T HE W AR ON P OVERTY ’ S E XPERIMENT IN P UBLIC M EDICINE : C OMMUNITY H EALTH C ENTERS AND THE M ORTALITY OF O LDER A MERICANS Martha J. Bailey 12 and.

THE WAR ON POVERTY’S EXPERIMENT IN PUBLIC MEDICINE:

COMMUNITY HEALTH CENTERS AND THE MORTALITY OF OLDER AMERICANS

Martha J. Bailey12 and Andrew Goodman-Bacon1

1 University of Michigan2 National Bureau of Economic Research

COMMUNITY HEALTH CENTERS (CHCS)1. Public provision of primary care and

subsidies for drugs

2. Model: Affordable, convenient, comprehensive and efficient delivery • located in poor communities• outreach, health education• in-house pharmacy• home visits and transportation• employed members of the community

3. Lofty ideals: Transform the model of health care for the poor

WHY CARE ABOUT CHCS?

• CHCS HAVE SURVIVED TO 2010– “Community”/“Federally-Qualified” Health Centers

• CHCS IMPORTANT IN HEALTH REFORM– ARRA: funds 126 new and 1,100 existing CHCs– ACA: over $11 billion – By 2015, CHCs should serve 40 million Americans

WHAT IS KNOWN ABOUT CHCS?

• Medical and Health Services Literature– Increase access (Hadley and Cunningham ‘04, LoSasso and Byck ‘10)– Reduce ER visits (Hochheiser, Woodward and Charney ‘71, Gold and

Rosenberg ‘74)– Reduce preventable hospital admissions (Epstein ‘01, Falik et al. ‘01)– Reduce health disparities (Politzer et al. ‘01, Regan et al. ‘03, Shin,

Jones and Rosenbaum ‘03, O’Malley et al. ‘05)– Improve infant health (Chabot ‘71, Goldman and Grossman ‘88)

• Limitations– Dated methods: Cross-sectional, interrupted timeseries, or case-control

methods– Substance: Almost all about health service use– Little data: no database on where, when, who for a longer-period

THIS PAPERQuestion: How have CHCs affected health?

• Empirical strategy:– Roll-out of the first CHC programs from 1965 to 1974

“It was an era that attracted some of the brightest social thinkers in the country but not those with the most well-organized minds. It was an era of great administrative confusion.”

~Robert A. Levine (1970)

• Focus on older adult mortality (ages 50 and older)– Accounts for an important and rising share of U.S. health care costs– Mortality is sensitive to the availability of medical care– Comprise >80% of deaths in the U.S from 1960 to 1990, mortality well

measured in every year at the county level from 1959 to 1988

• Long-term health effects of primary care/drugs– Cumulative impact up to 15 years after CHCs began

1. EXAMINE CHCS’ CUMULATIVE (15-YEAR) EFFECTS

2. DIRECT EVIDENCE ON CHCS’ HEALTH EFFECTS– Mortality of those 50 and older is a novel outcome and allows us to compare the effects

of CHCs for those with Medicare (65+) to those without– Mortality is the only county-level health outcome available annually, 1959-1988– Mortality is well measured and allows large sample sizes (~80-88% of deaths for 50+)

THIS PAPER’S CONCLUSIONS• CHCs associated with a 2% reduction in age-adjusted and age-specific

county-level mortality rates for those 50+o Due to reductions in cardiovascular-related causeso No effect on accident related mortality

• This is a plausible but large ITT effect for poor, 50+ patientso ATET implies a 6-8 percent reduction in mortality of the pooro ATET is 18-24 percent of the 1966 rich-poor mortality gap

• Mechanismso SHSUE: 20% increase in “regular source of care” and 40% reduction in the

likelihood of having any prescription drug expenditures among pooro Low cost (hypertension) drugs may account for 40% of the effecto Medicare awareness may contribute but no evidence of increases in per capita

Medicare expenditures in counties after CHCs arrive

BRIEF HISTORY OF THE CHC PROGRAM

JOHNSON’S WAR ON POVERTY

• Johnson declared an “unconditional war on poverty” in his first State of the Union address (Jan. 1964)– Landslide victory in ‘64 + most

liberal Congress since New Deal

1964 ECONOMIC OPPORTUNITY ACT

• Administered by the Office of Economic Opportunity (OEO)

• Sargent Shriver is “poverty czar”

• Almost 1 billion in outlays• Bulk of funds budgeted for VISTA, Head Start, Job Corps• 360 million for Community Action Programs (CAPs)

POOR HEALTH AS ANIMPEDIMENT TO WAR ON POVERTY

GOALS… We saw hundreds of people whose only hope of obtaining medical care was to become an emergency which could not be turned away. We heard countless stories of driving 50 or 100 miles to a city general hospital after refusal of care at a local hospital

…High blood pressure, diabetes, urinary tract infections, anemia, tuberculosis, gall bladder and intestinal disorders, eye and skin diseases were frequent findings among adults

~Dr. Raymond Wheeler, Citizen’s Board of Inquiry into Hunger, 1967-71

CHCS IN 116 COUNTIES, 1965-1974

Columbia Point

Mound Bayou, MS

OEO and DHEW funded CHCs in 40 states.

Two in Washington:1. 1969: King County (Community Health Board of Seattle)2. 1973: Adams County (Columbia Basin Health Association).

MANY ARE LARGE OPERATIONS

• The 1972 Directory of CHCs reports that– CHC in Denver had two centers and 11 satellite

centers, 800+ staff, targeted a population of 287k

– CHC in Oakland, CA provided 24/7 ambulatory care, employed 52 physicians and served 40k

– On average, 32k per county per year

WHAT DID CHCS DO AND WHOM DID THEY SERVE?

COMMON CHC SERVICES

EFFECTS ON “CLINIC” USE?

1963 and 1970 Survey of Health Services Utilization and Expenditures (SHSUE) surveyed same PSUs

Y: Use of a “clinic” as usual source of care

X: binary variable for sex, age groups, race, education and area size

I: income groups (<100%, 100-299, 300-449, 450 of poverty)

D*: 1 if CHC before 1970; Dt: 1 if 1970

Triple difference:Change over time in treated versus untreated PSUs for poorer versus richer respondents.

“CLINIC” AS USUAL SOURCE OF CARE

• Among the poorest households, clinic use increased over 170 % (0.16, s.e. 0.05) in PSUs receiving CHCs

• Higher income households (300-449 of poverty) had little increase (0.2 %) in PSUs receiving CHCs

EXPECTED EFFECTS OF CHCS ON MORTALITY?

MEASURING HEALTH

2400

2600

2800

3000

3200

Dea

ths

per

100,

000

1959 1963 1967 1971 1975 1979 1983 1987

A. Age-Adjusted (50+)

1000

1100

1200

1300

1400

Dea

ths

per

100,

000

1959 1963 1967 1971 1975 1979 1983 1987

B. 50-64

3000

3500

4000

4500

Dea

ths

per

100,

000

1959 1963 1967 1971 1975 1979 1983 1987Year

C. 65-79

1100

012

000

1300

014

000

1500

0D

eath

s pe

r 10

0,00

0

1959 1963 1967 1971 1975 1979 1983 1987Year

D. 80+

ALL-CAUSE MORTALITY, 1959-19881959-88:~900=28% 1959-88:

~436=30%

1959-88:~1300=28% 1959-88:

~3755=25%

WHY DID MORTALITY FALL?

• Largely due to decreases in cardiovascular mortality (diseases of the heart/stroke)– In 1960, CVD accounted for >50% of deaths– By 1988, deaths by diseases of the heart fell by

over 40% and strokes by over 60%

• Often attributed to introductions of anti-hypertensive drugs and information campaigns– Deaths due to cancer increased

CHCS’ EFFECTS ON MORTALITY?

• Reduced cost of primary care and prescription drugs– Increase use of care– Increase compliance with recommended treatments

(Medicare/Medicaid did not typically cover drugs)– Increase awareness about Medicaid and Medicare

• Diversion of care=changes in quality• Externalities

– Information spillovers, contagious diseases– Reductions in ER crowding

• Changes in population “frailty”

EMPIRICAL STRATEGY

COUNTY ROLL-OUT OF COMMUNITY HEALTH CENTERS, 1965-1974

Varies geographically and over time.

Cross-Sectional Differences (T1)Counties with

CHCs Counties without CHCs

T-test of difference (4)-(5) [p-value]

    (N=112) (N=2523)  Mean Population 400,506 44,449 <0.01 Mean Population Over 50 92,956 10,205 <0.01 Percent of population:

in urban area 90.95 55.99 <0.01 in rural area 1.11 11.85 <0.01in Northeast 29.87 23.63 <0.01in Midwest 26.51 32.23 <0.01in South 24.79 34.65 <0.01in West 18.82 9.49 <0.01under 5 years of age 11.18 11.22 0.2365 or older 9.15 10.04 <0.01Nonwhite 14.99 8.60 <0.01with >=12 years of education 43.05 40.69 0.25with <=4 years of education 7.50 8.70 <0.01in households with income <$3,000 16.46 25.53 <0.01in households with income > $10k 17.80 12.17 <0.01

Medical Resources: Total Active MDs (per 1,000 residents) 7.28 3.82 <0.01Any Med Students, 1969 0.71 0.03 <0.01

LOCAL CHC GRANTS• Any organization could apply to the OEO

Applications from “various and sundry groups” ~Theodore Berry, Assistant Director of the OEO

• No funding precedent; no guidelines

It was a wild sort of operation in those early days, making the first grants. We didn’t have any guidelines and didn’t have the time really to draft them to start out…

~Donald Baker, chief counsel for the OEO (Gillette 1996)

EVENT-STUDY FRAMEWORK

: county (j) f.e;

: urban group-by-year f.e.

: year (t) or state-by-year fixed effects

X: annual, county-level covariatesREIS: cash public assistance (AFDC, SSI, GA) and cash retirement and disability payments

OTHER:

1. 1960 county characteristics interacted with linear trends (proportion of population in urban area, nonwhite, with income under $3k, and the proportion of the county’s land that is rural or on a farm, number of active physicians in a county)

OR 2. county-specific linear time trends

EVENT-STUDY FRAMEWORK

Dj1(): Dummies for y years before & after CHC establishment (Tj*)

– Year before establishment is omitted: 1(t-Tj* = -1)

: Pre-effects– change in average difference in treated county outcomes y years before

establishment relative to the untreated counties– visual/statistical evaluation of pre-treatment trends (falsification test)

: Post-effects– Change in the average difference in treated county outcomes y years after

establishment relative to the untreated counties– describe dynamic, likely non-linear evolution of effects, reflecting set up

costs, changes in use, and changes in composition of population

IDENTIFYING ASSUMPTION

• CHC establishment uncorrelated with other determinants of mortality– Threats to internal validity must be correlated with

CHC counties and first-grant timing– Threats to internal validity vs. mechanisms

SENSITIVITY AND NOTES

1. Reweight sample of untreated counties– propensity scores balance observables between treated and

untreated (DiNardo et al. 1996, Heckman et al. 1998)– trim sample of counties <0.10 and >0.90 (Crump et al. 2009)

2. Technical notes– Estimated in levels (not logs)– Weighted by relevant 1960 population– Standard errors clustered by county– Sample limited to locations with at least 100 residents over the

age of 80 in every year

RESULTS

-150

-100

-50

050

-6 -4 -2 0 2 4 6 8 10 12 14

ALL-CAUSE AGE-ADJUSTED MORTALITY, 50+

-150

-100

-50

050

-6 -4 -2 0 2 4 6 8 10 12 14

ALL-CAUSE AGE-ADJUSTED MORTALITY, 50+Point estimates small in magnitude and statistically insignificant Trend break: clear, robust, and

negative beginning in the year the CHC was funded

-150

-100

-50

050

-6 -4 -2 0 2 4 6 8 10 12 14

ALL-CAUSE AGE-ADJUSTED MORTALITY, 50+

Phase-in consistent with setting up new facilities, hiring staff, attracting patients,effects accumulating with exposure

r

Years 0-4:~30-40 = 0.8-1.2%

r

Years 5-9:~60-70 = 1.9-2.2%

HOW LARGE IS THE EFFECT?

• We recover ITT (intention-to-treat effects); not everyone in a county used CHCs

• How large was the mortality reduction among the “treated” (ATET)?– Broad treatment definition: All elderly and poor

are treated: ~25%• Spill-overs in knowledge, eradication of contagious

disease, reductions of crowding in ER

– Narrow treatment definition: Only patients in the last five years:~18%

NARROW “TREATED” DEFINITION

• Users: ~1,358/100k in 1971 SHSUE (National)– Use in treated counties? 25 % of U.S. 50+ population

1357/. 25 = 5428 patients in treated counties– Underreporting? 39 to 50% report accurately

As high as 13,928 in treated counties/100,000– Cumulative use? 75% of individuals who saw a physician

visits in previous 5 years saw one in the last year

18,500 in treated counties/100,000

-Freeman et al. (1982) report that ~15% of 40+ use CHCs

• ATET: 60/0.0185 -323/100k for 50+ patients

MAGNITUDE OF ESTIMATES

• ATET of CHCs on older adult mortality– Relative to mortality of poor: 6-8 %– Relative to nonpoor/poor mortality gap: 18 -24 %– Relative to ‘60-’80 decline in AMR: 35 %– Relative to Chay et al. 2001’s Medicare effect 66%

EFFECT HETEROGENEITY

ALL-CAUSE 50-64 AMRYears 5-9:~15 = 1.5%

Not just a Medicare effect!

ALL-CAUSE 65-79 AMRYears 5-9:~100 = 2.3%

ALL-CAUSE 80+ AMRYears 5-9:~230 = 1.7%

1. By Cause: driven by cardiovascular causes.• Anti-hypertension meds?

2. Falsification: No effects on accidents.• Reassuring, since CHC provide primary care

3. Heterogeneity: largest effects in OEO centers, urban centers, counties with high pre-treatment mortality, counties with many doctors.• These groupings are correlated.

ADDITIONAL RESULTS SUMMARY

THREATS TO INTERNAL VALIDITY

THREATS TO INTERNAL VALIDITY?

• CHC establishment uncorrelated with changes in other determinants of mortality– Threats to internal validity must be concentrated in

CHC counties and correspond in timing to establishment dates

1. Other War on Poverty (WoP) grants packaged with CHCs grants

POTENTIAL CONFOUNDERS?

-1-.

50

.51

-6 -4 -2 0 2 4 6 8Years Since Treatment

A. Community Health Centers

OTHER WOP GRANTS

Funding probability equals 1 in year of first funding

Continued support. P(Subsequent CHC grant) <1 in a given year because of multi-year grants (few clinics closed)

-1-.

50

.51

-6 -4 -2 0 2 4 6

A. Community Health Centers

-1-.

50

.51

-6 -4 -2 0 2 4 6

B. Other Health

-1-.

50

.51

-6 -4 -2 0 2 4 6

C. CAP Administration

-1-.

50

.51

-6 -4 -2 0 2 4 6Years Since Treatment

D. Elderly Programs

-1-.

50

.51

-6 -4 -2 0 2 4 6Years Since Treatment

E. Head Start

-1-.

50

.51

-6 -4 -2 0 2 4 6Years Since Treatment

F. Legal

OTHER WOP GRANTS

1. Other War on Poverty (WoP) grants packaged with CHCs grants

2. CHC grants allocated based on mortality (cherry-picking).

POTENTIAL CONFOUNDERS?

CORRELATES OF CHC ESTABLISHMENT TIMING

1. Other War on Poverty (WoP) grants packaged with CHCs grants

2. CHC grants allocated based on mortality (cherry-picking).

3. Local initiatives to expand medical resources for the poor in treated counties that correspond in timing to CHC dates

POTENTIAL CONFOUNDERS?

-2.5

-1.2

50

1.25

2.5

-.02

5-.

0125

0.0

125

.025

-15 -12 -9 -6 -3 0 3 6 9 12 15

Hospitals per Thousand (Left) Beds per Thousand (Right)

CHANGES IN HOSPITAL CAPACITY

1. Other War on Poverty (WoP) grants packaged with CHCs grants

2. CHC grants allocated based on mortality (cherry-picking).

3. Local initiatives to expand medical resources for the poor in treated counties that correspond in timing to CHC dates

4. County-level, heterogeneous Medicaid effects.

POTENTIAL CONFOUNDERS?

-100

-50

050

-6 -4 -2 0 2 4 6 8 10 12 14

CHC with MxP CHC with MxMD

CHC with MxMS Baseline CHC Effects (Model 2)

CHC EFFECTS + MEDICAID CONTROLS

MECHANISMS

MECHANISMS

1. Reduced-cost/higher-quality primary care• Medical and nonmedical services provided by CHCs• Free medication: hypertension drugs

2. Medicare awareness• Information about Medicare

ACCESS TO PRIMARY CARE

ACCESS TO PRIMARY CARE

ANTI-HYPERTENSION MEDICATION

• Hypertension Detection and Follow-Up Program (HDFP 1973)– Randomized, community based trial (~10,000 patients)

• Treatment included screening, detailed follow-up, flexible scheduling and anti-hypertension medication

• Control group screened and obtained care from usual source.

• How large would CHCs’ effects be if they only provided these drugs? ~40% of CHC effect

MECHANISMS

1. Reduced-cost/higher-quality primary care• Medical and nonmedical services provided by CHCs• Free medication: hypertension drugs (40%)

2. Medicare awareness• Information about Medicare

FEDERAL, PER-CAPITA MEDICAL TRANSFERS

MECHANISMS

1. Reduced-cost/higher-quality primary care• Medical and nonmedical services provided by CHCs• Free medication: hypertension drugs (40%)

2. Medicare awareness• Information about Medicare• No evidence of greater p.c. Medicare expenditures

• Medicare mechanism isn’t driving results• CHCs reduce Medicare expenditures by referring more

patients

DID THE EXPERIMENT IN PUBLIC MEDICINE SUCCEED?

• New evidence that CHCs reduced mortality among adults 50 and older in the shorter and longer term– CHCs associated with ~2% reduction in AMR of 50+ in county– CHCs imply a ~6-8% reduction in AMR of 50+ patients below the poverty

line– Effect of CHCs equals 35% of 1966 poor/non-poor mortality gap

• Effects are large and present for those with Medicare

• Ongoing work– Study of effects for different groups

AcknowledgementsThis project was generously supported by:1. The West Coast Poverty Center (US DHHS, Office of the

Assistant Secretary for Planning and Evaluation, 3 U01 PE000004-03S3)

2. National Institute for Child Health and Human Development (R03-HD066145)

3. The University of California-Berkeley’s Center on the Economics and Demography of Aging (NIA 2P30AG012839)

4. The University of Michigan’s Robert Wood Johnson Foundation Health and Society Scholars Small Grant Program

5. The University of Michigan’s Population Studies Center’s Eva Mueller Award.

Extra slides

OPPOSITION TO THIS EXPERIMENT IN PUBLIC

MEDICINE• “a step toward socialism”• physicians feared competition from CHCs and

expressed concerns about “quality”• 1966 AMA president argued that the

government’s role in health care should be limited to “the overall stimulation and support of private enterprise, rather than undertaking specific operational and directional capacities ”