From non-agentic large language models to multi-agent systems in emergency medicine: a scoping review

This study aimed to conduct a scoping review of studies on non-agentic large language models (LLMs), LLM-based agents, and multi-agent systems reported in emergency medicine, and to identify current research trends and major gaps by analyzing their clinical application scope, system structures, evaluation approaches, and input data characteristics.

The Web of Science, Scopus, PubMed, and CINAHL databases were searched for literature published from March 8, 2021, to March 7, 2026. Among English full-text articles, studies addressing the application, evaluation, or benchmarking of non-agentic LLMs, LLM-based agents, or multi-agent systems in emergency medicine or the emergency department (ED) were included. Through reference tracking, 35 studies were analyzed.

Of the 35 included studies, 26 were application studies, 6 were framework studies, and 3 were benchmark studies. The studies were concentrated on a limited set of tasks, including triage, diagnostic and treatment decision support, and documentation. In terms of system type, non-agentic LLMs were the most common (n=25), followed by LLM-based agents (n=7) and multi-agent systems (n=3). Inputs were predominantly text-based, and evaluation mainly relied on expert comparison, retrospective record review, vignette-based comparison, and task-specific performance metrics. In contrast, workflow-level, prospective, and safety and trustworthiness-oriented evaluation were limited.

LLMs in emergency medicine have shown potential for task-level decision support and documentation. However, current literature remains focused on non-agentic LLM-based task support, while studies reflecting the dynamic workflow of real EDs remain limited. Future research should expand toward workflow-aware design, operational evaluation, multimodal data integration, multi-agent-based role coordination, and safety and trustworthiness validation.