Computational Health Economics for Health Care Spending

• Activity Number: 3 - Statistical Methods for Health Economics and Applied Econometrics in Health Policy
• Type: Invited
• Date/Time: Sunday, July 30, 2017 : 2:00 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Sherri Rose* and Savannah Bergquist and Tim Layton
• Companies: Harvard Medical School and Harvard University and Harvard Medical School
• Keywords: machine learning ; health economics ; ensembles ; health policy

Abstract

We take the hypothetical role of a profit-maximizing insurer attempting to design its health plans to attract profitable enrollees and deter unprofitable ones. Such an insurer would not be acting in the interests of providing socially efficient levels of care by offering plans that maximize the overall benefit to society, but rather intentionally distorting plan benefits in order to avoid high-cost enrollees to the detriment of both health and efficiency. We focus on a specific component of health plan design: the prescription drug formulary, one of the most important dimensions on which insurers can distort their plan benefits in response to selection incentives as other dimensions are now highly regulated (e.g., pre-existing conditions). Our computational health economics approach centers around developing an ensembled machine learning method to determine which drug classes are most predictive of a new measure of unprofitability we derive, and thus most vulnerable to distortions by insurers in the Health Insurance Marketplaces. This work is designed to highlight vulnerable unprofitable groups that may need special protection from regulators in health insurance market design.