Algorithmic Selection of Patients for Case Management: Alternative Proxies to Healthcare Costs

Abstract

Expected healthcare costs are commonly used in the United States as a proxy for health to select patients for case management. However, several recent studies have shown that AI algorithms for predicting costs exacerbate underlying disparities in the healthcare system and result in substantial bias against blacks, who have to be much sicker than whites to be chosen by these algorithms. We look at an alternative proxy based on emergency-room and inpatient utilization, and show that it results in more fair outcomes, reducing racial disparity while choosing patients truly in need for such services. We evaluate the effectiveness of this approach using the publicly available and nationally representative Medical Expenditure Panel Survey data collected annually by the U.S. Department of Health and Human Services.