The dramatic reductions in restenosis and repeat revascularization associated with coronary artery drug-eluting stents (DES) compared with their bare-metal stent (BMS) counterparts (1) prompted swift adoption into clinical practice (2). However, reports of late stent thrombosis (3- 4) and higher mortality (5- 6) resulted in release of 2 special Food and Drug Administration (FDA) advisories in 2006 (7- 8) as well as subsequent studies refining event rates (1,6,9- 13). The rarity of late DES complications means that extremely large sample sizes are required to clarify their frequency. Furthermore, the ability to examine rates of lower frequency complications in important patient subgroups is limited in smaller sample sizes (14).

Accordingly, the Agency for Healthcare Research and Quality and the U.S. FDA commissioned the formation of a nationally representative percutaneous coronary intervention (PCI) database to determine the safety and effectiveness of DES and BMS among a contemporary “real-world” cohort. This was accomplished through linkage of the American College of Cardiology–National Cardiovascular Data Registry (ACC-NCDR) with the Centers for Medicare and Medicaid Services national claims database. The resulting analyses will better inform national practice patterns overall and in important patient and lesion-level subgroups.

The national ACC-NCDR CathPCI Registry collects information for patients undergoing PCI procedures. We included all CathPCI patients ≥65 years of age undergoing an inpatient intracoronary stent procedure between January 1, 2004, and December 31, 2006. Patients receiving more than 1 stent type (i.e., both BMS and DES) were excluded (Figure 1). The Duke University Medical Center Institutional Review Board granted a waiver of informed consent and authorization for this study.

Because ACC-NCDR data are limited to a single episode of care we used the research-identifiable Medicare 100% inpatient fee-for-service claims file for longitudinal patient follow-up. The PCI procedure codes (International Classification of Diseases-Ninth Revision-Clinical Modification [ICD-9-CM] procedure codes: 00.66, 36.0x, 37.22, 37.23, and 88.5x, except 88.59) were used to identify potential index procedure matches in the Medicare files, which were then linked to NCDR with indirect identifiers (nonunique fields that when used in combination might identify unique hospital stays) to create unidentified longitudinal profiles and obtain up to 3 years' follow-up. Linking rules used a hierarchy of evidence approach such that rules with the most information were applied before those with less information. Once a match was achieved for a patient, no further rules were applied. Our linking rules contained combinations of information denoting the index PCI procedure site, patient date of birth (or components thereof) or age, admission date, discharge date, and sex. In the rare event that a single ACC-NCDR record could be matched with multiple Medicare records with the same rule, no linking occurred. Sites that did not match to Medicare records were excluded, as were patients whose index PCI procedure did not occur during a period of fee-for-service enrollment.

We evaluated 8 clinical end points: 5 events and 3 composites. Death was the only event defined both during the index PCI procedure (with ACC-NCDR information) and after discharge (with the Medicare denominator file). Clinical end points were defined with the Medicare claims file as the primary diagnosis for a hospital admission. The ICD-9-CM diagnosis codes used to identify events were: myocardial infarction (MI) (410.X1), stroke (430.X, 431.X, 432.X, 434.X), and bleeding (430-432, 578.X, 719.1X, 423.0, 599.7, 626.2, 626.6, 626.8, 627.0, 627.1, 786.3, 784.7, or 459.0). Revascularizations were identified with ICD-9-CM procedure codes (PCI: 36.00, 36.06, 36.07, 36.09; coronary artery bypass grafting [CABG]: 36.10-19). Only revascularizations occurring after discharge from the index hospital stay were included in the revascularization analysis. The composite events used in this study were: MI or death, MI or death or revascularization, and MI or death or stroke.

Baseline and propensity matching characteristics were categorized by stent type (DES vs. BMS) and summarized as counts and percentages for categorical variables and mean with standard deviations for continuous variables. Statistical significance was defined as p ≤ 0.05, with no correction for multiple comparisons, with SAS statistical software (version 9.1, SAS Institute, Cary, North Carolina) for all calculations.

We used propensity scores to adjust for between-treatment group differences in baseline characteristics (15). Propensity scores represent the estimated probabilities of patients receiving DES versus BMS in our population (15), in this case conditioned upon 102 observed covariates (6). Inverse probability weighted (IPW) estimators incorporating propensity scores were used to compare treatment groups (16). The propensity score model had a c-index of 0.690. In addition, the distribution of propensity scores for DES patients closely match those for BMS patients as evidenced by the 5-number summaries (min, 25th, 50th, 75th, max) describing the curves for patients receiving each type of stent: BMS (14.5%, 70.7%, 79.6%, 85.9%, and 99.1%) and DES (16.0%, 79.7%, 86.1%, 90.7%, and 99.5%). The overlap between the groups is excellent and suggests that the propensity score approach is statistically appropriate.

Inverse probability weighted estimators with monthly data partitions were used to calculate cumulative incidence rates for clinical end points (adjusted and unadjusted) (17- 18). Unadjusted estimates were based upon Kaplan-Meier estimates for treatment-specific censoring distributions, whereas adjusted estimates were based upon weights that were functions of Kaplan-Meier censoring estimates and propensity score estimates (17- 18). Adjusted hazard ratios (HRs) were calculated according to the IPW approach of Cole and Hernan (19). In particular, we calculated 2 IPW Cox proportional hazards models—1 with an indicator for DES as the only covariate and 1 with DES plus a selected group of clinically important variables including: sex, age, diabetes, renal disease, prior revascularization, prior MI, multivessel coronary artery disease, year of procedures, and race. From these models, we estimated the adjusted HR for DES versus BMS along with a 95% confidence interval (CI) on the basis of the sandwich estimated standard errors. To visually assess the proportional treatment effect assumptions, we plotted the monthly cumulative incidence rates over the 30-month follow-up period. Additionally, we plotted the treatment-group specific cumulative incidence rates excluding events from the first 6 and 12 months to identify the long-term component of the treatment effect. We refer to these latter analyses as 6- and 12-month landmark analyses.

A Cox proportional hazards mortality model (without propensity score weighting) was developed with backward selection of the propensity score variables with a selection threshold of p = 0.05. Forward selection was used in a sensitivity analysis for internal validation of the final model, which contained 60 covariates. These models served to validate the adjusted HR estimates from the IPW Cox regression model method of Cole and Hernan (19).

The PCI status included ST-segment elevation myocardial infarction (STEMI) (primary, rescue, or facilitated), urgent (non–ST-segment elevation myocardial infarction [NSTEMI] or unstable), and elective subgroups. Within the DES group, off- versus on-label use subgroups were examined. For patients enrolled in NCDR, with version 2 of the data collection form, off-label use was defined as intervention on ACC/AHA Type C lesion, PCI status of urgent or STEMI, intervention in a previously treated lesion, use of more than 2 stents in a lesion, treatment of a left main or graft segment, or multi-vessel PCI. For those enrolled using data collection form version 3, the off-label use definition was modified to also include device diameter ≤2.5 or >4 mm, total stented or lesion length ≥30 mm, and bifurcation lesions.

We conducted 2 sensitivity analyses. For the first analysis, each of the 5 main outcomes were examined in a subgroup of patients fitting the inclusion and exclusion criterion from the TAXUS IV (Treatment of de novo coronary disease using a single paclitAXel-elUting Stent IV) and SIRIUS (multicenter randomized double-blind study of the SIRolImUS-coated Bx Velocity stent in the treatment of patients with de novo coronary artery lesions) trials (n = 49,355) (20- 21) with a recalibrated propensity score including 76 clinical variables with a c-index of 0.71.

The second sensitivity analysis estimated “cause of death” after stent implantation according to the primary diagnosis of a hospital stay during which the patient expired or the most recent hospital stay within 6 months of death. With a previously validated list of ICD-9 codes (22), we examined the relative distribution of causes of death across DES and BMS patients.

Between January 2004 and December 2006, 390,973 NCDR patients ≥65 years of age underwent stent implantation, and 76% were linked to longitudinal Medicare records. After exclusions, the study population included 262,700 patients from 650 sites (Figure 1). Comparison of NCDR patients who did and did not match to Medicare records revealed nonmatch patients to be slightly younger (age 73 years vs. 74 years) and more likely to be men (62% vs. 58%) and to have commercial insurance (15% vs. 3%).

Overall, 45,025 patients received 1 or more BMS and 217,675 received 1 or more DES (54% paclitaxel-eluting, 46% sirolimus-eluting). Unadjusted baseline characteristics show significant differences between DES and BMS; these differences were reduced after propensity score weighting (Table 1). Sixty-nine percent of DES implantations were for non–FDA-approved indications. Mean follow-up for BMS patients was slightly longer (496 ± 371 days) than for DES patients (456 ± 302 days), due to the trends in stent use over the time period studied.

During the 30-month study period, 21,254 deaths occurred. Thirty-month overall mortality was higher in patients who received BMS than DES both before (17.9% vs. 12.9%; p < 0.0001) and after adjustment for population differences (16.5% vs. 13.5%, HR: 0.75; 95% CI: 0.72 to 0.79) (Table 2). The adjusted mortality difference was statistically significant in the initial 6 months after PCI and continued to increase throughout the 30-month follow-up period (Figure 2A). The estimated HR obtained with an unweighted Cox proportional hazards mortality model with backward variable selection was similar at 0.79 with a 95% CI (0.76 to 0.81). In addition to the use of DES, other factors favorably influencing 30-month post-PCI survival included female sex and prior PCI or CABG. As expected, mortality was higher in those with diabetes, renal failure, STEMI, or congestive heart failure.

There were 10,528 MIs during the study period. Unadjusted MI rates at 30 months were 10.0 of 100 patients in BMS versus 7.3 of 100 patients in DES (p < 0.0001) with similar results after adjustment (8.9 of 100 patients vs. 7.5 of 100 patients, HR: 0.77; 95% CI: 0.72 to 0.81) (Table 2). This result was driven by lower MI rates in DES patients during the first 12 months after PCI (Figure 2B), with no difference between 12 and 30 months of follow-up. In a secondary analysis, DES patients experienced a small increase in STEMI events beyond 12 months (Figure 3).

Revascularization (PCI or CABG) was performed in 34,751 patients, with a total of 40,427 revascularizations; 30-month unadjusted revascularization rates for BMS and DES populations were 24.5 of 100 patients and 23.0 of 100 patients, respectively (p = 0.007). With risk-adjustment, no difference in overall revascularization was observed in DES versus BMS patients at 30 months (23.5 of 100 patients vs. 23.4 of 100 patients, HR: 0.91; 95% CI: 0.87 to 0.96) (Table 2) (Figure 2C). However, revascularization rates were lower in DES patients 12 months after PCI (13.3 of 100 patients vs. 15.2 of 100 patients), followed by a late rebound in revascularization procedures in the DES group between 12 and 30 months (10.2 of 100 patients vs. 8.2 of 100 patients). When CABG and PCI revascularizations were examined separately, CABG was more common in BMS than DES over the 30-month follow-up period (3.7 of 100 patients vs. 2.5 of 100 patients), whereas the rate of PCI was similar.

During follow-up, 4,010 strokes and 5,120 major bleeding events required hospital stay, with 59% of strokes and 49% of bleeds occurring within 6 months after PCI. Unadjusted and adjusted stroke rates were roughly 3 of 100 patients at 30 months in each group (HR: 0.97; 95% CI: 0.88 to 1.07), and only a minimal difference was noted in bleeding (3.6 of 100 patients BMS vs. 3.4 of 100 patients DES, HR: 0.91; 95% CI: 0.84 to 1.00) (Table 2) ((Figure 2)D and Figure 2E).

Each of the composite end points tracked closely with its individual components, favoring DES- over BMS-treated patients both before and after statistical adjustment (Table 2). The unadjusted 30-month rates of death or MI (17% vs. 23%), death or MI or revascularization (32% vs. 38%), and death or MI or stroke (19% vs. 24%) were each lower in DES than BMS patients.

The 30-month DES survival advantage was present across all patient subgroups, independent of sex, age, comorbidities, and procedural indication or urgency (Figure 4A). This effect was somewhat less pronounced in those with a prior history of CABG and renal failure, with or without dialysis. Notably, patients receiving DES in 2005 and 2006 had a greater relative survival benefit than those receiving DES in 2004. Similarly, the 30-month risk of MI was lower in all patient subgroups except those with renal failure and insulin-dependent diabetes (Figure 4B).

Most patient subgroups experienced a slightly lower 30-month rate of revascularization with DES compared with BMS (Figure 4C). However, no benefit was observed in patients age >75 years or with diabetes, renal failure, heart failure, or 3-vessel disease. Revascularization rates were similar in patients undergoing PCI in 2006, in contradistinction to the slightly lower DES revascularization rates from 2004 and 2005 ((Figure 4)D and Figure 4E).

The 49,355 NCDR registry patients fitting the inclusion and exclusion criteria for the Taxus IV and SIRIUS DES randomized controlled trials (RCTs) had 30-month outcomes similar to those of the overall population such that those receiving DES had a lower 30-month risk of death (HR: 0.62; 95% CI: 0.55 to 0.70), MI (HR: 0.66; 95% CI: 0.55 to 0.80), death or MI (HR: 0.64; 95% CI: 0.57 to 0.70), and revascularization (HR: 0.87; 95% CI: 0.80 to 0.96) compared with BMS. No difference in stroke (HR: 0.97; 95% CI: 0.74 to 1.28) or major bleeding (HR: 0.87; 95% CI: 0.71 to 1.05) was noted between trial-eligible DES and BMS patients.

Presumed “cause” was extrapolated in 19,132 (90%) deaths with the algorithm described in the preceding text and included 8,451 inpatient and 10,591 outpatient deaths. Slightly more BMS deaths were attributable to MI (15.0% vs. 13.5%, p = 0.01) and malignancy (6.7% vs. 5.5%, p = 0.002), whereas more DES deaths were more attributable to chronic lung disease (2.5% vs. 1.9%, p = 0.01) and cerebrovascular disease (5.3% vs. 4.2%, p = 0.003). No significant differences were found for any of the remaining diagnoses. Overall, DES patients had a lower risk of cardiovascular-only (including congestive heart failure and MI) deaths compared with BMS patients (HR: 0.80; 95% CI: 0.74 to 0.86) as well as noncardiovascular death from all other causes (HR: 0.74; 95% CI: 0.70 to 0.78).

Our study is the largest-ever observational comparison of long-term outcomes in older patients receiving BMS or DES. A DES implantation was associated with lower risk of death and MI at 30 months as compared with BMS, whereas there were minor, if any, differences in bleeding, stroke, and overall revascularization. Our methodology allowed determination of comparative effectiveness in unselected individuals, in contemporaneous DES and BMS cohorts, with device selection and subsequent management of patients reflecting real-life, community practice.

Prior analyses comparing survival in DES- and BMS-treated patients from RCTs and smaller registries have produced conflicting results with relatively low precision. Whereas no difference in late survival was demonstrated in some RCTs (1,23), registries, and meta-analyses (9- 11,23- 28), other more recent studies have demonstrated a DES survival advantage with a point-estimate similar to that observed in our population (13,29- 32). The higher annualized mortality rates for patients in our population receiving either DES or BMS (5.4%/year vs. 6.6%/year) than previously reported in some registries (range 1.3%/year to 4.3%/year) (23- 27,33- 34) are likely due to higher risk in our elderly, inpatient population and are comparable to other Medicare cohorts (11,35).

Patients receiving DES experienced a 23% relative reduction in subsequent MI with no late increase in combined NSTEMI/STEMI risk, a result similar to several other analyses (1,9,11). Angiographic assessment of stent thrombosis was not possible in our dataset; however, isolated analysis of STEMI events revealed a slight increase in very late (>12 months) STEMI risk in DES patients, consistent with prior published reports on late stent thrombosis (6) and the expected time-course of clopidogrel discontinuation (36- 37).

Although DES have been associated with low revascularization rates (6,9,11,13,29- 32,34), recent registry reports suggest that they might actually be as high as 15% to 19% over a 2- to 3-year follow-up (6,11,26), with little difference between DES and BMS patients (6). The higher rate of repeat revascularization in our population (24%) might be due to not censoring patients after an event and to the inability to differentiate target lesion revascularization from nontarget lesion revascularization follow-up procedures with claims data. For example, a recent report from the Duke database identified a 2-fold higher rate of overall revascularization versus target vessel revascularization in DES patients at 2 years (12.0% vs. 6.6%) (26). Thus, the lack of anatomic data makes this database less than ideal for the comparison of revascularization between DES and BMS. An additional concern is the higher rate of late revascularization in DES compared with BMS, which tends to obscure the early benefit when examining overall DES-BMS HR. Our revascularization results should be interpreted cautiously.

Few differences in stroke or major bleeding rates requiring repeat hospital stay were observed between the overall DES and BMS populations. The anticipated greater use of clopidogrel in the DES group might have conferred a bleeding disadvantage, as has been seen in other studies (38- 39); however, no statistically significant difference was observed in our population. Unexpectedly, although a slightly higher unadjusted rate of anemia-associated deaths was observed in DES patients, no significant adjusted or unadjusted difference in gastrointestinal hemorrhage-associated deaths was evident at 30 months.

The differences in outcomes between registry and RCT analyses have been previously attributed to possible differences in DES performance in a real-world (registry) population as compared with a restricted RCT population, with the lack of a survival difference in RCTs being an artifact of their restricted patient populations. Because creation of a population subset fitting the inclusion and exclusion criteria of the Taxus IV and SIRIUS DES RCTs (20- 21) only sharpened the precision for each end point, differences in age, acuity, lesion characteristics, and off-label use in the registry and RCT populations are unlikely explanations of the observed differences in results.

Incomplete risk-adjustment after biased real-world stent selection might also contribute to the survival advantage noted in this and other registry analyses. Although we used both propensity analyses and Cox proportional hazards models to adjust for differences in baseline characteristics, it is possible that unmeasured baseline population differences remained. In fact, our “cause of death” sensitivity analysis did show slightly higher rates of death due to malignancy in BMS patients, suggesting that biased patient selection might have contributed to the overall mortality result, such that “sicker” patients with more comorbid disease received BMS. Although it is possible that the observed differences between DES and BMS patients are the sole product of unmeasured patient selection biases not reflected by this analysis, this explanation is less likely, given the large number of covariates used in our propensity matching.

This represents a novel large-scale linkage between a national procedural registry and a robust claims database, demonstrating that nationally representative analyses are feasible with clinically rich, procedural registries and a claims-based structure for follow-up. In combination, these 2 resources provide a powerful mechanism for tracking the post-marketing use and outcomes of novel devices and procedural innovations, at minimal cost. Importantly, the project was financed by the Agency for Healthcare Research and Quality and the Cardiovascular Consortium of Effective Healthcare Program and was independent of industry.

Data entered in NCDR are intended to be used as a quality improvement tool and undergo rigorous quality control (40), whereas the Medicare claims database captures all inpatient care episodes. Despite these disparate intents and the size and complexity of these 2 databases, our linkage rate was over 75%, adding to the generalizability of our results. We analyzed these data by 3 methods: IPW alone, IPW with Cox proportional hazards modeling, and standard Cox proportional hazards modeling to compare results between the different approaches. The high level of agreement between these methods enhances confidence in our findings.

Our data are observational and therefore dependent on the accuracy and completeness of the 2 matched datasets. Reliance on a claims database for outcomes might be fraught with under-reporting or misclassification of events. Such misclassification and underreporting should be nondifferential, however, and should bias our estimates toward the null value of no overall difference. Although differences in baseline characteristics were rigorously adjusted for using propensity weighting, it is possible that additional unadjusted differences between BMS and DES patients affected results, because our analyses are limited to the data collected by ACC-NCDR and Medicare. Thus we are unable to directly address some important questions that have been raised regarding the safety of stents, including whether repeat revascularization represented target lesion or vessel restenosis, the incidence of late stent thrombosis, and the impact of variations in thienopyridine use. Although the slight excess of STEMIs in DES patients after 12 months fits the time course of late stent thrombosis, these events did not translate to increased late mortality.

Although the linkage rate of 76% is incomplete, it is reflective of populations known to be absent from the Medicare data set, such as patients treated at Veterans Administration facilities, with Medicare Advantage insurance coverage, or undergoing outpatient procedures. Our findings are drawn from hospitalized patients over age 65 years, an age group that accounts for approximately one-half of all PCIs nationally. Although this cohort is likely sicker than the generally younger outpatient PCI population, the similarity in outcomes in those age 65, 75, and >75 years suggests that these differences might have been accounted for in the risk adjustment.

In summary, in this large population of Medicare beneficiaries undergoing PCI at facilities participating in the ACC-NCDR registry, patients who received DES had significantly lower mortality rates, including an early decrease in MI, than those who received BMS. No excess of major bleeding or stroke was noted. The survival advantage associated with DES was maintained across all subgroups analyzed and throughout the 30 months of follow-up. Drug-eluting stents seem to be safe and effective in community practice in the elderly population. Longer follow-up studies will need to be conducted to further support these results and to confirm the possible effects of antiplatelet agents.

For a supplementary table on the covariates used in the propensity score model, please see the online version of this article.

Clinical Effectiveness of Coronary Stents in Elderly Persons: Results From 262,700 Medicare Patients in the American College of Cardiology National Cardiovascular Data Registry