The Impact of Unmeasured Clinical Variables on the Accuracy of Hospital Report Cards: A Monte Carlo Study

Abstract

Hospital report cards are commonly produced using administrative data. The objective of this study was to determine the impact of unmeasured clinical data on the accuracy of hospitals' report cards. Monte Carlo simulations were based on both administrative and detailed clinical data for patients hospitalized with an acute myocardial infarction in Ontario, Canada. Data were simulated such that the true performance of each hospital was known. Both clinical and administrative risk scores were randomly generated for each patient. The ability of hospital report cards to correctly identify hospitals that truly had higher than acceptable mortality was compared when both clinical and administrative data were used and when only administrative data were used. By using Monte Carlo simulations, we were able to incrementally increase the divergence between the 2 risk scores. In a wide range of settings, sensitivity and specificity of hospital report cards was only negligibly greater when both administrative and clinical data were used compared to when only administrative data were used. Unmeasured clinical data have at most a minor impact on the accuracy of cardiac hospital report cards.