Computed tomography angiographic biomarkers help identify vulnerable carotid artery plaque

Objective: The current risk assessment for patients with carotid atherosclerosis relies primarily on measuring the degree of stenosis. More reliable risk stratification could improve patient selection for targeted treatment. We have developed and validated a model to predict for major adverse neurologic events (MANE; stroke, transient ischemic attack, amaurosis fugax) that incorporates a combination of plaque morphology, patient demographics, and patient clinical information. Methods: We enrolled 221 patients with asymptomatic carotid stenosis of any severity who had undergone computed tomography angiography at baseline and >= 6 months later. The images were analyzed for carotid plaque morphology (plaque geometry and tissue composition). The data were partitioned into training and validation cohorts. Of the 221 patients, 190 had complete records available and were included in the present analysis. The training cohort was used to develop the best model for predicting MANE, incorporating the patient and plaque features. First, single-variable correlation and unsupervised clustering were performed. Next, several multivariable models were implemented for the response variable of MANE. The best model was selected by optimizing the area under the receiver operating characteristic curve (AUC) and Cohen's kappa statistic. The model was validated using the sequestered data to demonstrate generalizability. Results: A total of 62 patients had experienced a MANE during follow-up. Unsupervised clustering of the patient and plaque features identified single-variable predictors of MANE. Multivariable predictive modeling showed that a combination of the plaque features at baseline (matrix, intraplaque hemorrhage [IPH], wall thickness, plaque burden) with the clinical features (age, body mass index, lipid levels) best predicted for MANE (AUC, 0.79), In contrast, the percent diameter stenosis performed the worst (AUC, 0.55). The strongest single variable for discriminating between patients with and without MANE was IPH, and the most predictive model was produced when IPH was considered with wall remodeling. The selected model also performed well for the validation dataset (AUC, 0.64) and maintained superiority compared with percent diameter stenosis (AUC, 0.49). Conclusions: A composite of plaque geometry, plaque tissue composition, patient demographics, and clinical information predicted for MANE better than did the traditionally used degree of stenosis alone for those with carotid atherosclerosis. Implementing this predictive model in the clinical setting could help identify patients at high risk of MANE.