Construction and validation of a machine learning-based prediction model for venous thromboembolism in lung transplant recipients supported by ECMO.

Introduction: This study investigated risk factors for venous thromboembolism (VTE) in lung transplant recipients receiving ECMO and developed a VTE risk prediction model based on machine learning (ML). Methods: We retrospectively reviewed the medical records of 189 patients who underwent elective lung transplantation at the Second Affiliated Hospital of Zhejiang University from May 2023 to November 2024. Recursive Feature Elimination was used to analyze risk factors for VTE during ECMO after lung transplantation. Six ML models were established. Grid search combined with 5-fold cross-validation identified optimal parameters for the models. The models' predictive performance was assessed using 5-fold cross-validation. The evaluation metrics included accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), receiver operating characteristic curve (ROC) and its area under the curve (AUC), as well as the F1 score. Six predictive factors were included in the model construction. Results: The Random Forest model performed best. The area under the ROC of the model on the validation set was 0.895 (95% CI: 0.788-1.000); the accuracy was 89.7%, sensitivity 89.7%, specificity 89.5%, PPV 0.946%, and negative predictive value 81.0%. The calibration curve showed strong agreement between predicted probabilities and observed outcomes; decision curve analysis indicated significant clinical utility across relevant threshold probabilities. Conclusion: The ML-derived VTE risk prediction model for lung transplant patients showed strong predictive ability and clinical utility confirmed by decision curve analysis.