Machine Learning Based Prediction of 28-Day Mortality in ECMO Patients: A Pilot Study Using MIMIC-IV Database

Area of Science:

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

Background:

• Extracorporeal membrane oxygenation (ECMO) is vital for severe cardiac/respiratory failure.
• Predicting ECMO patient outcomes is complex due to therapy's dynamic nature.
• Machine learning (ML) shows promise for prognostication by analyzing complex clinical data.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting 28-day mortality in ECMO patients.
• To identify key clinical predictors associated with mortality in this population.
• To assess the clinical utility of ML-based risk stratification for ECMO therapy.

Main Methods:

• Retrospective analysis of 162 ECMO patients from the MIMIC-IV v3.1 database.
• Feature selection using LASSO regression, followed by application of ML algorithms (Logistic Regression, Random Forest, XGBoost, SVM, Decision Tree).
• Model performance assessed via Area Under the Curve (AUC), calibration curves, and Decision Curve Analysis (DCA).

Main Results:

• The Random Forest model demonstrated the highest predictive performance with an AUC of 0.852.
• Key predictors of 28-day mortality identified include ACT, age, and Mean Arterial Pressure (MAP).
• Decision Curve Analysis confirmed substantial net clinical benefit, indicating practical utility.

Conclusions:

• Machine learning, particularly Random Forest, significantly enhances mortality prediction for ECMO patients.
• ML models provide more accurate and individualized risk stratification by integrating dynamic clinical variables.
• Future research should explore multi-center validation and time-series models for improved clinical applicability.