Internal to Common Carotid Artery Peak Systolic Velocity Ratios for Predicting North American Symptomatic Carotid Endarterectomy Trial Stenosis: Derivation/Validation Study Using a Machine Learning Technique

Abstract

The association between internal carotid artery/peak systolic velocity and stenosis severity as measured by the North American Symptomatic Carotid Endarterectomy Trial is known. The association of internal carotid artery peak systolic velocity to common carotid artery peak systolic velocity ratio is less well studied. We use a machine learning algorithm to study this association. We performed a meta-analysis of papers with point data showing graphs of internal carotid artery/peak systolic velocity ratio versus North American Symptomatic Carotid Endarterectomy Trial percent stenosis. We used a neural net algorithm to derive an equation relating internal carotid artery/common carotid artery peak systolic velocity to % stenosis in a derivation group (two thirds of the data points) and applied it to a validation subset (one third of the data points). Model performance was assessed by correlation coefficients and Bland-Altman analyses. We found 4 papers with appropriate data for a total of 775 data points. The mean % stenosis was 53% (26% SD) with a mean internal carotid artery/common carotid artery peak systolic velocity ratio of 3.9 (2.9 SD). The derivation data set (n = 516) showed an association with an r value of 0.76 (P < .0001) between predicted and measured stenosis. Applying the derived equation to the validation subset (n = 259) showed a similar association (r = 0.8; P < .0001). A machine learning algorithm gave a good approximation of the association between internal carotid artery/common carotid artery peak systolic velocity ratio and % stenosis on a continuous scale for the aggregate data of 4 published studies. These data could be used to study the accuracy of different cut-points for 50% and 70% stenosis in an unbiased fashion.