Carotid Artery Stenting vs Carotid Endarterectomy: Meta-analysis and Diversity-Adjusted Trial Sequential Analysis of Randomized Trials

Carotid artery stenting (CAS) has emerged as an alternative to carotid endarterectomy (CEA) for the treatment of carotid artery occlusive disease¹ and is endorsed by the American Heart Association/American Stroke Association guidelines as a reasonable strategy when performed by operators with established periprocedural morbidity and mortality of 4% to 6% (class IIa).²The European Society of Vascular Surgery recommends CAS instead of CEA in participants at high risk for cardiovascular disease if it is conducted in high-volume centers with documented low perioperative stroke and death rates.³

However, the safety and efficacy of CAS are controversial. The most recent Cochrane review⁴ (2009) on this topic found that CAS conferred significant reductions not only in cranial nerve injury but also in myocardial infarction (MI) (periprocedural); however, CAS was associated with a significant increase in 30-day death or stroke, which was no longer significant in a random-effects model. In the recently concluded International Carotid Stenting Study (ICSS), CEA was superior to CAS at 30 days following the procedure.⁵ These studies mainly evaluated periprocedural outcomes where the data are controversial. There are limited data on the intermediate to long-term outcomes of these revascularization strategies. Moreover, the results of the Carotid Revascularization Endarterectomy vs Stenting Trial showed equivalent long-term risks of the 2 procedures.⁶

The objective of our analysis was to evaluate the periprocedural (≤30-day) and intermediate to long-term outcomes of CAS vs CEA in participants with carotid artery disease.

We conducted PubMed, EMBASE, and Cochrane Central Register of Controlled Trials searches using the terms carotid artery disease, endarterectomy, and carotid artery stenting in humans until June 2010. We checked the reference lists of review articles, meta-analyses, and original studies identified by the electronic searches to find other eligible trials. There was no language restriction for the search. Authors of publications were contacted when results were unclear or when relevant data were not reported.

Eligible trials had to fulfill the following criteria to be included in this analysis: (1) randomized clinical trials (RCTs) of participants with or without symptomatic carotid artery stenosis comparing CAS with CEA, with or without an embolic protection device (EPD); and (2) reporting 30-day or longer-term outcomes.

Two of us (S.B. and S.K.) independently assessed trial eligibility and trial bias risk and extracted data. Disagreements were resolved by consensus. The bias risks of trials were assessed using the components recommended by the Cochrane Collaboration⁷: sequence generation of allocation; allocation concealment; blinding of participants, personnel, and outcome assessors; incomplete outcome data; selective outcome reporting; and other sources of bias. Of note, the studies did not differ for the quality components of incomplete outcome data, selective outcome reporting, or other sources of bias. For the component of blinding, given the nature of the trials (surgery vs less invasive procedure, done under different settings), none of the trials had blinding of participants or personnel. Thus, only blinding of outcome assessors was considered. Trials with high or unclear risk for bias for any 1 of the first 3 components were considered to have high risk of bias. Otherwise, they were considered to have low risk of bias.

Periprocedural (30-day) outcomes included death, MI, or stroke; death or stroke; any stroke; and MI.

The intermediate to long-term outcomes were those as in the Stenting and Angioplasty With Protection in Patients at High Risk for Endarterectomy (SAPPHIRE) trial⁸; composite of periprocedural death, MI, or stroke plus ipsilateral stroke or death thereafter; periprocedural death or stroke plus ipsilateral stroke thereafter; death or any stroke; and any stroke.

Other outcomes evaluated were cranial nerve injury and carotid restenosis. Of note, all the outcomes stated earlier include periprocedural events along with the intermediate to long-term events.

The outcomes were chosen based on the relevance to contemporary trials comparing CAS vs CEA. The composite of periprocedural death, MI, or stroke plus ipsilateral stroke or death after the periprocedural period (>30 days) captures events related to the stenting or CEA during the periprocedural period and avoids capturing contralateral strokes during the extended period of follow-up.

Intention-to-treat meta-analysis was performed in line with recommendations from the Cochrane Collaboration and the Preferred Reporting Items for Systematic Reviews and Meta-analyses Statement,^7,9 using standard software (Stata version 9.0 statistical software; StataCorp LP, College Station, Texas).¹⁰ Heterogeneity was assessed using the I² statistic.¹¹I² is the proportion of total variation observed between the trials attributable to differences between trials rather than sampling error (chance), with I² < 25% considered low and I² > 75% considered high. We used the Peto method for odds ratios (ORs).^12,13 The Peto OR is viewed as the most optimal approach when there are relatively few events in individual trials. Publication bias was estimated visually by funnel plots and using the Begg test and the weighted regression test of Egger.¹⁴ Primary analyses were performed after stratifying the studies based on the cohort enrolled (symptomatic vs asymptomatic patients with carotid artery disease). Three different analyses were performed: (1) evaluation of periprocedural events (30-day events); (2) evaluation of intermediate to long-term events; and (3) evaluation of intermediate to long-term events but excluding periprocedural events.

Subgroup analyses were performed comparing the following: trials with low risk of bias vs trials with high risk of bias; use of an EPD or not; US trials vs non-US trials (to examine for variability in practice patterns); and all trials vs trials after excluding the Carotid and Vertebral Artery Transluminal Angioplasty Study (CAVATAS) trial, in which most patients underwent balloon angioplasty alone, the ICSS trial (interim analysis), in which the follow-up was only 4 months for the intermediate to long-term analysis, and the CAVATAS for the periprocedural analysis. We estimated the difference between the estimates of the subgroups according to tests for interaction.¹⁵P < .05 indicates that the effects of treatment differ between the tested subgroups. For the intermediate to long-term analyses, to account for varying length of follow-up of different studies, various subgroup analyses were performed: 2 years or less, 2 to 4 years, and more than 4 years of follow-up. We also performed a metaregression analysis adjusting for duration of follow-up. In addition, a metaregression analysis was performed to evaluate the effect of age on outcomes. We used residual maximum likelihood to estimate the additive (between-study) component of variance τ² for the metaregression analysis. Bootstrap analyses were performed using a Monte Carlo permutation test for metaregression using 10 000 random permutations.¹⁶

In a single trial, interim analyses increase the risk of type I error. To avoid an increase of overall type I error, monitoring boundaries can be applied to decide whether a single randomized trial could be terminated early because of the P value being sufficiently small. Because no reason exists for the standards for a meta-analysis to be less rigorous than those for a single trial, analogous trial sequential monitoring boundaries (TSMBs) can be applied to meta-analysis as trial sequential analysis (TSA).^17,18 Cumulative meta-analysis of trials is at risk for producing random errors because of few data and repetitive testing of accumulating data, and the information size requirement analogous to the sample size of a single optimally powered clinical trial may not be met.^17,18

The underlying assumption for TSA is that significance testing is performed each time a new trial is published. The TSA depends on the quantification of the required information size. In this context, the smaller the required information size is, the more lenient the TSA is, thus the more lenient the criteria are for statistical significance.^17,18 A required diversity (D²)–adjusted information size was calculated, with D² being the relative variance reduction when the meta-analysis model is changed from a random-effects model to a fixed-effect model.¹⁹D² is the percentage that the between-trial variability constitutes of the sum of the between-trial variability and a sampling error estimate considering the required information size. D² is different from the intuitively obvious adjusting factor based on the common quantification of heterogeneity, the inconsistency (I²), which may underestimate the required information size.¹⁹

The TSA was performed with a desire to maintain an overall 5% risk of type I error, being the standard in most meta-analyses and systematic reviews, and we calculated the required information size (ie, the meta-analysis information size needed to detect or reject an intervention effect of a 20% [or 15%] relative risk increase [RRI] with a 20% risk of type II error and power of 80%).^17,18

We identified 263 RCTs through electronic and hand searches, of which a total of 13 RCTs^5,6,8,20-29 fulfilled the inclusion criteria and were chosen for this analysis (Figure 1). Eleven trials reported both periprocedural and long-term outcomes. The CAVATAS data were obtained from the updated analysis reported recently.²⁰

The baseline characteristics, inclusion criteria, exclusion criteria, and bias-risk assessment are summarized in Table 1 and Table 2. The 13 RCTs enrolled 7477 patients: 3754 patients (50.2%) were randomized to CAS and 3723 patients (49.8%) were randomized to CEA; the patients were followed up for a weighted mean (SD) of 2.7 (1.6) years.

Of the 13 trials, most included participants with symptomatic carotid artery disease except the trial by Brooks et al,²² the SAPPHIRE trial⁸ where 70% of participants were asymptomatic, the Carotid Revascularization Endarterectomy vs Stenting Trial⁶ where 47% of participants were asymptomatic, and the Trial of Endarterectomy vs Stenting for the Treatment of Carotid Atherosclerotic Stenosis in China²⁹ that included both cohorts. For most of our analyses, the symptomatic and asymptomatic cohorts of the Carotid Revascularization Endarterectomy vs Stenting Trial and the SAPPHIRE trial (data on file at Harvard Clinical Research Institute) were analyzed separately. In 6 trials, an EPD was used in most participants.^{5,6,8,23,29,31} In the Stent-Protected Angioplasty vs Carotid Endarterectomy trial, EPDs were used only in 27% of participants (considered with the no-EPD group for this analysis).²⁵

Among the 13 RCTs included in this analysis, 7 were considered trials with low risk of bias as described in “Methods”^{5,8,25,26,30,31} and the others were considered trials with unclear or high risk of bias (Table 2).

Carotid artery stenting was associated with a 31% increase in the risk of periprocedural death, MI, or stroke (Figure 2A) when compared with CEA (meta-analytic rate, 5.68% [95% confidence interval {CI}, 3.88%-7.48%] vs 4.73% [95% CI, 3.52%-5.94%], respectively), driven mainly by the analysis in the symptomatic cohort (P_interaction = .05). Carotid artery stenting as compared with CEA was associated with a 65% increase in the risk of periprocedural death or stroke (Figure 2B) (meta analytic rate, 5.50% [95% CI, 3.69%-7.30%] vs 3.81% [95% CI, 2.63%-5.00%], respectively) and a 67% increase in the risk of any periprocedural stroke (Figure 3A) (meta analytic rate, 4.97% [95% CI, 3.31%-6.63%] vs 3.19% [95% CI, 2.13%-4.25%], respectively). However, CEA as compared with CAS was associated with a 122% increase in the risk of periprocedural MI (Figure 3B) (meta analytic rate, 1.16% [95% CI, 0.45%-1.88%] vs 0.27% [95% CI, 0.10%-0.44%], respectively). The results were similar for both symptomatic and asymptomatic cohorts regarding these outcomes (P_interaction > .05). There was no or low to modest heterogeneity in the analyses and no evidence for publication bias.

For the outcome of death, MI, or stroke, the required diversity (D² = 68%)–adjusted information size (a priori diversity-adjusted information size = 26 359) was calculated based on a control event proportion (in the CEA group) of 9.2%, an RRI of 20%, α of 5%, and β of 20%. The cumulative z curve crossed the traditional boundary but not the TSMB, suggesting a lack of firm evidence for an RRI of 20% in the CAS group compared with the CEA group (Figure 4A). However, the cumulative z curve crossed both the traditional boundary and the TSMB for the outcomes of death or stroke (Figure 4B) and any stroke (Figure 4C), demonstrating firm evidence for a 20% RRI in the CAS group compared with the CEA group. However, for the outcome of MI, both the traditional boundary and the TSMB were crossed by the cumulative z curve, demonstrating firm evidence for a 15% relative risk reduction in the CAS group compared with the CEA group.

Analyzing only trials with low risk of bias, excluding 0-event trials, and/or the CAVATAS trial did not make any noticeable differences for the analyses.

Carotid artery stenting as compared with CEA was associated with a 19% increase in the risk for the SAPPHIRE-like composite outcome (Figure 5A), a 38% increase in the risk for the composite of periprocedural stroke or death plus ipsilateral stroke thereafter (Figure 5B), a 24% increase in the risk for death or stroke (Figure 6A), and a 48% increase in the risk for any stroke (Figure 6B). The results were similar for both symptomatic and asymptomatic cohorts (P_interaction > .05). There was no or low to modest heterogeneity in the analyses and no evidence for publication bias.

For the SAPPHIRE-like composite outcome (Figure 7A), the cumulative z curve crossed neither the traditional boundary nor the TSMB but was very close to the futility boundaries, suggesting a lack of firm evidence for a 20% RRI in the CAS group compared with the CEA group. Similarly, the cumulative z curve crossed neither the traditional boundary nor the TSMB for the outcome of periprocedural death or stroke plus ipsilateral stroke thereafter (Figure 7B). For the outcome of death or stroke, the cumulative z curve crossed the traditional boundary but not the TSMB, suggesting a lack of firm evidence for a 20% RRI in the CAS group compared with the CEA group (Figure 7C). For the outcome of any stroke, the TSMB was crossed by the cumulative z curve, suggesting firm evidence for at least a 20% RRI with CAS when compared with CEA (Figure 7D).

Carotid artery stenting as compared with CEA was associated with an 85% reduction in the risk of cranial nerve injury (all reported cranial nerve injuries were in the periprocedural period, and none occurred during longer follow-up) (OR = 0.15; 95% CI, 0.10-0.22; I² = 0.0%) (meta-analytic rate, 0.22% [95% CI, 0.05%-0.39%] vs 5.20% [95% CI, 4.40%-6.00%], respectively) but a 180% increase in the risk of intermediate to long-term carotid restenosis (OR = 2.80; 95% CI, 1.96-4.00; I² = 0.0%). Analysis of outcomes excluding periprocedural events showed that the 2 treatment strategies were comparable with similar event rates across the 2 groups (Table 3). However, the event rate for all of the outcomes outside the periprocedural period was low.

The results of the subgroup analyses are summarized in Table 4. Analysis stratified by bias risk (low vs high risk of bias), EPD use vs no use, US vs non-US trials, and excluding the CAVATAS and/or ICSS data made no noticeable difference from the main analyses (P_interaction > .05). To account for varying length of follow-up, additional analyses stratified by duration of follow-up did not make any noticeable difference to the analyses. Similarly, in a metaregression analysis evaluating the effect of length of follow-up, there was a trend toward no increased risk in the CAS group with longer length of follow-up (>40 months) for the outcomes of SAPPHIRE-like outcome (P = .06) (eFigure 1) and death or any stroke (P = .09) (eFigure 2) but not for the outcomes of periprocedural death or stroke plus ipsilateral stroke thereafter (P = .82) (eFigure 3) or any stroke (P = .92) (eFigure 4), where the increased risk with CAS persisted with follow-up duration. The P values for the trend were derived from Monte Carlo permutation using 10 000 random observations. There was no relationship between outcomes and mean age of participants for any of the outcome measures (data not shown).

The results of this analysis of currently available data using outcome measures considered standard for contemporary trials suggest that CAS is associated with significantly increased risk for both periprocedural and intermediate to long-term outcomes when compared with CEA (except for cranial nerve palsies and periprocedural MI). The TSMB was crossed by the cumulative z curve, suggesting firm evidence for at least a 20% RRI of periprocedural death or any stroke, any periprocedural stroke, and any intermediate to long-term stroke but at least a 15% reduction in MI with CAS compared with CEA.

Previous studies and meta-analyses have evaluated the efficacy and safety of CAS compared with CEA on periprocedural outcomes (≤30-day events). The most recent Cochrane review⁴ (2009) evaluated 10 randomized trials that enrolled 3178 participants and found that CAS conferred significant reductions in cranial nerve injury and MI but that it was associated with a significant increase in 30-day death or stroke, which was no longer significant in a random-effects model. More recent randomized trials not included in the Cochrane review suggest that CEA might be superior to CAS for 30-day events. In the recently concluded ICSS comparing CEA with CAS in 1710 participants with recently symptomatic carotid stenosis eligible for either procedure, CEA was superior to CAS at 30 days following the procedure, with nearly twice as many strokes associated with CAS when compared with CEA.⁵ Our updated analyses including the most recent trials showed an increased risk of periprocedural outcomes except for cranial nerve injury and MI.

To our knowledge, this is the largest meta-analysis evaluating the intermediate to long-term outcomes of CAS vs CEA. Our study differs from the analysis by Meier et al³³ in that we evaluated outcomes considered standard for evaluating intermediate to long-term outcomes with CAS, namely periprocedural death, MI, or stroke plus ipsilateral stroke or death thereafter and periprocedural death or stroke plus ipsilateral stroke thereafter. We included the results of the recently published or presented studies, ie, the Carotid Revascularization Endarterectomy vs Stenting Trial and the ICSS (interim analysis). The CAVATAS trial had a stroke definition more in line with other trials. Analysis was stratified based on the cohort enrolled (symptomatic vs asymptomatic), which is clinically more meaningful. Finally, a robust TSA was performed with monitoring boundaries.

Our results suggest that the periprocedural increased risk of CAS continues to be seen in the intermediate to long-term periods as well. These results were observed regardless of whether EPDs were used. The efficacy of an EPD in preventing stroke is controversial. In the magnetic resonance imaging substudy of the ICSS, new ischemic lesions were 3 times more common in the stenting group than in the endarterectomy group following treatment.³⁴ Moreover, EPDs were not effective in preventing cerebral ischemia during stenting. It is not clear whether this is due to embolization during deployment of the EPD itself. Similarly, investigators have questioned whether the operator experience has resulted in different results for studies in the United States compared with non-US studies.³⁵ However, this is controversial.³⁶ In the Endarterectomy vs Angioplasty in Patients With Symptomatic Severe Carotid Stenosis trial, participants treated by experienced interventionalists (who had performed >50 carotid stenting procedures) had a higher 30-day risk of stroke or death of 12.2% compared with 11.0% in participants treated by those who had performed 50 or fewer procedures and 7.1% in participants treated by interventionalists who were being proctored.³⁷ Our subgroup analysis based on US trials vs non-US trials did not show any significant interaction for any of the outcomes assessed. For intermediate to long-term outcomes, analysis of follow-up duration showed a trend toward equal risk with longer-term follow-up (>40 months) for only a few outcome measures, suggesting that the 2 strategies may be equivalent with longer follow-up. However, this was not observed for any stroke or death or for any stroke. Our analysis for outcomes excluding periprocedural outcomes suggests that the 2 procedures have similar risks outside the periprocedural period. This is likely due to the low event rate beyond the periprocedural period. Continued emphasis on reducing the periprocedural outcomes (as has been done more recently) is therefore more appropriate.

Our meta-analysis raises a number of important issues. Despite our findings, CAS is likely to be complementary to CEA (given the advantages of being less invasive, having a shorter recovery period, etc), especially in those who are poor surgical candidates or are at elevated risk for periprocedural MI. There is an urgent need to develop risk scores to select participants who have a low risk of periprocedural complications following CAS. Although randomized trials account for observed differences in baseline variables, unmeasured confounders may be missed. The question of the effect of relatively inexperienced operators performing stenting vs well-experienced surgeons performing CEA has not been resolved.³⁵ Moreover, the outcome measures used need more consideration with inclusion of cranial nerve palsies in the composite outcomes and perhaps only including disabling stroke. From a patient perspective, cranial nerve palsy is likely similar to a minor stroke. However, the trials available to date did not report these composite outcomes consistently, and these should be evaluated in a meta-analysis of individual patient data. There is also a need for innovation to prevent embolization during carotid stenting procedures, perhaps with proximal EPDs. In addition, the effects of carotid artery anatomy (bifurcation vs nonbifurcation) and lesion characteristics (calcified vs noncalcified, ulcerated vs not, thrombotic vs not), which are important factors for selection of patients in the real world, on the outcomes of these techniques have not been consistently reported in these trials. The questions of operator experience, patient selection, and optimal EPD use can be resolved only by a randomized trial designed to address these issues.

As in other meta-analyses, given the lack of data in each trial, we did not adjust our analyses for medications used during and following the procedure. Although detailed sensitivity analyses on many variables were undertaken, given heterogeneity in the study protocols, clinically relevant differences could have been missed and are best assessed in a meta-analysis of individual patient data. All of the trials did not report all of the outcomes. The subgroup analyses might suffer from multiple testing. The results of the sensitivity analyses are best described as secondary and hypothesis generating only. Of note, for the SAPPHIRE-like composite outcome, studies like the Stent-Protected Angioplasty vs Carotid Endarterectomy trial did not routinely measure periprocedural enzymes or electrocardiograms; hence, the periprocedural MI rate may be underestimated.

In this largest and most comprehensive meta-analysis of the available evidence from randomized trials to date, using outcomes considered the current standard for these trials, CAS was associated with a significant increase in the risk of short- and long-term outcomes compared with CEA. This was confirmed by a TSA in which the monitoring boundaries were crossed, suggesting an RRI of at least 20% with stenting. However, CAS was associated with a significant reduction of periprocedural MI or cranial nerve palsies. Thus, there is a need for identifying subsets of participants who are at low risk with CAS.

Correspondence: Deepak L. Bhatt, MD, MPH, VA Boston Healthcare System, 1400 VFW Pkwy, Boston, MA 02132 (dlbhattmd@post.harvard.edu).

Accepted for Publication: August 2, 2010.

Published Online: October 11, 2010. doi:10.1001/archneurol.2010.262

Author Contributions: Dr Bangalore had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Bangalore. Acquisition of data: Bangalore, Kumar, and Cutlip. Analysis and interpretation of data: Bangalore, Kumar, Wetterslev, Bavry, Gluud, and Bhatt. Drafting of the manuscript: Bangalore, Kumar, and Bavry. Critical revision of the manuscript for important intellectual content: Bangalore, Wetterslev, Gluud, Cutlip, and Bhatt. Statistical analysis: Bangalore, Wetterslev, Bavry, Gluud, and Bhatt. Administrative, technical, and material support: Wetterslev. Study supervision: Bangalore, Wetterslev, and Bhatt.

Financial Disclosure: Dr Bhatt has received research grants from AstraZeneca, Bristol-Myers Squibb, Eisai, Ethicon, Heartscape, Sanofi Aventis, and The Medicines Company.

Online-Only Material: The eFigures are available at http://www.archneurol.com.

Additional Contributions: Peter Ringleb, MD, and Markus Steinbauer, MD, provided clarification of their data.