Assessment of the Efficacy of Noninvasive Screening for Patients With Asymptomatic Neck Bruits
Background and Purpose Several recent clinical trials have shown that endarterectomy is efficacious in patients with asymptomatic carotid artery stenosis. The purpose of this study was to evaluate the effectiveness of various test strategies for screening and diagnosing carotid artery disease.
Methods We constructed a model of the natural history of carotid artery disease using literature-based estimates of the prevalence and incidence of carotid artery stenosis and associated morbidity and mortality. Markov cohort simulation was used to estimate the mean quality-adjusted life years and monetary costs associated with various management strategies.
Results Screening is cost-effective in the baseline model. Key parameters affecting the efficacy of screening are prevalence of operable lesions, benefit of surgery, surgical complication rates, quality of life with stroke, rate of stenosis progression, and excess morbidity and mortality.
Conclusions Asymptomatic patients with carotid bruits may benefit from screening if the prevalence rate is ≥20%, the benefits and risks associated with surgery are similar to those observed in the Asymptomatic Carotid Atherosclerosis Study, and the quality of life with stroke is considerably lower than the quality of life without stroke. Ultrasound followed by three-dimensional time-of-flight MR angiography, if indicated, is a promising test strategy.
The purpose of this study was to evaluate the effectiveness of screening for carotid artery disease, particularly among asymptomatic patients with carotid bruits. The question of screening arises in light of the results of three recent multicenter randomized clinical trials of asymptomatic patients with carotid artery stenosis.1 2 3 The first and smallest of the three studies (n=410) found no benefit to surgery in patients with 50% to 90% stenosis.1 However, patients with stenosis >90% at baseline were excluded in this trial, and nearly half of the patients in the medical arm underwent carotid endarterectomy during the course of the study. The two most recent studies (n=444 and n=1662) observed a respective 61% and 53% relative risk reduction in ipsilateral events attributable to surgery.2 3
The question of screening asymptomatic patients has been addressed previously. In 1988, Feussner and Matchar4 used decision analysis to assess the value of noninvasive screening for patients with asymptomatic neck bruits. Their study, however, was published before the completion of the clinical trials for asymptomatic patients1 2 3 and several similar trials for symptomatic patients.5 6 7 Thus, they were not able to use the most recent data on the risks and benefit of surgery.
Recently, Derdeyn and Powers8 built a decision model using data from recent clinical trials to assess the cost-effectiveness of screening with US, followed by catheter angiography if indicated. Similarly, we built a decision model to evaluate the cost-effectiveness of both US and 3D TOF MRA as screening tools for asymptomatic patients. We also assessed the cost-effectiveness of 3D TOF MRA as a presurgical test in lieu of CA.
We compared the costs and benefits of two management plans for asymptomatic patients with neck bruits: (1) noninvasive screening of the carotid arteries and (2) a wait-and-see approach, whereby only patients who become symptomatic are tested. The major advantage of screening over the wait-and-see approach is a reduction in the risk of unheralded strokes. The disadvantages of screening are its expense and the complications associated with early surgery.
For each of these two management plans, we compared the costs and benefits of four diagnostic/therapeutic test strategies. In the first test strategy, patients undergo duplex US, followed by CA if US detects ≥60% stenosis. In the second test strategy, patients undergo 3D TOF MRA followed by CA if MRA detects ≥60% stenosis. In these first two strategies, if CA identifies 60% to 99% stenosis with no clinically relevant tandem lesions, then the patient proceeds to surgery. In the third strategy, patients undergo 3D TOF MRA only. Finally, in the fourth strategy, patients undergo US, followed by 3D TOF MRA if the US detects ≥60% stenosis. In these last two strategies, if 3D TOF MRA detects 60% to 99% stenosis and does not detect any contraindications to surgery, then the patient proceeds to surgery.
Fig 1⇓ illustrates the decision model. The square decision node at the left indicates that a decision must be made as to screen or not screen. In the bottom branch, there is no testing unless a patient suffers a reversible cerebral event. In the top branch, patients undergo noninvasive screening. If the test result is “negative,” screening ends. If the test result is “positive,” then in test strategies 1 and 2 CA is performed. In strategy 3, the patients proceed directly to surgery. In strategy 4, 3D TOF MRA is performed.
We used Markov simulation9 to model the iterative risks of TIAs, strokes, and death over time. In this model a hypothetical cohort of asymptomatic patients moves from one state of health to another over time. Each cycle of the Markov model was 1 month in duration. There were three Markov states: well, (permanently) disabled, and dead. In addition, we allowed cycles of temporary disability after transient cerebral events (ie, TIAs or reversible strokes). Fig 2⇓ illustrates the allowed transitions between these states.
For each management plan (ie, screen and wait-and-see) and for each of the four diagnostic test strategies, we simulated an identical cohort of 20 000 patients. At the start of the simulation, all patients were asymptomatic. A program was written in Fortran to generate and track the patient cohorts over time until all patients reached death.
Assignment of Utilities
Our outcome measure was QALYs. QALYs describe the quality and quantity of life. Each Markov state was assigned a quality of life value (ie, utility). A utility value of 0 was given to the state of death, whereas a utility of 1 was given the state of well-being. Instead of assigning a single utility value to the disability states, we assigned a distribution of values to better reflect the diversity of stroke outcomes. Specifically, for each occurrence of a transient cerebral event or irreversible stroke, a random variable from a β distribution was generated. For irreversible strokes, the parameters of a β distribution were chosen such that the mean utility was 0.2, and 70% of the utilities were between 0.1 and 0.4. For transient cerebral events, the mean utility was 0.8, and 58% of the utilities were between 0.7 and 0.9. These mean utilities of 0.2 and 0.8 have been used previously to model “small strokes” and “large strokes,” respectively.10
To determine the duration of a transient cerebral event, we generated values from a log-normal distribution. The median duration of a transient event was 1 day, and 95% of durations were <2 weeks. Thus, 50% of reversible events fell under the definition of a TIA. This is consistent with most clinical definitions of TIAs and RINDs. A utility of 1 was assigned to cycles after a reversible event, unless another event occurred.
Baseline Estimates of Model Parameters
Table 1⇓ summarizes the baseline estimates of various parameters in the model. We define an operable lesion as one with 60% to 99% stenosis, as in the ACAS.3 In Table 2⇓ the annual risks of ipsilateral irreversible stroke and reversible events are summarized as functions of the amount of stenosis present and whether or not the patient has had previous symptoms. Whenever possible, we used estimates of risk based on the medical arm of recent clinical trials. When data were not available from clinical trials, we used other sources from the literature. (To estimate the risk of stroke for this analysis, it was necessary to use data from a variety of sources. Because of differences in patient mix and small sample sizes in some studies, our estimates of risk may be imprecise. In our sensitivity analysis, we tried to address this. Specifically, since the risk of unheralded, irreversible stroke is critical to the efficacy of screening, we assessed the effect of both lowering and raising the baseline estimates of this risk.) For patients who have just experienced a reversible cerebral event, we assume that the risk of an irreversible stroke in the following month is 4.4%.21 If a stroke does not occur in the following month, then the annual risk of an irreversible stroke is 13%6 and the annual risk of a reversible event is 14.5%.7 For patients who have just experienced an irreversible stroke, we assume that the risk of cerebrovascular death in the following month is 17%.22 If death does not occur, then the annual risk of another irreversible stroke is 14%.23
In ACAS,3 the ipsilateral stroke relative risk reduction attributable to surgery was estimated at 53% (absolute risk reduction of 5.9% over 5 years). This estimate includes surgery and angiography complications. For our assessment of QALYs, we must account for the timing of events; thus, we must keep separate accounts of the perioperative events and the events following an uncomplicated surgery. We estimate that for patients with uncomplicated surgeries, the relative risk reduction attributable to surgery is 73% for asymptomatic patients3 and 80% for symptomatic patients.6
The long-term benefit of surgery was not assessed in either the ACAS3 or NASCET.6 (In the ACAS, patients were followed for a median of 2.7 years; in the NASCET, the duration of follow-up averaged 1.5 years.) Bernstein et al,24 however, followed a consecutive series of 507 postoperative patients for up to 10 years. They found that the rate of restenosis decreases after the first postoperative year and that the risk of cerebrovascular events is inversely related to the amount of restenosis. These findings are supported by others.25 26 Thus, for our analysis, we assume that the relative risk reductions estimated in the ACAS and NASCET persist over the patient’s life.
Schedule of Screening Asymptomatic Patients
We first considered the scenario in which asymptomatic patients are screened only once. Under this first scenario, if the test result does not indicate that the patient has 60% to 99% stenosis, then the patient is not tested again unless he becomes symptomatic. We also considered the scenario in which asymptomatic patients whose test results show 30% to 59% stenosis return for annual screening up until the time the patient undergoes surgery or becomes symptomatic. For test strategies 1 and 4, US would be used for follow-up screening; for test strategies 2 and 3, MRA would be used.
Schedule of Testing Patients Once Symptomatic
Under both the wait-and-see and screening approaches to patient management, a subpopulation of the original cohort of asymptomatic patients will eventually become symptomatic. After each cerebrovascular event, the patient undergoes a workup under one of the four diagnostic test strategies. If a symptomatic patient is found to have <60% stenosis at workup, then we assume that the patient returns at regular intervals for follow-up testing. For symptomatic patients found to have 30% to 59% stenosis, follow-up testing is performed at 6 months and then annually unless another event occurs. For symptomatic patients found to have 1% to 29% stenosis, follow-up testing occurs annually. For test strategies 1 and 4, US would be used for follow-up testing; for test strategies 2 and 3, MRA would be used.
(1) Conventional angiography is 100% sensitive and specific for detecting an operable lesion and for detecting tandem lesions and aneurysms.
(2) A patient undergoes surgery only once.
(3) The combined frequency of tandem lesions and/or cerebral aneurysms is 5%.6 The accuracy of 3D MRA for detecting these lesions is as follows: sensitivity, 80%; specificity, 95%.27 28 For the baseline analysis, when surgery is performed on a patient with a tandem lesion or cerebral aneurysm, we assume that the perioperative risks are double those of a patient without a tandem lesion and that the patient obtains no benefit from surgery.29 30 31 32
(4) The frequency of patients with claustrophobia or metal devices that contraindicate MRA is 5%. When MRA is contraindicated, US is used for screening, followed by CA (ie, test strategy 1 is the default).
(5) The short-term morbidity associated with surgery is equivalent to a 1-week reduction in quality-adjusted survival.33
(6) Patients without a 60% to 99% stenosis who undergo surgery obtain no benefit from surgery.
(7) A 5% annual discount rate on utilities describes patients’ risk aversion.
(8) Twenty percent of patients with TIAs require hospital admission.
At our institution, we use Transition I (Transition Systems, Inc), a cost accounting software package modified for our use. This software allows us to approach an activity-based method of cost accounting. The software’s methodology uses a cost allocation model with two basic levels of cost accounting: the intermediate product level and the end-product or patient case level. Transition I integrates and applies both cost accounting methods through different subsystems. The strength of the system is that a manager can look both bottom up and top down in analyzing costs.
We used this cost accounting method to compute the mean direct professional and technical costs of diagnostic imaging and surgery (not billable charges) at our institution for 1995. The total direct cost of a duplex scan of the extracranial arteries was $181. The cost of MRA (head and neck) was $417. The cost of arch and carotid angiography was $1959. The cost of thromboendarterectomy of the carotid artery was $6991. In addition, at many institutions including ours, symptomatic patients receive CT of the brain, which costs $205.
The cost of a TIA and stroke was obtained by reviewing total cost data on patients hospitalized at our institution between January 1, 1995, and December 31, 1995. These were internal medicine or neurology admissions with diagnosis-related groups of 14 or 15. From these data, the mean cost of a TIA that required hospital admission, excluding diagnostic imaging and endarterectomy, was $16 800. The mean cost of a stroke, excluding diagnostic imaging and carotid endarterectomy, was $17 500.
We weighted the costs that were incurred over time by a 5% annual discount rate to reduce all costs to present value.34
Because of the uncertainties in the data points used in this analysis, we performed a sensitivity analysis to study the robustness of our conclusions. We focused on the parameters most likely to affect our conclusions. Specifically, we evaluated the following scenarios:
(1) Surgery is less beneficial. We examined the case in which surgery is less beneficial than that assumed in the baseline analysis. First, we examined the effect on our results if surgery is associated with a relative risk reduction, excluding surgical complications, as low as 30%. Note that a relative risk reduction, excluding surgical complications, of 50% is approximately equivalent to the lower end of the confidence interval for relative risk reduction, including surgical complications, in the ACAS.3 Second, we considered the scenario in which the benefits of surgery did not last the patient’s lifetime but rather persisted for shorter durations of 5 to 25 years.
(2) Higher surgical risks. We considered the effect of a 50% and a 100% increase in the surgical risks over those experienced in recent clinical trials. Thus, for prophylactic surgery, the mortality rate was assessed at 0.5% and 0.6% and the irreversible stroke rate at 2.6% and 3.4%, respectively. For symptomatic surgery, the mortality rate was assessed at 0.9% and 1.2% and the irreversible stroke rate at 7.8% and 10.4%, respectively.
(3) Frequency of unheralded, irreversible stroke is lower/higher. We considered the case in which the frequency of unheralded, irreversible strokes is 50% lower and 50% higher than the baseline estimates in Table 2⇑.
(4) 3D TOF MRA is less accurate at detecting operable lesions. Since our estimates of the accuracy of 3D TOF MRA were based on only a few available studies26 (unlike the estimates of the accuracy of US, which were based on a meta-analysis25 of 70 studies), we thought it was important to consider a scenario in which the sensitivity and specificity of MRA are as much as 10% lower than that used in the baseline analysis. We also considered the scenario in which the baseline estimates of accuracy are biased as a result of verification bias. For this assessment, we reduced the sensitivity of 3D TOF MRA by 10% and increased the specificity by up to 10%.
(5) Monetary costs are higher. We first considered a twofold increase in the cost of the diagnostic tests (ie, US, 3D TOF MRA, and CA). Second, we considered a twofold increase in the cost of carotid endarterectomy. Finally, we considered the effect on our results if all costs were increased twofold and if we factored in the chronic cost of stroke care. For the latter analysis we assumed that the cost of stroke was $10 000 annually.
(6) Stroke-free quality of life value is lower/Quality of life with stroke is higher. In patients with a high prevalence of asymptomatic disease, and even in normal people older than 50 years, the utility value associated with a stroke-free status may be less than 1.0.35 Thus, we considered the effect on our results if the utility associated with stroke-free survival was 0.90 instead of 1.0. We also considered the effect on our results if the utility value associated with stroke was greater than our baseline value of 0.2, as suggested in a study by Gage et al.36 For this analysis we set the mean utility for irreversible stroke equal to 0.4, similar to the estimate of Gage et al of 0.39 for “moderate to severe stroke.”
(7) Stenosis progression rate is lower. In the baseline analysis, we assumed that 23.5% of patients would progress one stenosis category annually. In the first year, this translates into approximately 9% of patients who had <60% stenosis at baseline progressing to >60% stenosis in the following year. In our sensitivity analysis, we reduced this rate from 23.5% to as low as 1%.
(8) Risk of “other” stroke and nonstroke mortality is higher. Patients with multiple risk factors for carotid artery disease are likely to have elevated “other” stroke (ie, contralateral or posterior circulation) and nonstroke mortality rates and to experience greater surgical risks. Thus, we considered the efficacy of screening patients with a high prevalence rate of carotid artery stenosis (ie, 30%) when the surgical risks and the risk of contralateral stroke and nonstroke mortality rates are as much as twice those in ACAS.
Comparison of the Cost-effectiveness of Different Strategies
Our strategy for comparing the two management plans was as follows. We first compared the cost-effectiveness of the diagnostic test strategies for the wait-and-see approach. We identified the best test strategy by first looking for simple dominance, whereby one test strategy is both more effective (in terms of QALYs) and less costly (in terms of dollars) than the alternatives. If the best test strategy could not be identified by simple dominance, then we identified the least expensive strategy (referred to as “A”) and compared it with the next least expensive strategy that was more effective (referred to as “B”). We computed the CER, defined as the additional cost of using strategy B instead of A per QALY gained by using B instead of A. If the marginal CER is <$50 000/QALY, which is a standard criterion used in decision analysis, then strategy B is preferred over A. Once we identified the best test strategy for the wait-and-see approach, we compared it with each of the various screening approaches. In this way, we assessed whether or not screening is cost-effective compared with the best wait-and-see approach.
The risk profile of the simulated patient cohorts at 6 months, 1 year, and 5, 10, and 20 years is summarized in Fig 3⇓. For this figure, the test strategy for the wait-and-see management plan is US followed by 3D TOF MRA (strategy 4). For the screening management plan, patients are screened with 3D TOF MRA (strategy 3); for the subcohort of patients who become symptomatic, test strategy 4 is used. The prevalence rate is 10%.
Initially, all patients are well. However, by 1 year 6% of screened patients and 7% of wait-and-see patients have moved to permanent disability or dead states. By 10 years, 60% of screened and 62% of wait-and-see patients have moved to permanent disability or dead states.
Comparison of Test Strategies for Wait-and-See Approach: Baseline Assessment
Table 3⇓ summarizes the QALYs and costs associated with the different test strategies for the wait-and-see management plan using the baseline estimates of risk. The mean number of QALYs is approximately 7 years. The mean cost per patient is between $10 000 and $13 000, depending on the prevalence of operable lesions and the test strategy used. Note that when the test strategies are compared, differences in mean QALYs of <0.08 year (≈1 month) are not statistically meaningful. Similarly, differences in mean cost of <$300 are not statistically meaningful. These minimally detectable differences are a function of the inherent variability among patients and the number of patients simulated.
Strategy 4 (US then 3D TOF MRA) is the preferred strategy for testing symptomatic patients, regardless of the prevalence rate. If the prevalence rate is 20%, it is the least expensive test strategy, and the marginal CERs of the other strategies are >$50 000. If the prevalence rate is 10% or 30%, strategy 4 is the second least expensive strategy, and its marginal CERs compared with the least expensive strategy are far below the $50 000 threshold (ie, $9250 and $7070 for prevalence rates of 10% and 30%, respectively). Furthermore, at many institutions including ours, it is much easier logistically to get a US on a symptomatic patient in the emergency department than it is to get an MRA study. When strategies 1 (US then CA) and 4 (US then 3D TOF MRA) are compared, as in Table 3⇑, strategy 4 is preferred to strategy 1.
Comparison of Management Plans for Asymptomatic Patients: Baseline Assessment
Table 4⇓ summarizes the mean QALYs and costs for various management plans for asymptomatic patients. At each prevalence rate we consider nine management strategies. The first possibility is the wait-and-see approach in which no screening is performed, but once a patient becomes symptomatic, then the strategy of US followed by 3D TOF MRA (strategy 4) is used to test and follow the patient, when applicable. In the second approach, patients are screened one time with US then CA, if needed. Later, if patients become symptomatic, strategy 4 is used to test and follow the patient, if applicable. In the third approach, patients are screened initially with US then CA, if needed. If the screening test indicates 30% to 59% stenosis, screening occurs annually with US unless the patient undergoes surgery or becomes symptomatic. If the patient becomes symptomatic, strategy 4 is used to test and follow the patient, if applicable. The other six screening approaches are similar but with different tests and test sequences.
If the prevalence rate is 10%, then only two screening strategies offer a significant gain over waiting: multiple screens with MRA followed by CA if needed (strategy 2) and multiple screens with MRA (strategy 3). With both of these screening plans, if the patient becomes symptomatic, then test strategy 4 is used. Since these two screening plans offer similar outcomes, strategy 3 is preferred since it is less expensive. The CER compared with no screening is $11 489. As a reference, the CER of a coronary artery bypass graft is $7300; the CER of treating mild hypertension is $32 600; and the CER of breast cancer screening in women aged 50 to 65 years is $20 000 to $50 000.37
If the prevalence rate is 20% or 30%, all of the screening strategies offer a gain of ≥1 month over no screening (between 33 and 77 days, on average). However, none of the screening strategies is significantly superior to the others in terms of gains in QALYs, and therefore the least expensive strategy is preferred. The least expensive strategies involve screening with US, then 3D TOF MRA if needed (strategy 4). Annual screening with US does not significantly increase the cost, and there may be a slight gain in QALYs, particularly if the prevalence rate is high. The CERs for multiple screens with strategy 4 are $4960 and $3128 for prevalence rates of 20% and 30%, respectively.
Fig 4⇓ depicts the (irreversible) stroke-free survival curves of the simulated cohorts in the wait-and-see (dashed curve), single screen (solid curve), and multiple screen (dotted curve) approaches. Here, the prevalence rate is set at 20%; strategy 4 is used to screen asymptomatic patients and to diagnose patients if they become symptomatic. The stroke-free survival is slightly greater with multiple screens than a single screen; both screening approaches offer greater stroke-free survival than no screening.
Table 5⇓ summarizes the results of the sensitivity analysis. The effectiveness of screening was sensitive to changes in several key parameters: benefit of surgery, surgical complication rates, quality of life with stroke, rate of stenosis progression, and excess morbidity and mortality rates.
The role of decision analysis in medical research is to provide insight into clinical problems. Decision analysis is not a substitute for a randomized controlled study but rather helps us to identify the key parameters in the decision. The results of this study suggest that the effectiveness of screening asymptomatic patients for carotid artery disease depends on the prevalence of disease, the benefit of surgery, the surgical complication rates, the quality of life with stroke, the rate of stenosis progression, and the excess morbidity and mortality rates.
If the relative risk reduction attributable to surgery is at the lower end of the confidence interval for risk reduction reported in the ACAS,3 then screening provides only marginal benefits at a 20% prevalence rate. At lower prevalence rates, screening is not effective. When the prevalence rate is low, the duration of benefits is particularly critical. Studies suggest that the benefits of surgery persist for ≥10 to 15 years.24 25 26 However, if the prevalence rate is <20%, the benefits may need to persist for ≥20 years for screening to be effective.
The clinical centers participating in ACAS and NASCET were selected for their low surgical complication rates. Mayberg and Winn38 warn that the surgical complication rates attained in ACAS may not be realized at other institutions. Our overall findings were not affected by modest increases in the surgical complication rates attained in ACAS and NASCET. However, when we doubled the rates (ie, stroke rates of 3.4% and 10.4% and mortality rates of 0.6% and 1.2% for asymptomatic and symptomatic patients, respectively), we found that screening is not effective if the prevalence rate is ≤10%.
The quality of life with stroke is another important parameter in the effectiveness of screening. If patients without stroke have a quality of life value of 0.9 and if stroke reduces that to 0.2, then screening is effective, according to our model. However, if stroke reduces the quality of life value only to 0.4, then screening may not provide any benefit over no screening unless the prevalence rate is ≥30%.
In our baseline model we considered a rapid rate of stenosis progression, equivalent to 9% of patients progressing from insignificant to significant disease annually. We found that screening is effective if this rate is ≥6%. However, if the rate is below 6%, screening is effective only if the prevalence rate is ≥20%; if the rate of progression is <1%, screening is effective only if the prevalence of disease is ≥30%.
Several risk factors for an increased prevalence of carotid artery stenosis, in addition to carotid bruits, have been noted in the literature and include peripheral vascular disease,39 40 smoking,40 41 42 43 44 hypertension,44 45 46 47 48 age,42 43 47 48 cholesterol/HDL,42 43 atrial fibrillation,49 diabetes mellitus,43 44 heart disease,40 44 and hyperhomocysteinemia.50 Furthermore, Barnett et al51 point out that NASCET patients with more than six of the following risk factors are at greater risk of stroke: male, aged >70 years, history of smoking, hypertension, myocardial infarction, congestive heart failure, diabetes, intermittent claudication, or high blood lipid levels. This might suggest that patients with multiple risk factors should be targeted for screening. However, our results are particularly sensitive to modest increases in the excess morbidity and mortality rates over those observed in ACAS.3 Yet, the ACAS patients had significant comorbidity: 64% were hypertensive, 25% were diabetic, and 69% had heart disease; in addition, 66% of patients were male, 36% were aged ≥70 years, and 28% were current smokers. More work is needed to understand the role of screening in patients with multiple risk factors in whom increased prevalence of operable lesions is counterbalanced by increased morbidity and mortality from other causes.
In our study we compared the efficacy of several diagnostic test strategies. We found that for both symptomatic and asymptomatic patients, 3D TOF MRA is more cost-effective as a presurgical tool than CA because of its reported high accuracy15 and ability to detect tandem lesions.27 28 The estimates we used for the accuracy of 3D TOF MRA were based on a few select institutions. A 5% reduction in our baseline estimates of accuracy did not change our conclusions; however, if the accuracy of 3D TOF MRA is 10% worse than reported, then the advantage of 3D TOF MRA over CA is negated.
Several other studies have been performed to investigate the effectiveness of screening for carotid artery disease. In a study by Feussner and Matchar,4 screening provided no benefit unless the prevalence of operable lesions was ≥35% and the risk of stroke from an untreated lesion was >3% annually. Our model differs from that of Feussner and Matchar in several key ways. First, the surgical complication rates in recent clinical trials were much lower than considered by Feussner and Matchar. Second, Feussner and Matchar did not consider the potential role of 3D TOF MRA as a presurgical tool. In our study, 3D TOF MRA was found to be more cost-effective than invasive angiography, which was responsible for a 0.2% mortality and 1.0% stroke rate in ACAS.3 Finally, Feussner and Matchar did not consider the dynamic nature of carotid atherosclerosis. We modeled the natural history of carotid stenosis progression and considered the efficacy of follow-up screenings.
In 1995, Derdeyn et al52 reported the results of their analysis of the efficacy of US screening for asymptomatic and symptomatic patients. Their study concluded that screening asymptomatic patients with US is efficacious. However, unlike our analysis and that of Feussner and Matchar,4 in which QALYs were computed, Derdeyn at al used the number of strokes prevented as their outcome unit. Derdeyn et al did not consider the effect of competing (noncerebrovascular) risks, the timing of events (ie, strokes occurring at 5 years were given weight equivalent to strokes occurring at surgery), or the monetary cost of screening.
Derdeyn and Powers8 recently published the results of a new study of the efficacy of US screening. In their new study, they compute QALYs and consider competing risks, the timing of events, and the monetary costs associated with testing and treatment. Like our model, they built in stenosis progression and consider follow-up screens. They conclude that a single screen may be cost-effective (CER of $35 130) but that annual screening is not cost-effective and may be detrimental. Their model differs from ours in several ways. Perhaps the most important difference is the control, or no-screen, cohort. In our study we compared the QALYs and costs of screening with the QALYs and costs of a wait-and-see approach in which patients initially are asymptomatic, but over time a subcohort becomes symptomatic and then are tested and may undergo surgery. We assumed that the risks and benefits of surgery for symptomatic patients are similar to the findings in NASCET.6 In our study we first investigated the best test strategy for the wait-and-see approach so that our screening cohort could be compared with the best outcomes achievable in the wait-and-see approach. In contrast, the control cohort in the study of Derdeyn and Powers is ill-defined. They state only that the costs and QALYs associated with the screening cohort are compared with “a natural history simulation.” It is unclear to us if this “natural history” control cohort is the relevant group for comparison with screening. Other differences between the model of Derdeyn and Powers and ours are as follows: (1) They did not consider 3D TOF MRA as a presurgical tool. Again, we found that 3D TOF MRA is more cost-effective than CA as a presurgical tool. (2) Some of their “costs” were based on reimbursement fees, which likely include not only costs but also profit building (ie, these are not costs, but rather charges). We used direct professional and technical costs. (3) Their strategy for follow-up screening required that all patients be followed annually. In our model, we followed only those patients who were found to have 30% to 59% stenosis on their first screen; patients with <30% stenosis on their first screen, which accounted for 62% of patients, were not followed. Finally, (4) they considered a very low rate of stenosis progression (0.5% annually). We also considered a rate that low, but in our baseline model the rate of progression from insignificant to significant disease was 9% annually.11
Several recent studies have compared different diagnostic test strategies in terms of accuracy, quality-adjusted survival, and monetary costs. Blakely et al14 performed a meta-analysis of the accuracy of US and MRA for detecting carotid artery disease. They found that duplex US and MRA have similar levels of accuracy. Unfortunately, these authors grouped together accuracy studies of 2D and 3D MRA. In our review of the MRA data, 3D MRA is much more specific than 2D MRA. Similarly, in a recent review of the MRA data, Bowen15 noted greater specificity with 3D MRA, necessitating separate evaluations of their accuracy.
Two studies have just been completed on the cost-effectiveness of various test strategies for symptomatic patients.53 54 In the study by Kent et al,53 the strategy offering the greatest quality-adjusted survival was one in which symptomatic patients undergo both US and 3D MRA. If the test results are disparate, the patient goes on to CA. Otherwise, if the stenosis is 70% to 99%, the patient proceeds directly to surgery. The next best strategy was US followed by CA if the US result is positive (similar to our strategy 1). The latter strategy was considerably cheaper than the first but resulted in a small loss in QALYs. The third best test strategy according to their model was US followed by MRA (similar to our strategy 4). However, the differences in QALYs between these three test strategies were very small, ie, <3 quality-adjusted days. Differences this small may not be meaningful clinically or statistically.
Vanninen et al54 studied 45 symptomatic patients with US, 3D MRA, and CA. From the accuracy results of this sample of patients, the authors compared the cost-effectiveness of seven test strategies for symptomatic patients. They concluded that the best test strategy for testing symptomatic patients was US followed by 3D MRA, if indicated. Our study arrived at this same conclusion, although our models differ considerably.
Our decision model has several limitations. First, we built the model based on estimates of prevalence and risk from the literature. These estimates are all potential sources of error; thus, our results are imprecise and may be biased. Second, our model assumed that the risk of stroke is a function of the degree of stenosis and the patient’s clinical symptoms. However, Bornstein et al55 have suggested that the risk of stroke may also be a function of the rate of stenosis progression. They define the “stenotic index” to describe the risk of stroke as a function of both the amount of stenosis and the rate of progression. To our knowledge, there are no outcome studies that have correlated the risk of stroke with the stenotic index. Thus, we did not consider this model for the natural history of stroke at this time. Finally, our model makes several assumptions, such as patients undergo surgery only once; such simplifying assumptions are often necessary in building decision models. A randomized controlled clinical trial would overcome the limitations of our study design; however, the monetary costs and duration of such a study limit the practicality of this approach.
Our model of the natural history of carotid artery stenosis suggests that asymptomatic patients with an elevated risk of carotid artery disease, such as patients with carotid bruits, may benefit from noninvasive screening if the prevalence of operable lesions is ≥20%, the benefits of surgery are similar to those observed in ACAS, and the quality of life with stroke is considerably lower than the quality of life without stroke. The test strategy of US followed by 3D TOF MRA, if indicated, is a promising approach to screening asymptomatic patients and to testing patients if they become symptomatic. Diagnostic testing should be performed where the technical and professional staff are adequately trained and experienced, the test protocols are standardized, and there is ongoing quality improvement.
Selected Abbreviations and Acronyms
|2D, 3D||=||two-dimensional, three-dimensional|
|ACAS||=||Asymptomatic Carotid Atherosclerosis Study|
|MRA||=||magnetic resonance angiography|
|NASCET||=||North American Symptomatic Carotid Endarterectomy Trial|
|QALYs||=||quality-adjusted life years|
|RIND||=||reversible ischemic neurological deficit|
|TIA||=||transient ischemic attack|
|TOF||=||time of flight|
This work was supported by National Institutes of Health grant 2-ROl-HL43812-04. The authors thank four anonymous reviewers and Daniel Kent, MD, for insightful, constructive suggestions; David Hesselink and Chris Sowl for compilation of the cost data; and Linda Placko for secretarial assistance.
- Received October 2, 1996.
- Revision received March 24, 1997.
- Accepted March 24, 1997.
- Copyright © 1997 by American Heart Association
The CASANOVA Study Group. Carotid surgery versus medical therapy in asymptomatic carotid stenosis. Stroke. 1991;22:1229-1235.
Feussner JR, Matchar DB. When and how to study the carotid arteries. Ann Intern Med.. 1988;109:805-818.
Derdeyn CP, Powers WJ. Cost-effectiveness of screening for asymptomatic carotid atherosclerotic disease. Stroke. 1996;27:1944-1950.
Sonnenberg FA, Beck JR. Markov models in medical decision making: a practical guide. Med Decis Making. 1993;13:322-338.
Matchar DB, Pauker SG. Transient ischemic attacks in a man with coronary artery disease: two strategies neck and neck. Med Decis Making. 1986,6:239-249.
Roederer GO, Langlois YE, Jager KA, Primozich JF, Beach KW, Phillips DJ, Strandness DE. The natural history of carotid arterial disease in asymptomatic patients with cervical bruits. Stroke. 1984;15:605-613.
Chambers BR, Norris JEW. Outcome in patients with asymptomatic neck bruits. N Engl J Med.. 1986;315:860-865.
Norris JEW, Zhu CZ, Bornstein NM, Chamber BR. Vascular risks of asymptomatic carotid stenosis. Stroke. 1991;22:1485-1490.
Hennerici M, Hulsbomer H, Rautenberg W, Hefter H. Spontaneous history of asymptomatic internal carotid occlusion. Stroke. 1986;17:718-722.
Harrison MJG, Marshall J. Prognostic significance of severity of carotid atheroma in early manifestations of cerebrovascular disease. Stroke. 1982;13:567-569.
Dennis M, Bamford J, Sandercock P, Warlow C. Prognosis of transient ischemic attacks in the Oxfordshire Community Stroke Project. Stroke. 1990;21:848-853.
Chambers BR, Norris JEW, Shurvell BL, Hachinski VC. Prognosis of acute stroke. Neurology. 1987;37:221-225.
Hier DB, Foulks MA, Swiontoniowski M, Sacco RL, Gorelick PB, Mohr JP, Price TR, Wolf PA. Stroke recurrence within 2 years after ischemic infarction. Stroke. 1991;22:155-161.
Sundt TM, Sandok BA, Whisnant JP. Carotid endarterectomy: complications and preoperative assessment of risk. Mayo Clin Proc.. 1975;50:310-306.
McCrory DC, Goldstein LB, Samsa GP, Oddone EZ, Landsman PB, Moore WS, Matchar DB. Predicting complications of carotid endarterectomy. Stroke. 1993;24:1285-1291.
Marzewski DJ, Furlan AJ, Louis PS, Little JR, Modic MT, Williams G. Intracranial internal carotid artery stenosis: long-term prognosis. Stroke. 1982;13:821-824.
Fryback DG, Dasbach EJ, Klein R, Klein BEK, Dorn N, Peterson K, Martin PA. The Beaver Dam Health Outcomes Study: initial catalog of health-state quality factors. Med Decis Making. 1993;13:89-102.
Tell GS, Polak JF, Ward BJ, Kittner SJ, Savage PJ, Robbins J. Relation of smoking with carotid artery wall thickness and stenosis in older adults: the Cardiovascular Health Study. Circulation. 1994;90:2905-2908.
Fine-Edelstein JS, Wolf PA, O’Leary DH, Poehlman H, Belanger AJ, Kase CS, D’Agostino RB. Precursors of extracranial carotid atherosclerosis in the Framingham Study. Neurology. 1994;44:1046-1050.
Fabris F, Zanocchi M, Bo M, Fonte G, Poli L, Bergoglio I, Ferrario E, Pernigotti L. Carotid plaque, aging, and risk factors: a study of 457 subjects. Stroke. 1994;25:1133-1140.
Sutton-Tyrrell K, Alcorn HG, Wolfson SK, Kelsey SF, Kuller LH. Predictors of carotid stenosis in older adults with and without isolated systolic hypertension. Stroke. 1993;24:355-361.
Pujia A, Rubba P, Spencer MP. Prevalence of extracranial carotid artery disease detectable by echo-Doppler in an elderly population. Stroke. 1992;23:818-822.
Vanninen R, Manninen H, Soimakallio S. Imaging of carotid artery stenosis: clinical efficacy and cost-effectiveness. AJNR Am J Neuroradiol. 1995;16:1875-1882.