Author + information
- Received February 14, 2000
- Revision received August 14, 2000
- Accepted November 2, 2000
- Published online February 1, 2001.
- Stephen E Kimmel, MD, MS∗,†,* (, )
- A.Russell Localio, MPH, MS, JD∗,‡,
- Ronald J Krone, MD, FACC§,
- Warren K Laskey, MD, FACC∥,
- for the Registry Committee of the Society for Cardiac Angiography and Interventions
- ↵*Reprint requests and correspondence: Dr. Stephen E. Kimmel, University of Pennsylvania School of Medicine, Center for Clinical Epidemiology and Biostatistics, 717 Blockley Hall, 423 Guardian Drive, Philadelphia, Pennsylvania 19104-6021
This study was designed to determine the effect of coronary stents on in-hospital mortality.
Despite extensive use of stents for percutaneous coronary interventions (PCIs), their effect on serious in-hospital events, especially mortality, is not well defined.
A cohort study was performed using 16,811 consecutive native-vessel PCI procedures performed on patients in the Society for Cardiac Angiography & Interventions Registry from July 1, 1996, through December 31, 1998. Patients undergoing balloon-only angioplasty were compared with those receiving a planned or unplanned stent. Procedures with other devices were excluded. Multivariable analyses adjusted for detailed clinical characteristics and for individual laboratory.
Stents were associated with a significant reduction in in-hospital mortality (0.3%) compared with balloon procedures (0.6%; multivariable odds ratio [OR] 0.55; 95% confidence interval [CI] 0.34, 0.89; p = 0.014). The risk of emergency coronary bypass also was reduced by stenting (0.3% vs. 0.7%; multivariable OR 0.47; 95% CI: 0.29, 0.76; p = 0.002). Adjustment for the use of glycoprotein IIb/IIIa inhibitors did not change the results, and the effects of stenting relative to balloon procedures were similar in those procedures with and without glycoprotein IIb/IIIa blockade (p = 0.94).
This study suggests that coronary stenting, compared with balloon procedures, reduces in-hospital mortality, independent of the clinical setting.
Despite the explosive growth in the use of coronary artery stenting for percutaneous coronary interventions (PCIs) (1), primarily justified by the reduction in restenosis demonstrated in randomized trials (2–7), few data have been available from routine clinical practice to determine the effects of stents on in-hospital events, especially mortality (8). Although randomized trials of coronary stents have not been specifically designed to examine in-hospital mortality (3–7,9), several trials, in the absence of adjuvant platelet glycoprotein IIb/IIIa (GP IIb/IIIa) receptor blockade, have shown a nonsignificant excess of cardiac events in patients receiving coronary stents (1,3,4,6). In contrast, a recent study using administrative data raised the possibility that stents reduce in-hospital mortality (8). Randomized trial data comparing stents plus abciximab with balloon angioplasty plus abciximab also demonstrated a reduction in six-month rates of death with stenting (10). However, the contribution of periprocedural mortality to these six-month findings is not known, nor are the effects of stenting versus balloon procedures independent of adjuvant therapies. Equally important, the effect of coronary stenting on mortality in broad-based practice may differ from the results of randomized trials that often include only carefully selected patients and experienced practitioners.
If stents do reduce mortality in actual practice, then further expansion of their use and reassessment of benchmark standards of short-term PCI outcomes should be employed. It is therefore critical for the cardiovascular community to clarify the effects of stenting on mortality in actual practice. The specific aims of this study were to examine the effects of coronary stenting in broad-based clinical practice on in-hospital mortality and the interaction of these effects with the use of GP IIb/IIIa inhibitors and with other clinically important groups of patients.
This study included consecutive patients undergoing PCI from July 1, 1996, through December 31, 1998, in the SCA&I Registry. Details of the SCA&I Registry data collection techniques and variable definitions have been reported previously (11,12). Briefly, detailed clinical characteristics of all patients undergoing PCI were entered into the Registry by support personnel in each laboratory prior to each patient’s procedure, and all in-laboratory outcomes and in-hospital deaths were recorded after the procedure. The data entry program included range and logic checks to prevent incorrect keying of data.
Patient population and laboratories
A total of 19,510 angioplasty procedures were performed in 33 centers participating in the SCA&I Registry. Fifteen centers contributed data to all 2½ years of the Registry. Procedures involving devices other than balloons or stents were excluded (1,354 procedures; mortality 0.5%, p = 0.7 compared with included patients).
The indications for stenting were prospectively defined and these definitions were used by all centers in the Registry. “Planned” stenting (n = 8,410) was defined as the intended use of a stent prior to beginning a procedure, whereas “unplanned” stents (n = 2,280) were those used after a balloon inflation because of unsatisfactory balloon results, including minor dissection. If a stent was used to treat the acute complication of either abrupt artery closure or impending closure following an initial procedure (regardless of the initial device used) it was considered to be a “bailout” procedure. These procedures did not use stents simply to improve the angiographic appearance of balloon angioplasty or to treat minor dissection.
The purpose of this study was to examine the effects of stents when used as planned therapy. However, because unplanned stents could be used to improve the angiographic appearance of a balloon angioplasty, they were considered part of “routine” PCI practice, and were therefore included in the study population. Secondary analyses excluded them.
Few stents were coded as “bailout” (n = 150, 1.2% of all procedures), reflecting the low incidence of major complications in this study; the majority (69%) were used to treat complications of balloon angioplasty (n = 104). However, abrupt vessel closure is itself a complication of angioplasty (13–15)and bailout stents are used to treat these complications. Therefore, it would be inappropriate to include bailout stents as predictors of the acute complications studied (16). It should be noted, however, that including procedures requiring bailout stenting did not change the study results.
Because PCI of coronary bypass grafts represents a different type of procedure and because of the limited number of graft procedures (330 balloon procedures with no mortality and 811 stent procedures with only two deaths), graft PCI procedures also were, a priori, excluded. The study population therefore consisted of subjects undergoing either isolated balloon angioplasty or stenting, planned or unplanned, of native coronary vessels.
The primary outcome variable was mortality, defined as death occurring any time during the hospitalization. Cause of death was not recorded. In addition, we examined emergency coronary artery bypass grafting (CABG), defined as CABG required immediately following the PCI procedure. This was further divided into “unstable” CABG (patients going to CABG with ongoing ischemia and/or hemodynamic instability) and “stabilized” CABG (patients going to CABG without ischemia or hemodynamic instability). Because of the lack of standardization of surveillance for myocardial infarction (MI), this outcome was not included.
Data on the type of device and indication for stenting (planned vs. unplanned) were available for 100% of procedures. Data were complete for mortality and CABG outcomes for 99.7% of procedures. Complete data also were available for >95% of all other study variables.
The demographic characteristics of subjects undergoing stent versus balloon procedures were compared using chi-square statistics. The unadjusted associations between stent use and the outcomes were estimated using odds ratios (ORs) and chi-square statistics.
To control for confounding by individual patient-level factors and also for the potential for confounding by laboratory (17,18)(because of enormous variation in the rates of use of stents across the participating hospitals), a two-stage procedure was used. First, a propensity score model was derived to predict the use of stents across hospitals (19,20). Propensity scores reduce the potential for selection bias and allow for simultaneous adjustment for multiple confounders without resulting in model overfitting (as would occur if each laboratory were included in a multivariable regression model) (19). In this initial model, the outcome was stent (vs. balloon procedures) and the predictors included laboratory, year, age, gender, heart failure, renal insufficiency, diabetes, hypertension, multivessel disease, peripheral vascular disease, unstable angina, American Heart Association/American College of Cardiology lesion type, MI prior to procedure, shock (systolic blood pressure <80 mm Hg and cardiac index <1.8 l/min/m2despite, or requiring, pharmacological or mechanical support), use of platelet GP IIb/IIIa blockers, use of thrombolytics, emergency and ad hoc procedures and prior CABG and PCI. Multi- versus single-vessel PCI was not included because unsatisfactory results of the first balloon dilation of a planned multivessel procedure could lead to the use of unplanned stenting and aborting further lesion dilations. This would make single-vessel PCI appear to predict stent use when, in fact, the need for the stent was the reason for performing only a single-vessel PCI. Vessel size was not included because: 1) it was not associated with mortality (p = 0.5); 2) it was missing in 30% of subjects; and 3) adjustment for vessel size in the subset of patients with known vessel size did not alter the findings (i.e., vessel size was not a confounder). The Hosmer–Lemeshow goodness-of-fit statistic (21)was used to assess the fit of the logistic model.
In the second stage, the patients were grouped into deciles of propensity to receive a stent based on the initial model. These deciles were then used as an indicator variable in logistic regression models that included only the propensity score and the device variable as independent variables. This method allows for determination of the effect of stents after adjustment for all variables included in the original propensity score model discussed earlier (19,20).
Interactions between device type and other risk factors were examined using multivariable logistic regression with the appropriate interaction terms, a propensity score that was derived using hospital only, and all other potential confounders. The a priori interaction subgroups were age, diabetes, use of GP IIb/IIIa inhibitors, MI within 24 h prior to PCI and lesion type.
Because outcomes may be correlated within laboratories (i.e., there is a potential lack of independence of the risk of outcomes within centers) (17), all analyses accounted for clustering of patients within laboratories using robust variance estimates (22). Analyses were performed using SPSS version 9.0.1 (SPSS Incorporated, Chicago, Illinois) and Stata version 5 (Stata Corporation, College Station, Texas).
Study population and predictors of stent use
A total of 16,811 procedures were performed among 33 laboratories. There were 6,121 balloon-only and 10,690 stent procedures.
The demographics of the population are shown in Table 1. Patients undergoing a balloon procedure were more likely to be female, undergo PCI within 24 h of an MI or as an emergency procedure, and have diabetes, renal insufficiency, multivessel disease, prior CABG, prior PCI and shock. Stent patients, however, were more likely to undergo an urgent procedure, receive a GP IIb/IIIa inhibitor, have a recent MI, receive thrombolytic therapy, have unstable angina and have complex lesion morphology.
Stent use increased from 44% in 1996 to 72% in 1998 (test for trend p < 0.001), and the proportion of stents that were unplanned decreased from 44% to 14% (test for trend p < 0.001). Individual hospitals varied widely in their use of stents (from 28% of procedures to 93%; p < 0.001). Although stent use was much more prevalent by 1998, there was still a range of stent use among laboratories (29% to 95%).
The propensity score model revealed several independent predictors of receiving a coronary stent (Table 2; Hosmer–Lemeshow goodness-of-fit statistic p = 0.7).
Overall, there were 76 deaths (0.5%) and 77 emergency CABG procedures (0.5%). Stents were associated with a significant reduction in both in-hospital mortality and emergency CABG (Table 3). Stents had their greatest effect on reducing “unstable” CABG procedures (Table 3). All results were essentially unchanged after adjustment for all other potential confounders (Table 3). Of the 76 deaths, 60 (79%) occurred early (≤24 h of the procedure). There was a statistically significant reduction in early mortality (0.3% in stent group vs. 0.5% in balloon group, p = 0.028) but not later (>24 h) mortality (0.1% in both groups, p = 0.1).
Among stent procedures, 8,410 (79%) were planned and 2,280 (21%) were unplanned. The mortality among planned stenting (0.3%) was lower than among unplanned stenting (0.7%, p = 0.01). Excluding unplanned stents, the effect of planned stenting alone versus balloon PCI was more pronounced: multivariable OR for death, 0.33 (95% CI: 0.21, 0.50); for CABG, 0.33 (95% CI: 0.19, 0.58); and for any death or CABG, 0.33 (95% CI: 0.22, 0.50).
The need for emergency CABG was associated with a significant increase in mortality (7.8% vs. 0.4% in patients not requiring emergency CABG; p < 0.001). In order to assess the contribution of reduced CABG incidence to the stent-mortality relationship, we determined the OR for mortality from stenting after adjusting for the difference in emergency CABG due to stents. If reduced CABG were the sole reason for reduced mortality, this analysis would produce an OR of 1. Although adjusting for emergency CABG somewhat reduced the OR, stents still significantly reduced the risk of death (multivariable OR 0.58; 95% CI: 0.36, 0.94). In addition, the beneficial effect of stents on mortality was similar for patients undergoing emergency CABG (multivariable OR 0.56) versus those not undergoing CABG (multivariable OR 0.59; test for interaction p = 0.97).
There were no significant interactions between device type and any of the a priori subgroups: age, diabetes, use of GP IIb/IIIa receptor blockers, MI within 24 h prior to PCI and lesion type (all p > 0.10). That is, the effect of stenting on reducing mortality was statistically indistinguishable between subgroups with and without these risk factors. With respect to GP IIb/IIIa inhibitors, the p-value of the test for interaction was 0.94 (i.e., stents had similar effects on mortality relative to balloons in those receiving, vs. not receiving, GP IIb/IIIa receptor blockers). In those receiving GP IIb/IIIa inhibitors, the adjusted OR for stents versus balloons was 0.57 (95% CI: 0.21, 1.56) and in those not receiving GP IIb/IIIa inhibitors the adjusted OR was 0.60 (95% CI: 0.33, 1.08).
This study demonstrated a significantly lower in-hospital mortality and emergency CABG risk with the use of coronary stenting compared with balloon angioplasty. Overall, stents were associated with a 49% reduction in the odds of death or CABG. This represents an absolute risk reduction for mortality and CABG of 0.3% and 0.4%, respectively. Although the absolute risk difference is small, the potential benefit to the population requiring coronary interventions could be great, given the tremendous number of procedures performed worldwide.
Randomized trials of stenting in the absence of platelet GP IIb/IIIa receptor blockade have been too small to exclude an independent effect of stents (positive or negative) on mortality (2–5,23). Some studies had more cardiac events in the stent group (3,4,6)and therefore some have argued that stenting, in the absence of GP IIb/IIIa receptor blockade, may actually increase risk (1). However, a recent observational study using administrative data from California demonstrated a significant reduction in in-hospital mortality in patients receiving stents compared with those not receiving stents (8). This study did not account for multiple potential confounders simultaneously, nor did it account for the effects of clustering. Clustering (or correlation of outcomes within centers) can lead to biased variance estimates and thus biased study inference (17). In addition, administrative data lack clinical, angiographic, medication (e.g., GP IIb/IIIa inhibitor use) and other device (e.g., atherectomy) information, making it impossible to adjust for potential differences between stented and nonstented patients on these factors. Despite these potential limitations, the finding of an approximately 50% reduction in in-hospital death in this prior study is consistent with our results.
Our study also suggests that stents themselves independently reduce periprocedural mortality relative to balloon angioplasty. Although the EPISTENT study demonstrated a reduction in six-month mortality using stents plus abciximab compared with balloons plus abciximab (10), the mechanism of this finding was not elucidated. A reduction in MI from stents did not appear to explain the findings (10). The relative contribution of stenting to periprocedural versus later mortality reduction also was not determined. Finally, the mortality benefit from stenting could have relied entirely on coadministration of abciximab. Our data now support a mortality benefit from stenting itself and suggest that at least some of the intermediate-term benefit from stenting derives from periprocedural mortality reduction. The mortality benefit of stenting relative to balloon angioplasty also appears to be the same for procedures that use GP IIb/IIIa inhibitors (i.e., stents with GP IIb/IIIa inhibitors vs. balloons with GP IIb/IIIa inhibitors) and those that do not (i.e., stents without GP IIb/IIIa inhibitors vs. balloons without GP IIb/IIIa inhibitors), although these analyses had limited power.
There are several important potential limitations inherent in any observational study. First, although it is possible that the mortality benefit was a chance finding, the consistency of benefit in different subgroups and the supportive data from California (8)suggest that these findings were not due to chance.
Second, stents could appear beneficial if they were used in lieu of balloons by lower risk individuals or by lower risk hospitals. However, extensive multivariable analyses that adjusted for numerous clinical and angiographic variables and for individual laboratories did not alter the magnitude of effect of stenting, suggesting that uncontrolled confounding did not produce these results. In addition, we included “unplanned” stents in the stent group to insure that the stent group included patients who had unsatisfactory results from balloon angioplasty. This inclusion also eliminated the possibility that operators who planned to use a stent miscoded it as “unplanned” only because predilation with the balloon did not produce adequate results. The fact that the percentage of stent procedures that were planned (79%) was almost identical to that found in the New Approaches to Coronary Interventions registry (82%) (24)suggests that substantial miscoding was unlikely. Regardless, a mortality benefit from stenting persisted despite the inclusion of unplanned procedures, which are associated with an increased risk of complications (24). Finally, although uncontrolled confounding is still possible (e.g., due to lesion complexity not captured by the lesion type variable), it is extremely unlikely to have produced the observed results. For example, an unmeasured confounder present in 50% of the population would have to be associated with a 25-fold increased risk of both stent use and mortality to have produced our results (25). We know of no such risk factor.
Third, the study could only address the effect of stents versus balloons when platelet GP IIb/IIIa blockers were used concomitantly with both devices and, separately, for stents versus balloons when IIb/IIIa blockers were not used. The study did not and could not address the effect of adding a platelet GP IIb/IIIa blocker to stent procedures (vs. stents without GP IIb/IIIa blockade) because of confounding by indication (platelet GP IIb/IIIa receptor blockers are used in higher risk individuals). The study does not, therefore, contradict the findings of EPISTENT (10). Rather, the lack of a significant interaction by GP IIb/IIIa receptor blockade suggests that stents may reduce mortality relative to balloon procedures when IIb/IIIa blockers are used with both types of devices and also may reduce mortality relative to balloon procedures when IIb/IIIa blockers are not used with both.
Fourth, registry data such as ours are not subject to strict quality assurance measures, making misclassification possible, albeit unlikely to alter the study conclusions.
Fifth, we could not examine MI outcomes or cause of death in our study. However, the difference in mortality by device type appeared to be due to differences in early mortality, suggesting that reductions in early complications of PCI from stents could explain our findings.
This study provides evidence that coronary stenting reduces the risk of periprocedural mortality compared with balloon angioplasty. This early reduction in mortality provides one possible explanation for the six-month mortality benefit seen in EPISTENT (10)and extends these findings to broad-based PCI practice. The results also add to recent administrative data analyses (8)by demonstrating a reduction in mortality from stents that is independent of case-mix, angiographic lesion characteristic, and individual laboratories. The mortality benefit from stents relative to balloon procedures provides additional justification for the addition of coronary stenting to percutaneous coronary interventional procedures, both those that use and do not use platelet GP IIb/IIIa blockade.
The authors acknowledge Sandra Barile for her assistance with manuscript preparation.
☆ The Registry is supported by grants from Cordis Corporation and Nycomed Amersham.
- coronary artery bypass grafting
- confidence interval
- myocardial infarction
- odds ratio
- percutaneous coronary intervention
- Society for Cardiac Angiography and Interventions
- Received February 14, 2000.
- Revision received August 14, 2000.
- Accepted November 2, 2000.
- American College of Cardiology
- Antoniucci D,
- Santoro G.M,
- Bolognese L,
- Valenti R,
- Trapani M,
- Fazzini P.F
- Detre K.M,
- Holmes D.R.J,
- Holubkov R,
- et al.
- Baim D.S,
- Kent K.M,
- King S.B,
- et al.
- de Feyter P.J,
- van den Brand M,
- Laarman G.J,
- van Domburg R,
- Serruys P.W,
- Suryapranata H
- Breslow N.E,
- Day N.E
- Hosmer D.W,
- Lemeshow S