Machine learning models and restricted cubic spline were employed to analyze and predict postoperative ischemic stroke in type A aortic dissection patients

Abstract Background Ischemic stroke remains a devastating postoperative complication in Type A aortic dissection (TAAD) patients, contributing significantly to elevate mortality rates. Identifying reliable predictors for ischemic stroke risk is crucial for implementing timely clinical interventions. This study endeavored to develop and validate a machine learning-based predictive model for ischemic stroke risk stratification in TAAD patients undergoing surgical treatment. Methods This retrospective cohort study analyzed 430 TAAD patients who underwent total aortic arch replacement with frozen elephant trunk implantation at Beijing Anzhen Hospital (2015–2021). The cohort was randomly partitioned into training (70%, n = 301) and validation (30%, n = 129) sets. We selected the top 8 outcome-relevant variables by ranking the intersecting features from the Boruta and LASSO algorithms by their AUC values. seven machine learning models were evaluated through receiver operating characteristic (ROC) curves, decision curve analysis (DCA), Precision-Recall (PR) curve and calibration plots. Model interpretability was enhanced via Shapley Additive Explanations (SHAP), while restricted cubic splines (RCS) elucidated potential non-linear/liner relationships between predictors and result. Results The Random Forest model demonstrated superior predictive performance over all other models, with a mean area under the curve (AUC) of 0.810 in the validation cohort and 0.806 in the test cohort. SHAP analysis identified key predictors of postoperative ischemic stroke, including Operative Time, Cardiopulmonary Bypass Time, Intraoperative Blood Loss, Intraoperative Plasma Transfusion ml, age, Myoglobin(Mb), Aortic Cross Clamp Time, and Left Subclavian Artery Perfusion. Additionally, restricted cubic splines (RCS) were independently applied to each continuous variable to examine their nonlinear relationships with the outcome. Finally, we subsequently developed a risk assessment calculator and made it publicly accessible online. Conclusion The Random Forest model model demonstrates the best predictive capacity for postoperative ischemic stroke in TAAD patients, offering clinicians a tool for early postoperative risk stratification and personalized therapeutic optimization.