Inference for Identifying Outlying Health Care Providers

• Activity Number: 504
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Michael Racz*+
• Companies: Albany College of Pharmacy and Health Sciences
• Keywords: Coronary artery bypass graft surgery ; credible regions ; posterior predictive p-value

Abstract

Provider profiling is the evaluation of the performance of hospitals, doctors, and other medical practitioners to enhance the quality of care. Many jurisdictions have released public report cards comparing hospital or physician-specific outcomes. Ohlssen, Sharples and Spiegelhalter present an extensive, thoughtful evaluation of provider profiling methodology. In particular they are concerned about whether a putative outlier is really an outlier or an observation in the tail of the common distribution for all practitioners, and present methodology to address this issue. Using both New York State bypass surgery data and simulated data of the same type we evaluate the Ohlssen et al. approach, comparing it with results from applying standard methodology. We also present and evaluate a third method. Here, a 100(1 - ?)% credible region is developed for the provider fixed effects and used to identify outliers. We also consider inference based on random effects models, commonly employed in provider profiling. Assuming a random effects model we present two methods to obtain 100(1 - ?)% credible regions which can be used to identify outlying units.