and most importantly, the lack of individual choice in school lunch participation at municipal schools allows little room for self-selection. This enables us to estimate population effect of actual participation in school lunch program and assess treatment heterogeneity by children’s socioeconomic background, both of which are infeasible in previous quasi-experimental studies. Second, strict nutritional standards embedded in Japanese school lunch program minimize program heterogeneity, and also foster understanding the effects of recent introduction of stricter than conventional nutritional requirements in UK and US. Third, there are no other large-scale food provision programs, unlike in US with various food assistance programs. Fourth, the absence of means testing minimizes concerns for stigma-induced reporting bias in school lunch participation. We find no evidence that school lunch affects body weight in full sample analysis. In subsamples of children with low socioeconomic background, however, we find significant negative effect of school lunch on BMI and obesity. We find no evidence that school lunch affects underweight prevalence. These findings are robust to propensity score trimming. Additionally, municipal school lunch provision for junior high schools is not significantly associated with growth or weight problems of younger children, inconsistent with endogeneity concerns. We find some evidence that the weight reduction effect for low SES children persists several years after they graduate from junior high school and stop having school

: Research shows that overweight adolescents become overweight adults, but do the same factors contribute to weight in adolescence as adulthood? Are extrinsic factors more important than intrinsic characteristics? This study identifies the correlation between BMI and various intrinsic and extrinsic covariates and evaluates their relative importance in BMI determination. Furthermore, it separates the sample into adolescents (12-20) and adults (21+) and

: Using 15 years of panel data, multi-level change models including both fixed and random effects assess the impact of extrinsic—environmental, biological, geographic and household—and intrinsic—sexual activity, substance use, desire to lose weight, etc.—characteristics on BMI. Separate adolescent and adult samples test for differences in relative BMI determinants.

: Results suggest that race and age are the most significant determinants of BMI at all ages. However, other determinants differ between adolescents are adults. In the early years of adolescents, intrinsic factors are highly deterministic, while extrinsic factors are insignificant. Intrinsic determinates of significance include age of first sexual encounter, tobacco experimentation, perspective on general health and desire to lose/stay the same weight play a significant role for adolescents. As youth age into adults, intrinsic factors decrease in importance, while extrinsic covariates become more deterministic.

: While biological/genetic attributes, are the largest determinants of BMI at every age, intrinsic factors appear to be more significant than extrinsic characteristics during adolescents. As individuals age, intrinsic determinants decrease in importance as extrinsic characteristics increase in significance. Thus, the weight determinants differ between adolescents and adults suggesting different methods of policy intervention be used to improve the health of both

In the United States, 13% of the population is food insecure, and one in seven Americans receive food assistance from the Supplemental Nutrition Assistance Program (SNAP). Half of all children born in the US every year receive supplemental nutrition support from the Special Supplemental Nutrition Program for Women, Infants and Children (WIC). SNAP and WIC improve short-term food security, but their effect on long-term food security is unknown. Nutritional knowledge is an important predictor of diet quality and healthy weight, however less is known about the relationship between parental nutritional knowledge, food security, and the risk of obesity in children as they transition to adulthood. The goal of this study is to examine the relationship between nutrition assistance program participation and parental nutritional knowledge during childhood with obesity and food insecurity risk in young adulthood.

Data for this study come from the 1997-2015 waves of the Panel Study of Income Dynamics (PSID) including the Child Development Supplement (CDS), and Transition into Adulthood Supplement (TAS). Begun in 1968, the PSID is a nationally representative, multigenerational household panel survey. The study sample includes a balanced panel of 2,796 families whose children were age 0-12 years in 1997 during the first wave of the CDS and age 17-31 in 2015. We created a ‘parental nutritional knowledge’ index based on five nutrition knowledge questions. To examine the relationship between SNAP and WIC receipt in childhood (in 1996-2000) and parental nutritional knowledge (in 1999) with weight status and food security outcomes in 2015, we estimated multinomial and logistic multivariable, fixed-effect regression models controlling for individual- and family-level covariates.

Preliminary results show that individuals who are food insecure during childhood have a higher risk of obesity in young adulthood (26% vs. 21%), while those who are persistently food insecure have an even larger disparity compared to those who are never food insecure (33% vs. 18%). In adjusted models, among low-income individuals with persistent food insecurity, food stamps participation during childhood reduces risk of obesity from 36% to 34%. Among low-income, food insecure children, SNAP participation increased the chances of becoming food secure in young adulthood from 50% to 54%. Parental nutritional knowledge analyses will also be presented.

Food insecurity during childhood has long-term implications for health, however, participation in nutritional assistance programs may mitigate that relationship and lower the risk of obesity in young adulthood.

Several cities in the U.S. have implemented taxes on sugar-sweetened beverages in an attempt to improve public health as well as raise revenue. On January 1, 2017, Philadelphia became the largest U.S. city to implement a beverage tax, which is 1.5 cents per ounce and applies to both caloric and non-caloric but artificially sweetened beverages. In this paper, we estimate the impact of the tax on retail prices, marketing, purchases, and child and adult consumption of taxed

The study includes an innovative nested data collection approach in which we collected: (1) prices and marketing from retailers, (2) purchase information from customers exiting the retailers, and (3) a follow-up household survey of those customers regarding household and child beverage purchases and consumption. We collected the information from a representative group of retailers and their consumers in Philadelphia and a matched group of retailers in surrounding counties that serve as a control group. We collected the information in the months prior to the implementation of the tax and again a year later. In addition, we examine detailed receipt data on beverage purchases in the six months prior to and after implementation of the tax for a sample of households in Philadelphia as well as for two control groups: residents in surrounding counties and a national matched comparison group. We use a difference-in-differences identification strategy to estimate the impact of the tax on prices, marketing, purchases, and consumption of taxed beverages. We also estimate the impacts on untaxed beverages to assess potential substitution by consumers. In addition, we examine differential impacts by proximity to the city border to determine whether spillover effects might dampen the impact on prices close to the border and whether residents increase travel across city borders to purchase beverages after the tax. Furthermore, we examine whether taxes have different impacts on prices and marketing by store type, brand, and beverage container volume, and whether the impacts on purchases and consumption vary by household income, age of the child, race, and ethnicity. Taken together, the findings illuminate how the tax in Philadelphia affected prices and how the changes in prices affected purchases and consumption.

Sugar-sweetened beverages (SSBs) have little nutritive value and add a nontrivial number of calories to diets. Frequent SSB consumption is associated with the incidence of obesity, diabetes, and other health conditions related to the cardiovascular system. Because excessive SSB consumption has a negative health impact, the purchase price of SSBs may not capture the full cost to society. To help internalize this cost, most states apply sales taxes to SSBs and some U.S

Numerous studies have examined the effect of SSB taxes on consumption and obesity and have arrived at mixed conclusions. However, most prior studies have estimated the contemporaneous impact of SSB taxes. A focus on the impact of contemporaneous SSB taxes may be insufficient to capture the entire long-run effect since stages of the relationship—the effect of SSB taxes on SSB consumption and the effect of any resulting change in SSB consumption on health—could occur gradually over time. There are at least two reasons why there may be a lag in the (full) consumption response to a SSB cost change. First, it may take time for some consumers to notice a change in SSB costs or to fully appreciate a cost change relative to their food budgets. Second, consumers may be slow to adjust their intake of SSBs because of the habit-forming nature of SSB consumption. In addition, the short- and long-run impact of SSB taxes may differ because

Using 1999-2014 restricted-use National Health and Nutrition Examination Survey (NHANES) data at the state level, we contribute to the literature by analyzing the short- and long-run effect of SSB taxes on caloric intake, body weight outcomes, blood sugar levels, blood cholesterol levels, blood pressure, as well as a metabolic syndrome index. A noteworthy outcome of interest is blood sugar levels since diabetes as an outcome has been studied relatively little in the literature despite it often being cited as a main reason for implementing a SSB tax. Our study will provide an up-to-date analysis of the health impacts of SSB taxes among children, adolescents, and adults, and will shed light on whether and how the sign and size of any health effects of SSB taxes depends on the time horizon of the analysis. The results of our study will help to inform future tax policy-making decisions by showing the importance of the long-run benefits of


Barton Willage [email protected]

Pimbucha Rusmevichientong [email protected]

Robin Gelburd [email protected]

Education is an important correlate of health and risky behaviors. However, determining the causal effect of education on health behaviors and health outcomes is difficult due to the endogeneity of education; health and education could be related through multiple pathways, including the potential effect of health on educational attainment. While many studies have examined variation in secondary schooling based on mandated school attendance for teenagers, fewer studies have examined post-secondary education. To test the causal effect of higher education on weight and general health, I exploit a shock in a cash transfer that was conditional on enrollment in college as an instrument for education. First I confirm this cash transfer is a strong and valid instrument for education; the instrument has meaningful effects on educational attainment and appears to be only related to health through education. Second, I find this change in

The instrument is based on Social Security benefits for children. Minor children of retired or disabled Social Security beneficiaries as well as children with deceased parents are eligible for their own Social Security benefit. Between the mid-1960s and the early-1980s, college-aged child recipients could continue to receive this benefit conditional on college enrollment. Dynarski (2003) showed that the termination of these benefits to college students caused a sharp decline in college-going among eligible young adults as compared to ineligible young adults. First, I confirm Dynarski’s finding that this program had a large and meaningful effect on education by utilizing (1) a larger dataset (National Health Interview Survey), (2) a more accurate measure of eligibility based on administrative records, and (3) a similar difference-in-differences strategy. I then use the exogenous variation in educational attainment caused by changes in Social Security

In preliminary analyses, I find that financial aid increased educational attainment for beneficiaries, including increasing the likelihood of attending any college and graduating from college. Interestingly, this effect is concentrated in women, with little to no effect on men. This heterogeneous effect by gender on education allows me to use men in a falsification test for the effect of education on health. Consistent with the first stage, I find meaningful benefits on women’s body

No recent study has examined whether college students consider the healthiness of a snack compared to other snack attributes, such as taste, convenience, or price. Furthermore, there is limited research on the relative importance of which snack nutritional ingredients influence college students’ snack choices. This study examined college students’ awareness of the healthiness of a snack relative to other snack attributes, and considered the ranking of nutritional ingredients that influence their snack choices to ultimately investigate whether nutrition education could foster healthier snack selections. To examine this, two Discrete Choice Experiments (DCEs) were constructed using a unique approach of Block Fractional Factorial Designs. The first DCE examined four three-level snack attributes (healthiness, taste, convenience and price). The second DCE measured six two-level nutritional ingredients (sugar, salt, calories, fat, “all natural,” and fiber). Nested logit models were used to analyze the DCEs and relative importance scores were calculated. A total of 1,624 undergraduate students from a large diverse public institution from the Orange County area in Southern California completed the DCEs. We found healthiness (55%) and sugar (24%) had the highest relative importance score of snack attributes and nutritional ingredients, respectively. College students tended to prefer quick and cheap snacks, but higher prices on healthy snacks did not affect their decision significantly. In addition, high-sugar snacks were less favored if college students had nutrition education. Interventions to increasing the availability of healthier

For more than two decades, an epidemic of obesity in the United States has been linked to rising rates of type 2 diabetes. Both conditions have been increasing among the nation’s youth. The increase in type 2 diabetes among young people portends many decades of treatment of the disease and its complications for a larger population than previously known, with all the attendant cost in healthcare resources. This study will analyze the cost to our

We analyzed data from our FAIR Health database of billions of privately billed healthcare claims to identify trends and patterns from 2011 to 2015 in obesity, type 2 diabetes and other obesity-related conditions in the nation’s

As obesity rates have increased in children, adolescents and young adults, our data show that type 2 diabetes also has increased in that population. The percent of claim lines with an obesity diagnosis increased annually in all age groups, from infants and toddlers to adults. In the age group 19 to 22 years, the increase in obesity claim lines was 154 percent. During the same period, the percent of claim lines with a type 2 diabetes diagnosis more than doubled in the pediatric population, increasing 109 percent. In most pediatric age groups, claim lines with an obesity diagnosis were more common for females than males, but claim lines with a type 2 diabetes diagnosis were more common for males

Using both charge and allowed amount data, we will compare the healthcare costs of pediatric patients diagnosed with type 2 diabetes to those of all pediatric patients. As an indicator of the future healthcare costs of this young

Type 2 diabetes associated with obesity in young people has been growing significantly; this study will document its economic impact on the healthcare system. That impact in the form of increasing healthcare costs will likely grow as this generation matures. By understanding the scope of this problem, payors, providers, government officials and policy makers can be better prepared to address and mitigate this public health issue of growing concern.

Chelsea Crain [email protected]

Juan Carlos Caro [email protected]

Sayaka Nakamura [email protected]

In an attempt to reduce the consumption of sugar-sweetened beverages (SSBs), stem the rise in obesity, and generate government revenue, taxes on SSBs have become increasingly common across the country. Boulder, Colorado, implemented the largest SSB tax to date on July 1, 2017, increasing the price of SSBs by two cents per ounce. To understand how the tax influcenes consumers, this paper uses multiple data sources to estimate the impact of the SSB tax in

The first data source is comprised of hand collected prices of beverage items from retailers and restaurants in April, June, August, and October of 2017. The second source is a collection of online restaurant menus from Grubhub.com and Orderup.com, which were collected weekly from March to October of 2017. The final data source is comprised of beverage prices from retailers listed on Instacart.com and Walmart.com, which were collected weekly from November, 2016 to October, 2017. Price data from each source contains observations both within Boulder and directly outside of Boulder. We compare the changes in the price per ounce of SSBs and non-SSBs over time in Boulder compared to the changes during the same period in Boulder County (outside the city) and Fort Collins. The multiple periods of data before implementation allow us to determine if Boulder County and Fort Collins are appropriate comparison groups to Boulder in that the parallel trends assumption is satisfied. Analysis demonstrates that the trends in prices over time prior to July 1 are similar in Boulder and in the rest of Boulder County and Fort Collins. Preliminary estimates indicate that, on average, 50 percent of the SSB tax was passed on to consumers. These pass through results are heterogeneous across firm type, beverage type, and beverage size, varying in magnitude between 30 to 80 percent pass-through. We also find that the price of non-SSB soft drinks significantly increase in Boulder restaurants after the implementation of the tax. Further, with the data dating back to the beginning of November, 2016, prior to

Neuroscience studies and observational analyses in the context of a modern average diet, suggest that low-calorie sweeteners (LCS) are linked with overweight and metabolic changes which enhance sweet preference (Murray, Tulloch, Criscitelli, & Avena, 2016; Swithers, 2013; Sylvetsky et al., 2017; Yang, 2010). Random controlled trials (RCT) and studies which match on dietary patterns suggest the opposite—no effect on food preferences or cardiometabolic outcomes (de Koning et al., 2012; Duffey, Gordon-Larsen, Steffen, Jacobs, & Popkin, 2010; Piernas, Tate, Wang, & Popkin, 2013; Tate et al., 2012). As consequence, systematic reviews reflect lack of consensus on the issue of addiction (Rogers et al.,

Overall, understanding LCS addiction relative to sugar is key from a public health perspective. If all sweetened beverages (SB) are similarly addictive, all of them should be regulated to reduce sweetness dependence, and thus, overweight and obesity prevalence. This study provides evidence on the addiction to LCS and sugar in sweetened beverages in Chile. In particular, we will use a random-coefficient discrete choice model, extending from Richards, Patterson, and Tegene (2007) to account for the addiction to products’ nutrients under the rational addiction framework, originally introduced by Becker and Murphy (1988). We test the relative addiction to LCS versus sugar in SB, allowing for several types of beverages and household characteristics. Data source comes from Kantar WorldPanel, which include weekly purchases of packaged foods and beverages from a representative sample of Chilean households (≈2,000) from the 2013-16. Purchase data is connected to nutrition facts panel data collected in several retail venues between 2015 and 2016, to calculate household purchases of specific nutrients, namely sugar and LCS. Chile is ideal as a case study for a rational addiction model of sweetened beverages. First, in contrast to the remarkable success reducing tobacco and risky alcohol consumption, obesity has increased sharply in the last decade, while diet and physical activity does not seem to improve (Ministerio de Salud, 2017). In 2017, three of every four adults are overweight, while obesity prevalence peaked at 34.4%. Additionally, Chile reports the largest consumption of sweetened beverages (Popkin & Hawkes, 2015), with a particularly large market participation of LCS beverages (Caro, Taillie, Ng, & Popkin, 2017). Finally, the last modification to the Chilean SB tax reduced the relative price of LCS beverages compared

Results of this study are critical for current food policy research and practice. First, we can address to which magnitude LCS are addictive in relation to sugar-sweetened beverages at the population level. Secondly, we can observe heterogeneity in addictive behavior based on household characteristics. Last, we can observe how tax policies affect households’ behavior in the context of rational addiction. Presence of LCS beverage addiction suggests that incentives to

We examine causal effects of Japanese school lunch program on 13- to 15-year-old children’s weight. We use individual level data drawn from the 1975-1994 National Nutrition Survey, a nationally-representative household survey with

Japanese compulsory education consists of six years of elementary school for 6- to 12-year-olds and three years of junior high school for 12- to 15-year-olds, where the majority attend municipal schools. Each municipality decides whether to provide school lunch at municipal elementary and junior high schools, and at a school where school lunch is provided, all students must eat school lunch in principle. While almost all municipalities provide school lunch for elementary schools, there is a large variation in municipal school lunch provision for junior high schools. We use a modified DID framework and compare differences between junior high school students and older (9- to 12-year-old) elementary school

We conduct subsample analysis by socioeconomic status (SES) to explore treatment heterogeneity, which has not been fully considered in the literature. School lunch alters children’s food intake at lunch, which might increase or decrease their weight, depending on the relative contents of school lunch and counterfactual lunch they would have eaten in the absence of school lunch. This implies possible treatment heterogeneity. Heterogeneous treatment effect might also

Studying Japan offers several merits. First, and most importantly, the lack of individual choice in school lunch participation at municipal schools allows little room for self-selection. This enables us to estimate population effect of actual participation in school lunch program and assess treatment heterogeneity by children’s socioeconomic background, both of which are infeasible in previous quasi-experimental studies. Second, strict nutritional standards embedded in Japanese school lunch program minimize program heterogeneity, and also foster understanding the effects of recent introduction of stricter than conventional nutritional requirements in UK and US. Third, there are no other large-scale food provision programs, unlike in US with various food assistance programs. Fourth, the absence of means testing minimizes concerns for stigma-induced reporting bias in school lunch participation. We find no evidence that school lunch affects body weight in full sample analysis. In subsamples of children with low socioeconomic background, however, we find significant negative effect of school lunch on BMI and obesity. We find no evidence that school lunch affects underweight prevalence. These findings are robust to propensity score trimming. Additionally, municipal school lunch provision for junior high schools is not significantly associated with growth or weight problems of younger children, inconsistent with endogeneity concerns. We find some evidence that the weight reduction effect for low SES children persists several years after they graduate from junior high school and stop having school

Molly Jacobs [email protected]

Noura Insolera [email protected]

David Jones [email protected]

: Research shows that overweight adolescents become overweight adults, but do the same factors contribute to weight in adolescence as adulthood? Are extrinsic factors more important than intrinsic characteristics? This study identifies the correlation between BMI and various intrinsic and extrinsic covariates and evaluates their relative importance in BMI determination. Furthermore, it separates the sample into adolescents (12-20) and adults (21+) and

: Using 15 years of panel data, multi-level change models including both fixed and random effects assess the impact of extrinsic—environmental, biological, geographic and household—and intrinsic—sexual activity, substance use,

: Results suggest that race and age are the most significant determinants of BMI at all ages. However, other determinants differ between adolescents are adults. In the early years of adolescents, intrinsic factors are highly deterministic, while extrinsic factors are insignificant. Intrinsic determinates of significance include age of first sexual encounter, tobacco experimentation, perspective on general health and desire to lose/stay the same weight play a

: While biological/genetic attributes, are the largest determinants of BMI at every age, intrinsic factors appear to be more significant than extrinsic characteristics during adolescents. As individuals age, intrinsic determinants decrease in importance as extrinsic characteristics increase in significance. Thus, the weight determinants differ between adolescents and adults suggesting different methods of policy intervention be used to improve the health of both

In the United States, 13% of the population is food insecure, and one in seven Americans receive food assistance from the Supplemental Nutrition Assistance Program (SNAP). Half of all children born in the US every year receive supplemental nutrition support from the Special Supplemental Nutrition Program for Women, Infants and Children (WIC). SNAP and WIC improve short-term food security, but their effect on long-term food security is unknown. Nutritional knowledge is an important predictor of diet quality and healthy weight, however less is known about the relationship between parental nutritional knowledge, food security, and the risk of obesity in children as they transition to adulthood. The goal of this study is to examine the relationship between nutrition assistance program participation and parental nutritional knowledge during childhood with obesity and food insecurity risk in young adulthood.

Data for this study come from the 1997-2015 waves of the Panel Study of Income Dynamics (PSID) including the Child Development Supplement (CDS), and Transition into Adulthood Supplement (TAS). Begun in 1968, the PSID is a nationally representative, multigenerational household panel survey. The study sample includes a balanced panel of 2,796 families whose children were age 0-12 years in 1997 during the first wave of the CDS and age 17-31 in 2015. We created a ‘parental nutritional knowledge’ index based on five nutrition knowledge questions. To examine the relationship between SNAP and WIC receipt in childhood (in 1996-2000) and parental nutritional knowledge (in 1999) with

Preliminary results show that individuals who are food insecure during childhood have a higher risk of obesity in young adulthood (26% vs. 21%), while those who are persistently food insecure have an even larger disparity compared to those who are never food insecure (33% vs. 18%). In adjusted models, among low-income individuals with persistent food insecurity, food stamps participation during childhood reduces risk of obesity from 36% to 34%. Among low-income, food insecure children, SNAP participation increased the chances of becoming food secure in young adulthood from 50% to 54%. Parental nutritional knowledge analyses will also be presented.

Food insecurity during childhood has long-term implications for health, however, participation in nutritional assistance programs may mitigate that relationship and lower the risk of obesity in young adulthood.

Several cities in the U.S. have implemented taxes on sugar-sweetened beverages in an attempt to improve public health as well as raise revenue. On January 1, 2017, Philadelphia became the largest U.S. city to implement a beverage tax, which is 1.5 cents per ounce and applies to both caloric and non-caloric but artificially sweetened beverages. In this paper, we estimate the impact of the tax on retail prices, marketing, purchases, and child and adult consumption of taxed

The study includes an innovative nested data collection approach in which we collected: (1) prices and marketing from retailers, (2) purchase information from customers exiting the retailers, and (3) a follow-up household survey of those customers regarding household and child beverage purchases and consumption. We collected the information from a representative group of retailers and their consumers in Philadelphia and a matched group of retailers in surrounding counties that serve as a control group. We collected the information in the months prior to the implementation of the tax and again a year later. In addition, we examine detailed receipt data on beverage purchases in the six months prior to and after implementation of the tax for a sample of households in Philadelphia as well as for two control groups: residents in surrounding counties and a national matched comparison group. We use a difference-in-differences identification strategy to estimate the impact of the tax on prices, marketing, purchases, and consumption of taxed beverages. We also estimate the impacts on untaxed beverages to assess potential substitution by consumers. In addition, we examine differential impacts by proximity to the city border to determine whether spillover effects might dampen the impact on prices close to the border and whether residents increase travel across city borders to purchase beverages after the tax. Furthermore, we examine whether taxes have different impacts on prices and marketing by store type, brand, and beverage container volume, and whether the impacts on purchases and consumption vary by household income, age of the child, race, and ethnicity. Taken together, the findings illuminate how the tax in Philadelphia affected prices and how the changes in prices affected purchases and consumption.

Jonathan Cantor [email protected]

Sugar-sweetened beverages (SSBs) have little nutritive value and add a nontrivial number of calories to diets. Frequent SSB consumption is associated with the incidence of obesity, diabetes, and other health conditions related to the cardiovascular system. Because excessive SSB consumption has a negative health impact, the purchase price of SSBs may not capture the full cost to society. To help internalize this cost, most states apply sales taxes to SSBs and some U.S

contemporaneous impact of SSB taxes. A focus on the impact of contemporaneous SSB taxes may be insufficient to capture the entire long-run effect since stages of the relationship—the effect of SSB taxes on SSB consumption and the effect of any resulting change in SSB consumption on health—could occur gradually over time. There are at least two reasons why there may be a lag in the (full) consumption response to a SSB cost change. First, it may take time for some consumers to notice a change in SSB costs or to fully appreciate a cost change relative to their food budgets. Second, consumers may be slow to adjust their intake of SSBs because of the habit-forming nature of SSB consumption. In addition, the short- and long-run impact of SSB taxes may differ because

Using 1999-2014 restricted-use National Health and Nutrition Examination Survey (NHANES) data at the state level, we contribute to the literature by analyzing the short- and long-run effect of SSB taxes on caloric intake, body weight outcomes, blood sugar levels, blood cholesterol levels, blood pressure, as well as a metabolic syndrome index. A noteworthy outcome of interest is blood sugar levels since diabetes as an outcome has been studied relatively little in the literature despite it often being cited as a main reason for implementing a SSB tax. Our study will provide an up-to-date analysis of the health impacts of SSB taxes among children, adolescents, and adults, and will shed light on whether and how the sign and size of any health effects of SSB taxes depends on the time horizon of the analysis. The results of our study will help to inform future tax policy-making decisions by showing the importance of the long-run benefits of



California State University Fullerton Sanam Rusmevichientong; Jessica Jaynes Complete

FAIR Health Complete

University of Iowa David Jones; John Cawley; David Frisvold Complete

CPC, UNC at Chapel Hill Juan C Salgado Complete

Nagoya University Shiko Maruyama Complete

East Carolina University Complete

Julia Wolfson; Alicia Cohen Complete

Mathematica Policy Research David Frisvold; John Cawley Complete

University of Michigan - Panel Study of Income Dynamics

RAND Corporation Brandon Restrepo Complete


Physician/Nurse Reimbursement, Training and Behavior

The effect of specialists' income on the application of the residency program


Effects of Episode-Based Payment on Health Care Spending and Utilization: Evidence from Perinatal Care in Arkansas


Factors Associated with the Acceptance of New TRICARE Standard and Medicare Patients by Health Care Providers


Evidence of Hot-Hand Behavior in Sports and Medicine


Malpractice Allegations and Physician Productivity: Evidence from the Emergency Department


Public Health Insurance Coverage Expansion and Physicians’ Working Hours


Equity in Health Care: A Field Experiment on the Effect of Socioeconomic Status on Access to Outpatient Care


Variation in Medical Prices and Outcomes of Injured Workers


Physician Payments and the Receipt of Human Papillomavirus Vaccine among Privately Insured Adolescents


Nurse Practitioner Prescriptive Authority and Prescription Opioid Use


Health Services as Credence Goods: A Field Experiment


Medicaid Physician Fees and Use of Primary Care Services: An Analysis of Recent Data Inclusive of ACA Fee Increase


Contract Physicians and the Quality of Primary Care: Evidence from China’s Iron Rice Bowl


The Spillover Impacts of a Medicare Payment Reform


Do State Regulations Restricting Prescriptions of Opioids Change Provider-Pharma Financial Ties? Evidence from Open Payments Data

Abstract

Objective In Korea, there is a controversy over the imbalance of supply and demand of medical specialists due to their preference of certain type of specialties. To help guide medical specialist workforce policy, we estimate the impact of the income of medical specialists and related variables, which are pointed out as the reasons behind the imbalance in application for different fields of specialization. Methods Using Korea Health Data from 2001 to 2013 for 26 specialist fields, we employ a panel analysis including fixed effects models and random effects models for estimating occupancy rate model of specialists. The human capital approach was used as a theoretical model for this study. Findings First, occupancy rate of specialists varies depending on the income of specialists, and it was projected that the rate increases as the income gets lager. Above all, effect of medical specialist income on their occupancy rates was shown to be greater in supporting fields or minor fields of specialization than in that of major. The income elasticity of the medical specialist was between 0.0377 and 0.09152. Second, the existence of medical specialist training subsidy, one of the variables that represent group effects of specific fields of specialization, appeared to affect the occupancy rate of medical specialists. Third, the difficulty of training and medical treatment/consultation, one of the variables that represent the characteristics of different fields of specialization, appears to affect the occupancy rate of medical specialists greatly. Fourth, the increase/decrease ratio of medical specialists, which was categorized as a variable that shows the characteristics of different fields of specialization, appeared to have an impact on the occupancy rate of medical specialists. In general, the occupancy rate decreased by 0.135% point when the quota of medical specialists was raised by 1%. It is seen to indicate the necessity to determine the appropriateness of the medical specialist quota and adjust it in order to enhance the medical specialist utilization rate. Conclusion The concentration of application in certain fields of specialization that leads to the imbalance in medical specialist occupancy rate can be partly explained by models based on the medical specialist income, job stability and characteristics of each field, and is partly attributed to unique characteristics of different fields that are not explained by these models. Therefore, it is necessary to advance into the direction of improving the inequity of group effects of different fields. There are policy means to improve the occupancy rate by adding a certain percentage to the health insurance fees or using policy variables to standardize the preference index of fields which have preference index.

In this project, we study how physicians respond to episode-based bundled payment (EBP), a prominent payment reform that pays a case rate for an entire episode of care. Unlike FFS reimbursement, EBP holds physicians responsible for all care within a discrete clinical episode, rewarding physicians not only for efficient use of their own services but also for efficient management of other health care inputs. While EBP programs are expanding, existing research is generally limited to evidence from voluntary demonstration projects, mainly in the Medicare market. We study the impact of EBP under the Arkansas Payment Improvement Initiative (APII), a multi-payer program that requires providers in the state to enter into EBP arrangements for perinatal care. Because of its multi-payer nature and the requirement that providers participate, the program covers the vast majority of births in the state. In a difference-in-differences analysis of commercial claims, we find that perinatal spending decreased by 3.8% overall in Arkansas after the introduction of EBP, compared to surrounding states. We find that the decrease was driven by reduced spending on non-physician health care inputs, specifically the prices paid for inpatient facility care. Reductions in the price of care could reflect referral patterns favoring low price facilities or lower negotiated rates at a given facility; we find preliminary evidence that a change in referral patterns is more likely. Our results are robust to a number of sensitivity tests, including alternate control groups, and we demonstrate that there was no effect among placebo conditions not subject to EBP. We additionally study quality of care under EBP by analyzing changes in screening rates for common perinatal conditions. Overall, we find that EBP was associated with a limited improvement in quality of care; out of six screening tests, we find increased utilization of only one under EBP.

Many Americans rely on public health insurance programs for their health care needs. This paper seeks to improve the understanding of provider acceptance rates of two commonly used sources of public health insurance: TRICARE, which serves military personnel and their family members and Medicare, which primarily serves individuals over age 65. We use data from a congressionally mandated survey of civilian physicians to address three main research questions: (1) How does the acceptance rate for new TRICARE Standard patients compare to the acceptance rate for new Medicare patients? (2) What are the similarities and differences in the reasons given by providers for not accepting TRICARE Standard or Medicare? And (3) what provider and local-area characteristics are associated with the decision to accept each insurance type? The TRICARE Standard Survey of Providers was administered to a nationally representative sample of civilian physicians and non-physician mental health providers from 2012 to 2015. Among other questions, providers were asked whether they accept any new patients, new TRICARE Standard patients, or new Medicare patients. The data also include information about the characteristics of the providers, such as their specialization, practice type, and age. In addition, we incorporate data from the American Community Survey to measure local area characteristics, such as the number of providers per 1,000 residents and per capita income. Our analysis reveals that both TRICARE Standard and Medicare have high acceptance rates among physicians, with almost three-quarters of physicians accepting both TRICARE and Medicare, only 3% accepting neither TRICARE nor Medicare and 4% not accepting any new patients. However, Medicare is more commonly accepted by physicians than TRICARE; 18% accept Medicare but not TRICARE but only 3% accept TRICARE but not Medicare. Acceptance rates are much lower for mental health providers than physicians, with 40% of mental health providers not accepting either insurance type. Reimbursement and providers’ belief that their specialty is not covered are commonly stated reasons by both physicians and mental health providers for not accepting TRICARE and/or Medicare, and those who have problems with reimbursement tend not to accept either insurance type. Mental health providers are more likely than physicians not to be aware of TRICARE. Finally, providers who accept new TRICARE but not Medicare are more likely than those making other acceptance choices to be found in areas where more individuals are eligible for TRICARE, which suggests that acceptance decisions may reflect the demand of the local area population. These findings provide insight into challenges and opportunities associated with attempts to increase provider acceptance of public insurance such as TRICARE and Medicare. For example, improving reimbursement is important for all public insurance. For smaller programs like TRICARE, informing providers about coverage offerings may increase acceptance.

This paper provides empirical evidence of hot-hand bias in two novel field settings: dart players' strategic choices, and physicians' decisions during childbirth. The “hot hand” is the notion that a person can enter a state in which her probability of success becomes higher than normal. “Hot-hand bias” refers to an exaggerated belief in the hot hand (whether it exists or not). First, I collect data of professional dart players from the 2016 World Darts Championship. The players are significantly more likely to hit after a successful shot, implying that players have a hot hand. Based on a precise estimate of the hot hand, I calculate the optimal strategy of a profit-maximizing dart player. I find that dart players are much more willing to take risks after a successful shot than what I calculate to be optimal. Second, I utilize 1.3 million hospital admissions for childbirth in New York State over 2010-2015. I find no evidence that physicians have a hot hand when performing obstetrical procedures. In the absence of hot hand, physicians are still 2% more likely to perform a C-section after a previous successful C-section. My empirical model includes physician fixed effects, and a large set of patient conditions that proxy for when a C-section is likely to maximize patient welfare. Across the two settings, robustness checks provide additional evidence consistent with decision-makers having a hot-hand bias. Generalizing the medical findings to the United States and assuming that the identified 2% increase in the C-section rate is unwarranted, the estimated health-care cost is $65 million per year.

While a substantial literature has studied the influence of malpractice pressure on physician behavior, existing research has not found that malpractice concerns influence physicians to any great extent. However, these studies generally focus on variation in state laws governing malpractice exposure. What is perhaps more important is the actual experience of being sued. In this project, we test how physicians respond to malpractice lawsuits – both those that are successful and those that are not. Despite the fact that the vast majority of physicians will face a malpractice claim during their career, there is little evidence on how experience with the liability system informs physicians’ assessments of malpractice pressure. We study the impact of malpractice allegations on the labor supply and treatment intensity decisions of Emergency Department (ED) physicians, combining physician-level data on malpractice claims with the universe of ED discharges in Florida. To address potential differences between physicians with different allegation histories, we exploit variation in the timing of unexpected malpractice allegations. We find that physician labor supply decreases sharply after malpractice allegations and that this reduction is persistent over time. We estimate a 10% reduction in physicians’ patient loads overall and find that our results are driven by intensive margin responses, specifically that physicians reduce the number of patients they treat but maintain practice in the state. We further find that physicians stop practicing at the hospital where the alleged negligence occurred, and that part of the decline in total patient volume is driven by physicians leaving high volume hospitals to practice at smaller facilities. Next, we find that physicians increase care intensity among their remaining patients, increasing total charges per patient by about 5% after an allegation. Lastly, we provide suggestive evidence that physicians do not respond optimally to malpractice allegations, showing that physicians respond equally to legitimate claims of negligence as to allegations that are ultimately dismissed. We additionally show that physicians adjust practice patterns equally for all patient types after a malpractice allegation, rather than using information from the allegation to adjust care for clinically relevant patients.

Over the past two decades, there have been two large public health insurance expansions, the State Children’s Health Insurance Program (SCHIP) and the Affordable Care Act (ACA). These insurance programs have significantly increased the number of patients with public health insurance and the demand for medical services, but it is not clear whether providers will supply additional services for newly-insured patients. Because, public health insurance programs provide relatively low reimbursement rates to physicians which discourage physicians to accept publicly-insured patients. This paper focuses on the labor supply response of physicians to these expansions. I use data from the Community Tracking Study (CTS) physician survey (1996-1997, 1998-1999, and 2000-2001) to examine the SCHIP expansion and the 2012-2015 American Community Study (ACS) to examine the ACA expansion. In response to the introduction of SCHIP, my estimates suggest that physicians reallocate their total working hours between patient care and non-patient care activities. The size of the impact was greater in areas with high level of physician concentration prior to the expansion. Physicians in high concentration areas tend to decrease time spent on direct patient care, but increase hours on non-direct patient care. In response to the ACA, physicians’ working hours did not increase, but working hours, the probability of being employed, and the probability of being employed increased for registered nurses. This suggests that physicians might utilize other healthcare providers to accommodate increases in demand for medical services after the expansion.

Inequality in access to health care may have different underlying reasons. Employing a randomised field experiment, we study the impact of socioeconomic status on health care access varying the patient’s educational level. We find that practice assistants favor patients with a degree over those without a degree in whether they offer an appointment. Physicians, in contrast, favor patients with a degree with respect to response time to the request for an appointment and in terms of waiting time. We argue that our results are consistent with implicit bias for practice assistants and statistical discrimination based on financial incentives for physicians.

Medical prices for workers compensation are set by medical fee schedules, and these fee schedule rates vary substantially across states (Fomenko and Liu, 2016). This large variation in fee schedule rates leads many policymakers and system stakeholders to express the following concerns about prices for medical care—when medical fee schedules are low and leading to relatively low prices for medical care, workers may have harder time finding desired medical providers. This may lead to problems getting access to timely care, prolonged recovery, lower rates of return to work and worse outcomes for injured workers. This study is a first comprehensive attempt to empirically examine relationship between prices of medical services and the outcomes that workers experience after a work-related injury. This study combines multiple sources of data on medical prices and outcomes of injured workers. We use medical billing information from Truven MarketScan® to determine prices for office visits for group health payors, and information from Workers Compensation Research Institute DBE databate to determine prices for office visits for workers’ compensation payors at MSA level. This helps describe an economically relevant variation in group health and workers’ compensation medical prices across different areas—when workers compensation prices are lower than prices paid by other payers, these providers may have fewer incentives to provide care to injured workers and instead focus their practice on patients with more generous insurance coverage. We then link this information to more than 6,600 surveys of injured workers collected by WCRI for injuries covering 2010 through 2013 across 14 states. We use these surveys to examine the relationship between medical prices and the outcomes including workers’ speed and sustainability of return to work, reported problems getting desired medical provider and desired medical care, and recovery of health and functioning. We supplement this analysis with the information from administrative records on the time to first non-emergency visit, mix of services provided (e.g. conservative versus specialty invasive care) and duration of disability payments. We find that that workers’ compensation prices are strongly associated with measures of access to care and nature of care received. Workers in areas with lower prices were more likely to report “big problems” getting the primary provider that they wanted, had longer time to first non-emergency office visit for evaluation and management services, higher likelihood of a surgery, and smaller in the number office visits. At the same time, we find that prices were not strongly related to recovery of health and functioning, measures of speed and rates of return to work, and measures of duration of temporary disability benefits.

In 2006, quadrivalent human papillomavirus vaccine (4vHPV) was licensed for use in females and in 2007 the Advisory Committee on Immunization Practices (ACIP) recommended routine use of 4vHPV in females aged 11 or 12 years. The licensure of HPV vaccine for use in males was approved in 2009 and the associated ACIP recommendations was published in 2011. Although HPV vaccines have shown promising results in reducing HPV infections, the uptake of the vaccine has been less than ideal. Financial concerns, such as high vaccine purchase costs and inadequate insurance reimbursements, are commonly cited as one of the key barriers to human papillomavirus (HPV) vaccination among healthcare providers. None of the existing studies has assessed physician payment for providing HPV vaccination services and its potential impact on the uptake of HPV vaccines among adolescents. This study uses the 2013-2014 MarketScan Commercial Claims and Encounters Database (CCAE). The sample includes adolescents 11-17 years continuously enrolled in a non-capitated private insurance plan in each year and excludes those who had received a HPV vaccine prior to each study year. Linear regressions are used to estimate the link between physician payments for a HPV vaccination visit and the probability of adolescents initiating the HPV vaccine series and the probability of adolescents receiving ≥ 2 HPV vaccine doses in the year. The key independent variable is state median payment for a HPV vaccination visit. The payment variable includes all aspects of the provider’s income incurring from providing a HPV vaccination service - insurance reimbursements for the HPV vaccine and vaccine administration and patients’ cost sharing for the visit (deductibles, coinsurance, and copayment). The findings show that physician payments are positively associated with HPV vaccine uptakes. For every $1 increase in payments, the probability of adolescents initiating the HPV vaccine series increased by 0.18 percentage point (an increase of 1.5% from the mean) and the probability of receiving ≥ 2 HPV vaccine doses increased by 0.10 percentage point (an increase of 2.1% from the mean).

This paper uses state differences in the nurse practitioner (NP) market to evaluate the effects of state laws allowing NPs to prescribe controlled substances on prescription opioid use. I study these effects by merging nationwide data from the Medical Expenditure Panel Survey (MEPS) over 18 years (1996-2013) with data on state laws. I then exploit variation in these laws over time to create a quasi-natural experiment and to estimate the causal impact of NP deregulation on prescription opioid use. I find, relative to patients living in more restrictive states, that patients who live in states with more flexible NP laws reduce their prescription opioid use by 7 percent to 9 percent. I also find that health outcomes either slightly improve or remain unaffected by the enactment of these laws. Taken together, these results indicate that NP deregulation slows the trend in prescription opioid growth while potentially improving patient outcomes. Furthermore, suggestive evidence implies that these effects may be even larger for the least restrictive states, opening the door for future reforms.

Information problems are a defining characteristic of health care markets. Despite their importance, there is little direct evidence of the impact of information problems between patients and physicians on the quality of treatment. Furthermore, the role of market conditions is not well understood. In this paper, we present the results from a field experiment in the market for dental care. A test patient who does not need treatment is sent to 180 dentists to receive treatment recommendations. In the experiment, we vary two factors: First, the information that the patient signals to the dentist. Second, we vary the perceived socioeconomic status (SES) of the test patient. Furthermore, we construct several measures of market conditions. Our study has two main contributions: First, we investigate physicians' provision of health care services on the level of individual patient-physician interactions. The design allows us to observe for each physician whether she/he provides the appropriate treatment recommendation or an overtreatment recommendation instead of observing only aggregate provision rates. Thus, we can provide direct evidence of overtreatment and thereby have a clean and simple measure of physician quality. Our micro approach allows us to not only observe the overtreatment behavior but also to control for the covariates on the individual level. We find that the patient receives an overtreatment recommendation in more than every fourth visit. A low waiting time for the next possible appointment, indicating excess capacities, is associated with significantly more overtreatment recommendations. Overall, results regarding the role of market conditions can be well explained based on short-term capacity considerations. Furthermore, we observe signicantly less overtreatment recommendations for the patient with higher SES compared to lower SES under standard information. More signalled information however does not signicantly reduce overtreatment.

We examine whether fees paid by Medicaid to primary care physicians affects the use of health care services with data that span the recent and large increases in fees mandated by the ACA. A difference-in-differences research design is used to obtain estimates of the association between Medicaid physician reimbursement and whether a person has a personal doctor (provider); whether a person had a routine check-up in the past year; whether a woman had a breast exam or a Pap test in the past year; whether a person had a dental visit or flu shot in the past year; whether a person has ever been diagnosed with asthma, diabetes, cardiovascular diseases, cancer, or chronic obstructive pulmonary disease; and whether a person had diabetes care in the past year. Data for the analysis are drawn from the Behavioral Risk Factor Surveillance System (BRFSS), which is one of the few sources of publicly available data that has the requisite information to conduct a study of this type. Results indicate that Medicaid fees for primary care are mostly unrelated to the use of services we examined with the exception that dental visits are positively, though modestly, associated with Medicaid fees. We conclude that Medicaid fees do not seem to have a major impact on the use of services; however, we note that our study looked at a narrow range of services due to data availability.

We present audit study evidence from China on the relative quality of outpatient care provided by physicians employed in civil service posts and those on fixed-term contracts in rural health centers. We use data from unannounced standardized patient (SP) interactions with physicians and within-clinic variation in contract status generated by China’s bianzhi system – a headcount quota system defining personnel assignments in public service units – to address bias arising from patient and doctor sorting. We find that employment on a fixed-term rather than civil service contract is associated with a large and significant improvement in physicians performance, despite physicians employed on fixed-term contracts being paid significantly less and having fewer formal qualifications. The estimated positive effect of fixed-term contracts on quality increases after controlling for observable physician characteristics. We also find direct evidence that these effects are due to increases in physician effort: comparisons between interactions with SPs and clinical vignettes testing knowledge of the same disease cases to the same doctors shows that civil service doctors exhibit greater underperformance relative to their knowledge of appropriate clinical practice.

By altering the incentives faced by providers, Medicare can indirectly affect care for the privately insured. Healthcare research on these spillover effects has focused on Medicare Advantage and traditional fee-for-service (FFS) Medicare or used payer mix variation to identify the effects of public insurance schemes, but scarce research assesses the indirect effects of Medicare on the population under age 65. This paper does just that, using a recent Medicare payment shock where some providers (but not all) moved from a traditional FFS payment model to a model that rewarded patient management and cost containment.

In 2016, the Centers for Medicare and Medicaid Services (CMS) introduced the Oncology Care Model (OCM) that included two distinct changes to provider reimbursement. First, participating providers were paid a monthly patient management fee to encourage care coordination. Second, a shared-savings program was established that rewarded providers when actual patient spending – administered at the patient-level – was below a projected benchmark level determined by CMS. Evidence suggests that the OCM program prompted practice reorganization, including the hiring of non-physician clinicians, and led to an increased emphasis on evidence-based treatment. Given that Medicare represents a significant portion of practice total reimbursement, it is conceivable that the OCM program may have induced changes in spending and health care use for patients with private insurance.

Using a unique dataset of medical claims from a large provider network, I estimate the spillover effects to the privately insured of the OCM program. These data are particularly suited to examining these issues as half of the provider practices in the sample participated in the OCM program and half did not. The practices themselves exhibit many features that make them compelling comparators: they are part of the same network and thus face similar input prices and care sensibilities, they have the same technologies by way of electronic medical records and financial management systems, they host several networking conferences throughout the year, and they share some common back-office staff. I use a differences-in-differences methodology to measure the effects of the OCM program on utilization and cost among patients with private insurance. I couple this analysis with an event study to assess the onset effects of the policy. Primary outcomes include changes in expenditures, such as drug-spend and total-spend, and secondary outcomes include changes in utilization of specific services such as regular visits to mitigate unnecessary hospitalizations or emergency room visits.

Preliminary results suggest that the Medicare change affected the utilization and cost outcomes for the privately insured population under age 65. Shared-savings models have elsewhere shown modest first year effects among those directly managed under those payment systems; this analysis will clarify whether, where, and to what magnitude these effects spillover to privately-insured patients uninvolved with the program.

Federal Open Payments data show that since August of 2013 through the end of 2016, US physicians explicitly received approximately $8 billion per year from drug and medical device manufacturers in various forms of research payments, investment interest, industry-sponsored meals, payments for education and training, consulting fees, travel and lodging, and entertainment. Prior studies have confirmed positive associations between these payments and rates of brand-name drug prescriptions (for instance, Yeh et al. 2016). The provider-pharma financial ties may also affect opioid prescribing practices, which have been a considerable focus of recent state regulatory changes. This paper utilizes the federal Open Payments database to investigate the impact of key state laws enacted to regulate opioid prescription practices, on provider-pharma financial ties. State laws of interest include state prescription drug monitoring programs, pill mill laws, and physical examination requirements, among others. This study aims to shed light on economic predictions of how regulations that reduce product demand could reduce promotional activities by industry. More specific to policy, this study examines potential conflicts of interest that could interfere with physicians’ responsibilities to their patients, and to the US public health efforts to combat the opioid epidemic.

Abstract

In Korea, there is a controversy over the imbalance of supply and demand of medical specialists due to their preference of certain type of specialties. To help guide medical specialist workforce policy, we estimate the impact of the income of medical specialists and related variables, which are pointed out as the reasons behind the imbalance in application for different fields of specialization.

Using Korea Health Data from 2001 to 2013 for 26 specialist fields, we employ a panel analysis including fixed effects models and random effects models for estimating occupancy rate model of specialists. The human capital

First, occupancy rate of specialists varies depending on the income of specialists, and it was projected that the rate increases as the income gets lager. Above all, effect of medical specialist income on their occupancy rates was shown to be greater in supporting fields or minor fields of specialization than in that of major. The income elasticity of the medical specialist was between 0.0377 and 0.09152. Second, the existence of medical specialist training subsidy, one of the variables that represent group effects of specific fields of specialization, appeared to affect the occupancy rate of medical specialists. Third, the difficulty of training and medical treatment/consultation, one of the variables that represent the characteristics of different fields of specialization, appears to affect the occupancy rate of medical specialists greatly. Fourth, the increase/decrease ratio of medical specialists, which was categorized as a variable that shows the characteristics of different fields of specialization, appeared to have an impact on the occupancy rate of medical specialists. In general, the occupancy rate decreased by 0.135% point when the quota of medical specialists was raised by 1%. It is seen to indicate the necessity to determine the appropriateness of the medical specialist quota and adjust it in order to enhance the medical specialist utilization rate.

The concentration of application in certain fields of specialization that leads to the imbalance in medical specialist occupancy rate can be partly explained by models based on the medical specialist income, job stability and characteristics of each field, and is partly attributed to unique characteristics of different fields that are not explained by these models. Therefore, it is necessary to advance into the direction of improving the inequity of group effects of different fields. There are policy means to improve the occupancy rate by adding a certain percentage to the health insurance fees or using policy variables to standardize the preference index of fields which have preference index.

In this project, we study how physicians respond to episode-based bundled payment (EBP), a prominent payment reform that pays a case rate for an entire episode of care. Unlike FFS reimbursement, EBP holds physicians responsible for all care within a discrete clinical episode, rewarding physicians not only for efficient use of their own services but also for efficient management of other health care inputs. While EBP programs are expanding, existing research is generally limited to evidence from voluntary demonstration projects, mainly in the Medicare market. We study the impact of EBP under the Arkansas Payment Improvement Initiative (APII), a multi-payer program that requires providers in the state to enter into EBP arrangements for perinatal care. Because of its multi-payer nature and the requirement that providers participate, the program covers the vast majority of births in the state. In a difference-in-differences analysis of commercial claims, we find that perinatal spending decreased by 3.8% overall in Arkansas after the introduction of EBP, compared to surrounding states. We find that the decrease was driven by reduced spending on non-physician health care inputs, specifically the prices paid for inpatient facility care. Reductions in the price of care could reflect referral patterns favoring low price facilities or lower negotiated rates at a given facility; we find preliminary evidence that a change in referral patterns is more likely. Our results are robust to a number of sensitivity tests, including alternate control groups, and we demonstrate that there was no effect among placebo conditions not subject to EBP. We additionally study quality of care under EBP by analyzing changes in screening rates for common perinatal conditions. Overall, we find that EBP was associated with a limited improvement in quality of

Many Americans rely on public health insurance programs for their health care needs. This paper seeks to improve the understanding of provider acceptance rates of two commonly used sources of public health insurance: TRICARE, which serves military personnel and their family members and Medicare, which primarily serves individuals over age 65. We use data from a congressionally mandated survey of civilian physicians to address three main research questions: (1) How does the acceptance rate for new TRICARE Standard patients compare to the acceptance rate for new Medicare patients? (2) What are the similarities and differences in the reasons given by providers for not accepting TRICARE Standard or Medicare? And (3) what provider and local-area characteristics are associated with the decision to accept each insurance type? The TRICARE Standard Survey of Providers was administered to a nationally representative sample of civilian physicians and non-physician mental health providers from 2012 to 2015. Among other questions, providers were asked whether they accept any new patients, new TRICARE Standard patients, or new Medicare patients. The data also include information about the characteristics of the providers, such as their specialization, practice type, and age. In addition, we incorporate data from the American Community Survey to measure local area characteristics, such as the number of providers per 1,000 residents and per capita income. Our analysis reveals that both TRICARE Standard and Medicare have high acceptance rates among physicians, with almost three-quarters of physicians accepting both TRICARE and Medicare, only 3% accepting neither TRICARE nor Medicare and 4% not accepting any new patients. However, Medicare is more commonly accepted by physicians than TRICARE; 18% accept Medicare but not TRICARE but only 3% accept TRICARE but not Medicare. Acceptance rates are much lower for mental health providers than physicians, with 40% of mental health providers not accepting either insurance type. Reimbursement and providers’ belief that their specialty is not covered are commonly stated reasons by both physicians and mental health providers for not accepting TRICARE and/or Medicare, and those who have problems with reimbursement tend not to accept either insurance type. Mental health providers are more likely than physicians not to be aware of TRICARE. Finally, providers who accept new TRICARE but not Medicare are more likely than those making other acceptance choices to be found in areas where more individuals are eligible for TRICARE, which suggests that acceptance decisions may reflect the demand of the local area population. These findings provide insight into challenges and opportunities associated with attempts to increase provider acceptance of public insurance such as TRICARE and Medicare. For example, improving reimbursement is important for all public insurance. For smaller programs like TRICARE, informing providers about coverage offerings may increase acceptance.

This paper provides empirical evidence of hot-hand bias in two novel field settings: dart players' strategic choices, and physicians' decisions during childbirth. The “hot hand” is the notion that a person can enter a state in which her probability of success becomes higher than normal. “Hot-hand bias” refers to an exaggerated belief in the hot hand (whether it exists or not). First, I collect data of professional dart players from the 2016 World Darts Championship. The players are significantly more likely to hit after a successful shot, implying that players have a hot hand. Based on a precise estimate of the hot hand, I calculate the optimal strategy of a profit-maximizing dart player. I find that dart players are much more willing to take risks after a successful shot than what I calculate to be optimal. Second, I utilize 1.3 million hospital admissions for childbirth in New York State over 2010-2015. I find no evidence that physicians have a hot hand when performing obstetrical procedures. In the absence of hot hand, physicians are still 2% more likely to perform a C-section after a previous successful C-section. My empirical model includes physician fixed effects, and a large set of patient conditions that proxy for when a C-section is likely to maximize patient welfare. Across the two settings, robustness checks provide additional evidence consistent with decision-makers having a hot-hand bias. Generalizing the medical findings to the United States and assuming that the identified 2% increase in the C-section rate is unwarranted, the estimated health-care cost is $65 million per year.

While a substantial literature has studied the influence of malpractice pressure on physician behavior, existing research has not found that malpractice concerns influence physicians to any great extent. However, these studies generally focus on variation in state laws governing malpractice exposure. What is perhaps more important is the actual experience of being sued. In this project, we test how physicians respond to malpractice lawsuits – both those that are successful and those that are not. Despite the fact that the vast majority of physicians will face a malpractice claim during their career, there is little evidence on how experience with the liability system informs physicians’ assessments of malpractice pressure. We study the impact of malpractice allegations on the labor supply and treatment intensity decisions of Emergency Department (ED) physicians, combining physician-level data on malpractice claims with the universe of ED discharges in Florida. To address potential differences between physicians with different allegation histories, we exploit variation in the timing of unexpected malpractice allegations. We find that physician labor supply decreases sharply after malpractice allegations and that this reduction is persistent over time. We estimate a 10% reduction in physicians’ patient loads overall and find that our results are driven by intensive margin responses, specifically that physicians reduce the number of patients they treat but maintain practice in the state. We further find that physicians stop practicing at the hospital where the alleged negligence occurred, and that part of the decline in total patient volume is driven by physicians leaving high volume hospitals to practice at smaller facilities. Next, we find that physicians increase care intensity among their remaining patients, increasing total charges per patient by about 5% after an allegation. Lastly, we provide suggestive evidence that physicians do not respond optimally to malpractice allegations, showing that physicians respond equally to legitimate claims of negligence as to allegations that are ultimately dismissed. We additionally show that physicians adjust practice patterns equally for all patient types after a malpractice allegation, rather than using information from the allegation to adjust care for

Over the past two decades, there have been two large public health insurance expansions, the State Children’s Health Insurance Program (SCHIP) and the Affordable Care Act (ACA). These insurance programs have significantly increased the number of patients with public health insurance and the demand for medical services, but it is not clear whether providers will supply additional services for newly-insured patients. Because, public health insurance programs provide relatively low reimbursement rates to physicians which discourage physicians to accept publicly-insured patients. This paper focuses on the labor supply response of physicians to these expansions. I use data from the Community Tracking Study (CTS) physician survey (1996-1997, 1998-1999, and 2000-2001) to examine the SCHIP expansion and the 2012-2015 American Community Study (ACS) to examine the ACA expansion. In response to the introduction of SCHIP, my estimates suggest that physicians reallocate their total working hours between patient care and non-patient care activities. The size of the impact was greater in areas with high level of physician concentration prior to the expansion. Physicians in high concentration areas tend to decrease time spent on direct patient care, but increase hours on non-direct patient care. In response to the ACA, physicians’ working hours did not increase, but working hours, the probability of being employed, and the probability of being employed increased for registered nurses. This suggests that physicians might utilize other healthcare providers to accommodate increases in demand for medical services after the

Inequality in access to health care may have different underlying reasons. Employing a randomised field experiment, we study the impact of socioeconomic status on health care access varying the patient’s educational level. We find that practice assistants favor patients with a degree over those without a degree in whether they offer an appointment. Physicians, in contrast, favor patients with a degree with respect to response time to the request for an appointment and in terms of waiting time. We argue that our results are consistent with implicit bias for practice assistants and statistical discrimination based on financial incentives for physicians.

Medical prices for workers compensation are set by medical fee schedules, and these fee schedule rates vary substantially across states (Fomenko and Liu, 2016). This large variation in fee schedule rates leads many policymakers and system stakeholders to express the following concerns about prices for medical care—when medical fee schedules are low and leading to relatively low prices for medical care, workers may have harder time finding desired medical providers. This may lead to problems getting access to timely care, prolonged recovery, lower rates of return to work and worse outcomes for injured workers. This study is a first comprehensive attempt to empirically examine relationship between prices of medical services and the outcomes that workers experience after a work-related injury. This study combines multiple sources of data on medical prices and outcomes of injured workers. We use medical billing information from Truven MarketScan® to determine prices for office visits for group health payors, and information from Workers Compensation Research Institute DBE databate to determine prices for office visits for workers’ compensation payors at MSA level. This helps describe an economically relevant variation in group health and workers’ compensation medical prices across different areas—when workers compensation prices are lower than prices paid by other payers, these providers may have fewer incentives to provide care to injured workers and instead focus their practice on patients with more generous insurance coverage. We then link this information to more than 6,600 surveys of injured workers collected by WCRI for injuries covering 2010 through 2013 across 14 states. We use these surveys to examine the relationship between medical prices and the outcomes including workers’ speed and sustainability of return to work, reported problems getting desired medical provider and desired medical care, and recovery of health and functioning. We supplement this analysis with the information from administrative records on the time to first non-emergency visit, mix of services provided (e.g. conservative versus specialty invasive care) and duration of disability payments. We find that that workers’ compensation prices are strongly associated with measures of access to care and nature of care received. Workers in areas with lower prices were more likely to report “big problems” getting the primary provider that they wanted, had longer time to first non-emergency office visit for evaluation and management services, higher likelihood of a surgery, and smaller in the number office visits. At the same time, we find that prices were not strongly related to recovery of health and functioning, measures of speed and rates of return to work, and measures of duration of temporary disability benefits.

In 2006, quadrivalent human papillomavirus vaccine (4vHPV) was licensed for use in females and in 2007 the Advisory Committee on Immunization Practices (ACIP) recommended routine use of 4vHPV in females aged 11 or 12 years. The licensure of HPV vaccine for use in males was approved in 2009 and the associated ACIP recommendations was published in 2011. Although HPV vaccines have shown promising results in reducing HPV infections, the uptake of the vaccine has been less than ideal. Financial concerns, such as high vaccine purchase costs and inadequate insurance reimbursements, are commonly cited as one of the key barriers to human papillomavirus (HPV) vaccination among healthcare providers. None of the existing studies has assessed physician payment for providing HPV vaccination services and its potential impact on the uptake of HPV vaccines among adolescents. This study uses the 2013-2014 MarketScan Commercial Claims and Encounters Database (CCAE). The sample includes adolescents 11-17 years continuously enrolled in a non-capitated private insurance plan in each year and excludes those who had received a HPV vaccine prior to each study year. Linear regressions are used to estimate the link between physician payments for a HPV vaccination visit and the probability of adolescents initiating the HPV vaccine series and the probability of adolescents receiving ≥ 2 HPV vaccine doses in the year. The key independent variable is state median payment for a HPV vaccination visit. The payment variable includes all aspects of the provider’s income incurring from providing a HPV vaccination service - insurance reimbursements for the HPV vaccine and vaccine administration and patients’ cost sharing for the visit (deductibles, coinsurance, and copayment). The findings show that physician payments are positively associated with HPV vaccine uptakes. For every $1 increase in payments, the probability of adolescents initiating the HPV vaccine series increased by 0.18 percentage point (an increase of 1.5% from the mean) and the probability of receiving ≥ 2 HPV vaccine

This paper uses state differences in the nurse practitioner (NP) market to evaluate the effects of state laws allowing NPs to prescribe controlled substances on prescription opioid use. I study these effects by merging nationwide data from the Medical Expenditure Panel Survey (MEPS) over 18 years (1996-2013) with data on state laws. I then exploit variation in these laws over time to create a quasi-natural experiment and to estimate the causal impact of NP deregulation on prescription opioid use. I find, relative to patients living in more restrictive states, that patients who live in states with more flexible NP laws reduce their prescription opioid use by 7 percent to 9 percent. I also find that health outcomes either slightly improve or remain unaffected by the enactment of these laws. Taken together, these results indicate that NP deregulation slows the trend in prescription opioid growth while potentially improving patient outcomes. Furthermore, suggestive evidence implies that these effects may be even larger for the least restrictive states, opening the door for future reforms.

Information problems are a defining characteristic of health care markets. Despite their importance, there is little direct evidence of the impact of information problems between patients and physicians on the quality of treatment. Furthermore, the role of market conditions is not well understood. In this paper, we present the results from a field experiment in the market for dental care. A test patient who does not need treatment is sent to 180 dentists to receive treatment recommendations. In the experiment, we vary two factors: First, the information that the patient signals to the dentist. Second, we vary the perceived socioeconomic status (SES) of the test patient. Furthermore, we construct several measures of market conditions. Our study has two main contributions: First, we investigate physicians' provision of health care services on the level of individual patient-physician interactions. The design allows us to observe for each physician whether she/he provides the appropriate treatment recommendation or an overtreatment recommendation instead of observing only aggregate provision rates. Thus, we can provide direct evidence of overtreatment and thereby have a clean and simple measure of physician quality. Our micro approach allows us to not only observe the overtreatment behavior but also to control for the covariates on the individual level. We find that the patient receives an overtreatment recommendation in more than every fourth visit. A low waiting time for the next possible appointment, indicating excess capacities, is associated with significantly more overtreatment recommendations. Overall, results regarding the role of market conditions can be well explained based on short-term capacity considerations. Furthermore, we observe signicantly less overtreatment recommendations for the patient with higher SES compared to lower SES under standard information. More signalled information however does not signicantly reduce overtreatment.


We present audit study evidence from China on the relative quality of outpatient care provided by physicians employed in civil service posts and those on fixed-term contracts in rural health centers. We use data from unannounced standardized patient (SP) interactions with physicians and within-clinic variation in contract status generated by China’s bianzhi system – a headcount quota system defining personnel assignments in public service units – to address bias arising from patient and doctor sorting. We find that employment on a fixed-term rather than civil service contract is associated with a large and significant improvement in physicians performance, despite physicians employed on fixed-term contracts being paid significantly less and having fewer formal qualifications. The estimated positive effect of fixed-term contracts on quality increases after controlling for observable physician characteristics. We also find direct evidence that these effects are due to increases in physician effort: comparisons between interactions with SPs and clinical vignettes testing knowledge of the same disease cases to the same doctors shows that civil service doctors exhibit

By altering the incentives faced by providers, Medicare can indirectly affect care for the privately insured. Healthcare research on these spillover effects has focused on Medicare Advantage and traditional fee-for-service (FFS) Medicare or used payer mix variation to identify the effects of public insurance schemes, but scarce research assesses the indirect effects of Medicare on the population under age 65. This paper does just that, using a recent Medicare payment shock where some providers (but not all) moved from a traditional FFS payment model to a model that rewarded patient management and cost containment.


Using a unique dataset of medical claims from a large provider network, I estimate the spillover effects to the privately insured of the OCM program. These data are particularly suited to examining these issues as half of the provider practices in the sample participated in the OCM program and half did not. The practices themselves exhibit many features that make them compelling comparators: they are part of the same network and thus face similar input prices and care sensibilities, they have the same technologies by way of electronic medical records and financial management systems, they host several networking conferences throughout the year, and they share some common back-office staff. I use a differences-in-differences methodology to measure the effects of the OCM program on utilization and cost among patients with private insurance. I couple this analysis with an event study to assess the onset effects of the policy. Primary outcomes include changes in expenditures, such as drug-spend and total-spend, and secondary outcomes include changes in utilization of specific services such as regular visits to mitigate unnecessary hospitalizations or


Federal Open Payments data show that since August of 2013 through the end of 2016, US physicians explicitly received approximately $8 billion per year from drug and medical device manufacturers in various forms of research payments, investment interest, industry-sponsored meals, payments for education and training, consulting fees, travel and lodging, and entertainment. Prior studies have confirmed positive associations between these payments and rates of brand-name drug prescriptions (for instance, Yeh et al. 2016). The provider-pharma financial ties may also affect opioid prescribing practices, which have been a considerable focus of recent state regulatory changes. This paper utilizes the federal Open Payments database to investigate the impact of key state laws enacted to regulate opioid prescription practices, on provider-pharma financial ties. State laws of interest include state prescription drug monitoring programs, pill mill laws, and physical examination requirements, among others. This study aims to shed light on economic predictions of how regulations that reduce product demand could reduce promotional activities by industry. More specific to policy, this study examines potential conflicts of interest that could interfere with physicians’ responsibilities to their patients, and to the US public health efforts to combat the opioid epidemic.


Youngho Oh [email protected]

Caitlin Carroll [email protected]

Yonatan Ben-Shalom [email protected]

In Korea, there is a controversy over the imbalance of supply and demand of medical specialists due to their preference of certain type of specialties. To help guide medical specialist workforce policy, we estimate the impact of

Using Korea Health Data from 2001 to 2013 for 26 specialist fields, we employ a panel analysis including fixed effects models and random effects models for estimating occupancy rate model of specialists. The human capital

First, occupancy rate of specialists varies depending on the income of specialists, and it was projected that the rate increases as the income gets lager. Above all, effect of medical specialist income on their occupancy rates was shown to be greater in supporting fields or minor fields of specialization than in that of major. The income elasticity of the medical specialist was between 0.0377 and 0.09152. Second, the existence of medical specialist training subsidy, one of the variables that represent group effects of specific fields of specialization, appeared to affect the occupancy rate of medical specialists. Third, the difficulty of training and medical treatment/consultation, one of the variables that represent the characteristics of different fields of specialization, appears to affect the occupancy rate of medical specialists greatly. Fourth, the increase/decrease ratio of medical specialists, which was categorized as a variable that shows the characteristics of different fields of specialization, appeared to have an impact on the occupancy rate of medical specialists. In general, the occupancy rate decreased by 0.135% point when the quota of medical specialists was raised by

The concentration of application in certain fields of specialization that leads to the imbalance in medical specialist occupancy rate can be partly explained by models based on the medical specialist income, job stability and characteristics of each field, and is partly attributed to unique characteristics of different fields that are not explained by these models. Therefore, it is necessary to advance into the direction of improving the inequity of group effects of different fields. There are policy means to improve the occupancy rate by adding a certain percentage to the health insurance fees or using policy variables to standardize the preference index of fields which have preference index.

In this project, we study how physicians respond to episode-based bundled payment (EBP), a prominent payment reform that pays a case rate for an entire episode of care. Unlike FFS reimbursement, EBP holds physicians responsible for all care within a discrete clinical episode, rewarding physicians not only for efficient use of their own services but also for efficient management of other health care inputs. While EBP programs are expanding, existing research is generally limited to evidence from voluntary demonstration projects, mainly in the Medicare market. We study the impact of EBP under the Arkansas Payment Improvement Initiative (APII), a multi-payer program that requires providers in the state to enter into EBP arrangements for perinatal care. Because of its multi-payer nature and the requirement that providers participate, the program covers the vast majority of births in the state. In a difference-in-differences analysis of commercial claims, we find that perinatal spending decreased by 3.8% overall in Arkansas after the introduction of EBP, compared to surrounding states. We find that the decrease was driven by reduced spending on non-physician health care inputs, specifically the prices paid for inpatient facility care. Reductions in the price of care could reflect referral patterns favoring low price facilities or lower negotiated rates at a given facility; we find preliminary evidence that a change in referral patterns is more likely. Our results are robust to a number of sensitivity tests, including alternate control groups, and we demonstrate that there was no effect among placebo conditions not subject to EBP. We additionally study quality of care under EBP by analyzing changes in screening rates for common perinatal conditions. Overall, we find that EBP was associated with a limited improvement in quality of

Many Americans rely on public health insurance programs for their health care needs. This paper seeks to improve the understanding of provider acceptance rates of two commonly used sources of public health insurance: TRICARE, which serves military personnel and their family members and Medicare, which primarily serves individuals over age 65. We use data from a congressionally mandated survey of civilian physicians to address three main research questions: (1) How does the acceptance rate for new TRICARE Standard patients compare to the acceptance rate for new Medicare patients? (2) What are the similarities and differences in the reasons given by providers for not accepting TRICARE

The TRICARE Standard Survey of Providers was administered to a nationally representative sample of civilian physicians and non-physician mental health providers from 2012 to 2015. Among other questions, providers were asked whether they accept any new patients, new TRICARE Standard patients, or new Medicare patients. The data also include information about the characteristics of the providers, such as their specialization, practice type, and age. In addition, we

Our analysis reveals that both TRICARE Standard and Medicare have high acceptance rates among physicians, with almost three-quarters of physicians accepting both TRICARE and Medicare, only 3% accepting neither TRICARE nor Medicare and 4% not accepting any new patients. However, Medicare is more commonly accepted by physicians than TRICARE; 18% accept Medicare but not TRICARE but only 3% accept TRICARE but not Medicare. Acceptance rates are

Reimbursement and providers’ belief that their specialty is not covered are commonly stated reasons by both physicians and mental health providers for not accepting TRICARE and/or Medicare, and those who have problems with reimbursement tend not to accept either insurance type. Mental health providers are more likely than physicians not to be aware of TRICARE. Finally, providers who accept new TRICARE but not Medicare are more likely than those making

These findings provide insight into challenges and opportunities associated with attempts to increase provider acceptance of public insurance such as TRICARE and Medicare. For example, improving reimbursement is important for all

Lawrence Jin [email protected]

Caitlin Carroll [email protected]

Younsoo Jung [email protected]

Christian Waibel [email protected]

Bogdan Savych [email protected]

This paper provides empirical evidence of hot-hand bias in two novel field settings: dart players' strategic choices, and physicians' decisions during childbirth. The “hot hand” is the notion that a person can enter a state in which her probability of success becomes higher than normal. “Hot-hand bias” refers to an exaggerated belief in the hot hand (whether it exists or not). First, I collect data of professional dart players from the 2016 World Darts Championship. The players are significantly more likely to hit after a successful shot, implying that players have a hot hand. Based on a precise estimate of the hot hand, I calculate the optimal strategy of a profit-maximizing dart player. I find that dart players are much more willing to take risks after a successful shot than what I calculate to be optimal. Second, I utilize 1.3 million hospital admissions for childbirth in New York State over 2010-2015. I find no evidence that physicians have a hot hand when performing obstetrical procedures. In the absence of hot hand, physicians are still 2% more likely to perform a C-section after a previous successful C-section. My empirical model includes physician fixed effects, and a large set of patient conditions that proxy for when a C-section is likely to maximize patient welfare. Across the two settings, robustness checks provide additional evidence consistent with decision-makers having a hot-hand bias. Generalizing the

While a substantial literature has studied the influence of malpractice pressure on physician behavior, existing research has not found that malpractice concerns influence physicians to any great extent. However, these studies generally focus on variation in state laws governing malpractice exposure. What is perhaps more important is the actual experience of being sued. In this project, we test how physicians respond to malpractice lawsuits – both those that are successful and those that are not. Despite the fact that the vast majority of physicians will face a malpractice claim during their career, there is little evidence on how experience with the liability system informs physicians’ assessments of malpractice pressure. We study the impact of malpractice allegations on the labor supply and treatment intensity decisions of Emergency Department (ED) physicians, combining physician-level data on malpractice claims with the universe

We find that physician labor supply decreases sharply after malpractice allegations and that this reduction is persistent over time. We estimate a 10% reduction in physicians’ patient loads overall and find that our results are driven by intensive margin responses, specifically that physicians reduce the number of patients they treat but maintain practice in the state. We further find that physicians stop practicing at the hospital where the alleged negligence occurred, and that part of the decline in total patient volume is driven by physicians leaving high volume hospitals to practice at smaller facilities. Next, we find that physicians increase care intensity among their remaining patients, increasing total charges per patient by about 5% after an allegation. Lastly, we provide suggestive evidence that physicians do not respond optimally to malpractice allegations, showing that physicians respond equally to legitimate claims of negligence as to allegations that are ultimately dismissed. We additionally show that physicians adjust practice patterns equally for all patient types after a malpractice allegation, rather than using information from the allegation to adjust care for

Over the past two decades, there have been two large public health insurance expansions, the State Children’s Health Insurance Program (SCHIP) and the Affordable Care Act (ACA). These insurance programs have significantly increased the number of patients with public health insurance and the demand for medical services, but it is not clear whether providers will supply additional services for newly-insured patients. Because, public health insurance programs provide relatively low reimbursement rates to physicians which discourage physicians to accept publicly-insured patients. This paper focuses on the labor supply response of physicians to these expansions. I use data from the Community Tracking Study (CTS) physician survey (1996-1997, 1998-1999, and 2000-2001) to examine the SCHIP expansion and the 2012-2015 American Community Study (ACS) to examine the ACA expansion. In response to the introduction of SCHIP, my estimates suggest that physicians reallocate their total working hours between patient care and non-patient care activities. The size of the impact was greater in areas with high level of physician concentration prior to the expansion. Physicians in high concentration areas tend to decrease time spent on direct patient care, but increase hours on non-direct patient care. In response to the ACA, physicians’ working hours did not increase, but working hours, the probability of being employed, and the probability of being employed increased for registered nurses. This suggests that physicians might utilize other healthcare providers to accommodate increases in demand for medical services after the

Inequality in access to health care may have different underlying reasons. Employing a randomised field experiment, we study the impact of socioeconomic status on health care access varying the patient’s educational level. We find that practice assistants favor patients with a degree over those without a degree in whether they offer an appointment. Physicians, in contrast, favor patients with a degree with respect to response time to the request for an appointment and

Medical prices for workers compensation are set by medical fee schedules, and these fee schedule rates vary substantially across states (Fomenko and Liu, 2016). This large variation in fee schedule rates leads many policymakers and system stakeholders to express the following concerns about prices for medical care—when medical fee schedules are low and leading to relatively low prices for medical care, workers may have harder time finding desired medical

This study is a first comprehensive attempt to empirically examine relationship between prices of medical services and the outcomes that workers experience after a work-related injury. This study combines multiple sources of data on medical prices and outcomes of injured workers. We use medical billing information from Truven MarketScan® to determine prices for office visits for group health payors, and information from Workers Compensation Research Institute DBE databate to determine prices for office visits for workers’ compensation payors at MSA level. This helps describe an economically relevant variation in group health and workers’ compensation medical prices across different areas—when workers compensation prices are lower than prices paid by other payers, these providers may have fewer incentives to provide care to injured workers and instead focus their practice on patients with more generous insurance coverage. We then link this information to more than 6,600 surveys of injured workers collected by WCRI for injuries covering 2010 through 2013 across 14 states. We use these surveys to examine the relationship between medical prices and the outcomes including workers’ speed and sustainability of return to work, reported problems getting desired medical provider and desired medical care, and recovery of health and functioning. We supplement this analysis with the information from administrative records on the time to first non-emergency visit, mix of services provided (e.g. conservative versus specialty invasive care) and duration of disability payments. We find that that workers’ compensation prices are strongly associated with measures of access to care and nature of care received. Workers in areas with lower prices were more likely to report “big problems” getting the primary provider that they wanted, had longer time to first non-emergency office visit for evaluation and management services, higher likelihood of a surgery, and smaller in the number office visits. At the same time, we find that prices were not

Yuping Tsai [email protected]

Morris Hamilton [email protected]

Wanda Mimra [email protected]

Cuiping Schiman [email protected]

Sean Sylvia [email protected]

In 2006, quadrivalent human papillomavirus vaccine (4vHPV) was licensed for use in females and in 2007 the Advisory Committee on Immunization Practices (ACIP) recommended routine use of 4vHPV in females aged 11 or 12 years. The licensure of HPV vaccine for use in males was approved in 2009 and the associated ACIP recommendations was published in 2011. Although HPV vaccines have shown promising results in reducing HPV infections, the uptake of the vaccine has been less than ideal. Financial concerns, such as high vaccine purchase costs and inadequate insurance reimbursements, are commonly cited as one of the key barriers to human papillomavirus (HPV) vaccination among healthcare providers. None of the existing studies has assessed physician payment for providing HPV vaccination services and its potential impact on the uptake of HPV vaccines among adolescents. This study uses the 2013-2014 MarketScan Commercial Claims and Encounters Database (CCAE). The sample includes adolescents 11-17 years continuously enrolled in a non-capitated private insurance plan in each year and excludes those who had received a HPV vaccine prior to each study year. Linear regressions are used to estimate the link between physician payments for a HPV vaccination visit and the probability of adolescents initiating the HPV vaccine series and the probability of adolescents receiving ≥ 2 HPV vaccine doses in the year. The key independent variable is state median payment for a HPV vaccination visit. The payment variable includes all aspects of the provider’s income incurring from providing a HPV vaccination service - insurance reimbursements for the HPV vaccine and vaccine administration and patients’ cost sharing for the visit (deductibles, coinsurance, and copayment). The findings show that physician payments are positively associated with HPV vaccine uptakes. For every $1 increase in payments, the probability of adolescents initiating the HPV vaccine series increased by 0.18 percentage point (an increase of 1.5% from the mean) and the probability of receiving ≥ 2 HPV vaccine

This paper uses state differences in the nurse practitioner (NP) market to evaluate the effects of state laws allowing NPs to prescribe controlled substances on prescription opioid use. I study these effects by merging nationwide data from the Medical Expenditure Panel Survey (MEPS) over 18 years (1996-2013) with data on state laws. I then exploit variation in these laws over time to create a quasi-natural experiment and to estimate the causal impact of NP deregulation on prescription opioid use. I find, relative to patients living in more restrictive states, that patients who live in states with more flexible NP laws reduce their prescription opioid use by 7 percent to 9 percent. I also find that health outcomes either slightly improve or remain unaffected by the enactment of these laws. Taken together, these results indicate that NP deregulation slows the trend in prescription opioid growth while potentially improving patient outcomes.

Information problems are a defining characteristic of health care markets. Despite their importance, there is little direct evidence of the impact of information problems between patients and physicians on the quality of treatment. Furthermore, the role of market conditions is not well understood. In this paper, we present the results from a field experiment in the market for dental care. A test patient who does not need treatment is sent to 180 dentists to receive treatment recommendations. In the experiment, we vary two factors: First, the information that the patient signals to the dentist. Second, we vary the perceived socioeconomic status (SES) of the test patient. Furthermore, we construct several measures of market conditions. Our study has two main contributions: First, we investigate physicians' provision of health care services on the level of individual patient-physician interactions. The design allows us to observe for each physician whether she/he provides the appropriate treatment recommendation or an overtreatment recommendation instead of observing only aggregate provision rates. Thus, we can provide direct evidence of overtreatment and thereby have a clean and simple measure of physician quality. Our micro approach allows us to not only observe the overtreatment behavior but also to control for the covariates on the individual level. We find that the patient receives an overtreatment recommendation in more than every fourth visit. A low waiting time for the next possible appointment, indicating excess capacities, is associated with significantly more overtreatment recommendations. Overall, results regarding the role of market conditions can be well explained based on short-term capacity considerations. Furthermore, we observe signicantly less overtreatment recommendations for the patient with


We present audit study evidence from China on the relative quality of outpatient care provided by physicians employed in civil service posts and those on fixed-term contracts in rural health centers. We use data from unannounced system – a headcount quota system defining personnel assignments in public service units – to address bias

arising from patient and doctor sorting. We find that employment on a fixed-term rather than civil service contract is associated with a large and significant improvement in physicians performance, despite physicians employed on fixed-term contracts being paid significantly less and having fewer formal qualifications. The estimated positive effect of fixed-term contracts on quality increases after controlling for observable physician characteristics. We also find direct evidence that these effects are due to increases in physician effort: comparisons between interactions with SPs and clinical vignettes testing knowledge of the same disease cases to the same doctors shows that civil service doctors exhibit

Brigham Walker [email protected]

Thuy Nguyen [email protected]

By altering the incentives faced by providers, Medicare can indirectly affect care for the privately insured. Healthcare research on these spillover effects has focused on Medicare Advantage and traditional fee-for-service (FFS) Medicare or used payer mix variation to identify the effects of public insurance schemes, but scarce research assesses the indirect effects of Medicare on the population under age 65. This paper does just that, using a recent Medicare payment shock


Using a unique dataset of medical claims from a large provider network, I estimate the spillover effects to the privately insured of the OCM program. These data are particularly suited to examining these issues as half of the provider practices in the sample participated in the OCM program and half did not. The practices themselves exhibit many features that make them compelling comparators: they are part of the same network and thus face similar input prices and care sensibilities, they have the same technologies by way of electronic medical records and financial management systems, they host several networking conferences throughout the year, and they share some common back-office staff. I use a differences-in-differences methodology to measure the effects of the OCM program on utilization and cost among patients with private insurance. I couple this analysis with an event study to assess the onset effects of the policy. Primary outcomes include changes in expenditures, such as drug-spend and total-spend, and secondary outcomes include changes in utilization of specific services such as regular visits to mitigate unnecessary hospitalizations or


Federal Open Payments data show that since August of 2013 through the end of 2016, US physicians explicitly received approximately $8 billion per year from drug and medical device manufacturers in various forms of research payments, investment interest, industry-sponsored meals, payments for education and training, consulting fees, travel and lodging, and entertainment. Prior studies have confirmed positive associations between these payments and rates of brand-name drug prescriptions (for instance, Yeh et al. 2016). The provider-pharma financial ties may also affect opioid prescribing practices, which have been a considerable focus of recent state regulatory changes. This paper utilizes the federal Open Payments database to investigate the impact of key state laws enacted to regulate opioid prescription practices, on provider-pharma financial ties. State laws of interest include state prescription drug monitoring programs, pill mill laws, and physical examination requirements, among others. This study aims to shed light on economic predictions of how regulations that reduce product demand could reduce promotional activities by industry. More specific to policy,


KIHASA Complete

Harvard University Complete

Mathematica Policy Research Priyanka Anand; Eric Schone Complete

Michael Chernew; Sherri Rose; Joe Thompson; A. Mark Fendrick


Harvard University Anupam Jena; David Cutler Complete

University of Iowa Complete

Eidgenössische Technische Hochschule Zurich Harald Stummer; Silvia Angerer Complete

Workers Compensation Research Institute Olesya Fomenko Complete


Abt Associates Complete

ETH Zürich Felix Gottschalk; Christian Waibel Complete

Northwestern University Anuj Gangopadhyaya; Robert Kaestner Complete

University of North Carolina at Chapel Hill Complete

Tulane University Complete

Indiana University Bloomington Kosali Simon Complete


Prescription Drugs Pharmacy Deserts and Medication Adherence

Prescription Drugs

Prescription Drugs

Prescription Drugs

Why is Research in Early-Stage Cancer Research so Low? A Re-assessment of Budish, Roin and Williams (2015)

First Opioid Prescription and Subsequent High-Risk Opioid Use

The Effect of Medical Marijuana Laws on Utilization of Prescribed Opioids and Other Prescription Drugs

Prescription Drugs

Prescription Drugs

Prescription Drugs

How Are Part D Plans Different? Evidence from Randomly Assigned Enrollees

The impact of new drug launches on hospitalization for 106 medical conditions in 15 OECD countries, 2002-2015: a triple-differences analysis

Are prescription drugs and preventive health behavior substitites ? Evidence from Medicare Part D

Prescription Drugs

Prescription Drugs

Prescription Drugs Fighting the opioid crisis: Lessons from Kentucky

Prices for and Spending on Specialty Drugs in Medicare Part D and Medicaid

Prescription Opioid Misuse and Labor Supply: Does the Level of Misuse Matter?

Prescription Drugs

Prescription Drugs Crime and Opioid Prescriptions

Prescription Drugs

Opioid Prescription Dispensing Restrictions and Unintended Consequences

Contagious Physician Prescribing: Opioid Physician Prescription Spill-overs and Labor Market Effects of Opioid Analgesics

Prescription Drugs

Prescription Drugs

Prescription Drugs

Patent Challenges and Follow-on Pharmaceutical Innovation

The Prescription Drug Paradox: Pipeline Pressure and Rising Prices

Nurse Practitioners’ Scope of Practice and Prescription Drug Abuse

Abstract

Poor medication adherence rates cost the US health care system roughly $290 billion per year, a sizable amount, especially given that poor medication adherence is largely preventable. Despite overwhelming evidence that prescription drugs are a highly effective form of treatment, medication adherence rates in the US are very low, likely in part because there are significant personal costs for filling prescriptions - both monetary and non-monetary. In this paper, I study the distance costs required to travel to the pharmacy, examining the extent to which access to pharmacies influences medication adherence. I use straightforward intent-to-treat measures of adherence: the total numbers of (1) pills dispensed, (2) patients, (3) new patients, and (4) pharmacy claims filed. Using All-Payer Pharmacy Claims data from Oregon, I first take an event-study approach around two different types of events. The first is local pharmacy openings and closings, the second is variation in insurance network status of a major pharmacy chain (Walgreens) in and out of the network of a major pharmacy benefits manager (Express Scripts) - essentially closing and re-opening Walgreens to Express Scripts enrollees. I find that pharmacy openings cause a 2% increase in the measures of adherence from a stable pre-trend for local patients, and that removing local Walgreens from the Express Script’s network causes a 5% decrease in the same outcome variables, for similarly stable pre-trends. I find the effect magnitudes decrease as patient distance-to-affected-pharmacy increases. I examine heterogeneous effects by drug type and insurance type, as well as by neighborhood characteristics. Notably, I find large effects for chronic drugs such as heart medication, cholesterol reducers, and beta blockers. This is a substantial result since it highlights the binding effect of distance costs on adherence to important medications. I then take these effects and combine them with observable characteristics of the zip code in a machine learning framework to extrapolate counterfactual effects to a national level. I estimate that counterfactually opening an additional pharmacy in each zip code across the US would lead to a median increase of roughly 3% in each of the outcome variables, but with significant heterogeneity across zip codes. These results highlight potential "pharmacy deserts" across the country, a fact important to policy makers aiming to reduce health care costs by improving medication adherence. Finally, I use a discrete-choice framework, combined with copay variation within drug/insurance-type bins but across pharmacies, to estimate patient willingness-to-pay for a one mile reduction in distance to nearest pharmacy. I find that, on average, patients are willing to pay $1 more in copays for a one mile distance reduction. The willingness-to-pay estimates are greater in rural areas with lower pharmacy access, and smaller in urban areas with higher pharmacy access. Combining these results with the effects of pharmacy openings allows me to estimate the social versus private value of increased medication adherence, and to approximate the societal value of additional pharmacies.

According to Budish, Roin and Williams (2015), R&D investment for early-stage cancer prescription drugs is low entirely because of missing financial incentives, and surrogates can overcome this distortion. We argue that technological barriers can also explain such low R&D activity. We summarize medical literature describing these barriers and augment their data to simultaneously assess the role of financial incentives versus technological barriers. Our results show that technological barriers suppress research in early-stage solid cancer. Surrogates may moderately increase R&D activities but have potential drawbacks and cannot address technological barriers.

Background: Recent guidelines and policies attach importance to containing the initial exposure to prescription opioids. Little is known about how prescribing decisions regarding the first opioid prescription are associated with high-risk opioid use in the long-term that puts patients at heightened risk of opioid misuse and overdose. Objective: To examine associations between features of the first opioid prescription and high-risk opioid use in the 18 months following the month of the first prescription. Methods: Retrospective cohort study. Two populations of interest are privately insured adults aged 18-64 and Medicare Advantage enrollees 65 or older who filled a first opioid prescription between 07/01/2011 and 06/30/2013. We used 2011-2014 data from a large, national commercial insurance claims database to identify individuals naive to opioid therapy (determined based on a six-month look-back with no opioids) and follow them for 18 months after the first opioid prescription. We considered three salient features of the first opioid prescription: long- vs. short- acting opioid formulation, daily dosage in morphine milligram equivalents (MMEs), and days of supply. High-risk opioid use in the long term was measured for each of six quarters (3-month intervals) with some opioid use following the first opioid prescription. Measures included 1) having opioid prescriptions overlapping for seven days or more, 2) having opioid and benzodiazepine prescriptions overlapping for seven days or more, 3) having filled opioid prescriptions from 3 or more prescribers, and 4) having a daily average MMEs exceeding 120. A secondary analysis examined how features of the first opioid prescription were associated with some use of opioid in each of the six quarters following the first prescription. Results: Our samples included 196,375 non-elderly, privately insured patients and 63,419 elderly, Medicare Advantage patients. All three features of the first prescription strongly predicted high-risk use. For example, for privately insured patients, receiving a long- (vs. short- ) acting opioid was associated with a 16.9-percentage-point increase (95% CI, 14.0-19.5), a daily MME of 50 or more (vs. less than 30) was associated with a 12.5-percentage-point increase (95% CI, 12.1-12.9), and a more-than 7-day supply (vs. 3 or fewer days) was associated with a 4.8-percentage-point increase (95% CI, 4.5-5.2), in the probability of having a daily dosage of 120 MMEs or more in the long term. Results for the Medicare Advantage patients were similar. Conclusions: Long-acting formulation, high daily dosage, and long duration (exceeding 7 days) of the first opioid prescription were associated with increased high-risk use of opioids in the long term. Implications for Policy and Practice: Our findings provide support to policies, clinical guidelines, and health care system interventions that direct first opioid prescriptions away from long-acting formulation, high daily dosage, and long duration. Caution needs to be exercised to guard against applying these policies and guidelines to all opioid prescribing (e.g., to long-term opioid users). Additional policies and considerations are needed to counteract barriers to continued opioid therapy after the initial prescription for patients in need.

More than half of the US population lives in a state that has adopted medical marijuana laws (MMLs). Studies show that most medical marijuana patients use marijuana for managing their pain with the overwhelming majority of them preferring it to opioids. Despite ongoing pro-marijuana policies and the growing trend of public acceptance, the evidence on how people change their prescription use due to the availability of marijuana as an alternative treatment is limited. Using the variations across state MMLs between 1996 and 2014 of Medical Expenditure Panel Survey (MEPS) this paper estimates the effects of MMLs on prescription drug utilization, with a focus on opioids. I find that MMLs lead to a $2.47 decrease in per person prescribed opioid spending among young adults (ages 18-39) over a year. Most of this decrease results from the intensive margin of use and MML states that allow home cultivation experience even larger decreases. Furthermore, the decreasing effects are persistent over time and they get stronger following the years of implementation. MMLs also decrease the number of opioid pill use among young adults. I do not find any discernible impact on older populations' opioid utilization. I then investigate the effects on other prescriptions for which marijuana can be a potential substitute and find the allowance of dispensaries is generally associated with decreases, although the effects depend on the the type of MML, the margin of use and age.

While substantial focus has been paid to the financial characteristics of prescription drug plans, far less attention has been given to the impact of non-financial features, including formulary composition, utilization restrictions, and pharmacy network composition. This is an important gap in the literature, given the substantial non-financial differences that exist across drug plans, and given broader questions that these non-financial characteristics could tie into. To examine the impact of non-financial features independently from financial characteristics of drug plans, we study how drug and non-drug spending and utilization differs across low-income exemption beneficiaries randomly auto-assigned to different Medicare Part D plans. These beneficiaries are not subject to cost sharing, eliminating the possibility that differences across Part D plans are driven by differences in deductibles, copayments, and other financial incentives. We also leverage a new data source indicating which low-income beneficiaries failed to actively choose a Part D plan and were randomly auto-assigned to a plan, allowing us to focus on this population and eliminating the possibility that differences across Part D plans are driven by endogenous sorting of beneficiaries to plans. We show that Part D plan assignment has important effects on drug spending and utilization, with formularies influencing the drugs beneficiaries take. Perhaps more importantly, we also show that formularies affect non-drug spending, providing new evidence for complementarities between these two types of healthcare utilization. Finally, we study the effects of Part D plan assignment on mortality and other outcomes and relate cross-plan differences in those outcomes to cross-plan differences in drug and non-drug utilization. In addition to its immediate implications, this study could be of broader academic and policy interest, by providing insights on important issues such as drug-driven medical offsets, behavioral determinants of plan choice, medication adherence, opioid use, and drug substitutability.

There are two types of prescription drug cost offsets. The first type of cost offset—from prescription drug use—is primarily about the effect of changes in drug quantity (e.g. due to changes in out-of-pocket drug costs) on other medical costs. Previous studies indicate that the cost offsets from prescription drug use may slightly exceed the cost of the drugs themselves. The second type of cost offset—the cost offset from prescription drug innovation—is primarily about the effect of prescription drug quality on other medical costs. Two previous studies (of a single disease or a single country) found that pharmaceutical innovation reduced hospitalization, and that the reduction in hospital cost from the use of newer drugs was considerably greater than the innovation-induced increase in pharmaceutical expenditure. In this study, I reexamine the impact that pharmaceutical innovation has had on hospitalization, using a “triple-differences,” or difference-in-difference-in-differences, research design: I estimate the impact that new drug launches had on hospitalization for 106 medical conditions in 15 OECD countries during the period 2002-2015. This design enables me to control for all determinants of hospitalization growth that are invariant across diseases within a country, and for all determinants that are invariant across countries within a disease. The relative number of new drugs launched for different diseases varies across countries. Hospitalization is not significantly related to the number of drugs launched 0-3 years earlier; this is not surprising since it takes 8-10 years for a drug to attain its peak level of utilization. However, both the number of hospital discharges and the number of hospital days are significantly inversely related to the number of drugs launched 6-15 years earlier. The estimates indicate that one additional drug launch reduces the number of hospital days 6-15 years later by about 4%. The shapes of the drug-age/drug utilization profiles and of the drug-age/hospital-days-effect profiles are very similar. The estimated reduction in 2015 hospital expenditure attributable to drugs launched during 1996-2009 is 2.5 times as large as the increase in 2015 drug expenditure attributable to those drugs, which implies that pharmaceutical innovation reduced overall medical expenditure. Cost offsets from drug innovation appear to be even larger than cost offsets from drug use.

Abstract

Using the National Health Interview Survey and Medical Expenditure Panel Survey, I examine whether prescription drug use substitutes investment in preventive health behaviors. To identify their causal relationship, I estimate the differences in the regression discontinuity of prescription drug uses and preventive health behavior at age 65 before and after the implementation of Medicare Part D. The resulting estimates indicate that the implementation of Medicare Part D increases prescription drug insurance coverage and prescription drug use. This led to a 10 percentage point reduction in the probability of engaging in moderate physical activity at the extensive margin, a 6.47% reduction in the probability of having healthy-weight and a 9.56 percentage point increases in the probability of being overweight. The effects on moderate physical activity at the intensive margin, vigorous strength physical activity--both at the intensive and extensive margin--cigarette smoking, body mass index and obesity are not statistically significant. The physical exercise, healthy-weight and overweight effect of prescription drug use is stronger among sub-group of individuals that experienced greater prescription drug coverage gain due to Medicare Part D . Key Words: Prescription Drug, Preventive Health Behavior, Medicare Part D, Regression Discontinuity

There has been considerable interest in the prices paid for specialty drugs and the amount of spending on such drugs. In particular, concerns have been raised regarding the effects of such drugs on federal spending and the out-of-pocket costs incurred by some beneficiaries. In this paper, we analyze trends in specialty drug pricing and spending (net of manufacturer rebates and discounts) in Medicare Part D and Medicaid over the 2010-2015 period. We also examine changes in out-of-pocket costs for Medicare Part D beneficiaries. We adopt the definition of specialty drugs developed by IMS Health. Under that definition, specialty drugs treat a chronic, complex, or rare disease and have at least four of seven additional characteristics (such as costing at least $6,000 per year and requiring special handling in the supply chain). We use beneficiary-level claims data for Medicare Part D to estimate total spending at retail prices and the number of units and prescriptions dispensed over the 2010-2015 period by drug. We also have data on total rebates and discounts paid by drug manufacturers under Part D by drug during that period, which we use to estimate net prices and spending for specialty drugs. For Medicaid, we use data on total spending at retail prices, the number of prescriptions and units dispensed, and the statutory rebate amounts by drug over the 2010-2015 period. We then combined those data with Redbook data (by NDC code), which has drug product characteristics as well as with a list of specialty drugs on the market in 2015 provided by IMS Health. Our results indicate that growth in spending on specialty drugs was a key driver of spending growth in both Medicare Part D and Medicaid’s outpatient drug benefit from 2010 to 2015. On a per capita basis, spending on specialty drugs increased substantially in Part D and more modestly in Medicaid, while spending on traditional drugs declined in both programs. Spending per capita for both specialty drugs and traditional drugs was higher in Medicare Part D than in Medicaid—and such spending grew at a faster rate under Medicare Part D—because of differences in how prices are determined in the two programs and differences in the mix of drugs used by beneficiaries in the two programs.

Retail prices paid for specialty drugs are similar under Part D and Medicaid. However for 50 top selling brand-name specialty drugs, net prices paid by Medicaid were much lower than those in Medicare Part D on average because the statutory rebates under Medicaid are much greater than the rebates Part D plans are able to negotiate from manufacturers. In addition, average net prices paid for brand-name specialty drugs increased much more quickly in Medicare Part D than in Medicaid over the 2010 to 2015 period.

We also use the Medicare Part D claims data to examine trends in the mean and distribution of beneficiaries’ out-of-pocket costs for specialty drugs over the 2010-2015 period.

Over the past two decades, misuse of prescription painkillers in the U.S. particularly opioids increased rapidly, to the point that in 2011 the CDC labeled the problem an "epidemic". Research investigated the negative health effects of prescription opioid misuse (POM), but little research examined the associations between POM and labor supply. This study investigates the associations between POM and labor force participation and, conditional on being in the labor force, employment. This study merged cross-sectional data from the 2010-2014 National Survey on Drug Use and Health for individuals aged 26 to 64 years old. The analysis examined any misuse, ‘infrequent misuse’ (1 – 199 days), and ‘frequent misuse’ (200 – 365 days) over past year. Prevalence estimates for labor force participation found that 84.8 % of individuals with any opioid misuse were in labor force compared to 81.2% of individuals with no misuse. Roughly, 84.1% of infrequent misusers were in labor force compared to 64.7% of frequent misusers. Unadjusted logistic regressions indicate a positive association between any misuse and labor force participation, but adjusted logistic regressions indicate no association. Both adjusted and unadjusted logistic results found frequent misusers to be less likely in the labor force compared to infrequent misusers. Prevalence estimates for employment indicate that 89.6% of individuals with opioid misuse were employed relative to 92.5% of individuals with no misuse. Roughly, 89.7% of infrequent misusers were employed compared to 81.8% of frequent misusers. Unadjusted and adjusted logistic results indicate a negative association between any misuse and employment, and no significant differences between infrequent and frequent misusers with respect to employment. The associations between POM and labor force participation and employment are not consistent throughout the opioid misuse spectrum. In the unadjusted logistic regression, labor force participation was positively associated with past year POM while frequent misusers were less likely to be in the labor force relative to infrequent misusers. Moreover, employment was negatively associated with past year POM whereas there were no differences in odds of employment between frequent and infrequent misusers. Focusing now on the multivariable analysis, the results indicate no association between past year POM and labor force participation; frequent misusers were less likely to be in the labor force compared to infrequent misusers. On the other hand, we found a negative association between past year POM and employment while there were no statistically significant differences between types of misusers and employment. Understanding the labor supply behavior of POMs is vital in formulating treatment and policy proposals that build upon work incentives. This study is among the first to use standard definitions of work status to enhance our understanding of the associations between POM and labor force participation and employment. These findings should be interpreted with caution. The data is self-reported with general validity and reliability issues. The surveys are cross-sectional, and thus, not appropriate to make causal inferences. Finally, we did not control for the potential endogeneity of POM in the labor supply specifications to avoid biases that might be associated with using weak instruments.

This study investigates the impacts of state-mandated emergency rules for opioid prescribing, in response to the ongoing opioid epidemic. We exploit the implementation of an emergency prescribing directive that came into immediate effect in Kentucky starting in July 2012, using the neighboring state of Indiana as a control. The analysis uses novel data from the prescription drug monitoring programs from Kentucky and Indiana that include the universe of all opioid scripts dispensed within the respective states between January 2012-November 2013. Individual prescriber- and pharmacy-level difference-in-difference analyses show that Kentucky's emergency rules significanlty lowered the number of patients being prescribed opioids, the number of opioid scripts written, and the number of refills and days of opioid supply authorized. The prescribing restrictions also affected patient sub-populations in heterogeneous ways. Opioid prescribing declined most sharply for younger patients, for men, and for those on high opioid doses. We further found evidence that after the implementation of the policy, prescribers began to transition patients to lower opioid doses and to opioid drugs like Buprenorphine, which are used for medication assisted treatment of opioid-use-disorders. Our results show that state-mandated opioid prescribing guidelines could provide significant supply side controls on high-risk opioid prescribing.

As the opioid epidemic has evolved overtime to refer not only to overdose deaths from prescription opioids, but also illicit opioids (ex. heroin, fentanyl), it is important to critically examine the potential role policies restricting access to prescription opioids had in this transition. This is particularly important in states such as Tennessee which have been hit particularly hard. In response to the epidemic Tennessee enacted laws to modify physician behaviors, paying special attention to pain clinics because of the role they are believed to play in diverting prescription opioids to the illegal market when they develop into a pill mill. Specifically, in 2014 the Tennessee General Assembly passed public chapter 700 and 983, which prohibited pain clinics and providers from directly dispensing opioids expect under specific circumstance. This dispensing restriction may influence their general prescribing habits (i.e. decreasing the number of prescriptions, quantity prescribed, etc.), which could decrease the supply of opioids in Tennessee. While these laws most directly affect the supply of opioids, their ultimate purpose is to address the opioid epidemic more broadly and thus their effect on the supply and demand for opioids must be considered. The purpose of this paper is to utilize patient level data from Tennessee Control Substance Monitoring Database (CSMD) for the years 2013-2016 to evaluate the effect of the 2014 laws on the supply of prescription opioids. An interrupted time series model (ITS) will be used to evaluate the ability of the law to reduce the supply of opioids, which will be measured by two outcomes: the monthly total days of opioids prescribed and the monthly average total morphine milligram equivalent (MME). Since not all patients are equally likely to transition to the illicit market, subgroup analysis or quantile regression will be performed to identify any possible heterogeneous treatment effects across the distribution of the outcomes. However, the CSMD only captures information for the market for legally prescribed opioids which have close substitutes in the illegal market in the form of both diverted prescription opioids, as well as illegal substitutes such as heroin. Thus, any decline in the supply of prescription opioids could be offset by substituting for opioids available through the illegal market, which could undermine the overall effectiveness of these laws. Thus secondary analysis will be conducted using visits to the emergency department for an opioid overdose from the Hospital Discharge Dataset to proxy as a measure for total illicit opioid demand in an ITS model. To allow for any readjustments between legal and illegal markets, the policy indicator will be lagged by one year. Overall, this analysis will provide further insight into the ability of supply side interventions to affect an epidemic that is influenced by legal and illegal markets. Developing policy evaluations that are sensitive to both markets is important given the continuing rise in heroin deaths which suggests the importance substitution to the illegal market plays.

The United States is facing an opioid crisis. With annual overdose deaths in the tens of thousands, the nation faces billions of dollars in lost productivity. Despite a growing body of research on the link between opioid prescriptions and substance use disorders, the possible link between prescriptions and crime remains unexplored. The omission is noteworthy because reducing violent crime is an important health policy target. The criminology literature has established a positive correlation between drug usage and crime rates. According to one theory, addiction and substance use increase the risk of criminal behavior through several channels, including the financial pressure to support habitual drug use and the incentive of illegal suppliers to control markets through violence. Recent studies suggest opioid prescriptions contribute to addiction, and if addiction leads to crime, then changes in prescribing behavior should effect changes in crime rates. Using novel data from California over the years 2011-2017, this study is the first to test whether increases in opioid prescriptions lead to greater crime rates. The California data are especially useful for two reasons. First, the state collects detailed data on opioid prescription counts, which are made available by the state's Prescription Drug Monitoring Program (PDMP). Second, the state recently experienced an exogenous shock to criminal populations, providing a natural experiment. In May 2011, the US Supreme Court ruled the overcrowding of California prisons and jails to be unconstitutional (Brown v. Plata 2011), and the state was required to reduce its population of inmates over the next two years. The variation is statistically valuable because opioid prescriptions may be related to crime for certain subpopulations but not others. In particular, prescriptions may lead to more crime in locations with more criminal offenders. The unexpected release of inmates helps identify any relationship because, without the variation, other factors might explain any observed correlation between prescription counts and offender populations. Data on prison and jail admissions and releases from the California Department of Corrections and Rehabilitation (CDCR) and the Board of State and Community Corrections (BSCC) allow for a test of any interaction between opioid prescriptions and the in/outflow of offender populations in a community. The study will employ panel data regression methods. The county is the unit of observation, and the time dimension is monthly. Crime data are taken from the FBI's Uniform Crime Reports. Opioid prescription counts, broken down by sex, crude age group, and drug Schedule classification, are provided by California's PDMP (known by the acronym CURES). Prison and jail populations are provided by California's CDCR and BSCC. Socioeconomic control variables include population demographics from the Census Bureau and unemployment rates from the Bureau of Labor Statistics. After constructing the panel, the study will test whether lagged prescriptions in a county predict future crime, along with any possible interaction effects among offender populations. Should the results find a positive relationship between opioid prescriptions and crime, they should also contribute to our understanding of the magnitude of the current opioid crisis.

In the United States, 64,000 people were killed by drug overdosis in 2016. This is up 22 percent from the 2015 level of 52,404 deaths, and have prompted policy makers to declare the opioid epidemic a national emergency. Previous literature provide evidence that physicians matter for the treatment choices of their patients, implying that policies targeting this particular group can limit the current opioid epidemic. However, a thorough understanding of the mechanisms of physician prescribing behavior is needed to grasp the full potential of such strategies. In this paper I contribute to the literature on physician behavior and the opioid epidemic along several important dimensions. Using Danish registry data on the full population linking providers and patients, I document that the opioid prescribing behavior of physicians is dynamic and mutable in nature, and constitutes an important determinant of opioid treatment intensity of individuals. Specifically, I provide evidence that opioid prescribing leniency is partly the result of a diffusion process across physician clinics, spreading from high leniency areas, and effectively constituting a roll-out of opioid leniency across the country. To estimate the diffusion process, I adopt a peer effects framework. I construct physician specific networks and utilize clinic closures due to physician retirement to obtain within-physician variation in composition of peers over time. This allows me to estimate robust spill-over effects. The estimated spill-over effects are non-negligible: an increase in leniency of one standard deviation by a random peer leads to an increase of 5% of a standard deviation in prescribing leniency. I estimate impacts on labor market outcomes of opioid consumption instrumenting individual opioid consumption leveraging the within physician variation in exposure to the diffusion process. Estimating the labor market effects of opioid consumption is important as it constitutes an understudied branch of costs to the increased opioid use worldwide. I estimate significant negative impacts of opioid usage on labor market income: a one standard deviation increase in opioid use results in a reduction of 7.8 percentage points in the labor market income percentile rank. Moreover, I estimate that opioid usage, despite its intent, actually causes increases in short-term disability: a one percentage increase in opioid use results in an increase of 0.16 percentage point’s short-term disability. The results have wide implications for policy makers, as they add to the understanding of the dynamics of the contagious nature of the opioid epidemic and extend the cost side of opioid usage beyond effects on health and health care expenditures. Furthermore, the results suggest that the effect of policies targeting individual physicians are magnified, as local reductions in opioid leniency would diffuse to nearby physicians due to the presence of spill-overs.

The 1984 Hatch-Waxman Act offers market exclusivities to innovators for their new drugs and follow-on improvements and to first-time generic competitors for challenging innovators' patents. Using data on small-molecule New Chemical Entities (NCEs) from 1985 to 2016, patent challenges they face, and their follow-on innovation, I study the interplay between generic pressure and follow-on innovation. I find that on average follow-on innovation per NCE nearly triples in anticipation of first-time generic competition. However, NCEs approved from 2003 to 2006 show a decline in follow-on indications but an increase in other follow-on innovations relative to NCEs approved from 1985 to 2002. Importantly, follow-on indications reduce the probability of settlement in patent challenges by an average of five and a half percent. This result translates into an average externality value to consumers of about $50 million to $115 million per follow-on indication. By contrast, other follow-on innovations have no effect on the probability of settlement. This study also describes follow-on innovation launching and patenting dynamics that ensue, and it shows that the ratio of patents per follow-on innovation has consistently increased overtime. This suggests caution when using patent counts as a proxy for innovative output.

Economic literature has extensively studied how prices for incumbent firms respond to competition after entry, especially in prescription drug markets following generic entry. However, less attention has been paid to firm behavior prior to entry. We contribute to this gap in the literature by both developing a model of pricing strategies for incumbent drug manufacturers under health insurance and empirically assessing pricing adjustments for incumbent firms, using the insulin market as a natural experiment. We consider the price strategy of incumbent firms among branded, horizontally-differentiated drugs under tiered insurance, which predicts rising prices for incumbent drugs for a range of elasticities as the likelihood of entry increases. Empirically, we exploit exogenous variation in a potential entrant's completion of clinical trials to identify the effect of drug pipeline pressure on the prices of incumbent drugs. Results suggest that pipeline pressure significantly increases the prices of incumbent drugs, and potential biosimilar entry may drive this effect.

As the health care market expanded rapidly over time, provider shortages came to the forefront of policy making. To address this problem, alongside regulations targeting the cost of physician practice the policy makers currently consider alternative delivery methods, and expanding the scope of practice of various types of health care providers. The National Council of State Legislatures (NCSL) reported tracking 827 bills to redefine health providers’ scope of practice in 2012 alone.[1] This paper investigates the impact of regulations allowing nurse practitioners (NPs) to prescribe schedule II prescription drugs such as opioids and stimulants. This topic is of particular importance in light of the recent increase in prescription (Rx) drug abuse. According to 2014 National Survey on Drug Use and Health approximately 54 million people, the equivalent of more than 20 percent of people over 12 years old, have used prescription drugs for nonmedical reasons at least once in their lifetime.[2] Many of them obtained the drugs from a health care provider; for instance, approximately 17.3% of people who abuse prescription painkillers report that a physician prescribed the drugs.[3] Laws allowing NPs to prescribe schedule II drugs effectively expand the number of providers allowed to prescribe these drugs but also change the characteristics of the labor force engaged in drug prescribing. If limiting the ability of nurse practitioners to prescribe drugs enhances service quality, an expansion of prescribing rights of NPs might lead to more prescription drug misuse. In addition, an increase in the number of providers eases access to prescription drugs possibly aggravating of the current drug diversion problem. On the other hand, physician shortages could be the reason behind the high rate of misuse of prescription drugs. Existing physicians might not have enough time and resources to establish a close relationship with their patients or investigate the possibility of fraudulent claims. In this case increasing the number of drug-prescribing providers could alleviate the pressure and reduce illicit access to prescription drugs. Using difference-in-difference and triple-difference econometric models we find that legislation allowing nurse practitioners to prescribe schedule II drugs is associated with less opioid misuse but no statistically significant change in stimulants misuse. The results are robust across several datasets including Treatment Episode Data Set (TEDS) and Mortality Multiple Causes of Death files.

[1] Source: http://www.ncsl.org/research/health/scope-of-practice-overview.aspx accessed 1/21/2017 [2] Source: https://www.samhsa.gov/data/sites/default/files/NSDUH-DetTabs2014/NSDUH-DetTabs2014.pdf, accessed 1/20, 2017 [3] Substance Abuse and Mental Health Services Administration. Results from the 2010 National Survey on Drug Use and Health: volume 1: summary of national findings.

Abstract

Poor medication adherence rates cost the US health care system roughly $290 billion per year, a sizable amount, especially given that poor medication adherence is largely preventable. Despite overwhelming evidence that prescription drugs are a highly effective form of treatment, medication adherence rates in the US are very low, likely in part because there are significant personal costs for filling prescriptions - both monetary and non-monetary. In this paper, I study the distance costs required to travel to the pharmacy, examining the extent to which access to pharmacies influences medication adherence. I use straightforward intent-to-treat measures of adherence: the total numbers of (1) pills dispensed, (2) patients, (3) new patients, and (4) pharmacy claims filed. Using All-Payer Pharmacy Claims data from Oregon, I first take an event-study approach around two different types of events. The first is local pharmacy openings and closings, the second is variation in insurance network status of a major pharmacy chain (Walgreens) in and out of the network of a major pharmacy benefits manager (Express Scripts) - essentially closing and re-opening Walgreens to Express Scripts enrollees. I find that pharmacy openings cause a 2% increase in the measures of adherence from a stable pre-trend for local patients, and that removing local Walgreens from the Express Script’s network causes a 5% decrease in the same outcome variables, for similarly stable pre-trends. I find the effect magnitudes decrease as patient distance-to-affected-pharmacy increases. I examine heterogeneous effects by drug type and insurance type, as well as by neighborhood characteristics. Notably, I find large effects for chronic drugs such as heart medication, cholesterol reducers, and beta blockers. This is a substantial result since it highlights the binding effect of distance costs on adherence

I then take these effects and combine them with observable characteristics of the zip code in a machine learning framework to extrapolate counterfactual effects to a national level. I estimate that counterfactually opening an additional pharmacy in each zip code across the US would lead to a median increase of roughly 3% in each of the outcome variables, but with significant heterogeneity across zip codes. These results highlight potential "pharmacy deserts" across the country, a fact important to policy makers aiming to reduce health care costs by improving medication adherence. Finally, I use a discrete-choice framework, combined with copay variation within drug/insurance-type bins but across pharmacies, to estimate patient willingness-to-pay for a one mile reduction in distance to nearest pharmacy. I find that, on average, patients are willing to pay $1 more in copays for a one mile distance reduction. The willingness-to-pay estimates are greater in rural areas with lower pharmacy access, and smaller in urban areas with higher pharmacy access. Combining these results with the effects of pharmacy openings allows me to estimate the social versus private value of increased medication adherence, and to approximate the societal value of additional pharmacies.


: Recent guidelines and policies attach importance to containing the initial exposure to prescription opioids. Little is known about how prescribing decisions regarding the first opioid prescription are associated with high-risk

: To examine associations between features of the first opioid prescription and high-risk opioid use in the 18 months following the month of the first prescription. : Retrospective cohort study. Two populations of interest are privately insured adults aged 18-64 and Medicare Advantage enrollees 65 or older who filled a first opioid prescription between 07/01/2011 and 06/30/2013. We used

2011-2014 data from a large, national commercial insurance claims database to identify individuals naive to opioid therapy (determined based on a six-month look-back with no opioids) and follow them for 18 months after the first opioid prescription. We considered three salient features of the first opioid prescription: long- vs. short- acting opioid formulation, daily dosage in morphine milligram equivalents (MMEs), and days of supply. High-risk opioid use in the long term was measured for each of six quarters (3-month intervals) with some opioid use following the first opioid prescription. Measures included 1) having opioid prescriptions overlapping for seven days or more, 2) having opioid and benzodiazepine prescriptions overlapping for seven days or more, 3) having filled opioid prescriptions from 3 or more prescribers, and 4) having a daily average MMEs exceeding 120. A secondary analysis examined how features of the first opioid prescription were associated with some use of opioid in each of the six quarters following the first prescription.

: Our samples included 196,375 non-elderly, privately insured patients and 63,419 elderly, Medicare Advantage patients. All three features of the first prescription strongly predicted high-risk use. For example, for privately insured patients, receiving a long- (vs. short- ) acting opioid was associated with a 16.9-percentage-point increase (95% CI, 14.0-19.5), a daily MME of 50 or more (vs. less than 30) was associated with a 12.5-percentage-point increase (95% CI, 12.1-12.9), and a more-than 7-day supply (vs. 3 or fewer days) was associated with a 4.8-percentage-point increase (95% CI, 4.5-5.2), in the probability of having a daily dosage of 120 MMEs or more in the long term. Results for the Medicare

: Long-acting formulation, high daily dosage, and long duration (exceeding 7 days) of the first opioid prescription were associated with increased high-risk use of opioids in the long term. : Our findings provide support to policies, clinical guidelines, and health care system interventions that direct first opioid prescriptions away from long-acting formulation, high daily dosage, and long

duration. Caution needs to be exercised to guard against applying these policies and guidelines to all opioid prescribing (e.g., to long-term opioid users). Additional policies and considerations are needed to counteract barriers to continued

More than half of the US population lives in a state that has adopted medical marijuana laws (MMLs). Studies show that most medical marijuana patients use marijuana for managing their pain with the overwhelming majority of them preferring it to opioids. Despite ongoing pro-marijuana policies and the growing trend of public acceptance, the evidence on how people change their prescription use due to the availability of marijuana as an alternative treatment is limited. Using the variations across state MMLs between 1996 and 2014 of Medical Expenditure Panel Survey (MEPS) this paper estimates the effects of MMLs on prescription drug utilization, with a focus on opioids. I find that MMLs lead to a $2.47 decrease in per person prescribed opioid spending among young adults (ages 18-39) over a year. Most of this decrease results from the intensive margin of use and MML states that allow home cultivation experience even larger decreases. Furthermore, the decreasing effects are persistent over time and they get stronger following the years of implementation. MMLs also decrease the number of opioid pill use among young adults. I do not find any discernible impact on older populations' opioid utilization. I then investigate the effects on other prescriptions for which marijuana can be a potential substitute and find the allowance of dispensaries is generally associated with decreases, although

While substantial focus has been paid to the financial characteristics of prescription drug plans, far less attention has been given to the impact of non-financial features, including formulary composition, utilization restrictions, and pharmacy network composition. This is an important gap in the literature, given the substantial non-financial differences that exist across drug plans, and given broader questions that these non-financial characteristics could tie into. To examine the impact of non-financial features independently from financial characteristics of drug plans, we study how drug and non-drug spending and utilization differs across low-income exemption beneficiaries randomly auto-assigned to different Medicare Part D plans. These beneficiaries are not subject to cost sharing, eliminating the possibility that differences across Part D plans are driven by differences in deductibles, copayments, and other financial incentives. We also leverage a new data source indicating which low-income beneficiaries failed to actively choose a Part D plan and were randomly auto-assigned to a plan, allowing us to focus on this population and eliminating the possibility that differences across Part D plans are driven by endogenous sorting of beneficiaries to plans. We show that Part D plan assignment has important effects on drug spending and utilization, with formularies influencing the drugs beneficiaries take. Perhaps more importantly, we also show that formularies affect non-drug spending, providing new evidence for complementarities between these two types of healthcare utilization. Finally, we study the effects of Part D plan assignment on mortality and other outcomes and relate cross-plan differences in those outcomes

In addition to its immediate implications, this study could be of broader academic and policy interest, by providing insights on important issues such as drug-driven medical offsets, behavioral determinants of plan choice, medication

There are two types of prescription drug cost offsets. The first type of cost offset—from prescription drug use—is primarily about the effect of changes in drug quantity (e.g. due to changes in out-of-pocket drug costs) on other medical costs. Previous studies indicate that the cost offsets from prescription drug use may slightly exceed the cost of the drugs themselves.

—is primarily about the effect of prescription drug quality on other medical costs. Two previous studies (of a single disease or a single country) found that pharmaceutical innovation reduced hospitalization, and that the reduction in hospital cost from the use of newer drugs was considerably greater than the innovation-induced increase in pharmaceutical expenditure. In this study, I reexamine the impact that pharmaceutical innovation has had on hospitalization, using a “triple-differences,” or difference-in-difference-in-differences, research design: I estimate the impact that new drug launches had on hospitalization for 106 medical conditions in 15 OECD countries during the period 2002-2015. This design enables me to control for all determinants of hospitalization growth that are invariant across diseases within a country, and for all determinants that are invariant across countries within a disease. The relative number of new drugs launched for different diseases varies across countries. Hospitalization is not significantly related to the number of drugs launched 0-3 years earlier; this is not surprising since it takes 8-10 years for a drug to attain its peak level of utilization. However, both the number of hospital discharges and the number of hospital days are significantly inversely related to the number of drugs launched 6-15 years earlier. The estimates indicate that one additional drug launch reduces the number of hospital days 6-15 years later by about 4%. The shapes of the drug-age/drug utilization profiles and of the drug-age/hospital-days-effect profiles are very similar. The estimated reduction in 2015 hospital expenditure attributable to drugs launched during 1996-2009 is 2.5 times as large as the increase in 2015 drug expenditure attributable to those drugs, which implies that pharmaceutical

Using the National Health Interview Survey and Medical Expenditure Panel Survey, I examine whether prescription drug use substitutes investment in preventive health behaviors. To identify their causal relationship, I estimate the differences in the regression discontinuity of prescription drug uses and preventive health behavior at age 65 before and after the implementation of Medicare Part D. The resulting estimates indicate that the implementation of Medicare

10 percentage point reduction in the probability of engaging in moderate physical activity at the extensive margin, a 6.47% reduction in the probability of having healthy-weight and a 9.56 percentage point increases in the probability of being overweight. The effects on moderate physical activity at the intensive margin, vigorous strength physical activity--both at the intensive and extensive margin--cigarette smoking, body mass index and obesity are not statistically significant. The physical exercise, healthy-weight and overweight effect of prescription drug use is stronger among sub-group of individuals that

Key Words: Prescription Drug, Preventive Health Behavior, Medicare Part D, Regression Discontinuity

There has been considerable interest in the prices paid for specialty drugs and the amount of spending on such drugs. In particular, concerns have been raised regarding the effects of such drugs on federal spending and the out-of-pocket costs incurred by some beneficiaries. In this paper, we analyze trends in specialty drug pricing and spending (net of manufacturer rebates and discounts) in Medicare Part D and Medicaid over the 2010-2015 period. We also examine

We adopt the definition of specialty drugs developed by IMS Health. Under that definition, specialty drugs treat a chronic, complex, or rare disease and have at least four of seven additional characteristics (such as costing at least $6,000 per year and requiring special handling in the supply chain). We use beneficiary-level claims data for Medicare Part D to estimate total spending at retail prices and the number of units and prescriptions dispensed over the 2010-2015 period by drug. We also have data on total rebates and discounts paid by drug manufacturers under Part D by drug during that period, which we use to estimate net prices and spending for specialty drugs. For Medicaid, we use data on total spending at retail prices, the number of prescriptions and units dispensed, and the statutory rebate amounts by drug over the 2010-2015 period. We then combined those data with Redbook data (by NDC code), which has drug product characteristics as well as with a list of specialty drugs on the market in 2015 provided by IMS Health. Our results indicate that growth in spending on specialty drugs was a key driver of spending growth in both Medicare Part D and Medicaid’s outpatient drug benefit from 2010 to 2015. On a per capita basis, spending on specialty drugs increased substantially in Part D and more modestly in Medicaid, while spending on traditional drugs declined in both programs. Spending per capita for both specialty drugs and traditional drugs was higher in Medicare Part D than in Medicaid—and such spending grew at a faster rate under Medicare Part D—because of differences in how prices are determined in the two programs and differences in the mix of drugs used by beneficiaries in the two programs.

Retail prices paid for specialty drugs are similar under Part D and Medicaid. However for 50 top selling brand-name specialty drugs, net prices paid by Medicaid were much lower than those in Medicare Part D on average because the statutory rebates under Medicaid are much greater than the rebates Part D plans are able to negotiate from manufacturers. In addition, average net prices paid for brand-name specialty drugs increased much more quickly in Medicare

We also use the Medicare Part D claims data to examine trends in the mean and distribution of beneficiaries’ out-of-pocket costs for specialty drugs over the 2010-2015 period.

Over the past two decades, misuse of prescription painkillers in the U.S. particularly opioids increased rapidly, to the point that in 2011 the CDC labeled the problem an "epidemic". Research investigated the negative health effects of prescription opioid misuse (POM), but little research examined the associations between POM and labor supply. This study investigates the associations between POM and labor force participation and, conditional on being in the labor force, employment. This study merged cross-sectional data from the 2010-2014 National Survey on Drug Use and Health for individuals aged 26 to 64 years old. The analysis examined any misuse, ‘infrequent misuse’ (1 – 199 days), and ‘frequent misuse’ (200 – 365 days) over past year. Prevalence estimates for labor force participation found that 84.8 % of individuals with any opioid misuse were in labor force compared to 81.2% of individuals with no misuse. Roughly, 84.1% of infrequent misusers were in labor force compared to 64.7% of frequent misusers. Unadjusted logistic regressions indicate a positive association between any misuse and labor force participation, but adjusted logistic regressions indicate no association. Both adjusted and unadjusted logistic results found frequent misusers to be less likely in the labor force compared to infrequent misusers. Prevalence estimates for employment indicate that 89.6% of individuals with opioid misuse were employed relative to 92.5% of individuals with no misuse. Roughly, 89.7% of infrequent misusers were employed compared to 81.8% of frequent misusers. Unadjusted and adjusted logistic results indicate a negative association between any misuse and employment, and no significant differences between infrequent and frequent misusers with respect to employment. The associations between POM and labor force participation and employment are not consistent throughout the opioid misuse spectrum. In the unadjusted logistic regression, labor force participation was positively associated with past year POM while frequent misusers were less likely to be in the labor force relative to infrequent misusers. Moreover, employment was negatively associated with past year POM whereas there were no differences in odds of employment between frequent and infrequent misusers. Focusing now on the multivariable analysis, the results indicate no association between past year POM and labor force participation; frequent misusers were less likely to be in the labor force compared to infrequent misusers. On the other hand, we found a negative association between past year POM and employment while there were no statistically significant differences between types of misusers and employment. Understanding the labor supply behavior of POMs is vital in formulating treatment and policy proposals that build upon work incentives. This study is among the first to use standard definitions of work status to enhance our understanding

These findings should be interpreted with caution. The data is self-reported with general validity and reliability issues. The surveys are cross-sectional, and thus, not appropriate to make causal inferences. Finally, we did not control for the potential endogeneity of POM in the labor supply specifications to avoid biases that might be associated with using weak instruments.

This study investigates the impacts of state-mandated emergency rules for opioid prescribing, in response to the ongoing opioid epidemic. We exploit the implementation of an emergency prescribing directive that came into immediate effect in Kentucky starting in July 2012, using the neighboring state of Indiana as a control. The analysis uses novel data from the prescription drug monitoring programs from Kentucky and Indiana that include the universe of all opioid scripts dispensed within the respective states between January 2012-November 2013. Individual prescriber- and pharmacy-level difference-in-difference analyses show that Kentucky's emergency rules significanlty lowered the number of patients being prescribed opioids, the number of opioid scripts written, and the number of refills and days of opioid supply authorized. The prescribing restrictions also affected patient sub-populations in heterogeneous ways. Opioid prescribing declined most sharply for younger patients, for men, and for those on high opioid doses. We further found evidence that after the implementation of the policy, prescribers began to transition patients to lower opioid doses and to opioid drugs like Buprenorphine, which are used for medication assisted treatment of opioid-use-disorders. Our results show that state-mandated opioid prescribing guidelines could provide significant supply side controls on high-risk

As the opioid epidemic has evolved overtime to refer not only to overdose deaths from prescription opioids, but also illicit opioids (ex. heroin, fentanyl), it is important to critically examine the potential role policies restricting access to prescription opioids had in this transition. This is particularly important in states such as Tennessee which have been hit particularly hard. In response to the epidemic Tennessee enacted laws to modify physician behaviors, paying special attention to pain clinics because of the role they are believed to play in diverting prescription opioids to the illegal market when they develop into a pill mill. Specifically, in 2014 the Tennessee General Assembly passed public chapter 700 and 983, which prohibited pain clinics and providers from directly dispensing opioids expect under specific circumstance. This dispensing restriction may influence their general prescribing habits (i.e. decreasing the number of prescriptions, quantity prescribed, etc.), which could decrease the supply of opioids in Tennessee. While these laws most directly affect the supply of opioids, their ultimate purpose is to address the opioid epidemic more broadly and thus their effect on the supply and demand for opioids must be considered. The purpose of this paper is to utilize patient level data from Tennessee Control Substance Monitoring Database (CSMD) for the years 2013-2016 to evaluate the effect of the 2014 laws on the supply of prescription opioids. An interrupted time series model (ITS) will be used to evaluate the ability of the law to reduce the supply of opioids, which will be measured by two outcomes: the monthly total days of opioids prescribed and the monthly average total morphine milligram equivalent (MME). Since not all patients are equally likely to transition to the illicit market, subgroup analysis or quantile regression will be performed to identify any possible heterogeneous treatment effects across the

However, the CSMD only captures information for the market for legally prescribed opioids which have close substitutes in the illegal market in the form of both diverted prescription opioids, as well as illegal substitutes such as heroin. Thus, any decline in the supply of prescription opioids could be offset by substituting for opioids available through the illegal market, which could undermine the overall effectiveness of these laws. Thus secondary analysis will be conducted using visits to the emergency department for an opioid overdose from the Hospital Discharge Dataset to proxy as a measure for total illicit opioid demand in an ITS model. To allow for any readjustments between legal and illegal

Overall, this analysis will provide further insight into the ability of supply side interventions to affect an epidemic that is influenced by legal and illegal markets. Developing policy evaluations that are sensitive to both markets is important given the continuing rise in heroin deaths which suggests the importance substitution to the illegal market plays.

The United States is facing an opioid crisis. With annual overdose deaths in the tens of thousands, the nation faces billions of dollars in lost productivity. Despite a growing body of research on the link between opioid prescriptions and substance use disorders, the possible link between prescriptions and crime remains unexplored. The omission is noteworthy because reducing violent crime is an important health policy target. The criminology literature has established a positive correlation between drug usage and crime rates. According to one theory, addiction and substance use increase the risk of criminal behavior through several channels, including the financial pressure to support habitual drug use and the incentive of illegal suppliers to control markets through violence. Recent studies suggest opioid prescriptions contribute to addiction, and if addiction leads to crime, then changes in

Using novel data from California over the years 2011-2017, this study is the first to test whether increases in opioid prescriptions lead to greater crime rates. The California data are especially useful for two reasons. First, the state collects detailed data on opioid prescription counts, which are made available by the state's Prescription Drug Monitoring Program (PDMP). Second, the state recently experienced an exogenous shock to criminal populations, providing a natural experiment. In May 2011, the US Supreme Court ruled the overcrowding of California prisons and jails to be unconstitutional (Brown v. Plata 2011), and the state was required to reduce its population of inmates over the next two years. The variation is statistically valuable because opioid prescriptions may be related to crime for certain subpopulations but not others. In particular, prescriptions may lead to more crime in locations with more criminal offenders. The unexpected release of inmates helps identify any relationship because, without the variation, other factors might explain any observed correlation between prescription counts and offender populations. Data on prison and jail admissions and releases from the California Department of Corrections and Rehabilitation (CDCR) and the Board of State and Community Corrections (BSCC) allow for a test of any interaction between opioid prescriptions and the in/outflow of offender populations in a community. The study will employ panel data regression methods. The county is the unit of observation, and the time dimension is monthly. Crime data are taken from the FBI's Uniform Crime Reports. Opioid prescription counts, broken down by sex, crude age group, and drug Schedule classification, are provided by California's PDMP (known by the acronym CURES). Prison and jail populations are provided by California's CDCR and BSCC. Socioeconomic control variables include population demographics from the Census Bureau and unemployment rates from the Bureau of Labor Statistics. After constructing the panel, the study will test whether lagged prescriptions in a county predict future crime, along with any

Should the results find a positive relationship between opioid prescriptions and crime, they should also contribute to our understanding of the magnitude of the current opioid crisis.

In the United States, 64,000 people were killed by drug overdosis in 2016. This is up 22 percent from the 2015 level of 52,404 deaths, and have prompted policy makers to declare the opioid epidemic a national emergency. Previous literature provide evidence that physicians matter for the treatment choices of their patients, implying that policies targeting this particular group can limit the current opioid epidemic. However, a thorough understanding of the mechanisms of physician prescribing behavior is needed to grasp the full potential of such strategies. In this paper I contribute to the literature on physician behavior and the opioid epidemic along several important dimensions. Using Danish registry data on the full population linking providers and patients, I document that the opioid prescribing behavior of physicians is dynamic and mutable in nature, and constitutes an important determinant of opioid treatment intensity of individuals. Specifically, I provide evidence that opioid prescribing leniency is partly the result of a diffusion process across physician clinics, spreading from high leniency areas, and effectively constituting a roll-out of opioid leniency across the country. To estimate the diffusion process, I adopt a peer effects framework. I construct physician specific networks and utilize clinic closures due to physician retirement to obtain within-physician variation in composition of peers over time. This allows me to estimate robust spill-over effects. The estimated spill-over effects are non-negligible: an increase in leniency of one standard deviation by a random peer leads to an increase of 5% of a standard deviation in prescribing leniency. I estimate impacts on labor market outcomes of opioid consumption instrumenting individual opioid consumption leveraging the within physician variation in exposure to the diffusion process. Estimating the labor market effects of opioid consumption is important as it constitutes an understudied branch of costs to the increased opioid use worldwide. I estimate significant negative impacts of opioid usage on labor market income: a one standard deviation increase in opioid use results in a reduction of 7.8 percentage points in the labor market income percentile rank. Moreover, I estimate that opioid usage, despite its intent, actually causes increases in short-term disability: a one percentage increase in opioid

The results have wide implications for policy makers, as they add to the understanding of the dynamics of the contagious nature of the opioid epidemic and extend the cost side of opioid usage beyond effects on health and health care expenditures. Furthermore, the results suggest that the effect of policies targeting individual physicians are magnified, as local reductions in opioid leniency would diffuse to nearby physicians due to the presence of spill-overs.

The 1984 Hatch-Waxman Act offers market exclusivities to innovators for their new drugs and follow-on improvements and to first-time generic competitors for challenging innovators' patents. Using data on small-molecule New Chemical Entities (NCEs) from 1985 to 2016, patent challenges they face, and their follow-on innovation, I study the interplay between generic pressure and follow-on innovation. I find that on average follow-on innovation per NCE nearly triples in anticipation of first-time generic competition. However, NCEs approved from 2003 to 2006 show a decline in follow-on indications but an increase in other follow-on innovations relative to NCEs approved from 1985 to 2002. Importantly, follow-on indications reduce the probability of settlement in patent challenges by an average of five and a half percent. This result translates into an average externality value to consumers of about $50 million to $115 million per follow-on indication. By contrast, other follow-on innovations have no effect on the probability of settlement. This study also describes follow-on innovation launching and patenting dynamics that ensue, and it shows that the ratio of patents per follow-on innovation has consistently increased overtime. This suggests caution when using patent counts as a proxy for innovative output.

Economic literature has extensively studied how prices for incumbent firms respond to competition after entry, especially in prescription drug markets following generic entry. However, less attention has been paid to firm behavior prior to entry. We contribute to this gap in the literature by both developing a model of pricing strategies for incumbent drug manufacturers under health insurance and empirically assessing pricing adjustments for incumbent firms, using the insulin market as a natural experiment. We consider the price strategy of incumbent firms among branded, horizontally-differentiated drugs under tiered insurance, which predicts rising prices for incumbent drugs for a range of elasticities as the likelihood of entry increases. Empirically, we exploit exogenous variation in a potential entrant's completion of clinical trials to identify the effect of drug pipeline pressure on the prices of incumbent drugs. Results suggest that pipeline pressure significantly increases the prices of incumbent drugs, and potential biosimilar entry may drive this effect.

As the health care market expanded rapidly over time, provider shortages came to the forefront of policy making. To address this problem, alongside regulations targeting the cost of physician practice the policy makers currently consider alternative delivery methods, and expanding the scope of practice of various types of health care providers. The National Council of State Legislatures (NCSL) reported tracking 827 bills to redefine health providers’ scope of practice in 2012

This paper investigates the impact of regulations allowing nurse practitioners (NPs) to prescribe schedule II prescription drugs such as opioids and stimulants. This topic is of particular importance in light of the recent increase in prescription (Rx) drug abuse. According to 2014 National Survey on Drug Use and Health approximately 54 million people, the equivalent of more than 20 percent of people over 12 years old, have used prescription drugs for nonmedical reasons at least once in their lifetime.[2] Many of them obtained the drugs from a health care provider; for instance, approximately 17.3% of people who abuse prescription painkillers report that a physician prescribed the drugs.[3] Laws allowing NPs to prescribe schedule II drugs effectively expand the number of providers allowed to prescribe these drugs but also change the characteristics of the labor force engaged in drug prescribing. If limiting the ability of nurse practitioners to prescribe drugs enhances service quality, an expansion of prescribing rights of NPs might lead to more prescription drug misuse. In addition, an increase in the number of providers eases access to prescription drugs possibly aggravating of the current drug diversion problem. On the other hand, physician shortages could be the reason behind the high rate of misuse of prescription drugs. Existing physicians might not have enough time and resources to establish a close relationship with their patients or investigate the possibility of fraudulent claims. In this case increasing the number of drug-prescribing providers could alleviate the pressure and reduce illicit access to prescription drugs. Using difference-in-difference and triple-difference econometric models we find that legislation allowing nurse practitioners to prescribe schedule II drugs is associated with less opioid misuse but no statistically significant change in stimulants misuse. The results are robust across several datasets including Treatment Episode Data Set (TEDS) and Mortality Multiple Causes of Death files.

[1] Source: http://www.ncsl.org/research/health/scope-of-practice-overview.aspx accessed 1/21/2017 [2] Source: https://www.samhsa.gov/data/sites/default/files/NSDUH-DetTabs2014/NSDUH-DetTabs2014.pdf, accessed 1/20, 2017 [3] Substance Abuse and Mental Health Services Administration. Results from the 2010 National Survey on Drug Use and Health: volume 1: summary of national findings.


Grant Gannaway [email protected]

Stephan Lindner [email protected]

Yuhua Bao [email protected]

Pelin Ozluk [email protected]

Despite overwhelming evidence that prescription drugs are a highly effective form of treatment, medication adherence rates in the US are very low, likely in part because there are significant personal costs for filling prescriptions - both monetary and non-monetary. In this paper, I study the distance costs required to travel to the pharmacy, examining the extent to which access to pharmacies influences medication adherence. I use straightforward intent-to-treat measures of adherence: the total numbers of (1) pills dispensed, (2) patients, (3) new patients, and (4) pharmacy claims filed. Using All-Payer Pharmacy Claims data from Oregon, I first take an event-study approach around two different types of events. The first is local pharmacy openings and closings, the second is variation in insurance network status of a major pharmacy chain (Walgreens) in and out of the network of a major pharmacy benefits manager (Express Scripts) - essentially closing and re-opening Walgreens to Express Scripts enrollees. I find that pharmacy openings cause a 2% increase in the measures of adherence from a stable pre-trend for local patients, and that removing local Walgreens from the Express Script’s network causes a 5% decrease in the same outcome variables, for similarly stable pre-trends. I find the effect magnitudes decrease as patient distance-to-affected-pharmacy increases. I examine heterogeneous effects by drug type and insurance type, as well as by neighborhood characteristics. Notably, I find large effects for chronic drugs such as heart medication, cholesterol reducers, and beta blockers. This is a substantial result since it highlights the binding effect of distance costs on adherence

I then take these effects and combine them with observable characteristics of the zip code in a machine learning framework to extrapolate counterfactual effects to a national level. I estimate that counterfactually opening an additional pharmacy in each zip code across the US would lead to a median increase of roughly 3% in each of the outcome variables, but with significant heterogeneity across zip codes. These results highlight potential "pharmacy deserts" across the

Finally, I use a discrete-choice framework, combined with copay variation within drug/insurance-type bins but across pharmacies, to estimate patient willingness-to-pay for a one mile reduction in distance to nearest pharmacy. I find that, on average, patients are willing to pay $1 more in copays for a one mile distance reduction. The willingness-to-pay estimates are greater in rural areas with lower pharmacy access, and smaller in urban areas with higher pharmacy access. Combining these results with the effects of pharmacy openings allows me to estimate the social versus private value of increased medication adherence, and to approximate the societal value of additional pharmacies.


: Recent guidelines and policies attach importance to containing the initial exposure to prescription opioids. Little is known about how prescribing decisions regarding the first opioid prescription are associated with high-risk

: Retrospective cohort study. Two populations of interest are privately insured adults aged 18-64 and Medicare Advantage enrollees 65 or older who filled a first opioid prescription between 07/01/2011 and 06/30/2013. We used 2011-2014 data from a large, national commercial insurance claims database to identify individuals naive to opioid therapy (determined based on a six-month look-back with no opioids) and follow them for 18 months after the first opioid prescription. We considered three salient features of the first opioid prescription: long- vs. short- acting opioid formulation, daily dosage in morphine milligram equivalents (MMEs), and days of supply. High-risk opioid use in the long term was measured for each of six quarters (3-month intervals) with some opioid use following the first opioid prescription. Measures included 1) having opioid prescriptions overlapping for seven days or more, 2) having opioid and benzodiazepine prescriptions overlapping for seven days or more, 3) having filled opioid prescriptions from 3 or more prescribers, and 4) having a daily average MMEs exceeding 120. A secondary analysis examined how features of the first

: Our samples included 196,375 non-elderly, privately insured patients and 63,419 elderly, Medicare Advantage patients. All three features of the first prescription strongly predicted high-risk use. For example, for privately insured patients, receiving a long- (vs. short- ) acting opioid was associated with a 16.9-percentage-point increase (95% CI, 14.0-19.5), a daily MME of 50 or more (vs. less than 30) was associated with a 12.5-percentage-point increase (95% CI, 12.1-12.9), and a more-than 7-day supply (vs. 3 or fewer days) was associated with a 4.8-percentage-point increase (95% CI, 4.5-5.2), in the probability of having a daily dosage of 120 MMEs or more in the long term. Results for the Medicare

: Long-acting formulation, high daily dosage, and long duration (exceeding 7 days) of the first opioid prescription were associated with increased high-risk use of opioids in the long term. : Our findings provide support to policies, clinical guidelines, and health care system interventions that direct first opioid prescriptions away from long-acting formulation, high daily dosage, and long

duration. Caution needs to be exercised to guard against applying these policies and guidelines to all opioid prescribing (e.g., to long-term opioid users). Additional policies and considerations are needed to counteract barriers to continued

More than half of the US population lives in a state that has adopted medical marijuana laws (MMLs). Studies show that most medical marijuana patients use marijuana for managing their pain with the overwhelming majority of them preferring it to opioids. Despite ongoing pro-marijuana policies and the growing trend of public acceptance, the evidence on how people change their prescription use due to the availability of marijuana as an alternative treatment is limited. Using the variations across state MMLs between 1996 and 2014 of Medical Expenditure Panel Survey (MEPS) this paper estimates the effects of MMLs on prescription drug utilization, with a focus on opioids. I find that MMLs lead to a $2.47 decrease in per person prescribed opioid spending among young adults (ages 18-39) over a year. Most of this decrease results from the intensive margin of use and MML states that allow home cultivation experience even larger decreases. Furthermore, the decreasing effects are persistent over time and they get stronger following the years of implementation. MMLs also decrease the number of opioid pill use among young adults. I do not find any discernible impact on older populations' opioid utilization. I then investigate the effects on other prescriptions for which marijuana can be a potential substitute and find the allowance of dispensaries is generally associated with decreases, although

Boris Vabson [email protected]


Abraham Asfaw [email protected]

While substantial focus has been paid to the financial characteristics of prescription drug plans, far less attention has been given to the impact of non-financial features, including formulary composition, utilization restrictions, and pharmacy network composition. This is an important gap in the literature, given the substantial non-financial differences that exist across drug plans, and given broader questions that these non-financial characteristics could tie into. To examine the impact of non-financial features independently from financial characteristics of drug plans, we study how drug and non-drug spending and utilization differs across low-income exemption beneficiaries randomly auto-assigned to different Medicare Part D plans. These beneficiaries are not subject to cost sharing, eliminating the possibility that differences across Part D plans are driven by differences in deductibles, copayments, and other financial incentives. We also leverage a new data source indicating which low-income beneficiaries failed to actively choose a Part D plan and were randomly auto-assigned to a plan, allowing us to focus on this population and eliminating the

We show that Part D plan assignment has important effects on drug spending and utilization, with formularies influencing the drugs beneficiaries take. Perhaps more importantly, we also show that formularies affect non-drug spending, providing new evidence for complementarities between these two types of healthcare utilization. Finally, we study the effects of Part D plan assignment on mortality and other outcomes and relate cross-plan differences in those outcomes

In addition to its immediate implications, this study could be of broader academic and policy interest, by providing insights on important issues such as drug-driven medical offsets, behavioral determinants of plan choice, medication

(e.g. due to changes in out-of-pocket drug costs) on other medical

on other medical costs. Two previous studies (of a single disease or a single country) found that pharmaceutical innovation reduced hospitalization, and that the reduction in hospital cost from the use of newer drugs was considerably greater than the innovation-induced increase in pharmaceutical expenditure. In this study, I reexamine the impact that pharmaceutical innovation has had on hospitalization, using a “triple-differences,” or difference-in-difference-in-differences, research design: I estimate the impact that new drug launches had on hospitalization for 106 medical conditions in 15 OECD countries during the period 2002-2015. This design enables me to control for all determinants of hospitalization growth that are invariant across diseases within a country, and for all

Hospitalization is not significantly related to the number of drugs launched 0-3 years earlier; this is not surprising since it takes 8-10 years for a drug to attain its peak level of utilization. However, both the number of hospital discharges and the number of hospital days are significantly inversely related to the number of drugs launched 6-15 years earlier. The estimates indicate that one additional drug launch reduces the number of hospital days 6-15 years later by about 4%.

The estimated reduction in 2015 hospital expenditure attributable to drugs launched during 1996-2009 is 2.5 times as large as the increase in 2015 drug expenditure attributable to those drugs, which implies that pharmaceutical

Using the National Health Interview Survey and Medical Expenditure Panel Survey, I examine whether prescription drug use substitutes investment in preventive health behaviors. To identify their causal relationship, I estimate the differences in the regression discontinuity of prescription drug uses and preventive health behavior at age 65 before and after the implementation of Medicare Part D. The resulting estimates indicate that the implementation of Medicare

reduction in the probability of engaging in moderate physical activity at the extensive margin, a 6.47% reduction in the probability of having healthy-weight and a 9.56 percentage point increases in the probability of being overweight. The effects on moderate physical activity at the intensive margin, vigorous strength physical activity--both at the intensive and extensive margin--cigarette smoking, body mass index and obesity are not statistically significant. The physical exercise, healthy-weight and overweight effect of prescription drug use is stronger among sub-group of individuals that

Anna Anderson-Cook [email protected]

Pierre Alexandre [email protected]

Sumedha Gupta [email protected]

There has been considerable interest in the prices paid for specialty drugs and the amount of spending on such drugs. In particular, concerns have been raised regarding the effects of such drugs on federal spending and the out-of-pocket costs incurred by some beneficiaries. In this paper, we analyze trends in specialty drug pricing and spending (net of manufacturer rebates and discounts) in Medicare Part D and Medicaid over the 2010-2015 period. We also examine

We adopt the definition of specialty drugs developed by IMS Health. Under that definition, specialty drugs treat a chronic, complex, or rare disease and have at least four of seven additional characteristics (such as costing at least $6,000 per year and requiring special handling in the supply chain). We use beneficiary-level claims data for Medicare Part D to estimate total spending at retail prices and the number of units and prescriptions dispensed over the 2010-2015 period by drug. We also have data on total rebates and discounts paid by drug manufacturers under Part D by drug during that period, which we use to estimate net prices and spending for specialty drugs. For Medicaid, we use data on total spending at retail prices, the number of prescriptions and units dispensed, and the statutory rebate amounts by drug over the 2010-2015 period. We then combined those data with Redbook data (by NDC code), which has drug

Our results indicate that growth in spending on specialty drugs was a key driver of spending growth in both Medicare Part D and Medicaid’s outpatient drug benefit from 2010 to 2015. On a per capita basis, spending on specialty drugs increased substantially in Part D and more modestly in Medicaid, while spending on traditional drugs declined in both programs. Spending per capita for both specialty drugs and traditional drugs was higher in Medicare Part D than in Medicaid—and such spending grew at a faster rate under Medicare Part D—because of differences in how prices are determined in the two programs and differences in the mix of drugs used by beneficiaries in the two programs.

Retail prices paid for specialty drugs are similar under Part D and Medicaid. However for 50 top selling brand-name specialty drugs, net prices paid by Medicaid were much lower than those in Medicare Part D on average because the statutory rebates under Medicaid are much greater than the rebates Part D plans are able to negotiate from manufacturers. In addition, average net prices paid for brand-name specialty drugs increased much more quickly in Medicare

Over the past two decades, misuse of prescription painkillers in the U.S. particularly opioids increased rapidly, to the point that in 2011 the CDC labeled the problem an "epidemic". Research investigated the negative health effects of prescription opioid misuse (POM), but little research examined the associations between POM and labor supply. This study investigates the associations between POM and labor force participation and, conditional on being in the labor force, employment. This study merged cross-sectional data from the 2010-2014 National Survey on Drug Use and Health for individuals aged 26 to 64 years old. The analysis examined any misuse, ‘infrequent misuse’ (1 – 199 days), and ‘frequent misuse’ (200 – 365 days) over past year. Prevalence estimates for labor force participation found that 84.8 % of individuals with any opioid misuse were in labor force compared to 81.2% of individuals with no misuse. Roughly, 84.1% of infrequent misusers were in labor force compared to 64.7% of frequent misusers. Unadjusted logistic regressions indicate a positive association between any misuse and labor force participation, but adjusted logistic regressions indicate no association. Both adjusted and unadjusted logistic results found frequent misusers to be less likely in the labor force compared to infrequent misusers. Prevalence estimates for employment indicate that 89.6% of individuals with opioid misuse were employed relative to 92.5% of individuals with no misuse. Roughly, 89.7% of infrequent misusers were employed compared to 81.8% of frequent misusers. Unadjusted and adjusted logistic results indicate a

The associations between POM and labor force participation and employment are not consistent throughout the opioid misuse spectrum. In the unadjusted logistic regression, labor force participation was positively associated with past year POM while frequent misusers were less likely to be in the labor force relative to infrequent misusers. Moreover, employment was negatively associated with past year POM whereas there were no differences in odds of employment between frequent and infrequent misusers. Focusing now on the multivariable analysis, the results indicate no association between past year POM and labor force participation; frequent misusers were less likely to be in the labor force compared to infrequent misusers. On the other hand, we found a negative association between past year POM and employment while there were no statistically significant differences between types of misusers and employment. Understanding the labor supply behavior of POMs is vital in formulating treatment and policy proposals that build upon work incentives. This study is among the first to use standard definitions of work status to enhance our understanding

These findings should be interpreted with caution. The data is self-reported with general validity and reliability issues. The surveys are cross-sectional, and thus, not appropriate to make causal inferences. Finally, we did not control for the

This study investigates the impacts of state-mandated emergency rules for opioid prescribing, in response to the ongoing opioid epidemic. We exploit the implementation of an emergency prescribing directive that came into immediate effect in Kentucky starting in July 2012, using the neighboring state of Indiana as a control. The analysis uses novel data from the prescription drug monitoring programs from Kentucky and Indiana that include the universe of all opioid scripts dispensed within the respective states between January 2012-November 2013. Individual prescriber- and pharmacy-level difference-in-difference analyses show that Kentucky's emergency rules significanlty lowered the number of patients being prescribed opioids, the number of opioid scripts written, and the number of refills and days of opioid supply authorized. The prescribing restrictions also affected patient sub-populations in heterogeneous ways. Opioid prescribing declined most sharply for younger patients, for men, and for those on high opioid doses. We further found evidence that after the implementation of the policy, prescribers began to transition patients to lower opioid doses and to opioid drugs like Buprenorphine, which are used for medication assisted treatment of opioid-use-disorders. Our results show that state-mandated opioid prescribing guidelines could provide significant supply side controls on high-risk

Jackie Yenerall [email protected]

Alexander Lundberg [email protected]

Peter Thingholm [email protected]

As the opioid epidemic has evolved overtime to refer not only to overdose deaths from prescription opioids, but also illicit opioids (ex. heroin, fentanyl), it is important to critically examine the potential role policies restricting access to

In response to the epidemic Tennessee enacted laws to modify physician behaviors, paying special attention to pain clinics because of the role they are believed to play in diverting prescription opioids to the illegal market when they develop into a pill mill. Specifically, in 2014 the Tennessee General Assembly passed public chapter 700 and 983, which prohibited pain clinics and providers from directly dispensing opioids expect under specific circumstance. This dispensing restriction may influence their general prescribing habits (i.e. decreasing the number of prescriptions, quantity prescribed, etc.), which could decrease the supply of opioids in Tennessee. While these laws most directly affect the

The purpose of this paper is to utilize patient level data from Tennessee Control Substance Monitoring Database (CSMD) for the years 2013-2016 to evaluate the effect of the 2014 laws on the supply of prescription opioids. An interrupted time series model (ITS) will be used to evaluate the ability of the law to reduce the supply of opioids, which will be measured by two outcomes: the monthly total days of opioids prescribed and the monthly average total morphine milligram equivalent (MME). Since not all patients are equally likely to transition to the illicit market, subgroup analysis or quantile regression will be performed to identify any possible heterogeneous treatment effects across the

However, the CSMD only captures information for the market for legally prescribed opioids which have close substitutes in the illegal market in the form of both diverted prescription opioids, as well as illegal substitutes such as heroin. Thus, any decline in the supply of prescription opioids could be offset by substituting for opioids available through the illegal market, which could undermine the overall effectiveness of these laws. Thus secondary analysis will be conducted using visits to the emergency department for an opioid overdose from the Hospital Discharge Dataset to proxy as a measure for total illicit opioid demand in an ITS model. To allow for any readjustments between legal and illegal

Overall, this analysis will provide further insight into the ability of supply side interventions to affect an epidemic that is influenced by legal and illegal markets. Developing policy evaluations that are sensitive to both markets is important

The United States is facing an opioid crisis. With annual overdose deaths in the tens of thousands, the nation faces billions of dollars in lost productivity. Despite a growing body of research on the link between opioid prescriptions and substance use disorders, the possible link between prescriptions and crime remains unexplored. The omission is noteworthy because reducing violent crime is an important health policy target. The criminology literature has established a positive correlation between drug usage and crime rates. According to one theory, addiction and substance use increase the risk of criminal behavior through several channels, including the financial pressure to support habitual drug use and the incentive of illegal suppliers to control markets through violence. Recent studies suggest opioid prescriptions contribute to addiction, and if addiction leads to crime, then changes in

The California data are especially useful for two reasons. First, the state collects detailed data on opioid prescription counts, which are made available by the state's Prescription Drug Monitoring Program (PDMP). Second, the state recently experienced an exogenous shock to criminal populations, providing a natural experiment. In May 2011, the US Supreme Court ruled the overcrowding of California prisons and jails to be unconstitutional (Brown v. Plata 2011), and the state was required to reduce its population of inmates over the next two years. The variation is statistically valuable because opioid prescriptions may be related to crime for certain subpopulations but not others. In particular, prescriptions may lead to more crime in locations with more criminal offenders. The unexpected release of inmates helps identify any relationship because, without the variation, other factors might explain any observed correlation between prescription counts and offender populations. Data on prison and jail admissions and releases from the California Department of Corrections and Rehabilitation (CDCR) and the Board of State and Community Corrections (BSCC) allow for a test of any

The study will employ panel data regression methods. The county is the unit of observation, and the time dimension is monthly. Crime data are taken from the FBI's Uniform Crime Reports. Opioid prescription counts, broken down by sex, crude age group, and drug Schedule classification, are provided by California's PDMP (known by the acronym CURES). Prison and jail populations are provided by California's CDCR and BSCC. Socioeconomic control variables include population demographics from the Census Bureau and unemployment rates from the Bureau of Labor Statistics. After constructing the panel, the study will test whether lagged prescriptions in a county predict future crime, along with any

In the United States, 64,000 people were killed by drug overdosis in 2016. This is up 22 percent from the 2015 level of 52,404 deaths, and have prompted policy makers to declare the opioid epidemic a national emergency. Previous literature provide evidence that physicians matter for the treatment choices of their patients, implying that policies targeting this particular group can limit the current opioid epidemic. However, a thorough understanding of the

In this paper I contribute to the literature on physician behavior and the opioid epidemic along several important dimensions. Using Danish registry data on the full population linking providers and patients, I document that the opioid prescribing behavior of physicians is dynamic and mutable in nature, and constitutes an important determinant of opioid treatment intensity of individuals. Specifically, I provide evidence that opioid prescribing leniency is partly the result of a diffusion process across physician clinics, spreading from high leniency areas, and effectively constituting a roll-out of opioid leniency across the country. To estimate the diffusion process, I adopt a peer effects framework. I construct physician specific networks and utilize clinic closures due to physician retirement to obtain within-physician variation in composition of peers over time. This allows me to estimate robust spill-over effects. The estimated spill-over effects

I estimate impacts on labor market outcomes of opioid consumption instrumenting individual opioid consumption leveraging the within physician variation in exposure to the diffusion process. Estimating the labor market effects of opioid consumption is important as it constitutes an understudied branch of costs to the increased opioid use worldwide. I estimate significant negative impacts of opioid usage on labor market income: a one standard deviation increase in opioid use results in a reduction of 7.8 percentage points in the labor market income percentile rank. Moreover, I estimate that opioid usage, despite its intent, actually causes increases in short-term disability: a one percentage increase in opioid

The results have wide implications for policy makers, as they add to the understanding of the dynamics of the contagious nature of the opioid epidemic and extend the cost side of opioid usage beyond effects on health and health care expenditures. Furthermore, the results suggest that the effect of policies targeting individual physicians are magnified, as local reductions in opioid leniency would diffuse to nearby physicians due to the presence of spill-overs.

Ruben Jacobo-Rubio [email protected]

Alice Ellyson [email protected]

Anca Grecu [email protected]

The 1984 Hatch-Waxman Act offers market exclusivities to innovators for their new drugs and follow-on improvements and to first-time generic competitors for challenging innovators' patents. Using data on small-molecule New Chemical Entities (NCEs) from 1985 to 2016, patent challenges they face, and their follow-on innovation, I study the interplay between generic pressure and follow-on innovation. I find that on average follow-on innovation per NCE nearly triples in anticipation of first-time generic competition. However, NCEs approved from 2003 to 2006 show a decline in follow-on indications but an increase in other follow-on innovations relative to NCEs approved from 1985 to 2002. Importantly, follow-on indications reduce the probability of settlement in patent challenges by an average of five and a half percent. This result translates into an average externality value to consumers of about $50 million to $115 million per follow-on indication. By contrast, other follow-on innovations have no effect on the probability of settlement. This study also describes follow-on innovation launching and patenting dynamics that ensue, and it shows that the ratio of patents per

Economic literature has extensively studied how prices for incumbent firms respond to competition after entry, especially in prescription drug markets following generic entry. However, less attention has been paid to firm behavior prior to entry. We contribute to this gap in the literature by both developing a model of pricing strategies for incumbent drug manufacturers under health insurance and empirically assessing pricing adjustments for incumbent firms, using the insulin market as a natural experiment. We consider the price strategy of incumbent firms among branded, horizontally-differentiated drugs under tiered insurance, which predicts rising prices for incumbent drugs for a range of elasticities as the likelihood of entry increases. Empirically, we exploit exogenous variation in a potential entrant's completion of clinical trials to identify the effect of drug pipeline pressure on the prices of incumbent drugs. Results suggest that

As the health care market expanded rapidly over time, provider shortages came to the forefront of policy making. To address this problem, alongside regulations targeting the cost of physician practice the policy makers currently consider alternative delivery methods, and expanding the scope of practice of various types of health care providers. The National Council of State Legislatures (NCSL) reported tracking 827 bills to redefine health providers’ scope of practice in 2012

This paper investigates the impact of regulations allowing nurse practitioners (NPs) to prescribe schedule II prescription drugs such as opioids and stimulants. This topic is of particular importance in light of the recent increase in prescription (Rx) drug abuse. According to 2014 National Survey on Drug Use and Health approximately 54 million people, the equivalent of more than 20 percent of people over 12 years old, have used prescription drugs for nonmedical reasons at least once in their lifetime.[2] Many of them obtained the drugs from a health care provider; for instance, approximately 17.3% of people who abuse prescription painkillers report that a physician prescribed the drugs.[3] Laws allowing NPs to prescribe schedule II drugs effectively expand the number of providers allowed to prescribe these drugs but also change the characteristics of the labor force engaged in drug prescribing. If limiting the ability of nurse practitioners to prescribe drugs enhances service quality, an expansion of prescribing rights of NPs might lead to more prescription drug misuse. In addition, an increase in the number of providers eases access to prescription drugs possibly aggravating of the current drug diversion problem. On the other hand, physician shortages could be the reason behind the high rate of misuse of prescription drugs. Existing physicians might not have enough time and resources to establish a close relationship with their patients or investigate the possibility of fraudulent claims. In this case increasing the number of drug-prescribing providers could alleviate the pressure and reduce illicit access to prescription drugs. Using difference-in-difference and triple-difference econometric models we find that legislation allowing nurse practitioners to prescribe schedule II drugs is associated with less opioid misuse but no statistically significant change in


University of Chicago Complete

Oregon Health & Science University Vinay Prasad Complete

Weill Cornell Medical College Complete

Andrew Young School of Policy Studies Complete

Phyllis Johnson; Yongkang Zhang; Jessica Ancker; Bruce Schackman; Lisa Witkin; M. Carrington Reid; Philip Jeng

Stanford University-SIEPR Daniel Prinz; Timothy Layton Complete


Tulane University Complete

Congressional Budget Office Jared Maeda; Lyle Nelson Complete

Florida Atlantic University Patrick Richard; Valeria Paz Complete

Indiana University, Purdue University Morhaf Achkar Complete

Tennessee Department of Health Complete

West Virginia University Complete

Aarhus University Complete

FDA Complete

CHOICE Institute, University of Washington Anirban Basu Complete

Seton Hall University Andrew Friedson Complete


Theory or Econometric Advances



Understanding the Distributional Impacts of Health Insurance Reforms: Evidence from a Consumer-Cost Sharing Program

Estimating Hospital Quality with Quasi-experimental Data

Random Effect Difference-in-Differences Models and Their Applications



Random Effect Difference-in-Differences Models and Their Applications

Generalization of the Difference-in-Differences Model and its Advantages for Applied Research

Abstract

In this paper, we examine the heterogeneous effects of a health insurance reform on the distribution of total medical spending for the California Public Employees' Retirement System (CalPERS). The reform, called reference pricing, changes the relative prices faced by CalPERS members when receiving care between higher-priced and lower-priced healthcare providers for several medical procedures. Specifically, the reference pricing program establishes a maximum reimbursable amount (i.e. the reference price) for procedures done in higher-priced settings while none is set for patients who have medical procedures performed in lower-priced settings. Using medical claims data for CalPERS and an unaffected control group, we estimate the quantile treatment effects of the program to capture treatment effect heterogeneity. To do this, we employ the method detailed in Firpo (2007), which relies on the assumption that selection into treatment is based only on observable characteristics. Results vary by the medical procedure considered. However, we generally find large, negative effects at higher quantiles of the spending distributions, with smaller effects at lower quantiles. Our results suggest that the reform leads to a relative right-tail reduction in the distribution of total medical spending and that the reform does not lead to a simple, location shift in the distributions of spending. These effects are not captured by mean estimates but have important policy implications

Non-random sorting can bias outcome-based measures of institutional quality and distort the growing set of quality-based policies intended to incentivize institutions and inform consumers. I develop an alternative framework for quasi-experimental quality estimation that accommodates selection on unobservables, nonlinear causal effects, institutional comparative advantage, and Roy selection. I then use this approach to compute empirical Bayes quality posteriors from the 30-day mortality rates of a large set of U.S. hospitals. These posteriors optimally combine estimates from quasi-experimental ambulance company assignment and predictions from observational risk-adjustment models (RAMs) along the lines of a conventional bias-variance tradeoff. I find that higher-spending, higher-volume, and privately-owned hospitals have better quality posteriors, and that most emergency healthcare markets exhibit positive selection-on-gains with patients being admitted to more appropriate hospitals on average. I then quantify the effects of this non-random selection by simulating Medicare reimbursement and consumer guidance policies that use quality posteriors instead of RAMs. The types of hospitals subsidized by performance-linked payment schemes (e.g. Value-based Purchasing) are largely unchanged when quasi-experimental data is incorporated, but existing transfers are magnified. My admission policy simulations, however, highlight the limitations of consumer guidance programs in settings with significant selection on match-specific quality.

Motivation Regression-based difference in difference (DID) analysis is widely used for impact evaluations. Pooled ordinary-least-squares (POLS) DID is the simplest specification. Fixed-effects (FE) DID, another common specification, produces estimates that are more precise than POLS DID. We introduce a third specification, the random-effects (RE) DID, and discuss its properties relative to the others. To the best of our knowledge, RE DID models have not been introduced in the literature before. This may be because, unlike RE, FEs can accommodate correlation between individual-specific effects and regressors. However, as our results show, this feature is not useful in DID applications. Methodology The Pooled DID equation is Y=αP+βP

T*T+βPA*A+βP

TA*TA+ε where Y is the dependent variable, αP a constant, T takes 1 for the treatment group and 0 otherwise, A takes 1 for the after-intervention period and 0 otherwise, TA takes 1 for the treatment group in the after-intervention period and 0 otherwise, and ε is a disturbance term. The FE DID equation is Y=αF+βF

A*A+βFTA*TA+ε

where the fixed effects αF represent unobserved individual-specific effects potentially correlated with the regressors, and ε are iid [0, σε2].

We define the RE DID specification as Y=αR+βR

T*T+βRA*A+βR

TA*TA+ε where the random effects αR are unobservable individual-specific effects that are distributed independently of the regressors. We assume αR are iid [0, σα

2] and ε are iid [0, σε2].

We run the base POLS, FE, and RE models above using CMS’s National Health Expenditure data (balanced panel). We use 1999 as baseline and 2000 as the year of a hypothetical intervention. We also study an extended model including a time invariant covariate. We compare POLS, FE, and RE DID estimates. Findings Table 1: Pooled OLS, FE, and RE DID regression estimates

Base Model

POLS FE

After 67.253 67.253

(37.064) (8.816)

Treat 115.755

(54.030)

Aft&Treat 34.862 34.862

(76.409) (18.175)

Time Invariant Covariate

Notes: Values in parenthesis are standard errors. N=102. 51 groups. 2 periods. The first three columns in Table 1 show that RE and FE give identical point estimates and standard errors (s.e.) for the parameter of interest, Aft&Treat. This implies that RE and FE outperform POLS. The last three columns show that the RE model allows the identification of time invariant characteristics (note the estimate is identical to POLS) while giving same point estimates and s.e. for Aft&Treat as the FE model.

Conclusions RE DID is preferred to both POLS and FE DID because it allows the inclusion of both time-invariant characteristics (which FE cannot) and unit-specific effects (which POLS cannot). This is achieved at no cost in terms of the identifying assumption. A main application of RE DID is out-of-sample prediction of intervention effects based on time invariant characteristics (e.g., region) to inform the scalability of interventions. POLS DID would produce less efficient estimates and FE DID would not allow to condition the predictions on time invariant characteristics.

Introduction Difference-in-differences (DD) models are common in health policy evaluation. They allow causal inference using observational data under the strong identifying assumption of parallel trends, i.e., treatment and control groups would have had parallel outcome trends in the absence of treatment. We discuss one particular specification of triple difference (TD) models, which we call generalized difference-in-differences (GDD). GDD relaxes the parallel trends assumption by allowing the outcome trends of the treatment and control groups to differ up to a linear term. Related models were introduced decades ago[1], inspired by Donald T. Campbell’s comparative interrupted time series (CITS) work.[2] Our conjecture is that these models were not widely applied by economists, despite their superiority to DD, due to the absence of an econometric specification that generalizes DD and a rigorous discussion of the required assumptions and formula to estimate the treatment effect. Very recently, Mora and Reggio[3] presented a fully flexible generalization of DD along with a discussion of the underlying assumptions and treatment effects. The aim of this paper is to present a specialization of their model that can be easily used by applied researchers to take full advantage of the relaxation of the parallel trends assumption. We compare the model to DD and explain its advantages. Methodology We build the GDD econometric specification (eq. 1) from the standard DD model by adding terms that control for differences in pre and post intervention trends for both the treatment and control groups. Y=b0+b1*treat*post*time+b2*treat*post+b3*treat*time+b4*post*time+b5*treat+b6*time+b7*post+e (eq. 1) where Y is the outcome of interest, treat is an indicator for treatment group, post is an indicator for post-treatment period, and time represents the time period. Terms in bold are additions to the standard DD model, which consists only of the terms including b0, b2, b5, and b7. Then, we use simulations to show how point estimates from the GDD and DD models compare to each other, under different scenarios. We combine the simulations with theoretical insights to discuss the contribution of each added term in achieving a valid estimate. The point estimate for the intervention effect is calculated as

b2+b1*[(1+2+…+N)/N] where N is the number of post-intervention periods. Findings GDD generalizes DD in the sense that its identifying assumption holds under less restrictive conditions. When the parallel trend assumption was satisfied, the GDD and DD models produced the same point estimates. When the trends of the treatment and comparison groups differed linearly at baseline, only GDD produced consistent estimates. Implications GDD relaxes the identifying assumption of DD at little cost. This is a strong reason for applied researchers to consider favoring GDD over DD for any application that satisfies the slightly longer pre-period requirement of the model.

References [1] Simonton, D.K. (1977). Cross-sectional time-series experiments: some suggested statistical analyses. Psychological Bulletin 84(3), 489-502 [2] Campbell, DT and Stanley, JC. (1966). Experimental and quasi-experimental designs for research. Chicago: Rand McNally: 1966. [3] Mora R. and Reggio I. (2017). Alternative diff-in-diffs estimators with several pretreatment periods. Econometric Reviews, 1-22

Abstract

In this paper, we examine the heterogeneous effects of a health insurance reform on the distribution of total medical spending for the California Public Employees' Retirement System (CalPERS). The reform, called reference pricing, changes the relative prices faced by CalPERS members when receiving care between higher-priced and lower-priced healthcare providers for several medical procedures. Specifically, the reference pricing program establishes a maximum reimbursable amount (i.e. the reference price) for procedures done in higher-priced settings while none is set for patients who have medical procedures performed in lower-priced settings. Using medical claims data for CalPERS and an unaffected control group, we estimate the quantile treatment effects of the program to capture treatment effect heterogeneity. To do this, we employ the method detailed in Firpo (2007), which relies on the assumption that selection into treatment is based only on observable characteristics. Results vary by the medical procedure considered. However, we generally find large, negative effects at higher quantiles of the spending distributions, with smaller effects at lower quantiles. Our results suggest that the reform leads to a relative right-tail reduction in the distribution of total medical spending and that the reform does not lead to a simple, location shift in the distributions of spending. These effects are not captured by mean estimates but have important policy implications

Non-random sorting can bias outcome-based measures of institutional quality and distort the growing set of quality-based policies intended to incentivize institutions and inform consumers. I develop an alternative framework for quasi-experimental quality estimation that accommodates selection on unobservables, nonlinear causal effects, institutional comparative advantage, and Roy selection. I then use this approach to compute empirical Bayes quality posteriors from the 30-day mortality rates of a large set of U.S. hospitals. These posteriors optimally combine estimates from quasi-experimental ambulance company assignment and predictions from observational risk-adjustment models (RAMs) along

I find that higher-spending, higher-volume, and privately-owned hospitals have better quality posteriors, and that most emergency healthcare markets exhibit positive selection-on-gains with patients being admitted to more appropriate hospitals on average. I then quantify the effects of this non-random selection by simulating Medicare reimbursement and consumer guidance policies that use quality posteriors instead of RAMs. The types of hospitals subsidized by performance-linked payment schemes (e.g. Value-based Purchasing) are largely unchanged when quasi-experimental data is incorporated, but existing transfers are magnified. My admission policy simulations, however, highlight the limitations of consumer guidance programs in settings with significant selection on match-specific quality.

Regression-based difference in difference (DID) analysis is widely used for impact evaluations. Pooled ordinary-least-squares (POLS) DID is the simplest specification. Fixed-effects (FE) DID, another common specification, produces estimates that are more precise than POLS DID. We introduce a third specification, the random-effects (RE) DID, and discuss its properties relative to the others. To the best of our knowledge, RE DID models have not been introduced in the literature before. This may be because, unlike RE, FEs can accommodate correlation between individual-specific effects and regressors. However, as our results show, this feature is not useful in DID applications.

a constant, T takes 1 for the treatment group and 0 otherwise, A takes 1 for the after-intervention period and 0 otherwise, TA takes 1 for the treatment group in the after-intervention period and 0

represent unobserved individual-specific effects potentially correlated with the regressors, and ε are iid [0, σε2].

are unobservable individual-specific effects that are distributed independently of the regressors. We assume αR are iid [0, σα2] and ε are iid [0, σε

2]. We run the base POLS, FE, and RE models above using CMS’s National Health Expenditure data (balanced panel). We use 1999 as baseline and 2000 as the year of a hypothetical intervention. We also study an extended model including a

Model with Time Invariant Covariate

RE POLS FE

67.253 67.253 67.253

(8.816) (35.831) (8.816)

115.755 -4.748

(54.030) (67.646)

34.862 34.862 34.862

(18.175) (73.868) (18.175)

-31.021

(11.066)

The first three columns in Table 1 show that RE and FE give identical point estimates and standard errors (s.e.) for the parameter of interest, Aft&Treat. This implies that RE and FE outperform POLS. The last three columns show that the RE model allows the identification of time invariant characteristics (note the estimate is identical to POLS) while giving same point estimates and s.e. for Aft&Treat as the FE model.

RE DID is preferred to both POLS and FE DID because it allows the inclusion of both time-invariant characteristics (which FE cannot) and unit-specific effects (which POLS cannot). This is achieved at no cost in terms of the identifying assumption. A main application of RE DID is out-of-sample prediction of intervention effects based on time invariant characteristics (e.g., region) to inform the scalability of interventions. POLS DID would produce less efficient estimates

Difference-in-differences (DD) models are common in health policy evaluation. They allow causal inference using observational data under the strong identifying assumption of parallel trends, i.e., treatment and control groups would have

We discuss one particular specification of triple difference (TD) models, which we call generalized difference-in-differences (GDD). GDD relaxes the parallel trends assumption by allowing the outcome trends of the treatment and control

Related models were introduced decades ago[1], inspired by Donald T. Campbell’s comparative interrupted time series (CITS) work.[2] Our conjecture is that these models were not widely applied by economists, despite their superiority to DD, due to the absence of an econometric specification that generalizes DD and a rigorous discussion of the required assumptions and formula to estimate the treatment effect. Very recently, Mora and Reggio[3] presented a fully flexible generalization of DD along with a discussion of the underlying assumptions and treatment effects. The aim of this paper is to present a specialization of their model that can be easily used by applied researchers to take full advantage of the relaxation of the parallel trends assumption. We compare the model to DD and explain its advantages.

We build the GDD econometric specification (eq. 1) from the standard DD model by adding terms that control for differences in pre and post intervention trends for both the treatment and control groups. *post+e (eq. 1)

is an indicator for post-treatment period, and time represents the time period. Terms in bold are additions to the standard DD model, which consists only of

Then, we use simulations to show how point estimates from the GDD and DD models compare to each other, under different scenarios. We combine the simulations with theoretical insights to discuss the contribution of each added term

GDD generalizes DD in the sense that its identifying assumption holds under less restrictive conditions. When the parallel trend assumption was satisfied, the GDD and DD models produced the same point estimates. When the trends of the treatment and comparison groups differed linearly at baseline, only GDD produced consistent estimates.

GDD relaxes the identifying assumption of DD at little cost. This is a strong reason for applied researchers to consider favoring GDD over DD for any application that satisfies the slightly longer pre-period requirement of the model.

[1] Simonton, D.K. (1977). Cross-sectional time-series experiments: some suggested statistical analyses. Psychological Bulletin 84(3), 489-502 [2] Campbell, DT and Stanley, JC. (1966). Experimental and quasi-experimental designs for research. Chicago: Rand McNally: 1966. [3] Mora R. and Reggio I. (2017). Alternative diff-in-diffs estimators with several pretreatment periods. Econometric Reviews, 1-22


Marion Aouad [email protected]

Peter Hull [email protected]

Guido Cataife [email protected]

In this paper, we examine the heterogeneous effects of a health insurance reform on the distribution of total medical spending for the California Public Employees' Retirement System (CalPERS). The reform, called reference pricing, changes the relative prices faced by CalPERS members when receiving care between higher-priced and lower-priced healthcare providers for several medical procedures. Specifically, the reference pricing program establishes a maximum

Using medical claims data for CalPERS and an unaffected control group, we estimate the quantile treatment effects of the program to capture treatment effect heterogeneity. To do this, we employ the method detailed in Firpo (2007), which relies on the assumption that selection into treatment is based only on observable characteristics. Results vary by the medical procedure considered. However, we generally find large, negative effects at higher quantiles of the spending distributions, with smaller effects at lower quantiles. Our results suggest that the reform leads to a relative right-tail reduction in the distribution of total medical spending and that the reform does not lead to a simple, location

Non-random sorting can bias outcome-based measures of institutional quality and distort the growing set of quality-based policies intended to incentivize institutions and inform consumers. I develop an alternative framework for quasi-experimental quality estimation that accommodates selection on unobservables, nonlinear causal effects, institutional comparative advantage, and Roy selection. I then use this approach to compute empirical Bayes quality posteriors from the 30-day mortality rates of a large set of U.S. hospitals. These posteriors optimally combine estimates from quasi-experimental ambulance company assignment and predictions from observational risk-adjustment models (RAMs) along

I find that higher-spending, higher-volume, and privately-owned hospitals have better quality posteriors, and that most emergency healthcare markets exhibit positive selection-on-gains with patients being admitted to more appropriate hospitals on average. I then quantify the effects of this non-random selection by simulating Medicare reimbursement and consumer guidance policies that use quality posteriors instead of RAMs. The types of hospitals subsidized by performance-linked payment schemes (e.g. Value-based Purchasing) are largely unchanged when quasi-experimental data is incorporated, but existing transfers are magnified. My admission policy simulations, however, highlight the

Regression-based difference in difference (DID) analysis is widely used for impact evaluations. Pooled ordinary-least-squares (POLS) DID is the simplest specification. Fixed-effects (FE) DID, another common specification, produces estimates that are more precise than POLS DID. We introduce a third specification, the random-effects (RE) DID, and discuss its properties relative to the others. To the best of our knowledge, RE DID models have not been introduced in the literature before. This may be because, unlike RE, FEs can accommodate correlation between individual-specific effects and regressors. However, as our results show, this feature is not useful in DID applications.

a constant, T takes 1 for the treatment group and 0 otherwise, A takes 1 for the after-intervention period and 0 otherwise, TA takes 1 for the treatment group in the after-intervention period and 0

We run the base POLS, FE, and RE models above using CMS’s National Health Expenditure data (balanced panel). We use 1999 as baseline and 2000 as the year of a hypothetical intervention. We also study an extended model including a

RE

67.253

(8.816)

-4.748

(79.842)

34.862



(18.175)

-31.021

(15.488)

The first three columns in Table 1 show that RE and FE give identical point estimates and standard errors (s.e.) for the parameter of interest, Aft&Treat. This implies that RE and FE outperform POLS. The last three columns show that the RE

RE DID is preferred to both POLS and FE DID because it allows the inclusion of both time-invariant characteristics (which FE cannot) and unit-specific effects (which POLS cannot). This is achieved at no cost in terms of the identifying assumption. A main application of RE DID is out-of-sample prediction of intervention effects based on time invariant characteristics (e.g., region) to inform the scalability of interventions. POLS DID would produce less efficient estimates

Difference-in-differences (DD) models are common in health policy evaluation. They allow causal inference using observational data under the strong identifying assumption of parallel trends, i.e., treatment and control groups would have

We discuss one particular specification of triple difference (TD) models, which we call generalized difference-in-differences (GDD). GDD relaxes the parallel trends assumption by allowing the outcome trends of the treatment and control

Related models were introduced decades ago[1], inspired by Donald T. Campbell’s comparative interrupted time series (CITS) work.[2] Our conjecture is that these models were not widely applied by economists, despite their superiority to DD, due to the absence of an econometric specification that generalizes DD and a rigorous discussion of the required assumptions and formula to estimate the treatment effect. Very recently, Mora and Reggio[3] presented a fully flexible generalization of DD along with a discussion of the underlying assumptions and treatment effects. The aim of this paper is to present a specialization of their model that can be easily used by applied researchers to take full advantage of

We build the GDD econometric specification (eq. 1) from the standard DD model by adding terms that control for differences in pre and post intervention trends for both the treatment and control groups.

represents the time period. Terms in bold are additions to the standard DD model, which consists only of

Then, we use simulations to show how point estimates from the GDD and DD models compare to each other, under different scenarios. We combine the simulations with theoretical insights to discuss the contribution of each added term

GDD generalizes DD in the sense that its identifying assumption holds under less restrictive conditions. When the parallel trend assumption was satisfied, the GDD and DD models produced the same point estimates. When the trends of

GDD relaxes the identifying assumption of DD at little cost. This is a strong reason for applied researchers to consider favoring GDD over DD for any application that satisfies the slightly longer pre-period requirement of the model.


Copenhagen Business School Timothy Brown; Christopher Whaley Complete

Microsoft Research Complete

IMPAQ International, LLC Complete




Tobacco, Alcohol, and Illegal Substances




Gun Violence in Black and White: Evidence from Policy Reform in Missouri

Medical Marijuana Availability, Price, and Product Variety and Adolescents’ Marijuana Use

Tobacco-21 Laws: Impacts on Late Adolescent Smoking

The effect of “ride-sharing” on risky behaviors and traffic fatalities






Keg registration laws, alcohol consumption, and alcohol-involoved traffic fatalities among adolescents

Defining a Substance Market: Substitution and Complementarities between Marijuana, Alcohol, and Tobacco

The Impacts of Potency, Warning Messages, and Price on Preferences for Marijuana Products

The Impact of Ridesharing Apps on Personal Alcohol Consumption

Differential Rational Addiction Choices among Youth and Young Adults




Outdoor smoking patterns and frequency of exposure to Environmental Tobacco Smoke (ETS): Evidence from NYC streets.

Pseudo-Mature Behaviors, School Activities, and Early Adult Outcomes

An Instrumental Variables Approach to Estimating the Effects of Changes in the Heroin Market on Overdose in the US




Cigarette Smoking and Selective Migration: Are Tobacco Control Policies Less Effective in Rural America?

Can Public Transportation Reduce Accidents? Evidence from the Introduction of Late-Night Buses in Israeli Cities

The Effects of Tobacco Policy on Youth Smoking and Physical Activity

Tobacco, Alcohol, and Illegal SubstancesAre There Gateway Drugs? An Analysis of Opioid Dependence Pathways in National Surveys

Abstract

The role of state-level background check requirements for private firearm sales in reducing gun violence remains controversial in both the empirical literature and gun control policy debate. On August 28, 2007 the Missouri General Assembly repealed an 86 year-old “permit- to-purchase" (PTP) law requiring that handgun purchasers possess a permit, and subsequently undergo a background check, for all sales. The vast racial disparities in firearm homicide within Missouri raises important questions concerning the disproportionate impact of the repeal on Black communities throughout the state. Using generalized synthetic control estimation, this paper finds that the PTP repeal led to a modest increase in county-level gun ownership in addition to substantial evidence of increased firearm homicide in the early years of the 2007- 2013 post-repeal period. In particular, state-level effects suggests that overall Black firearm homicide increases on average by an additional five deaths per 100,000 while the same rates for Black victims ages 15-24 rise by 29 deaths per 100,000. County-level estimates also show considerable increases in firearm homicide in Black communities within the more urban regions of the state. Treatment effect estimates for state-level Black firearm homicide translate into approximately an additional 260 deaths attributable to the change in the law over the 2007-2013 period.

PURPOSE: To examine the availability of medical marijuana dispensaries, price of medical marijuana products, and variety of medical marijuana products in school neighborhoods and their associations with adolescents’ use of marijuana and susceptibility to use marijuana in the future. METHODS: A representative sample of 8th, 10th, and 12th graders (N=46,646) from 117 randomly selected schools in California participated in the cross-sectional 2015-16 California Student Tobacco Survey (CSTS). Characteristics of medical marijuana dispensaries in California were collected and combined with school locations to compute availability, price, and product variety of medical marijuana in school neighborhoods. Multilevel logistic regressions with random intercepts at school level were conducted to test the associations, accounting for individual and school socioeconomic characteristics. RESULTS: The distance from school to the nearest medical marijuana dispensary (within 0-1 mile and 1-3 mile bands) was not associated with adolescents’ use of marijuana in the past month or susceptibility to use marijuana in the future, nor was the weighted count of medical marijuana dispensaries within the 3-mile band of school. Neither the product price nor the product variety in the dispensary nearest to school was associated with marijuana use or susceptibility to use. The results were robust to different specifications of medical marijuana measures. CONCLUSIONS: There was no evidence supporting the associations of medical marijuana availability, price, or product variety around school with adolescents’ marijuana use and susceptibility to use.

In the past five years, 3 states and over 100 localities have raised their tobacco sales ages to 21. This paper is the first nationally representative analysis to estimate this policy’s effects on late adolescent smoking. Specifically, difference-in-differences analyses use the Behavioral Risk Factor Surveillance System’s Selected Metropolitan/Micropolitan Area Risk Trends (SMART) data to test how age-21 tobacco sales restrictions impact conventional cigarette use among 18 to 20 year olds. The analytic sample is restricted to metropolitan statistical areas and metropolitan divisions included in every year of the 2011-2015 SMART data. Specifically, these are areas for which the nationally representative BRFSS surveys interviewed at least 500 respondents in each survey-year. As all states adopting age-21 laws implemented their policies in 2016 or later, the analyses presented here are based on policies implemented at the sub-state level. Limiting consideration to 18 to 20 year olds interviewed between 2011 and 2015 yields a sample size of 22,995 respondents, 22,131 of whom responded to the survey’s current smoking questions. The outcome variable of interest is a binary “current smoker” indicator. For each respondent, the percent of the population covered by an age-21 tobacco sales restriction is calculated for their MMSA-by-state as of their interview date. Difference-in-differences regressions evaluate whether having a higher likelihood of exposure to an age-21 tobacco purchasing restriction yields a differential likelihood of current smoking, controlling for geographic unit and year fixed effects, as well as respondent demographics and an array of other policy variables. This specification passes the requisite parallel trend tests. Baseline findings indicate that exposure to an age-21 tobacco sales restriction at interview yields a statistically significant 4.8 percentage point drop in one’s likelihood of being a current smoker. Since the vast majority of age-21 laws were implemented in the Northeast and Mid-Atlantic Census Divisions, specification checks restrict the sample to these areas and find a slightly lower but still statistically significant 2.4 percentage point drop in current smoking associated with policy exposure. Notably, among respondents living in areas with non-zero exposure to tobacco-21 restrictions, the mean likelihood of exposure is only 9.0% in the full sample, as compared to 38.3% in the Northeast and Mid-Atlantic Census Divisions. Thus, for the average respondent who was exposed to these policies, we would expect a 0.4 percentage point drop in smoking relative to unexposed respondents nation-wide (0.048*0.09 = 0.004), versus a 0.9 percentage point drop relative to unexposed respondents in the Northeast and Mid-Atlantic Census Divisions (0.024*0.383 = 0.009). Falsification tests repeat the main analysis using 23 to 25 year-old respondents, a group not bound by the age-21 restrictions. In this case, age-21 tobacco restrictions yield statistically insignificant effects on current smoking. Overall, these results indicate that restricting tobacco sales to individuals under age-21 yields a statistically significant reduction in smoking among 18 to 20 year olds, on the order of 0.4 to 0.9 percentage points.

During past few years, the “ride-sharing” company like Uber has become remarkably polarizing and is growing exponentially both in United States and all around the world. Uber has changed people’s public transportation choices by launching a phone application that links individual’s transportation needs with private drivers that offer riding services. Such wide availability and lower cost of Uber enable people to ask for riding services instead of taking the risk of drunk-driving after drinking. In this paper, we evaluate the treatment effects of this recent controversial debating public program, “ride-sharing”, on risky behaviors like drinking and smoking, and traffic fatalities by investigating Uber’s expansion during 2010-2015 and exploiting the variations in Uber launch dates across different counties in United States. We firstly use synthetic control method with panel data on drinking and smoking outcomes as well as county-level demographic characteristics to identify treatment effect of Uber launch on probabilities of drinking and smoking. We find little evidence that Uber launch increases people's intention to drinking, but slightly increases the probability of smoking in Uber counties. Furthermore, using different-in-difference methodology and a panel data constructed from Fatality Analysis Reporting System (FARS) , we identify that the launch of Uber service is associated with 11.3 percentage points decrease in total fatality rate and 3.2 percentage points decrease in alcohol-involved traffic fatalities. These results are robust to difference model specifications and falsification tests.

This paper identifies the causal treatment effect of KR laws on underage alcohol consumption and related outcomes: alcohol-related traffic fatalities, by exploiting the substantial variations in the timing of the introduction of these laws across different states at different times. Keg registration (KR) laws require alcohol retailers or wholesalers attach a registered label to the beer kegs that they sell. These laws aim to reduce underage illegal alcohol consumption by imposing liability on adults who purchase beer kegs or host keg parties. Although, in recent years, an increasing number of states have adopted the KR laws to control underage alcohol use and abuse, empirical evidence of the effectiveness of this policy is quite limited. To correct for slection bias and policy endogeneity problem, we matched our treatment samples with a comparable control samples using propensity score matching method. Using the matched sample that created, we conducted difference-in-difference analysis and suggest that the introduction of the KR laws is associated with up to a 2.3 percentage point reduction in binge drinking among minors. This statistical significant reduction is mainly driven by male minors. Furthermore, our results show that strict KR laws can significantly decrease the number of alcohol-involved traffic fatalities by 0.292 among 17 year-old minors and 0.319 among 15-17 year-old minors. Our results are robust to alternative model specifications.

State-level efforts to legalize marijuana have given consumers new choices in the market for substances beyond alcohol and tobacco. Setting optimal tax policies in these states, for public health or revenue motives, may require an understanding of the degree to which consumers are will to substitute between different sin goods. We use detailed data on marijuana, tobacco, and alcohol sales in Washington state to estimate a demand system for substances that allows for flexible substitution patterns. We estimate the price elasticity of demand for marijuana in -1.32, and the income elasticity is 1.88. The legalization of marijuana in Washington reduced demand for alcohol by 0.8% and reduced demand for tobacco by 0.2%. Post-legalization, we find that an increase in marijuana prices increases alcohol sales slightly but does not affect tobacco sales. Our results suggest that the market for sin goods can effectively be modeled as a set of independent markets for individual substances.

Objective. To examine the impacts of tetrahydrocannabinol (THC) level, cannabidiol (CBD) level, warning messages, and price on adults’ preference for marijuana products. Design. An online discrete choice experiment was implemented in October 2017. Each participant was randomly assigned to 1 out of 6 choice sets, each including 12 randomly-ordered choice tasks. Each choice task asked the participant to choose 1 out of 3 alternative marijuana products sold in recreational marijuana stores, with varying levels in 4 attributes (THC level, CBD level, warning message, and price). An opt-out option was also offered in each choice task. The impacts of the attributes on the probability of choosing marijuana products were analyzed using conditional logit regressions, controlling for individual sociodemographic characteristics. Past-year marijuana users and nonusers were analyzed separately. Setting and Participants. A general population sample of 2,398 participants (1,200 past-year marijuana users and 1,198 nonusers) aged 18 years or older living in states that had passed the laws to legalize recreational marijuana (California, Colorado, Massachusetts, Nevada, Oregon, and Washington) were recruited to complete the discrete choice experiment from an online panel. Results. Among both marijuana users and nonusers, higher price was associated with lower probability of choosing marijuana products (p<.001); compared to CBD-free, very low CBD level (0.4%) was associated with lower probability of choosing marijuana products (p<.001) whereas very high CBD level (15%) was associated with higher probability of choosing marijuana products (p<.001). Higher THC level increased the probability of choosing marijuana products among users but had no impact on nonusers. Compared to no warning label, text warning message currently adopted by Washington and Colorado increased the probability of choosing marijuana products among users (p<.05) and FDA not-approved disclaimer reduced the probability of choosing marijuana products among nonusers (p<.001). The results did not differ by purpose of use (primarily medical or recreational in the past year) among users.

Conclusion. Restricting potency level, imposing tax, and adopting warning messages that emphasize no approval from FDA may reduce marijuana purchase among adults.

This study evaluates the indirect effects of the availability of the ride-sharing service, Uber, on alcohol consumption in the United States. Uber is a smartphone application that immediately connects riders to willing drivers and provides them with: real-time location tracking of the vehicle, advance information on the cost of the ride and a convenient phone payment system. Given the ease of use, Uber provides a convenient alternative to drunk driving, one of the largest contributors to traffic accidents and traffic deaths in the United States, especially among 21-34 olds. Recent evidence in working-papers by Martin-Buck (2017) and Dills and Mulholland (2017) has shown that there have been reductions in drunk driving fatalities and DUI/DWI arrests associated with the ride-sharing service availability, using national data from the Fatality Analysis Reporting System and the Uniform Crime Reporting Program. Additionally, they find a decrease in physical and sexual violence, suggesting that the ride-sharing service may reduce altercations over transportation and/or impaired decision-making over the course of a night out. However, to date the research has overlooked the first order, albeit indirect effects of Uber on the act of drinking itself. There are two potential avenues of influence on alcohol consumption. First, there may be an increase at the extensive margin: the total number of individuals who consume alcohol on a given night. The magnitude of this effect relies on the number of drivers that are freed up from their responsibilities as a sober driver to participate in the evening’s festivities. Second, there could be an increase in the intensity of alcohol consumption. We use a difference-in-difference framework to test the effect of Uber’s arrival on the frequency and intensity of drinking, and frequency of drinking and driving as self-reported in the Behavior Risk Factor Surveillance System (BRFSS) between 2007 and 2015. Through these analyses we can glean insights into the metamorphosis of social behaviors in the wake of the new technology. We find that two forms of Uber’s service, UberBlack and UberX, are not associated with changes in drinking behavior while another option is associated with decreases in alcohol consumption. On the other hand, we find the Uber option that provides the lowest average cost to consumers, UberXL, is associated with large increases in excessive drinking, particularly among those aged 25-39.

We estimated a rational addiction model using a 10 year-panel from the National Longitudinal Survey of Youth 1997 Cohort. After controlling for individual heterogeneity, we find that drinking and smoking show a somewhat rational addiction pattern, with a moderate degree of addiction. The level of present bias or time preference differs by behaviors. Drinking participation is likely a rational choice, with a corresponding 2.8% estimated interest rate. In contrast, heavy drinking, smoking, and daily smoking, despite responding to future participation, are also subject to present bias and a heavy discount of future consequences. Instrumental variable results are more likely pertaining to myopic users who are not forward-looking and respond to policies. In sum, regulatory policies are less likely to influence rational users and are more likely to influence myopic users who have self-control or heavy discounting issues. Policy makers should take account of these decision biases in their costs and benefits of regulatory actions.

Over past several decades, smoking rates in the USA have substantially declined, however, the public health dangers from smoking are still prevalent as approximately 36.5 million individuals continue to smoke daily. Smokers aren’t the only individuals affected by smoking; nonsmokers face Environmental Tobacco Smoke (ETS) -- commonly referred to as Secondhand Smoke -- which is responsible for numerous deaths worldwide and is known to cause major health problems for adults and children. To combat the threat of ETS, many localities have enacted smoke-free laws banning smoking in a variety of public and work places. These smoke-free laws have had little effect on decreasing smoking, but have generally lead to a reduction in exposure to ETS by nonsmokers and have altered the behavior of smokers by displacing them from their prior smoking locations to private homes and other outdoor locations (Adda and Cornaglia, 2010, Carpenter, Postolek, and Warman, 2011). This behavioral change in smokers and the displacement of ETS requires further understanding as to how smoke-free laws impacted patterns of smoking, as ETS is threatening populations that prior to smoke-free laws were less affected. In this paper, I use a novel observational dataset collected over several months on New York City streets to investigate smoking displacement and smoking patterns. Since ETS has greatest negative health impact on children, I collected the data to document smoking patterns at times when more children were walking to and from schools. I find that nonsmokers walking in the city are exposed to ETS approximately once every 1.9 city blocks. Upon matching the observational data to city street characteristics, I find that smoking patterns on city streets are nonrandom and are associated with the presence of institutions that banned smoking due to smoke-free laws. I find strong, positive associations between the number of smokers located in proximity to schools, hotels, and health clinics. While my observational smoking data shows an association between smoking location patterns and places that banned smoking, it does not establish a causal link. To further investigate smoking displacement and smoking patterns, I use the NY Youth Smoking Survey and NY Adult Smoking Survey from the NY State Department of Health to examine smoking location changes over the past fifteen years. Unlike my primary data, the survey questions do not provide great detail on where individuals smoke, however, the surveys show that among active smokers, smoking has declined at prohibited locations. This supports the smoking displacement hypothesis and bolsters the evidence I observe through my primary data. Although previous research has documented smoking displacement, the main contribution of my paper is to show that smoking displacement may be occurring in previously undocumented locations and that further data collection and changes to survey questions need to take place in order to better understand which populations are affected by ETS, how much they are affected, and what policy needs to be implemented to further limit negative impacts on population health.

This paper explores the impacts of pseudo-mature behaviors (PMBs), school activities, and social identity on early adult outcomes, specifically the likelihood of college attendance and annual earnings. We define the PMBs to include (binge) drinking alcohol, cigarette smoking, and sexual activity. A student's school activities involve his/her participation in sports, non-sports, and mixed clubs. The incorporation of such measures into our analysis highlights the importance of social identity, a concept that is popular in sociology but often not traditionally emphasized in economics. We use data obtained from the Restricted-Use National Longitudinal Study of Adolescent Health (Add Health). A two-step estimation procedure is used that allows for the possible endogeneity of the PMBs. Much of the literature has struggled to find good identifying instruments; French and Popovici (2011) provide a nice summary of the approaches undertaken by researchers to identify non-weak instruments (IVs). Correspondingly, we use maternal binge drinking and popularity as per Renna (2007, 2009) and Mundt and French (2013), respectively, and also control for a single parent household, which is supported by Peer Cluster Theory, but new to the literature. Our analysis also controls for peer-group effects with the incorporation of grade fixed effects and robust errors that are clustered at the school-level. The primary empirical work is conducted using relative PMB measures (e.g., drinking in excess of one’s school peers). Many (e.g., Balsa et al., 2010; Gaviria and Raphael, 2001) in the literature have argued that the relevant peer group for a teen is his/her classmates. We perform a few robustness checks whereby we investigate the possibility that the effects of PMBs vary by grade and we also consider alternative measures of drinking (e.g., binge drinking) and adult outcomes (e.g., receipt of high school diploma). These exercises support our preliminary conclusions that excessive drinking has a statistically significant effect for females on educational outcomes but the sign varies by choice of IV. We find a negative (positive) and statistically significant effect of excessive smoking (past sexual activity) on male (female) academic accomplishments. The inclusion of school-level activities yield positive, significant, and sizable effects, however. This suggests that participation in extra curricular activities are important in determining an adolescent’s long-run trajectory and may even offset any negative effects associated with PMBs. These results have implications for public policy both in terms of education and the continued funding of extra curricular activities.

We investigate a major supply shock to the US heroin market in the 1990s, the introduction of Colombian-sourced heroin, that led to a substantial rise in heroin overdose admissions. The instrumental variables approach uses the interaction of the timing of the supply shock with city-level pre-shock characteristics that have been shown to facilitate or hinder the introduction of new heroin sources. The estimation strategy allows us to disentangle the causal effects of multiple city-level heroin market factors that are correlated with overdose: the price per pure gram of heroin, the coefficient of variation of purity of heroin, and the proportion of heroin of Colombian origin. We find that changes in the price per pure gram and country of origin have substantial effects on heroin overdose admissions, explaining about two-thirds of the rise in heroin overdoses over the 1990s. The results have important implications for understanding the effects of changing heroin market conditions on the current US heroin epidemic.

In 1956, 52% of urban men and 42% of rural men smoked cigarettes. By 2010, while the prevalence of smoking was dramatically lower for both groups, the disparity had flipped: 24.7% of urban men and 30.6% of rural men smoked and similar patterns exist for women. This paper puts forth a new potential explanation for the observed urban/rural smoking trends: selective internal migration. Between 1950 and 2010, the share of the United States population living in an urban area increased from 64% to 81%. If relatively more educated, higher-SES individuals are driving urbanization, as seems likely from the economics literature on migration, then migration and the well-known correlation between smoking and education may generate relatively fewer urban smokers. On the other hand, if smokers are more likely to move from rural to urban areas, perhaps seeking better job opportunities, then observed disparities between urban and rural populations may be understated. If smoking is correlated with clear determinants of migration (e.g., job opportunities, education, income, etc.), then migration flows will alter the smoking composition of local areas over time. The potential for selective migration matters for the evaluation of tobacco control policies, both overall and by location. The vast majority of studies of cigarette taxes and indoor smoking bans on smoking prevalence use repeated cross-sectional data and research designs which assume that the composition of urban and rural populations remains fixed over time. These designs typically use within-state variation in smoking prevalence to identify the effect of changes in policy on smoking prevalence. If net migration changes the smoking composition of a state or county, then the estimated effect of changes in tobacco control policy will reflect both the policy and the smoking composition change. To address the potential for selective migration to a.) explain the change in urban/rural smoking disparity and b.) bias our understanding of tobacco control policy, I simulate a Roy model of selective migration. The model demonstrates the conditions under which migration shifts the smoking composition of a local area. Next, I merge data from the Current Population Survey's Tobacco Use Survey and Annual Social and Economic Supplement from 1993 through 2015, and I demonstrate a strongly positive correlation between smoking and migration. Finally, I estimate the effect of state-level cigarette taxes and indoor smoking bans on smoking behavior after controlling for nonrandom population movement (i.e., selective migration). My approach is to estimate the probability of migration into a given state conditional on socio-economic characteristics. The estimated probabilities then enter a control function which, in a regression of smoking on these local area policies, nets out the effect of shifting population characteristics. The result of this analysis will allow for me to decompose the within-state trend in smoking to a.) tobacco control policies, b.) selective migration.

The notion that public transportation can mitigate accidents has been widely claimed but to-date empirical evidence that supports this relationship in a causal manner is scarce. We present results from difference-in-differences (DID) and triple differences (DDD) frameworks that exploit the introduction of late-night buses (night buses) into cities in Israel beginning in 2007. Our preferred DDD estimation utilizes spatial, temporal, and time-of-day variation in estimating the effect of late-night bus frequencies on accident outcomes. The results show a reduction in accidents involving young drivers in response to night buses, on the order of magnitude of 37% in the mean metropolitan area served by night buses. Injuries resulting from these accidents also decrease by 24%. Our results are robust to alternative DDD estimations, which utilize variation in the day of the week that night buses operate. The reduction in the number of injuries is less than the reduction in overall accidents, despite the analysis being based on data that documents only accidents involving at least one injury. Due to this finding, we proceed to evaluate the effect of late-night buses on the severity of accidents and find that the number injured per accident increased following the introduction of late-night buses, despite the overall reduction in accidents. This suggests that late-night buses are increasing the propensity to consume alcohol both among late-night bus users and among private car users not utilizing late-night buses. Thus, late-night buses are generating a positive alcohol consumption externality, which in turn increases the severity of accidents occurring. Overall, the results suggest that public transportation - and in particular late-night public transportation - can entail substantial benefits in terms of road accident reductions. Nevertheless, while there is a reduction in the number of accidents and in the number of injured, the accidents that do end up occurring are more severe, due to the alcohol consumption externality that late-night buses intended for late-night outings generate.

A major goal of tobacco control policies is to improve the health of teenagers by affecting their smoking behavior. Past economic research suggests these policies reduce youth smoking, although the effect may have waned in recent years. However, little is known about the spillover effects these policies may have on other health behaviors in teenagers, and the limited existing work has focused on obesity. Our research is the first to examine the effects of tobacco policies on teenagers’ physical activity. Both smoking and exercise behaviors are habitual and likely formed during adolescence, making this age group of particular interest. Moreover, the benefits of physical activity for teenagers are well established and include building healthy bones and muscles, controlling weight, reducing anxiety/stress and increasing self-esteem. The spillover effects of tobacco policies on teenagers’ physical activity therefore have strong implications for the efficacy of these policies in improving overall health. Our research builds on the conceptual framework of Conway and Niles (2017), which reveals the theoretically ambiguous spillover effects of tobacco policies on adult exercise. A tax-induced reduction in smoking, for example, may increase one’s ability to exercise making exercising more desirable than before. Likewise, exercise and smoking may both be considered strategies for weight management, such that increased exercise is a preventive measure against the weight gain associated with reduced smoking. Conversely, if people believe that physical activity reduces some of the ill effects of smoking, then they may use physical activity as a way to compensate for the harms caused by smoking. Finally, an increase in the cost of cigarettes could have income effects. Examining teenagers’ physical activity introduces two new challenges. One is that required physical education classes mean that exercise may not always be voluntary. The second is the heightened ability to join a sports team, which likely facilitates physical activity but may serve other purposes for the teen (e.g., social).

Our empirical analyses address these two challenges by investigating teenagers’ participation in Physical Education classes and sports teams, as well as their smoking and exercise behaviors. Using repeated cross-sectional data from the 1991 to 2015 Youth Risk Behavior Survey (YRBS), combined with state-level policies and controls, our research is the first to provide evidence on how tobacco control policies, including cigarette taxes, smoking bans and anti-tobacco spending, affect these physical activity behaviors in adolescents. This time period is of particular relevance as it includes substantial changes in cigarette taxes and smoking bans, as well as the Master Settlement Agreement (MSA) of 1998 that resulted in funds targeted to preventing new smokers. Our preliminary smoking results agree with past work; increased cigarette taxes has a negative effect on smoking but the effect is fairly weak in recent years. Results for spillover effects suggest that increased cigarette taxes have offsetting effects on two key conduits for teen physical activity -- an increase in sports team membership but a decrease in days of Physical Education. On net, cigarette taxes appear to have a weakly negative effect on teen physical activity.

Over the past 15 years, opioid abuse and overdose has more than tripled in the United States. Additionally, programs like DARE have been teaching students about the dangers of gateway drugs and how they can lead to opioid abuse later in life. My research examines whether gateway drugs are actually common precursors to later opioid dependence in observational data, for what subpopulations, and whether this has changed over time. Future research using different methods would be needed to understand the causal effects of gateway drugs, but currently there is not widespread knowledge of how often these patterns are actually observed in practice. I use data from several years of the National Survey on Drug Use and Health (NSDUH), a Centers for Drug Control and Prevention (CDC) annual survey of approximately 70,000 individuals that asks about current and historical drug use as well as a comprehensive set of demographic and economic characteristics. In particular, NSDUH questions respondents about age of first use for cigarettes, alcohol, marijuana, opioids, heroin, meth etc. allowing me to examine frequency of different drug pathways, as well as to document other correlates of opioid dependence. By analyzing NSDUH data over the period 1979-2016, I present evidence on common paths observed to opioid abuse later in life, enabling an initial assessment of the role of traditional gateway drugs in eventual opioid dependence.

Abstract

The role of state-level background check requirements for private firearm sales in reducing gun violence remains controversial in both the empirical literature and gun control policy debate. On August 28, 2007 the Missouri General Assembly repealed an 86 year-old “permit- to-purchase" (PTP) law requiring that handgun purchasers possess a permit, and subsequently undergo a background check, for all sales. The vast racial disparities in firearm homicide within Missouri raises important questions concerning the disproportionate impact of the repeal on Black communities throughout the state. Using generalized synthetic control estimation, this paper finds that the PTP repeal led to a modest increase in county-level gun ownership in addition to substantial evidence of increased firearm homicide in the early years of the 2007- 2013 post-repeal period. In particular, state-level effects suggests that overall Black firearm homicide increases on average by an additional five deaths per 100,000 while the same rates for Black victims ages 15-24 rise by 29 deaths per 100,000. County-level estimates also show considerable increases in firearm homicide in Black communities within the more urban regions of the state. Treatment effect estimates for state-level Black firearm homicide translate into approximately an additional 260 deaths attributable to the change in the law over the 2007-2013

: To examine the availability of medical marijuana dispensaries, price of medical marijuana products, and variety of medical marijuana products in school neighborhoods and their associations with adolescents’ use of marijuana

graders (N=46,646) from 117 randomly selected schools in California participated in the cross-sectional 2015-16 California Student Tobacco Survey (CSTS). Characteristics of medical marijuana dispensaries in California were collected and combined with school locations to compute availability, price, and product variety of medical marijuana in school neighborhoods. Multilevel logistic regressions with random intercepts at school level were conducted to test the associations, accounting for individual and school socioeconomic characteristics.

: The distance from school to the nearest medical marijuana dispensary (within 0-1 mile and 1-3 mile bands) was not associated with adolescents’ use of marijuana in the past month or susceptibility to use marijuana in the future, nor was the weighted count of medical marijuana dispensaries within the 3-mile band of school. Neither the product price nor the product variety in the dispensary nearest to school was associated with marijuana use or susceptibility to

: There was no evidence supporting the associations of medical marijuana availability, price, or product variety around school with adolescents’ marijuana use and susceptibility to use.

In the past five years, 3 states and over 100 localities have raised their tobacco sales ages to 21. This paper is the first nationally representative analysis to estimate this policy’s effects on late adolescent smoking. Specifically, difference-in-differences analyses use the Behavioral Risk Factor Surveillance System’s Selected Metropolitan/Micropolitan Area Risk Trends (SMART) data to test how age-21 tobacco sales restrictions impact conventional cigarette use among 18 to 20

The analytic sample is restricted to metropolitan statistical areas and metropolitan divisions included in every year of the 2011-2015 SMART data. Specifically, these are areas for which the nationally representative BRFSS surveys interviewed at least 500 respondents in each survey-year. As all states adopting age-21 laws implemented their policies in 2016 or later, the analyses presented here are based on policies implemented at the sub-state level. Limiting consideration to 18 to 20 year olds interviewed between 2011 and 2015 yields a sample size of 22,995 respondents, 22,131 of whom responded to the survey’s current smoking questions. The outcome variable of interest is a binary “current smoker” indicator. For each respondent, the percent of the population covered by an age-21 tobacco sales restriction is calculated for their MMSA-by-state as of their interview date. Difference-in-differences regressions evaluate whether having a higher likelihood of exposure to an age-21 tobacco purchasing restriction yields a differential likelihood of current smoking, controlling for geographic unit and year fixed effects, as well as respondent demographics and an array of other policy variables. This specification passes the requisite parallel trend tests. Baseline findings indicate that exposure to an age-21 tobacco sales restriction at interview yields a statistically significant 4.8 percentage point drop in one’s likelihood of being a current smoker. Since the vast majority of age-21 laws were implemented in the Northeast and Mid-Atlantic Census Divisions, specification checks restrict the sample to these areas and find a slightly lower but still statistically significant 2.4 percentage point drop in current smoking associated with

Notably, among respondents living in areas with non-zero exposure to tobacco-21 restrictions, the mean likelihood of exposure is only 9.0% in the full sample, as compared to 38.3% in the Northeast and Mid-Atlantic Census Divisions. Thus, for the average respondent who was exposed to these policies, we would expect a 0.4 percentage point drop in smoking relative to unexposed respondents nation-wide (0.048*0.09 = 0.004), versus a 0.9 percentage point drop relative to unexposed respondents in the Northeast and Mid-Atlantic Census Divisions (0.024*0.383 = 0.009). Falsification tests repeat the main analysis using 23 to 25 year-old respondents, a group not bound by the age-21 restrictions. In this case, age-21 tobacco restrictions yield statistically insignificant effects on current smoking. Overall, these results indicate that restricting tobacco sales to individuals under age-21 yields a statistically significant reduction in smoking among 18 to 20 year olds, on the order of 0.4 to 0.9 percentage points.


This paper identifies the causal treatment effect of KR laws on underage alcohol consumption and related outcomes: alcohol-related traffic fatalities, by exploiting the substantial variations in the timing of the introduction of these laws across different states at different times. Keg registration (KR) laws require alcohol retailers or wholesalers attach a registered label to the beer kegs that they sell. These laws aim to reduce underage illegal alcohol consumption by imposing liability on adults who purchase beer kegs or host keg parties. Although, in recent years, an increasing number of states have adopted the KR laws to control underage alcohol use and abuse, empirical evidence of the effectiveness of this policy is quite limited. To correct for slection bias and policy endogeneity problem, we matched our treatment samples with a comparable control samples using propensity score matching method. Using the matched sample that created, we conducted difference-in-difference analysis and suggest that the introduction of the KR laws is associated with up to a 2.3 percentage point reduction in binge drinking among minors. This statistical significant reduction is mainly driven by male minors. Furthermore, our results show that strict KR laws can significantly decrease the number of alcohol-involved traffic fatalities by 0.292 among 17 year-old minors and 0.319 among 15-17 year-old minors. Our results are

State-level efforts to legalize marijuana have given consumers new choices in the market for substances beyond alcohol and tobacco. Setting optimal tax policies in these states, for public health or revenue motives, may require an understanding of the degree to which consumers are will to substitute between different sin goods. We use detailed data on marijuana, tobacco, and alcohol sales in Washington state to estimate a demand system for substances that allows for flexible substitution patterns. We estimate the price elasticity of demand for marijuana in -1.32, and the income elasticity is 1.88. The legalization of marijuana in Washington reduced demand for alcohol by 0.8% and reduced demand for tobacco by 0.2%. Post-legalization, we find that an increase in marijuana prices increases alcohol sales slightly but does not affect tobacco sales. Our results suggest that the market for sin goods can effectively be modeled as a

. To examine the impacts of tetrahydrocannabinol (THC) level, cannabidiol (CBD) level, warning messages, and price on adults’ preference for marijuana products. . An online discrete choice experiment was implemented in October 2017. Each participant was randomly assigned to 1 out of 6 choice sets, each including 12 randomly-ordered choice tasks. Each choice task asked the participant to

choose 1 out of 3 alternative marijuana products sold in recreational marijuana stores, with varying levels in 4 attributes (THC level, CBD level, warning message, and price). An opt-out option was also offered in each choice task. The impacts of the attributes on the probability of choosing marijuana products were analyzed using conditional logit regressions, controlling for individual sociodemographic characteristics. Past-year marijuana users and nonusers were

. A general population sample of 2,398 participants (1,200 past-year marijuana users and 1,198 nonusers) aged 18 years or older living in states that had passed the laws to legalize recreational marijuana (California, Colorado, Massachusetts, Nevada, Oregon, and Washington) were recruited to complete the discrete choice experiment from an online panel.

. Among both marijuana users and nonusers, higher price was associated with lower probability of choosing marijuana products (p<.001); compared to CBD-free, very low CBD level (0.4%) was associated with lower probability of choosing marijuana products (p<.001) whereas very high CBD level (15%) was associated with higher probability of choosing marijuana products (p<.001). Higher THC level increased the probability of choosing marijuana products among users but had no impact on nonusers. Compared to no warning label, text warning message currently adopted by Washington and Colorado increased the probability of choosing marijuana products among users (p<.05) and FDA not-approved disclaimer reduced the probability of choosing marijuana products among nonusers (p<.001). The results did not differ by purpose of use (primarily medical or recreational in the past year) among users.

. Restricting potency level, imposing tax, and adopting warning messages that emphasize no approval from FDA may reduce marijuana purchase among adults.

This study evaluates the indirect effects of the availability of the ride-sharing service, Uber, on alcohol consumption in the United States. Uber is a smartphone application that immediately connects riders to willing drivers and provides them with: real-time location tracking of the vehicle, advance information on the cost of the ride and a convenient phone payment system. Given the ease of use, Uber provides a convenient alternative to drunk driving, one of the largest contributors to traffic accidents and traffic deaths in the United States, especially among 21-34 olds. Recent evidence in working-papers by Martin-Buck (2017) and Dills and Mulholland (2017) has shown that there have been reductions in drunk driving fatalities and DUI/DWI arrests associated with the ride-sharing service availability, using national data from the Fatality Analysis Reporting System and the Uniform Crime Reporting Program. Additionally, they find a decrease in physical and sexual violence, suggesting that the ride-sharing service may reduce altercations over transportation and/or impaired decision-making over the course of a night out. However, to date the research has overlooked the first order, albeit indirect effects of Uber on the act of drinking itself. There are two potential avenues of influence on alcohol consumption. First, there may be an increase at the extensive margin: the total number of individuals who consume alcohol on a given night. The magnitude of this effect relies on the number of drivers that are freed up from their responsibilities as a sober driver to participate in the evening’s festivities. Second, there could be an increase in the intensity of alcohol consumption. We use a difference-in-difference framework to test the effect of Uber’s arrival on the frequency and intensity of drinking, and frequency of drinking and driving as self-reported in the Behavior Risk Factor Surveillance System (BRFSS) between 2007 and 2015. Through these analyses we can glean insights into the metamorphosis of social behaviors in the wake of the new technology. We find that two forms of Uber’s service, UberBlack and UberX, are not associated with changes in drinking behavior while another option is associated with decreases in alcohol consumption. On the other hand, we find the Uber option that provides the lowest average cost to consumers, UberXL, is associated with large increases in excessive drinking, particularly among those aged 25-39.

We estimated a rational addiction model using a 10 year-panel from the National Longitudinal Survey of Youth 1997 Cohort. After controlling for individual heterogeneity, we find that drinking and smoking show a somewhat rational addiction pattern, with a moderate degree of addiction. The level of present bias or time preference differs by behaviors. Drinking participation is likely a rational choice, with a corresponding 2.8% estimated interest rate. In contrast, heavy drinking, smoking, and daily smoking, despite responding to future participation, are also subject to present bias and a heavy discount of future consequences. Instrumental variable results are more likely pertaining to myopic users who are not forward-looking and respond to policies. In sum, regulatory policies are less likely to influence rational users and are more likely to influence myopic users who have self-control or heavy discounting issues. Policy makers should

Over past several decades, smoking rates in the USA have substantially declined, however, the public health dangers from smoking are still prevalent as approximately 36.5 million individuals continue to smoke daily. Smokers aren’t the only individuals affected by smoking; nonsmokers face Environmental Tobacco Smoke (ETS) -- commonly referred to as Secondhand Smoke -- which is responsible for numerous deaths worldwide and is known to cause major health problems for adults and children. To combat the threat of ETS, many localities have enacted smoke-free laws banning smoking in a variety of public and work places. These smoke-free laws have had little effect on decreasing smoking, but have generally lead to a reduction in exposure to ETS by nonsmokers and have altered the behavior of smokers by displacing them from their prior smoking locations to private homes and other outdoor locations (Adda and Cornaglia, 2010, Carpenter, Postolek, and Warman, 2011). This behavioral change in smokers and the displacement of ETS requires further understanding as to how smoke-free laws impacted patterns of smoking, as ETS is threatening populations

In this paper, I use a novel observational dataset collected over several months on New York City streets to investigate smoking displacement and smoking patterns. Since ETS has greatest negative health impact on children, I collected the data to document smoking patterns at times when more children were walking to and from schools. I find that nonsmokers walking in the city are exposed to ETS approximately once every 1.9 city blocks. Upon matching the observational data to city street characteristics, I find that smoking patterns on city streets are nonrandom and are associated with the presence of institutions that banned smoking due to smoke-free laws. I find strong, positive associations between the

While my observational smoking data shows an association between smoking location patterns and places that banned smoking, it does not establish a causal link. To further investigate smoking displacement and smoking patterns, I use the NY Youth Smoking Survey and NY Adult Smoking Survey from the NY State Department of Health to examine smoking location changes over the past fifteen years. Unlike my primary data, the survey questions do not provide great detail on where individuals smoke, however, the surveys show that among active smokers, smoking has declined at prohibited locations. This supports the smoking displacement hypothesis and bolsters the evidence I observe through my primary data. Although previous research has documented smoking displacement, the main contribution of my paper is to show that smoking displacement may be occurring in previously undocumented locations and that further data collection and changes to survey questions need to take place in order to better understand which populations are affected by ETS, how much they are affected, and what policy needs to be implemented to further limit negative impacts

This paper explores the impacts of pseudo-mature behaviors (PMBs), school activities, and social identity on early adult outcomes, specifically the likelihood of college attendance and annual earnings. We define the PMBs to include (binge) drinking alcohol, cigarette smoking, and sexual activity. A student's school activities involve his/her participation in sports, non-sports, and mixed clubs. The incorporation of such measures into our analysis highlights the importance of social identity, a concept that is popular in sociology but often not traditionally emphasized in economics. We use data obtained from the Restricted-Use National Longitudinal Study of Adolescent Health (Add Health). A two-step estimation procedure is used that allows for the possible endogeneity of the PMBs. Much of the literature has struggled to find good identifying instruments; French and Popovici (2011) provide a nice summary of the approaches undertaken by researchers to identify non-weak instruments (IVs). Correspondingly, we use maternal binge drinking and popularity as per Renna (2007, 2009) and Mundt and French (2013), respectively, and also control for a single parent household, which is supported by Peer Cluster Theory, but new to the literature. Our analysis also controls for peer-group effects with the incorporation of grade fixed effects and robust errors that are clustered at the school-level. The primary empirical work is conducted using relative PMB measures (e.g., drinking in excess of one’s school peers). Many (e.g., Balsa et al., 2010; Gaviria and Raphael, 2001) in the literature have argued that the relevant peer group for a

We perform a few robustness checks whereby we investigate the possibility that the effects of PMBs vary by grade and we also consider alternative measures of drinking (e.g., binge drinking) and adult outcomes (e.g., receipt of high school diploma). These exercises support our preliminary conclusions that excessive drinking has a statistically significant effect for females on educational outcomes but the sign varies by choice of IV. We find a negative (positive) and statistically significant effect of excessive smoking (past sexual activity) on male (female) academic accomplishments. The inclusion of school-level activities yield positive, significant, and sizable effects, however. This suggests that participation in extra curricular activities are important in determining an adolescent’s long-run trajectory and may even offset any negative effects associated with PMBs. These results have implications for public policy both in terms of education and the

We investigate a major supply shock to the US heroin market in the 1990s, the introduction of Colombian-sourced heroin, that led to a substantial rise in heroin overdose admissions. The instrumental variables approach uses the interaction of the timing of the supply shock with city-level pre-shock characteristics that have been shown to facilitate or hinder the introduction of new heroin sources. The estimation strategy allows us to disentangle the causal effects of multiple city-level heroin market factors that are correlated with overdose: the price per pure gram of heroin, the coefficient of variation of purity of heroin, and the proportion of heroin of Colombian origin. We find that changes in the price per pure gram and country of origin have substantial effects on heroin overdose admissions, explaining about two-thirds of the rise in heroin overdoses over the 1990s. The results have important implications for understanding the

In 1956, 52% of urban men and 42% of rural men smoked cigarettes. By 2010, while the prevalence of smoking was dramatically lower for both groups, the disparity had flipped: 24.7% of urban men and 30.6% of rural men smoked and similar patterns exist for women. This paper puts forth a new potential explanation for the observed urban/rural smoking trends: selective internal migration. Between 1950 and 2010, the share of the United States population living in an urban area increased from 64% to 81%. If relatively more educated, higher-SES individuals are driving urbanization, as seems likely from the economics literature on migration, then migration and the well-known correlation between smoking and education may generate relatively fewer urban smokers. On the other hand, if smokers are more likely to move from rural to urban areas, perhaps seeking better job opportunities, then observed disparities between urban and rural populations may be understated. If smoking is correlated with clear determinants of migration (e.g., job opportunities, education, income, etc.), then migration flows will alter the smoking composition of local areas over time. The potential for selective migration matters for the evaluation of tobacco control policies, both overall and by location. The vast majority of studies of cigarette taxes and indoor smoking bans on smoking prevalence use repeated cross-sectional data and research designs which assume that the composition of urban and rural populations remains fixed over time. These designs typically use within-state variation in smoking prevalence to identify the effect of changes in policy on smoking prevalence. If net migration changes the smoking composition of a state or county, then the estimated effect of changes in tobacco control policy will reflect both the policy and the smoking composition change. To address the potential for selective migration to a.) explain the change in urban/rural smoking disparity and b.) bias our understanding of tobacco control policy, I simulate a Roy model of selective migration. The model demonstrates the conditions under which migration shifts the smoking composition of a local area. Next, I merge data from the Current Population Survey's Tobacco Use Survey and Annual Social and Economic Supplement from 1993 through 2015, and I

correlation between smoking and migration. Finally, I estimate the effect of state-level cigarette taxes and indoor smoking bans on smoking behavior after controlling for nonrandom population movement (i.e., selective migration). My approach is to estimate the probability of migration into a given state conditional on socio-economic characteristics. The estimated probabilities then enter a control function which, in a regression of smoking on these local area policies, nets out the effect of shifting population characteristics. The result of this analysis will allow for me to decompose the within-state trend in smoking to a.) tobacco control policies, b.) selective migration.

The notion that public transportation can mitigate accidents has been widely claimed but to-date empirical evidence that supports this relationship in a causal manner is scarce. We present results from difference-in-differences (DID) and triple differences (DDD) frameworks that exploit the introduction of late-night buses (night buses) into cities in Israel beginning in 2007. Our preferred DDD estimation utilizes spatial, temporal, and time-of-day variation in estimating the effect of late-night bus frequencies on accident outcomes. The results show a reduction in accidents involving young drivers in response to night buses, on the order of magnitude of 37% in the mean metropolitan area served by night buses. Injuries resulting from these accidents also decrease by 24%. Our results are robust to alternative DDD estimations, which utilize variation in the day of the week that night buses operate. The reduction in the number of injuries is less than the reduction in overall accidents, despite the analysis being based on data that documents only accidents involving at least one injury. Due to this finding, we proceed to evaluate the effect of late-night buses on the severity of accidents and find that the number injured per accident increased following the introduction of late-night buses, despite the overall reduction in accidents. This suggests that late-night buses are increasing the propensity to consume alcohol both among late-night bus users and among private car users not utilizing late-night buses. Thus, late-night buses are generating a positive alcohol consumption externality, which in turn increases the severity of accidents occurring. Overall, the results suggest that public transportation - and in particular late-night public transportation - can entail substantial benefits in terms of road accident reductions. Nevertheless, while there is a reduction in the number of accidents and in the number of injured, the accidents that do end up occurring are more severe, due to the alcohol consumption externality that late-night buses intended for late-night outings

A major goal of tobacco control policies is to improve the health of teenagers by affecting their smoking behavior. Past economic research suggests these policies reduce youth smoking, although the effect may have waned in recent years. However, little is known about the spillover effects these policies may have on other health behaviors in teenagers, and the limited existing work has focused on obesity. Our research is the first to examine the effects of tobacco policies on teenagers’ physical activity. Both smoking and exercise behaviors are habitual and likely formed during adolescence, making this age group of particular interest. Moreover, the benefits of physical activity for teenagers are well established and include building healthy bones and muscles, controlling weight, reducing anxiety/stress and increasing self-esteem. The spillover effects of tobacco policies on teenagers’ physical activity therefore have strong implications for the

Our research builds on the conceptual framework of Conway and Niles (2017), which reveals the theoretically ambiguous spillover effects of tobacco policies on adult exercise. A tax-induced reduction in smoking, for example, may increase one’s ability to exercise making exercising more desirable than before. Likewise, exercise and smoking may both be considered strategies for weight management, such that increased exercise is a preventive measure against the weight gain associated with reduced smoking. Conversely, if people believe that physical activity reduces some of the ill effects of smoking, then they may use physical activity as a way to compensate for the harms caused by smoking. Finally, an increase in the cost of cigarettes could have income effects. Examining teenagers’ physical activity introduces two new challenges. One is that required physical education classes mean that exercise may not always be voluntary. The second is the heightened ability to join a sports team, which likely facilitates physical activity but may serve other purposes for the teen (e.g., social).

Our empirical analyses address these two challenges by investigating teenagers’ participation in Physical Education classes and sports teams, as well as their smoking and exercise behaviors. Using repeated cross-sectional data from the 1991 to 2015 Youth Risk Behavior Survey (YRBS), combined with state-level policies and controls, our research is the first to provide evidence on how tobacco control policies, including cigarette taxes, smoking bans and anti-tobacco spending, affect these physical activity behaviors in adolescents. This time period is of particular relevance as it includes substantial changes in cigarette taxes and smoking bans, as well as the Master Settlement Agreement (MSA) of 1998 that resulted in funds targeted to preventing new smokers. Our preliminary smoking results agree with past work; increased cigarette taxes has a negative effect on smoking but the effect is fairly weak in recent years. Results for spillover effects suggest that increased cigarette taxes have offsetting effects on two key conduits for teen physical activity -- an increase in sports team membership but a decrease in days of Physical Education. On net, cigarette taxes appear to

Over the past 15 years, opioid abuse and overdose has more than tripled in the United States. Additionally, programs like DARE have been teaching students about the dangers of gateway drugs and how they can lead to opioid abuse later in life. My research examines whether gateway drugs are actually common precursors to later opioid dependence in observational data, for what subpopulations, and whether this has changed over time. Future research using different methods would be needed to understand the causal effects of gateway drugs, but currently there is not widespread knowledge of how often these patterns are actually observed in practice. I use data from several years of the National Survey on Drug Use and Health (NSDUH), a Centers for Drug Control and Prevention (CDC) annual survey of approximately 70,000 individuals that asks about current and historical drug use as well as a comprehensive set of demographic and economic characteristics. In particular, NSDUH questions respondents about age of first use for cigarettes, alcohol, marijuana, opioids, heroin, meth etc. allowing me to examine frequency of different drug pathways, as well as to document other correlates of opioid dependence. By analyzing NSDUH data over the period 1979-2016, I present evidence on common paths observed to opioid abuse later in life, enabling an initial assessment of the role of traditional


Morgan Williams Jr. [email protected]

Yuyan Shi [email protected]

Abigail Friedman [email protected]


The role of state-level background check requirements for private firearm sales in reducing gun violence remains controversial in both the empirical literature and gun control policy debate. On August 28, 2007 the Missouri General Assembly repealed an 86 year-old “permit- to-purchase" (PTP) law requiring that handgun purchasers possess a permit, and subsequently undergo a background check, for all sales. The vast racial disparities in firearm homicide within Missouri raises important questions concerning the disproportionate impact of the repeal on Black communities throughout the state. Using generalized synthetic control estimation, this paper finds that the PTP repeal led to a modest increase in county-level gun ownership in addition to substantial evidence of increased firearm homicide in the early years of the 2007- 2013 post-repeal period. In particular, state-level effects suggests that overall Black firearm homicide increases on average by an additional five deaths per 100,000 while the same rates for Black victims ages 15-24 rise by 29 deaths per 100,000. County-level estimates also show considerable increases in firearm homicide in Black communities within the more urban regions of the state. Treatment effect estimates for state-level Black firearm homicide translate into approximately an additional 260 deaths attributable to the change in the law over the 2007-2013

: To examine the availability of medical marijuana dispensaries, price of medical marijuana products, and variety of medical marijuana products in school neighborhoods and their associations with adolescents’ use of marijuana

graders (N=46,646) from 117 randomly selected schools in California participated in the cross-sectional 2015-16 California Student Tobacco Survey (CSTS). Characteristics of medical marijuana dispensaries in California were collected and combined with school locations to compute availability, price, and product variety of medical marijuana in school neighborhoods. Multilevel logistic regressions with random

: The distance from school to the nearest medical marijuana dispensary (within 0-1 mile and 1-3 mile bands) was not associated with adolescents’ use of marijuana in the past month or susceptibility to use marijuana in the future, nor was the weighted count of medical marijuana dispensaries within the 3-mile band of school. Neither the product price nor the product variety in the dispensary nearest to school was associated with marijuana use or susceptibility to

: There was no evidence supporting the associations of medical marijuana availability, price, or product variety around school with adolescents’ marijuana use and susceptibility to use.

In the past five years, 3 states and over 100 localities have raised their tobacco sales ages to 21. This paper is the first nationally representative analysis to estimate this policy’s effects on late adolescent smoking. Specifically, difference-in-differences analyses use the Behavioral Risk Factor Surveillance System’s Selected Metropolitan/Micropolitan Area Risk Trends (SMART) data to test how age-21 tobacco sales restrictions impact conventional cigarette use among 18 to 20

The analytic sample is restricted to metropolitan statistical areas and metropolitan divisions included in every year of the 2011-2015 SMART data. Specifically, these are areas for which the nationally representative BRFSS surveys interviewed at least 500 respondents in each survey-year. As all states adopting age-21 laws implemented their policies in 2016 or later, the analyses presented here are based on policies implemented at the sub-state level. Limiting

The outcome variable of interest is a binary “current smoker” indicator. For each respondent, the percent of the population covered by an age-21 tobacco sales restriction is calculated for their MMSA-by-state as of their interview date. Difference-in-differences regressions evaluate whether having a higher likelihood of exposure to an age-21 tobacco purchasing restriction yields a differential likelihood of current smoking, controlling for geographic unit and year fixed

Baseline findings indicate that exposure to an age-21 tobacco sales restriction at interview yields a statistically significant 4.8 percentage point drop in one’s likelihood of being a current smoker. Since the vast majority of age-21 laws were implemented in the Northeast and Mid-Atlantic Census Divisions, specification checks restrict the sample to these areas and find a slightly lower but still statistically significant 2.4 percentage point drop in current smoking associated with

Notably, among respondents living in areas with non-zero exposure to tobacco-21 restrictions, the mean likelihood of exposure is only 9.0% in the full sample, as compared to 38.3% in the Northeast and Mid-Atlantic Census Divisions. Thus, for the average respondent who was exposed to these policies, we would expect a 0.4 percentage point drop in smoking relative to unexposed respondents nation-wide (0.048*0.09 = 0.004), versus a 0.9 percentage point drop

Falsification tests repeat the main analysis using 23 to 25 year-old respondents, a group not bound by the age-21 restrictions. In this case, age-21 tobacco restrictions yield statistically insignificant effects on current smoking. Overall, these results indicate that restricting tobacco sales to individuals under age-21 yields a statistically significant reduction in smoking among 18 to 20 year olds, on the order of 0.4 to 0.9 percentage points.



Boyoung Seo [email protected]

Yuyan Shi [email protected]

Jennifer Trudeau [email protected]

Ce Shang [email protected]

This paper identifies the causal treatment effect of KR laws on underage alcohol consumption and related outcomes: alcohol-related traffic fatalities, by exploiting the substantial variations in the timing of the introduction of these laws across different states at different times. Keg registration (KR) laws require alcohol retailers or wholesalers attach a registered label to the beer kegs that they sell. These laws aim to reduce underage illegal alcohol consumption by imposing liability on adults who purchase beer kegs or host keg parties. Although, in recent years, an increasing number of states have adopted the KR laws to control underage alcohol use and abuse, empirical evidence of the effectiveness of this policy is quite limited. To correct for slection bias and policy endogeneity problem, we matched our treatment samples with a comparable control samples using propensity score matching method. Using the matched sample that created, we conducted difference-in-difference analysis and suggest that the introduction of the KR laws is associated with up to a 2.3 percentage point reduction in binge drinking among minors. This statistical significant reduction is mainly driven by male minors. Furthermore, our results show that strict KR laws can significantly decrease the number of alcohol-involved traffic fatalities by 0.292 among 17 year-old minors and 0.319 among 15-17 year-old minors. Our results are

State-level efforts to legalize marijuana have given consumers new choices in the market for substances beyond alcohol and tobacco. Setting optimal tax policies in these states, for public health or revenue motives, may require an understanding of the degree to which consumers are will to substitute between different sin goods. We use detailed data on marijuana, tobacco, and alcohol sales in Washington state to estimate a demand system for substances that allows for flexible substitution patterns. We estimate the price elasticity of demand for marijuana in -1.32, and the income elasticity is 1.88. The legalization of marijuana in Washington reduced demand for alcohol by 0.8% and reduced demand for tobacco by 0.2%. Post-legalization, we find that an increase in marijuana prices increases alcohol sales slightly but does not affect tobacco sales. Our results suggest that the market for sin goods can effectively be modeled as a

. An online discrete choice experiment was implemented in October 2017. Each participant was randomly assigned to 1 out of 6 choice sets, each including 12 randomly-ordered choice tasks. Each choice task asked the participant to choose 1 out of 3 alternative marijuana products sold in recreational marijuana stores, with varying levels in 4 attributes (THC level, CBD level, warning message, and price). An opt-out option was also offered in each choice task. The impacts of the attributes on the probability of choosing marijuana products were analyzed using conditional logit regressions, controlling for individual sociodemographic characteristics. Past-year marijuana users and nonusers were

. A general population sample of 2,398 participants (1,200 past-year marijuana users and 1,198 nonusers) aged 18 years or older living in states that had passed the laws to legalize recreational marijuana (California,

. Among both marijuana users and nonusers, higher price was associated with lower probability of choosing marijuana products (p<.001); compared to CBD-free, very low CBD level (0.4%) was associated with lower probability of choosing marijuana products (p<.001) whereas very high CBD level (15%) was associated with higher probability of choosing marijuana products (p<.001). Higher THC level increased the probability of choosing marijuana products among users but had no impact on nonusers. Compared to no warning label, text warning message currently adopted by Washington and Colorado increased the probability of choosing marijuana products among users (p<.05) and FDA not-approved disclaimer reduced the probability of choosing marijuana products among nonusers (p<.001). The results did not differ by purpose of use (primarily medical or recreational in the past year) among users.

This study evaluates the indirect effects of the availability of the ride-sharing service, Uber, on alcohol consumption in the United States. Uber is a smartphone application that immediately connects riders to willing drivers and provides them with: real-time location tracking of the vehicle, advance information on the cost of the ride and a convenient phone payment system. Given the ease of use, Uber provides a convenient alternative to drunk driving, one of the largest contributors to traffic accidents and traffic deaths in the United States, especially among 21-34 olds. Recent evidence in working-papers by Martin-Buck (2017) and Dills and Mulholland (2017) has shown that there have been reductions in drunk driving fatalities and DUI/DWI arrests associated with the ride-sharing service availability, using national data from the Fatality Analysis Reporting System and the Uniform Crime Reporting Program. Additionally, they find a decrease

However, to date the research has overlooked the first order, albeit indirect effects of Uber on the act of drinking itself. There are two potential avenues of influence on alcohol consumption. First, there may be an increase at the extensive margin: the total number of individuals who consume alcohol on a given night. The magnitude of this effect relies on the number of drivers that are freed up from their responsibilities as a sober driver to participate in the evening’s festivities. Second, there could be an increase in the intensity of alcohol consumption. We use a difference-in-difference framework to test the effect of Uber’s arrival on the frequency and intensity of drinking, and frequency of drinking and driving as self-reported in the Behavior Risk Factor Surveillance System (BRFSS) between 2007 and 2015. Through these analyses we can glean insights into the metamorphosis of social behaviors in the wake of the new technology. We find that two forms of Uber’s service, UberBlack and UberX, are not associated with changes in drinking behavior while another option is associated with decreases in alcohol consumption. On the other hand, we find the Uber option that

We estimated a rational addiction model using a 10 year-panel from the National Longitudinal Survey of Youth 1997 Cohort. After controlling for individual heterogeneity, we find that drinking and smoking show a somewhat rational addiction pattern, with a moderate degree of addiction. The level of present bias or time preference differs by behaviors. Drinking participation is likely a rational choice, with a corresponding 2.8% estimated interest rate. In contrast, heavy drinking, smoking, and daily smoking, despite responding to future participation, are also subject to present bias and a heavy discount of future consequences. Instrumental variable results are more likely pertaining to myopic users who are not forward-looking and respond to policies. In sum, regulatory policies are less likely to influence rational users and are more likely to influence myopic users who have self-control or heavy discounting issues. Policy makers should

Gregory Pac [email protected]

Tracy Regan [email protected]

Daniel Rosenblum [email protected]

Over past several decades, smoking rates in the USA have substantially declined, however, the public health dangers from smoking are still prevalent as approximately 36.5 million individuals continue to smoke daily. Smokers aren’t the only individuals affected by smoking; nonsmokers face Environmental Tobacco Smoke (ETS) -- commonly referred to as Secondhand Smoke -- which is responsible for numerous deaths worldwide and is known to cause major health problems for adults and children. To combat the threat of ETS, many localities have enacted smoke-free laws banning smoking in a variety of public and work places. These smoke-free laws have had little effect on decreasing smoking, but have generally lead to a reduction in exposure to ETS by nonsmokers and have altered the behavior of smokers by displacing them from their prior smoking locations to private homes and other outdoor locations (Adda and Cornaglia, 2010, Carpenter, Postolek, and Warman, 2011). This behavioral change in smokers and the displacement of ETS requires further understanding as to how smoke-free laws impacted patterns of smoking, as ETS is threatening populations

In this paper, I use a novel observational dataset collected over several months on New York City streets to investigate smoking displacement and smoking patterns. Since ETS has greatest negative health impact on children, I collected the data to document smoking patterns at times when more children were walking to and from schools. I find that nonsmokers walking in the city are exposed to ETS approximately once every 1.9 city blocks. Upon matching the observational data to city street characteristics, I find that smoking patterns on city streets are nonrandom and are associated with the presence of institutions that banned smoking due to smoke-free laws. I find strong, positive associations between the

While my observational smoking data shows an association between smoking location patterns and places that banned smoking, it does not establish a causal link. To further investigate smoking displacement and smoking patterns, I use the NY Youth Smoking Survey and NY Adult Smoking Survey from the NY State Department of Health to examine smoking location changes over the past fifteen years. Unlike my primary data, the survey questions do not provide great detail on where individuals smoke, however, the surveys show that among active smokers, smoking has declined at prohibited locations. This supports the smoking displacement hypothesis and bolsters the evidence I observe through my primary data. Although previous research has documented smoking displacement, the main contribution of my paper is to show that smoking displacement may be occurring in previously undocumented locations and that further data collection and changes to survey questions need to take place in order to better understand which populations are affected by ETS, how much they are affected, and what policy needs to be implemented to further limit negative impacts

This paper explores the impacts of pseudo-mature behaviors (PMBs), school activities, and social identity on early adult outcomes, specifically the likelihood of college attendance and annual earnings. We define the PMBs to include (binge) drinking alcohol, cigarette smoking, and sexual activity. A student's school activities involve his/her participation in sports, non-sports, and mixed clubs. The incorporation of such measures into our analysis highlights the importance of social identity, a concept that is popular in sociology but often not traditionally emphasized in economics. We use data obtained from the Restricted-Use National Longitudinal Study of Adolescent Health (Add Health). A two-step estimation procedure is used that allows for the possible endogeneity of the PMBs. Much of the literature has struggled to find good identifying instruments; French and Popovici (2011) provide a nice summary of the approaches undertaken by researchers to identify non-weak instruments (IVs). Correspondingly, we use maternal binge drinking and popularity as per Renna (2007, 2009) and Mundt and French (2013), respectively, and also control for a single parent household, which is supported by Peer Cluster Theory, but new to the literature. Our analysis also controls for peer-group effects with the incorporation of grade fixed effects and robust errors that are clustered at the school-level. The primary empirical work is conducted using relative PMB measures (e.g., drinking in excess of one’s school peers). Many (e.g., Balsa et al., 2010; Gaviria and Raphael, 2001) in the literature have argued that the relevant peer group for a

We perform a few robustness checks whereby we investigate the possibility that the effects of PMBs vary by grade and we also consider alternative measures of drinking (e.g., binge drinking) and adult outcomes (e.g., receipt of high school diploma). These exercises support our preliminary conclusions that excessive drinking has a statistically significant effect for females on educational outcomes but the sign varies by choice of IV. We find a negative (positive) and statistically significant effect of excessive smoking (past sexual activity) on male (female) academic accomplishments. The inclusion of school-level activities yield positive, significant, and sizable effects, however. This suggests that participation in extra curricular activities are important in determining an adolescent’s long-run trajectory and may even offset any negative effects associated with PMBs. These results have implications for public policy both in terms of education and the

We investigate a major supply shock to the US heroin market in the 1990s, the introduction of Colombian-sourced heroin, that led to a substantial rise in heroin overdose admissions. The instrumental variables approach uses the interaction of the timing of the supply shock with city-level pre-shock characteristics that have been shown to facilitate or hinder the introduction of new heroin sources. The estimation strategy allows us to disentangle the causal effects of multiple city-level heroin market factors that are correlated with overdose: the price per pure gram of heroin, the coefficient of variation of purity of heroin, and the proportion of heroin of Colombian origin. We find that changes in the price per pure gram and country of origin have substantial effects on heroin overdose admissions, explaining about two-thirds of the rise in heroin overdoses over the 1990s. The results have important implications for understanding the

Michael Darden [email protected]

Shirlee Lichtman-Sadot [email protected]

Rebecca Sen Choudhury [email protected]

In 1956, 52% of urban men and 42% of rural men smoked cigarettes. By 2010, while the prevalence of smoking was dramatically lower for both groups, the disparity had flipped: 24.7% of urban men and 30.6% of rural men smoked and similar patterns exist for women. This paper puts forth a new potential explanation for the observed urban/rural smoking trends: selective internal migration. Between 1950 and 2010, the share of the United States population living in an urban area increased from 64% to 81%. If relatively more educated, higher-SES individuals are driving urbanization, as seems likely from the economics literature on migration, then migration and the well-known correlation between smoking and education may generate relatively fewer urban smokers. On the other hand, if smokers are more likely to move from rural to urban areas, perhaps seeking better job opportunities, then observed disparities between urban and rural populations may be understated. If smoking is correlated with clear determinants of migration (e.g., job opportunities, education, income, etc.), then migration flows will alter the smoking composition of local areas over time. The potential for selective migration matters for the evaluation of tobacco control policies, both overall and by location. The vast majority of studies of cigarette taxes and indoor smoking bans on smoking prevalence use repeated cross-sectional data and research designs which assume that the composition of urban and rural populations remains fixed over time. These designs typically use within-state variation in smoking prevalence to identify the effect of changes in policy on smoking prevalence. If net migration changes the smoking composition of a state or county, then the estimated effect of changes in tobacco control policy will reflect both the policy and the smoking composition change. To address the potential for selective migration to a.) explain the change in urban/rural smoking disparity and b.) bias our understanding of tobacco control policy, I simulate a Roy model of selective migration. The model demonstrates the conditions under which migration shifts the smoking composition of a local area. Next, I merge data from the Current Population Survey's Tobacco Use Survey and Annual Social and Economic Supplement from 1993 through 2015, and I

correlation between smoking and migration. Finally, I estimate the effect of state-level cigarette taxes and indoor smoking bans on smoking behavior after controlling for nonrandom population movement (i.e., selective migration). My approach is to estimate the probability of migration into a given state conditional on socio-economic characteristics. The estimated probabilities then enter a control function which, in a regression of smoking on these local area policies, nets out the effect of shifting population characteristics. The result of this analysis will allow for me to decompose the within-state trend in smoking to a.) tobacco control policies, b.) selective migration.

The notion that public transportation can mitigate accidents has been widely claimed but to-date empirical evidence that supports this relationship in a causal manner is scarce. We present results from difference-in-differences (DID) and triple differences (DDD) frameworks that exploit the introduction of late-night buses (night buses) into cities in Israel beginning in 2007. Our preferred DDD estimation utilizes spatial, temporal, and time-of-day variation in estimating the effect of late-night bus frequencies on accident outcomes. The results show a reduction in accidents involving young drivers in response to night buses, on the order of magnitude of 37% in the mean metropolitan area served by night buses. Injuries resulting from these accidents also decrease by 24%. Our results are robust to alternative DDD estimations, which utilize variation in the day of the week that night buses operate. The reduction in the number of injuries is less than the reduction in overall accidents, despite the analysis being based on data that documents only accidents involving at least one injury. Due to this finding, we proceed to evaluate the effect of late-night buses on the severity of accidents and find that the number injured per accident increased following the introduction of late-night buses, despite the overall reduction in accidents. This suggests that late-night buses are increasing the propensity to consume alcohol both among late-night bus users and among private car users not utilizing late-night buses. Thus, late-night buses are generating a positive alcohol consumption externality, which in turn increases the severity of accidents occurring. Overall, the results suggest that public transportation - and in particular late-night public transportation - can entail substantial benefits in terms of road accident reductions. Nevertheless, while there is a reduction in the number of accidents and in the number of injured, the accidents that do end up occurring are more severe, due to the alcohol consumption externality that late-night buses intended for late-night outings

A major goal of tobacco control policies is to improve the health of teenagers by affecting their smoking behavior. Past economic research suggests these policies reduce youth smoking, although the effect may have waned in recent years. However, little is known about the spillover effects these policies may have on other health behaviors in teenagers, and the limited existing work has focused on obesity. Our research is the first to examine the effects of tobacco policies on teenagers’ physical activity. Both smoking and exercise behaviors are habitual and likely formed during adolescence, making this age group of particular interest. Moreover, the benefits of physical activity for teenagers are well established and include building healthy bones and muscles, controlling weight, reducing anxiety/stress and increasing self-esteem. The spillover effects of tobacco policies on teenagers’ physical activity therefore have strong implications for the

Our research builds on the conceptual framework of Conway and Niles (2017), which reveals the theoretically ambiguous spillover effects of tobacco policies on adult exercise. A tax-induced reduction in smoking, for example, may increase one’s ability to exercise making exercising more desirable than before. Likewise, exercise and smoking may both be considered strategies for weight management, such that increased exercise is a preventive measure against the weight gain associated with reduced smoking. Conversely, if people believe that physical activity reduces some of the ill effects of smoking, then they may use physical activity as a way to compensate for the harms caused by smoking. Finally, an increase in the cost of cigarettes could have income effects. Examining teenagers’ physical activity introduces two new challenges. One is that required physical education classes mean that exercise may not always be voluntary. The second

Our empirical analyses address these two challenges by investigating teenagers’ participation in Physical Education classes and sports teams, as well as their smoking and exercise behaviors. Using repeated cross-sectional data from the 1991 to 2015 Youth Risk Behavior Survey (YRBS), combined with state-level policies and controls, our research is the first to provide evidence on how tobacco control policies, including cigarette taxes, smoking bans and anti-tobacco spending, affect these physical activity behaviors in adolescents. This time period is of particular relevance as it includes substantial changes in cigarette taxes and smoking bans, as well as the Master Settlement Agreement (MSA) of 1998 that resulted in funds targeted to preventing new smokers. Our preliminary smoking results agree with past work; increased cigarette taxes has a negative effect on smoking but the effect is fairly weak in recent years. Results for spillover effects suggest that increased cigarette taxes have offsetting effects on two key conduits for teen physical activity -- an increase in sports team membership but a decrease in days of Physical Education. On net, cigarette taxes appear to

Anna Harris [email protected]

Over the past 15 years, opioid abuse and overdose has more than tripled in the United States. Additionally, programs like DARE have been teaching students about the dangers of gateway drugs and how they can lead to opioid abuse later in life. My research examines whether gateway drugs are actually common precursors to later opioid dependence in observational data, for what subpopulations, and whether this has changed over time. Future research using different methods would be needed to understand the causal effects of gateway drugs, but currently there is not widespread knowledge of how often these patterns are actually observed in practice. I use data from several years of the National Survey on Drug Use and Health (NSDUH), a Centers for Drug Control and Prevention (CDC) annual survey of approximately 70,000 individuals that asks about current and historical drug use as well as a comprehensive set of demographic and economic characteristics. In particular, NSDUH questions respondents about age of first use for cigarettes, alcohol, marijuana, opioids, heroin, meth etc. allowing me to examine frequency of different drug pathways, as well as to document other correlates of opioid dependence. By analyzing NSDUH data over the period 1979-2016, I present evidence on common paths observed to opioid abuse later in life, enabling an initial assessment of the role of traditional


Complete

University of California, San Diego Shu-hong Zhu; Sharon Cummins Complete

Yale University Complete


CUNY Graduate Center Economics Department/National Bureau of Economic Research


Kelley School of Business, Indiana University Keaton Miller Complete

University of California, San Diego Rosalie Pacula; Ying Cao Complete

Sacred Heart University Benjamin Brewer Complete

IHRP, UIC Frank Chaloupka Complete

Alfred University Complete

Boston College Choon Sung Lim Complete

Dalhousie University Daniel Ciccarone; Jay Unick Complete

George Washington University Complete

Ben-Gurion University of the Negev Complete

University of Nwe Hampshire Complete

CompleteIndiana University, School of Public and Environmental Affairs

ashecon.orgashecon.org/wp-content/uploads/2017/12/S… · XLS file · Web view · 2017-12-07Program Title Abstract Presenting Author Presenting Author Email Address Presenting Author

Documents