Machine Learning-Based Predictive Model for Grade 3 Primary Graft Dysfunction Following Lung Transplantation: A Retrospective Cohort Study.

Background: This study aimed to identify key predictors for Grade 3 Primary Graft Dysfunction (PGD) after lung transplantation. Machine learning (ML) algorithm models were constructed for early clinical identification of high-risk PGD patients based on these predictors. Methods: A total of 297 lung transplant recipients from December 2018 to December 2024 were retrospectively enrolled. Patient classification followed the 2016 International Society for Heart and Lung Transplantation (ISHLT) criteria. Results: The area under the receiver operating characteristic curve (AUC) values for the logistic regression (LR), K-Nearest neighbors (KNN), random forest (RF), and decision tree (DT) models in validation cohort were 0.6960, 0.6307, 0.9989, and 0.9138, respectively. The RF algorithm was selected as the optimal predictive model for Grade 3 PGD risk after lung transplantation. The RF model showed a maximum net benefit of 0.2837 at a threshold probability of 0.6 in the training set. In the test cohort, the maximum net benefit was 0.3034 at a threshold probability of 0.5. The net benefit difference between the two datasets was minimal (mean difference: -0.0034). This finding reflected robust generalization capability for the RF model. The most influential features for the RF model's predictions were intraoperative red blood cell transfusion volume, preoperative oxygenation index, donor cold ischemia time, preoperative NT-proBNP, white blood cell count, use of cardiopulmonary bypass (CPB) during surgery, and CRP level. Conclusion: The RF model effectively predicted the risk of Grade 3 PGD after lung transplantation. This model showed potential for providing decision support in the early identification of high-risk patients.