Predicting ESI triage score using ensemble machine learning algorithms for patients in emergency medicine at Tehran hospitals

Background: The Emergency Severity Index (ESI) is the most common triage systems for assessingthe urgency of emergency patients. The existing nurse and physician opinion-based triagesystem to measure ESI show some weakness. It is highly recommended switching from triagebased on expert opinion to machine learning-based ESI evaluation. In this study we aim to applyensemble machine learning algorithms to predict ESI score for patients attending emergency departmentsin Tehran city.Methods: Data of ۱۲۰۶۵۰۰۰ patients admitted at the emergency department (ED) at ۴۶ teachingand non-teaching hospitals of Tehran, affiliated to Iran University of Medical Sciences, wereretrieved (۲۰۱۸–۲۰۲۲) from the central database. The predictors used for the ensemble ML werethose variables taken at the time of the initial assessment as part of the triage in the emergencydepartment. In total, ۲۰ input data variables include Pulse rate (PR), Systolic blood pressure(BPS), diastolic blood pressure (BPD), Body temperature, Respiration rate (rate of breathing) RR,Oxygen saturation levels (SpO۲) were used to analyze the data and build the prediction model.Results: The three most common methods were used as base classifier in stacking model. thecomparison of different classification methods in models trained with triage information resultedin a test AUC of ۰.۷۵ for SVM (۹۵% CI ۰.۷۶-۰.۷۷), ۰.۸۹ for MLP (۹۵% CI ۰.۸۸-۰.۸۹) and ۰.۸۷ forXGBoost (۹۵% CI ۰.۹۰-۰.۹۱) for single classifier and the stack final predictor model resulted inAUC of .۰۹۸. other result showed that vital signs PR and BPS are the main predictors of patientsEIS score in the emergency room.Conclusion: If potential admissions are known at an early stage, bed management and administrationcan be informed accordingly and react accordingly. In addition, the prediction of inpatientadmission can also be used as a placeholder for the severity of the disease, the need for emergencymeasures and other subsequent decisions. It was shown that the ensemble algorithms performbetter than the single ones for ESI triage prediction.

کلیدواژه ها:

Ensemble learning ، Emergency Medicine ، Triage system