Seminar: Dynamic Medical Decision-Making to Define Monitoring Policies for Cardiovascular Disease Prevention.

Abstract

Physicians seek to prevent chronic diseases by tracking and nudging patients’ healthcare behavior despite the limited time, resources, and diverse needs of a heterogeneous panel of patients. Therefore, there is a need to schedule patients optimally. If programmed incorrectly, patients may forgo needed treatment and suffer disease-related adverse events. Moreover, patients may not adhere to medication and follow-up recommendations despite the physician’s efforts. We develop a guided set of solutions to improve patients’ health. First, we propose a finite horizon and finite-state Markov decision process to define monitoring policies. Second, we develop a discrete Monte-Carlo simulation model built using electronic health records to test policies. Third, we build a dynamic logistic regression model that identifies low-adherent patients. Fourth, we will discuss a preliminary version of a multi-armed bandit model to prioritize high-risk patients within a limited budget scenario. We test and validate using the Veterans Affairs health system longitudinal data for cardiovascular diseases. Finally, we include race and gender effects, given the one-size-fits-all nature of the current national cholesterol guidelines.