Multiple Regression Predictive Model For Carotid Artery Stenosis Using Duplex Ultrasound
Christian L. Dohring, BS, Joshua T. Geiger, BS, Adam J. Doyle, MD.
University of Rochester Medical Center, Rochester, NY, USA.
OBJECTIVES: Carotid duplex ultrasound (CDUS) is the most common imaging modality used to assess carotid artery stenosis. Many velocity criteria have been proposed with different cutoff values for varying degrees of stenosis. We sought to produce a multiple linear regression predictive model to predict the percentage of carotid artery stenosis. METHODS: A retrospective review of all patients between 2010 and 2019 who underwent CDUS and a neck CTA within 6 months was conducted. Vessel diameter and corresponding CDUS data were recorded. Data from this cohort were added to a previously reported deidentified data set from patients between 2000 and 2009. A multiple linear regression model was created using peak systolic velocity (PSV), end-diastolic velocity (EDV), and PSV to common carotid artery PSV ratio (PSVR) were used as predictors. The model was compared the to the Society of Radiologists in Ultrasound Consensus Conference (SRUCC) criteria by applying each to the dataset for assessing >50% and >70% stenosis.
RESULTS: A total of 905 vessels were included. Regression analysis created a model for predicting percent stenosis: y = -69.54 + 50.55*log [PSV] + 16.89*log [PSVR] (adjusted R2 =0.67). Using y = 50 as a cutoff for 50% stenosis yielded a sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and overall accuracy of 81.7%, 91.6%, 93.7%, 76.7%, and 85.6, respectively. Using y = 70 as a cutoff for 70% stenosis yielded a sensitivity, specificity, PPV, NPV, and overall accuracy of 69.0%, 92.2%, 80.5%, 86.5%, and 84.9%, respectively. Performance statistics for varied thresholds are represented in Table I. SRUCC criteria showed to have a sensitivity, specificity, PPV, NPV, and overall accuracy of 95.6%, 73.0%, 84.3%, 91.6%, and 86.6, respectively, for >50% stenosis and 88.5%, 77.5%, 64.6%, 93.5%, and 80.9%, respectively, for >70% stenosis.
CONCLUSIONS: Our predictive regression model is not restricted to a single cutoff value, providing user flexibility in determining an appropriate carotid artery stenosis for surgery. The model also provides an increased specificity compared to the SRUCC criteria. If applied, this model would eliminate potential false positives. 
Back to 2021 ePosters