Predicting the Future in Trauma: Trauma and Injury Severity Score Loses Accuracy and Validity for Late Deaths After Injury.

Abstract

The Trauma and Injury Severity Score (TRISS) is widely used to predict mortality in trauma patients, but its performance metrics have not been analyzed for early vs later deaths. Therefore, we aimed to investigate the impact of time to death on the accuracy of TRISS. Patients from 2013 to 2018 American College of Surgeons Trauma Quality Improvement Program database were included. We compared predicted survival by TRISS using the areas under receiver operating characteristic curves (AUCs) and calibration curves between different cut-off times and subgroups. We further compared early (≤72hr) and late (>72hr) deaths based on mechanisms and severity. Among the 1,180,745 patients, the total mortality rate was 6.4%, with 59% early deaths and 41% late deaths. The AUC of TRISS for all patients was .919 (95% CI: .918-.921) for ≤72hr survival and .845 (95% CI: .843-.848) for >72hr survival. Significant discrepancies in AUCs between the early and late death groups existed in all cohorts based on blunt/penetrating mechanisms and severity. TRISS predicted well in early survival of penetrating injury but was less reliable in late survival of penetrating injury and all blunt injury. TRISS tended to underestimate survival, particularly for patients with lower probability of survival, with increased discrepancies seen for predicting late deaths. The predictive ability of TRISS varies significantly based on the timing of deaths and key injury factors. TRISS may be best utilized in predicting early survival in penetrating injury, but its reliability and accuracy diminish when predicting late deaths for all kinds of injury.