Exposures

The primary exposure was hospital-specific sepsis mortality. Hospital-specific sepsis mortality was reported as an O:E mortality ratio based strictly on in-hospital mortality for the final hospital in which a patient was treated, calculated for the study period 2013–2014. A multivariable logistic regression model was constructed with an outcome of in-hospital mortality and a priori-defined patient-level predictors, including age, race, sex, comorbidities, infection source, organ dysfunction, skilled nursing facility residence prior to admission, community factors (percent Black, percent Hispanic, percent with high school degree or higher, percent below poverty line), ICU services in hospital, and teaching hospital. All variables were selected based on theory and the proposed relationship with exposures and outcomes, and the same adjustment variables were used for all models [5, 17, 26]. All continuous variables were modeled in categories, and interactions were tested. The predicted probability of mortality was generated for each case in the data set, then the sum of observed in-hospital mortality and predicted in-hospital mortality for each facility was calculated. The ratio of the observed mortality to the predicted mortality was the O:E ratio and was the primary predictor in this analysis.