The prognostic value of repeated measurements may depend on the initial risk for stroke, the effect of stenosis progression on risk [2, 7, 12, 13, 18, 37, 38], the rate of progression [7, 13, 18, 38], the accuracy and frequency of the measurements, and the presence of other risk factors. Cerebrovascular event rates declined over time in two asymptomatic study groups [39, 40], but other studies have suggested that the risk for these events increases with time [8, 14]. Exploring the role of stenosis progression may improve our understanding of the natural history of asymptomatic carotid disease and the prognostic utility of ultrasonography of the carotid arteries. If repeated measurements of carotid stenosis are useful for predicting cerebrovascular events, the following hypotheses should be true: 1) An updated measurement of stenosis is a more powerful predictor of outcome than is the initial measurement, 2) information about stenosis progression improves outcome prediction that is based on the initial measurement of stenosis, and 3) the predictive ability of the initial stenosis measurement decreases over time.

To test these hypotheses, we did a secondary analysis of data from the Asymptomatic Cervical Bruit Study [41, 42], which is one of the largest studies to date on the natural history of asymptomatic carotid stenosis. We focused on the ability of initial and repeated ultrasonography of the carotid arteries to predict the occurrence of TIA, stroke, and death.

Methods

Top Methods Results Discussion Author & Article Info References

Study Design

The multicenter Asymptomatic Cervical Bruit Study [41] combined a prospective natural history study of asymptomatic carotid stenosis with a placebo-controlled, double-blind, randomized trial [42] to evaluate aspirin for the prevention of vascular events in persons with carotid stenosis greater than 50%. Accrual of participants lasted from 1988 to 1991 for the natural history study (follow-up continued until 1994) and from 1988 to 1994 for the randomized trial; the study was stopped in 1994 after interim analysis indicated that aspirin had no effect [42]. Carotid stenosis was assessed by duplex ultrasonography at 6-month intervals. Details of the study design are published elsewhere [41].

Study Sample

Persons with cervical bruits, referred by general practitioners and specialists in the community and in teaching hospitals, were eligible. Exclusion criteria were previous symptomatic cerebrovascular disease or carotid endarterectomy (or both); valvular heart disease except for mitral valve prolapse; myocardial infarction in the previous 3 months; atrial fibrillation; use of anticoagulant agents or aspirin or other antiplatelet agents; nonarterial cervical bruits; intolerance to aspirin; or a life expectancy less than 5 years. Persons with baseline carotid stenosis of 50% or more received placebo or aspirin (325 mg/d). Persons with carotid stenosis less than 50% entered the natural history arm of the study and were eligible for the aspirin-placebo trial if stenosis progressed to 50% or more. For this analysis, all participants were included because removing those who received aspirin did not alter the results.

Measurement of Stenosis

At the initial evaluation and at each follow-up point, duplex ultrasonography was done on the right and left common and internal carotid arteries. Six categories of carotid stenosis (expressed as the percentage decrease in artery diameter) were defined using Doppler spectral features described elsewhere [18]: 0% (normal), 1% to 15%, 16% to 49%, 50% to 79%, 80% to 99%, and 100% (complete occlusion). Participants were classified according to the highest degree of stenosis identified (maximum stenosis) [18, 41]. For nine participants who did not have stenosis measured at enrollment, baseline stenosis was defined as the first available follow-up measurement. During the study, ultrasonographic measurements were validated against angiography in a sample of 257 patients from all participating centers; excellent overall agreement was seen ( = 0.75 [95% CI, 0.64 to 0.96]). Because other studies suggested that risk for ischemic cerebrovascular events increased with stenosis of 50% or more or 80% or more [3, 5, 6, 14], analyses performed with carotid stenosis divided into six categories were repeated with baseline stenosis dichotomized at 50% and 80%. Measurements of stenosis updated at each follow-up point were represented as time-dependent variables [43].

Progression of maximum stenosis was determined from updated duplex ultrasonographic readings and defined in three ways: 1) as any change to a higher category, 2) as progression to stenosis of 50% or more, and 3) as progression to stenosis of 80% or more. Progression was represented as a time-dependent dichotomous (yes/no) variable. Regression of carotid plaque was classified as "no progression."

Other Risk Factors

Baseline evaluation included a complete history and physical examination done by a neurologist and a laboratory profile [41]. Information on history of smoking and peripheral vascular disease (intermittent claudication) were obtained by using questionnaires. Hypertension, hyperlipidemia, and diabetes were defined by history or a baseline blood pressure of at least 160/95 mm Hg, a cholesterol level of 6.5 mmol/L (250 mg/dL) or more, or a blood glucose level of 8.2 mmol/L (150 mg/dL) or more, respectively [42]. Heart disease was defined by history or by evidence of old or recent myocardial infarction on the baseline electrocardiogram. Use of aspirin was defined by enrollment in the aspirin arm of the randomized trial.

Clinical End Points

The clinical end points recorded were TIAs, ischemic and hemorrhagic strokes, myocardial infarction, unstable angina, vascular deaths, and other deaths. Follow-up continued after nonfatal events occurred [41]. End points were ascertained during follow-up visits or telephone interviews and confirmed through review of hospital records, death certificates, and autopsy reports by a blinded adjudication committee (in cases of disagreement, these records were reviewed by external adjudicators). Transient ischemic attack, defined according to specific criteria, included amaurosis fugax [44]. Stroke severity was assessed by using the Barthel index [45]. In this analysis, we used the outcomes of 1) stroke, 2) TIA or stroke, and 3) stroke or death. Because of the small number of strokes that occurred, findings are reported primarily for the aggregate outcomes.

Statistical Analysis

Methods for censored survival data were used with survival time defined as the time from enrollment to the first event. Participants were censored at the time of carotid endarterectomy, unstable angina, myocardial infarction, loss to follow-up, or study closure. All hypotheses were tested at the 0.05 significance level. Univariate analyses rely on the Kaplan-Meier method and the log-rank test, with the study cohort stratified according to initial stenosis category.

The association between stenosis and outcome is explored with four variations of the multivariable Cox proportional-hazards model [43] that adjust for the same a priori selected baseline risk factors: age, sex, heart disease, hypertension, hyperlipidemia, diabetes, claudication, smoking, and use of aspirin. The predictive ability of stenosis is expressed by its relative risk (RR) (adjusted hazard ratio) [46]. To verify whether an updated measurement of stenosis is a more powerful predictor of outcome than the initial measurement, we compared one Cox model containing baseline stenosis with another containing updated measurement of stenosis, represented by a time-dependent covariate.

Furthermore, we differentiated between participants with severe baseline stenosis who remained ultrasonographically stable and participants who had progression. Thus, a third model includes two separate stenosis-related variables: baseline stenosis and a time-dependent covariate indicating whether, at a given point in time, progression had occurred. If information about stenosis progression is useful for predicting outcome, progression should be a significant independent risk factor when adjusted for baseline stenosis.

To determine whether the predictive ability of initial stenosis decreases over time, we used a flexible generalization of the Cox model to estimate relative risk [47, 48]. Unlike the proportional hazards model, which assumes that relative risks remain constant, the flexible model permits estimation of changes over time. Relative risks are plotted as a function of follow-up time, and the significance of changes is assessed with a likelihood ratio test, comparing how the flexible and constant models fit the data [46, 47]. The risk factor is considered a significant predictor of outcome if the 95% CI for the relative risk function excludes 1.0 [47].

Multivariable analyses were repeated for the subset of participants with mild to moderate (1% to 79%) baseline stenosis.

Approximate sensitivity, specificity, and positive and negative predictive values for progression as a predictor of TIA or stroke within 2 years were estimated for a hypothetical population with distribution of risk factors, rate of progression, relative risk for progression, and annual event rate corresponding to those in our sample. The proportions of progressors and nonprogressors expected to have an event within 2 years were estimated, assuming a parametric proportional hazards model with exponentially distributed time-to-event.

Descriptive statistics and univariate survival analyses were done using SAS software (SAS Institute, Cary, North Carolina). We used BMDP Module 2L (BMDP Statistical Software, Inc., University of California Press, Berkeley, California) for multivariable Cox models. Software for flexible time-varying analyses is written in the C programming language [49].

Results

Top Methods Results Discussion Author & Article Info References

The Asymptomatic Cervical Bruit Study enrolled 715 participants (60.4% were women) 23 to 91 years of age (mean age, 65.2 years; median age, 66 years). The mean duration of follow-up was 3.2 years (3.4 years for participants with baseline stenosis of 1% to 79%); follow-up ranged from 1 day to 5.6 years (median follow-up, 3.5 years). Eleven participants (1.5%) were lost to follow-up, and 8 (1.1%) had carotid endarterectomy. The mean time from enrollment to the last ultrasonographic examination was 2.2 years (median, 2.1 years), and the mean time from the last ultrasonographic examination to an event or the end of follow-up was 0.98 years (median, 0.5 years). The mean number of ultrasonographic examinations per patient was 5.1 (range, 1 to 11; median, 4.4).

Clinical Characteristics and Carotid Stenosis

Table 1 shows the distribution of risk factors and carotid stenosis at study enrollment. One participant who did not have ultrasonography was excluded. Half of the participants had initial stenosis of 50% or more, and 21% had initial stenosis of 80% or more. Advanced stenosis is over-represented in our sample because accrual of participants (with stenosis 50%) for the aspirin trial continued for 2 years longer than did accrual for the natural history study [42].

Progression of Carotid Stenosis

Table 2 shows the proportion of participants with stenosis progression. Overall, 21.3% of participants had progression of stenosis and 9.1% had regression of stenosis. A total of 8.9% of participants with baseline stenosis less than 80% had progression to stenosis of 80% or more over an average of 2.4 years of ultrasonographic follow-up.

Risk for Cerebrovascular Events and Death

Table 3 shows the occurrence of primary events in our sample; 523 participants (73%) had no event. The annualized event rate for TIA or stroke (not preceded by another event) was 4.3%. No fatal strokes or cases of intracerebral hemorrhage occurred. Of a total of 38 strokes, 15 (39.5%) occurred in participants whose updated measurement of stenosis was 80% or more. Three of the 4 disabling strokes (Barthel score < 70) that were seen occurred in participants with stenosis less than 80%, and 1 occurred in a patient who had progression from the 50% to 79% stenosis category (at enrollment) to the 80% to 99% stenosis category (before the event). Mortality included 24 vascular and 18 nonvascular deaths.

Association between Initial Stenosis and Outcome

Figure 1 shows the evolution of absolute risks for each baseline stenosis category. In this univariate survival analysis, more severe stenosis was associated with a significantly increased risk for TIA or stroke (P < 0.001) (Figure 1), stroke alone (P = 0.0356), and stroke or death (P < 0.001).

View larger version (18K): [in this window] [in a new window]   Figure 1. Probability of remaining free of transient ischemic attack (TIA) or stroke. Univariate Kaplan-Meier estimates of the probability of remaining free of TIA or stroke for each of six categories of initial maximum stenosis are shown. Numbers below the x-axis are numbers of patients at risk in each category (category A = no occlusion; category B = stenosis of 1% to 15%; category C = stenosis of 16% to 49%; category D = stenosis of 50% to 79%; category E = stenosis of 80% to 99%; and category F = complete occlusion) in years 0, 1, 3, and 5.

Predictive Ability of Initial and Updated Measurements of Stenosis

Table 4 compares, in multivariable analyses adjusting for all baseline risk factors, the ability of baseline and updated measurements of stenosis to predict the occurrence of TIA or stroke. Initial and updated measurements of stenosis (whether expressed as a linear trend, stenosis 50%, or stenosis 80%) were all significant predictors of TIA or stroke and of stroke or death (P < 0.003 in all cases). Baseline stenosis of any degree was not a significant predictor of stroke alone (RR, 1.2 [CI, 0.9 to 1.6]), but an updated stenosis measurement of 80% or more was a significant predictor (RR, 2.5 [CI, 1.2 to 5.1]). Heart disease at enrollment significantly increased the risk for all outcomes and tripled the risk for stroke (RR, 3.2 [CI, 1.6 to 6.5]). Hypertension carried a relative risk of 2.4 (CI, 1.2 to 4.8) for stroke and 1.6 (CI, 1.05 to 2.4) for TIA or stroke. The adjusted relative risk estimates for stenosis, heart disease, and hypertension were very similar in the baseline and time-dependent models. The same pattern of findings was seen for stroke or death and for participants with baseline stenosis of 1% to 79% (data not shown).

Predictive Ability of Stenosis Progression

The last column of Table 4 shows the relative risks for TIA or stroke associated with stenosis progression when adjusted for baseline stenosis. The relative risks associated with progression to a higher category were slightly greater for stroke (RR, 2.2 [CI, 1.02 to 4.9]) and stroke or death (RR, 2.0 [CI, 1.1 to 3.5]) than for TIA or stroke. Regardless of the participants or events considered, progression to stenosis of 50% or more was not statistically significant (P > 0.2 in each case). In contrast, for the sample as a whole, having progression to stenosis of 80% or more increased the risk for stroke (RR, 4.2 [CI, 1.4 to 12.9]), TIA or stroke (RR, 3.0 [CI, 1.3 to 6.7]), and stroke or death (RR, 3.4 [CI, 1.5 to 7.8]) compared with having the same baseline stenosis but no such progression. In all cases, the relative risk estimates and the significance of initial stenosis remained substantially unchanged when progression was added to the model. Progression analyses for the subgroup of patients with baseline stenosis less than 80% yielded relative risks of similar magnitude but of varying statistical significance.

Table 5 summarizes the estimated approximate screening test characteristics, derived from our data, of using progression for the prediction of TIA or stroke within 2 years. Regardless of the definition of progression used, the sensitivity and positive predictive value are very low.

Does the Predictive Ability of the Initial Measurement Decrease over Time?

Figure 2 shows, for all participants, the estimated relative risk over time for TIA or stroke associated with the initial measurement of stenosis. The plot and CIs from the flexible model suggest that baseline stenosis remains a significant independent predictor during the first 4 years of follow-up, with an increase of approximately 50% in risk associated with each successive category of stenosis. There is no evidence of a significant decrease in the predictive ability of the baseline measurement of stenosis (P > 0.2) [42, 46-48]. Similar patterns were seen for stroke or death and for participants with baseline stenosis of 1% to 79% (data not shown).

View larger version (10K): [in this window] [in a new window]   Figure 2. Predictive ability of the initial stenosis category over time. The solid curve (hazard ratio) is the estimated relative risk over time, adjusted for baseline risk factors (age, sex, heart disease, hypertension, hyperlipidemia, diabetes, claudication, smoking, and use of aspirin), for transient ischemic attack (TIA) or stroke associated with an increase in baseline stenosis by one category compared with other patients. Dotted curves indicate the 95% Cls for the relative risk. The dashed horizontal line represents a reiative risk of 1.0, corresponding to equal risk in all groups. Areas below this line represent risk reduction; areas above this line represent increases in risk.

Discussion

Top Methods Results Discussion Author & Article Info References

We sought to determine whether repeated measurements of carotid stenosis done using duplex ultrasonography would improve our ability to predict the occurrence of ischemic cerebrovascular events. Our analysis explicitly investigated the role of stenosis progression in a multivariable model with other risk factors. Our results are based on 3664 duplex ultrasonographic examinations done in 714 asymptomatic patients over 2285 patient-years; this is one of the largest available data sets of this type. This study group had low rates of loss to follow-up and endarterectomy and an annualized stroke rate of 1.6%, which is similar to the rate reported by others [2, 3, 18, 38]. Although we found evidence of increasing risk for TIA, stroke, and death with progression to high-grade carotid stenosis, our data overall do not support the use of frequent ultrasonographic examinations in asymptomatic patients for the prevention of disability and death due to stroke.

The clinical usefulness of serial ultrasonography in asymptomatic carotid disease may be limited by the low incidence of stroke, inadequate understanding of plaque progression and structure, technical issues, and the absence of an effective intervention for the prevention of stroke. In studies of asymptomatic carotid disease, the most clinically relevant outcome is the occurrence of unheralded stroke [26, 31, 50]. Even in our high-risk sample, no stroke-related deaths occurred and most strokes, occurring in the absence of severe stenosis, could not have been predicted by ultrasonography. We could not confirm stenosis to be a significant independent risk factor for stroke alone. In contrast, heart disease was a highly significant predictor of stroke.

Several natural history studies [2, 6, 7, 18, 37, 38] have found an association between stenosis progression and risk for cerebrovascular events, but these studies did not consider other clinical risk factors. In our analyses that controlled for common clinical conditions, risk was not increased by progression to stenosis of 50% or more, but progression to stenosis of 80% or more was associated with a significantly higher risk for stroke, TIA or stroke, and stroke or death. However, even with an average of 2.4 years between the first and the last ultrasonographic examinations, only 8.9% of participants who had had stenosis less than 80% at enrollment had progression to stenosis of 80% or more. Although 46 participants who had had stenosis of 50% to 79% had progression to stenosis of 80% or more, this progression preceded an unheralded disabling stroke in only one case. The baseline stenosis measurement retained its predictive ability over an average clinical follow-up period of more than 3 years, and, in the subgroup of 500 patients with mild to moderate stenosis at enrollment, we could not confirm progression to be a significant predictor of cerebrovascular events. Finally, the low sensitivities and positive predictive values for progression, which reflect the low rates of progression and of cerebrovascular events, respectively, further illustrate the poor performance of serial ultrasonography for routine follow-up in asymptomatic patients.

Because progression to stenosis of 80% or more appears to increase risk for ischemic cerebrovascular events, the diagnostic yield of serial ultrasonography could improve if this procedure is used in patients who have a particularly high risk for progression and subsequent disabling stroke. Smoking, coronary artery disease, diabetes, peripheral vascular disease, hypertension, age, and increased levels of low-density lipoprotein cholesterol and fibrinogen have been implicated as risk factors for stenosis progression [18, 37, 51-53], whereas treatment of hypertension may slow progression [51].

We focused on the prognostic usefulness of repeated ultrasonographic measurements in clinical practice rather than on the underlying biology of progressive atherosclerosis. Each event that occurred was attributed to the most recent maximum stenosis measurement; the actual degree of stenosis just before an event is not known, a situation that reflects clinical reality. If unheralded strokes are precipitated by the sudden narrowing of carotid arteries [7], current knowledge does not allow us to predict when this will occur. Moreover, although specific features of plaque structure and events such as rupture, hemorrhage, and ulceration may be important determinants of TIA and stroke (much like the mechanisms associated with coronary occlusion [54, 55]), these findings are unlikely to be documented by intermittent ultrasonographic examinations. Research in this area is limited [22, 56, 57].

Technical considerations also limit the usefulness of duplex ultrasonography for predicting stroke. One concern is the lack of a universally accepted classification of carotid stenosis. Fourteen observational studies of asymptomatic carotid disease [1-710, 14, 18, 37, 38, 40, 58] all used different classification categories. Furthermore, although duplex ultrasonography has been validated against angiography in research centers [9, 15-20], repeated measurements amplify the risk for error. If measurement error is considerable in clinical practice, the predictive value of repeated measurement may be compromised, even if true progression is important. Further research on the role and causes of stenosis progression would benefit from standard definitions for carotid stenosis, stenosis progression, and stenosis regression and from appropriate choice of outcomes, including the severity and type of stroke [59].

The limitations of our study include the small number of strokes that occurred; our results for stroke alone should be considered approximate. As have others [2, 3, 5, 6, 14, 18, 24, 27, 28, 40], we used aggregate outcomes (such as "TIA or stroke") to increase statistical power and account for competing risks that may preclude occurrence of a stroke ("stroke or death") [60]. Our stenosis classification system, which groups moderate stenoses of 50% to 79% together, precludes further analysis within this range and limits comparison of findings with those of more recent studies [28]. Although early cessation of the randomized aspirin-placebo trial limited follow-up for some participants, our methods accounted for the variation in duration of follow-up. Nonetheless, we cannot comment on the usefulness of repeated ultrasonographic measurements over periods longer than 3 to 4 years. Analysis of whether strokes were ipsilateral or contralateral to the maximum stenosis lesion is addressed elsewhere [61].

Frequent measurements of carotid stenosis entail substantial costs for the health care system [34, 62] and, in the era of advancing technology and shrinking resources, the question of the prognostic utility of repeated measurements in chronic disease is likely to receive increasing attention. Our results suggest that the diagnostic yield of carotid ultrasonography is too low to warrant frequent testing of patients with asymptomatic carotid disease, particularly as long as the benefit of endarterectomy for stroke prevention in these patients remains uncertain [24-36]. For participants at moderate to high risk for TIA and stroke, patient education, attention to treatable risk factors, and close clinical follow-up to ensure detection of the earliest symptoms of cerebrovascular ischemia will enable the clinician to offer appropriate intervention in a timely fashion.

Drs. Abrahamowicz and Battista: Division of Clinical Epidemiology, The Montreal General Hospital, 1650 Cedar Avenue, Montreal, Quebec H3G 1A4, Canada.

Dr. Cote: Department of Neurology, The Montreal General Hospital, 1650 Cedar Avenue, Montreal, Quebec H3G 1A4, Canada.