Table 2. The main effects of AI and ML in ED triage.

AI, artificial intelligence; ML, machine learning; ED, emergency department; ESI, emergency severity index; LASSO, Least Absolute Shrinkage and Selection Operator; AUC, area under the ROC curve; XGBoost, extreme gradient boosting; LR, logistic regression

Author	The main effects of AI and ML in Ed triage
Chang, 2022 [38]	A high degree of accuracy and decreased time needed to provide services to critical patients.
Chen, 2023 [34]	The generated prediction model outperforms other prediction models and has better sensitivity and accuracy compared to emergency physicians.
Choi, 2022 [35]	The triage model used in the study could be used to identify patients with a low risk of bacteremia immediately after initial emergency room triage and showed excellent discriminative performance in facilitating early emergency room disposition decisions.
Feretzakis, 2022 [42]	The utilization of AI may have a favorable impact on the future of emergency medicine.
Fernandes, 2020 [36]	The XGBoost model proved to have a higher sensitivity in identifying patients assigned to the Manchester Triage System-3 and suggested complementing the already existing triage system with a ML model and avoiding under-triaging.
Gao, 2022 [22]	AI reduced the number of triage nurses at emergency triage stations and to some extent improved their working efficiency.
Goto, 2018 [43]	The use of ML significantly improved the ability to predict two clinical outcomes (critical care and hospitalization) over the traditional approach using ESI information.
Goto, 2019 [25]	ML in emergency room triage improved the discriminative ability to predict clinical outcomes compared with conventional triage and had a high sensitivity for predicting critical care outcomes to reduce the number of under-triaged children in higher triage levels and avoid over-triaging children who are less ill.
Hong, 2018 [44]	ML can predict hospital admission at emergency room triage. These features can be used to create a model that can be implemented into electronic health records systems as clinical decision support.
Hsu, 2021 [26]	These algorithms provided evidence from ML to aid in clinical decision-making, increase provider awareness of clinical prognosis, and possibly prevent death.
Hwang, 2022 [27]	ML models using a nationwide database can predict critical hospitalizations of pediatric ED visits more effectively than the conventional triage method.
Ivanov, 2021 [32]	ML models demonstrated significantly higher ESI accuracy, measured against the consensus of the expert clinicians supported by the ESI Handbook, than the nurses for all gold records from each study site and for each triage acuity level.
Joseph, 2020 [28]	Deep learning approaches to identifying critically ill patients at ED triage, neural network, and gradient‐boosting models demonstrated significantly higher accuracy than traditional methods of triage, suggesting that these models have the potential to significantly enhance the triage process.
Klang, 2021 [45]	A deep learning model could be used by neurocritical care experts and clinical stakeholders, such as ED clinicians and nursing managers, to identify patients who might need a neuroscience intensive care unit bed early in the ED triage process.
Lee, J. 2021 [23]	This model can increase ED physicians’ confidence in their decisions regarding patient disposition and allow for quicker initiation of hospital admission.
Lee, S. 2023 [48]	Showed promise in predicting length of stay for patients presenting with syncope with a fair to good performance in the AUC ranging from the same day discharge to long length of stay.
Leonard, 2022 [24]	Models that predict admission and discharge can be used for additional decision support information, allow nurses to request beds while waiting for clinical decision-making, or fast track for discharge. Also helps patient flow and hospital overcrowding.
Levartovsky, 2021 [39]	A ML decision support system can be implemented into electronic health records to guide physician clinical decisions regarding imaging, discharging, and admission.
Lin, 2021 [40]	The ML model outperformed the pre-existing conventional tools in identifying sepsis patients.
Liu, 2021 [13]	In the model, the mode of arrival was the most important triage feature; pulse pressure and shock index were found to be beneficial.
Lu, 2022 [41]	ML models can successfully predict ED cardiac arrest based on the clinical features available at triage, aiding the identification of high-risk patients and preventing deaths.
Pai, 2022 [46]	ML can shorten admission time and decrease overall ED wait time by making efficient movement of patients out of the ED.
Patel, 2018 [47]	ML models can be used in triage to differentiate low- and high-risk patients to improve efficiency.
Raita, 2019 [30]	Demonstrates the superior predictive abilities of modern ML models over the conventional model in a large population of adults in the emergency room and reduces under-triaging of critically ill patients.
Sundrani, 2023 [49]	The models that used features from 15 minutes of passive monitoring significantly outperformed models restricted to conventional triage features. This approach could be used to improve the triage of initially stable patients at risk for decompensation and could be applied continuously for real-time estimates of near-term clinical deterioration.
Tu, 2022 [37]	Results showed that the LR algorithm was the best algorithm to predict the mortality risk in patients with TBI in the emergency room triage setting.
Wolff, 2019 [31]	From the model viewpoint, their results showed successful experimentation of different ML techniques that have a recent interest in scientific research with great potential to be used as a triage decision support tool.
Wu, 2021 [50]	Compared to other clinical risk scores, their Least Absolute Shrinkage and Selection Operator (LASSO) regression model had a superior performance in predicting critical care outcomes in patients with chest pain and minimized the potential over-predicted and under-predicted critical care outcomes that could result in excessive resource allocation to low-risk patients and insufficient treatment of high-risk patients.
Xiao, 2023 [33]	Suggest that the proposed models significantly reduce the under-triage and over-triage issues compared to manual triage, outperform reference models, and can accurately predict patient severity level and clinical department.