Value of Medication Adherence in Chronic Vascular Disease: Fixed Effects Models of Health Services Utilization and Cost

©2008 Caremark. All rights reserved. This presentation contains confidential and proprietary information of Caremark and cannot be reproduced, distributed or printed without written permission from Caremark.©2008 Caremark. All rights reserved. This presentation contains confidential and proprietary information of Caremark and cannot be reproduced, distributed or printed without written permission from Caremark.

Value of Medication Adherence in Chronic Vascular Disease: Fixed Effects Models of Health Services Utilization and Cost

M. Christopher Roebuck Director, Health Economics

Co-Authors: Liberman, J.; Gemmill-Toyama, M.; Brennan, T.

American Society of Health Economists3rd Biennial ConferenceCornell University, Ithaca, NYMonday, June 21, 20108:30-10:00am (Hollister 306)

©2008 Caremark. All rights reserved. Caremark proprietary and confidential information. Not for distribution.2

Background• Almost half of all Americans (approximately 133 million) suffer from at least

one chronic condition1

• Although medication adherence enhances health and reduces adverse health events, average compliance rates are just 50%2

• Because medication adherence increases pharmacy expenditures, payers and policymakers are interested in knowing whether lower medical costs from adherence offset these higher pharmacy costs

• If so, policies and programs to encourage medication adherence (e.g., value-based insurance design) may be well-worth their investment

• Despite the critical importance of estimating the value of medication adherence, the existing literature is surprisingly scant and methodologically challenged

1Centers for Disease Control and Prevention (CDC). Chronic disease overview [CDC website]. November 20, 2008. Available at: http://www.cdc.gov/NCCdphp/overview.htm. Accessed September 29, 2009.

2World Health Organization (WHO). Adherence to long-term therapies: evidence for action [World Health Organization website]. 2003. Available at: http://www.who.int/chp/knowledge/publications/adherence_report/en/index.html. Accessed September 29, 2009.


Objectives

• To estimate the impact of medication adherence in four chronic vascular conditions (congestive heart failure (CHF), hypertension, diabetes, and dyslipidemia) on health services utilization and cost

• To examine whether adherence effects are different for seniors or by gender


Literature: Main Findings• Clinical trials routinely document drug cost-effectiveness usually via reduced

hospitalizations and emergency room (ER) visits, however, results may not be applicable to “real world” treatment settings

– Controlled environment likely different than eventual community-based settings– Provide treatment versus non-treatment and dose-response effect estimates– Individuals aren’t randomized to adherent versus non-adherent cohorts– Observational data more readily available and allows for hypothetical treatments

• Observational studies generally find higher adherence associated with– Increased pharmacy costs– Usually increased outpatient visits – Often lower ER use and hospitalization– But its impact on total healthcare costs is not always a net benefit

• Sokol et al. (2005) report non-seniors have total healthcare cost savings from adherence to CHF, diabetes, dyslipidemia, and hypertension drugs

©2008 Caremark. All rights reserved. Caremark proprietary and confidential information. Not for distribution.

Literature: Challenges• Potential endogeneity of adherence

– Findings from observational studies are questionable since unobserved characteristics may be the real cause, thereby leading to biased estimates

– Adherent individuals may engage in other healthy behaviors such as regular exercise that are unmeasured and correlated with health services utilization and cost (i.e., the “healthy user” effect)

• Cross sectional studies do not allow one to determine the direction of causality (i.e., individuals may become adherent as a result of an adverse medical event)

• It is also difficult to determine the timing and duration of adherence effects

5


Data• Integrated pharmacy and medical claims data on 135,008 patients from 9

employers• Annual panel dataset of individuals (age ≥ 18) with continuous eligibility

throughout study period: 7/1/05–6/30/08 (i.e., 3 observations per individual)• With one or more of the following diagnoses (ICD9CM):

– CHF: 428.x – Diabetes: 250.x – Dyslipidemia: 272.0, 272.2

– Hypertension: 401.x - 405.x • Final sample sizes:

– CHF:16,353– Diabetes: 42,080– Dyslipidemia: 53,041– Hypertension: 112,757


Measuring AdherenceGoals:• Create a single adherence measure for each condition based on the

commonly used Medication Possession Ratio (MPR)• Considered adherence distribution and functional form of expected effect

– Using continuous MPR may not be clinically appropriate due to non-linear effects (i.e., is a movement from 0.10 to 0.30 clinically the same as 0.60 to 0.80?)

– Relatively arbitrary threshold of 0.80 generally referred to as “adherent”• Account for primary non-compliance

Steps followed:• For each of three 1-year observations• Calculated MPR by therapeutic class (TC)• Rolled up to condition-level as MPR mean, weighted by TC days’ supply• Created dichotomous measures of “optimally adherent” where MPR≥ 0.80• Time period started as of condition diagnosis date (i.e., MPR=0 for

individuals without medication)


MPR Histograms0

510

1520

De

nsity

0 .2 .4 .6 .8 1HFMPR

05

1015

2025

De

nsity

0 .2 .4 .6 .8 1HPMPR

05

1015

2025

De

nsity

0 .2 .4 .6 .8 1DIMPR

05

1015

2025

De

nsity

0 .2 .4 .6 .8 1CHMPR

Congestive Heart Failure Hypertension

Diabetes Dyslipidemia


Exploring Functional Form ofMPR→Total Healthcare Costs

20

00

030

00

040

00

050

00

060

00

0

0 .2 .4 .6 .8 1

Lower/Upper Limits of CILocal est.

LOCPR of MPR on THCG for HF

12

00

014

00

016

00

018

00

020

00

0

0 .2 .4 .6 .8 1


LOCPR of MPR on THCG for HP

2000

025

000

3000

0

0 .2 .4 .6 .8 1


LOCPR of MPR on THCG for DI

15

00

016

00

017

00

018

00

019

00

020

00

0

0 .2 .4 .6 .8 1


LOCPR of MPR on THCG for CH


VariablesDependent variables• Health services utilization:

– Annual inpatient hospital days– Annual emergency department visits– Annual outpatient physician visits

• Health services costs:– Annual gross pharmacy costs– Annual gross medical costs– Annual gross total healthcare costs

Independent variables– Optimally adherent– Charlson Comorbidity Index– Gender– Senior (≥65)– Vector of annual time dummies


Methods

• We estimated 6 linear fixed effects models

• Included adherent main effect and interaction terms for male*adherent and senior*adherent in the models

• Used Driscoll & Kray heteroskedasticity-robust standard errors


Sample MeansVariable CHF Hypertension Diabetes Dyslipidemia

Male 0.550 0.487 0.532 0.502

Age 77.301 68.401 67.872 65.096

Senior (Age ≥ 65) 0.872 0.614 0.615 0.517

Charlson Comorbidity Index 2.025 1.112 1.696 1.001

Annual Inpatient Hospital Days 11.901 3.291 4.255 2.239

Annual Emergency Dept Visits 0.613 0.318 0.353 0.265

Annual Outpatient Physician Visits 11.651 8.506 9.407 8.660

Annual Pharmacy Costs $3,780 $2,867 $3,624 $2,920

Annual Medical Costs $39,076 $14,813 $17,955 $12,688

Annual Total Healthcare Costs $42,856 $17,680 $21,580 $15,608

Medication Possession Ratio (MPR) 0.400 0.591 0.513 0.522

Optimally Adherent (MPR ≥ 0.80) 0.340 0.505 0.412 0.426


Bivariate Results: Health Services Utilization and Cost by Medication Adherence StatusCondition

Adherence

Status

Inpatient Hospital

Days

Emergency Department

Visits

Outpatient Physician

Visits

Pharmacy Costs

Medical Costs

Total Healthcare

Costs

CHF

Non-adherent 13.220 0.648 11.344 $3,274 $42,549 $45,823

Adherent 8.046 0.569 14.339 $4,649 $33,113 $37,762

Hypertension

Non-adherent 4.245 0.372 8.690 $2,171 $16,835 $19,006

Adherent 1.613 0.256 8.708 $3,251 $11,041 $14,292

Diabetes

Non-adherent 4.695 0.372 9.236 $2,615 $18,501 $21,116

Adherent 2.520 0.313 10.228 $4,586 $14,725 $19,311

Dyslipidemia

Non-adherent 2.055 0.258 8.137 $1,932 $10,880 $12,812

Adherent 1.621 0.248 9.616 $3,850 $12,479 $16,329


Fixed Effects Results: Impatient Hospital Days

Table 4. Estimated Effects of Medication Adherence on Annual Health Services Utilization by Chronic Vascular Condition

Health Services Utilization CategoryAdherence Comparison

Congestive Heart Failure(n = 16,353)

Hypertension(n = 112,757)

Diabetes(n = 42,080)

Dyslipidemia(n = 53,041)

Annual inpatient hospital days

Adherent (vs. non-adherent)-5.715*** -2.135*** -2.394*** -1.177***

Adherent female (vs. non-adherent female)-6.461*** -2.218*** -2.376*** -1.145***

Adherent male (vs. non-adherent male)-5.114*** -2.046*** -2.410*** -1.209***

Adherent senior (vs. non-adherent senior)-5.868*** -3.143*** -3.407*** -1.881***

Adherent non-senior (vs. non-adherent non-senior)-4.737*** -0.572*** -0.834*** -0.442***

Notes: Presented are marginal effect estimates from linear fixed effects models of health services utilization. All models included a weighted Charlson Comorbidity Index; two year indicator variables; dummy variables for senior, male, and adherent; and interaction terms for adherent with male and senior. Statistical significance based on robust Driscoll-Kraay standard errors denoted as follows: *** p<0.01; ** p<0.05; * p<0.10.


Fixed Effects Results: Total Healthcare Costs

Table 5. Estimated Effects of Medication Adherence on Annual Health Services Costs by Chronic Vascular Condition

Health Services Cost CategoryAdherence Comparison

Congestive Heart Failure(n = 16,353)

Hypertension(n = 112,757)

Diabetes(n = 42,080)

Dyslipidemia(n = 53,041)

Annual total healthcare costs

Adherent (vs. non-adherent) -$7,823*** -$3,908*** -$3,756*** -$1,258***

Adherent female (vs. non-adherent female) -$11,506*** -$3,797*** -$3,335*** -$1,213***

Adherent male (vs. non-adherent male) -$4,860*** -$4,026*** -$4,126*** -$1,303***

Adherent senior (vs. non-adherent senior) -$7,893*** -$5,824*** -$5,170*** -$1,847***

Adherent non-senior (vs. non-adherent non-senior) -$7,374*** -$939*** -$1,576*** -$644***

Notes: Presented are marginal effect estimates from linear fixed effects models of health services utilization. All models included a weighted Charlson Comorbidity Index; two year indicator variables; dummy variables for senior, male, and adherent; and interaction terms for adherent with male and senior. Statistical significance based on robust Driscoll-Kraay standard errors denoted as follows: *** p<0.01; ** p<0.05; * p<0.10.


Discussion• Optimal medication adherence in CHF, hypertension, diabetes, and

dyslipidemia was associated with:– Increases in gross pharmacy costs and physician office visits– Decreases in emergency department visits and inpatient hospital days

• Higher pharmacy costs were more than offset by lower medical costs

• Average benefit-cost ratios were:– 8:1 for CHF– 10:1 for hypertension– 7:1 for diabetes– 3:1 for dyslipidemia– Highest was 13:1 for seniors with hypertension– Lowest was 2:1 for non-seniors with dyslipidemia

• Adherence effects are more pronounced for the elderly

• Adherence effects did not significantly differ by gender

16


Limitations

• Endogeneity still possible– Reverse causality– Time-variant unobservables correlated with adherence and utilization/cost

• Difficult to determine the timing and duration of adherence effects

• Non-linear, two-part models perhaps more appropriate– Probit / negative binomial for count measures– Probit / gamma-log link GLM for cost data

• However, linear models have some advantages:– Fixed effects estimation is easier (e.g., fixed effects gamma/log GLM)– More easily explainable to medical journal readers (e.g., Health Affairs)

17


Thank You!

• Comments and suggestions are welcomed

M. Christopher RoebuckCVS CaremarkDirector, Health Economics11311 McCormick Road, Suite 230Hunt Valley, MD [email protected]

Value of Medication Adherence in Chronic Vascular Disease: Fixed Effects Models of Health Services Utilization and Cost

Health & Medicine

higher adherence

value of medication

impact of medication

adherence distribution

adherence goals

duration of adherence

health economics

single adherence measure