Prospective validation and comparison of clinical prediction models for early trauma care: A multicentre cohort study

Abstract Objective We aimed to prospectively validate and compare published prediction models and clinician-assigned triage categories for early trauma care. Design Prospective multicentre cohort study. Setting Three public hospitals in urban India: one secondary care hospital in Mumbai and one tertiary care teaching hospitals in Delhi and Kolkata each. Participants Adult patients aged over 18 years presenting to the emergency department with a history of trauma between 2016 and 2022. A total of 13,041 patients were included in the final analysis. Methods We externally validated five published trauma prediction models (GAP, Gerdin, KTS, MGAP, and RTS) and clinician-assigned triage categories based on initial assessment. The primary outcome was 30-day all-cause mortality. Models were recalibrated using a separate updating sample prior to evaluation, and model performance was assessed in terms of discrimination (AUC), calibration (calibration slope and plots), and decision curve analysis. Results All models and clinician gestalt-based triage demonstrated excellent discrimination (AUC range: 0.90-0.96) and good calibration after updating. The GAP model achieved the highest AUC (0.96, 95% 0.94-0.97), and RTS demonstrated the highest sensitivity (0.70). Conclusion Simple, physiology-based prediction models and clinician gestalt both demonstrated excellent performance in predicting 30-day mortality among adult trauma patients in Indian emergency departments. These findings provide a practical foundation for further development of trauma triage systems.