Thomas Schober (Johannes Kepler University Linz)

There is an increasing interest in measuring and comparing the quality of care in hospitals. Widely used risk adjustment methods rely on observable characteristics to account for patient selection, but are often criticized for their inability to fully control for differences in patients across hospitals. We assess hospital performance using exogenous variation shaped by the institutional setting of inpatient care in Upper Austria. Hospitals have agreed on a rotating schedule, where on each day, one or two hospitals are primarily responsible for the admission of inpatients. For patients in need of acute care, this schedule creates a quasi-random allocation into different hospitals. We use this variation in an instrumental variable (IV) framework to estimate hospital performance, and compare the results to traditional risk adjustment methods. We use patient mortality and readmissions as quality indicators and draw on administrative data from Upper Austria with hospital visits from the years 2005 to 2018. We find substantial differences between IV and risk adjustment estimates, and show that increasing the number of variables used to control for patient differences often does not provide more credible results.