Conversational AI in tactical combat casualty care: Baseline GPT-4o improves medic decision-making

• Published in: Clinical simulation in nursing Vol. 107; p. 101803
• Main Authors: Páleník, Jan, Soták, Michal, Černý, Martin, Komarc, Martin, Svoboda, Norbert, Hayu, Daor, Tyll, Tomáš, Netuka, David, Masopust, Václav, Roubík, Karel, Májovský, Martin
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.10.2025
• Subjects: Artificial intelligence; Combat medicine; Decision-making; Large language models; Mechanical ventilation; Simulation-based training; Tactical combat casualty care; Decision-making; Large language models; Tactical combat casualty care; Simulation-based training; Combat medicine; Artificial intelligence; Mechanical ventilation
• ISSN: 1876-1399
• DOI: 10.1016/j.ecns.2025.101803

•AI support improved ventilator setting accuracy by 79.7% in combat simulation.•GPT-4o provided consistent, guideline-based ventilator recommendations.•Tidal volume accuracy improved most, by 164.2%, due to IBW calculations.•Less experienced and female medics benefited most from AI support.•GPT-4o outperformed human-AI collaboration in ventilator adjustments. Mechanical ventilation is vital in Tactical Combat Casualty Care (TCCC), yet many combat medics lack sufficient training in ventilator management. Although artificial intelligence (AI) shows promise in emergency medicine, its use in combat scenarios remains largely unexplored. This study evaluated AI-assisted ventilator management under simulated battlefield conditions. This prospective, simulation-based observational study using a within-subject design was conducted in February 2025 in the Czech Republic with 42 Czech Armed Forces combat medics from four units. Participants adjusted ventilator settings (Vt, RR, FiO₂, PEEP, I:E) across ten scenarios, five with GPT-4o AI assistance and five without. The AI was customized to TCCC protocols and Joint Trauma System guidelines. Performance was scored objectively and analyzed using two-way repeated-measures ANOVA. AI assistance significantly improved parameter accuracy (p < 0.001), with an overall performance increase of 79.7%. The greatest gains were observed in Vt selection (164.2%), followed by FiO₂ (105.9%) and PEEP (60.6%), while I:E ratio adjustments showed marginal improvement (19%). Less experienced and female medics benefited most. Notably, AI-generated recommendations outperformed AI-assisted human decisions. AI-enhanced support significantly improved ventilator management in combat simulations, suggesting potential to optimize real-time decision-making in austere environments. However, this system is not ready for operational deployment, and results should be interpreted as exploratory pending further field validation. Human oversight and medic accountability remain essential.