A Statistical Framework for Severity Adjustment of Hospital Mortality Rates

This Note presents a statistical framework and associated data analyses that should inform the interpretation of hospital death rates for Medicare patients. It develops statistical modeling and analysis designed to define how much variation in hospital mortality rates for Medicare patients is attributable to each of these causes — small sample fluctuations, mix in severity of patient condition, and hospital quality of care. The authors compare alternative estimators of hospital-specific mortality rates and their standard errors with the currently used ones in a simulation study; they put observed mortality rates into a statistical framework that allows them to estimate how much hospitals vary in their underlying mortality rates both with and without severity adjustment. They present ways to measure how accurately this underlying variation can be estimated. It is clear from the national data that hospitals differ in their underlying death rates after accounting for sampling variation but not for patient severity, and this study's variance component models confirm this finding across four medical conditions (stroke, pneumonia, myocardial infarction, and congestive heart failure).