Beyond the Categories: A Formula-Driven Prediction of Carotid Stenosis

Abstract

Purpose

Current methods to predict carotid stenosis from ultrasound duplex criteria involve assigning a category of stenosis on the basis of an individual laboratory-defined combination of peak systolic velocity (PSV), end diastolic velocity (EDV), and ratio of internal carotid artery velocity to common carotid artery velocity. This study will define a formula by use of regression analysis of the duplex ultrasound criteria compared with the angiographic results. This study will then compare the formula predictions of stenosis with the current means of combining categories to determine whether there is an increase in accuracy and correlation with angiographic findings.

Methods

A retrospective review of the duplex scans and NASCET-defined angiogram results from 209 patent carotid arteries in 114 patients over the course of 4 yr at a single institution was performed. Regression analysis comparing each of the PSV, EDV, and internal carotid artery/common carotid artery ratios (RATIO) with angiographic stenosis was performed. Simple and multiple linear regression equations were obtained. The equation was tested for validity. The data were then reanalyzed by use of the formulas, and predicted stenoses from the formulas were obtained. The formula-predicted stenoses (F1 and F2), category-based stenoses (READAS), and angiographic stenoses were compared. A determination was then made of their statistically significant difference by use of the Wilcoxon signed rank test and receiver operator curve (ROC) analysis.

Results

An r² value of 0.7231, 0.6341, and 0.7262 was obtained, respectively, for the equations comparing PSV, EDV, and ICA/CCA ratio with angiographic stenosis. Limiting the data to stenosis >30% resulted in correlation coefficients between the regression formula predicted data and the angiographic data of 0.71. A statistically significant difference was demonstrated between the category results and angiography (p < 0.0001). No statistically significant difference was demonstrated between the formula-predicted data and the angiographic data. ROC analysis and Area (AZ) test demonstrated a statistically significant difference and better prediction of a >60% stenosis by the regression equation than by the current category method (p < 0.05).

Conclusion

Regression analysis of duplex data versus NASCET-defined angiographic findings allows formation of a model to predict carotid stenosis. This can be done with greater accuracy than the commonly accepted means of categorizing the duplex results.