Emergency Department Triage Hospitalization Prediction Based on Machine Learning and Rule Extraction

Le résumé fourni par la source

The objective of this work was to investigate the usefulness of explainable AI (XAI) functionality in the Emergency Department (ED) triage hospitalization prediction modelling based on machine learning and rule extraction. Prediction modelling was carried out using the MIMIC-IV-ED dataset that contains over 400,000 ED visits for predicting hospitalisation Yes vs No (discharged). The Gradient Boosting (GB) prediction modelling gave the best performance achieving an accuracy, sensitivity and specificity of 83%, 82% and 84% respectively. These results are comparable with previous studies published in the literature. Subsequently, interpretable rules were extracted with a high precision threshold, demonstrating the rules’ overall accuracy in replicating the model’s behaviour. Further work is needed to validate the rules extracted with the ED medical experts as well as to benchmark the findings with other models and studies.Clinical Relevance: The use of explainable AI (XAI) in predicting hospitalization outcomes in the Emergency Department (ED) using machine learning and rule extraction, has the potential to improve clinical decision-making.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.