Using emergency department triage for machine learning-based admission and mortality prediction

Area of Science:

• Emergency Medicine
• Artificial Intelligence
• Clinical Decision Support

Background:

• Patient admission decisions in emergency departments often rely on limited data.
• Accurate prediction of patient disposition and mortality is crucial for resource allocation and patient care.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting hospital admission (ward observation vs. intensive care) and 30-day mortality.
• To identify key features influencing these predictions in emergency department patients.

Main Methods:

• A retrospective cohort study of 58,323 adult patients was conducted.
• Machine learning models, including Random Forests and Neural Networks, were trained to predict outcomes.
• Permutation feature importance was used to analyze feature relevance.

Main Results:

• Models achieved high predictive accuracy: AUC-ROC of 0.842 for ward admission, 0.819 for intensive care admission, and 0.925 for 30-day mortality.
• Key predictors for ward admission included age and chief complaint.
• Age and general ward admission were most important for predicting 30-day mortality.

Conclusions:

• Machine learning models demonstrate significant potential in predicting patient disposition and 30-day mortality in the emergency department.
• These predictive capabilities can enhance clinical decision-making for patients triaged using the Manchester Triage System.