Machine learning-based prediction of ischemic cardio-cerebrovascular events after endovascular or microsurgical treatment of unruptured intracranial aneurysms and risk stratification by the early post-treatment triglyceride-glucose index.

Background: Ischemic cardio-cerebrovascular events (ICCEs), including acute coronary syndrome and ischemic cerebral infarction, remain clinically important complications after endovascular or microsurgical treatment of unruptured intracranial aneurysms (UIAs). However, early identification of patients at high post-treatment ischemic risk remains challenging, and reliable risk-stratification tools are lacking. Objective: To develop a machine learning-based framework for predicting ischemic cardio-cerebrovascular events (ICCEs) within 6 months after treatment in patients with unruptured intracranial aneurysms (UIAs) and to evaluate the risk-stratification value of the early post-treatment triglyceride-glucose (TyG) index. Methods: A total of 1,954 patients with UIAs who underwent microsurgical or endovascular treatment between December 2021 and December 2024 were enrolled from the China Treatment Trial for Unruptured Intracranial Aneurysm (ChTUIA) registry. Nine predictive models, including logistic regression as a baseline comparator, were evaluated after feature selection using least absolute shrinkage and selection operator regression and the Boruta algorithm. The synthetic minority over-sampling technique was used to address class imbalance. Model performance was assessed by discrimination, calibration, and clinical utility metrics, and the optimal model was interpreted using SHapley Additive exPlanations. The association between the post-treatment day-3 TyG index and ICCEs was analyzed using multivariable Cox regression, restricted cubic spline analysis, and subgroup analyses. Results: During the 6-month follow-up, 240 of 1,954 patients (12.28%) developed ICCEs. Of the included patients, 1,343 underwent endovascular treatment and 611 underwent microsurgical treatment. Among all models, CatBoost achieved the best overall performance, with an accuracy of 0.875 and an area under the receiver operating characteristic curve (AUROC) of 0.945 (95% CI, 0.927-0.963). SHAP analysis identified the post-treatment TyG index as one of the most influential predictors. In multivariable analysis, each 1-unit increase in TyG was associated with a 2.61-fold higher hazard of ICCEs (HR = 2.61, 95% CI: 2.29-2.96, Conclusion: The CatBoost model demonstrates strong predictive performance for post-treatment ICCEs in UIA patients. The early post-treatment TyG index is independently and nonlinearly associated with ICCE risk and may serve as a simple, practical metabolic marker for individualized perioperative risk stratification.