OUP user menu

★ Editor's choice ★

Off-pump vs. on-pump coronary artery bypass surgery: an updated meta-analysis and meta-regression of randomized trials

Jonathan Afilalo, Mandana Rasti, Samuel M. Ohayon, Avi Shimony, Mark J. Eisenberg
DOI: http://dx.doi.org/10.1093/eurheartj/ehr307 1257-1267 First published online: 10 October 2011


Aims The benefits of off-pump coronary artery bypass (OPCAB) continue to be debated, in part due to the fact that pooled effects fail to consider differences in trial and patient characteristics. We sought to analyse the contemporary evidence for OPCAB vs. conventional coronary artery bypass (CCAB), incorporating recent larger trials, and adjusting for differences in trials using a technique known as meta-regression.

Methods and results We systematically reviewed MEDLINE, EMBASE, and the Cochrane database for published and unpublished randomized trials of OPCAB vs. CCAB in which 30-day or in-hospital clinical outcomes were reported. The outcomes of interest were: all-cause mortality, stroke, and myocardial infarction. In addition to measuring the pooled treatment effects using a random effects meta-analysis model, we measured the effect of selected trial-level factors on the effects observed using the meta-regression technique. Fifty-nine trials were included, encompassing 8961 patients with a mean age of 63.4 and 16% females. There was a significant 30% reduction in the occurrence of post-operative stroke with OPCAB [risk ratio (RR) 0.70, 95% CI: 0.49–0.99]. There was no significant difference in mortality (RR: 0.90, 95% CI: 0.63–1.30) or myocardial infarction (pooled RR: 0.89, 95% CI: 0.69–1.13). In the meta-regression analysis, the effect of OPCAB on all of the clinical outcomes was similar regardless of mean age, proportion of females in the trial, number of grafts per patient, and trial publication date.

Conclusion Our meta-analysis incorporating recent trials suggests that there appears to be a beneficial effect of OPCAB on stroke. Moreover, our meta-regression does not support the hypothesis that differences in study populations are responsible for the observed outcomes, although pooled individual patient-data would be better suited to confirm these findings.

  • Coronary artery bypass
  • Off-pump
  • Meta-analysis
  • Meta-regression
  • Mortality
  • Stroke


The theoretical benefits of OPCAB were and continue to be centred around the avoidance of cardiopulmonary bypass and aortic manipulation. Human and animal studies have documented the negative effects of cardiopulmonary bypass on markers of inflammation, coagulation, microembolization, thermoregulation, acid–base balance, and regional perfusion.1 Observational studies have suggested that, by avoiding these negative effects, OPCAB substantially reduces rates of mortality and morbidity when compared with conventional coronary artery bypass surgery (CCAB). The encouraging initial studies were followed by a sharp rise in OPCAB procedures performed in the USA, reaching 25% of coronary artery bypass surgeries in 2001. Since then, a number of randomized trials have failed to support the theoretical and observational benefits of OPCAB, showing no effect on mortality and mixed effects on morbidity. These mixed effects are generally divided between positive effects on bleeding and atrial fibrillation, and non-significant effects on myocardial infarction and stroke. In light of these reports, the enthusiasm for OPCAB has been tempered, with the proportion of OPCAB procedures plateauing at ∼20% in recent years.2

The evidence base surrounding OPCAB remains equivocal, ridden with biased observational studies, and relatively small randomized trials. Attempts to systematically pool the evidence from randomized trials have been limited because of two main reasons. First, the number of patients included in prior systematic reviews and meta-analyses has been small and consequently the confidence intervals (CIs) generated have been wide. Prior meta-analyses included between 3369 and 3996 patients and their reported 95% CIs for mortality were wide (0.58–1.80 and 0.58–1.60, respectively)3,4 A number of larger randomized trials of OPCAB vs. CCAB have recently been published but have yet to be incorporated into meta-analyses. Second, given the heterogeneity between individual patients, the use of meta-analysis to globally pool data and provide a single overarching conclusion has been criticized. Certain patient subsets are thought to benefit more from OPCAB, including those with advanced age, female sex, prior stroke, renal impairment, pulmonary disease, high-risk features, and patients in whom a higher number of off-pump grafts can be performed to achieve complete revascularization.2

Therefore, we sought to systematically review the contemporary evidence for OPCAB vs. CCAB, incorporating recent larger trials, and adjusting for factors known to modulate the effect of OPCAB using a technique known as meta-regression. Meta-regression allows us to evaluate the effect of OPCAB as a function of trial-level factors such as mean age, sex, and number of grafts performed. There have been no previous meta-regression analyses on this subject, and we hypothesized that this could add an informative dimension to enable clinicians to better individualize care for their patients when considering OPCAB.


Search strategy

MEDLINE, EMBASE, the Cochrane Database, the Internet (www.clinicaltrials.gov, www.clinicaltrialresults.org, www.cardiosource.com, www.medscape.com, www.theheart.org), and abstracts from major cardiology conferences until February 2011 were queried to identify published and unpublished trials. The following search string was used in PubMed: (‘off-pump’ OR ‘cardiopulmonary bypass’) AND (‘randomized’ OR ‘clinical trial’). The search was limited to human studies in any language. To ensure that the search was complete, reference lists from retrieved manuscripts and PubMed's related articles search feature were used. Investigators were contacted to obtain unpublished trial-level data from their respective trials when needed.

Selection criteria

The inclusion criteria for our analysis were: (i) randomized clinical trial, (ii) experimental group allocated to OPCAB via the standard sternotomy approach, (iii) control group allocated to CCAB via the standard sternotomy approach. The exclusion criteria were: (i) concomitant surgical intervention other than coronary artery bypass, (ii) concomitant medical intervention in one but not both of the two groups; for example, OPCAB plus a given drug vs. CCAB. Three reviewers (J.A., M.R., S.O.) reviewed the eligible trials and determined whether they met the selection criteria. The manuscript was prepared in accordance with standards set forth by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement.5

Study variables

The variables of interest spanned domains related to the baseline characteristics, operative course, and post-operative outcomes. The outcome measures were all-cause mortality within the first 30 days after the index surgery, post-operative stroke, and post-operative myocardial infarction. The pre-determined modulating factors to be examined were: age, sex, number of bypass grafts, and trial publication date. Age was represented as the mean age of the patients participating in the trial. Sex was represented as the proportion of females in the trial. Number of bypass grafts was represented as the difference between the mean number of grafts (arterial and venous combined) performed in the OPCAB group minus the mean number of grafts performed in the CCAB group in the trial, termed graft differential. Moreover, an analysis of the occurrence of post-operative stroke as a function of prior history of stroke was planned. Data were extracted in duplicate by two investigators (M.R., S.O.) and independently verified by a third investigator (J.A.). Disagreements were resolved by consensus.

Validity assessment

The Cochrane Collaboration's tool for assessing risk of bias was systematically applied.6 Trials were graded based on: (i) sequence generation, (ii) allocation concealment, (iii) blinding, (iv) incomplete outcome data, (v) selective outcome reporting, and (vi) other sources of bias. The global assessment for each trial was summarized as low, unclear, or high risk of bias and presented in Table 1.

View this table:
Table 1

Trial and baseline patient characteristics

YearnAge% Female# GraftsPrior strokeRisk of bias
BHACAS 124200220062.−0.10%Unclear
BHACAS 224200220162.516.42.93.0−0.10%Unclear
MASS III45201030860.−0.5NALow
Sousa Uva68201014765.316.

Statistical analysis

The statistical approach was divided into two parts. To determine the overall effect of OPCAB, the trial data were pooled and a meta-analysis was performed using a random-effects model. In this type of model, trials with zero events in both groups are excluded from the pooled analysis. Some have added a correction factor of 0.5 to both zero event groups in order to force these trials into the analysis; this results in a substantial number of ‘pseudoevents’ and dubious validity biased towards the null hypothesis. More transparent approaches to overcome zero event trials have been suggested.7 Both the collapsed 4-by-4 table and fixed-effects meta-analysis approaches were performed as a sensitivity analysis. Heterogeneity was assessed using the I2 test. Results were reported as risk ratios (RRs) with their 95% CIs.

Next, meta-regression analyses were performed to determine whether the effects of OPCAB were modulated by the pre-specified factors. Meta-regression graphs depict the effect of OPCAB on the outcome (plotted as a log RR on the y-axis) as a function of a given factor (plotted as a mean or proportion of that factor on the x-axis). Meta-regression coefficients show the estimated increase in log RR per unit increase in the covariate. Since log RR >0 corresponds to RR >1 and log RR <0 corresponds to RR <1, a negative coefficient would indicate that as a given factor increases, the RR decreases, i.e. OPCAB is more beneficial in reducing the outcome of interest. All statistical analyses were performed with the STATA 11 statistical software package (College Station, TX, USA).


Of 475 potentially relevant studies, 59 randomized clinical trials met the selection criteria and were included (Figure 1). These trials encompassed a total of 8961 patients of which 4461 were randomized to OPCAB and 4500 to CCAB. The mean age was 63.4 ± 3.8 years (range 56.0–75.4 years) and the proportion of females was 16% ± 8% (range 0–39%). The mean number of grafts was 2.8 ± 0.6 in the OPCAB group and 3.0 ± 0.6 in the CCAB group, representing a mean graft differential of −0.2 grafts. Trial and baseline patient characteristics are shown in Table 1.

Figure 1

Flow diagram for study selection. When multiple publications from the same study were identified, the primary manuscript that reported the outcome of interest (mortality, stroke, myocardial infarction) was selected for inclusion in the meta-analysis. One unpublished trial31 met the selection criteria and was included. RCT, randomized clinical trial.

All-cause deaths were observed in 26 trials comprising a total of 117 events among 6832 patients. The incidence of post-operative mortality was 1.6% in the OPCAB group and 1.9% in the CCAB group, representing a non-significant difference (RR: 0.90, 95% CI: 0.63–1.30) (Figure 2). Meta-regression coefficients were not statistically significant for mean age (coefficient: −0.04, 95% CI: −0.11 to 0.04), proportion of females (coefficient: −0.02, 95% CI: −0.05 to 0.01), trial date (coefficient: −0.01, 95% CI: −0.14 to 0.12), and graft differential (coefficient: 0.86, 95% CI: −1.12 to 2.83).

Figure 2

Forest plot for all-cause mortality. Fifty-four deaths were observed among 3431 off-pump coronary artery bypass patients compared with 63 among 3401 conventional coronary artery bypass patients, representing a non-significant difference.

Post-operative strokes were observed in 27 trials comprising a total of 125 events among 7194 patients. The incidence of post-operative stroke was 1.4% in the OPCAB group and 2.1% in the CCAB group, representing a significant 30% reduction (RR: 0.70, 95% CI: 0.49–0.99) (Figure 3). Meta-regression coefficients were not statistically significant for mean age (coefficient: 0.00, 95% CI: −0.07 to 0.07), proportion of females (coefficient: −0.02, 95% CI: −0.06 to 0.01), trial date (coefficient: 0.06, 95% CI: −0.07 to 0.19), and graft differential (coefficient: 0.84, 95% CI: −1.11 to 2.78). The proportion of patients with a prior history of stroke did not affect the observed RRs for post-operative stroke (coefficient: 8.28, 95% CI: −2.90 to 19.46).

Figure 3

Forest plot for stroke. Forty-nine strokes were observed among 3605 off-pump coronary artery bypass patients compared with 76 among 3589 conventional coronary artery bypass patients, representing a 30% relative risk reduction.

Post-operative myocardial infarctions were observed in 29 trials comprising a total of 253 events among 6961 patients. The incidence of post-operative myocardial infarction was 3.4% in the OPCAB group and 3.9% in the CCAB group, representing a non-significant difference (RR: 0.89, 95% CI: 0.69–1.13) (Figure 4). Meta-regression coefficients were not statistically significant for mean age (coefficient: 0.03, 95% CI: −0.02 to 0.07), proportion of females (coefficient: −0.01, 95% CI: −0.04 to 0.02), trial date (coefficient: 0.02, 95% CI: −0.07 to 0.11), and graft differential (coefficient: 0.58, 95% CI: −1.11 to 2.27).

Figure 4

Forest plot for myocardial infarction. One hundred and eighteen myocardial infarctions were observed among 3482 off-pump coronary artery bypass patients compared with 135 among 3479 conventional coronary artery bypass patients, representing a non-significant difference.

There was no statistical heterogeneity for the outcomes of interest (I2 = 0% for all-cause mortality, stroke, and myocardial infarction). Begg's and Egger's tests did not reveal any evidence of publication bias, with the exception of post-operative myocardial infarction for which Egger's test was statistically significant (Figure 5). In sensitivity analyses, the beneficial effect of OPCAB on stroke was preserved after incorporating zero event trials using a collapsed 4-by-4 analysis (RR: 0.64, 95% CI: 0.46–0.94) and after employing a fixed-effects model (RR: 0.67, 95% CI: 0.48–0.94) (Figure 6).

Figure 5

Representative meta-regression plots. (top) Meta-regression plot for mortality by proportion of females included in the trial. The coefficient was not statistically significant (coefficient: –0.02, 95% CI: –0.05 to 0.01). (middle) Meta-regression plot for stroke by mean age in the trial. The coefficient was not statistically significant (coefficient: 0.00, 95% CI: –0.07 to 0.07). (bottom) Meta-regression plot for myocardial infarction by mean graft differential in the trial (mean number of coronary bypass grafts in OPCAB minus CCAB). The coefficient was not statistically significant (coefficient: 0.58, 95% CI: –1.11 to 2.27).

Figure 6

Publication bias funnel plots. No evidence of statistically significant publication bias for trials reporting all-cause mortality (top; Begg's test P = 0.48, Egger's test P = 0.12) or stroke (middle; Begg's test P = 0.10, Egger's test P = 0.33). Possible element of publication bias for trials reporting myocardial infarction (bottom; Begg's test P = 0.63, Egger's test P = 0.006).


To our knowledge, this is the largest meta-analysis of randomized data performed to date, providing incremental value by demonstrating that OPCAB reduces the incidence of post-operative stroke compared with CCAB. Furthermore, this analysis confirms that OPCAB does not significantly reduce the incidence of short-term all-cause mortality and post-operative myocardial infarction. The potential benefits of OPCAB on these outcomes do not appear to be globally determined by patient age, sex, trial date, and grafts performed.

The effect of OPCAB on stroke has been a topic of controversy, with recent reports offering opposite conclusions, some showing a reduction in stroke4 but most showing no effect.3,810 The largest trial to date showed no effect or trend for reduction in stroke.63 A post hoc sample size calculation sheds light on why this is not surprising—10 448 patients per group would be required to demonstrate a 30% reduction in stroke with a two-tailed alpha of 0.05 and a power of 80% (assuming an incidence rate of 1.4%). Thus, even the largest trial was not powered to detect differences in this infrequent outcome. Notwithstanding, there is a sound epidemiological and theoretical rationale to support the benefit of OPCAB on stroke. Conventional coronary artery bypass mandates at least four transgressions to the ascending aorta: anastomosis of the proximal grafts, cross-clamping, cannulation for return of oxygenated blood, and cannulation for cardioplegia. Depending on the technique used, OPCAB eliminates two to three aortic transgressions and therefore reduces the risk of atheromatous emboli from the aorta to the brain. Gaseous emboli and activation of coagulation pathways also increase the risk of stroke with CCAB compared with OPCAB.1114 Large propensity-matched observational studies have shown that OPCAB reduces the incidence of stroke compared with CCAB,15 and in particular, that it is the early strokes of embolic cause that are reduced and not the delayed strokes of variable causes.16 The specific association between CCAB and early embolic strokes suggests a causal link involving aortic manipulation that should empirically be reduced by OPCAB.

Although there have been a number of meta-analyses on this subject, this analysis is unique for several reasons. The pooled sample size was three-fold larger than previously published meta-analyses. The larger sample size translated into greater statistical power and narrower CIs, reducing the amount of uncertainty surrounding treatment effects. Several recently published and unpublished trials had not been included in prior meta-analyses and were included in this analysis (trials from 2009 to 2010 contributed 4085 out of the 8961 patients). In comparison with other recent efforts that have focused on a smaller subset of trials and a specific subdomain such as graft patency,17 this meta-analysis synthesized the global body of knowledge as it pertains to hard endpoints with the potential to influence clinical practice on a large scale. Moreover, a broad meta-regression had yet to be reported in the medical literature, a component that adds to the field.

According to some sources, female patients and those with advanced age are thought to face higher risks associated with cardiopulmonary bypass and therefore benefit more from OPCAB. The benefit of OPCAB is thought to be greater when more grafts are performed and when contemporary technique is used (with trial date as an indirect index for the latter). The meta-regression analysis in this study was unable to confirm these hypotheses that differences in study population or operative technique are responsible for the treatment effects observed across trials.


The list of factors evaluated in the meta-regression was limited to age, sex, grafts, and trial date because these factors were judged to be of primary importance and were reported in the majority of trials. Other factors such as renal function, pulmonary disease, surgical risk score, and aortic atheroma would have been interesting to evaluate but were not consistently reported. The meta-regression technique reflects trial-level characteristics rather than individual patient data. Individual patient-level data, when available, is preferable to evaluate potential treatment interactions. A sizeable proportion of trials were designed to capture biomarker changes or radiographic parameters rather than clinical endpoints, resulting in trials with no or few accrued clinical events and/or suboptimal ascertainment of these events (uncertain risk of bias). This is particularly relevant for stroke, which was found to be sensitive to inclusion or exclusion of certain trials, with the upper limit of the 95% CI bordering the null. Although the statistical heterogeneity between trials was nil, the differences between trials in terms of operative technique and volume (often not explicitly reported) may have led to an element of clinical heterogeneity which could not be accounted for. Lastly, secondary outcomes such as bleeding and atrial fibrillation were not evaluated in this meta-analysis, and differences in these outcomes may be important considerations to guide the clinical application of OPCAB.

In conclusion, OPCAB reduces the incidence of post-operative stroke by 30% and has no notable effect on mortality or myocardial infarction. Of note, modest effects on mortality and myocardial infarction to the order of 10–20% cannot be ruled out with this pooled sample size (post hoc sample size calculations suggested that this pooled sample size would be powered to detect 38 and 27% relative differences in mortality and myocardial infarction, respectively). Given the apparent reduction in stroke, the use of OPCAB should be pursued, especially in patients at higher risk of this complication. Patient age, sex, number of grafts performed, trial date, and prior history of stroke do not appear to explain the effect of OPCAB on stroke or other outcomes, although pooled individual patient-data would be better suited to address this issue. Other factors such as aortic atheroma (as identified by computed tomography or epiaortic ultrasound) appear promising to predict which patients will derive the greatest benefit from OPCAB for stroke reduction18,19 and deserve further study.


M.J.E. is a National Researcher supported by the Canadian Institutes of Health Research (CIHR) and the Fonds de la recherche en santé du Québec (FRSQ).

Conflict of interest: none declared.


  • See page 1181 for the editorial comment on this article (doi:10.1093/eurheartj/ehr374)


View Abstract