Online Program

Inferring Medication Adherence Using Health Outcomes with Bayesian State-Space Models (306528)

Keywords: Bayesian analysis, state-space models, sequential Monte Carlo, medication adherence

Patients' non-adherence to their prescribed medication is a serious obstacle to successful medication therapy and a widespread problem in clinical care. Current methods to summarize medication adherence are generally not practical or accurate enough to be useful in clinical settings. We develop an approach to infer medication adherence rates from commonly-collected clinical data, including: (1) health outcomes measured over time that are likely to be directly impacted by differential adherence, and (2) baseline health characteristics and sociodemographic data. Our approach uses efficient Bayesian computational methods for the goal of inferring recent adherence behavior, and uses information not typically utilized in adherence models. First, we fit a Bayesian State-Space Model (SSM) to health outcomes as a function of time-varying adherence. Second, we infer a particular patient's medication adherence given their observed health outcomes and baseline health and sociodemographic information using a Sequential Monte Carlo (SMC) algorithm, which accomplishes efficient sampling in high dimensional spaces.