Performance of composite versus all-or-none compliance statistics for medical evidence-based core measure sets

Core measure sets, composite bundles, and all-or-none (AON) statistics are becoming increasingly common for measuring patient care process quality and compliance to evidence-based medicine. Although dozens of such measures have been proposed for evaluating, incentivizing, and compensating hospital performance, their relative statistical properties remain largely uninvestigated. We compute and compare the sensitivity, specificity, and operating characteristics of each method for detecting within-hospital changes and between-hospital differences, with composite measures usually performing better than their AON counterparts - contrary to current trends and suggestions in the literature, with measurement implications on how to best accelerate improvement and reward performance differences.