Carnicelli Anthony P, Stone Jonathan J, Doyle Adam, Chowdhry Amit, Gillespie David L, Chandra Ankur
University of Rochester School of Medicine and Dentistry, Rochester, NY 14642.
University of Rochester School of Medicine and Dentistry, Rochester, NY 14642.
Ann Vasc Surg. 2014 Aug;28(6):1548-55. doi: 10.1016/j.avsg.2014.02.010. Epub 2014 Feb 12.
Carotid duplex ultrasound (CDUS) is commonly used to screen for carotid artery stenosis. Specificities of CDUS criteria however are lower than sensitivities, potentially resulting in false-positive examinations with subsequent unnecessary imaging or surgery. Our objective was to establish a multivariate logistic regression to increase the specificity of CDUS for high-grade (≥70%) stenosis.
A retrospective review collected CDUS velocities and radiographic measurements from patients who underwent both CDUS and computed tomography angiography (CTA). After stratification with standard CDUS criteria, a logistic regression was created using peak systolic velocity (PSV), end diastolic velocity (EDV), and PSV ratio (PSV of internal carotid artery [ICA]/PSV of common carotid artery [CCA]) as predictor variables. A receiver operating characteristic curve was generated to test the model's predictive ability. A cutoff probability for unequivocal high-grade stenosis was chosen based on optimal specificity. The regression model was applied to patients with equivocal high-grade stenosis. Probabilities for detection of high-grade stenosis were calculated. Descriptive statistics were generated to quantify the accuracy of the model.
A total of 244 vessels were included. Standardized velocity criteria for ≥70% stenosis yielded a sensitivity of 90.6% (95% confidence interval [CI], 82.3-95.6%), specificity of 63.5% (95% CI, 55.4-70.5%), positive predictive value (PPV) of 57.0% (95% CI, 48.8-65.5%), and negative predictive value (NPV) of 92.7% (95% CI, 85.8-96.5%). Regression analysis produced a model for predicting the probability of high-grade stenosis defined as probability = logit(-1) (-4.97 + [0.00938 × PSV] + [0.0135 × EDV] + [0.103 × PSV ICA/CCA ratio]). A cutoff probability of 0.65 for high-grade stenosis yielded a sensitivity of 54.7% (95% CI, 43.9-65.0%), specificity of 94.3% (95% CI, 89.3-97.2%), PPV of 83.9% (95% CI, 71.6-91.9%), and NPV of 79.3% (95% CI, 72.8-84.5%). A cutoff PSV of 400 cm/sec was chosen for unequivocal stenosis of ≥70%. A total of 94 patients were found to meet criteria for high-grade stenosis (PSV ≥ 230 cm/sec) but fall short of criteria for unequivocal high-grade stenosis (PSV < 400 cm/sec). Application of the regression model resulted in identification of 15 patients with probability ≥0.65 for high-grade stenosis and 79 patients with probability <0.65. This resulted in a 16% potential reduction in CTA scans.
Our regression model provides increased specificity of CDUS for high-grade stenosis in patients who have met initial highly sensitive screening criteria. Application of this model may limit the need for additional imaging and increase the threshold for operative intervention in asymptomatic patients with equivocal high-grade carotid stenosis.
颈动脉双功超声（CDUS）常用于筛查颈动脉狭窄。然而，CDUS标准的特异性低于敏感性，这可能导致假阳性检查，进而引发不必要的影像学检查或手术。我们的目标是建立一个多变量逻辑回归模型，以提高CDUS对重度（≥70%）狭窄的特异性。
一项回顾性研究收集了接受CDUS和计算机断层血管造影（CTA）检查患者的CDUS血流速度和影像学测量数据。根据标准CDUS标准进行分层后，以收缩期峰值速度（PSV）、舒张末期速度（EDV）和PSV比值（颈内动脉[ICA]的PSV/颈总动脉[CCA]的PSV）作为预测变量建立逻辑回归模型。生成受试者工作特征曲线以测试该模型的预测能力。根据最佳特异性选择明确重度狭窄的临界概率。将回归模型应用于疑似重度狭窄的患者。计算检测重度狭窄的概率。生成描述性统计数据以量化模型的准确性。
共纳入244条血管。≥70%狭窄的标准化速度标准的敏感性为90.6%（95%置信区间[CI]，82.3 - 95.6%），特异性为63.5%（95% CI，55.4 - 70.5%），阳性预测值（PPV）为57.0%（95% CI，48.8 - 65.5%），阴性预测值（NPV）为92.7%（95% CI，85.8 - 96.5%）。回归分析得出一个预测重度狭窄概率的模型，定义为概率 = logit(-1)（-4.97 + [0.00938×PSV] + [0.0135×EDV] + [0.103×PSV ICA/CCA比值]）。重度狭窄的临界概率为0.65时，敏感性为54.7%（95% CI，43.9 - 65.0%），特异性为94.3%（95% CI，89.3 - 97.2%），PPV为83.9%（95% CI，71.6 - 91.9%），NPV为79.3%（95% CI，72.8 - 84.5%）。对于明确的≥70%狭窄，选择临界PSV为400 cm/秒。共有94例患者符合重度狭窄标准（PSV≥230 cm/秒）但未达到明确重度狭窄标准（PSV < 400 cm/秒）。应用回归模型后，识别出15例重度狭窄概率≥0.65的患者和79例概率<0.65的患者。这使得CTA扫描潜在减少16%。
我们建立的回归模型提高了CDUS对已符合初始高敏感性筛查标准患者重度狭窄的特异性。应用该模型可能会减少额外影像学检查需求，并提高疑似重度颈动脉狭窄无症状患者的手术干预阈值。
