Abstract 153: Estimating Medication Adherence Effects Using Part D Copayment as an Instrumental Variable in Heart Failure Patients

Abstract

Purpose: Better adherence to medications for heart failure patients may be associated with lower rates of readmission following hospital discharge. Treatment selection could bias estimated effects of medication adherence, but using variation in the Part D copayment within and between patients over time as an instrumental variable (IV) may help identify medication adherence effects. Methods: We use data from the Atherosclerosis Risk in Communities (ARIC) Study to construct person-month records for up to 12 months following discharge for 952 hospitalizations for 541 ARIC cohort participants with heart failure between 2006-2012 who were enrolled in Part D. We measure medication adherence from Part D claims as “proportion ambulatory days covered” (PADC) each month. Adherence is defined as PADC≥80%. We use bivariate probit to jointly estimate models for medication adherence and outcomes (monthly rate of hospital readmission or death) for months 2-12 following discharge. The adherence model uses dummy variables to control for Part D copayment phase ($250 deductible, 25%, 100% donut hole, or 5%) interacted with Medicaid copayment coverage as IV. Outcomes are modeled as a function of adherence in the prior month. Both equations also control for patient demographics and case mix including comorbidities and clinic/lab values. Separate models are run for three medications (ACEI/ARB, beta blockers or diuretics). We also estimate outcome logit equations without IV. Results: Average adherence per month was 49%, 51% and 52% for ACEI/ARB, beta blockers and diuretics, respectively. Readmission and death rates per month were 11% and 1%, respectively; 60% of patients who were not readmitted within the first 30 days were readmitted at some point during the year following discharge. Approximately 23% of follow-up months were for patients in the “donut hole” with 100% copayment. Part D copayment for persons without Medicaid drug coverage was a strong instrument in predicting adherence. In single equation models, being adherent to medication in one month was not associated with a reduced likelihood of readmission or death next month. However, the association was substantively and statistically associated in bivariate probit models using Part D copay as an IV. For example, without an IV, ACEI/ARB adherence did not predict readmission or death in the following month (OR 0.94; 95% CI 0.73-1.22); among patients whose adherence depends partly on cost-sharing, the IV estimation showed a lower likelihood of readmission or death (OR 0.54; 95% CI 0.31-0.92). Conclusions: Medication adherence after hospitalization is suboptimal, and readmission rates during the year following discharge are high. Associations between medication adherence and readmission/death corrected for selection suggest that patients with heart failure may benefit from continued interventions to improve medication adherence.