A Bayesian Approach to Variation Assessment of Clinical Outcomes among Hospitals
Keywords: Bayesian hierarchical model, clinical outcome, variation assessment
Information on the variation of clinical outcomes is useful both for hospitals to direct quality improvement and for patients to seek the best healthcare provider. Statistical analysis in variation of clinical outcomes among hospitals is challenging due to possible biases in outcome assessment, clinical practices, and sampling variability. An Australian group developed a frequentist shrunken model to examine variation of the clinical outcome severe intraventricular hemorrhage (IVH) in pre-term infants. Alternatively, we used a Bayesian approach to assess the same data. We proposed a hierarchical Bayesian model and assessed the variation of severe IVH among 24 hospitals. In the model, we let the number of severe IVH cases in each hospital follow a binomial distribution. We set the prior of the rate of severe IVH of a hospital to follow a Beta distribution and the hyperpriors to follow a Gamma distribution. The Markov chain Monte Carlo (MCMC) method was used to estimate the posteriors. Sensitivity tests demonstrated that the Bayesian model was robust with good fitness to the data. Compared with the shrunken model, the Bayesian model yielded a smaller excess number O-E, the difference between the observed value and the expected one, for each hospital. The distribution of the excess number O-E was less jagged. Our model found two more hospitals having event rates significantly varying from the estimated overall rate, in addition to the two shown in the shrunken model. These results suggest that the hierarchical Bayesian model is an appropriate alternative approach to assess variation of clinical outcomes.