Emergency Severity Index Protocol with Machine Learning

Effective triage in emergency departments is vital for optimizing patient outcomes and resource use, especially in resource-limited contexts like Burkina Faso. This study presents an automated triage system using machine learning (ML) to predict patient priority levels and appropriate medical services based on the Emergency Severity Index (ESI) protocol. We analyzed electronic hospital records from 23,695 patients across three major Burkina Faso health centers collected from 2021 to 2024. Data, including physiological measures (e.g., blood pressure, temperature, SpO2, pulse) and patient complaints, were preprocessed using TF-IDF vectorization. The supervised ML algorithms XGBoost, LightGBM, RandomForest, Logistic Regression, and SVM were developed and tested in a loop to select the best performing model after each training session. Our results highlighted the potential of ML to streamline emergency triage in Burkina Faso (85% accuracy (precision) for priority class (1 to 5) and 74% accuracy (precision) for services class). Particularly for improving the detection of critical cases through further data integration and model refinement.