Although there has been extensive study of air medicine's role in traumatic brain injury (TBI), the outcome remains unclear. Relationships between data set variables can be identified using learning algorithms, such as decision trees, support vector machines (SVMs) and artificial neural network (ANNs), but are not empirically useful for hypothesis testing. The authors' county trauma registry identified patients with Head Abbreviated Injury Score 3+. Using ANN, SVM, and decision trees, predictive models were generated. Differential survival values for each patient (actual and predicted outcome) were calculated using the three best-performing ANN models. To identify those who benefit from air transport, predicted survival values were calculated for each patient with transport mode artificially input as "ground" or "air" as well. X sq was used for ratio of unexpected survivors to unexpected deaths comparison for air- and ground-transported patients, for SVM comparison. Exploration of various transport mode indications in optimized survival algorithms was performed using decision tree analysis. The study included a total of 11,961 patients. An air transport survival benefit across patients was predicted by all three learning algorithms, especially for patients with hypotension, lower Glasgow Coma Scale scores, or higher Injury Severity Score or Head Abbreviated Injury Score. In TBI, a survival advantage seems to be conferred by air medical response, especially in more critically injured patients.
Abstract