Objective To establish a risk warning model for adverse events related to cerebral ischemia in the early stage after carotid endarterectomy (CEA) based on the XGBoost algorithm.

Methods A total of 208 symptomatic carotid artery stenosis patients treated with CEA from January 2021 to May 2024 were selected as the training set. The patients were divided into the occurrence group and non-occurrence group based on whether the adverse events related to cerebral ischemia occurred within 30 days after the operation. Another 89 patients from June 2024 to March 2025 were selected as the external validation set. The clinical data of the occurrence group and non-occurrence group were collected. Univariate analysis was used to screen the predictive factors of early cerebral ischemia-related adverse events after CEA. Four common machine learning algorithms, namely XGBOOT, decision tree, random forest and support vector machine, were used for classification prediction and comparison to select the best predictive model. The SHAP values were used to conduct interpretability analysis on the important features screened out and external validation, and the predictive value of model. was evaluated.

Results The age, BMI, proportion of hyperlipidemia, proportion of contralateral carotid artery stenosis and hs-CRP in the occurrence group were higher than those in non-occurrence group (P ＜ 0.05). The accuracy and AUC of the XGBOOT model were 87.96% and 0.933 (95%CI: 0.890–0.963), respectively, which were significantly higher than those of decision tree61.73% and 0.816 (95%CI: 0.756–0.866), random forest82.42% and 0.885 (95%CI: (0.834–0.925) and support vector machine72.80% and 0.842 (95%CI: 0.785–0.889) (P ＜ 0.05). The XGBOOT model had the best comprehensive predictive performance. The top two important characteristic variables were hs-CRP and hyperlipidemia, which were the ultimate predictive factors for the risk of cerebral ischemia-related adverse events in the early postoperative period. The external validation results showed that the sensitivity and specificity of the XGBOOT model in predicting the risk of early postoperative cerebral ischemia-related adverse events were 85.71% and 91.46%, respectively. The consistency between this model and actual clinical results was 91.01%, and the Kappa value was 0.554 (P ＜ 0.05). ConclusionsThe comprehensive predictive performance of the XGBOOT model for predicting the risk of early cerebral ischemia-related adverse events after CEA is the best. The hs-CRP and hyperlipidemia are its key predictive factors. Based on this, actively taking prevention and treatment measures is helpful to reduce the incidence of early postoperative cerebral ischemia-related adverse events.