Encounter Decisions for Patients With Diverse Sociodemographic Characteristics: Predictive Analytics of EMR Data From a Large Chain of Clinics

Abstract

Managing chronic diabetes care is a major challenge faced by healthcare organizations because it requires resource commitment over a long duration, high levels of patient engagement in the care process, and the socioeconomic and racial diversity of the patient population significantly affect care outcomes. Therefore, it is important to personalize chronic care treatment to improve chronic care outcomes. We propose a decision framework for the predictive management of diabetes that can help reduce the population-level risk of diabetes. We use machine learning on clinical measures, demographics, and socioeconomic status of a large patient population from a chain of clinics in the Midwestern United States to predict the future health conditions of individual diabetes patients. Furthermore, we use the predictive analytic model outcome to build a decision analytic framework to optimally allocate encounters to individual patients. Also, we propose a heuristic solution to the optimal resource allocation model for implementation purposes. We make theoretical and methodological contributions by identifying and combining clinical, demographic, and socioeconomic factors to predict future diabetes risk for patients and demonstrate the use of the predicted risks for optimal resource utilization. Another significant contribution is demonstrating that a data-driven predictive encounter allocation, considering the socioeconomic and demographic factors influencing health risks across patient populations, can promote more equitable healthcare delivery. Finally, we discuss implementation issues and actions.