Bacteriological profiling and antimicrobial resistance of combat-related infections: analysis of war wounds from Ukraine during 2024–2025

Combat-related wound infections in the Russo-Ukrainian war pose high morbidity due to the prevalence of multidrug-resistant (MDR) organisms. The aim of the study was to characterize pathogen distribution, antimicrobial resistance (AMR) dynamics, and risk factors (evacuation and sampling timing) with the intent to develop a prognostic model for resistance in wounded military personnel. A retrospective analysis of 203 patients with ballistic injuries treated at a Role 3 military hospital (Zaporizhzhia, Ukraine) was performed. Microbiological sampling yielded 685 unique strains. Isolates were categorized by year (2024 vs. 2025), sampling timing (≤ 48 h vs. > 48 h), and evacuation interval (< 72 h vs. ≥ 72 h). Susceptibility testing followed EUCAST v.14–15. Bacteriological profiling was made for Multidrug-resistant (MDR) bacteria, Extensively Drug-Resistant (XDR), and Pan-drug-Resistant (PDR) microbes. Data was managed in WHONET Microbiology Laboratory Database Software (2025); statistical analyses used R and AMR packages. A supervised machine-learning workflow using a standard Random Forest algorithm was implemented to predict phenotypic resistance risk, with model performance evaluated by ROC-AUC analysis. Gram-negative bacteria predominated, led by A. baumannii , K. pneumoniae , E. coli , and E. cloacae . Overall resistance rate was 51.2%. True patient-level carbapenem resistance was high in K. pneumoniae (imipenem 83.1%) and A. baumannii (imipenem 91.5%); colistin retained activity. Patient-corrected MDR prevalence significantly escalated from 72.9% in 2024 to 82.3% in 2025 ( p = 0.022), confirming a deepening ecological resistance reservoir. Polymicrobial infections were common (93.6% of episodes), often including MDR/XDR pathogens. Resistance was highest in the early-evacuation/early-sampling (Groups 1, 73.6% MDR) and late-evacuation/late-sampling (Groups 4, 97.1% MDR) groups. Random Forest predicted resistance with an overall AUC = 0.718, demonstrating reliable performance specifically within high-volume cohorts (Groups 1 and 4). Combat wounds demonstrate a severe burden of resistant Gram-negative pathogens with stationary chronological resistance plateaus linked to evacuation and sampling timelines. The rapid colonization kinetics observed during early transit continuum indicate that the institutional reservoir is pervasive across the evacuation network. The machine-learning model provides reliable automated risk stratification for high-volume echelons to optimize empirical selection, while early clinical management must prioritize aggressive surgical debridement and source control over premature antimicrobial escalation.