Predictive Factors Driving Positive Awake Test in Carotid Endarterectomy Using Machine Learning.

Background: Positive neurologic awake testing during the carotid cross-clamping may be present in around 8% of patients undergoing carotid endarterectomy (CEA). The present work aimed to assess the accuracy of an artificial intelligence (AI)-powered risk calculator in predicting intraoperative neurologic deficits (INDs).

Methods: Data was collected from carotid interventions performed between January 2012 and January 2023 under regional anesthesia. Patients with IND were selected along with consecutive controls without IND in a case-control study design. A predictive model for IND was developed using machine learning, specifically Extreme Gradient Boosting (XGBoost) model, and its performance was assessed and compared to an existing predictive model. Shapley Additive exPlanations (SHAP) analysis was employed for the model interpretation.

Results: Among 216 patients, 108 experienced IND during CEA. The AI-based predictive model achieved a robust area under the curve of 0.82, with an accuracy of 0.75, precision of 0.88, sensitivity of 0.59, and F1Score of 0.71. High body mass index (BMI) increased contralateral carotid stenosis, and a history of limb paresis or plegia were significant IND risk factors. Elevated preoperative platelet and hemoglobin levels were associated with reduced IND risk.

Conclusions: This AI model provides precise IND prediction in CEA, enabling tailored interventions for high-risk patients and ultimately improving surgical outcomes. BMI, contralateral stenosis, and selected blood parameters emerged as pivotal predictors, bringing significant advancements to decision-making in CEA procedures. Further validation in larger cohorts is essential for broader clinical implementation.