Using Machine Learning to Predict Cerebral Hyperperfusion Following Carotid Revascularization
Recommended Citation
Halabi M, Chamseddine H, Shepard A, Rashid A, Nypaver T, Weaver M, Peshkepija A, Kavousi Y, Onofrey K, Kabbani L. Using Machine Learning to Predict Cerebral Hyperperfusion Following Carotid Revascularization. J Vasc Surg 2025; 81(6):e268-e269.
Document Type
Conference Proceeding
Publication Date
6-1-2025
Publication Title
J Vasc Surg
Abstract
Objective: Cerebral hyperperfusion syndrome (CHS) is a rare but serious complication of carotid revascularization that may result in neurological dysfunction. Although the precise etiology of CHS remains unclear, this study aims to employ machine learning techniques to predict its occurrence to develop a comprehensive model identifying clinical features associated with this condition. Methods: Patients undergoing carotid artery stenting (CAS) and carotid endarterectomy (CEA) were identified in the Vascular Quality Initiative between 2005 and 2024. Preoperative and intraoperative variables were collected along with outcomes, including patient demographics and procedural details. Preoperative characteristics were assessed for the entire population, followed by stratification by procedure type to evaluate intraoperative characteristics. Machine learning models were developed to predict the occurrence of CHS. Feature importance was assessed using SHapley Additive exPlanations (SHAP) to identify key predictors, and model performance was evaluated using area under the receiver operating characteristic curve (AUCROC). Results: A total of 247,542 patients were analyzed, including 61,700 undergoing transcarotid artery revascularization (TCAR), 41,550 transfemoral CAS (tfCAS), and 144,292 CEAs. Preoperative characteristics were highly predictive of CHS (AUC = 0.97) across all revascularization modalities. Contralateral stenosis and coronary artery disease were the strongest risk factors, as indicated by high positive SHAP values (Fig. 1). Ipsilateral stenosis and emergency presentation also significantly increased CHS risk. Conversely, protective factors such as preoperative statin and ACE inhibitor use had negative SHAP values, reducing the risk of CHS. Intraoperative characteristics for CEA and TCAR were less predictive, with AUCs of 0.50 and 0.76, respectively. For CEA, shunt use and operative time were the most influential factors, with longer operative times associated with higher CHS risk. In TCAR, contrast volume and procedure time were the top predictors, with marginal associations to increased CHS risk. In contrast, tfCAS showed the strongest intraoperative predictive value (AUC = 0.98). Fluoroscopy time, procedure time, and contrast volume were the most significant contributors, with longer fluoroscopy times and higher contrast volumes markedly increasing the risk of CHS. Conclusions: Preoperative characteristics were highly predictive of CHS, whereas intraoperative predictors for TCAR and CEA had limited value. In contrast, tfCAS showed strong associations with procedural factors such as fluoroscopy time and contrast volume, highlighting its higher risk for CHS. High-risk patients should avoid tfCAS, with alternative strategies such as TCAR or CEA being more appropriate to minimize CHS risk. [Formula presented] [Formula presented]
Volume
81
Issue
6
First Page
e268
Last Page
e269
