Can Bruits Distinguish High-Grade from Moderate Symptomatic Carotid Stenosis?

Methods

We analyzed the clinical profiles from 1268 patients enrolled in the NASCET [1], a randomized, controlled trial conducted in 57 centers across Canada and the United States, between its inception in 1988 and November 1991. Patients were considered for entry if they were younger than 80 years of age, deemed fit for surgery, and had had recent single or recurrent transient ischemic attacks or nondisabling strokes judged to arise from carotid artery disease. Patients were excluded if they had a concurrent illness likely to cause death within 5 years or had other probable cause for their symptoms (such as a cardiac disorder likely to produce emboli).

All eligible patients had systematic history taking and physical examinations, the latter including auscultation of the orbits, supraclavicular area, and neck for focal or diffuse bruits. Although not formally defined, focal carotid bruits were generally taken to be bruits that could only be heard over a well-localized area of the bifurcation of the artery in the region of the upper border of the thyroid cartilage, whereas diffuse bruits could be heard over most of the artery's length. Patients also had other diagnostic studies, including computed tomography (CT) of the head, neck ultrasonography, and bilateral selective carotid angiography. All angiograms were reviewed and graded, using linear diameter measurement from films and comparing the narrowest residual luminal diameter with that of the normal internal carotid well beyond the bulb. This was done by one neuroradiologist, whose intraobserver agreement over chance (κ) was 0.89 for distinguishing 70% or greater stenosis from lesser degrees of stenosis. Patients with surgically amenable stenoses of 30% to 99% on the symptomatic side were then randomly assigned to either best medical therapy (including aspirin, 1300 mg/d) or best medical therapy plus carotid endarterectomy. The baseline findings of these randomized patients constitute the data for this report.

Using standard methods [4, 5], we determined the sensitivity, specificity, and likelihood ratios of bruits at all three sites (orbits and supraclavicular and neck areas), ipsilateral and contralateral, singly and in combination, for distinguishing high-grade (at least 70%) from lesser degrees of stenosis. We did a logistic regression on all baseline clinical and laboratory data (including diagnostic imaging) to identify all statistically significant predictors of 70% to 99% stenosis. Various combinations of bruits and other baseline features were then examined for their ability to distinguish high-grade from moderate stenosis. We estimated the accuracy of the fitted logistic regression model by the “proportion correctly classified” as described by Hosmer and Lemeshow [6].

Results

Of the 1268 patients studied, 68% were men, 29% had had a retinal ischemic qualifying event, 69% had a hemispheric insult, and 2% had both; 61% of the qualifying events were transient ischemic attacks and 39% were minor strokes. The presence of bruits at the baseline examination is shown in Table 1. Orbital bruits were omitted from further analysis because of their extremely low occurrence (0.9%).

The relation between carotid bruits and the degree of stenosis of the underlying internal carotid artery at angiography is shown in Figure 1. Focal carotid bruits became more frequent with increasing degrees of stenosis, peaked in the 70% to 89% range, and became much less common as the degree of stenosis passed 90%.

View larger version:

• In this window
• In a new window

Figure 1. Frequency of focal ipsilateral bruit by stenosis.

As a result, the accuracy of bruits for ruling in and ruling out high-grade stenosis was not high (Table 2). The best performance was shown by the ipsilateral focal carotid bruit, with a sensitivity of 63% (95% CI, 59% to 66%), a specificity of 61% (CI, 57% to 65%), a positive likelihood ratio of 1.61 (CI, 1.4 to 1.8), and a negative likelihood ratio of 0.61 (CI, 0.54 to 0.69) for 70% to 99% stenosis of the underlying carotid artery. Accordingly, a decision rule to do carotid angiography only in patients with carotid bruits, although reducing angiography by 49%, would have excluded 37% of patients with high-grade stenoses from diagnostic confirmation and beneficial surgery.

A logistic regression analysis on all available clinical and laboratory parameters from our data set identified the following five clinically independent predictors for the presence of 70% to 99% stenosis: a qualifying event that was either transient in duration (transient ischemic attack) or solely retinal in distribution, the presence of a focal ipsilateral carotid bruit, evidence of cerebral infarction on computed tomography, and an estimated stenosis of more than 90% on ultrasonography. The following features did not distinguish between high- and lower-grade stenosis: age; sex; ethnicity; employment status; handedness; old cerebrovascular disease; side of qualifying event; or a history of hypertension, diabetes, hyperlipidemia, smoking, claudication, angina pectoris, previous myocardial infarction, heart failure, valvular heart disease, atrial fibrillation, vertebrobasilar insufficiency, seizures, alcohol dependency, or previous aspirin use; or abnormal laboratory determinations of hemoglobin, leukocyte count, platelet count, or serum creatinine.

Table 3 shows the predicted probability of high-grade stenosis, given the presence or absence in any combination of the five predicting factors. Thus, a patient with (transient) amaurosis fugax, a focal carotid bruit on the same side, evidence of previous infarction on computed tomography, and an ultrasound scan indicating greater than 90% stenosis has a 94% probability of high-grade stenosis, substantially higher than the 63% probability generated by the bruit alone. The model classifies 12% of patients as having a high (≥ 80%) and 13% as having a low (≥ 20%) probability of high-grade stenosis. When the model's predictions (≥ 50% or <50%) for high-grade stenosis are compared with those obtained with angiography, however, only 69% of patients are correctly classified in this training set. A decision rule to do carotid angiography only in patients in whom the model predicted high-grade stenosis as 50% or more, although reducing angiography by 55%, would have excluded 36% of patients with high-grade stenoses from diagnostic confirmation and beneficial surgery.

When ultrasonography and computed tomographic scans are removed from the logistic regression, the model's predictive ability changes. Use of the same decision rule would reduce angiography by only 39%, while still excluding 26% of patients with high-grade stenoses from diagnostic confirmation and surgery.

Discussion

Cervical bruits alone are not sufficiently sensitive predictors of high-grade carotid artery stenosis to assist in electing symptomatic patients for angiography. A decision rule to do carotid angiography only in patients with carotid bruits, although reducing angiography by 49%, would have excluded 37% of patients with high-grade stenoses from diagnostic confirmation and beneficial surgery. As shown in Figure 1, not only are bruits found in significant proportions with every degree of stenosis, but the frequency sharply diminishes with stenoses of 90% to 99%, that group of patients who benefit most from surgery. This drop in bruit frequency with stenoses of more than 90% could be the result of a change in hemodynamics, from a state of turbulence at lesser degrees of stenoses to actual compromise in flow at higher ranges of stenoses, causing the intensity of bruits to fade.

By adding four other independent clinical factors identified by logistic regression to bruit, it is possible to generate a wide spectrum of predicted probabilities for the presence of high-grade stenosis. Two of the model's components, the presence of a bruit and an ultrasound scan suggesting high-grade stenosis, make biological sense. That hemispheric (rather than retinal) events are associated with lesser degrees of stenosis is curious, and the divergence of infarction on computed tomography (associated with high-grade stenosis) and minor strokes (associated with less severe stenoses than transient ischemic attacks) is difficult to explain, and underscores how biological sense and statistical findings may have very little in common and why any statistical model such as this one needs to be tested in a second, independent patient group. In any event, the predictive model's ability to correctly classify high-grade compared with moderate stenosis is only 69%, and a decision rule based on it would exclude 36% of patients with high-grade stenosis from confirmatory angiography and beneficial surgery. The lack of stability present in the model also seriously hampers its usefulness.

A shortcoming in our study arises from essential and appropriate elements of the design of its parent trial [1]. For good reason, NASCET did not enroll patients with stenoses of less than 30% on initial angiogram reading and, because from Figure 1 we see that these patients are least likely to exhibit carotid bruits, the exclusion of these patients reduced the “true-negative” entries into our accuracy estimates, causing us to underestimate specificity and the positive likelihood ratio for carotid bruits. For the same reason, the accuracy of our five-element prediction rule may have been underestimated. However, the exclusion of low-grade stenoses cannot affect sensitivity, and it is the low sensitivity of both bruits and the prediction model that limits their clinical usefulness.

A second source of bias could arise if patients with bruits were preferentially referred to NASCET [1]. If so, this would inflate both “true positives” and “false positives,” overestimating sensitivity and underestimating specificity. Again, however, this would not alter our conclusions, for the sensitivity we observed, even if inflated, is still too low for clinical usefulness.

Possible concerns could also arise from the clinical examination within our trial setting. First, neurovascular assessments were limited to auscultation for carotid, ophthalmic, and supraclavicular bruits. Pulse qualities of the carotid and other cervicofacial arteries were not recorded. Second, although we have no data on the assessment reliability for cervical bruits, the literature indicates reasonable interobserver reliability with a κ of 0.67 [7]. The lack of perfect agreement may be partially explained from a physiologic standpoint because the intensity of bruits can fluctuate over time with the heart rate and stroke volume.

Finally, even if the clinical examination for bruits was done with previous knowledge of angiographic results, the resulting inflation of accuracy remained below any threshold for clinical usefulness.

Compared with the rest of the angiographic literature, it appears our results underestimate the accuracy of cervical bruits for high-grade carotid stenosis (Table 4). These higher positive likelihood ratios were attained in populations with lesser prevalence of high-grade disease, leading to overall lower post-test probabilities. For example, Hankey and colleagues [11] found that the presence of a bruit in their study increased the probability of a 75% to 99% stenosis by 22%, from a pretest value of 15% to a post-test value of 37%, again showing the limited predictive accuracy of cervical bruits for high-grade stenosis.

However, there still remain other reasons to examine patients for the presence of cervical bruits. As shown in two community-based studies [12, 13], the presence of an asymptomatic carotid bruit is associated with increased long-term incidence of stroke, myocardial infarction, and death. Bruits are markers of systemic vascular disease and should therefore remain part of every general physical examination.

Although an association exists between a focal ipsilateral carotid bruit and the degree of stenosis of its underlying carotid artery, it is not strong enough to make the presence of a bruit, by itself, an adequate clinical predictor of high-grade disease. Even more important, in an era of seeking ways to restrict the application of expensive or risky diagnostic procedures when they are unlikely to yield useful results, neither the absence of a carotid bruit nor any result of applying a more complex predictive model can identify patients in whom it is appropriate to withhold angiography; too many patients with high-grade stenosis who could benefit from endarterectomy would be denied the definitive diagnostic test. Given that Haynes and colleagues [14] recently reported similarly low agreement between Doppler ultrasonography and carotid angiography, it appears that surgically fit patients presenting with anterior circulation transient ischemic attacks or minor strokes should be considered for cerebral angiography. Meanwhile, the search continues for accurate, noninvasive approaches for determining who should have surgery.