PREDICTIONS OF HOSPITAL MORTALITY-RATES - A COMPARISON OF DATA SOURCES

Background: Comparing hospital mortality rates requires accurate adjus tment for patients' intrinsic differences. Commercial severity systems require either administrative data that omit vital clinical facts abo ut patients' conditions at hospital admission or costly, time-consumin g abstraction of medical records. The validity of supplementing admini strative data with laboratory data has not been assessed. Objective: T o compare risk-adjusted mortality predictions using administrative dat a alone; administrative data plus laboratory values; and the combinati on of administrative, laboratory, and clinical data. Design: Retrospec tive cohort study. Setting: 30 acute care hospitals. Patients: 46 769 patients hospitalized with acute myocardial infarction, cerebrovascula r accident, congestive heart failure, or pneumonia. Measurements: Each patient's probability of dying was estimated by using administrative data only (unrestricted administrative models), administrative data re stricted to secondary diagnoses that are unlikely to be hospital-acqui red complications (restricted administrative models), restricted admin istrative data plus laboratory data (laboratory models), and restricte d administrative data plus laboratory and abstracted clinical data (cl inical models). Results: The unrestricted administrative models predic ted death better than the restricted administrative models (average ar eas under the receiver-operating characteristic [ROC] curves, 0.87 and 0.75, respectively) and as well as the laboratory models and the clin ical models (average areas under the ROC curves, 0.86 and 0.87, respec tively). The good mortality predictions obtained by using the unrestri cted administrative models result from inclusion of hospital-acquired complications that commonly precede death. The laboratory models ranke d 93% of patients and 95% of hospitals in a manner similar to the clin ical models; in comparison, rankings provided by the laboratory models were similar to those provided for 75% of patients and 69% of hospita ls by the unrestricted administrative models and for 72% of patients a nd 77% of hospitals by the restricted administrative models. Conclusio ns: Adding laboratory data (often available electronically) to restric ted administrative data sets can provide accurate predictions of inpat ient death from acute myocardial infarction, cerebrovascular accident, congestive heart failure, or pneumonia. This alternative avoids the c ost of data abstraction and the serious errors associated with using a dministrative data alone.