Assessing the Predictive Accuracy of Trauma Scoring Systems for In-Hospital Mortality in Resource-Limited Settings

Introduction: Rapid, reliable prognostication is essential for guiding triage and clinical management in resource‑constrained trauma centers. We evaluated three widely used scoring systems—the Injury Severity Score , Revised Trauma Score , and Trauma and Injury Severity Score—to determine their accuracy in predicting in‑hospital mortality. Methods: In this single‑center cohort study, we retrospectively analyzed 1,027 consecutive trauma admissions to the University Hospital for Traumatic Injuries in Albania (October 2024–January 2025). Of these, 429 patients (41.8%) required hospitalization. We calculated Injury Severity Score, Revised Trauma Score, and Trauma and Injury Severity Score for each hospitalized patient and compared predictive performance using means ± SD, p‑values, receiver operating characteristic (ROC) curves, and calibration analyses. Results: The mean Injury Severity Score was 21.9 ± 6.1 (p < 0.001), indicating strong association with mortality but limited definitive prognostication. Revised Trauma Score averaged 7.43 ± 1.20 (p < 0.001), demonstrating utility for rapid triage with a clear inverse correlation to mortality. Trauma and Injury Severity Score yielded a mean predicted survival probability of 94.3 ± 4.2 (p < 0.001), outperforming Injury Severity Score and Revised Trauma Score in discrimination (highest area under ROC curve) and calibration, though its accuracy decreased in patients with significant comorbidities. Conclusions: All three scoring systems significantly predicted in‑hospital mortality (p < 0.001), with Trauma and Injury Severity Score offering the most comprehensive prognostic value by integrating anatomical, physiological, age, and injury‑type variables. Injury Severity Score remains useful for surgical urgency planning, and Revised Trauma Scale excels in immediate triage. However, none fully account for comorbid conditions, underscoring the need for enriched datasets and AI‑driven models in future large‑scale research.