Systematic Review: The Comparative Effectiveness of Percutaneous Coronary Interventions and Coronary Artery Bypass Graft Surgery

Context The relative benefits and harms of coronary artery bypass graft surgery (CABG) versus percutaneous coronary intervention (PCI) are sometimes unclear. Contribution This systematic review of 23 randomized trials found that survival at 10 years was similar for CABG and PCI, even among diabetic patients. Procedural strokes and angina relief were more common after CABG (risk difference, 0.6% and about 5% to 8%, respectively), whereas repeated revascularization procedures were more common after PCI (risk difference, 24% at 1 year). Caution Only 1 small trial used drug-eluting stents. Few patients with extensive coronary disease or poor ventricular function were enrolled. The Editors Coronary artery bypass graft (CABG) surgery and catheter-based percutaneous coronary intervention (PCI), with or without coronary stents, are alternative approaches to mechanical coronary revascularization. These 2 coronary revascularization techniques are among the most common major medical procedures performed in North America and Europe: In 2005, 261000 CABG procedures and 645000 PCI procedures were performed in the United States alone (1). However, the comparative effectiveness of CABG and PCI remains poorly understood for patients in whom both procedures are technically feasible and coronary revascularization is clinically indicated. In patients with left main or triple-vessel coronary artery disease with reduced left ventricular function, CABG is generally preferred because randomized, controlled trials (RCTs) have shown that it improves survival compared with medical therapy (2, 3). In patients with most forms of single-vessel disease, PCI is generally the preferred form of coronary revascularization (4), in light of its lower clinical risk and the evidence that PCI reduces angina and myocardial ischemia in this subset of patients (5). Most RCTs comparing CABG and PCI have been conducted in populations with coronary artery disease between these extremes, namely patients with single-vessel, proximal left anterior descending disease; most forms of double-vessel disease; or less extensive forms of triple-vessel disease. We sought to evaluate the evidence from RCTs on the comparative effectiveness of PCI and CABG. We included trials using balloon angioplasty or coronary stents because quantitative reviews have shown no differences in mortality or myocardial infarction between these PCI techniques (6, 7). We also included trials using standard or minimally invasive CABG or both procedures (8, 9). We sought to document differences between PCI and CABG in survival, cardiovascular complications (such as stroke and myocardial infarction), and freedom from angina. Finally, we reviewed selected observational studies to assess the generalizability of the RCTs. Methods Data Sources We searched the MEDLINE, EMBASE, and Cochrane databases for studies published between January 1966 and August 2006 by using such terms as angioplasty, coronary, and coronary artery bypass surgery, as reported in detail elsewhere (10). We also sought additional studies by reviewing the reference lists of included articles, conference abstracts, and the bibliographies of expert advisors. We did not limit the searches to the English language. Study Selection We sought RCTs that compared health outcomes of PCI and CABG. We excluded trials that compared PCI alone or CABG alone with medical therapy, those that compared 2 forms of PCI, and those that compared 2 forms of CABG. The outcomes of interest were survival, myocardial infarction, stroke, angina, and use of additional revascularization procedures. Two investigators independently reviewed titles, abstracts, and the full text as needed to determine whether studies met inclusion criteria. Conflicts between reviewers were resolved through re-review and discussion. We did not include results published solely in abstract form. Data Extraction and Quality Assessment Two authors independently abstracted data on study design; setting; population characteristics (sex, age, race/ethnicity, comorbid conditions, and coronary anatomy); eligibility and exclusion criteria; procedures performed; numbers of patients screened, eligible, enrolled, and lost to follow-up; method of outcome assessment; and results for each outcome. We assessed the quality of included trials by using predefined criteria and graded their quality as A, B, or C by using methods described in detail elsewhere (10). In brief, a grade of A indicates a high-quality trial that clearly described the population, setting, interventions, and comparison groups; randomly allocated patients to alternative treatments; had low dropout rates; and reported intention-to-treat analysis of outcomes. A grade of B indicates a randomized trial with incomplete information about methods that might mask important limitations. A grade of C indicates that the trial had evident flaws, such as improper randomization, that could introduce significant bias. Data Synthesis and Analysis We used random-effects models to compute weighted mean rates and SEs for each outcome. We computed summary risk differences and odds ratios between PCI and CABG and the 95% CI for each outcome of interest at annual intervals. Because the results did not differ materially when risk differences and odds ratios (10) were used and the low rate of several outcomes (for example, procedural mortality) made the risk difference a more stable outcome metric (11, 12), we report here only the risk differences. We assessed heterogeneity of effects by using chi-square and I 2 statistics (13). When effects were heterogeneous (I 2 > 50%), we explored the effects of individual studies on summary effects by removing each study individually. We assessed the possibility of publication bias by visual inspection of funnel plots and calculated the number of missing studies required to change a statistically significant summary effect to not statistically significant (11). We performed analyses by using Comprehensive Meta-Analysis software, version 2.0 (Biostat, Englewood, New Jersey). Inclusion of Observational Studies We also searched for observational data to evaluate the generalizability of the RCT results, as reported in detail elsewhere (10). In brief, we included observational studies from clinical or administrative databases that included at least 1000 recipients of each revascularization procedure and provided sufficient information about the patient populations (such as demographic characteristics, preprocedure coronary anatomy, and comorbid conditions) and procedures performed (such as balloon angioplasty vs. bare-metal stents vs. drug-eluting stents). Role of the Funding Source This project was supported by the Agency for Healthcare Research and Quality. Representatives of the funding agency reviewed and commented on the study protocol and drafts of the manuscript, but the authors had final responsibility for the design, conduct, analysis, and reporting of the study. Results We identified 1695 potentially relevant articles, of which 204 merited full-text review (Appendix Figure). A total of 113 articles reporting on 23 unique RCTs met inclusion criteria (Table 1 [14126]). These trials enrolled a total of 9963 patients, of whom 5019 were randomly assigned to PCI and 4944 to CABG. Most trials were conducted in Europe, the United Kingdom, or both locations; only 3 trials were performed in the United States. The early studies (patient entry from 1987 to 1993) used balloon angioplasty as the PCI technique, and the later studies (patient entry from 1994 to 2002) used stents as the PCI technique. Only 1 small trial of PCI versus CABG used drug-eluting stents (116). Nine trials limited entry to patients with single-vessel disease of the proximal left anterior descending artery, whereas the remaining 14 trials enrolled patients with multivessel disease, either predominantly (3 trials) or exclusively (11 trials). Appendix Figure. Study flow diagram. CABG= coronary artery bypass grafting; CAD= coronary artery disease; PCI= percutaneous coronary intervention; RCT= randomized, controlled trial. Table 1. Overview of Randomized, Controlled Trials The quality of 21 trials was graded as A, and 1 trial (117) was graded as B. One trial (116) was graded as C because randomization may not have been properly executed (details are available elsewhere [10]). We performed sensitivity analyses by removing these studies from the analysis, and our summary results did not change statistically significantly. The average age of the trial participants was 61 years, 27% were women, and most were of European ancestry. Roughly 20% had diabetes, half had hypertension, and half had hyperlipidemia. Whereas approximately 40% of patients had a previous myocardial infarction, few had heart failure or poor left ventricular function. Among studies that enrolled patients with multivessel coronary disease, most had double-vessel rather than triple-vessel disease. Revascularization procedures were performed by using standard methods for the time the trial was conducted (Table 1). Among patients with multivessel disease, more grafts were placed during CABG than vessels were dilated during PCI. Among patients assigned to PCI, stents were commonly used in the recent studies, but in the earlier trials, balloon angioplasty was standard. Among patients assigned to CABG, arterial grafting with the left internal mammary artery was frequently done, especially in more recent trials. Some studies used minimally invasive, direct coronary artery bypass and off-pump operations to perform CABG in patients with single-vessel left anterior descending disease (Table 1). Short-Term and Procedural Outcomes Survival (within 30 days of the procedure) was high for both procedures: 98.9% for PCI and 98.2% for CABG. When data from all trials were combined, the survival difference between PCI and CABG was small and not statistically significant (0.2% [95% CI, 0.3% to 0.6%]) (Figure 1