Background: A few recent studies have compared the abilities of different injury severity measures to predict inpatient mortality. This study extended previous studies in that it used a registry with noncenters as well as centers, and examined the relative marginal abilities of competing severity measures to predict mortality when physiologic data also are available.
Methods: Several methods for assessing injury severity of trauma patients were compared in terms of their ability to predict mortality with and without the addition of additional demographic and physiologic information using logistic regression models. Separate determinations also were made for all patients and for three groups of patients with blunt trauma resulting from motor vehicle crashes, low falls, and other blunt injuries. Statistical models were compared using measures of discrimination and calibration.
Results: The International Classification of Disease-Based Severity Score (ICISS) had the best discrimination for each of the eight models examined, and it was significantly better than all the other measures in relation to the models for all patients and for victims of motor vehicle crashes. The ICISS also had the best calibration in half of the models with and half without demographic and physiologic information. The New Injury Severity Score had the best calibration in relation to two of the remaining four models. Physiologic data add substantially to the ability to predict mortality regardless of the anatomic injury severity measure used.
Conclusions: On the average, the ICISS had the best discrimination of all of the measures, as well as a slight edge with respect to calibration in predicting trauma mortality with or without the aid of demographic or physiologic measures.