Real-time machine learning model to predict short-term mortality in critically ill patients: development and...

Area of Science:

• Critical care medicine
• Machine learning in healthcare
• Predictive modeling for patient outcomes

Background:

• Real-time prediction of short-term mortality in critically ill patients is crucial for timely intervention.
• Existing models require validation across diverse clinical settings and ethnicities.
• Development of accurate predictive models is essential for improving patient care in intensive care units.

Purpose of the Study:

• To develop and validate an ensemble machine learning model for predicting short-term mortality in critically ill patients.
• To assess the model's performance using routinely collected clinical variables.
• To compare the model's predictive accuracy against the National Early Warning Score (NEWS).

Main Methods:

• An ensemble model combining deep learning and light gradient boosting machine was developed.
• Internal validation used data from a South Korean academic hospital (2007-2021).
• External validation utilized large datasets from MIMIC, eICU-CRD, and AmsterdamUMCdb.

Main Results:

• The developed model (iMORS) achieved high internal AUROC of 0.964 and external AUROCs ranging from 0.870 to 0.890.
• iMORS significantly outperformed NEWS across all validation datasets (p < 0.001).
• The model demonstrated robust predictive performance in diverse international cohorts.

Conclusions:

• The iMORS model offers excellent real-time prediction of short-term mortality in critically ill patients.
• Validated internally and externally, the model shows significant potential for clinical application.
• This machine learning tool can serve as a valuable decision-support system for intensive care unit clinicians.