Harnessing generative artificial intelligence for emergency and disaster medicine: A systematic review of applications, challenges, and future directions
American Journal of Disaster Medicine · 2026-09-01 · 21 (3) · 295-313
DOI: 10.5055/ajdm.0535 ↗Abstract
Objective: Generative artificial intelligence (GAI) offers transformative potential for emergency and disaster medicine but lacks comprehensive evaluation in chaotic, resource-limited settings. This review assesses GAI’s efficacy, limitations, and ethical considerations in disaster response.Methods: Following the Preferred Reporting Items for Systematic reviews and Meta-Analyses guidelines, 30 studies (2015-2025) from five databases were systematically reviewed. Eligible studies evaluated GAI for triage, diagnostics, or resource allocation in disasters. Risk of bias was assessed via the Cochrane Risk of Bias Tool for randomized controlled trials and Risk of Bias In Nonrandomized Studies-of Exposure for observational studies. Meta-analyses used randomeffects models; subgroup analyses explored disaster types and artificial intelligence (AI) classes.Results: Included studies demonstrated GAI’s accuracy in triage (78-92 percent) and diagnostics (91-94 percent sensitivity for imaging). Natural disasters showed consistent benefits (effect size [ES] = 0.85, I2 = 0 percent), while anthropogenic events had variable effects (ES = 1.06, I2 = 96.12 percent). Image-based AI (ES = 0.93) outperformed large language models (LLMs) (ES = 0.87) in reliability. Bias (73 percent of studies) and infrastructure gaps (50 percent) were key barriers, with only 40 percent addressing regulatory needs.Conclusion: GAI enhances disaster response but requires context-specific validation, bias mitigation, and policy frameworks. Image-based AI is currently the most dependable solution, whereas LLMs need further refinement.
EUVIMED