*Result*: Sustainable electronic trauma registry with NLP/ML-enhanced ICD-10 classification: 13-year insights from violence-heavy KwaZulu-Natal, South Africa.

*Background: The Electronic Surgical Registry (ESR) was implemented at Harry Gwala Regional Hospital, Pietermaritzburg, South Africa in 2012. Leveraging natural language processing (NLP) and machine learning (ML) for enhanced ICD-10 mechanism classification, its streamlined design has contributed to its longevity. This study evaluates data extracted from the registry (2012-2024), demonstrating the ESR's utility as a sustainable model for resource-constrained settings while providing a comprehensive audit of service delivery and identifying priority trauma cohorts requiring preventative campaigns.
Methods: The data of 13,664 trauma patients were retrospectively analyzed. Analysis included demographics, mechanism of injury, Abbreviated Injury Scale (AIS) and Injury Severity Score (ISS), operative data, patients' utilization of ICU and/or high care, morbidity classification, and patient outcomes. Mechanism of injury was classified using ICD-10 codes via NLP/ML. Missing data occurred in <5 % of cases for key variables and were handled via listwise deletion for descriptive analyses and multiple imputation for comparisons.
Results: The median age was 28 years (IQR 20-38, range 0-92) with 84.6 % male predominance. Classification revealed multi-mechanism patterns of injury (i.e., injuries involving both blunt and penetrating components, often from complex assaults) in 9776 cases (71.5 %). Isolated penetrating mechanisms comprised 2906 cases (21.3 %), isolated blunt mechanisms in 946 cases (6.9 %), and bite injuries included 36 cases (0.3 %). Sharp weapons were used in 48.3 % of assault cases. Female patients (n = 2103, 15.4 %) showed multi-mechanism assault affecting 1531 (72.8 %). Pediatric trauma comprised 1816 cases (13.3 %) with 1269 assault cases (69.9 %). Median ISS was 9 (IQR 4-16) (vs. 8 [IQR 4-13] in 2012, p < 0.001 via Mann-Whitney U test). Operations were required in 3583 patients (26.2 %), primarily with laparotomy (2066 procedures). ICU admission was required in 13.5 % of patients and high care in 3.6 %. Morbidity was identified in 15 % with surgical site infection (SSI) predominating (20.8 %). Mortality was 8.5 % (1161 patients).
Conclusion: The ESR, enhanced by NLP/ML for nuanced ICD-10 classification, reveals KwaZulu-Natal's unique violence-dominated epidemiology as an illustrative example of its utility. Female and pediatric assault data require urgent prevention strategies. The ESR is a dependable tool for resource-constrained trauma services, informing public health priorities through reliable epidemiological surveillance.
(Copyright © 2025 The Authors. Published by Elsevier Ltd.. All rights reserved.)*

*Declaration of competing interest The authors declare no conflicts of interest related to this study.*