Interpretable Machine Learning-Based Triage For Decision Support in Emergency Care

In medicine, triage is a process that rations health services towards patients who are most in need of immediate care. Traditionally, a nurse will carry out an evaluation to ascertain a patient's condition and assign an Emergency Severity Index(ESI) to represent the urgency level needed to attend to them, which is time-consuming and subject to human error and bias. Mistriage could potentially lead to delayed medical attendance and ultimately to death if the condition is not managed, while no triage could overwhelm the finite hospital resources and medical personnel in trying to attend to the huge patient volumes. Existing research in the domain largely focuses on the prediction of patient disposition(whether or not one will be admitted). This work proposes using Explainable AI to generate insights that can support human-in-the-loop Machine Learning triage of patients at the emergency department level. This research utilized Machine learning classification algorithms like the Decision Trees classifier, Random forest classifier, Histogram-Based Gradient Boosting classifier, XGBoost classifier, and ELM Classifier to predict the Emergency Severity Index of a patient based on their medical data. Upon hyperparameter tuning, the models are evaluated and the Histogram-Based Gradient Boosting classifier topped them all as the best-performing model with an average Area Under the Curve(AUC) score of 91 % over the 5 classes and an F1 score of 70%. Explainable AI(XAI) techniques like Local Interpretable Model-Agnostic Explanations(LIME) and Shapley Additive Explanations(SHAP) were used to create more transparent and trustworthy algorithms for Intelligent Healthcare.