Temporal Changes in Periprocedural Events in the Carotid Revascularization Endarterectomy Versus Stenting Trial
Background and Purpose—Post-hoc, we hypothesized that over the recruitment period of the Carotid Revascularization Endarterectomy Versus Stenting Trial (CREST), increasing experience and improved patient selection with carotid stenting, and to a lesser extent, carotid endarterectomy would contribute to lower periprocedural event rates.
Methods—Three study periods with approximately the same number of patients were defined to span recruitment. Composite and individual rates of periprocedural stroke, myocardial infarction, and death rate were calculated separately by treatment assignment (carotid stenting/carotid endarterectomy). Temporal changes in unadjusted event rates, and rates after adjustment for temporal changes in patient characteristics, were assessed.
Results—For patients randomized to carotid stenting, there was no significant temporal change in the unadjusted composite rates that declined from 6.2% in the first period, to 4.9% in the second, and 4.6% in the third (P=0.28). Adjustment for patient characteristics attenuated the rates to 6.0%, 5.9%, and 5.6% (P=0.85). For carotid endarterectomy–randomized patients, both the composite and the combined stroke and death outcome decreased between periods 1 and 2 and then increased in period 3.
Conclusions—The hypothesized temporal reduction of stroke+death events for carotid stenting–treated patients was not observed. Further adjustment for changes in patient characteristics between periods, including the addition of asymptomatic patients and a >50% decrease in proportion of octogenarians enrolled, resulted in practically identical rates.
Periprocedural stroke and death event rates associated with carotid endarterectomy (CEA) in symptomatic patients enrolled in multicenter clinical trials have declined from >5.8% reported in 1991 by the North American Symptomatic Carotid Endarterectomy Trial (NASCET)1 to rates of 3.9% reported in 2006 by the Endarterectomy Versus Angioplasty in Patients With Symptomatic Severe Carotid Stenosis Trial (EVA-3S),2 3.4% reported in 2010 by the International Carotid Stenting Study (ICSS),3 and 3.2% reported in 2010 by the Carotid Revascularization Endarterectomy Versus Stenting Trial (CREST).4 Likewise, periprocedural stroke and death rates in asymptomatic patients treated with CEA have declined from 2.3% reported by the Asymptomatic Carotid Atherosclerosis Study (ACAS)5 and 3.1% reported by the Asymptomatic Carotid Surgery Trial (ACST)6 to 1.4% reported in 2010 by CREST.4 Similar declines in stroke and death rates have occurred in symptomatic patients treated with carotid stenting (CAS), where rates of 12.1% reported in 2001 in the WALLSTENT study7 and 10.0% reported in 2001 by the Carotid and Vertebral Artery Transluminal Angioplasty Study (CAVATAS)8 have declined to 6.0% reported by CREST in 2010.4 Four primary factors could be contributing to these declines: (1) improvements in techniques contributing to a lowering of the periprocedural risk, (2) improvements in the design and construction of stents, (3) improved patient selection for appropriate candidates, or (4) better periprocedural medical management.
The recruitment period for CREST spanned from December 2000 to July 2008,4,9 encompassing much of this period of declining event rates. When CREST began, CAS was a relatively new procedure with few experienced operators. In comparison, CEA was an established procedure with decades of substantial experience and training to identify, and either avoid or manage, patients at high risk for periprocedural events. In an attempt to compensate for operator inexperience in the CAS procedure, in CREST, a lead-in CAS registry was used to provide insights into the ability for operators to provide high-quality treatment with low complication rates.10,11 Regardless, many physicians performing CAS in the study were relatively inexperienced with both technical and patient selection components of the procedure. Accordingly, it was hypothesized that over time improvements in CAS would be associated with larger reductions in periprocedural events than for CEA.
This study aims to formally assess if there were temporal changes in the composite and individual rates of periprocedural stroke, myocardial infarction (MI), and death for patients randomized to CAS and CEA over the recruitment period in CREST. In addition, we will assess if the changes in the procedural event rates were attributable to temporal changes in the characteristics of the patients recruited into the CREST trial.
Details of the design and primary results of CREST have been reported.4,9 Participants were enrolled from December 2000 to July 2008 at 117 clinical centers in the United States and Canada. The protocol was approved by the institutional research/ethics review boards at participating sites. All participants provided signed informed consent. Although the primary end point for CREST was the composite of any stroke, MI, or death during the periprocedural period and ipsilateral stroke within 4 years after randomization, in this analysis we focus only on the events during the periprocedural period. The definition of stroke was an acute neurological event with focal symptoms and signs, lasting ≥24 hours, consistent with focal cerebral ischemia. MI was defined by a creatinine kinase MB or troponin level twice the upper limit of normal range or higher according to the site’s laboratory plus either chest pain or symptoms consistent with ischemia or ECG evidence of ischemia, including elevation of >1 mm in ≥2 contiguous leads according to the core laboratory or ST-segment depression. For those where the procedure was performed within 30 days, the definition of the periprocedural period was defined as the period from randomization through 30 days after the procedure. If the procedure was not performed within 30 days after randomization, the periprocedural period was defined as the period from randomization through 36 days after randomization (6 additional days were added as this was the average length of time between randomization and the procedure). Stroke and MI were adjudicated by specialty committees masked to treatment assignment.
Analyses to assess temporal changes in the event rates were complicated by several factors. There was an initial slow recruitment rate in the period between 2000 and 2005, requiring making the first temporal period include this entire 6-year period. In addition, there was a protocol change in 2005 permitting the inclusion of asymptomatic patients (Figure). For this analysis, event rates (composite, stroke and death, and MI) were calculated in 3 temporal periods defined by tertile of date of recruitment for the symptomatic patients: period 1 was January 2001 through January 3, 2006, period 2 from January 4, 2006 through March 19, 2007, and period 3 from March 20, 2007 through July 2008 (Figure). This approach was adopted (1) to have a clearly defined early period (ie, largely before 2006) and a late period (largely after first quarter of 2007), so that if there were reductions in event rates present, they could be more easily detected; and (2) to have as large a sample size as possible within each period to reduce noise in the estimation process. Differences in event rates were assessed using both an intention-to-treat and per-protocol populations. Although the research question of whether there were temporal changes in event rates represents an analysis not specified during the protocol development (ie, a post hoc hypothesis), the determination of the statistical approach using these 3 periods was an a priori design where we specified the analytic approach before examining the data. Additional analyses using different time periods and different study populations (such as per-treatment received, ie, per-protocol) are provided in the online-only Data Supplement.
For both the intention-to-treat analysis and the per-protocol analysis, for each treatment group, within each temporal period, the unadjusted event rate and exact 95% confidence intervals were calculated. The focus of analysis was whether there was a significant change over time in event rates, assessed by Poisson regression models using a test of linear trend across these estimated crude rates for the 3 periods. To remove the influence of temporal changes in the characteristics of the patients recruited to CREST, additional analyses assessed the change in even rates after adjustment for temporal changes in the patient profile. Poisson regression was chosen as the analysis approach because it allowed the estimation of these adjusted event rates that could be compared with unadjusted rates. Changes over time in the distribution of demographic factors (age and sex) and major risk factors (as defined by medical record: previous coronary artery bypass surgery, diabetes mellitus, hypertension, cigarette smoking, and dyslipidemia) were assessed between periods using the χ2 test (for the categorical variables) and ANOVA (for age in years). Because the number of events was relatively small, patient characteristics that had substantial temporal changes were identified and only these factors were included in the Poisson models. Finally, CAS-versus-CEA differences in the risk factor–adjusted event rates were assessed using the same Poisson regression models.
Of the 2502 participants randomized, 832 were recruited in period 1, 839 in period 2, and 831 in period 3 (Figure). A description of the study population is provided in Table 1. There was a significant decrease in the age of patients enrolled in the study over the recruitment periods (P=0.0007), with an average age of 69.9 years and 14.1% octogenarians during period 1, when compared with an average age of 68.2 and only 6.7% octogenarians for period 3. Because of the inclusion of asymptomatic patients in 2005, there was an increase in the proportion of asymptomatic subjects in the latter parts of the recruitment period (P<0.0001). There was an increase in the proportion of patients with dyslipidemia, increasing from 80.0% in period 1 to 86.5% in period 3 (P=0.0002) and marginal evidence of an increase in the proportion of male subjects, increasing from 64.4% to 68.0% (P=0.092). On the basis of these findings, event rates in subsequent Poisson regression models were adjusted for age, sex, symptomatic status, and dyslipidemia.
Table 2 provides the number of events and event rates for patients randomized to CAS (n=1262) and CEA (n=1240) for the 3 periods (ie, intention-to-treat population). For the composite outcome, for those randomized to CAS, the period 1 composite event rate was 6.2% (95% confidence interval [CI], 4.1%–9.0%), with a nonsignificant (P=0.28) decrease to 4.9% (95% CI, 3.1%–7.4%) and 4.6% (95% CI, 2.8%–7.0%) in periods 2 and 3. Adjustment for the 4 covariates substantially mediated the decrease, with adjusted event rates relatively unchanged across the periods: 6.0% (95% CI, 3.9%–9.3%), 5.9% (95% CI, 3.6%–9.7%), and 5.6% (95% CI, 3.3%–9.3%).
For the combined stroke and death outcome (Table 2), there was a similar nonsignificantly decreasing trend for the patients randomized to CAS (P=0.18), with event rates of 5.5% (95% CI, 3.5%–8.1%), 4.0% (95% CI, 2.3%–6.3%), and 3.6% (95% CI, 2.0%–5.9%). Similarly, covariate adjustment substantially mediated the decrease with rates of 5.2% (95% CI, 3.2%–8.4%), 5.1% (95% CI, 2.9%–9.0%), and 4.7% (95% CI, 2.6%–8.4%).
Table 2 also shows the temporal pattern of event rates for the patients randomized to CEA to be less consistent but did not show a trend for change in event rates either unadjusted (P=0.59) or after adjustment with covariates (P=0.82), with a composite end point event rate of 5.8% (95% CI, 3.8%–8.5%) in period 1, a decrease to 2.7% (95% CI, 1.3%–4.7%) in period 2, and an increase back to 5.1% (95% CI, 3.2%–7.6%) in period 3. Covariate adjustment did not substantially affect this pattern with rates of 5.7% (95% CI, 3.6%–8.9%), 3.2% (95% CI, 1.7%–6.0%), and 6.2% (95% CI, 3.8%–10.1%). Similar patterns were observed for the stroke–death outcome and MI outcome.
Analysis of CAS patients who received their assigned treatment within 30 days of randomization (ie, per-protocol population) demonstrated a nonsignificant trend toward a decreasing pattern of adverse outcomes (P=0.38; Table 3). In period 1, the combined stroke and death event rate was 5.7% (95% CI, 3.6%–8.6%), period 2 was 4.2% (95% CI, 2.4%–6.7%), and period 3 was 3.5% (95% CI, 1.9%–5.9). This represents an absolute 2.2% (relative 39%) reduction in stroke and death events. The pattern of temporal changes in the per-protocol analysis of the CEA-treated patients (Table 3) was similar to that observed in the intention-to-treat analysis (Table 2).
Analysis by symptomatic status demonstrated a nonsignificant (P=0.44) pattern of decline for asymptomatic patients randomized to CAS, the stroke, and death rate was 3.7% (95% CI, 1.0%–9.2%) in the first third of the study and 2.1% (95% CI, 0.7%–4.9%) in the final third (Table 4). This downward trend in event rates remained somewhat persistent after adjustment (for age, sex, symptomatic status, and dyslipidemia) with rates of 2.7% (95% CI, 0.9%–9.3%), 1.9% (95% CI, 0.7%–4.7%) and 1.7% (95% CI, 0.7%–4.5%). In contrast, event rates were stable (P=0.80) for CAS-treated symptomatic patients; 6.1% (95% CI, 3.7%–9.4%), 6.2% (95% CI, 3.1%–10.8%), and 5.6% (2.7%–10.0%). After adjustment for covariates, these rates were virtually identical (P=0.90): 5.0% (95% CI, 2.9%–8.5%), 5.6% (95% CI, 3.0%–10.8%), and 5.1% (2.6%–10.0%).
Supplemental analysis supported these observations when reported by study periods defined by calendar year (2003–2004, 2005–2006, and 2007–2008) rather than tertile of recruitment (Table I in the online-only Data Supplement). In addition, we hypothesized that sites admitted to the study early could have potentially had greater CAS experience, and as such, we organized the sites by their order of inclusion in the study. However, analysis is restricted to the first 40 sites admitted to the study (Table II in the online-only Data Supplement). Finally, to address the concern that more experienced operators were included in the initial period of the study, with less experienced sites included later in the study, an analysis was done by wave (first 40 sites, versus second 40 sites, versus last 39 sites) of sites admitted to the study (Table III in the online-only Data Supplement). This analysis failed to show lower rates for the more experienced sites admitted to the trial first.
We had hypothesized that within the CAS group, periprocedural event rates would decline over time because of the increasing technical and procedural experience of the physicians performing CAS. Although there were nominal temporal declines in the crude stroke+death (and composite end point), these declines failed to reach a level of statistical significance. Hence, the hypothesis of a declining periprocedural stroke+death risk in CAS-treated patients over the CREST recruitment period was not supported by these data. There was no decline in the rate of MI for CAS and an uneven trend for CEA.
There were also substantial changes in the composition of the patient cohort over the study period. Because of changes in the protocol made in 2005 to admit asymptomatic patients, the last 2 periods included a larger proportion of asymptomatic patients. It is well documented that periprocedural event rates for CEA-treated patients are lower in asymptomatic than in symptomatic patients.12 CREST has previously reported that this is true for both CAS-treated and CEA-treated patients.4 As such, the inclusion of a higher proportion of asymptomatic patients should result in lower event rates for both treatment groups.
We also observed that the average age of the patients decreased by 1.5 years over the recruitment period, a difference that was associated with a decrease in the proportions of octogenarians from 14.1% in the first period to only 6.7% in the third. This decline could have been attributable to a growing knowledge of the higher risk from stenting in the elderly based on results of the CREST lead-in13 and other sources.14 Subsequent data from the randomized CREST series have also shown that procedures in the elderly are associated with higher composite and stroke and death risk for CAS-treated patients, but there is no similar increase in risk for CEA-treated patients.10,11,15 As such, a reduction in the representation of elderly patients should be associated with a declining number of events among the CAS-treated, but not in the CEA-treated, patients.
We observed increased representation of men as the study progressed, where in the first period, there were 64.4% men and the latter period there was 68.0% men. Although this minor increase in the number of men would be unlikely to have a large effect on event rates, our a priori analysis plan was to adjust for all covariates that showed significant temporal change over time. We have shown that for CAS-treated patients, there is a lower periprocedural risk for men compared with women, but no differences between men and women treated with CEA.16 As such, the lower proportion of women in the latter recruitment periods should reduce the risk of events in patients treated with CAS but not in patients treated with CEA.
Finally, the initial sites included in the study were more experienced, whereas less experienced sites admitted later to the study had to establish their experience by enrolling patients in the CREST Lead-in Registry.11,17 We were concerned that the event rates could be higher for these less experienced sites; however, an analysis separating sites into those admitted early, middle, and late in the study failed to detect significant change (Table III in the Data Supplement).
In summary, the observed changes in the study population should have resulted in lower event rates over time in CAS-treated and CEA-treated patients. However, significant declines were not observed, and the changes in the population attenuated the nominal decline in rates for the CAS-treated patients that were observed: the adjusted stroke+death event rates in patients treated with CAS were nearly identical in the 3 temporal periods. This attenuation of the nominal decline associated with the adjustment for patient characteristics suggests any potential decline in stroke+death risk in CAS-treated patients is largely associated with changes in the population treated rather than improvements in the procedure itself. The pattern for CEA-treated asymptomatic patients showed a decrease-then-increasing pattern. We do not have a good explanation for this pattern and speculate that it is only a chance pattern.
There are many strengths and weaknesses of this report. CREST is the largest randomized trial comparing CAS and CEA, making the stratification by recruitment period possible. However, for regulatory reasons, the single embolic protection/stent system approved for use in CREST remained static over the 2000 to 2008 time period. During this time, numerous second generation, lower profile, and more user friendly devices were approved by the Food and Drug Administration but not incorporated into CREST. Accordingly, the apparent absence of technical improvement as opposed to improvement associated with experience may not necessarily be transferrable to the community at large where incorporation of second generation devices and techniques has been widespread. Recruitment was slow during the earliest years the study was active, for example, there were only 136 patients recruited between January 2001 and December of 2003 (Figure). This slow recruitment makes it difficult to conduct comparisons before 2003, and forced the first tertile to include a longer period than the subsequent 2 tertiles (Figure). In addition, the inclusion of the asymptomatic patients in 2005 resulted in lower event rates after 2005, and this reduction in events limited the statistical power to detect differences between periods. Importantly, the detection of temporal changes in procedure risk may have also been limited by the expansion of CREST centers, from the original 40 to 117 because interventionists of differing experience were credentialed into CREST throughout the enrollment period. Overall, the number of stroke and death events was small, and accordingly, subgroup analyses were limited and must be interpreted with caution.
In conclusion, we hypothesized that the temporal decline in periprocedural risk for CAS-randomized patients would be more rapid than for CEA-randomized patients. We observed only a nominal and nonsignificant decrease in the crude event rates for CAS-randomized patients. Further adjustment for changes in patient characteristics between periods, including the addition of asymptomatic patients and a >50% decrease in proportion of octogenarians enrolled, resulted in practically identical rates.
Sources of Funding
The study was supported by the National Institute of Neurological Disorders and Stroke of the National Institutes of Health (U01 NS 038384) with supplemental funding from Abbott Vascular, Inc.
L.N. Hopkins received grants/research support from Toshiba; he has ownership interest in Boston Scientific, Valor Medical, Claret Medical Inc, Augmenix, Endomation, Silk Road, Ostial, Apama, StimSox, Photolitec, ValenTx, Ellipse, Axtria, Nextplain and MedinaMed; he is an advisory board member for Claret Medical, Inc; he is a member of speaker’s bureau for Abbott Vascular and Toshiba; he received honoraria from Cordis, Memorial Healthcare System, Complete Conf. Management and Covidien. B.T. Katzen is an advisory board member for WL Gore, Medtronic Vascular, Boston Scientific, The Medicines Company. E. Chakhtoura is an advisory board member for Abbott Vascular. The other authors report no conflicts.
Guest Editor for this article was Markku Kaste, MD, PhD.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.115.008898/-/DC1.
- Received April 20, 2015.
- Revision received June 8, 2015.
- Accepted June 10, 2015.
- © 2015 American Heart Association, Inc.
- Ederle J,
- Dobson J,
- Featherstone RL,
- Bonati LH,
- van der,
- Worp HB,
- de Borst GJ,
- et al
- Halliday A,
- Mansfield A,
- Marro J,
- Peto C,
- Peto R,
- Potter J,
- et al
- Alberts MJ.
- Hopkins LN,
- Roubin GS,
- Chakhtoura EY,
- Gray WA,
- Ferguson RD,
- Katzen BT,
- et al
- Rothwell PM,
- Slattery J,
- Warlow CP.
- Hobson RW II,
- Howard VJ,
- Roubin GS,
- Brott TG,
- Ferguson RD,
- Popma JJ,
- et al
- Voeks JH,
- Howard G,
- Roubin GS,
- Malas MB,
- Cohen DJ,
- Sternbergh WC III,
- et al
- Howard VJ,
- Voeks JH,
- Lutsep HL,
- Mackey A,
- Milot G,
- Sam AD II,
- et al
- Howard G,
- Voeks JH,
- Meschia JF,
- Howard VJ,
- Brott TG.