Hospital profiling using Bayesian decision theory

Abstract

AbstractBackgroundFor evaluating the quality of care provided by hospitals, special interest lies in the identification of performance outliers. We study a setting where the decision to classify hospitals as performance outliers or non-outliers is based on the observed result of a single binary quality indicator.MethodsWe propose to embed the classification of providers into a Bayesian decision theoretical framework which enables the derivation of optimal decision rules with respect to the expected decision consequences. We argue that these consequences depend upon for which pathway to quality improvement the profiling of hospitals takes place. We propose paradigmatic utility functions for the two pathways external reporting and change in care delivery and compare the resulting optimal decision rules with regard to their threshold values, sensitivity and specificity. We further apply them to the area of hip replacement surgeries by analyzing data from the mandatory German hospital profiling program. Based on five quality indicators, we re-evaluate the performance of 1,277 hospitals which treated over 180,000 patients for hip-replacement surgeries during 2017.ResultsBased on the utilities we assigned to the classification decisions, the decision rule for change in care delivery classifies more high-volume providers as outliers compared to the decision rule for external reporting, with consequences for both sensitivity and specificity. The re-evaluation of the five quality indicators illustrates that classification decisions are highly dependent upon the underlying utilities.ConclusionAnalyzing the classification of hospitals as a decision theoretic problem and considering pathway-specific consequences of decisions can help to derive an appropriate decision rule. Contrasting decision rules with regard to their underlying assumptions about the utilities of classification consequences can be helpful to make implicit assumptions transparent and justifiable.