Development and validation of a pediatric model predicting trauma-related mortality

Abstract Objectives To develop a prediction model of mortality in pediatric trauma-based injuries. Our secondary objective was to transform this model into a translational tool for clinical use. Study design A retrospective cohort study of children ≤ 18 years was derived from the National Trauma Data Bank between the years of 2007 to 2015. The goal was to identify clinical or physiologic variables that would serve as predictors for pediatric death. Data was split into a development cohort (80%) to build the model and then tested in an internal validation cohort (20%) and a temporal cohort. The area under the receiver operating characteristic curve (AUC) was assessed for the new model. Results In 693,192 children, the mortality rate was 1.4% (n = 9,785). Most subjects were male (67%), White (65%), and incurred an unintentional injury (92%). The proposed model had an AUC of 96.4% (95% CI: 95.9%-96.9%). In contrast, the Injury Severity Score yielded an AUC of 92.9% (95% CI: 92.2%-93.6%), while the Revised Trauma Score resulted in an AUC of 95.0% (95% CI: 94.4%-95.6%). Conclusion The TRAGIC + Model (Temperature, Race, Age, GCS, Injury Type, Cardiac-systolic blood pressure + Mechanism of Injury and Sex) is a new pediatric mortality prediction model that leverages variables easily obtained upon trauma admission.