Association for Public Policy Analysis & Management

Research Objective: Our objective is to estimate the effect of health insurance coverage on medication adherence. Poor adherence is associated with potentially avoidable adverse health outcomes and significant spending (Cutler and Everett 2010). Most prior studies on adherence use data from a single source of coverage (such as Medicare Part D). Some potentially important drivers of adherence – such as coverage itself – are difficult to assess empirically in these data. The 2014 Affordable Care Act’s (ACA) coverage expansion provisions resulted in nearly 18 million newly-insured individuals. We used variation in coverage due to the ACA and routine churn in coverage over time to assess the relationship between coverage and adherence.

Methods: We leveraged within-patient variation in coverage over time in panel data. Our main results are from regression models with patient-level fixed effects to control for unobserved, time-invariant patient-level characteristics. We also estimated models with state-level fixed effects to explore relationships between adherence and time-invariant demographic characteristics. Finally, we compared adherence in a 2012 “pre-ACA” cohort and a 2015 “post-ACA” cohort cross-sectionally controlling for changes in patient-level observables. For all analyses, we calculated a proportion of days covered (PDC) measure for each 30-day period in our data using a “daily diary” approach and defined periods with PDC greater than 80% as adherent.

Data: We used multi-payer prescription transaction data from IQVIA which covers nearly 90% of US retail pharmacy transactions and includes identifiers to track fills for individual patients over time. We used all 2012-2016 transactions for separate samples of five million non-elderly patients with 2012 and 2015 prescription fills, respectively, to treat chronic conditions (ACE/ARBs, SSRI/SNRIs, statins, antihyperglycemics excluding insulin, and drugs to treat asthma/COPD). We estimated the fixed effects models using data from the 2012 cohort. The pre/post-ACA cross-sectional comparison used data from both cohorts.

Results: In models with patient fixed-effects, the share of adherence periods was between 7.8% and 28.6% higher across drug categories for patients with Medicaid coverage compared with patients without coverage. Other sources of coverage had a similar relationship with measured adherence. In models with state fixed-effects, we found higher adherence rates for older patients, male patients (in four of five drug categories), patients with more total prescription fills, and patients with health insurance coverage regardless of source. Periods for patients in the 2015 (post-ACA) cohort compared to the 2012 (pre-ACA) cohort were between 15 and 20 percent more likely to be adherent, although it is not possible to determine whether these differences were driven by the coverage expansion or other factors.

Conclusions: Our results confirm the intuition that health coverage is an important driver of medication adherence. Our results suggest that adherence increased after the ACA coverage expansion. We also show that patient characteristics – like age, gender, and measures of prescription volume – as strongly associated with measured adherence. Our results highlight an important benefit of insurance coverage that should be considered in future health reform deliberations. They can also help inform the design and targeting of policies to improve adherence.