Evaluating hospital performance.

Patient selection remains a major challenge in evaluating hospital performance. We exploit the quasi-random assignment of patients to hospitals, based on a rotation schedule between hospitals in the Upper Austrian capital of Linz. In an instrumental variable (IV) framework, we use high-quality administrative data and estimate hospital performance with respect to in-hospital mortality, 30-day mortality, and 30-day readmission. We contrast these results with those of traditional risk adjustment models based on patient observables. We find that the assessment of hospital performance is sensitive to the inclusion of patient observables and that increasing the number of socio-economic covariates to better control for patient risk profiles does not always help bring risk-adjusted estimates closer to IV estimates. The divergence between methods is most pronounced for readmissions, where risk-adjustment models imply large and statistically significant differences between hospitals, whereas IV estimates are substantially smaller and not statistically significant. Our results suggest that common risk adjustment does not adequately control for patient differences between hospitals and that hospital quality indicators based on common administrative data should be interpreted with caution. The trend toward personalized medicine may support the process of collecting more clinical information at the individual level, thus allowing for better quality comparisons between hospitals.