Understanding Medication Adherence Using Stated-Preference Data

OBJECTIVE: More than half of people who have experienced a myocardial infarction (MI) are not adherent to their medication regimen, which leads to poorer health outcomes and greater health care utilization. We used a stated-preference (SP) study to examine factors that could influence patient compliance to prophylactic cardiovascular treatments, and discuss practical issues in using SP methods to explain medication adherence as well as how economic theory helps us interpret the SP results.

METHODS: Preference data for treatments that lower the risk of cardiovascular events were collected from respondents in the United States with self-reported history of MI using a discrete-choice experiment (DCE). To control for compliance bias and endogeneity in the stated-adherence data, we designed two question formats derived from a household-production model of adherence behavior. All respondents answered 11 judgment questions that presented a pair of virtual patients who were prescribed different treatments defined by: reduction in the risks of nonfatal MI and fatal MI, treatment-related risk of serious infection, mode and frequency of administration, and monthly medication cost. One format asked respondents to select the treatment to which they would most likely be nonadherent. The second format asked respondents to state which of two virtual patients was better off after learning how adherent each was to each medication. Limited dependent-variable models were used to estimate weights indicating the impact of treatment and respondent characteristics on stated-adherence and quantifying the stated impact of nonadherence on respondents’ well-being.

RESULTS: 464 respondents completed the survey. The analysis of the first question format indicated that reduction in the risk of a nonfatal MI had the largest effect on stated adherence, followed by medication cost, the risk of serious infection , and mode and frequency of administration. Reductions in the risk of a fatal MI did not have a statistically significant impact on stated adherence. Utility-theoretic joint analysis of the two question formats found that adherence was always preferred to nonadherance, and that the reductions in compliance considered in the study had a significant impact on the perceived overall benefits of prophylactic treatments.      

CONCLUSIONS: Historically, the use of SP methods to understand patient compliance has been limited by lack of an analytical framework based on a utility-theoretic behavioral model, possible compliance bias, and the potential for endogeneity. Our study explores whether two SP question formats linked to a common economic framework can overcome some of these limitations to help explain treatment adherence. We find that both clinical and nonclinical factors can impact treatment adherence, suggesting that the flexibility to include a variety of factors with SP models can be useful in understanding patient compliance.