Abstract
A machine learning model was developed and validated to predict postoperative complications in patients with acute type A aortic dissection (ATAAD) who underwent total arch replacement combined with frozen elephant trunk (TAR + FET), with the goal of improving postoperative survival quality and guiding clinical treatment. We retrospectively analyzed data from 635 ATAAD patients who underwent TAR + FET surgery at our institution between January 2018 and October 2023. Based on the International Aortic Arch Surgery Study Group definition of Major Adverse Outcomes (MAO), the entire dataset was divided into 160 patients with MAO and 475 patients without MAO. We utilized 66 variables to train 190 machine learning models. The SHAP method identified 11 strong predictors to create a simplified model. We evaluated the predictive performance and clinical utility of both models using receiver operating characteristic (ROC) curves, precision-recall curves (PRC), calibration plots, and clinical decision curves. The combination of Random Survival Forest (RSF) and Gradient Boosting Machine (GBM) was identified as the best predictive model. Both the full model and the simplified model achieved an area under the ROC curve above 0.85 and an area under the PRC curve above 0.703. The Brier values for the simplified model's calibration outcomes in the training and validation sets were 0.124 and 0.138, respectively, with a clinical utility risk threshold probability range of 0.2 to 0.9. A web-based simplified prediction model was developed (https://pmodel.shinyapps.io/pmodel/), enabling the prediction of complication risk in ATAAD patients undergoing TAR + FET surgery, thereby guiding clinical treatment decisions. The combination model of RSF and GBM effectively predicts the risk of postoperative complications in ATAAD patients, helping surgeons identify high-risk individuals and implement personalized perioperative management.