Comparison Of Bypass Surgery And Drug-eluting Stenting In Diabetics With Left Main And/or Multivessel Disease:a Meta-analysis Of Randomized And Nonrandomized Studies

1. Background and Aims:Diabetes mellitus is a powerful, independent risk factor for cardiovascular disease and account for about25%of all patients requiring myocardial revascularization. Patients with diabetes have more extensive and diffuse coronary artery disease (CAD) than nondiabetic patality and morbidity after revascularization procedures, including myocardial infarction (MI), and restenosis after balloon angioplasty and bare-metal stenting. Despite significant improvements in the CAD mortalities in the past decades, CAD because of atherosclerosis resulting in MI remains the leading cause of death worldwide. CABG was generally regarded as a preferred revascularization strategy for patients with LM and/or MVD. However, the advances in the interventional field, especially the advent and development of drug-eluting stents(DES) which significantly reduced restenosis and the need for subsequent repeat revascularizations as compared with bare metal stents (BMS), have largely cut back one of the major limitations of PCI. Several RCTs, exclusively comparing coronary artery bypass grafting surgery (CABG) and percutaneous coronary intervention with drug-eluting stents (PCI-DES) for the diabetic subset with left main (LM) and/or multivessel disease (MVD), have reported medium-and long-term outcomes, but given high-selected patients in the RCTs, their applicability to the general population is unknown. The OCTs, unlike the RCTs, can reflect daily clinical practice in the real world. We conducted a systematic review and meta-analysis of randomized and nonrandomized studies to establish clinical efficacy and safety of PCI-DES versus CABG in patients with diabetes and LM and/or MVD both in the real world (OCTs) and in high-selected population(RCTs).2. Methods2.1. Search StrategyWe systematically searched PubMed, EMBASE, the Cochrane Central Register of Controlled Trials, Google Scholar and SinoMed for relevant studies reported from January2002(the year when the drug-eluting stents were introduced to clinical practice), to December2013, without language and publication restriction. To achieve the maximum sensitivity of the search strategies and identify all trials comparing PCI-DES with CABG in diabetic subset, we appropriately used both free text and thesaurus terms, including:multivessel disease, left main, diabetes mellitus, percutaneous coronary intervention, drug-eluting stents, and coronary artery bypass. We also performed a systematical search from reference lists of selected articles, conference proceedings, and personal files for relevant citations.2.2. Inclusion and exclusion CriteriaStudies were included in this meta-analysis if they met the following criteria:1) randomized controlled trials (RCT), observational controlled trials (OCT) and prespecified subgroup analyses comparing CABG with PCI-DES for diabetics with LM and/or MVD;2) studies published in peer-reviewed journals with full available text; and3) follow-up period≥12months. Studies were excluded if they met any one of the following criteria:1) the subjects were not exclusively diabetics with LM and/or MVD,2) using only BMS or involving BMS with DES in one PCI subject,3) duplicate publication,4) less than50patients in each cohort.2.3. Studies selectionTwo reviewers screened the citations and abstracts identified by the search strategies. Full text reviews were also conducted by two reviewers to establish eligibility when screening reviewers believed that a citation potentially met inclusion criteria. Disagreements regarding inclusion were resolved via consensus.2.4. Data ExtractionThree reviewers independently extracted data from the eligible studies. The following information were extracted from each study:first author, year of publication, duration of follow-up, number of participants in each group (CABG or PCI-DES), baseline characteristics, and outcome events including:all-cause mortality(the primary outcome), non-fatal myocardial infarction, non-fatal stroke and repeat revascularization. Each OCT was named by the family name plus the publication year (family name+year) respectively, and the RCT was presented as its own study name. For studies that were reported in>1publications, we extracted data from the most complete publication and used other publications as supplements. We also tried our best to contact the authors by email for information, if their articles did not report the information in detail.2.5. Statistical MethodsData were summarized using descriptive statistics. Discrete variables were presented as proportions (%, count/sample size) and compared by the χ2test or Fisher’s exact test, where appropriate. Continuous variables were presented as mean±standard deviation and compared using the Student’s t-test. The verified data were analyzed using Revman software (version5.2). The endpoints of each study were analyzed using risk ratio (RR) with95%confidence interval (CI). The Cochrane chi-square (Cochrane Q) test was used to assess the between-trial heterogeneity. The I2statistic was calculated as a measure of the proportion of the overall variation attributable to the between-trial heterogeneity rather than to chance, and we used the reported guidelines for low (I2=25-49%), moderate (I2=50-74%), and high (I2>75%) heterogeneity. The overall effects size (risk ratio RR) was calculated by fixed-effect model with the Mantel-Haenszel method when there was no significant heterogeneity (p>0.10or I2<50%), or with DerSimonian-Laird weights for the random-effects model when there was a significant heterogeneity (p≦0.10or I2≧50%). Forest plots were then created for graphical presentations of clinical outcomes. Publication bias with respect to the primary outcome (all-cause death) was assessed visually using a funnel plot. When there is no publication bias, studies of all sizes are scattered equally right and left of the line indicating the pooled estimate of natural log RR. For each endpoint, we conducted subgroup analyses in RCTs and OCTs respectively, apart from an overall analysis. A sensitivity analysis was performed when the between-trial heterogeneity was significant, so as to evaluate the influence of removing individual studies on the pooled RR.3. ResultsWe identified19eligible studies (4randomized and15nonrandomized)(Figure1) enrolling5805patients in OCT subgroup (PCI-DES:2961, CABG:2844) and3060in RCT subgroup (PCI-DES:1541, CABG:1519) respectively. The mean follow-up durations ranged from1year to5.6years. In the OCTs with the exception of the propensity score-matched studies, the C ABG cohorts had the higher prevalence of the triple vessel and/or LM disease, higher EuroSCORE and/or higher SYNTAX Score, but not in the RCTs (Table2). The CARDia trial included patients undergoing PCI with BMS initially (BMS31%, DES69%). The VA CARDS trial was stopped because of slow recruitment after enrolling only25%of the intended sample size, leaving it severely underpowered for the primary composite endpoint of death plus nonfatal MI.3.1. All-cause DeathRandom-effects meta-analysis yielded two different outcomes for all-cause death in RCTs and OCTs subgroup. The prevalence of death in RCTs subgroup was14.0%in the PCI-DES cohort and9.6%in CABG cohort with an RR of1.51(95%[CI],[1.09,2.10], P=0.01), which demonstrated a risk ratio reduction of51%for death in CABG cohort. A moderate heterogeneity among the RCTs, mainly driven by including the VA CARDS trial, was revealed by sensitivity analysis (I2=52%, P=0.10). After excluding VA CARDS trial, random-effects model generated an RR1.36[1.11,1.66] for death (P=0.003) with no residual heterogeneity (I2=0%, P=0.53).While the OCT subgroup analysis indicated a comparable mortality between PCI-DES and CABG (10.3%DES vs.8.8%CABG, RR1.08,95%CI [0.83,1.42], P=0.55). Heterogeneity analysis revealed a low heterogeneity among the OCTs (P=45%, P=0.03) which was largely due to the inclusion of the Javaid2007trial. After the exclusion of this trial, random-effects meta-analysis yielded an RR1.02[0.85,1.22] for death (P=0.80) with no residual heterogeneity (I2=0%P=0.73).When pooling all the RCTs and OCTs, random-effects model yielded an RR1.23(11.7%DES vs.9.1%CABG,95%CI [1.00,1.53], P=0.06) for death. After excluding the VA CARDS trial and the Javaid2007trial, the overall mortalities of the two arms reached statistical difference (RR1.15,95%CI [1.01,1.31], P=0.04) with no residual heterogeneity between trials (I2=0%, P=0.51)(Figure2, Table3). RCT patients risked a higher mortality than OCT patients (9.6%OCTs vs.11.9%RCTs, RR0.81,95%CI [0.71,0.92], P=0.001).There was an asymmetry of the points on visual estimation of the funnel plot,which indicated the possibility of publication bias with respect to the primary outcome (all-cause death).(Figure3)3.2. Nonfatal Myocardial InfarctionThe RCT subgroup analysis demonstrated no difference in MI incidence (10.3%DES vs.5.9%CABG RR1.44,95%CI [0.79,2.60], P=0.23), with a significant between-trial heterogeneity (I2=75%, P=0.007). Sensitivity analysis show that this heterogeneity was largely contributed by the inclusion of the VA CARDS trial, where MI rate after CABG was much higher than that after PCI-DES, unlike other RCTs. When excluding this trial, Random-effects meta-analysis demonstrated a statistical difference (10.6%DES vs.5.3%CABG, RR2.0195%CI1.54to2.62, P<0.00001) with no residual heterogeneity (I2=0%, P=0.83). The OCT subgroup analysis showed that MI after PCI-DES was more prevalent as opposed to CABG (7.4%DES vs.3.7%CABG, RR1.82,95%CI [1.15,2.86], P=0.01) with a moderate heterogeneity (I2=54%, P=0.01), for which no special trial was mostly responsible.Regardless of the inclusion or exclusion of the VA CARDS trial, random-effects meta-analysis showed that the overall MI rate was consistently higher in PCI-DES ’patients (inclusion:8.5%DES vs.4.6%CABG, RR1.68,95%CI [1.20,2.37], P=0.003; exclusion:8.6%DES vs.4.3%CABG, RR1.91,95%CI [1.43,2.57], P<0.0001).(Figure4,Table3)3.3. Nonfatal StrokeFixed-effects meta-analysis revealed that, both in the RCTs and OCTs subgroup, PCI-DES was associated with much lower risk of stroke compared with CABG (RCTs:2.3%DES vs.3.8%CABG, RR0.59,95%CI [0.39,0.90], P=0.01; OCTs:1.8%DES vs.4.0%CABG, RR0.46,95%CI [0.32,0.66], P<0.0001) with no heterogeneity between trials (RCTs:I2=0%,P=0.94; OCTs:I2=0%,P=0.57). When pooling two subgroups, PCI-DES patients kept consistently a lower frequency of stroke (2.0%DES vs.3.9%CABG, RR0.51,95%CI [0.39,0.67], P<0.00001) with no heterogeneity between trials (I2=0%, P=0.80).(Figure5)3.4. Repeat RevascularizationThe RCT subgroup analysis showed that patients after PCI-DES risk a several-fold higher rate of subsequent revascularization (17.4%DES vs.7.4%CABG, RR2.11,95%CI [1.41,3.15], P=0.0003), even if including the VA CARDS trial which primarily contributed to the significant heterogeneity among RCTs(I2=71%, P=0.01). After the exclusion of this trial, the point estimate for repeat revascularization reached a statistical significance (17.32%DES vs.6.57%CABG, RR2.61,95%CI [2.09,3.27], P<0.00001) with no residual heterogeneity (I2=0%, P=0.89). The OCT subgroup analysis suggested patients after PCI-DES was at a substantially higher risk of repeat revascularization (19.8%DES vs.5.7%CABG, RR3.22,95%CI [2.73,3.80], P=0.0001) with no heterogeneity among OCTs (I2=0%, P=0.53).Random-effects meta-analysis showed that the overall incidence of subsequent revascularization was consistently much higher in PCI-DES cohort regardless of the inclusion or exclusion of the VA CARDS trial (inclusion:19.0%DES vs.6.3%CABG, RR2.9595%CI [2.46,3.55] P<0.00001; exclusion:19.0%DES vs.6.0%CABG, RR2.99,95%CI [2.62,3.42], P<0.0001).(Figure6, Table3).4. ConclusionCABG for patients with diabetes mellitus and LM and/or MVD consistently had the evident advantages over PCI-DES in death from any cause, nonfatal MI, and repeat revascularization, but the substantial disadvantage in nonfatal stroke. Given comparable mortality between two strategies for the OCT patients and the lower overall mortality in OCT subgroup, it was reasonable that CABG was preferred to diabetic patients with higher-risk lesion in daily clinical practice.