ID:70 FC-PC1-8 Machine Learning Models for Predicting Outcomes in Patients Receiving Extracorporeal Membrane Oxygenation

Background: Prognostic assessment in patients receiving extracorporeal membrane oxygenation (ECMO) remains challenging due to heterogeneous clinical presentations and complex disease trajectories. Machine learning (ML) approaches may improve outcome prediction by integrating multiple clinical variables. Methods: We retrospectively analyzed data from 342 ECMO patients. Clinical variables included demographics, laboratory values, comorbidities, and ECMO-specific parameters. Five models were developed and compared: Logistic Regression, Support Vector Machine (SVM), Random Forest, CatBoost, and XGBoost. The dataset was randomly divided into training and testing cohorts. Model performance was evaluated using accuracy, area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value (PPV), and F1-score. Receiver operating characteristic (ROC) curves were generated for comparative analysis. Results: Among the evaluated models, XGBoost demonstrated the highest discriminative performance with an AUC of 0.678, followed by CatBoost (AUC 0.635) and Random Forest (AUC 0.607) (Figure). Tree-based ensemble models consistently outperformed Logistic Regression and SVM. CatBoost achieved the most balanced classification performance, with a sensitivity of 0.708 and an F1-score of 0.660. ROC curve analysis showed improved separation between survivors and non-survivors using ensemble-based methods. Conclusion: Machine learning models, particularly gradient boosting algorithms like XGBoost, provide a viable framework for predicting in-hospital mortality in ECMO patients. These models may support clinical risk stratification, while external validation in larger multicenter cohorts is warranted.