Artificial Intelligence-Assisted Triage in Emergency Departments: A Scoping Review of Clinical Applications, Outcomes, and Implementation Challenges.

OBJECTIVE: Emergency department (ED) triage determines patient prioritization, early risk recognition, and allocation of limited resources. Artificial intelligence (AI) has been explored to support triagerelated decision-making, but the evidence remains heterogeneous. This scoping review aimed to map applications of AI-assisted triage in EDs and summarize reported outcomes, safety, equity, and implementation challenges. METHODS: This scoping review followed Arksey and O'Malley's framework. Scopus, PubMed, the Cumulative Index to Nursing and Allied Health Literature (CINAHL), Web of Science, and manual searching were used. Original empirical studies published in English between 2015 and 2026 were included if they evaluated AI tools supporting ED triage, risk stratification, resource prediction, or patient-flow decision-making. RESULTS: Of 1,865 records identified, 27 studies met the inclusion criteria. Included studies used machine learning, deep learning, natural language processing (NLP), artificial neural networks, interpretable machine learning, AI-informed decision-support systems, and large language models (LLMs). AI was applied to acuity classification, admission prediction, intensive care unit (ICU) admission prediction, mortality prediction, sepsis detection, waiting-time estimation, and patient-flow optimization. Conventional machine learning and NLP models generally reported promising predictive performance, particularly when structured triage variables were combined with unstructured clinical text. However, fewer studies evaluated prospective validation, workflow integration, measurable clinical and operational impact, equity, or post-deployment monitoring. Evidence on LLMs remains preliminary, with concerns about undertriage, inconsistency, hallucination, local adaptability, and the need for supervised use. CONCLUSION: AI-assisted triage may support ED decision-making and patient-flow management. Future implementation should prioritize supervised human-AI collaboration, prospective validation, explainability, fairness assessment, clinician training, workflow integration, and continuous monitoring.