Early Prediction of Patient Disposition for Emergency Department Visits Using Machine Learning

This is the authors' abstract. We don't add key points for this paper.

Overcrowding within emergency departments (ED) affects patient satisfaction and quality of care. The leading causes of ED overcrowding are systematic delays between procedures and patient disposition after ED treatment. Early prediction of patient disposition can improve patient flow and optimize allocation of hospital resources. While studies for predicting disposition using machine learning methods have been actively conducted abroad, few have been conducted in South Korea in spite of the lagging emergency medical environment. Previous studies are limited to binary predictions; either hospital admission or discharge. In this study, we attempted to predict disposition (discharge, general ward admission, ICU admission) of patients using initial information of ED patients from the Korean national emergency department information system (NEDIS). We used five machine learning methods including logistic regression, decision tree, random forest, CatBoost, and TabNet. The results showed that CatBoost yielded the best performance. This result can aid in decision making by providing standard indicators for hospital admission.