Predicting Survival After Extracorporeal Membrane Oxygenation by Using Machine Learning.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Critical Care Medicine

Background:

• Venoarterial extracorporeal membrane oxygenation (VA-ECMO) is life-saving but linked to significant patient morbidity and mortality.
• Clinical decision-making for VA-ECMO patients requires enhanced support tools.
• High resource utilization necessitates improved patient management strategies.

Purpose of the Study:

• To develop a machine learning (ML) algorithm for augmenting clinical decision-making in VA-ECMO patients.
• To predict survival to discharge using initial laboratory values during VA-ECMO support.
• To compare the ML model's performance against the established SAVE score.

Main Methods:

• Retrospective review of 282 adult patients undergoing VA-ECMO from May 2011 to October 2018.
• Utilized laboratory values from the initial 48 hours of VA-ECMO support.
• Trained a deep neural network on 70% of data, validated on 15%, and tested on 15%, comparing AUC with the SAVE score.

Main Results:

• The ML model achieved 82% accuracy in predicting survival to discharge in the testing cohort.
• Key predictors identified included lactate, age, total bilirubin, and creatinine.
• The ML model demonstrated a significantly higher area under the curve (0.92) compared to the SAVE score (0.65; P = .01).

Conclusions:

• Machine learning models show potential to enhance clinical decision-making for VA-ECMO patients.
• This proof-of-concept study highlights the utility of AI in critical care settings.
• Further validation using multi-institutional data is recommended for broader applicability.