A Machine Learning Prediction Model for In-Hospital Mortality in Postoperative Patients With Type A Aortic Dissection

Backgrounds: Scoring systems are increasingly used for risk stratification in the postoperative management of type A aortic dissection (TAAD), but no consensus exists regarding the optimal tool. Thus, this study aimed to develop and validate a machine learning-based model to predict in-hospital mortality among postoperative patients with TAAD. Methods: Data from postoperative TAAD patients treated at two centers were analyzed retrospectively. The primary outcome was in-hospital all-cause mortality. Clustering methods were implemented for feature selection. A random forest algorithm was selected from a range of machine learning methods to develop a postoperative mortality risk prediction tool. Receiver operating characteristic (ROC) curves were employed to assess predictive accuracy and reliability. Model calibration was evaluated by plotting calibration curves. SHapley Additive exPlanations (SHAP) values were calculated to quantify the contribution of each feature. Model performance was compared with the European System for Cardiac Operative Risk Evaluation II (EuroSCORE II). Results: A total of 983 patients who underwent surgical repair for TAAD were included in the study. Eight features were identified as significant predictors of in-hospital mortality. On postoperative day (POD) 1, the final random forest model achieved an area under the receiver operating characteristic curve (AUROC) of 0.8899 (95% confidence interval (CI): 0.7988-0.9469). External validation on POD 1 yielded an AUROC of 0.8331 (95% CI: 0.7722-0.8871). The model demonstrated significantly higher discriminative performance than EuroSCORE II. Conclusions: This study developed and primarily validated an accurate machine learning model to predict the all-cause in-hospital mortality of patients with TAAD undergoing surgical repair.