VISUALIZATION OF COMBAT INJURIES: TEMPORAL STRUCTURE AND DEMOGRAPHIC DETERMINANTS

Abstract

Background. Modern warfare with widespread use of body armour has shifted the spectrum of combat injuries. The study aims to describe the pattern of combat-related injuries from computed tomography (CT) imaging during the full-scale invasion of Ukraine and to assess the predictive value of age alone employing machine learning methods. Materials and methods. Opportunistic retrospective cohort study of 606 consecutively evacuated adult males who underwent CT at a second-level medical centre from April 2022 to September 2025. Multiclass automated machine learning (H2O AutoML) was trained using age as the sole predictor of CT-diagnosed injury category. Results and discussion. No acute pathology was found in 50.3 % of scans. The most common findings were metallic foreign bodies in soft tissues (25.5 %) and extremity fractures. Penetrating torso and severe traumatic brain injuries were rare (< 0.3 %). The age patterns strongly influenced injury pattern: soft-tissue shrapnel wounds predominated in patients < 40 years, whereas fractures and degenerative changes prevailed in older combatants. Over four years, the proportion of chronic and combined injuries increased 2–3-fold. The bestperforming generalised linear model achieved R² = 0.9996, but log-loss remained high (5.04) in middle-aged groups, confirming limited predictive power of age alone. Conclusion. CT remains a gold standard in stratifying combat injuries. Machine-learning models using demographic variables are promising as clinical decision-support tools in resource-constrained wartime settings.