Does Off-pump Coronary Artery Bypass Reduce Mortality, Morbidity, and Resource Utilization When Compared with Conventional Coronary Artery Bypass? A Meta-analysis of Randomized Trials

This meta-analysis of randomized trials was performed in accordance with state-of-the-art methodologic recommendations, including the Quality of Reporting of Meta-Analysis (fig. 1) Consensus Statement and Cochrane Collaboration recommendations9,10and according to a protocol that prespecified outcomes, search strategies, inclusion criteria, and statistical analyses. A search was undertaken in accordance with Cochrane Collaboration recommendations to identify all published or unpublished randomized trials of OPCAB versus  CCAB. English and non-English articles were included. MEDLINE, Cochrane CENTRAL, EMBASE, Current Contents, DARE, NEED, and INAHTA databases were searched from the date of their inception to the end of May 2004. Search terms included variants of “off-pump,”“minimally invasive,”“beating heart,” and “coronary artery bypass.” Tangential electronic exploration of related articles and hand searches of bibliographies, scientific meeting abstracts, and related journals were also performed.

Studies were included if they met each of the following conditions: randomized allocation to OPCAB on the beating heart versus  CCAB on the asystolic heart with cardiopulmonary bypass circuit, adult patients undergoing single or multiple vessel bypass, and reporting at least one pertinent clinical or economic outcome. Off-pump studies using minimally invasive direct coronary artery bypass with thoracotomy and studies allowing for ventricular assist devices were included with the intent of subanalyzing these groups. Blinded and unblinded studies were included. Hybrid (i.e. , OPCAB plus balloon angioplasty) and robotically assisted surgery studies were excluded.

Three authors independently identified trials for inclusion and extracted information on demographics, interventions, and outcomes. Authors of included trials were contacted when necessary to clarify data and to identify multiple publications. When later publications added more information to original publications of randomized trials, the updated trial was included. Two reviewers independently assigned each trial a Jadad quality score that evaluates randomization, blinding, and completeness of follow-up (maximum score, 5).14Disagreements were resolved by consensus.

The primary outcome was defined as all-cause mortality at 30 days, 6 months, and greater than 1 yr. Secondary outcomes included postoperative incidence of stroke, acute myocardial infarction, atrial fibrillation, renal failure, need for inotropes, need for intraaortic balloon pump, mediastinitis/wound infection, respiratory infections, angina recurrence, reintervention for ischemia, restenosis, need for transfusions, reexploration for bleeding, neurocognitive dysfunction, duration of ventilation, intensive care unit (ICU) length of stay (LOS), hospital LOS, hospital costs, and quality of life (QOL). Atrial fibrillation was defined according to study author definitions. Acute myocardial infarction was defined per study author definitions of new onset infarction using electrocardiogram or enzymatic criteria. Mediastinitis/wound infection was defined as deep or superficial wound infections of the chest, excluding leg wound infections. Respiratory infection was defined according to author definitions, whether or not confirmed by chest radiograph. Need for transfusion was defined as the number of patients requiring blood product transfusion (usually defined as red blood cell transfusion) during the intraoperative and postoperative period combined. When data were presented separately for the intraoperative and postoperative periods (i.e. , no combined estimate was provided), we preferentially extracted only the postoperative transfusion requirements to avoid double-counting patients. Renal failure was defined as a new increase in serum creatinine of >50%, a decline in creatinine clearance of >50%, or requirement for dialysis. Reintervention was defined as the need for repeat surgical coronary artery bypass grafting or the need for balloon angioplasty. Restenosis and patency were defined as per the study authors. QOL was defined as per the study authors, with the intent of combining summary scales such as the EuroQOL or Short Form-36 when numbers were provided. Because there is no standard definition for neurocognitive dysfunction, we planned to include only studies reporting neurocognitive dysfunction dichotomously and when tests in accordance with the statement of consensus were used.15Duration of ventilation was measured from end of surgery to time of tracheal extubation. Intensive care LOS and hospital LOS were measured from end of surgery to ICU or hospital discharge, respectively.

Outcomes were analyzed as dichotomous variables, with the exception of duration of ventilation and LOS, which were analyzed as continuous variables when the mean and SD were provided. For dichotomous variables, odds ratios (OR) and 95% confidence intervals (CI) were calculated. For continuous variables, the weighted mean difference (WMD) was calculated. When significant differences were found for proportions, the number needed to treat (NNT) was calculated∥using the aggregate odds ratio and incorporating the summary estimate of baseline risk from the conventional revascularization arm (95%CI).16–18The NNT provides an estimate of the number of patients that would have to undergo off-pump revascularization rather than conventional revascularization for one additional patient to benefit.

Heterogeneity was explored using the Q-statistic.19Recognizing that the Q-test is often underpowered to detect statistically significant heterogeneity, particularly when there are few trials in the analysis, the relatively conservative threshold of P < 0.10 was chosen to suggest statistically significant heterogeneity across trials. In addition to the Q statistic, the I²was calculated to quantify the degree of heterogeneity across trials that could not be attributable to chance alone. As the I²indicates the proportion of variability between trials that cannot be attributable to chance alone, it provides an improved measure of heterogeneity between trials and is not limited by power.9,20,21 

For each outcome, the fixed effect (Mantel-Haenszel for dichotomous variables and inverse variance for continuous variables) or random effects (DerSimonian and Laird for dichotomous and continuous variables) model was used when the Q-statistic suggested lack or presence of heterogeneity, respectively. Pooled effect estimates and heterogeneity between studies were analyzed by use of Comprehensive MetaAnalysis® (Biostat, 2002, Englewood, New Jersey) and Review Manager (RevMan for Windows, version 4.2.2, 2003, The Cochrane Collaboration, Oxford, England). Other than for the Q statistic, statistical significance was defined as P < 0.05. All tests of statistical significance were two-sided. Whenever possible, data analysis was by intention-to-treat.

Subanalyses defined a priori  included outcomes in patients who were elderly (age >70 yr), undergoing urgent or redo bypass, or had chronic obstructive pulmonary disease or renal failure at baseline, requiring conversion from OPCAB to CCAB, and those in whom ventricular assist devices were used. For each outcome, data were subgrouped by trials entering patients with single, multiple, or mixed single/multiple vessel disease before plotting the graphs. In subgroup analyses, the differences in relative size of effect were tested using a chi-square test for interaction or a chi-square test for trend. Sensitivity analysis was planned to explore the potential effect of trial quality (Jadad Score <3 versus  3 or greater), publication status (published versus  unpublished), and patients excluded in nonintent-to-treat trials using a worst-case scenario assumption. Post hoc  sensitivity analyses were performed for each outcome with an I²confidence interval that included 50% or greater to investigate the impact on heterogeneity of eliminating studies that differed by important clinical or methodological characteristics. Post hoc  sensitivity analysis was also performed for NNT, to explore the effect of replacing the baseline risk derived from the aggregate risk estimates given in our meta-analyses with “real world” baseline risks given in observational studies.16–18Publication bias was explored through visual inspection of funnel plots in which the inverse of the estimated variance of the natural logarithm of the adjusted relative risk was plotted against the natural logarithm of the adjusted relative risk for each outcome.22 

Figure 1outlines the search results. Of more than 2,920 citations screened, 95 apparently relevant randomized trials were identified. Of these, 25 were excluded during online screening of the abstracts for the following reasons: nonrandom comparison,23–33combined procedure,34,35robotic procedure,36or no direct comparison of OPCAB and CCAB.37–47A total of 70 potentially relevant randomized trials were retrieved for evaluation. On further examination of these retrieved papers, 16 were subsequently excluded for the following reasons: nonrandom design,48–55combined procedure,56no conventional revascularization group,57no relevant outcomes reported,58–61and no extractable data reported.62,63Of the remaining reports, we identified 37 independent randomized trials reported in 54 published papers.64–117One paper reported two trials.64No unpublished trials were found. Therefore, a total of 3,369 patients in 37 original trials provided data for this meta-analysis (table 1).Baseline characteristics were comparable between groups (table 2), except for the mean number of distal vessels anastomosed (2.6 ± 0.6 for OPCAB and 2.8 ± 0.7 for CCAB surgery, P = 0.0001).

Two included studies reported data for patients with chronic obstructive pulmonary disease,75,83and in two studies a ventricular assist device was used.73,92Three trials used minimally invasive direct coronary artery bypass technique.82–84Although most studies stated that they excluded high-risk patients (table 1), one study included only high-risk patients, defined as at least three of the following: age greater than 65 yr, high blood pressure, diabetes, serum creatinine > 2.0 mg/dl, ejection fraction <45%, pulmonary disease, unstable angina, congestive heart failure, repeat bypass surgery, anemia, and significant carotid atherosclerosis.74The median Jadad score was 2 (range, 1–3), as none of the trials were double-blinded.14No statistically significant heterogeneity was found for all-cause mortality; however, statistically significant heterogeneity was found for atrial fibrillation, transfused patients, neurocognitive dysfunction at 30 days, ventilation time, and ICU and hospital LOS (tables 3 and 4). Funnel plots showed no clear evidence of publication bias for any endpoint.

Table 3and figure 2outline primary and secondary outcomes. All-cause mortality at 30 days was not statistically significantly reduced (OR, 1.02; 95%CI, 0.58–1.80) (fig. 2a). Similarly, all-cause mortality at 1–2 yr was not statistically significantly reduced (OR, 0.88; 95%CI, 0.41–1.88) but was reported in only six trials. Six-month mortality was not reported. Postoperative 30-day stroke (OR, 0.68; 95%CI, 0.33–1.40) (fig. 2b) and acute myocardial infarction (OR, 0.77; 95%CI, 0.48–1.26) (fig. 2c) were not statistically significantly reduced. Stroke at 1–2 yr was not reduced (OR, 0.50; 95%CI, 0.15–1.70); however, clinically important differences cannot be ruled out because of the wide confidence intervals.118 

In contrast, OPCAB compared with CCAB significantly reduced postoperative atrial fibrillation (OR, 0.58; 95%CI, 0.44–0.77; NNT = 11; 95%CI, 8–17) (fig. 2d), patients transfused (OR, 0.43; 95%CI, 0.29–0.65; NNT = 7; 95%CI, 6–10) (fig. 2e), respiratory infections (OR, 0.41; 95%CI, 0.23–0.74; NNT = 19; 95%CI, 12–52), need for inotropes (OR, 0.48; 95%CI, 0.32–0.73; NNT = 12; 95%CI, 8–21), duration of ventilation (WMD, −3.4 h; 95%CI, −5.0 to −1.7 h), ICU LOS (WMD, −0.3 days; 95% CI, −0.6 to −0.1 days), and hospital LOS (WMD, −1.0 days; 95% CI, −1.5 to −0.5 days) (fig. 2f).

No significant difference was found for renal dysfunction, intraaortic balloon pump, mediastinitis, and reexploration for bleeding within 30 days or for angina recurrence or reintervention within 30 days or up to 2 yr. Although no significant difference was found for 30-day neurocognitive dysfunction (OR, 0.57; 95%CI, 0.21–1.54), significant reduction was found at 2 to 6 months (OR, 0.56; 95%CI 0.35–0.89), but this effect was not maintained beyond 12 months (OR, 0.91; 95%CI, 0.57–1.46).

Patency was reported in four randomized trials.87,91,98,101The timepoint of angiography (before discharge to 1 yr postbypass) and the completeness of follow-up (64–100%) varied considerably across trials. The results of angiographically measured patency also differed across trials (table 5). One trial reported decreased patency at 3 months in the OPCAB group compared with the CCAB group.87Three larger trials reported no significant differences in patency at varying timepoints (table 5).91,98,101Inadequate data were available to allow for meta-analysis of patency results.

Quality of life was reported in four trials. Puskas et al.  reported EuroQOL-6 and Short Form-36 at 30 days and 1 yr and showed that QOL improved similarly in OPCAB and CCAB groups over time.101van Dijk and colleagues reported EuroQOL scores (original version) at baseline and 1, 3, 6, and 12 months and showed similar improvements in QOL in both groups over time.98,108Ascione et al.  reported EuroQOL-5 and Short Form-36 scores for the Beating Heart Against Cardioplegic Arrest Studies 1 and 2 trials at a median follow-up of 3 yr and showed that scores for both groups were very similar and did not differ significantly for all dimensions evaluated.71Pooled analysis of QOL data were not feasible as a result of differences in reporting QOL measures and duration of follow-up.

When subgroup analysis by number of coronary arteries grafted was performed, no statistically significant differences in effect sizes were identified between subgroups except for patients transfused when evaluated by number of grafts performed (test for interaction, P = 0.001). However, this subgroup analysis should be interpreted cautiously because many trials provided only aggregate estimates for single-vessel and multiple-vessel disease that were not separable into their respective groups. As a result of insufficient data, subgroup analysis was not possible for age, renal dysfunction, pulmonary disease, and patients undergoing redo or urgent bypass. Sensitivity analysis showed that results did not change when trials allowing for ventricular assist devices or minimally invasive direct coronary artery bypass trials were excluded. Adding excluded patients in prespecified sensitivity analysis showed that the results were robust across reasonable assumptions. On average, of the 20 studies (total 2,579 patients) that reported conversion rates, 8.0% of OPCAB patients and 1.7% of CCAB patients required conversion to on-pump bypass and off-pump bypass, respectively. However, no studies reported outcomes for this subset. No unpublished trials were found; therefore, sensitivity analysis by publication status was unnecessary. Sensitivity analysis by Jadad score showed no clear association between trial quality and outcome (test for interaction, P > 0.05 for each outcome).

Post hoc  sensitivity analysis was performed for each outcome with high heterogeneity (i.e, I²confidence intervals including 50% or greater) to explore potential clinical and methodologic contributors. To evaluate heterogeneity, the I²statistic is superior to the Q statistic, as it is independent of the number of trials included in the analysis and thereby mitigates erroneous conclusion of homogeneity in meta-analyses of low power. Possible values of I²range from 0% to 100%, where 0% would indicate absence of heterogeneity and higher values indicate increasing degrees of heterogeneity. Atrial fibrillation, transfusion, neurocognitive dysfunction, respiratory infection, inotrope use, LOS, and ventilation time had high heterogeneity as indicated by the I²test (table 3). In a sensitivity analysis of atrial fibrillation, van Dijk et al.  107(the only trial in which routine atrial fibrillation prophylaxis was employed, with sotalol) and Gerola et al.  81(significantly more OPCAB patients had a history of myocardial infarction than the CCAB group at baseline) were eliminated from the analysis to explore the effect on heterogeneity and outcomes. The resulting I²de creased from 36% (95%CI, 0–68%) to 27% (95%CI, 8–54%) and the resulting OR was 0.53 (95%CI, 0.39–0.69). Other sensible sensitivity analyses could not be defined based on identifiable clinical or methodologic differences among studies. In a sensitivity analysis of inotrope use, when van Dijk et al.  107was removed (the only trial reporting use of thoracic epidurals in the OPCAB group only), I²decreased from 43% to 33% and the resulting OR became 0.43 (95%CI, 0.28–0.67). Thus, the original inclusion of van Dijk et al.  107resulted in a more conservative estimate of benefit attributable to OPCAB with respect to atrial fibrillation and inotropes use. For sensitivity analysis of patients transfused, heterogeneity decreased when only studies reporting red blood cell transfusions (rather than the aggregate of any blood product transfusion) were included, suggesting that varying definitions of transfusion contributed most of the heterogeneity for this outcome. Because every trial reporting neurocognitive dysfunction used a different definition for neurocognitive decline and different combinations of tests were employed, sensitivity analysis based on these factors was not possible. For LOS, no clinical or methodological differences were evident to allow for sensitivity analysis. A high I²would be expected for LOS and ventilation time given the lack of defined clinical pathways across trials.

It is recognized that NNTs derived from meta-analyses should be interpreted with caution, as the baseline risk may vary across the trials included in the meta-analysis. It may be preferable to use baseline risks that are more reflective of the real world setting coupled with the OR from the meta-analysis (because there is empiric evidence that the OR is more likely to be consistent across trials and across patient risk groups) to calculate the NNT. When we used baseline risks from observational studies and registries,3–5,119,120the estimates of NNT did not change materially because the baseline risk of atrial fibrillation, transfusions, inotrope use, respiratory infections, and neurocognitive dysfunction in our meta-analysis and “real world” studies were similar.

Five trials reported economic outcomes.69,89,98,101,104Each trial found OPCAB to be less costly versus  CCAB, with a range in reported in-hospital cost reductions of approximately 15–35% (table 6). Pooled analysis of economic outcomes was not practical as a result of differences in methods of collecting and reporting costs.

Compared with conventional bypass surgery, OPCAB decreased the odds of postoperative atrial fibrillation by 42% (95%CI, 23–56%), transfusions by 57% (95%CI, 35–71%), need for inotropes by 52% (95%CI, 27–68%), respiratory infections by 59% (95%CI, 26–77%), neurocognitive dysfunction at 2 to 6 months by 44% (95%CI, 11–65%), average ventilation time by 3.4 h (95%CI, 1.7–5.1 h), average ICU stay by 0.3 days (95%CI, 0.1–0.6 days), and average hospital LOS by 1.0 day (95%CI, 0.5–1.5 days) at no measurable increased risk to the patient. These results were similar across single-vessel and multi-vessel subgroups and whether or not trials using ventricular assist devices were included.

This study has significant implications when projected over a large population. For example, given the aggregate estimates of baseline risk and reduction in odds in this meta-analysis, for every 1,000 patients undergoing off-pump instead of on-pump surgery, there would be approximately 91 (95%CI, 59–125) fewer patients with postoperative atrial fibrillation, 143 (95%CI, 103–179) fewer patients requiring blood transfusions, 83 (95%CI, 48–123) fewer patients requiring inotropes, 53 (95%CI, 19–87) fewer respiratory infections, 100 (95%CI, 91–167) fewer with neurocognitive dysfunction at 2–6 months, and 1,000 (95%CI, 500–1500) fewer hospital days, 300 (95%CI, 100–600) fewer ICU days, and 3,400 (95%CI, 1700–5100) fewer ventilation hours at no known additional risk to patients and likely at reduced overall in-hospital cost.

The reductions in rates of atrial fibrillation, inotrope requirement, blood transfusion, ventilation time, and respiratory infection are congruent with evidence that exposure to cardiopulmonary bypass pump induces a systemic inflammatory response that may predispose to increased myocardial irritability, heart failure, platelet dysfunction, respiratory failure, and immune dysfunction.121,122However, the lack of statistically significant differences in mortality, strokes, renal dysfunction, and early neurocognitive dysfunction are discordant with previous nonrandomized studies and reviews.8,12,123,124The fact that statistically significant differences were not found for these outcomes, however, does not prove that these outcomes are equivalent for off-pump versus  on-pump surgery. It simply suggests that there was insufficient power to prove whether or not true differences exist.118 

Two previous meta-analyses have examined the effect of OPCAB versus  CCAB in randomized and nonrandomized trials. Reston et al.  12suggested significant benefit for OPCAB versus  CCAB for a number of clinical outcomes including death, myocardial infarction, stroke, and atrial fibrillation. However, these results were derived mainly from nonrandomized trials. Caution must be exercised in interpreting these results because of the selection bias inherent in nonrandomized trials (i.e. , healthier patients or those with less complex lesions may be preferentially selected for off-pump surgery). In addition, Reston et al.  provide no information regarding baseline patient characteristics or number of grafts performed, making the determination of similarity between groups difficult. Parolari et al. ,11in a recent meta-analysis of OPCAB versus  CCAB trials, used a combined endpoint of stroke, myocardial infarction, and mortality from randomized trials and concluded no clinical superiority of either surgical technique (OR, 0.48; 95%CI, 0.21–1.09). Although there was a trend that appeared to favor OPCAB versus  CCAB for this composite outcome, the results need to be interpreted with caution, as three of nine included trials were duplicate reports of the same population. A similar error including three duplicated reports of the same population occurred in Reston et al. ,12which serves to highlight the continued risk of inadvertent repeated publications contributing more than once to meta-analyses, even when care is taken to attempt to distinguish independent reports.125In our meta-analysis, we attempted to exclude all repeated data by contacting the authors of included studies. Overall, nearly 30% of all originally identified randomized trials included data previously reported. In a number of cases, it was difficult to detect duplicate data, whereas in other cases the original publication was explicitly referenced in subsequently published papers.

No statistical heterogeneity was identified for the primary outcome of mortality or for most secondary endpoints. However, even in the absence of statistical heterogeneity (i.e. , test for heterogeneity P > 0.10), important methodological and clinical differences between trials should be explored to assess their potential impact on effect sizes. As expected, some endpoint definitions varied across individual trials. In cases where high heterogeneity (I²>50%) is discovered it may be useful to examine individual trials for clinical or methodological differences. For example, in van Dijk et al.  107all study participants were given prophylactic sotalol and thoracic epidurals were administered to patients randomized to the OPCAB arm only. Therefore, our aggregate analysis for atrial fibrillation prevention is a conservative estimate with the inclusion of this trial and is applicable largely to nonprophylaxed patients. Inotrope requirements were decreased in OPCAB patients in all studies except van Dijk et al.  107It is possible that the use of thoracic epidurals with narcotic and local anesthetics in the OPCAB group may have led to the increased use of inotropes, thus negating the potential for benefit from OPCAB surgery.

The high variability identified for neurocognitive dysfunction identified in this analysis is not surprising because of the lack of consistency in testing for neurocognitive dysfunction and the definition of neurocognitive decline used in these trials. Although this analysis showed significantly reduced neurocognitive decline at 2 to 6 months, significant reductions were not found at 1 month or 12 months. As only 4 studies that met our inclusion criteria contributed data to this outcome at any time point, the results are underpowered to rule out clinically important differences. In addition, significant heterogeneity was observed across studies for neurocognitive dysfunction at 30 days. For these reasons, no reliable conclusions can be drawn from this data. More studies with standard definitions and analyses are required to delineate the true effect on neurocognitive outcome. Furthermore, evaluation of neurocognitive changes on QOL would add to the clinical relevance of these findings.

We elected not to combine economic outcomes because the results would be nonsensical based on the diversity of costs included in the reported estimates. Only one trial specified the use of a clinical pathway to manage time to extubation and ICU or hospital discharge,100and therefore variability is expected across trials as a result of differing processes of care.126Given the clinical diversity of these trials and given the lack of consistent standards of care, such as surgical techniques (i.e. , method of myocardial stabilization, arterial versus  venous conduits, experience in off-pump techniques), anesthetic regimens, concomitant medications (i.e. , prophylaxis for atrial fibrillation, antianginals, antiplatelets), blood-conserving techniques (heparinization, antifibrinolytics, cell salvage), and clinical pathways to manage ventilation time and length of stay, it is unknown to what degree the benefits of OPCAB found in this analysis will apply incrementally to the current standard of practice.

This analysis must be interpreted in light of the strengths and limitations of included trials. Although most trials were conducted in recent years and likely represent contemporaneous anesthetic and surgical practices (with the exception that van Dijk et al.  107used thoracic epidurals in the OPCAB group), their generalizability ultimately depends on surgeon expertise.8Most trials specified that a single surgeon who was experienced in off-pump surgery performed all revascularizations; therefore, these results may not apply to inexperienced surgeons.127–130 

In general, only patients with lesions amenable to OPCAB were included, and many trials excluded patients with diseased circumflex beyond the first obtuse margin. Although on average 0.2 fewer grafts (95%CI, 0.1–0.3) were performed in the OPCAB group, the clinical significance of this may be questioned. This imbalance did not lead to a significantly increased risk of adverse clinical outcomes (myocardial infarction, angina recurrence, reintervention rate), although trials reporting outcomes beyond 30 days of follow-up were few and provided inadequate power to measure important differences in these outcomes. Although the more relevant issue would be completeness of revascularization, this was reported in only eight trials: five trials reported equivalence or no difference in completeness,75,87,90,100two trials reported incomplete revascularization,74,78and one trial reported that complete revascularization was achieved more often with OPCAB than with CCAB (84% versus  76% achieving complete revascularization).98The recent publication of Khan et al.  87has drawn attention to concerns of reduced mid-term graft patency after OPCAB. However, the recent publication of 1-yr follow-up by Puskas et al.  101showed no significant difference in graft patency over the longer term in a larger patient population than that of Khan et al.  Similarly, two additional trials have shown no significant difference in patency at 3 months to 12 months.91,98Whether the discrepant results of Khan et al.  represent issues related to surgeon inexperience, type of conduits performed, or other factors remains unknown.127–130In our meta-analysis, statistical analysis of patency was attempted, but deemed not possible for the following reasons: 1) few articles reported graft patency, 2) large numbers of patients were lost to follow-up, 3) patency outcomes were reported “per graft” rather than “per patient,” and 4) length of follow-up was generally short. Additional trials will be required to address concerns about differences in the quality of revascularization after OPCAB surgery. More importantly, the more relevant endpoint of need for reintervention over the long term should be explored in future research. Clearly, the issue of patency is likely to be surgeon-dependent and has been inadequately explored in randomized studies to allow for conclusions at this time.

Although the baseline demographics suggest that trials, on average, included relatively young and healthy patients, some recent trials included higher risk patients with multiple vessel disease.72,74,75,78,79,83,87,116Nevertheless, some high-risk groups (age >70, renal dysfunction, pulmonary disease, aortic disease, ongoing ischemia) were under-represented. This is notable, as it is precisely these patients that are purported to benefit from OPCAB.8,74 

Although the aggregate estimates for absolute risk of stroke (1.0%) and myocardial infarction (2.8%) are within expected rates for CCAB,5,119the absolute risk of mortality for the CCAB group (1.0%) is lower than expected (2–3%).120This may reflect the highly selected population found in the majority of these trials, which was generally younger and had fewer overall grafts than the United States national average.2The rate of mediastinitis/wound infection was high, at 4.8% overall. However, trials that identified mediastinitis separately from wound infection reported rates of 1–2%. In other studies, superficial wound infection data were not separable from mediastinitis for analysis of the subcomponents.

The rigor of this meta-analysis, as evidenced by comprehensive searches or randomized trials of all relevant outcomes and comparisons in any language, and the adherence to “Quality of Reports of Meta-Analyses” recommendations10serve to increase confidence that this represents a complete summary of the best available evidence. Most conclusions were robust across sensitivity analyses, including estimates of NNT when real-world baseline estimates were imputed. Although the median Jadad quality score was 2 of 5, this is common for meta-analysis of randomized trials131,132and does not necessarily mean the trials were of low quality but rather that key methodological details simply were not reported.133Indeed, in this meta-analysis, tests for interaction found no significant association between effect size and study quality. Because blinding requires considerable effort in this type of trial the most common reason for Jadad quality scores of 3 or less was the lack of blinding. The absence of blinding may have led to biases in outcomes such as length of stay and ventilation time if off-pump patients were preferentially fast-tracked as a result of previous perceptions of the relative increased safety of doing so in off-pump patients, and this caveat should be incorporated into the interpretation of the results.

Although our analysis delineates the landscape of existing evidence, it also serves to highlight gaps that remain. Most notable is the lack of research defining long-term clinical, economic, and QOL outcomes associated with differing revascularization techniques. Accordingly, theoretical concerns about the quality and patency of grafts performed on the beating heart have not been adequately addressed by randomized trials to date, although survival and rates of reintervention for ischemia at 1–2 yr have been encouraging (table 3). Also notable is the lack of research in high-risk patients.

In setting the future research agenda for off-pump revascularization, it is prudent to note that studies exploring differences in mortality, acute myocardial infarction, and stroke may not be worth pursuing in patient populations similar to those represented in our analysis, as the 95% confidence intervals rule out clinically important differences. For example, the aggregate estimate for absolute reduction in mortality is 0.2%, with 95% confidence that the true value lies within the range of approximately −0.7% to + 0.7%. Measuring a difference of this magnitude is not practical because of large sample size requirements (table 7). Similarly, as a result of the small absolute differences between OPCAB and CCAB for stroke (0.6%) and acute myocardial infarction (0.8%), randomized trials of impractical sizes would be required to prove whether clinically small, but statistically significant, differences of this magnitude exist (table 7).

To maximally inform decision-making, clinicians need to understand how the expected net treatment benefit varies according to patient risk group. In meta-analyses of clinical trials, it is common to explore whether treatment benefits vary according to the underlying risk of the patients in different trials, with the hope of defining which patient populations benefit most and which benefit least.134However, as was found in this meta-analysis, it is common for data at the trial level to provide inadequate detail to allow for subanalysis by patient risk group. In the future, the results of this meta-analysis should be complemented by explorations of the net treatment benefit according to well-defined patient risk groups. Although some of this information might be retrievable through meta-analysis of individual patient data, the time and resource requirement to perform such an analysis may preclude this from occurring in the near future. Ideally, resources should be directed toward randomized trials in high-risk patients who are purported to benefit most from off-pump revascularization.

Given that these results represent the best available estimation of the benefits and risks of off-pump versus  conventional and interventional revascularization techniques, the decision between OPCAB or CCAB should be governed by careful consideration of the clinical importance of these potential benefits in light of the uncertainties (operator experience, contemporary practice patterns such as arrhythmia prophylaxis and blood conserving strategies, and clinical pathways for expedited discharge). The decision maker will need to incorporate the following important caveats into decision-making: if reducing atrial fibrillation, respiratory infections, inotropes, and transfusions is clinically relevant despite the availability of antiarrhythmics and blood-conserving techniques, then off-pump surgery may be preferred if surgeon skills and expertise are available to improve these clinical outcomes and possibly reduce resource utilization. If preventing atrial fibrillation (which is often transient, and of questionable clinical relevance), inotropes, respiratory infections, and transfusions is deemed less important in light of the current practice environment, physicians may choose to continue with CCAB, particularly if it is more familiar to their surgeons and especially in light of the paucity of long-term evaluation of patency and other clinical outcomes.

In conclusion, OPCAB did not significantly reduce mortality, stroke, myocardial infarction, and renal dysfunction compared with CCAB. On the other hand, this analysis suggests OPCAB may improve selected 30-day clinical outcomes (atrial fibrillation, inotrope requirements, respiratory infections, and blood transfusion) without measurable increased risk to the patient while reducing resource utilization (ventilation time, ICU LOS, and total hospital LOS) and potentially reducing in-hospital costs compared with CCAB. Clinical and economic outcomes beyond 1 yr are encouraging but remain to be confirmed in further trials. Similarly, neurocognitive outcomes are promising, but require further study.

Overall, this meta-analysis demonstrates that OPCAB improves only selected short-term and mid-term clinical outcomes other than mortality and may improve resource utilization compared with CCAB. Future clinical trials are unlikely to find a clinically significant difference in mortality, stroke, or acute myocardial infarction with off-pump versus  conventional surgical revascularization in low to medium risk patients similar to those included in this analysis.

The authors are grateful for the excellent work of Jeanette Mikulic, Administrative Assistant, Department of Anesthesia & Perioperative Medicine, University of Western Ontario, in preparing the manuscript and Larry Stitt, M.Sc., Assistant Director, Biostatistical Support Unit, University of Western Ontario, London, Ontario, Canada, for offering his statistical expertise. We would also like to acknowledge all study authors who responded to our request for additional information.