Meta-analysis of secure randomised controlled trials of β-blockade to prevent perioperative death in non-cardiac surgery
- International Centre for Circulatory Health, National Heart and Lung Institute, Imperial College London, London, UK
- Correspondence to Dr Sonia Bouri, International Centre for Circulatory Health, National Heart and Lung Institute, Imperial College London, St Mary's Hospital, 59-61 North Wharf Road, London W2 1LA, UK;
- Received 11 May 2013
- Revised 11 July 2013
- Accepted 12 July 2013
- Published Online First 31 July 2013
Background Current European and American guidelines recommend the perioperative initiation of a course of β-blockers in those at risk of cardiac events undergoing high- or intermediate-risk surgery or vascular surgery. The Dutch Echocardiographic Cardiac Risk Evaluation Applying Stress Echocardiography (DECREASE) family of trials, the bedrock of evidence for this, are no longer secure. We therefore conducted a meta-analysis of randomised controlled trials of β-blockade on perioperative mortality, non-fatal myocardial infarction, stroke and hypotension in non-cardiac surgery using the secure data.
Methods The randomised controlled trials of initiation of β-blockers before non-cardiac surgery were examined. Primary outcome was all-cause mortality at 30 days or at discharge. The DECREASE trials were separately analysed.
Results Nine secure trials totalling 10 529 patients, 291 of whom died, met the criteria. Initiation of a course of β-blockers before surgery caused a 27% risk increase in 30-day all-cause mortality (p=0.04). The DECREASE family of studies substantially contradict the meta-analysis of the secure trials on the effect of mortality (p=0.05 for divergence). In the secure trials, β-blockade reduced non-fatal myocardial infarction (RR 0.73, p=0.001) but increased stroke (RR 1.73, p=0.05) and hypotension (RR 1.51, p<0.00001). These results were dominated by one large trial.
Conclusions Guideline bodies should retract their recommendations based on fictitious data without further delay. This should not be blocked by dispute over allocation of blame. The well-conducted trials indicate a statistically significant 27% increase in mortality from the initiation of perioperative β-blockade that guidelines currently recommend. Any remaining enthusiasts might best channel their energy into a further randomised trial which should be designed carefully and conducted honestly.
Physicians across Europe are still advocated by guidelines to initiate a course of perioperative β-blockade in three classes of patients:
‘… [those] who have known IHD or myocardial ischaemia according to pre-operative stress testing’,
‘… [those] scheduled for high-risk surgery’ and
‘… [those] scheduled for intermediate-risk surgery’. The joint guidelines produced by the American College of Cardiology Foundation and the American Heart Association (ACCF/AHA) also endorse perioperative β-blockade in patients undergoing vascular or intermediate-risk surgery with coronary artery disease (CAD), or with more than one risk factor for CAD, or with pre-existing β-blockade (table 1).
The principal evidence for mortality benefit has been the Dutch Echocardiographic Cardiac Risk Evaluation Applying Stress Echocardiography (DECREASE)2 family of studies which were discredited almost 2 years ago3 and subsequently underwent lengthy internal investigation, the results of which have been public for some time.4 Nevertheless, neither the European Society of Cardiology (ESC) nor the AHA guidelines have been retracted.
All studies investigated in the DECREASE family for which data had not been lost were found to be insecure because of serious flaws (table 2). In one case it was clear that the entire study dataset had been fabricated. DECREASE I,5 published in 1999, escaped investigation as the terms of the investigation only reached back 10 years.
Individual clinicians may feel powerless to act independently in contravention of guidelines. The ESC has recently reiterated that its guideline was ‘based on the contributions of many European experts and on available evidence-based medicine including many studies from different nations. They are, therefore, the result of a group discussion and not of an individual position’.6
We therefore conducted a meta-analysis of the remaining secure intention-to-treat randomised controlled trial (RCT) data on the initiation of a course of β-blockade for the prevention of all-cause mortality and other secondary endpoints in the perioperative period for patients undergoing non-cardiac surgery.
We included published RCTs that compared the initiation of a course of β-blocker therapy in the preoperative period with placebo in adults undergoing non-cardiac surgery. There were no language restrictions. We searched Medline (1966 to 1 April 2013), the Cochrane Central Register of Randomised Controlled Trials, the WHO International Clinical Trials Registry Platform Search Portal (http://apps.who.int/trialsearch/), Excerpta Medica Database (EMBASE) and the Cumulative Index to Nursing and Allied Health Literature (CINAHL) using the search terms available in the online supplement on 23 March 2013 (see online supplementary appendix 1). We also hand-searched previous reviews and meta-analyses for other studies. We excluded non-randomised studies, studies comparing β-blockers with another treatment, studies using a one-off dose preoperatively rather than a course of β-blockers extending into the postoperative period and studies which did not report intention-to-treat data.
All-cause mortality on intention-to-treat-basis
The primary endpoint was all-cause mortality from the date of randomisation without excluding the in-hospital postoperative window. The time point was 30 days or, if this was not available, until hospital discharge. The secondary endpoints were non-fatal myocardial infarction (MI), stroke and hypotension.
Data extraction was performed in duplicate by MJS and SB with any disagreements resolved by DPF.
We performed the meta-analysis excluding studies from the DECREASE family because every study in it that had enough documentation to be investigated was found to be insecure (table 2).
We used the I2 statistic to measure the level of heterogeneity.7 A random effects model was used to synthesise the data with Mantel–Haenszel risk ratios calculated. Review Manager V.5.2.1 was used to perform the meta-analysis.8
Assessment of quality of trials
All studies were assessed for quality using the Cochrane ‘risk of bias’ tool9 which considers the risk of selection, performance, detection, attrition and reporting bias. Publication bias was assessed using a funnel plot.
Identification of trials
We identified 300 publications; 265 were initially found on PubMed (see online supplementary appendix 2), 3 from the Cochrane Central Register of Randomised Controlled Trials, 19 from the CINAHL, 1 from EMBASE and 12 via hand searching of references. A total of 282 were excluded after reading the abstract (of which 39 received premedication only) and a further seven were excluded after reading the full text for the following reasons: two were duplicates of other studies,10 ,11 one could not be included because 10 patients were excluded after randomisation including one who had pulmonary oedema in the metoprolol arm,12 three did not meet the time point criteria of 30 days or until discharge,13–15 and one initiated the β-blockade postoperatively16 (see online supplementary appendix 2). A total of 11 RCTs met the eligibility criteria (figure 1), of which two were from the DECREASE family (DECREASE I and DECREASE IV5 ,17).
The β-blocker administered varied between studies. Three trials used bisoprolol,5 ,14 ,17 five metoprolol,18–22 two atenolol23 ,24 and one propranolol.25 β-Blockers were initiated between 37 days5 and 30 min23 before surgery and continued between 525 and 30 days21 after surgery (table 3). Nine studies had 30-day all-cause mortality available. One of these studies23 separated post-discharge from in-hospital mortality and therefore it was necessary to sum the two time windows to obtain events from the time of randomisation (see online supplementary appendix 3). In two studies all-cause mortality data were only available to discharge.24 ,25
Assessment of quality of trials
The risk of bias is shown in table 4. Publication bias was assessed using a funnel plot (see online supplementary appendix 4) which did not show significant asymmetry, but this cannot definitively exclude publication bias.
In total there were 10 529 patients in nine secure trials, with 162 deaths in 5264 patients randomised to β-blockers and 129 deaths in 5265 patients randomised to placebo. In the nine secure studies, β-blockers caused a statistically and clinically significant increase in mortality of 27% (RR 1.27, 95% CI 1.01% to 1.60%, p=0.04). There was little heterogeneity between studies (I2=0%, p=0.68; figure 2).
We conducted a separate meta-analysis of the two insecure studies (figure 3). These show a consensus effect of a non-statistically significant decrease in mortality by more than half (RR 0.42, 95% CI 0.15 to 1.23, p=0.11). There was moderate heterogeneity between the two studies (I2=44%, p=0.18).
The contrast between the secure and the DECREASE studies was statistically significant (p=0.05, figure 4).
Six secure trials provided data for MI; β-blockade was reported to reduce non-fatal MI (RR 0.73, 95% CI 0.61 to 0.88, p=0.001). The DECREASE studies also reported a reduction (RR 0.21, 95% CI 0.03 to 1.61, p=0.13), with no significant contrast between the secure and the DECREASE studies (p=0.23, figure 5).
Six secure studies provided data for stroke; β-blockers significantly increased stroke (RR 1.73, 95% CI 1.00 to 2.99, p=0.05). DECREASE I reported stroke (RR 1.33, 95% CI 0.30 to 5.93, p=0.71), which was not significantly different from the secure studies (p=0.75, figure 6).
Six secure studies reported hypotension, which occurred more frequently in the β-blocker group than in the control group (15.2% vs 10.0%, RR 1.51, 95% CI 1.37 to 1.67, p<0.00001, figure 7).
The initiation of a course of β-blockers preoperatively in patients undergoing non-cardiac surgery increases mortality by 27%, which is both statistically and clinically significant. DECREASE trials I and IV report findings inconsistent with the intention-to-treat results of secure RCTs on the initiation of β-blockers on perioperative all-cause mortality.
Although β-blockers reduce non-fatal MI, they also increase hypotension and stroke. It is conceivable that an increase in death due to hypotension or stroke was overcoming a reduction in death from MI, leaving a net increase in deaths. However, there are insufficient quantitative data on the subclasses of death to be certain: deaths are relatively few and difficult to classify reliably by cause.
Cardiologists might be tempted to hope that careful uptitration of β blockade (rather than initiation directly to a standard maintenance dose as in the POISE trial) might give benefits without inducing harm. However, the principal grounds for this hope are the DECREASE trials. The investigation committee established that there was no evidence that the published β-blocker uptitration was really done.3 ,4
Cardiologists might also hope that the 100 mg dose that increased mortality in the POISE trial might have been excessive and that commoner dosages such as metoprolol 25 mg three times a day instead might be beneficial rather than harmful. However, the higher headline value of 100 mg in the POISE trial is of metoprolol CR/XL which is a slow-release once-daily preparation with a bioavailability 25–30% lower than that of standard metoprolol.26 ,27 Thus, the dosage of 100 mg CR/XL in the POISE trial is equivalent to the 75 mg/day that accrues with 25 mg three times a day of immediate-release metoprolol, whose initiation cardiologists might consider conventional. The POISE trial was therefore not high-dose.
Within the ESC guidelines and associated meta-analysis, the inclusion of non-secure data caused them to reach a conclusion that β-blockers had a neutral effect on mortality and allowed them to focus on the reduction of non-fatal MI as a surrogate endpoint. This resulted in β-blockers receiving a class I/IIa recommendation, despite secure trials indicating that they increase mortality.
The β-blocker section of the 2009 ESC Guidelines for Perioperative Cardiac Risk Assessment and Management1 requires reconsideration; without the DECREASE studies the profound adverse findings of the large POISE trial are the dominant contributor.
The POISE trial had a protocol of initiating a dose considered by some to be high (100 mg extended-release metoprolol) shortly (2–4 h) before surgery. This has been argued to be unrepresentative of clinical practice, but not been borne out by surveys of practice28 and is similar to the total daily metoprolol dose from other regimens such as 25 mg three times a day which might not be considered dramatic. Nevertheless, POISE-like regimes now have no reason for continuance.
If the appropriateness of the POISE protocol is doubted, then the remaining secure data are not sufficient to guide physicians either way.
Although there is a retrospective study reporting that β-blockade is associated with lower mortality in high-risk but not in low-risk patients,29 the lead author of the most reliable prospective RCT stated that ‘the groups at highest risk looked like they benefited the least, not the most. The notion of targeting high-risk people is not supported by POISE’.30
Opportunity to prevent perioperative deaths
In the present analysis the RR of mortality from randomisation to β-blockade for non-cardiac surgery is 1.27 (95% CI 1.01 to 1.60) or, conversely, randomisation to not having β-blockade has a RR of 0.79 (95% CI 0.63 to 0.99), indicating a 21% reduction. In the UK,31 for example, almost 2.5 million high- or intermediate-risk procedures are performed per year, with deaths at 30 days totalling 47 286.
Could we have found this earlier?
Any one of three considerations might have opened this opportunity earlier. First, with a strong pointer in 2008 that the introduction of β-blockers before surgery increases mortality, we could have avoided the siren call of reduction in non-fatal MI. If a patient succumbs after intervention, knowing that he or she was prevented from having a MI is no consolation.
Second, we could have realised that not all trial data are of equal reliability. The POISE21 investigators prominently carried out anti-bias steps including record-keeping and scrutiny for anomalies which were acted on. For example, they flew to Colombia and Iran to investigate suspicious returns, resulting in invalidation of data from one centre in Colombia and the entire dataset for Iran.
Third, we could have acted on a 2008 meta-analysis33 flagging DECREASE I to be at high risk of bias 4 years before the DECREASE family was formally declared insecure.
Could DECREASE I have been valid?
Data storage appeared to be haphazard for the DECREASE family of studies.4 Of the five investigated DECREASE studies, the only one for which raw data existed was DECREASE VI, but the investigation concluded that this was ‘fictitious data’.
Even for recent studies such as DECREASE V, not a single case record form (CRF) could be found in any location for the 101 patients. In DECREASE IV the key data required to judge outcomes were missing and the adjudication committee was fabricated. A review of the hospital computer information system found that ‘in a large number of cases a myocardial infarct which the researchers had recorded could not be confirmed’ in the hospital records. DECREASE III could not be investigated as all the source documentation was lacking. For DECREASE II only half the CRFs were found and study outcomes were again realised not to have been assessed as described in the publication. The investigation did not attempt to evaluate the distant DECREASE I.
All we know is that the later DECREASE family of studies fell far short of the standards assumed by clinical readers. There are two hypotheses. The first is that standards started high in the uninvestigated DECREASE I and then declined subsequently as more experience was gained as an international perioperative clinical research centre, ultimately reaching the depths of the entirely fictitious DECREASE VI. The alternative hypothesis is that that honesty was low throughout.
This meta-analysis can only include data of which we are aware. There may be further unreported trials. Our group has no direct knowledge of the process that went on in the DECREASE family other than what has been reported by the two investigations conducted by the Board of the Erasmus Medical Centre.3 ,4
While there was minimal evidence that heterogeneity was assessed by Cochrane's I2 among the secure trials, this measure may be low powered to detect such a difference. In addition to vascular surgery, the studies included a wide range of surgeries including abdominal, orthopaedic, urological, gynaecological and plastic surgery, among others. It may therefore be difficult to see if the initiation of a course of β-blockers before surgery in certain patient groups is beneficial.
The meta-analysis is heavily influenced by the POISE trial.21 However, this is appropriate because the POISE trial is by far the largest study and it was well conducted. Without it there is little remaining evidence base.
A statistically significant increase in all-cause mortality has overwhelming clinical significance which cannot be compensated for by a simultaneous reduction in non-fatal events. The use non-fatal MI as a surrogate for death therefore may not be valid for perioperative β-blockade.
Patient safety being paramount, guidelines for perioperative β-blocker initiation should be retracted without further delay. Future guidelines should be accompanied by a commitment from named individuals to retract them immediately if the advice given is later revealed to be harmful.
Routine initiation of β-blockers for this indication should not be recommended, except in the context of RCTs which should be designed carefully, conducted honestly and reported truthfully.
Contributors SB, MJS and DPF conceived the study, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. SB is the guarantor. SB, MJS, GC, JM and DPF drafted and revised the manuscript.
Funding DPF is supported by the British Heart Foundation (FS/10/038).
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 3.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/3.0/