Comparative Effectiveness of Carotid Stenting to Medical Therapy Among Patients With Asymptomatic Carotid Stenosis

Abstract

Background:

No completed trials have compared carotid artery stenting (CAS) to medical therapy (MT). We examined the effectiveness of CAS compared with MT in patients with asymptomatic carotid stenosis.

Methods:

We conducted a retrospective cohort study of 219 979 Veterans ≥65 years who received carotid imaging for asymptomatic carotid stenosis between 2005 and 2009 in the US Veterans Health Administration. We constructed a sample of patients who received MT (n=2509) and comparable patients who received CAS (n=551) and followed them for 5 years. Using target trial methodology, we computed weighted Kaplan-Meier curves and estimated the risk of fatal and nonfatal stroke in each group over 5 years of follow-up. We also estimated the cumulative incidence functions for fatal and nonfatal stroke accounting for nonstroke deaths as competing risks.

Results:

Five hundred fifty-one patients received CAS, and 2509 patients received MT. The observed rate of stroke or death (perioperative complications) within 30 days in the CAS arm was 2.2%. Using the target trial methodology, the 5-year risk of fatal and nonfatal stroke was similar among patients assigned to CAS (6.9%) compared with patients assigned to MT (7.1%; risk difference, −0.1% [95% CI, −2.6% to 2.7%]). In an analysis that incorporated the competing risk of death, the risk difference between the two arms remained nonsignificant (risk difference, −1.5% [95% CI, −3.0% to 0.3%]).

Conclusions:

In this sample of older male adults, we found no difference between MT and CAS in the treatment of asymptomatic carotid stenosis. Future studies in other settings are needed to confirm these findings.

Carotid artery stenosis accounts for ≈11% of strokes worldwide. Randomized controlled trials (RCTs) conducted over 2 decades established that carotid endarterectomy (CEA) is beneficial in preventing stroke in both asymptomatic^1–3 and symptomatic patients^4,5 with carotid stenosis. However, no completed trial is available that directly compared carotid artery stenting (CAS) to medical therapy (MT). One Europe-based trial was halted due to inadequate recruitment.⁶

See related article, p 1167

Carotid revascularization among asymptomatic patients is a primary prevention procedure. Revascularization is a trade-off between higher perioperative short-term risks which include stroke and death in exchange for a lower long-term risk of stroke. Stroke risk from carotid artery stenosis has been declining since the original asymptomatic carotid trials were completed due to marked improvements in MT.⁷ Stroke has now dropped in ranking from the third to the fifth leading cause of death in the United States.⁸ Given the declining stroke risk from asymptomatic carotid artery stenosis, the National Institute of Neurological Disease and Stroke funded CREST-2 (Carotid Revascularization and Medical Management for Asymptomatic Carotid Stenosis Trial) to determine whether revascularization offers any benefit beyond aggressive risk factor control.⁹ CREST-2 was launched in 2014 and is an ongoing trial that involves 2 parallel RCTs: (1) CEA plus intensive MT versus intensive MT alone and similarly (2) CAS plus intensive MT versus intensive MT alone. CREST-2 has a projected completion date of 2025 although the pandemic may affect this timeline.^9,10 While we await these trial results, it is also important to understand the comparative effectiveness of CAS compared with MT outside of a trial setting because patient selection may not be as optimal, interventionalists may have higher complication rates than those who participate in a trial, and, therefore, the risk-benefit ratio of intervention may be less favorable outside of RCTs.

In prior work, we evaluated the comparative effectiveness of CEA versus MT using the US National Department of Veterans Affairs (VA) and Medicare data.¹¹ Using the same methods and date sources,¹¹ we examined whether CAS is superior to initial MT in a community setting. We hypothesized that MT is a more favorable treatment strategy compared with CAS among patients with asymptomatic carotid stenosis.

Methods

Data sets will be deidentified and made available in accordance with VA policies after those requesting data sign a Letter of Agreement.

Study Population

Using VA national data, we identified 219 979 patients ≥65 years with at least one primary care (eg, general medicine, women clinic) or subspecialty visit (eg, geriatrics, cardiology, endocrine, pulmonary, neurology) in the VA who received diagnostic carotid imaging (carotid ultrasound, computed tomography angiography, or magnetic resonance angiography) between 2005 and 2009 using Current Procedural Terminology codes and International Classification of Diseases-9 codes (Table S1). We required at least one visit to the VA in the year before the carotid imaging test to ensure the presence of basic baseline data. To create a cohort of patients with asymptomatic carotid stenosis, we excluded all patients with carotid imaging tests who had a stroke or transient ischemic attack (TIA) in the 6 months before the imaging test using a previously developed highly sensitive algorithm based on International Classification of Diseases-9 codes that was modified for this study to include retinal strokes and TIA (Table S2).¹² We also excluded all patients who had received a CEA or CAS in the 6 months before index carotid image to ensure we excluded all individuals who may have received intervention for TIA or stroke.

Data Sources

All patient demographic data, administrative data (eg, comorbid conditions, utilization data), vital statistics, laboratory data, and pharmacy data were retrieved from the VA Corporate Data Warehouse and Medicare data.¹³ Text notes for carotid image reports were identified from the VA Corporate Data Warehouse. Data on carotid revascularization were retrieved from the VA Corporate Data Warehouse using International Classification of Diseases-9 codes (CAS:00.61, 00.63; CEA:38.12), and Current Procedural Terminology codes (CAS:37215 and 37216; CEA:35301). Vital status, death, and cause of death were obtained from the Suicide Data Repository, which is jointly administered by the Department of Defense and the VA. The Suicide Data Repository was created to track deaths by suicide by military personnel; it also includes comprehensive cause of death data for all Veterans.¹⁴

Identifying Patients With Carotid Stenosis

Using a previously developed and validated natural language processing algorithm designed to reduce the burden of chart review,¹⁵ we excluded all carotid imaging reports with <50% stenosis or hemodynamically insignificant stenosis (Figure 1). We imported all text reports for patients with at least one carotid image with ≥50% stenosis (n=71 839) into a database for human screening. Each note was screened by one reviewer who was trained in assessing carotid stenosis (details on training are provided in Table S3). In this preliminary screening process, we identified all patients who had carotid imaging text reports indicating >70% carotid stenosis or notes with a qualitative description of near occlusion, critical stenosis, or severe stenosis. At this step, we included any patient who could potentially have severe stenosis. Subsequently, medical record review was used to confirm stenosis. We identified 13 371 potential Veterans with ≥70% stenosis (Figure 1).

Figure 1.

Cohort construction: identifying patients who received intervention and medical therapy (MT). CAS indicates carotid artery stenting; CEA, carotid endarterectomy; and TIA, transient ischemic attack.

Identifying Patients Who Received Initial MT

We considered time zero to be the time of the first carotid image demonstrating stenosis. Administrative data on baseline characteristics (in the year before index carotid image) demonstrated that patients who receive intervention differed from patients who did not in their prevalence of baseline comorbidities (Table S4). The first step in our analytic strategy was to construct a sample of MT patients who were similar to those undergoing intervention using propensity score matching. Many patients are incidentally identified with carotid stenosis during a workup that involves head and neck imaging. Such individuals may not be suitable for intervention due to comorbidities (eg, head and neck or esophageal cancer). Therefore, we fit a generalized boosted model for treatment (intervention versus MT) using a set of baseline covariates (Table 1)¹⁶ to identify groups that were largely similar and could be eligible for MT or intervention. Using the estimated propensity scores, we used a nearest neighbor match to select MT patients who were similar to intervention patients based on measured covariates. At this stage, we retained all matched MT patients who were similar to intervention patients on measured covariates because we expected some patients might be excluded during the medical record review phase. The result of this first step in our analytic strategy was to construct a cohort of potential MT and intervention patients who were largely similar in measured baseline covariates. The propensity score method does not account for all the differences between the two arms because subsequent medical record review will result in exclusions that may create differences between the two groups.

Table 1.

Baseline Characteristics of the Carotid Cohort Based on Actual Treatment Received

Medical Record Review

Next, we reviewed each potentially eligible patient’s chart to confirm both the degree of carotid stenosis and symptom status (hemorrhagic or ischemic stroke or TIA in 6 months before the index imaging). We excluded patients with a history of stroke or TIA in the 6 months before index imaging. Patients with carotid stenosis between 50% and 70% were also excluded because recent trials have included patients with a higher threshold of stenosis and practice patterns have moved toward patients perceived at higher risk of stroke; therefore, our goal was to include patients with at least 70% but <100% carotid stenosis.¹⁷ We also collected data on disease severity that were not available in national VA data (eg, ejection fraction, pulmonary function). We collected data on receipt of aspirin at baseline because this is an over-the-counter medication, which may not be dispensed by VA pharmacies. We also collected data on whether the patient received a CEA or CAS and whether the patient had a history of remote stroke. For details on medical record review protocol, see Table S3.

Among the 3795 cases designated as intervention patients and the 3888 propensity-matched MT patients, 823 patients were excluded because of previous stroke or TIA, and 872 patients were excluded because they did not meet stenosis criteria. We also excluded 139 patients for whom there was ambiguity on group classification because the chart review suggested they had received a procedure from an outside hospital; however, we were unable to verify the procedure in VA or Medicare data. Finally, we excluded 77 patients from the MT arm when a provider specifically documented that the patient was not a surgical candidate in the medical record. After applying exclusion criteria, we had 551 CAS patients, 2712 CEA patients, and 2509 MT patients (Figure 1). The comparative effectiveness of CEA compared with MT was reported in a separate manuscript mainly because the sample size of participants who received CAS was small, limiting our ability to conduct a 3-way comparison.¹¹ In addition, patients who received CAS in this real-world sample were sicker than patients with CEA limiting our ability to identify suitable triplet matches (Table S3). In this study, we compare CAS to initial MT.

Emulating Trials Comparing Intervention to MT

When assessing the comparative effectiveness of early intervention (CAS) versus initial MT on the outcome of fatal and nonfatal stroke, we chose to emulate the analyses used in the ACST (Asymptomatic Carotid Surgery Trial)—the last completed trial to compare carotid intervention to MT.² Because of potential delays in receiving treatment after carotid imaging and the possibility that patients in the initial MT arm may go on to have CAS or CEA due to the development of a subsequent event/indication for these procedures (ie, stroke or TIA),¹⁸ similar to ACST, we operationally defined the intervention as “receipt of intervention (CAS) within the first year after carotid imaging.” Thus, our target trial was of early intervention (CAS within 1 year) versus initial medical management (initial MT).

Outcome Measure and Follow-Up

The primary outcome was fatal and nonfatal stroke. Stroke events during the follow-up period were assessed using the Tirschwell high-specificity algorithm from the national VA and Medicare data.¹⁹ All patients were followed from the date of the first carotid image exhibiting ≥70% stenosis for up to 5 years.

Statistical Analysis

First, we computed Kaplan-Meier estimates of 5-year stroke risk (and associated 95% CIs) from the date of index imaging to allow comparisons of fatal and nonfatal stroke risk in this sample to other populations. Second, we used a recently described methodological framework, called the target trial method,¹¹ for use with observational studies in which the active intervention can be applied at varying times post-baseline.^20–22 Specifically, we randomly assigned each subject at baseline to 1 of the 2 treatment groups (CAS or initial MT), possibly contrary to the treatment that the patient actually received. Subjects were then followed using the observational data until the first of 4 events: (1) the occurrence of a fatal or nonfatal stroke (outcome), (2) the end of the study period (administrative censoring 5 years after baseline), (3) their treatment status became inconsistent with their assigned treatment status (treatment switching), or (4) patients enrolled in Medicare Advantage (health maintenance organization) between baseline and 1 year (censored because procedures and diagnoses recorded under those plans were not available).²³ Thus, a patient who was randomly assigned to the MT arm and who subsequently underwent CAS within 1 year would be censored due to treatment switching on the date they underwent CAS because, once they underwent CAS (within 1 year), their current treatment was no longer consistent with the arm to which they were assigned (initial MT). Similarly, a patient who was randomly assigned to the CAS arm and who had not undergone CAS within 1 year would be censored after 1 year (because their treatment at 1 year was, at that point, no longer consistent with the arm to which they were randomly assigned). Patients who received CEA before CAS, concurrent coronary artery bypass grafting with CAS, or concurrent aortic valve replacement with CAS were also censored because their treatment was not consistent with the arm to which they were assigned (CAS alone). We considered a concurrent procedure to be one that occurred within 1 day of the CAS.

Because the actual choice of treatment (CAS versus MT) may have depended on patients’ baseline and postbaseline risk and prognostic factors, we accounted for informative censoring due to treatment switching by incorporating inverse probability of censoring weights (IPCW) in our analysis. These methods were used in a previously published article comparing CEA to initial MT.¹¹ The use of IPCW in this context is similar to the use of IPCW in RCTs to account for informative censoring (eg, due to treatment switching or noncompliance with treatment).²⁴ We estimated the IPCW using a Cox model that included measured baseline (time invariant) and time-varying covariates to account for the baseline differences observed in Table 1 and to account for the possibility that patients may develop a condition (eg, cancer) between index and treatment assignment that may make them ineligible for the arm they were assigned. Variables included in the censoring model are included in Table S5. We then fit a Cox proportional hazards model for the outcome (fatal and nonfatal stroke) as a function of an indicator variable for the randomly assigned treatment group (CAS versus MT) and incorporated the time-varying IPCW. Regression diagnostics revealed violations of the proportional hazards assumption (including crossing CAS and MT survival curves), and thus we focused on estimating and reporting 5-year fatal and nonfatal stroke risk. Specifically, we estimated the 5-year risk of fatal and nonfatal stroke in the CAS and MT groups, incorporating the IPCW, and then computed the risk difference comparing the CAS to the MT group. We computed our final point and 95% CI estimates of the 5-year risk and risk difference from the 50th, 2.5th, and 97.5th percentiles of the bootstrap distribution for the risks and risk difference. Finally, we accounted for competing risks in our analysis.²⁵ We computed the 5-year risk difference using methods that account for competing risks (deaths due to causes other than stroke). To do so, we estimated the cumulative incidence functions for fatal and nonfatal stroke in the CAS and MT groups, accounting for nonstroke deaths as competing risks. These cumulative incidence functions were weighted by the time-varying IPCW as above. Some patients were missing data on baseline characteristics (eg, body mass index, blood pressure). The small number of patients with missing data were categorized as unknown in the categorical variables (Table 1) and included in the propensity model and censoring model.

This study was approved by the University of California, San Francisco Institutional Review Board. Given the large sample and retrospective nature of the data collected, we were granted a waiver of informed consent.

Results

Baseline Characteristics of Sample

Between 2005 and 2009, 551 cases received CAS and 2509 received MT within 1 year after the index carotid image.

Table 1 provides baseline characteristics of the propensity-matched sample comparing CAS to MT. The baseline sociodemographic characteristics of the CAS versus MT patients were similar, with no difference in mean age, race, ethnicity, marital status, and Medicaid enrollment. Standardized differences were <20% on most variables indicating successful matching. There were some notable differences between the CAS and MT patients in terms of comorbidity. For example, CAS patients were more likely to have ischemic heart disease (61.5% versus 45.7%), chronic obstructive pulmonary disease (17.1% versus 8.7%), and use antianginals (31% versus 21.6%). These differences were accounted for in the statistical analysis by estimating the IPCW using a Cox model that included the measured baseline (time invariant) and time-varying covariates.

Observed Risk of Stroke and Death the Community Sample Based on Actual Treatment Received

The observed risk of stroke or death (perioperative complications) within 30 days in the CAS arm was 2.2% ([95% CI, 1.2%–3.8%] Table 2). The observed unadjusted 5-year risk of fatal or nonfatal stroke among patients with carotid stenosis was 10.3% (95% CI, 7.8%–13.4%) in the CAS group and 6.9% (95% CI, 5.8%–8.1%) in the MT group (Table 2). Five-year survival was 61.7% (95% CI, 57.7%–65.8%) and 66.9% (95% CI, 65.0%–68.7%) in the CAS and MT arms.

Table 2.

Five-Year Stroke Risks in Patients Receiving CAS Versus Initial MT

Target Trial Results: 5-Year Stroke-Free Survival (Patients Who Were Randomized to CAS Compared With Those Randomized to MT, With Censoring for Treatment Switching)

After incorporating the IPCW, the 5-year risk of fatal and nonfatal stroke was the same for patients randomized to CAS and to MT (6.9% versus 7.1%; risk difference, −0.1% [95% CI, −2.6% to 2.7%]; Table 2; Figure 2).

Figure 2.

Target trial results (emulating the design of ACST [Asymptomatic Carotid Surgery Trial]): 5-y stroke-free survival among patients who were randomized to carotid artery stenting compared with those randomized to initial medical therapy after adjusting for baseline characteristics and incorporating inverse probability of censoring weights in the analysis.

Figure 3 displays the 5-year cumulative incidence function accounting for competing risks comparing CAS versus MT patients. The stroke risk at 5 years of follow-up was 4.7% in the CAS group and 6.2% in the MT group. The risk difference between the two groups when the competing risk of death was taken into consideration was −1.5% (95% CI, −3.0% to 0.3%), suggesting no difference between CAS and MT.

Figure 3.

Five-year cumulative incidence of stroke accounting for competing risks among patients who were assigned to carotid artery stenting (CAS) compared with those assigned initial medical therapy (MT) after adjusting for baseline characteristics and incorporating inverse probability of censoring weights in the analysis.

Discussion

We used national VA and Medicare data to mimic an RCT comparing CAS to MT. In our main analysis, we found no benefit from CAS compared with MT in improving stroke-free survival in this older male population. The differences between CAS and MT after 5 years of follow-up were smaller than the observed upfront perioperative complication risk in our study (2.2%) and in other community settings.²⁶ Both the main analysis and the analysis including competing risks suggest that CAS did not offer an advantage in terms of stroke reduction compared with initial MT alone. In studies comparing CAS to CEA, younger individuals had better outcomes with CAS. Our study was limited to older adults, and, therefore, we were unable to examine whether CAS was superior to MT among younger Veterans.

Our results although limited by a modest sample provide some information on the comparative effectiveness of CAS to initial MT in real-world settings where patients may have higher perioperative complications, higher comorbidity, or poor risk factor control. Although the perioperative complications in our study were <2.2%, perioperative stroke or death as high as 4.0% among asymptomatic patients undergoing CAS have been reported outside of trials, which is more than the stroke risk reduction we observed between 2009 and 2014 (the period of this study).²⁶ One of 3 patients in this study did not survive 5 years. If patients selected for CAS have a short life expectancy, then they may not accrue enough long-term stroke-free years to offset the short-term increased risk of the procedure itself. Finally, we found significant room for improvement in risk factor control both among patients who received intervention and those who received initial MT. About one-fifth of patients were not on antiplatelet therapy, a third were not adherent to statins, and there was room for improvement in blood pressure control. Despite substantial room for improvement in terms of risk factor control, the 5-year stroke risk in the MT arm was 6.9%, which was lower than the 11% stroke risk reported in ACST—the last published trial examining carotid revascularization.²⁷

The analytic approach also deserves comment. A key issue to consider is that the intervention (CAS) can be applied at any time after imaging. A successful analytic strategy thus accounts for and avoids immortal time bias. First, we considered the use of a landmark analysis, in which subjects were followed from 1 year after the first carotid image (the landmark date equals the image data plus 365 days).^28,29 Subjects for whom an event occurred between the imaging date and the landmark date would be excluded. Treatment status would have been based on what occurred between the imaging date and the landmark date. While this approach avoids immortal time bias and allows for a clearly defined exposure, it results in the exclusion of all of the early events. Consequently, it would artificially make the CAS look safer because the early perioperative stroke events would be removed. Furthermore, this approach provides limited information to inform clinical decision-making because all information on early complications is excluded. Second, we considered using a Cox proportional hazards regression model in the full sample, treating the intervention as a time-varying covariate and reporting the association between CAS and the outcome using a hazard ratio. Subjects would have been classified in the MT arm until they underwent CAS, at which time they would have been classified as being in the CAS group. Although this approach avoids immortal time bias and is analytically simple to conduct, it has a primary limitation. This approach would result in an estimation of a relative hazard ratio and does not allow estimation of absolute treatment effects. Using the target trial method, we were able to estimate the absolute reduction in the risk of the outcome, which are important quantities for medical decision-making. In contrast, the Cox model only allows for estimation of relative changes in the hazard of the outcome. Therefore, we chose the target trial methods to enable our analysis to both mimic an RCT and account for immortal time bias and time-varying confounding between index image date and treatment assignment.

Several limitations are noted. First, although perioperative strokes were accounted for in the outcome of fatal and nonfatal stroke over 5 years of follow-up, we were unable to incorporate the perioperative complication of death as a result of CAS into the analysis. Second, we used national VA, Medicare, and Suicide Data Repository data to assess fatal and nonfatal stroke outcomes. It is possible that despite using standard, established methods for assessing stroke in the administrative data, we may have missed stroke events that were not coded.^12,19 Third, it is possible that we missed stroke documentation in the medical record review; we strived to minimize any abstraction errors by rigorous training of the abstractors and quality control of the abstracted charts (Supplemental Material). Fourth, our study was conducted among older male Veterans, and our findings may not be generalizable to non-Veteran populations although older male Veterans are similar to male Medicare enrollees in terms of overall comparability and men receive the majority of these procedures in the Medicare program.^30,31 Fifth, the implementation of our analytic strategy required censoring subjects at the time at which their actual treatment became discordant with that to which they had been randomized per the target trial methodology. This informative censoring was accounted for using model-based IPCWs, and as with most methods used for the analysis of observational studies, the success of our approach rests, in part, on the assumption that we have measured an adequate set of variables to account for the informative censoring. Although we had access to a data source with detailed clinical data, it is possible that subjects’ treatment became discordant with that to which they were randomized for reasons that we were unable to capture.¹⁷ Sixth, as in all observational studies, we cannot account for confounding due to unmeasured covariates. Seventh, the sample of patients undergoing CAS was modest; future studies with a larger sample in other community settings are needed to confirm these findings.

In conclusion, in this sample of older male adults, we found no difference between MT and CAS in the treatment of asymptomatic carotid stenosis. Future studies in other settings are needed to confirm these findings.

Article Information

Sources of Funding

This project was supported by NIH R01 HL114563-01A1. Support for Veterans Affairs (VA)/Centers for Medicare and Medicaid Services data was provided by the Department of Veterans Affairs, VA Health Services Research and Development Service, and VA Information Resource Center (project numbers SDR 02-237 and 98-004).

Disclosures

Dr Johanning reports holding a patent to FutureASSURE LLC, which holds intellectual property related to frailty assessment. Dr Cheng reports travel support from the American Academy of Neurology, travel support from the American Academy of Neurology, and grants from the National Institutes of Health (NIH). The funding source (NIH) had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. The other authors report no conflicts.

Supplemental Material

Tables S1–S5