Abstract
Emergency departments (EDs) are critical points of entry in healthcare systems where timely and accurate decision-making is vital. Machine learning (ML) offers promising capabilities to enhance patient triage, optimize resource allocation, and improve clinical outcomes in these high-pressure environments. This systematic review investigates the application of ML in EDs, identifies commonly used algorithms and tools, examines their limitations, and provides recommendations for improvement. A structured literature search was conducted across 5 major databases: Google Scholar, Scopus, Web of Science, IEEE Xplore, and PubMed, yielding 1257 peer-reviewed articles. Studies were included if they were published between 2017 and 2024, written in English, and focused on ML applications in EDs within the fields of Computer Science, Engineering, Decision Science, or Mathematics. Exclusion criteria eliminated articles under 6 pages, inaccessible full texts, non-ML-focused studies, and publications such as proposals, abstracts, or book reviews. After screening and quality assessment by 2 independent reviewers, 27 studies were selected for in-depth analysis. Of these, 88.9% were journal articles, 7.4% book chapters, and 3.7% conference proceedings. Findings reveal that the various ML algorithms applied in EDs are context-dependent and use various evaluation metrics, while tools for data extraction and analysis include Python, Keras, TensorFlow, SQL, MATLAB, RStudio, and IBM SPSS. The identified limitations involved data complexity, model accuracy, lack of generalizability, and incomplete datasets. Recommendations across studies emphasized the need to broaden data sources, integrate additional predictors, and improve algorithmic comparisons. This review contributes to the growing body of knowledge on ML in emergency care by synthesizing current practices, highlighting critical challenges, and offering practical directions for future research and implementation.