A Medline search from October 2002 to October 2012 included the following keywords: ‘percutaneous coronary intervention’, ‘coronary artery bypass grafting’, ‘comparison’, ‘multi-vessel’, ‘left-main coronary artery’, ‘stent’, ‘CABG’, and ‘PCI’. Reference lists of articles were reviewed for additional papers. Additionally, we performed a manual search of the table of contents of journals known to publish relevant content and contacted key researchers in the field to inquire about manuscripts in-press.
Inclusion criteria included the following: 1) Studies published in English with full text available; 2) Retrospective comparisons between CABG and PCI; 3) > 1 year of follow-up; and 4) multi-vessel CAD. Exclusion criteria included: 1) ACS/MI within 24 hours of intervention; 2) left main coronary artery disease (LMCA); 3) RCTs; and 4) studies with the main goal of a specific subpopulation comparison. Review articles, editorials, and other non-peer reviewed manuscripts or abstracts were excluded. Studies also were assessed for scientific rigor (e.g., peer reviewed, impact factor for journal), inclusion of relevant independent and outcome-related variables, appropriate sample size, statistical heterogeneity of results, validity (internal, external), similarity of hypotheses across studies, evidence of a sufficient knowledge base for statistical integration, and consistency of evidence. A scoping review was undertaken as a means of refining the specific question for the systematic integration of the studies [7
]. Quality of manuscripts was assessed using a domain-based evaluation [8
]. We considered studies of greater than 1 year to be “long-term.”
Two reviewers independently conducted literature searches and discrepancies were resolved by consensus. Abstracts of relevant articles were evaluated for inclusion in this study. Data concerning study characteristics and comorbid conditions were recorded.
Source information was tabulated for all studies including, publication year, country of data collection, report type, and language in which the study was published. Statistical analyses were conducted using SAS®
Software (Version 9.3, Cary, NC). Hazard ratios (HR) and 95% confidence intervals (CI)were individually plotted to visualize differences between studies. Summary HRsand95%CIswere computed by adapting standard Mantel-Haenszel (M-H) methods for determining weighted log-normal relative effect measures [9
]. Homogeneity of HRs was tested using Tarone’s approximate score method [10
]. Although unpublished studies were not included in the current analysis, we collected basic information on these studies when available to help determine possible publication bias (file-drawer effect). A methods moment was employed to assess the sensitivity of results to hypothetical unpublished studies [11