Enhancing Survival Prediction After Venoarterial Extracorporeal Membrane Oxygenation Using Machine Learning.

Area of Science:

• Cardiology
• Critical Care Medicine
• Medical Informatics

Background:

• High in-hospital mortality persists for patients requiring venoarterial extracorporeal membrane oxygenation (VA-ECMO).
• Accurate prediction of survival is crucial for optimizing patient management and resource allocation in VA-ECMO therapy.

Purpose of the Study:

• To compare the predictive performance of the Survival after Venoarterial ECMO (SAVE) score against machine learning (ML) models.
• To evaluate ML models that incorporate extensive electronic medical record data for predicting survival in VA-ECMO patients.

Main Methods:

• Retrospective review of 194 adult patients undergoing VA-ECMO from 2016-2022.
• Development and validation of ML models (CatBoost algorithm) using vital signs, lab values, and ventilator settings.
• Comparison of ML model performance (pre-ECMO and on-ECMO data) against the SAVE score using area under the receiver-operator characteristics curves (AUC).

Main Results:

• The on-ECMO ML model achieved a higher AUC (0.83) than the SAVE score (0.73).
• ML models incorporating ML-identified variables demonstrated stepwise AUC improvements, reaching 0.89 for a combined model.
• On-ECMO variables significantly enhanced predictive performance and identified novel survival-associated factors.

Conclusions:

• Interpretable ML models offer comparable or superior accuracy to the SAVE score for predicting VA-ECMO survival.
• Integrating on-ECMO data into predictive models substantially improves their performance.
• ML approaches hold promise for refining prognostication and guiding clinical decisions in VA-ECMO.