Machine learning methods applied to triage in emergency services: A systematic review

Area of Science:

• Emergency Medicine
• Health Informatics
• Artificial Intelligence

Background:

• Accurate symptom assessment in emergency services is crucial for patient care.
• Technology, specifically machine learning (ML), can enhance predictions from health records and patient flow.
• Improving triage accuracy is a key objective in emergency medicine.

Purpose of the Study:

• To evaluate the effectiveness of machine learning (ML) systems in emergency department (ED) triage.
• To compare the predictive capabilities of ML triage systems against established triage scales and scores.
• To identify the performance of different ML models in predicting patient outcomes in the ED.

Main Methods:

• A systematic review was conducted following PRISMA guidelines.
• Searches were performed across CINAHL, Cochrane, Cuiden, Medline, and Scopus databases.
• The search strategy combined terms: "Machine learning", "triage", and "emergency".

Main Results:

• Eleven studies met the inclusion criteria for the review.
• ML methods consistently predicted critical outcomes such as mortality, critical care needs, and hospital admission.
• XGBoost and Deep Neural Networks demonstrated superior prediction accuracy compared to other ML models, including Logistic Regression.

Conclusions:

• Machine learning (ML) serves as a valuable tool for improving the emergency triage process.
• ML models accurately predict significant emergency variables, including mortality risk and the need for critical care or hospitalization.
• The findings support the integration of ML into emergency department workflows for enhanced decision-making.