Cureus. 2026 Jul 1;18(7):e111866. doi: 10.7759/cureus.111866. eCollection 2026 Jul.
ABSTRACT
Conventional triage systems in emergency departments (EDs) face limitations including subjectivity and variability. Artificial intelligence (AI) and machine learning (ML) offer potential solutions. This systematic review evaluates AI/ML models for ED triage, focusing on model types, performance, and comparisons with conventional systems. A systematic search of PubMed, Scopus, Web of Science, and Embase was conducted (January 2021-April 2025). Studies applying AI/ML for ED triage with reported performance metrics were included. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines were followed, and the Prediction Model Risk of Bias Assessment Tool (PROBAST) was used for risk of bias assessment. The protocol was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO) (registration number: CRD420261407520). Narrative synthesis was performed. Eight retrospective or cross-sectional studies were included (n=2,000 to >2.6 million patients). XGBoost was the most consistently top-performing model (AUC: 0.76-0.96). Natural language processing (NLP) improved predictive accuracy when added to structured data. AI/ML outperformed conventional triage scales in direct comparisons (e.g., area under the receiver operating characteristic curve (AUROC) 0.991 vs. 0.844 for pediatric critical illness). PROBAST showed low bias in six studies, unclear in one, and high in one. No prospective implementation studies were identified. AI/ML models, especially XGBoost and NLP-enhanced architectures, show strong predictive performance for ED triage outcomes. However, retrospective designs, lack of external validation, and absent prospective data limit current evidence. Rigorous implementation research is needed before clinical adoption.
PMID:42540130 | PMC:PMC13425186 | DOI:10.7759/cureus.111866