Machine Learning-Based Model to Classify Emergency Severity Index Levels 1-3 in Febrile Patients With Tachycardia:...

Area of Science:

• Emergency Medicine
• Health Informatics
• Machine Learning in Healthcare

Background:

• Febrile patients with tachycardia present complex triage profiles.
• The Emergency Severity Index (ESI) has inter-rater variability issues in Thailand.
• Machine learning (ML) offers potential to improve triage reliability using existing data.

Purpose of the Study:

• Develop and evaluate ML models for predicting ESI levels 1-3 in febrile, tachycardic adults.
• Identify the optimal ML model for clinical triage decision support.
• Enhance the consistency and accuracy of emergency department triage.

Main Methods:

• Diagnostic prediction study involving 500 febrile adults with tachycardia (≥ 37.6 °C, pulse > 100 bpm).
• Data split into 80:20 development and testing sets; expert-assigned ESI levels as outcome.
• Compared Random Forest, XGBoost, and Gradient Boosting Machine models using cross-validation and class-weighting.

Main Results:

• XGBoost achieved the highest discrimination with Area Under the ROC Curve (AuROC) values of 1.00, 0.94, and 0.97 for ESI levels 1-3.
• XGBoost models demonstrated excellent calibration with minimal misclassification.
• The model showed strong predictive consistency across different ESI categories.

Conclusions:

• XGBoost was selected as the best performing model for predicting ESI levels.
• The model will be integrated into the Smart ER system as the Thailand Triage Prediction System (TTPS).
• The TTPS aims to improve real-time triage accuracy and workflow efficiency.