The Impact of Misspecified Prior Distributions in Hospital Profiling Under an Empirical Bayes Framework

• Activity Number: 328
• Type: Topic Contributed
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Sheng Wang*+ and Alex Bohl and Hali Hambridge and Frank Yoon and David Jones
• Companies: Mathematica Policy Research and Mathematica Policy Research and Mathematica Policy Research and Mathematica Policy Research and Mathematica Policy Research
• Keywords: Empirical Bayes ; Hospital quality ; Misspecified model ; Simulation

Abstract

Empirical Bayes methods have gained popularity in hospital profiling. The methods offer gains in precision over traditional frequentist methods for estimating risk-adjusted rates, especially for hospitals with small sample sizes. However, an incorrectly specified parametric model for the prior distribution could jeopardize these gains. An outstanding statistical and policy debate is whether underlying hospital quality depends on hospital characteristics, such as teaching status and volume, and how this relationship should be specified in the prior distribution. We evaluate the performance of hospital profiling models based on misspecified prior distributions on simulated discharge populations. In particular, we simulate discharge populations with patient outcomes and known relationships with hospital characteristics and estimate hospital risk-adjusted rates of patient safety events. We assess and compare the performance of true and misspecified models under various simulation settings and also provide contextual examples of how misspecification may affect important policy programs, such as pay-for-performance.