@article{JMAI11468,
author = {Sam Hoang Hong and Kathryn Crowder and Hossein Piri and Deepa Suryanrayan and Sana Sajjad and Jon Rokne and Gouri Ginde},
title = {AI-enabled decision support systems for patient triage: a scoping review},
journal = {Journal of Medical Artificial Intelligence},
volume = {9},
number = {0},
year = {2026},
keywords = {},
abstract = {Background: Nurse-led triage in emergency departments (EDs) is critical for prioritizing care but remains prone to mistriage. While artificial intelligence (AI)-enabled Clinical Decision Support Systems (CDSS) aim to reduce this variability, their real-world integration into nursing workflows and impact on clinical outcomes remain poorly defined. This scoping review aimed to map the current landscape of AI- and machine learning (ML)-enabled CDSS for nurse-led ED triage, addressing two complementary research questions: (RQ1) what AI/ML models are applied and how is their performance reported; and (RQ2) what facilitators and barriers to real-world implementation are identified.Methods: Guided by the PRISMA-ScR framework, this scoping review searched six databases (Embase, MEDLINE, CINAHL, PubMed, Scopus, PsycINFO) covering January 2020–November 2025, supplemented by a Web of Science search [2020–April 2026] and backward snowballing [2018–2019]. Inclusion criteria encompassed primary empirical studies evaluating AI/ML-based CDSS in patient triage contexts, published in English; studies addressing only diagnostic or prognostic tasks, or conducted outside healthcare settings, were excluded.Results: Forty primary empirical studies were synthesized. Traditional ML provides high consistency for acuity scoring, while multimodal architectures integrating text, vital signs, and imaging demonstrate strong performance across settings. Generative AI offers superior speed but faces “reasoning traps” and an over-triage bias that risks resource strain. A significant “explainability paradox” exists: clinicians find technical visualizations burdensome, preferring natural language summaries aligned with clinical guidelines. AI’s inability to quantify “soft skills” and non-verbal cues remains a major barrier to professional trust.Conclusions: This scoping review has mapped the rapidly evolving landscape of AI and ML applications in CDSS and nurse-led patient triage, covering selected studies from 2018 to April 2026, including database searches from 2020–2025, backward snowballing from 2018–2019, and supplementary Web of Science coverage through April 2026. It also provides a knowledge gap and perspective from professionals and clinical staff on the AI integration. Through this scoping review, researchers can potentially find knowledge gaps and information for further development and reference for their current work.},
issn = {2617-2496}, url = {https://jmai.amegroups.org/article/view/11468}
}