Using machine learning tools to predict outcomes for emergency department intensive care unit patients

Area of Science:

• Critical Care Medicine
• Health Informatics
• Machine Learning in Healthcare

Background:

• Global increase in critically ill patients necessitates improved mortality prediction for effective emergency resource allocation.
• Traditional scoring systems struggle with the complex interactions of risk factors in critically ill patients.
• A need exists for accurate predictive models tailored to emergency department critically ill patients.

Purpose of the Study:

• To develop and evaluate machine learning-based scoring systems for predicting 7-day mortality in emergency department critically ill patients.
• To compare the performance of machine learning models against traditional scoring systems.

Main Methods:

• Retrospective cohort study of 1624 critically ill patients (≥16 years) admitted to an emergency department intensive care unit.
• Utilized prospective factors including previous diseases, physiological parameters, and laboratory results.
• Developed and evaluated Support Vector Machine (SVM), Gradient Boosting Decision Tree (GBDT), XGBoost, and logistic regression models for 7-day mortality prediction.

Main Results:

• XGBoost model achieved the highest Area Under the Curve (AUC) of 0.849.
• Other machine learning models (SVM, GBDT, logistic regression) also demonstrated strong predictive capabilities with AUCs ranging from 0.794 to 0.840.
• Machine learning models, especially XGBoost, outperformed the traditional SAPS 3 model (AUC = 0.826) in discriminatory capability.

Conclusions:

• Machine learning models, particularly XGBoost, offer a more reliable approach to predicting mortality in emergency department critically ill patients.
• These advanced models can enhance the accuracy of risk stratification and resource management in critical care settings.
• The findings support the integration of machine learning into clinical decision-making for emergency critical care.