A Causal Tree Approach for Personalized Health Care Outcome Analysis

Abstract

Using patient-level data from 35 hospitals for 6 cardiovascular surgeries in New York, we provide empirical evidence that outcome differences between health care providers are heterogeneous across different groups of patients. We then use a causal tree approach to identify patient groups that exhibit significant differences in outcome. By quantifying these differences, we demonstrate that a large majority of patients can achieve better expected outcomes by selecting providers based on patient-centric outcome information. We also show how patient-centric outcome information can help providers to improve their processes and payers to design effective pay-for-performance programs.