A Bayesian approach to risk-adjusted outcome monitoring in healthcare.

Clinical outcomes are commonly monitored in healthcare practices to detect changes in care providers' performance. One key challenge in outcome monitoring is the need of adjustment for patient base-line risks. Various control charting methods have been developed to conduct risk-adjusted outcome monitoring, but they all rely on the availability of a large number of historical data. We propose a Bayesian approach to this type of monitoring for cases where historical data are not available. In our approach, detection of change is formulated as a model-selection problem and solved using a popular Bayesian tool for variable selection, the Bayes factor. Issues in decision-making about whether there is a change point in the observed patient outcomes are addressed, including specification of priors and computation of Bayes factors. This approach is applied to a real data set on cardiac surgeries, and its performance under different parameter scenarios is studied through simulations.