Accurately predicting the risk of early mortality after trauma can guide appropriate use of resources. This study aims to create a pragmatic mortality prediction from prehospital data. The Linking Investigators in Trauma and Emergency Service Task Order One (LITES TO1) database was used to identify predictors of mortality at hour 3, hour 24, and day 30 after trauma. Individual characteristics were assessed using a bivariate logistic regression model. The independent effect of characteristics significantly associated with mortality in a bivariate setting were assessed using a machine learning recursive partitioning model. Initial Glasgow Coma Scale motor score (GCSm) and worst GCS were the strongest predictors of mortality at all time points. Both were predictive of all three most common causes of death: traumatic brain injury/herniation, prehospital/traumatic arrest, and uncontrolled hemorrhage. This is the first predictive machine-learned model tot demonstrate that initial prehospital GSCm strongly predicts mortality after trauma. Using this measure as indication for transport to trauma-designated hospitals could improve resource allocation.
Brito et al. (Thu,) studied this question.