Accuracy of different survival prediction models in a trauma population

Background: There is growing demand for a simple accurate scoring model to evaluate the quality of trauma care. This study compared different trauma survival prediction models with regard to their performance in different trauma populations. Methods: The probability of survival for 10 777 trauma patients admitted to hospital was calculated using the formulas of the following models: the Major Trauma Outcome Study (MTOS), the Trauma Audit and Research Network (TARN) and the Base Excess Injury Severity Scale (BISS). Updated coefficients were calculated by logistic regression analysis based on a Dutch data set. Different models were compared for several subsets of patients, according to age and injury type and severity, using the area under the receiver operating characteristic (ROC) curve (AUC). Calibration for the updated models was presented graphically. Results: Most of the models had an AUC exceeding 0.8. For the total population, the TARN Ps07 model with updated coefficients had the highest AUC (0.924); for the subset of patients in whom all parameters were available, the BISS model including the Glasgow Coma Scale had the highest AUC (0.909). All of the models had high discriminative power for patients aged less than 55 years. However, in older or intubated patients and in those with severe head injuries the discriminative power of the models dropped. The TARN model showed the best accuracy. Conclusion: The investigated models predict mortality fairly accurately in a Dutch trauma population. However, the accuracy of the models depends greatly on the patients included. Severe head injuries and greater age are likely to lead to a decrease in the accuracy of survival prediction.