OUP user menu

Clinical outcomes in randomized trials of off- vs. on-pump coronary artery bypass surgery: systematic review with meta-analyses and trial sequential analyses

Christian H. Møller, Luit Penninga, Jørn Wetterslev, Daniel A. Steinbrüchel, Christian Gluud
DOI: http://dx.doi.org/10.1093/eurheartj/ehn335 2601-2616 First published online: 15 July 2008

Abstract

Aims To assess the clinical outcomes of off- vs. on-pump coronary artery bypass surgery in randomized trials.

Methods and results We searched electronic databases and bibliographies until June 2007. Trials were assessed for risk of bias. Outcome measures were all-cause mortality, myocardial infarction, stroke, atrial fibrillation, and renewed coronary revascularization at maximum follow-up. We applied trial sequential analysis to estimate the strength of evidence. We found 66 randomized trials. There was no statistically significant differences regarding mortality [relative risk (RR) 0.98; 95% confidence interval (CI) 0.66–1.44], myocardial infarction (RR 0.95; 95% CI 0.65–1.37), or renewed coronary revascularization (RR 1.34; 95% CI 0.83–2.18). We found a significant reduced risk of atrial fibrillation (RR 0.69; 95% CI 0.57–0.83) and stroke (RR 0.53; 95% CI 0.31–0.91) in off-pump patients. However, when continuity correction for zero-event trials was included, the reduction in stroke became insignificant (RR 0.62; 95% CI 0.32–1.19). Trial sequential analysis demonstrated overwhelming evidence supporting that off-pump bypass surgery reduces atrial fibrillation.

Conclusion Off-pump surgery reduces the risks of postoperative atrial fibrillation compared with on-pump surgery. For death, myocardial infarction, stroke, and renewed coronary revascularization, the evidence is still weak and more low-bias risk trials are needed.

Keywords
  • Surgery
  • Systematic review
  • Meta-analysis
  • Cardiovascular disease
  • Cardiopulmonary bypass

Introduction

Coronary artery bypass grafting (CABG) is a well-established treatment for ischaemic heart disease and has traditionally been performed using cardiopulmonary bypass (on-pump CABG). Adverse clinical events associated with coronary artery bypass surgery have been attributed to cardiopulmonary bypass, cardiac arrest, and aortic cannulation. To avoid these adverse events, CABG performed without cardiopulmonary bypass (off-pump CABG) has gained increased interest. Randomized trials comparing off- vs. on-pump CABG have been conducted, but results have varied and most trials have been underpowered. To compensate for this, meta-analyses are needed. Previous meta-analyses found either no significant differences or a reduction in atrial fibrillation and stroke associated with off-pump CABG.15

The objective of this systematic review was to assess the benefit and harm of off- vs. on-pump CABG focusing on all-cause mortality, myocardial infarction, stroke, coronary reintervention, and atrial fibrillation. As trials with low methodological quality have high risk of ‘bias’ (systematic error) and a tendency to overestimate the intervention effect, we considered the bias risk of the included trials, by evaluating components of randomization, blinding, and intention-to-treat analysis.69 Meta-analyses may also overestimate treatment effect because of ‘play of chance’ (random error).10 Especially, randomized trials and meta-analysis with few participants or outcomes are at risk of producing random error. To minimize random error, we calculated an information size (i.e. a required meta-analysis sample size) as large as that of an adequately powered randomized trial and applied trial sequential analysis to the meta-analysis.10

Methods

Search strategy

We searched The Cochrane Central Register of Controlled Trials (CENTRAL) in The Cochrane Library (Issue 2, 2007), MEDLINE (1966 to June 2007), EMBASE (1980 to June 2007), CINAHL (1982 to June 2007), and the Science Citation Index Expanded (1945 to June 2007). We scanned bibliographies of relevant articles for additional trials. We used no restrictions to language or publication status.

Trial selection and characteristics

The present review follows The Cochrane Collaboration methodology.6 We included all randomized trials comparing off- vs. on-pump CABG for isolated coronary artery disease using cardioplegia in the on-pump group. We excluded trials that compared off- vs. on-pump beating-heart bypass surgery and trials where cardiac arrest was conducted with ventricular fibrillation.

We included trials independently of type of stabilization system in off-pump CABG and type of cardioplegic solution and cardiopulmonary bypass in on-pump CABG. Trials with collateral interventions were included if the interventions were used equally in the intervention groups.

Outcomes were all-cause mortality, myocardial infarction, stroke, atrial fibrillation, and renewed coronary revascularization (CABG or percutaneous coronary intervention). Outcomes from trials with several publications were included from the publication with maximum follow-up and with largest number of patients reported. We used the definitions of the outcome measures used in the individual trial.

Data extraction and quality assessment

Two of three authors (C.H.M. and L.P. or D.A.S.) independently assessed trial eligibility. Excluded trials were listed with the reason for exclusion. Disagreement was solved by discussion or in consultation with a third author (C.G.). We contacted authors of the trials if information about methodology or data was unclear or missing.

From each trial we recorded first author or trial name; country of origin; trial design; number of participants; inclusion and exclusion criteria; characteristics of patients: age, diabetes mellitus, left ventricular ejection fraction, patients with three-vessel disease, and number of distal anastomoses; type of cardioplegia; patient temperature during cardiopulmonary bypass; dose of heparin in off-pump CABG; reversal of heparin effects with protamine in off-pump CABG; duration of follow-up; intention-to-treat analysis; surgical conversion rate; and primary and secondary outcomes for the trial.

As a result of the risk of overestimating intervention effects, analyses were stratified according to the risk of bias.69 Trials with adequate generation of allocation sequence, adequate allocation concealment, blinded outcome assessment, and intention-to-treat analysis were considered low-bias risk trials. When one or more of the quality components were unclear or inadequate, trials were classified accordingly. Generation of the allocation sequence was considered adequate when generated by a computer, random-number table, shuffling of cards, or similar. Allocation concealment was considered adequate when allocation of patients involved a central independent unit; on-site locked computer, sealed envelopes, or similar. Blinded outcome assessment was considered adequate when the authors stated that outcomes were assessed by independent assessors, who were unaware of which intervention the patients underwent or similar. Analysis according to intention-to-treat was considered adequate, when all randomized patients were included in the analysis (i.e. when no patient exclusion was reported and all patients seemed included in the data analysis, this was considered adequate). Post-randomization exclusion of patients was registered. When possible we converted per-protocol to intention-to-treat analysis. Bias risk was assessed without blinding by two authors (CHM and LP or DAS). Consensus was reached through discussion or arbitration by a third author (C.G.). High inter-rater agreement between blinded and unblinded assessments as well as between two independent assessors has previously been found.7

Quantitative data synthesis

We used The Cochrane Collaboration software (RevMan 4.2.10).11 Data were analysed with a random-effects model and presented as relative risk (RR) and with 95% confidence interval (CI).

Data from each trial were entered in the meta-analyses as intention-to-treat when possible. For trials with factorial design, we based our results on at-margins analysis, comparing all groups that underwent off-pump CABG with all groups that underwent on-pump CABG.

Heterogeneity was assessed with I2 that describes the percentage of total variation across trials because of between-trial variance rather than sampling error.6,12 I2 lies between 0% (no heterogeneity) and 100% (maximal heterogeneity).6 To assess publication bias and other types of bias, funnel plots were created in which a trial's RR was logarithmic transformed and plotted against the standard error.6

The RevMan 4.2.10 software is unable to handle trials with zero events in both intervention groups when meta-analyses are performed as RR or odds ratios. It seems unjustified to exclude zero-event trials13 and potentially create the risk of inflating the magnitude of the pooled treatment effects.14 We therefore performed a random-effects meta-analysis with empirical continuity correction of 0.01 in zero-event trials.

We performed sensitivity analyses on trials that did not report outcomes on all randomized patients assuming a best–best case scenario; a worst–best case scenario; a best–worst case scenario, and a worst–worst case scenario, as well as on length of follow-up (≤30 days, >30 days).

Subgroup analyses were planned regarding the type of stabilization system and dose of heparin in the off-pump group and temperature during cardiopulmonary bypass and type of cardioplegic solution in the on-pump group. Intervention effects in the subgroups were compared by test of interaction.15

Trial sequential analysis

Cumulative meta-analyses are at risk of producing random errors because of repetitive testing on accumulating data.10 To minimize random errors we calculated the required information size (i.e. the number of participants needed in a meta-analysis to detect or reject a certain intervention effect).10 Information size calculation also accounted for the heterogeneity present in the meta-analysis. In our meta-analysis, information size was based on the assumption of a plausible RR reduction of 20% or on the RR reduction observed in trials with low bias risk.10 The underlying assumption of trial sequential analysis is that significance testing may be performed each time a new trial is added to the meta-analysis. We added the trials according to the year of publication and if more than one trial was published in a year, trials were added alphabetically according to the last name of the first author. On the basis of the required information size and risk for type I and type II errors trial sequential monitoring boundaries were constructed.10 These boundaries will determine the statistical inference one may draw regarding the cumulative meta-analysis that has not reached the required information size; if a trial sequential monitoring boundary is crossed before the required information size is reached in a cumulative meta-analysis, firm evidence may have been established and further trials superfluous. On the other hand, if the boundaries are not surpassed, it is most probably necessary to continue doing trials in order to detect or reject a certain intervention effect. We used as default a type I error of 5%, type II error of 20%, and adjusted information size for heterogeneity unless otherwise stated.10

Results

Database searches yielded 1650 references. Exclusion of duplicates and irrelevant references left 138 references describing 86 trials. After review additional 22 references (20 trials) were excluded. We included 116 references describing 66 randomized trials fulfilling our inclusion criteria (Figure 1). There was excellent inter-rater agreement between the review authors regarding data extraction and bias risk assessment. Authors of 62 of the trials were contacted for additional information. Sixty percent responded. Trial design was parallel groups in 65 trials (62 with two groups, one with three groups,16 and two with four groups17,18) and 2 × 2 factorial in one trial.19

Figure 1

Flow diagram of identification of randomized trials for inclusion.

A total of 5537 participants were randomly assigned in the 66 trials. Demographic data of each trial are given in Tables 13. The number of participants in each trial ranged from 20 to 400. The mean age was 63 years. The mean proportion of women was 22% in the 60 trials reporting sex, and the mean proportion of patients with diabetes was 26% in the 37 trials reporting diabetic status. Patients with ejection fraction <0.30 were not represented in 34 trials. Sixty-eight patients with ejection fraction <0.30 were included in six trials. The number of patients with three-vessel disease was reported in 19 trials (1232 patients). Two trials included only patients with one-vessel disease16,20 and one trial mainly patients with one-vessel disease.21 Surgeons' experience in the two procedures was reported in <25% of the trials.

View this table:
Table 1

Characteristics of trials with adequate randomization and blinded outcome assessment

Source/trial nameNo. of patientsMean age, yearsFemale, %EFDiabetes, %Follow-up
BHACAS I35–45200622080% > 0.502025 months
BHACAS II35,42201631677% > 0.503113.7 months
OCTOPUS23,46–58281613178% > 0.50185 years
SMART59–63200622349% > 0.453112 months
Al-Ruzzeh et al.64168631768% > 0.50216 months
  • EF, ejection fraction (reported percent above 0.50).

View this table:
Table 2

Characteristics of trials with adequate randomization, but lack blinding

Source/trial nameNo. of patientsMean age, yearsFemale, %EFDiabetes, %Follow-up
Gu et al.206261340.538In-hospital
Diegeler et al.654065NR0.59NRIn-hospital
Gulielmos et al.17,664062230.67NR3 months
Matata et al.67206015NR0In-hospital
Covino et al.6837NR110.5316In-hospital
Penttilä et al.692259NR95% > 0.50NRIn-hospital
Caputo et al.70406310NR0In-hospital
Zamvar et al.71606313NRNR10 weeks
Carrier et al.72657023NRNRIn-hospital/30-day mortality
Raja et al.733006424NRNRIn-hospital
Alwan et al.747064300.59NRIn-hospital/30-day mortality
Gerola et al.75,761605934NR21In-hospital/30-day mortality
Khan et al.77103631368% > 0.50273 months
Legare et al.78–80300632086% > 0.50333.8 years
Lingaas et al.81–8612065220.721712 months
Motallebzadeh et al.873564946% > 0.5040In-hospital
PRAGUE-488,8940063190.582812 months
Selvanayagam et al.90–92606113NR25In-hospital
Ascione et al.93206215NR0In-hospital
Kobayashi et al.9414760150.5660> 30 days
Ascione et al.95406313NRNRIn-hospital
Michaux et al.965063160.583230 days
Motallebzadeh et al.97,98212651148% > 0.50246 months
Niranjan et al.1980671884% > 0.5014In-hospital
Paparella et al.9931NRNRNRNRIn-hospital
Tatoulis et al.30,100,101100652289% > 0.503230 days
Jares et al.102206430NRNR24 h
Kunes et al.1033468240.65NR7 days
Ozkara et al.1044459250.5659In-hospital
  • EF, ejection fraction (reported as either percent above 0.50 or mean of all participants in the trial); NR, not reported.

View this table:
Table 3

Characteristics of trials with unclear randomization and/or lack blinding

Source/trial nameNo. of patientsMean age, yearsFemale, %EFDiabetes, %Follow-up
Vural et al.21504812100% > 0.50NR2 months
Czerny et al.1053064230.67NRIn-hospital
Kochamba et al.1065859220.6441In-hospital
Wandschneider et al.1071196621NRNRIn-hospital
Wildhirt et al.108–111336519NRNR3 days
Baker et al.33266419NR196 months
Czerny et al.1128064160.64NR13 months
Guler et al.165855NRNRNR2 months
Tang et al.113456320NR0In-hospital
Al-Ruzzeh et al.11420641530% > 0.50NR4 days
Kherani et al.11546NRNRNRNR30 days
Lee et al.1166066180.55NR12 months
Muneretto et al.1171766739NR4112 months
Parolari et al.1182561240.61NRIn-hospital
Sahlman et al.1195063180.5612In-hospital
Vedin et al.120–12574652081% > 0.50196 months
Velissaris et al.126,127546220100% > 0.500In-hospital
Dorman et al.1285263440.6138In-hospital
El-din1293056100.48NRIn-hospital
Gasz et al.130206325NRNRIn-hospital
Synnergren et al.1315262350.62NRIn-hospital
Anderson et al.132506722NRNRIn-hospital
Blacher et al.22286236NR11In-hospital
Celik et al.1336067320.62010 days
Cavalca et al.1345066240.56NRIn-hospital
Gönenc et al.135426521NRNR24 h
Malik et al.1365058160.5326In-hospital
Mariscalco et al.1377065190.5437In-hospital
Nesher et al.1381256825NR21In-hospital
Quaniers et al.18806310NRNRIn-hospital
Rachwalik et al.1394258290.63NRIn-hospital
Schmid et al.1403068300.670In-hospital
  • EF, ejection fraction (reported as either percent above 0.50 or mean of all participants in the trial); NR, not reported.

In off-pump CABG the Octopus system (Medtronic Inc., Minneapolis, MN) was used in 24 trials, CardioThoracic system (Cupertino, CA) in 10 trials, Guidant system (Guidant, Santa Clara, CA) in four trials, both Octopus and CardioThoracic systems in two trials, and the stabilization system was not specified in 26 trials. Three trials performed off-pump surgery as minimal invasive surgery through an anterolateral incision16,17,20 and in one of these trials on-pump surgery was also performed through an anterolateral incision.17 None of trials stated that off-pump surgery was exclusively performed as no-touch aorta. Also, the use of intracoronary shunt was insufficiently reported. The amount of heparin used in off-pump patients and target ‘activated clotting time’ varied between trials. In four trials heparin was not reversed with protamine (see Supplementary material online, Appendix A). Epidural analgesia was used for patients after off-pump surgery in two trials.22,23

Cardiopulmonary bypass was conducted under normothermia in 13 trials, mild hypothermia (30–36°C) in 43 trials, moderate hypothermia (down to 28–29°C) in four trials, and six trials did not report on this issue (see Supplementary material online, Appendix A). Blood cardioplegic solution was used in 45 trials (62% as cold blood cardioplegia), cold crystalloid cardioplegia in 12 trials, and type of cardioplegic solution was not specified in nine trials.

The conversion from off-pump to on-pump CABG was 5%, range 0–27%. The conversion from on- to off-pump CABG was 0.9%, range 0–7% (see Supplementary material online, Appendix B). Significantly fewer distal anastomoses were performed after off-pump surgery (weighted mean difference −0.29, −0.46 to −0.13, 41 trials).

One trial reported 24 h follow-up, 38 trials in-hospital follow-up (three with 30-day mortality), seven trials between 4 and 30 days follow-up, and 20 trials more than 30-day follow-up (up to 5 years).

Methodological quality of included trials

Five of the 66 trials (7.6%, 1050 patients) had low-bias risk, i.e. had adequately generated allocation sequence, adequate allocation concealment, blinded outcome assessment, and intention-to-treat analysis. None of these five were zero-event trials. Twenty-five trials lacked blinded outcome assessment. The remaining trials lacked blinded outcome assessment and were unclear in one or both the randomization components (see Supplementary material online, Appendix B).

In the SMART trial, three patients (1.5%) were excluded because of concomitant mitral valve surgery. This trial was considered fulfilling intention-to-treat.24 In 21 trials intention-to-treat analyses were either not followed or unclear (see Supplementary material online, Appendix B).

Mortality

Mortality was reported in 57 trials (5202 patients). The meta-analysis showed no significant difference in mortality (RR 0.98; 95% CI 0.66–1.44) (Figure 2). No discrepancy was found between low- and high-bias risk trials. There was no heterogeneity between trials (I2 = 0%). However, 36 trials (2031 patients) were zero-event trials (without death in any of the groups). These trials are not included in a traditional meta-analysis. When all zero-event trials were given a continuity correction of 0.01 the point estimate changed slightly (RR 1.07; 95% CI 0.70–1.62). Using trial sequential analysis, the required information size is 242 657 patients to demonstrate or reject a 5% RR reduction of mortality based on the meta-analysed estimate from low-bias risk trials and a pooled event rate of 4.79% in the on-pump group. This indicates that only 2% of the required number of patients has been included until now.

Figure 2

Intervention effect of off- vs. on-pump coronary artery bypass grafting on mortality. Error bars indicate 95% confidence intervals.

Myocardial infarction

Occurrence of myocardial infarction was reported in 44 trials including 4303 patients. No significant difference occurred between off- and on-pump surgery (RR 0.95; 95% CI 0.65–1.37) (Figure 3). This result was independent of risk of bias. No heterogeneity was found (I2 = 0%). Nineteen trials (909 patients) were zero-event trials. When zero-event trials were continuity corrected, there was no noticeable change (RR 0.94; 95% CI 0.63–1.39). Using trial sequential analysis, the required information size is 4671 patients to demonstrate or reject a 33.3% RR reduction of myocardial infarction based on the meta-analysed estimate from low-bias risk trials and a pooled event rate of 4.88% in the on-pump group. With 4303 randomized patients the cumulative z-curve has not crossed any of the trial sequential monitoring boundaries, despite the accrual of 92% of the required information size (see Supplementary material online, Appendix C).

Figure 3

Intervention effect of off- vs. on-pump coronary artery bypass grafting on myocardial infarction. Error bars indicate 95% confidence intervals.

Stroke

Occurrence of stroke was reported in 47 trials including 4535 patients of which 21 were zero-event trials. The meta-analysis of the effect of off- vs. on-pump CABG on stroke not including zero-event trials showed a statistically significant reduction of stroke (RR 0.53; 95% CI 0.31–0.91, P = 0.02) (Figure 4). However, with an empirical continuity correction of 0.01 to the zero-event trials, the effect of off-pump CABG was not statistically significant (RR 0.62; 95% CI 0.32–1.19, P = 0.15). There was no between-trial heterogeneity (I2 = 0%). In low-bias risk trials (1050 patients), the RR reduction in off-pump CABG did not reach statistical significance (RR 0.65; 95% CI 0.27–1.54). Using trial sequential analysis, the required information size is 7155 patients to demonstrate or reject an intervention effect of 34.7% RR reduction of stroke based on the meta-analysed estimate from low-bias risk trials and a pooled event rate of 2.94% in the on-pump group. Until now, only 4535 patients are included and the cumulative z-curve has not crossed any of the trial sequential monitoring boundaries, despite the accrual of 63% of the required information size (see Supplementary material online, Appendix D).

Figure 4

Intervention effect of off- vs. on-pump coronary artery bypass grafting on stroke. Error bars indicate 95% confidence intervals.

Atrial fibrillation

Occurrence of atrial fibrillation was reported in 30 trials including 3634 patients (two zero-event trials). Off-pump CABG was according to the traditional meta-analysis associated with a significant RR reduction (RR 0.69; 95% CI 0.57–0.83, continuity corrected for zero-event trial RR 0.69; 95% CI 0.58–0.83) (Figure 5). In low-bias risk trials (1050 patients), the effect was insignificant (0.63, 0.35–1.13) and with high degree of heterogeneity (I2 = 83.3%). Using trial sequential analysis, the required information size is 1403 patients to demonstrate or reject an intervention effect of 36.9% RR reduction of atrial fibrillation based on the meta-analysed estimate from low-bias risk trials and an a pooled event rate of 32.5% in the on-pump group (see Supplementary material online, Appendix E). To demonstrate or reject a priori anticipated intervention effect of 20% RR reduction the required information size is 5069 patients (see Supplementary material online, Appendix F). In the trial sequential analysis 3634 patients are included and the cumulative z-curve crosses both the trial sequential monitoring boundary according to a 20% RR reduction and the boundary according to an RR reduction of 36.9 suggested by low-bias risk trials, even though both calculations were based on an even more conservative significance level (type I error of 1% and type II error of 10%).

Figure 5

Intervention effect of off- vs. on-pump coronary artery bypass grafting on atrial fibrillation. Error bars indicate 95% confidence intervals.

Renewed cardiac revascularization

Fifteen trials (2215 patients) reported cardiac reintervention and in eight of them follow-up were for 1 year or more. No significant difference between off- and on-pump CABG was found (without continuity correction RR 1.34; 95% CI 0.83–2.18, with continuity correction RR 1.28; 95% CI 0.79–2.09) (Figure 6). Analysing only low-bias risk trials resulted in wider CIs (RR 1.40; 95% CI 0.68–2.89). Using trial sequential analysis, the required information size is 5846 patients to demonstrate or reject an intervention effect of 40% RR reduction of coronary reintervention based on the meta-analysed estimate from low-bias risk trials and a pooled event rate of 3.86% in the on-pump group. Until now, only 38% of the required information size has been included. The cumulative z-curve has not crossed any of the trial sequential monitoring boundaries.

Figure 6

Intervention effect of off- vs. on-pump coronary artery bypass grafting on renewed cardiac revascularization. Error bars indicate 95% confidence intervals.

Sensitivity analyses

We did sensitivity analyses on the few patients excluded after randomization and without outcomes reported. Regarding stroke, the effect of off-pump CABG changed from a statistically significant effect in the best-best case scenario analyses into an insignificant effect in the worst–best case scenario analysis. All other sensitivity analyses did not affect our results noticeably. Sensitivity analysis on length of follow-up had no noticeable influence on the results (see Supplementary material online, Appendix G).

Subgroup analyses

In off-pump CABG, the type of stabilization system was stratified into Octopus, CardioThoracic, Guidant, or not specified or mixed. Heparin dose was stratified into <300 IU/kg or ≥300 IU/kg. None of the tests of interaction revealed significant effect of stabilization system or heparin dose on intervention effects on any of our outcome measures.

In on-pump CABG, temperature during cardiopulmonary bypass was stratified into normothermia (>36°C) and hypothermia (≤36°C). Cardioplegic solution was stratified by three groups; (warm, isotherm, tepid blood), (cold blood), or (cold crystalloid). None of the tests of interaction revealed significant effects of temperature or cardioplegic solution on intervention effects on any of our outcomes.

Funnel plots

Funnel plots for each of our outcomes did not show any noticeable asymmetry.

Discussion

Principal findings

Our systematic review contains several findings. Very few randomized trials comparing off- vs. on-pump CABG are conducted with high methodological quality, which gives ample room for systematic errors (‘bias’).69 Second, even though more than 5500 patients had been included in the trials the number of outcomes was sparse. This increases the risks of random errors (‘play of chance’), both for finding differences when no differences exist and for observing no differences even if important differences exist.10 We found no evidence that off-pump CABG reduces mortality. The mortality proportion for off- and on-pump was 1.9 and 2.2%, respectively. According to trial sequential analysis more than 250 000 patients would be needed to demonstrate or reject such an intervention effect. Evidence that off-pump CABG reduces myocardial infarction was not found, but as for mortality the CIs are wide. However, trial sequential analysis showed that <370 patients are needed to reach the required information size to demonstrate or reject an RR reduction of 33% regarding myocardial infarction. On the basis of the traditional meta-analysis of all randomized trials, off-pump CABG was associated with a reduced risk of stroke, but the significant result disappeared when empirical continuity correction for zero-event trials was conducted. The low number of strokes in both groups impairs the reliability of this outcome.25 On the basis of the trial sequential analysis, 7155 randomized patients would be needed to demonstrate or reject an RR reduction of 35% regarding stroke. Trial sequential analysis demonstrates overwhelming evidence that off-pump CABG reduces the occurrence of atrial fibrillation with a 20% RR reduction. The cumulative z-curve crosses the trial sequential monitoring boundary after the 14th trial. The boundary was constructed for an a priori heterogeneity adjusted information size of 5069 with a type I error of 1% and a type II error of 10%.10 Our use of trial sequential monitoring boundaries applied to cumulative meta-analysis, adapted from interim monitoring boundaries of a single trial, showed that the current evidence for a clinically relevant effect of off-pump CABG on atrial fibrillation is conclusive. The quality of the bypass anastomoses and completeness of revascularization have been questioned after off-pump CABG. We could not show a significantly increased risk of renewed coronary revascularization after off-pump CABG, but CIs are wide and the number of randomized patients has still not reached the required information size. We excluded five trials using on-pump beating-heart or on-pump ventricular fibrillation instead of cardiac arrest. However, this did not affect our results (unpublished data). As a sensitivity analysis, we analysed each outcome according to short- and long-term follow-up. We found no noticeable time-related variation between the two interventions for any of our outcomes.

Strengths

Our review offers a number of strengths. Before initiating the review we wrote a protocol according to the guidelines of The Cochrane Collaboration.6 Our systematic review represents a comprehensive review of the topic, including 66 randomized trials with 5537 participants, as well as a thorough assessment of the trial methodology following the recommendations of The Cochrane Collaboration6 and findings of methodological studies.79 None of the previous meta-analyses have performed continuity correction of the many zero-event trials and have thereby excluded several randomized patients resulting in an artificially high proportion of outcomes. Since the previous meta-analyses, results of medium- and long-term follow-up from several larger trials have been published. With 20 trials reporting results from more than 30 days of follow-up in our meta-analysis, we almost doubled this number compared with the previous meta-analyses.15 A part from the analysis of atrial fibrillation, heterogeneity was absent in our meta-analyses and subgroup analyses showed no significant effect on any of our outcome measures. Furthermore, we have conducted trial sequential analysis, which we consider a useful instrument to evaluate when firm evidence may be reached.10

Limitations

Our systematic review also has limitations. Our findings and interpretations are limited by the quality and quantity of available evidence. The included patient population in randomized trials may not be representative for a general patient population. Trials have mainly included low-risk patients, that is, only ∼50% had three-vessel disease, the majority had preserved left ventricular ejection fraction, and patients with renal and pulmonary dysfunction were in general excluded. Accordingly, the external validity or generalizability may be compromised. The methodological quality of some of the trials was assessed based only on published reports, which may not reflect the actual design and bias risk of the trial. Authors were contacted when information was unclear or lacking. More than half of the authors responded to our requests for further information. Only five trials including 1050 participants fell in the group of low-bias risk trials. Even among the low-bias risk trials, we are not able to exclude bias as these trials could also be influenced by the lack of blinding, which opens for collateral intervention bias, reporting bias, and other biases.69 This highlights the uncertainty that our and others meta-analyses are encumbered with.

Surgeons' experience in the two procedures was described in few trials. As for every comparison between a new and an old procedure, the intervention effect of the new procedure may be underestimated because of a learning curve. We cannot exclude that some trials are influenced by surgeons’ learning curves,26 which may explain the observed variation in conversion rate from off- to on-pump. However, the between-trial variation (heterogeneity) was for most of the outcomes zero, indicating that the results are consistent irrespective of the speculations above. Off-pump CABG is believed to be more beneficial in high-risk patient, that is, high age or patients with renal and pulmonary dysfunction.27 To enlighten these issues, our meta-analyses, with exception of atrial fibrillation, cannot conclude on the obtainable evidence.

Relation to previous studies

The meta-analyses by Parolari et al.4 and van der Heijden et al.5 used a composite outcome of stroke, myocardial infarction, or death. Both found a point estimate of the odds ratio favouring off-pump surgery, but their results were not statistically significant. Trials with high-risk patients or average number of grafts per patients <2 were excluded in the study by Parolari et al.4 and some patients were included more than once, because of double publication of some trials. Although an effect on the composite outcome measure would be of interest, very few trials reported their outcomes in such a detail that stroke, myocardial infarction, and death can be pooled as a composite outcome. Investigators examining the effect of cardiac surgery should adopt a standardized way of reporting and analysing results.

Two meta-analyses of randomized trials published in 2005 found no significant difference with respect to mortality, myocardial infarction and stroke.1,3 Both the meta-analyses found significantly reduced risk of atrial fibrillation after off-pump surgery. Wijeysundera et al.3 conducted a meta-analysis of observational studies (293 617 patients) and found that off-pump CABG significantly favoured 30-day mortality, stroke, myocardial infarction, and atrial fibrillation. As observational studies have a lower ranking in the evidence hierarchy and are under influence of confounding by indication, these results should be interpreted with caution.28

In 2006 Sedrakyan et al. published a meta-analysis concluding that off-pump CABG is significantly associated with reduced risk of stroke and atrial fibrillation.2 From our systematic review including more than 1500 extra patients and correcting for zero-event trials, we do not find a significant reduction in the risk of stroke. As mentioned, the total number of 54 strokes is too low to reach firm conclusions, and from the RR reduction found in the low-bias risk trials an information size of 7155 patients is needed and at the moment only 63% of this number has been randomized.

Implications

To increase the strength of evidence on which method to prefer in CABG, we recommend inclusion of consecutive high-risk patients into large randomized trials with longer follow-up and blinded outcome assessment. Such trials ought to be registered.29

Even after the publication of 66 randomized trials with more than 5500 patients included, we do not know for sure if off-pump CABG is better or worse compared with on-pump CABG. Only 23% of the trials reported sample size estimation, and the vast majority of these trials had either overestimated the occurrence of outcomes in the on-pump group or overestimated the relative benefit of off-pump intervention. During this systematic review we have identified five ongoing trials expected to include another 3500 patients.3034 Our systematic review has to be updated when results from these trials are available.

Conclusion

Considering the methodological quality of the included trials and all the evidence provided, we did not find convincing evidence that off-pump coronary artery bypass reduces any of the major clinical outcomes. Firm evidence is demonstrated that off-pump CABG reduces the rate of post-operative atrial fibrillation. However, we do not find a beneficial effect on this singular outcome reliable for recommending off-pump instead of on-pump CABG, as the evidence on the primary outcomes is inconclusive. Unless ongoing trial results combined with already obtained evidence demonstrate superiority of one of the CABG methods, both interventions may be offered to patients.

Funding

Funding was received from The Danish Heart Foundation: (08-4-R64-A2029-B948-22480), The Danish Medical Research Council, The Copenhagen Hospital Corporatiońs Medical Research Council, The Rigshospitalet Research Council, and The Copenhagen Trial Unit. The funding sources had no role in the conduct of the study, collection of data, management, analysis, interpretation of the data, or preparation of the manuscript.

Acknowledgements

We thank the patients who participated in the trials and the investigators who conducted the trials. We acknowledge the authors who kindly responded to our requests for further information on the trials they were involved in and Sarah Louise Klingenberg from the Cochrane Hepato-Biliary Group for assistance with the database literature searches.

Authors Contribution: Møller had full access to all data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Møller, Penninga, Steinbrüchel, and Gluud.

Acquisition of data: Møller, Penninga, Steinbrüchel, and Gluud.

Analysis and interpretation of data: Møller, Penninga, Wetterslev, Steinbrüchel, and Gluud.

Drafting of the manuscript: Møller, Penninga, Wetterslev, Steinbrüchel, and Gluud.

Critical revision of the manuscript for important intellectual content: Møller, Penninga, Wetterslev, Steinbrüchel, and Gluud.

Statistical analysis: Møller, Wetterslev, and Gluud.

Conflict of interest: none declared.

References

View Abstract