Abstract:
|
Hospital benchmarking typically uses regression models to adjust for patient case-mix. However, hospitals with non-overlapping patient populations may not be compared fairly. An alternative method uses a template of patients and matches patients from each hospital to the template to compare outcomes directly, thus minimizing non-overlap on observed characteristics. However, template patients may not be representative of each hospital’s population. We test a hospital-specific template in the VA to answer the question: how well does each hospital treat its specific patients relative to peer hospitals that see similar patients?
We used one year of nationwide VA data (556,341 patients at 122 hospitals). For each hospital, we selected a representative template of 300 patients from that hospital's patient population. We matched patients at other hospitals to the index hospital's template on demographics, diagnoses, and acute physiology. The index hospital was compared to hospitals matching all 300 patients. Initial results indicate that there is wide variation in hospital populations and that a hospital-specific template may be a more credible method of hospital benchmarking in VA.
|