Impact of case volume on hospital performance assessment.

BACKGROUND: Process performance measures are increasingly used to assess and reward hospital quality. The impact of small hospital case volumes on such measures is not clear. METHODS: Using data from the Hospital Quality Alliance, we examined hospital performance for 8 publicly reported process measures for acute myocardial infarction (AMI) from 3761 US hospitals during the reporting period of January to December 2005. For each performance measure, we examined the association between hospital case volume, process performance, and designation as a "top hospital" (performance at or above the 90% percentile score). RESULTS: Sample sizes available for process performance assessment varied considerably, ranging from a median of 3 patients per hospital for timely administration of thrombolytics therapy to 62 patients for aspirin given on arrival at the hospital. In aggregate, hospitals with larger AMI case volumes had better process performance; for example, use of beta-blockers at arrival rose from 72% of patients at hospitals with less than 10 AMI cases to 80% of patients at hospitals with more than 100 cases (P < .001 for volume trend). In contrast, owing to an artifact of wide sampling variation in sites with small denominators, classification of a center as a top hospital actually declined rapidly with increasing case volume using current analytic methods (P < .001). This unexpected association persisted after excluding very low volume centers (<25 cases) and when using Achievable Benchmarks of Care. Using hierarchical models removed the paradoxical association but may have introduced a bias in the opposite direction. CONCLUSIONS: Large-volume hospitals had better aggregate performance but were less likely to be identified as top hospitals. Methods that account for small and unequal denominators are needed when assessing hospital process measure performance.

Duke Scholars

Author Sean Michael O'Brien Biostatistics & Bioinformatics, Division of Biostatistics

Author Elizabeth Ray DeLong Biostatistics & Bioinformatics, Division of Biostatistics