|Home | About | Journals | Submit | Contact Us | Français|
Carotid arterial stent systems (CAS) are an alterative to carotid endarterectomy for the treatment of moderate to severe carotid stenosis, but the effectiveness of CAS compared to endarterectomy in preventing stroke and death is uncertain. This study’s objective was to compare the clinical outcomes among Medicare beneficiaries undergoing carotid revascularization before and after CAS became widely available.
Observational, retrospective cohort study of 46,784 patients undergoing carotid revascularization from August, 2005–March 2006 (the coverage era) compared to propensity-score-matched patients undergoing carotid revascularization between October, 2002–September, 2004 (the pre-coverage era), prior to widespread Medicare coverage of CAS.
Mortality at 90 and 270 days after revascularization, as well as the combined outcomes of peri-procedural acute myocardial infarction and any stroke or death within 90 and 270 days after revascularization, between the two eras. Comparisons were also made between localities with high (23% of carotid procedures being CAS) and lower (9% of carotid procedures being CAS) adoption rates of carotid stents during the coverage era.
There were no significant differences in 90-day mortality (2.2% vs. 2.2%, P=0.79), 90-day combined outcomes (4.5% vs. 4.3%, P=0.13), or 270-day mortality (4.8% vs. 4.6%, P=0.17) between the coverage and pre-coverage eras, but there were more 270-day combined outcomes in the coverage era (7.7% vs. 7.3%, P=0.03). In localities with higher adoption of carotid stents, there was higher 90-day mortality (adjusted odds ratio [OR] 1.15, P=0.16), 90-day combined outcomes (OR=1.17, P=0.03), 270-day mortality (OR=1.13, P=0.07), and 270-day combined outcomes (OR=1.10, P=0.09) in the coverage era. There were no differences in event rates between eras in areas with lower carotid stent adoption.
The adoption of carotid stents for treatment of carotid stenosis was associated with increased rates of adverse clinical outcomes following carotid revascularization.
Carotid endarterectomy (CEA) reduces the incidence of stroke and death among patients with symptomatic carotid arterial stenosis,1, 2 yet the perioperative complication rate is not insignificant,3 and the American Academy of Neurology recommends endarterectomy be avoided in patients with >6% risk of perioperative complications or with an expected survival of <5 years.4 Carotid arterial stenting with cerebral protection (CAS) was approved by the Food and Drug Administration (FDA) in 2004 as an alternative therapy to endarterectomy among patients at high risk for perioperative complications.5 Clinical trials comparing CAS to carotid endarterectomy (CEA) have yielded conflicting results.6–11 Two ongoing, large multicenter randomized controlled trials comparing endarterectomy to CAS—the Carotid Revascularization Endarterectomy Versus Stenting Trial (CREST) and the International Carotid Stenting Study (ICSS)—are completing enrollment and may further establish whether CAS efficacy is comparable to endarterectomy.12, 13
Nevertheless, it remains uncertain how CAS performs in comparison to CEA outside the context of clinical trials. Prior investigations of carotid revascularization procedures among the Medicare population have revealed important differences in the clinical outcomes achieved by hospitals that were involved in the clinical trials compared to hospitals that did not participate in clinical trials.14 Recognizing the variability in hospitals’ ability to deliver optimal outcomes from carotid revascularization, the Centers for Medicare and Medicaid Services (CMS) reimburses CAS only for hospitals that meet qualifying requirements (e.g., prior procedural experience, appropriate ancillary services, adequate case volume, etc.).15 It is uncertain whether this coverage policy resulted in comparable CEA and CAS clinical outcomes.
The goal of our research was to compare the effectiveness of CAS to CEA in a non-experimental, nationwide population of Medicare beneficiaries receiving carotid revascularization following the March, 2005 Medicare coverage expansion for CAS.
Observational studies comparing CEA to CAS are complicated by the strong selection effects inherent in the non-randomized assignment of patients to therapy. Direct comparisons between CEA and CAS recipients are likely to be confounded by factors that are unobservable, even with highly detailed clinical databases. To surmount this difficulty, we assessed the aggregate clinical outcomes of patients receiving either CAS or CEA within small geographic regions. From 2005–2006, the percentage of all carotid revascularizations that were CAS varied across localities from 0–25%. Had there been differences in the effectiveness of CEA versus CAS, there should have likewise been differences in clinical outcomes among carotid procedure patients across regions, with adverse event rates correlated with CAS adoption. A similar approach has been used in other studies utilizing observational data.11
To control for the possibility that high-CAS utilization areas were systematically different from low-CAS utilization area in health care quality or patient characteristics, we compared outcomes within localities between two distinct eras (i.e., the era immediately prior to Medicare coverage of CAS, and the era subsequent to wider Medicare coverage of CAS—Figure 1), to control for time-invariant differences in carotid procedural outcomes across regions. This analytic technique of comparing the temporal differences in outcomes within fixed spatial units is known as “difference-in-difference” estimation.16
We examined fee-for-service Medicare claims for patients receiving either CEA or CAS during two “eras.” The “pre-coverage” era was defined as October 1, 2002–September, 30, 2004, when CAS was only covered for Medicare beneficiaries enrolled in clinical trials. The “coverage era” was defined as August 1, 2005 through March 31, 2006, after the CMS implemented a national coverage decision expanding eligibility for CAS (March 17, 2005). Carotid stent utilization was determined by International Classification of Diseases, 9th revision procedure codes 00.55 or 39.90 (prior to 10/1/2004) or codes 0.061 and 0.063 (starting on 10/1/2004) with concurrent diagnosis codes 433.10, 433.11, 433.30, or 433.31. Endarterectomy was determined by procedure code 38.12. We included only patients who were older than age 66 to provide at least 1 year of prior claims for each patient. We excluded patients who received carotid revascularization outside of the 50 states or the District of Columbia, or who had received prior carotid revascularization within 12 months. Due to coding limitations, we could not assess whether patients had symptomatic carotid disease at the time of revascularization, nor could we assess medication use.
Demographic information was obtained from Medicare’s enrollment database. Clinical comorbidities among carotid revascularization patients was determined by examining diagnosis codes included in the claim in which the carotid procedure was performed, as well as all other inpatient and outpatient claims in the 12 months preceding the carotid procedure. Information about the hospital performing the carotid procedure was obtained from Medicare Hospital Cost Report Information System annual reports. Members of the American Association of Medical College’s Council of Teaching Hospitals and Health Systems were designated as academic hospitals. Hospitals located in metropolitan statistical areas with > 1 million inhabitants were designated as urban hospitals.17
We selected the Dartmouth Atlas for Healthcare’s Hospital Referral Region (HRRs) as the geographic unit of analysis, as the availability of CAS in 2005–2006 varied markedly across HRRs.18 These 306 contiguous geographic units were originally defined as localities within which Medicare beneficiaries receive the vast majority of their hospital care. We examined the ratio of CAS procedures to all carotid revascularization (CAS+CEA) procedures among patients in each HRR in both the pre-coverage and coverage eras. We calculated the percentage point increase in CAS use in each HRR between eras as a measure of the extent of CAS adoption in each locality. HRRs were divided into quintiles of HRR adoption rates during the coverage era.
In order to further decrease the bias introduced by comparing outcomes among carotid revascularization recipients between eras, we used propensity scores19, 20 to match carotid revascularization recipients within HRRs across eras. Matching patients within HRRs assured balance of time-invariant HRR-level factors across comparison groups. We thus fitted logistic regression models with pre-coverage versus coverage era as the dependent variable, and age, race, sex, clinical comorbidities (Elixhauser method),21 and hospital characteristics (academic center, urban location, cardiovascular volume) as independent variables. We then matched pre-coverage and coverage era patients within each HRR, using the closest Mahalanobis match within propensity score calipers, permitting pre-coverage era patients to control for more than one coverage era patient.19, 20 In our primary analysis we purposefully did not match patients based on their actual receipt of CAS versus CEA to avoid introducing the confounding-by-indication bias described above.
Clinical outcomes were assessed using Medicare claims in the 270 days after procedure receipt. Medicare’s “Denominator” file was used to assess mortality—death records are cross-linked to the Social Security Death Master File, and a high level of accuracy in mortality reporting has been confirmed. The occurrence of acute myocardial infarction (AMI) was determined by the appearance of diagnosis code 410.xx, and the occurrence of stroke was determined by the appearance of diagnosis codes 433.01, 433.11, 433.21 433.31, 433.81, 433.91, 434.01, 434.11, 434.91, or 436. Following a formula for combining clinical outcomes often used in carotid revascularization clinical trials,6 we combined outcomes of any stroke (claims do not indicate laterality, thus we included all strokes rather than ipsilateral stroke) or death after the revascularization procedure, or any AMI within 30 days of the procedure.6 We calculated 90-day and 270-day mortality and combined adverse outcomes across quintiles of CAS adoption rates. We then estimated a logistic regression model, using generalized estimating equations to account for matched observations, with 90-day mortality as the dependent variable and propensity score, treatment in the coverage era, treatment in a high-CAS-adoption HRR, and the interaction between coverage era and high-CAS-adoption HRR as independent variables. We estimated similar logistic regression models for 90-day combined adverse events (death, stroke, or AMI), 270-day mortality, and 270-day combined adverse events. All analyses were performed with SAS 9.2 (Cary, NC) or STATA 10.1 (College Station, TX).
Among Medicare beneficiaries CAS use nearly quadrupled between early 2004 (266 procedures/month) and late 2006 (1,015 procedures/month, p<.001) while carotid endarterectomy utilization dropped during the same periods from 7,820 procedures/month to 6,519 procedures/month (p<.001). The overall number of carotid revascularization procedures remained statistically unchanged between October, 2003 and March, 2006 (average change in procedures per month =−17.6, P=.11 for trend). Geographic utilization of CAS was highly variable (Figure 2).
We initially identified 138,885 pre-coverage-era and 47,386 coverage era carotid revascularization patients, with the ratio reflective of the pre-coverage era being 24 months compared to the 8-month coverage era. After implementing the propensity score match, 46,784 coverage era patients (99%) could be matched to 35,778 unique pre-coverage-era patients (Table I), with 8,359 pre-coverage-era patients matching more than one coverage-era patient (average number of matches per multiply-matched pre-coverage patient = 2.3). Comparisons of the standardized difference (i.e., difference in means divided by the pooled standard deviation) between matched groups indicated both patient and hospital characteristics were highly similar.
Patients in the matched cohorts were predominantly white and nearly 60% were men. Few patients (<4%) had a prior diagnosis of stroke or transient ischemic attach, although nearly half of the patients had coronary artery disease, 25% had diabetes, 15–16% had non-carotid peripheral vascular disease, and 19–21% had chronic pulmonary disease. Approximately 40–50% of patients were treated at urban hospitals, and 18–21% of patients were treated at academic hospitals.
Overall rates of death among pre-coverage and coverage era patients were comparable at both 90 days (pre-coverage: 2.2%, coverage: 2.2%, P=.79) and at 270 days (pre-coverage: 4.6%, coverage: 4.8%, P=.17) subsequent to carotid revascularization. Rates of combined outcomes (i.e., early AMI, any stroke/death) were similar in the two cohorts at 90 days (pre-coverage: 4.3%, coverage: 4.5%, P=.13), but at 270 days but there was a significantly higher rate of the combined outcomes in the coverage era (pre-coverage: 7.3%, coverage: 7.7%, P=.03).
We then calculated the CAS rate in each HRR during the pre-coverage and coverage eras, and we divided the matched cohorts across quintiles of HRR adoption rates for CAS. Utilization rates of CAS across these 5 quintiles ranged from 1% in the lowest quintile to 23% in the highest quintile (Table II). There were no significant differences between unadjusted rates of pre-coverage-era and coverage era mortality or combined adverse clinical outcomes in the lower 4 tertiles for CAS adoption. In the highest tertile for CAS adoption, there were higher rates of 90-day mortality (rate ratio[RR]=1.08, P=.16), 270-day mortality (RR=1.07, P=.06), 90-day combined outcomes (RR=1.09, P=.03) and 270-day combined outcomes (RR=1.05, P=.09).
We then separately fit logistic regression models predicting 90-day and 270-day mortality, as well as the 90-day and 270-day combined adverse outcomes. Each model contained a predictor variable indicating the coverage era, a predictor indicating whether the patient was in a high-CAS-adopting HRR (i.e., the top quintile of HRRs for CAS adoption), an interaction term between these predictors, and covariates for which the standardized difference in the matched cohorts exceeded 2%. We observed minimal differences between adjusted pre-coverage-era and coverage era outcomes among patients who were not in the highest CAS-adoption HRR quintiles, with odds ratios ranging from 0.96 to 1.03 (all P-values>.33, Figure 3). Conversely, adjusted odds ratios for adverse outcomes in the coverage era in the highest CAS-adoption HRR quintile were higher, with the odds ratio (95% confidence interval) for 90-day mortality = 1.15 (0.95 to 1.41, P=.15), 270-day mortality = 1.13 (0.99 to 1.30, P=.07), 90-day combined adverse outcome = 1.17 (1.02 to 1.35, P=.03), and 270-day combined adverse outcome = 1.10 (0.98 to 1.23, P=.09).
Qualitatively similar results were obtained when we selected different quantiles (e.g., quartiles, deciles, etc.) for dividing patients geographically. Our results also did not substantially change when additional covariates were added to the regression models.
To compare actual outcomes among CAS and CEA patients, we performed an additional propensity-score match of CAS to CEA patients in the coverage era. The model predicting CAS receipt included age, race, sex, prior stroke/transient ischemic attack, presence of bilateral carotid disease, congestive heart failure, coronary artery disease, non-carotid peripheral vascular disease, valvular heart disease, hypertension, chronic pulmonary disease, diabetes, and renal disease. Overlap of propensity scores between the two cohorts was 87% (Figure 4), indicating only modest observable differences between patient populations. Unsurprisingly, CAS recipients had much higher adjusted 90-day (2.7% versus 1.5%, P<.001) and 270-day (6.5% versus 4.2%, P<.001) mortality, as well as higher adjusted rates of combined outcomes at 90 days (5.7% versus 4.0%, P<.001) and 270 days (10.6% versus 7.6%, P<.001).
In a matched analysis of patients receiving carotid revascularization prior to versus subsequent to widespread CAS coverage, we observed higher mortality and adverse clinical outcomes during the coverage era. Across localities, higher rates of CAS adoption were associated with higher rates of adverse clinical outcomes. While some differences were not statistically significant at the.05 level, the magnitude and direction of the differences remained consistent even after multivariable statistical adjustment.
Carotid stent systems with embolic protection extend the clinical population that may benefit from revascularization due to the lower periprocedural risk of catheter-based therapy compared to surgery. Nevertheless, were carotid stents to have worse downstream clinical outcomes than endarterectomy, their up-front clinical benefit could be nullified, making stents a suboptimal therapeutic choice. Among several potential explanations for our findings is the possibility that CAS patients had modestly worse outcomes than comparable CEA patients. While it is also possible that carotid stents were used in the coverage era among patients who would not have undergone carotid endarterectomy in the pre-coverage era, we observed no change in the overall volume of carotid revascularization procedures between the pre-coverage and coverage eras, rather than the increase in total procedures that would have been expected had CAS been predominantly used in patients who previously would not have been surgical candidates. Our propensity score match on multiple potential confounders further reduced the possibility that selection of higher risk patients during the coverage era explained the observed results.
Medicare covered CAS under a more restrictive coverage policy than carotid endarterectomy. Only 635 hospitals used CAS among Medicare beneficiaries during the coverage era compared to over 2000 hospitals that performed CEA. Medicare’s qualification criteria for CAS hospitals may have improved outcomes among patients receiving CAS by essentially requiring CAS patients to obtain care at high-volume centers with extensive procedural experience. Nevertheless, our results suggest that this systematic advantage in the provision of CAS care did not result in comparable CAS-CEA outcomes.
The results of our difference-and-difference models, matching patients across eras, suggest more modest difference in CEA/CAS outcomes than in our direct comparison of CEA to CAS patients in the coverage era. The larger differences observed in the direct comparison likely resulted from patients being selected for CAS during the coverage era who had unobservable risk factors. The benefit of our primary analytic approach, in which we investigated changes in outcomes among the combined group of CAS/CEA recipients across eras, is that the undesirable biases of both therapy selection (so-called “confounding by indication”), as well as time-invariant, unobserved confounders (e.g., the proportion of the CAS/CEA population who smoked) were greatly diminished.
While the results of this observational study are not directly comparable to clinical trials, it is notable that prior studies comparing CAS to CEA have yielded varying results. The SPACE trial demonstrated no difference in 30-day rates of stroke and death among patients who received CAS versus CEA (6.9% versus 6.4%). At two years, the rate of ipsilateral stroke combined with any periprocedural stroke or death was 9.5% for CAS and 8.8% for CEA (P=.62).9 These results were similar to those of the SAPPHIRE trial, in which there were no significant differences in periprocedural stroke or death plus ipsilateral stroke or neurological death at 1 year (5.5% CAS versus 8.4% CEA, P=.36).6 Similarly, the CARESS trial demonstrated no significant difference in combined death, stroke, or AMI at 30 days (3.6% CEA versus 2.1% CAS, P=.41) or at 1 year (13.6% CEA versus 10.0% CAS, P=.30).7 Conversely, the EVA-3S trial was stopped early because of excess 30-day incidence of stroke or death among CAS compared to endarterectomy patients (9.6% CAS versus 3.9% CEA, P=.01).8 As such, the comparative effectiveness of CAS compared to CEA remains controversial, and few studies have used non-experimental data to compare outcomes.
The current study contributes to the CAS-CEA comparative effectiveness evidence base by reporting outcomes among a nationwide cohort of patients who received CAS or CEA under non-experimental conditions. Results of clinical trials may not be generalizable to patient populations that differ in important aspects from trial populations. In particular, the Medicare patients we studied were older (average age = 76) than patients in the SAPPHIRE (average age = 72.5), EVA-3S (average age = 70), or SPACE (average age = 68) trials. Furthermore, our study is more representative of care outside of major academic hospitals—patients in our study received care in over 2000 different U.S. hospitals, while the SAPPHIRE trial (for example) was conducted at 29 major academic centers.
A particular strength of this study is the use of the geographic variability in the adoption of CAS to partially replicate the effect of randomization in a clinical trial. We assumed that a locality’s tendency to use CAS more frequently as a therapeutic option for carotid revascularization was not correlated with time-varying clinical characteristics of populations eligible for carotid revascularization, or in time-varying health system characteristics. Our difference-in-difference analysis adjusted for the residual time-invariant differences in clinical populations and health system characteristics, as did our use of propensity score matching.
Several of our outcomes comparisons resulted in differences with two-sided P-values between .05–.10. While such results have typically been classified in binary fashion as statistically insignificant,22 the consistency of the findings and the occurrence of some results with P-values <.05 suggest that chance alone is unlikely to explain our results.
We were unable to include patients with moderate or severe carotid disease who received medical therapy alone. It is possible that there was a decline in the numbers of such “medical therapy” patients in localities where CAS was rapidly adopted, thus potentially biasing our analyses due to unobserved changes over time in the procedure recipient population. More generally, changes in either carotid procedure populations or health care systems between the pre-coverage and coverage eras may have biased our results if these changes were more common in high-CAS-adopting localities, and if the changes were not “captured” by matching patients across treatment eras within localities.
Our study was also limited by relatively limited use of CAS relative to endarterectomy even in areas that were robust adopters of CAS. Even in the top decile of HRRs for CAS adoption, the overall rate of CAS use during the coverage era was only 27%. It is possible that modest differences in clinical outcomes for CAS compared to CEA would not have been detected, since outcomes would have differed for only a fraction of patients receiving carotid revascularization across eras. Furthermore, the rates of adverse outcomes in this study were uniformly lower for both CEA and CAS recipients compared to clinical trial enrollees. Although this could have been caused by under-recording of clinical outcomes in the Medicare database, it is also possible that the cohort of patients we examined had, on average, less severe carotid arterial disease (e.g., asymptomatic stenoses) than clinical trial enrollees for whom rigorous inclusion criteria were necessarily applied.
There may have been undercoding or miscoding of CAS during the pre-coverage era, however CAS with cerebral protection was only FDA-approved for marketing in the last month of the pre-coverage era. It is likely, therefore, that the vast majority of CAS use in pre-coverage area was in the context of clinical trials, and CMS did reimburse CAS used in clinical trials throughout the pre-coverage era. It is also possible that CAS in the pre-coverage era was conducted without cerebral protection, however this should have resulted in improved outcomes during the coverage era in high-adopting localities due to providers switching to cerebral protection CAS; instead we observed worsened outcomes.
Finally, our results could have been caused by provider inexperience with CAS relative to CEA, although this possibility was lessened somewhat by the terms of Medicare CAS coverage, which required hospitals to certify either prior participation in CAS clinical trials or the establishment of CAS clinical programs with multiple required quality components (e.g. advanced radiological services, interventionalist certification, etc.). Changes over time in the type of specialists performing carotid interventions may likewise have resulted in different clinical outcomes during the coverage era, although we had no method of determining the specialty of CAS interventionalists.
In this observational study of Medicare beneficiaries undergoing carotid revascularization, we detected increases in mortality and in the combined rates of AMI, death, and stroke in areas where carotid stenting was frequent, compared to areas where it was infrequent.
The authors gratefully acknowledge the administrative and design contributions of Janell Olah, MFA, and Mollie Epstein, MA in the production of the Tables and Figures for this manuscript. Dr Groeneveld 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. This research was supported by an unrestricted grant from the Institute for Health Technology Studies, Washington, DC. This project was also funded, in part, under a grant from the Pennsylvania Department of Health, which specifically disclaims responsibility for any analyses, interpretations, or conclusions. Dr Groeneveld was additionally supported by a Research Career Development Award from the Department of Veterans Affairs’ Health Services Research and Development Service, Washington, DC.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
None of the authors have any personal or financial conflicts of interest in regard to this study.