Predicting Successful Weaning from Veno-Arterial ECMO Using Machine Learning

Area of Science:

• Cardiology
• Medical Informatics
• Critical Care Medicine

Background:

• Extracorporeal membrane oxygenation (ECMO) provides vital cardiopulmonary support for acute heart failure.
• Weaning patients from veno-arterial (V-A) ECMO presents significant clinical challenges and risks.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting successful V-A ECMO weaning.
• To identify key clinical predictors associated with successful ECMO weaning.

Main Methods:

• Retrospective analysis of 122 patients undergoing V-A ECMO at Rennes University Hospital (Jan 2020-Jan 2023).
• Training and evaluation of multiple machine learning algorithms (Random Forest, XGBoost, KNN, SVM, logistic regression) using eHOP data.
• Performance assessment using Area Under the Curve (AUC) metrics.

Main Results:

• Machine learning models demonstrated strong predictive performance, with AUCs ranging from 0.84 to 0.86.
• XGBoost achieved the highest AUC of 0.86 (95% CI: 0.72-0.96).
• Significant predictors of successful weaning included ECMO flow rate, fraction of inspired oxygen (FmO2), and ECMO duration.

Conclusions:

• Machine learning models show promise in assisting clinicians with V-A ECMO weaning decisions.
• Further external validation is necessary to integrate these predictive tools into clinical practice.
• Identifying key predictors can optimize patient management during ECMO support.