Machine learning outperforms the Canadian Triage and Acuity Scale (CTAS) in predicting need for early critical care

Area of Science:

• Emergency Medicine
• Artificial Intelligence
• Health Informatics

Background:

• Emergency department (ED) triage is crucial for prioritizing patient care.
• The Canadian Triage Acuity Scale (CTAS) is a widely used but potentially improvable triage system.
• Identifying patients needing critical care early is vital for improving outcomes.

Purpose of the Study:

• To compare the predictive performance of machine learning (ML) models against the CTAS for identifying patients requiring critical care within 12 hours of ED arrival.
• To investigate the potential of ML to enhance ED triage accuracy.

Main Methods:

• Developed and evaluated three ML models (LASSO regression, gradient-boosted trees, deep learning) using retrospective data from 670,841 ED visits.
• Compared ML model performance against CTAS using area under the receiver-operator characteristic curve (ROC) and precision-recall curve (PRC) metrics.
• Utilized Shapley additive explanation scores to analyze predictor importance.

Main Results:

• ML models demonstrated superior performance: deep learning (ROC 0.926), gradient-boosted trees (ROC 0.912), and LASSO regression (ROC 0.892).
• CTAS achieved a lower ROC of 0.804.
• ML models also showed higher precision-recall curve values compared to CTAS.

Conclusions:

• Machine learning models significantly outperform CTAS in identifying patients likely to need early critical care at ED triage.
• ML models hold potential for improving the discrimination and reliability of triage algorithms.
• Future validation studies are recommended for incorporating ML into revised CTAS protocols.