Background: The present study explored a range of variables to identify predictors of mortality and morbidity and to develop prediction models based on these variables.
Methods: Tools for predicting mortality, hospital length of stay and a patient's destination post-hospital discharge were developed using logistic regression in one dataset (design) and evaluated for prediction performance in a separate dataset (validation). The performance of the mortality model was compared to the trauma and injury severity score (TRISS) and a severity characterization of trauma (ASCOT).
Results: The profile of variables contributing to the final prediction models developed from the design dataset varied across the different outcomes of interest although age, injury severity score, development of complications and triage category were common predictors of all three outcomes. The performance of the new mortality prediction model was superior to both TRISS and ASCOT in the validation dataset. Overall, the new models did not meet the prespecified performance criteria.
Conclusions: The present study identified key predictors of mortality and morbidity (length of hospital stay and discharge destination). The newly developed mortality model out-performed published trauma scoring methods. However, further development and trial of the new prediction models is required before implementation as definitive audit and benchmarking tools could be recommended.