The ACC and the American Heart Association have released a new, updated guideline for minimizing the risk of cardiovascular complications before, during and after a patient undergoes non-cardiac surgery. The "2014 ACC/AHA Guideline on Perioperative Cardiovascular Evaluation and Management of Patients Undergoing Noncardiac Surgery," was published Aug. 1 in the Journal of the American College of Cardiology, and was accompanied by a systematic review of the guideline on perioperative beta blockade in non-cardiac surgery. The updated guideline was based on a thorough evidence review that analyzed randomized controlled trials, case series, systematic reviews, cohort studies and registries. It puts particular emphasis on preoperative evaluation, offering recommendations on when clinical testing – such as 12-lead ECG, assessment of left ventricular function, coronary angiography and stress testing – may be warranted. Additionally proposed is a new testing algorithm that incorporates a risk calculator while continuing to emphasize that stable patients undergoing low-risk surgery or practice good exercise tolerance rarely need further cardiac studies. Among its various other recommendations, the guideline recommends that beta blockers should be continued in patients undergoing non-cardiac surgery who have been on the drugs chronically. Also noted is that it may be reasonable to begin perioperative beta blockers for patients with intermediate or high risk myocardial ischemia, or for patients with three or more Revised Cardiac Risk Index risk factors such as heart failure, coronary artery disease, renal insufficiency, diabetes mellitus, or even cerebrovascular accident. Regardless, initiation of therapy should be long enough in advance to assess the safety and tolerability of any beta blocker before surgery. These recommendations for perioperative beta blocker use were based on the results of the accompanied systematic review of 17 studies that found that "perioperative beta blockade started within one day or less before non-cardiac surgery prevents nonfatal myocardial infarction but increases risk of stroke, mortality, hypotension and bradycardia." The authors of the review note that moving forward, additional multicenter random control trials are needed to address the knowledge gap of insufficient data on beta blockade started two or more days prior to surgery. Further recommendations in the updated guideline address elective non-cardiac surgery, which should be delayed 14 days after balloon angioplasty, 30 days after bare-metal stent implantation, and optimally 365 days after drug-eluting stent implantation. Coronary artery bypass graft surgery should only be performed if it would be indicated independent of a non-cardiac surgery. Also analyzed is dual antiplatelet therapy, with the writing committee recommending that its management be determined by a committee of the patient, and his or her surgeon, anesthesiologist, and cardiologist to weigh the relative risk of bleeding versus prevention of stent thrombosis. In patients that have received coronary stents and must undergo a surgical procedure that mandates the discontinuation of P2Y12 platelet receptor inhibitor therapy, it is recommended that the patient continue with aspirin and restart P2Y12 as soon as possible. The writing committee notes that moving forward, "future research on perioperative evaluation and management span the spectrum from randomized controlled trials to regional and national registries to focus on patient outcomes." They add that "future research will also need to understand how information regarding perioperative risk is incorporated into patient decision-making." In a related editorial comment, Jeffrey Anderson, MD, FACC, et al. note, "the clinical practice guidelines on cardiovascular care in the perioperative period represents a fresh and objective review of old and new evidence in this important clinical arena." They add that, "Clinicians will find the recommendations in these revised clinical practice guidelines useful in their daily work and can be reassured that the recommendations have been vetted thoroughly by the most rigorous scientific process. Furthermore, the recommendations in both documents are fundamentally aligned, so that cardiovascular clinicians worldwide may deliver optimal, standardized care." Keywords: Exercise Tolerance, Stroke, Myocardial Infarction, Ventricular Function, Left, Drug-Eluting Stents, Hypotension, Risk Factors, Perioperative Period, Electrocardiography, Registries, Renal Insufficiency, Coronary Angiography, Metals, Thrombosis, Heart Failure, Bradycardia, Coronary Artery Bypass, Cardiac Surgical Procedures, Diabetes Mellitus, Exercise Test < Back to Listings
Randomized controlled trials (RCTs) have yielded conflicting results regarding the ability of beta‐blockers to influence perioperative cardiovascular morbidity and mortality. Thus routine prescription of these drugs in an unselected population remains a controversial issue. A previous version of this review assessing the effectiveness of perioperative beta‐blockers in cardiac and non‐cardiac surgery was last published in 2018. The previous review has now been split into two reviews according to type of surgery. This is an update, and assesses the evidence in non‐cardiac surgery only. To assess the effectiveness of perioperatively administered beta‐blockers for the prevention of surgery‐related mortality and morbidity in adults undergoing non‐cardiac surgery. We searched CENTRAL, MEDLINE, Embase, CINAHL, Biosis Previews and Conference Proceedings Citation Index‐Science on 28 June 2019. We searched clinical trials registers and grey literature, and conducted backward‐ and forward‐citation searching of relevant articles. We included RCTs and quasi‐randomized studies comparing beta‐blockers with a control (placebo or standard care) administered during the perioperative period to adults undergoing non‐cardiac surgery. If studies included surgery with different types of anaesthesia, we included them if 70% participants, or at least 100 participants, received general anaesthesia. We excluded studies in which all participants in the standard care control group were given a pharmacological agent that was not given to participants in the intervention group, studies in which all participants in the control group were given a beta‐blocker, and studies in which beta‐blockers were given with an additional agent (e.g. magnesium). We excluded studies that did not measure or report review outcomes. Two review authors independently assessed studies for inclusion, extracted data, and assessed risks of bias. We assessed the certainty of evidence with GRADE. We included 83 RCTs with 14,967 participants; we found no quasi‐randomized studies. All participants were undergoing non‐cardiac surgery, and types of surgery ranged from low to high risk. Types of beta‐blockers were: propranolol, metoprolol, esmolol, landiolol, nadolol, atenolol, labetalol, oxprenolol, and pindolol. In nine studies, beta‐blockers were titrated according to heart rate or blood pressure. Duration of administration varied between studies, as did the time at which drugs were administered; in most studies, it was intraoperatively, but in 18 studies it was before surgery, in six postoperatively, one multi‐arm study included groups of different timings, and one study did not report timing of drug administration. Overall, we found that more than half of the studies did not sufficiently report methods used for randomization. All studies in which the control was standard care were at high risk of performance bias because of the open‐label study design. Only two studies were prospectively registered with clinical trials registers, which limited the assessment of reporting bias. In six studies, participants in the control group were given beta‐blockers as rescue therapy during the study period. The evidence for all‐cause mortality at 30 days was uncertain; based on the risk of death in the control group of 25 per 1000, the effect with beta‐blockers was between two fewer and 13 more per 1000 (risk ratio (RR) 1.17, 95% confidence interval (CI) 0.89 to 1.54; 16 studies, 11,446 participants; low‐certainty evidence). Beta‐blockers may reduce the incidence of myocardial infarction by 13 fewer incidences per 1000 (RR 0.72, 95% CI 0.60 to 0.87; 12 studies, 10,520 participants; low‐certainty evidence). We found no evidence of a difference in cerebrovascular events (RR 1.65, 95% CI 0.97 to 2.81; 6 studies, 9460 participants; low‐certainty evidence), or in ventricular arrhythmias (RR 0.72, 95% CI 0.35 to 1.47; 5 studies, 476 participants; very low‐certainty evidence). Beta‐blockers may reduce atrial fibrillation or flutter by 26 fewer incidences per 1000 (RR 0.41, 95% CI 0.21 to 0.79; 9 studies, 9080 participants; low‐certainty evidence). However, beta‐blockers may increase bradycardia by 55 more incidences per 1000 (RR 2.49, 95% CI 1.74 to 3.56; 49 studies, 12,239 participants; low‐certainty evidence), and hypotension by 44 more per 1000 (RR 1.40, 95% CI 1.29 to 1.51; 49 studies, 12,304 participants; moderate‐certainty evidence). We downgraded the certainty of the evidence owing to study limitations; some studies had high risks of bias, and the effects were sometimes altered when we excluded studies with a standard care control group (including only placebo‐controlled trials showed an increase in early mortality and cerebrovascular events with beta‐blockers). We also downgraded for inconsistency; one large, well‐conducted, international study found a reduction in myocardial infarction, and an increase in cerebrovascular events and all‐cause mortality, when beta‐blockers were used, but other studies showed no evidence of a difference. We could not explain the reason for the inconsistency in the evidence for ventricular arrhythmias, and we also downgraded this outcome for imprecision because we found few studies with few participants. The evidence for early all‐cause mortality with perioperative beta‐blockers was uncertain. We found no evidence of a difference in cerebrovascular events or ventricular arrhythmias, and the certainty of the evidence for these outcomes was low and very low. We found low‐certainty evidence that beta‐blockers may reduce atrial fibrillation and myocardial infarctions. However, beta‐blockers may increase bradycardia (low‐certainty evidence) and probably increase hypotension (moderate‐certainty evidence). Further evidence from large placebo‐controlled trials is likely to increase the certainty of these findings, and we recommend the assessment of impact on quality of life. We found 18 studies awaiting classification; inclusion of these studies in future updates may also increase the certainty of the evidence.
This review assessed evidence from randomized controlled trials (RCTs) on whether beta‐blockers reduce deaths or other serious events when given to people undergoing surgery other than heart surgery. The findings for heart surgery are covered in another review. Background Surgery increases stress in the body, which responds by releasing the hormones adrenaline and noradrenaline. Stress from surgery can lead to death or other serious events such as heart attacks, stroke, or an irregular heartbeat. For surgery that does not involve the heart, an estimated 8% of people may have injury to their heart around the time of surgery. Beta‐blockers are drugs that block the action of adrenaline and noradrenaline on the heart. Beta‐blockers can slow down the heart, and reduce blood pressure, and this may reduce the risk of serious events. However, beta‐blockers may lead to a very low heart rate or very low blood pressure which could increase the risk of death or a stroke. Prevention of early complications after surgery is important, but using beta‐blockers to prevent these complications is controversial. Study characteristics The evidence is current to 28 June 2019. We included 83 RCTs with 14,967 adults who were undergoing different types of surgery other than heart surgery. Eighteen studies are awaiting classification (because we did not have enough details to assess them), and three studies are ongoing. The types of beta‐blockers used in the studies were: propranolol, metoprolol, esmolol, landiolol, nadolol, atenolol, labetalol, oxprenolol, and pindolol. Studies compared these beta‐blockers with either a placebo (disguised to look like a beta‐blocker but containing no medicine) or with standard care. Key results Beta‐blockers may make little or no difference to the number of people who die within 30 days of surgery (16 studies, 11,446 participants; low‐certainty evidence), have a stroke (6 studies, 9460 participants; low‐certainty evidence), or experience ventricular arrhythmias (irregular heartbeat rhythms, starting in the main chambers of the heart, that are potentially life‐threatening and may need immediate medical treatment; 5 studies, 476 participants; very low‐certainty evidence). We found that beta‐blockers may reduce atrial fibrillation (an irregular heartbeat, starting in the atrial chambers of the heart, that increases the risk of stroke if untreated; 9 studies, 9080 participants; low certainty‐evidence), and the number of people who have a heart attack (12 studies, 10,520 participants; low‐certainty evidence). However, taking beta‐blockers may increase the number of people who experience a very low heart rate (49 studies, 12,239 participants; low‐certainty evidence), or very low blood pressure (49 studies, 12,304 participants; moderate‐certainty evidence), around the time of surgery. In a few studies, we also found little or no difference in the number of people who died after 30 days, who died because of a heart problem, or had heart failure. We found no evidence of whether beta‐blockers alter the length of time in hospital. No studies assessed whether people who were given beta‐blockers had a better quality of life after heart surgery. Certainty of the evidence The certainty of the evidence in this review was limited by including some studies that were at high risk of bias, and we noticed that some of our findings were different if we only included placebo‐controlled studies or studies that reported how participants were randomized. We also found one large, well‐conducted, international study that had different findings to the smaller studies. It showed a reduction in heart attacks and an increase in stroke and all‐cause mortality when beta‐blockers were used, whilst the other studies did not show a clear effect. We were also less certain of the findings for outcomes with few studies, such as for ventricular arrhythmias. Conclusion Although beta‐blockers may make little or no difference to the number of people who die within 30 days, have a stroke, or have ventricular arrhythmias, they may reduce atrial fibrillation and heart attacks. Taking beta‐blockers may increase the number of people with a very low heart rate or very low blood pressure around the time of surgery. Further evidence from large, placebo‐controlled trials is likely to increase the certainty of these findings, and we recommend the assessment of impact on quality of life. GRADE Working Group grades of evidence Cardiovascular mortality and morbidity are prevalent and costly in people undergoing cardiac and non‐cardiac surgery. Worldwide, major perioperative complications are responsible for a third of deaths during the perioperative period (Devereaux 2015). Complications include myocardial infarction and myocardial injury (Sellers 2018), and in non‐cardiac surgery, up to 8% of patients may show signs of myocardial injury during the perioperative period (VISION 2014). Postoperative arrhythmias, such as atrial fibrillation, occur in 3% of people (Sellers 2018). Prevention of postoperative complications remains a major issue (Jørgensen 2018). There are various pharmacological agents that may be used to protect people undergoing surgery against adverse cardiovascular events (Lewis 2018), and alpha‐2 adrenergic agents have also been evaluated for their use perioperatively (Duncan 2018). In addition, the choice of anaesthetic drugs and techniques can also affect cardiovascular outcomes (Hristovska 2017). This review, however, evaluates the effectiveness of beta‐adrenoceptor blocking agents, or beta‐blockers, for this purpose. These pharmacological agents block the actions of the stress hormones epinephrine (adrenaline) and norepinephrine (noradrenaline). They are typically used to manage abnormal heart rhythms, heart failure, coronary heart disease, and their effectiveness has been assessed for hypertension (Wiysonge 2017) and for secondary prevention of stroke (De Lima 2014). Cardiac adverse events appear to be related to the persistently exaggerated sympathetic response that is associated with substantial increases in heart rate and myocardial oxygen consumption. Drugs that block beta‐adrenergic receptors, and thus the sympathetic response, are capable of preventing cardiac complications in people with acute myocardial infarction, silent ischaemia and heart failure (Jørgensen 2018). There is thus a pharmacological rationale to support the use of beta‐blockers in the perioperative period (Oprea 2019). Although several trials have provided encouraging findings demonstrating a reduced perioperative incidence of death from cardiac causes and non‐fatal cardiovascular complications (Mangano 1990; Wallace 1998), the routine administration of beta‐blocking agents in an unselected population before major surgery is still discussed as a controversial topic, especially after the results of the POISE trial were published (POISE 2008), which demonstrated an increase of all‐cause mortality and stroke with the use of beta‐blockers. Until recently, recommendations were largely based on the results of four randomized trials (DECREASE‐IV 2009; POISE 2008; Poldermans 1999; Wallace 1998). After the DECREASE (Dutch Echocardiographic Cardiac Risk Evaluation Applying Stress Echo) trial family, which showed beneficial effects of beta‐blockers on mortality and on prevention of myocardial infarction in the perioperative setting, was discredited in 2011 (Bouri 2014; Chopra 2012), perioperative use of beta‐blockers became once again a matter of controversy. The previous version of this review, which assessed the effectiveness of beta‐blockers in both cardiac and non‐cardiac surgery, excluded one retracted study but did not include an updated search (Blessberger 2018). The previous version has now been split into two reviews according to type of surgery. This is an update, and incorporates new evidence in non‐cardiac surgery only. Evidence for cardiac surgery is reported in (Blessberger 2019). To assess the effectiveness of perioperatively administered beta‐blockers for the prevention of surgery‐related mortality and morbidity in adults undergoing non‐cardiac surgery. We included randomized controlled trials (RCTs) and quasi‐randomized studies in which investigators used methods to allocate participants to groups such as hospital record number or date of birth. We included studies that assessed the effects of beta‐blockers on adults who were 18 years of age or older and who were undergoing non‐cardiac surgery under general anaesthesia. If studies included surgery under different types of anaesthesia, we included the study if more than 100 randomly assigned participants received general anaesthesia, or if more than 70% of participants received general anaesthesia. We excluded trials investigating procedures that required local or regional anaesthesia only. We included studies in which beta‐adrenoceptor‐blockers (beta‐blockers) were administered during the perioperative period; we defined the perioperative period as 30 days before surgery to 30 days after surgery. Beta‐blockers could be started before surgery, during surgery or at the latest by the end of the first day after surgery. Beta‐blockers were given intravenously, orally or via a feeding tube, and were compared with a control (placebo or standard care). We excluded studies in which all participants in the standard care control group were given a pharmacological agent that was not given to participants in the intervention group; similarly, we excluded studies in which all participants in the control group were given a beta‐blocker. We excluded studies (or intervention groups within a multi‐arm study) in which the beta‐blocker was given with a supplementary agent (e.g. magnesium) unless the agent was given in both groups as part of standard care management. We excluded studies that did not measure review outcomes (see Differences between protocol and review). Except for long‐term all‐cause mortality, length of hospital stay, and quality of life, we aimed to collect outcome data that were measured within 30 days postoperatively or before hospital discharge (whichever occurred later). We removed some outcomes in this update (see Differences between protocol and review).
We identified RCTs through literature searching with systematic and sensitive search strategies, as outlined in Chapter 6.4 of theCochrane Handbook for Systematic Reviews of Interventions (Lefebvre 2011). We applied no restrictions to language or publication status. We searched the following databases for relevant trials:
In consultation with the Information Specialist for Cochrane Anaesthesia, we developed a subject‐specific and sensitive search strategy in MEDLINE and other listed databases. We subtracted the previous review's search results (up to and including 2012) from the new search. Search strategies can be found in: Appendix 1; Appendix 2; Appendix 3; Appendix 4; Appendix 5; Appendix 6; Appendix 7. We scanned the following clinical trials registers for ongoing and unpublished trials: We carried out forward‐citation searching of identified included studies published since 2012 in Web of Science on 22 March 2019 (apps.webofknowledge.com). We conducted a search of grey literature using Opengrey on 5 April 2019 (www.opengrey.eu/). In addition, we scanned reference lists of relevant systematic reviews published since 2015. Two review authors (SL, and LF or MP) independently selected studies and extracted data from new included studies. We compared decisions at each stage. In cases of disagreement, we reassessed the respective studies to reach consensus, and if necessary included a third review author (HB) for resolution. We used reference management software to collate the results of searches and to remove duplicates (Endnote). We used Covidence 2019 software to screen results of the search of titles and abstracts and to identify potentially relevant studies. We sourced the full texts of all potentially relevant studies and considered whether they met the inclusion criteria (see Criteria for considering studies for this review). We reviewed abstracts at this stage and included them in the review only if they provided sufficient information and relevant results that included denominator figures for the intervention and control groups. We recorded the number of papers retrieved at each stage and report this information using a PRISMA flow chart (see Figure 1). We report in the review brief details of closely related but excluded papers. We used a data extraction form to collect information and outcome data from studies (Appendix 8). We collected the following information.
We considered the applicability of information from individual studies and generalizability of the data to our intended study population (i.e. the potential for indirectness in our review). In multi‐arm studies, we did not collect data on intervention agents that were not eligible for inclusion in the review. We assessed study quality, study limitations, and the extent of potential bias using the Cochrane 'Risk of bias' tool (Higgins 2017). We considered the following domains.
For each domain, two review authors (SRL, and MP or LF) judged whether study authors made sufficient attempts to minimize bias in their study design. We made judgements using three measures ‐ high, low, or unclear risk of bias. We recorded this in 'Risk of bias' tables and present a 'Risk of bias' graph and a summary 'Risk of bias' figure (see Figure 2 and Figure 3). For other potential risks of bias, we considered the effect of beta‐blockers given as 'rescue therapy' to treat specified conditions. We judged studies to have a high risk of other bias if administration of such 'rescue therapy' had the potential to influence outcome data. We collected dichotomous data for mortality outcomes, acute myocardial infarction, cerebrovascular events, ventricular arrhythmias, atrial fibrillation and atrial flutter, bradycardia, hypotension, and congestive heart failure. We collected continuous data for length of hospital stay. We report dichotomous data as risk ratios (RRs) to compare groups, and continuous data as mean differences (MDs). We report 95% confidence intervals (CI). For multi‐arm studies, which included different types of beta‐blockers or different doses of beta‐blockers, we combined dichotomous data to create a single beta‐blocker group, and we used these composite data in the primary analysis. During subgroup analysis by type of beta‐blocker, we included data separately for each type of beta‐blocker, and used the 'halving method' with data in the control group to avoid a unit of analysis error (Deeks 2017). If multi‐arm studies had included continuous data for length of stay, we planned to calculate combined mean and standard deviation values according to the formula provided in Chapter 7.7.3.8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). If information on both study group allocation and respective outcomes was available, we re‐included withdrawn participants in keeping with the intention‐to‐treat principle. If information was not available, we performed an available case analysis. We did not perform imputation techniques. We excluded continuous data that assessed length of stay, if a range of dispersion (standard deviation or standard error) was not provided along with mean values. When both measures of spread (standard deviation and standard error) were presented, we used the standard deviation as the measure of choice. We did not apply imputation techniques. We attempted contact with some study authors for additional information; we report this information in the Notes section of the Characteristics of included studies. We assessed whether evidence of inconsistency was apparent in our results by considering heterogeneity. We assessed clinical and methodological heterogeneity by comparing similarities in our included studies between study designs, participants, interventions, and outcomes, and used the data collected from the full‐text reports (as stated in Data collection and analysis). We explored clinical and methodological heterogeneity through subgroup analysis. We assessed statistical heterogeneity by calculating the ChI2 test or I2 statistic (Higgins 2003), and judged any heterogeneity above an I2 statistic value of 40% and a Chi2 P value of 0.05 or less to indicate moderate to substantial statistical heterogeneity (Deeks 2017). We did not conduct meta‐regression to explore heterogeneity in this updated review (see Differences between protocol and review). As well as looking at statistical results, we considered point estimates and overlap of CIs. If CIs overlap, then results are more consistent. However, combined studies may show a large consistent effect but with significant heterogeneity. We, therefore, planned to interpret heterogeneity with caution (Guyatt 2011a). We attempted to source published protocols for each of our included studies by using clinical trials registers. We planned to compare published protocols with published study results, to assess the risk of selective reporting bias. In addition, we appraised reporting bias through visual assessment of funnel plots (Egger 1997). We only included figures of funnel plots in the review in which we identified possible reporting bias based on visual assessment. We presented a statistical summary of treatment effects in the absence of significant clinical or methodological heterogeneity. We used the statistical calculator in Review Manager 5 (RevMan 5) to perform meta‐analysis (Review Manager 2014). For dichotomous outcomes, we used the Mantel‐Haenszel random‐effects model to account for potential variability in participant conditions between studies (Borenstein 2010). For continuous outcomes, we used inverse variance using a random‐effects model. We calculated CIs at 95%, and used a P value of 0.05 or less to judge whether a result was statistically significant; for statistically significant results, we also reported the number needed to treat for an additional beneficial outcome (NNTB) or the number needed to treat for an additional harmful outcome (NNTH). We considered imprecision in the results of analyses by assessing the CI around effect measures; a wide CI would suggest a higher level of imprecision in our results. A small number of studies may also reduce precision (Guyatt 2011b). In subgroup analysis, we evaluated the following factors that may influence the results.
In multi‐arm studies of more than one type of beta‐blocking agent, we compared each type of beta‐blocker using the 'halving method' for the control group data (Higgins 2011b); thus, we avoided a unit of analysis error. We planned to complete subgroup analysis in which we found more than 10 studies (Deeks 2017), for the following outcomes.
We explored the potential effect of decisions made as part of the review process. In each sensitivity analysis, we compared the effect estimate with the main analysis. We reported these effect estimates only if they indicated a difference in interpretation of the effect. We performed the following sensitivity analyses.
In addition to sensitivity analyses reported in an earlier version of the review (Blessberger 2018), we also used sensitivity analysis to explore the potential effect of studies in which the control group were given beta‐blockers as a 'rescue therapy'. The review included several studies published before 2000, in which clinical management may differ from current standards. We, therefore, made a post‐hoc decision to explore the potential effect of these early studies on the outcomes; in sensitivity analysis, we excluded studies published before 2000. We calculated RRs using a random‐effects model for all analyses in the review. Although the random‐effects model accounted for potential variation in the population, this statistical tool did not account for outcomes with rare events. In sensitivity analysis, we evaluated the effect of outcomes with events fewer than 1% using Peto odds ratio (Higgins 2011). We assessed the effect of these potential biases on the following outcomes.
One review author (SL) used the GRADE system to assess the certainty of the body of evidence and construct a 'Summary of findings' table associated with the following outcomes (Guyatt 2008).
The GRADE approach appraises the certainty of a body of evidence based on the extent to which we can be confident that an estimate of effect or association reflects the item being assessed. Evaluation of the certainty of a body of evidence considers within‐study risk of bias, directness of the evidence, heterogeneity of the data, precision of the effect estimates, and risk of publication bias. We constructed a 'Summary of findings' table using GRADEpro GDT software (gradepro.org). We used the GRADE approach to appraise the certainty of the body of evidence for the remaining outcomes, but we did not construct a 'Summary of findings' table for these outcomes. After the removal of duplicates from the search results, we screened 5972 titles and abstracts for this update, which included forward‐ and backward‐citation searches, clinical trials registers and grey literature. We sourced 312 full‐text reports to assess eligibility. See Figure 1.
See Characteristics of included studies. We included 83 RCTs with 14,967 participants (Ali 2015; Alkaya 2014; Aoyama 2016; Apipan 2010; Bayliff 1999; Bhattacharjee 2016; Burns 1988; Ceker 2015; Chung 1992; Cucchiara 1986; DIPOM 2006; Do 2012; El‐Shmaa 2016; Gibson 1988; Goyagi 2005; Gupta 2011; Harasawa 2006; Helfman 1991; Horikoshi 2017; Inada 1989; Inoue 2010; Jakobsen 1986; Jakobsen 1992; Jakobsen 1997; Jangra 2016; Kao 2017; Kawaguchi 2010; Kindler 1996; Lai 2006; Lee 1994; Lee 2010; Lee 2015; Lee 2017; Lim 2000; Liu 1986; Liu 2006; Louizos 2007; Magnusson 1986; Mallon 1990; Meftahuzzaman 2014; Menigaux 2002; Mikawa 1991; Miller 1990; Miller 1991; Miyazaki 2009; Moon 2011; Neary 2006; Ohri 1999; Ojima 2017; Oxorn 1990; Park 2009; POBBLE 2005; POISE 2008; PRESAGE 2016; Raby 1999; Safwat 1984; Shailaja 2013; Sharma 1996; Sharma 2018; Shrestha 2011; Shukla 2010; Singh 1995; Singh 2010; Singh 2012; Srivastava 2015; Stone 1988; Sugiura 2007; Tendulkar 2017; Ugur 2007; Unal 2008; Urias 2016; Van den Berg 1998; Verma 2018; Wajima 2011; Wallace 1998; Ward‐Booth 1983; White 2003; Whitehead 1980; Yamazaki 2005; Yang 2006; Yang 2008; Yoshida 2017; Zaugg 1999). We found no quasi‐randomized studies. We included one study for which we could only source the abstract and this limited the details of study characteristics that we were able to extract (Urias 2016). We sourced the full text of all remaining studies. This review included 52 new non‐cardiac surgery studies (Ali 2015; Alkaya 2014; Aoyama 2016; Bhattacharjee 2016; Ceker 2015; Chung 1992; Do 2012; El‐Shmaa 2016; Goyagi 2005; Harasawa 2006; Helfman 1991; Horikoshi 2017; Inoue 2010; Jakobsen 1986; Jangra 2016; Kao 2017; Kindler 1996; Lee 1994; Lee 2015; Lee 2017; Lim 2000; Louizos 2007; Mallon 1990; Meftahuzzaman 2014; Menigaux 2002; Mikawa 1991; Miyazaki 2009; Ohri 1999; Ojima 2017; Park 2009; PRESAGE 2016; Safwat 1984; Shailaja 2013; Sharma 1996; Sharma 2018; Shrestha 2011; Singh 1995; Singh 2010; Singh 2012; Srivastava 2015; Sugiura 2007; Tendulkar 2017; Ugur 2007; Unal 2008; Urias 2016; Van den Berg 1998; Verma 2018; Wajima 2011; Ward‐Booth 1983; White 2003; Yamazaki 2005; Yoshida 2017). The remaining studies were previously included in Blessberger 2018. Participants were scheduled for the surgery, which we categorized as high risk, medium risk, and low risk (Kristensen 2014). We categorized studies in which study authors did not specify types of surgery as medium risk. Twenty‐two studies included high‐risk surgeries which were as follows.
Forty‐eight studies included medium‐risk surgeries, which were as follows.
Eleven studies included low‐risk surgeries, which were as follows. All trials included participants under general anaesthesia only, except four studies, which included participants who received other types of anaesthesia (DIPOM 2006; POISE 2008; Raby 1999; Yang 2006); at least 100 participants, or 70% of participants in each group received general anaesthesia. Study authors in these four trials did not report subgroup data according to type of anaesthesia. We collected data from study reports on additional risk factors for included participants; we used information reported in the baseline characteristics tables and in the study inclusion and exclusion criteria. We summarized the most commonly reported factors in a table (Appendix 9). Thirty studies excluded participants who were taking beta‐blockers preoperatively (Alkaya 2014; Bayliff 1999; DIPOM 2006; Gibson 1988; Horikoshi 2017; Inada 1989; Jangra 2016; Lai 2006; Lee 2015; Lim 2000; Liu 2006; Mallon 1990; Menigaux 2002; Miller 1990; Neary 2006; Oxorn 1990; POBBLE 2005; POISE 2008; PRESAGE 2016; Safwat 1984; Sharma 2018; Singh 2010; Singh 2012; Srivastava 2015; Sugiura 2007; Unal 2008; Verma 2018; Yang 2006; Yang 2008; Zaugg 1999). Whilst no studies specifically included participants who were already taking beta‐blockers, four studies reported in baseline characteristics tables that at least some participants were taking beta‐blockers pre‐operatively (Helfman 1991; Raby 1999; Van den Berg 1998; Wallace 1998); the remaining studies did not report this information. Six studies included only participants who had hypertension (Ceker 2015; Magnusson 1986; Shailaja 2013; Sharma 1996; Stone 1988; Sugiura 2007), whilst 12 studies excluded participants who had a history of hypertension (Bhattacharjee 2016; Chung 1992; El‐Shmaa 2016; Goyagi 2005; Lee 2015; Liu 1986; Meftahuzzaman 2014; Menigaux 2002; Sharma 2018; Singh 2012; Srivastava 2015; Verma 2018). Seventeen studies reported in baseline characteristics tables that at least some participants had a history of hypertension (Bayliff 1999; DIPOM 2006; Gibson 1988; Helfman 1991; Horikoshi 2017; Inada 1989; Kawaguchi 2010; Lai 2006; Lim 2000; Miller 1991; Ojima 2017; POISE 2008; PRESAGE 2016; Safwat 1984; Van den Berg 1998; Wallace 1998; Zaugg 1999). In addition, we noted that one study included only participants with diabetes (DIPOM 2006), and one study included only participants who were cigarette smokers (Louizos 2007). Four studies specified inclusion of only elderly participants (Lai 2006; Liu 2006; Miyazaki 2009; Zaugg 1999), and one included only participants described as middle‐aged to elderly (Van den Berg 1998). However, because the surgical population in this review was mixed, we noted that some studies also typically included an elderly population. All studies were conducted in a hospital setting. Studies did not always report whether they were conducted in a single centre or in multiple centres. Six studies reported that they were multi‐centre studies (Cucchiara 1986; DIPOM 2006; Miller 1991; POBBLE 2005; POISE 2008; Yang 2006), and we assumed that the remaining studies were all single‐centre studies. Twenty‐two studies were multi‐arm studies and included: Types of beta‐blockers assessed were: In nine studies, beta‐blockers were titrated according to heart rate or blood pressure (Ali 2015; DIPOM 2006; Harasawa 2006; Kawaguchi 2010; POBBLE 2005; Raby 1999; Wallace 1998; Yang 2008; Zaugg 1999). In the remaining studies, beta‐blockers were given at a fixed dose. In 18 studies, administration was started before surgery, and we defined this time point as any time up to 15 minutes before the induction of anaesthesia (Ali 2015; Apipan 2010; Bayliff 1999; Burns 1988; DIPOM 2006; Gupta 2011; Jakobsen 1986; Jakobsen 1992; Jakobsen 1997; Magnusson 1986; POBBLE 2005; POISE 2008; Shrestha 2011; Shukla 2010; Ward‐Booth 1983; Whitehead 1980; Yang 2006; Yang 2008). In six studies, administration was started after surgery (Harasawa 2006; Ojima 2017; Park 2009; PRESAGE 2016; Raby 1999; Wallace 1998). Urias 2016 did not report time of administration, and Zaugg 1999 was a multi‐arm study that included administration pre‐ and postoperatively and administration intraoperatively. In the remaining studies, administration was during surgery; we defined this time point as starting immediately before, and up to 15 minutes before, induction of anaesthesia until emergence from anaesthesia. Duration of administration varied across studies. All studies included at least one review outcome as this formed part of the inclusion criteria for this review. However, we found that we could not use outcome data in 15 studies because they had not clearly reported data (Alkaya 2014; Chung 1992; Do 2012; Gibson 1988; Helfman 1991; Jakobsen 1986; Jangra 2016; Kindler 1996; Oxorn 1990; Sharma 1996; Singh 2010; Van den Berg 1998; Verma 2018; Wajima 2011; White 2003). We reported numbers of studies for each outcome in Effects of interventions. We found that most studies did not report sources of support or conflicts of interest. Seven studies reported support from pharmaceutical companies (Inada 1989; Jakobsen 1986; Lee 2017; Mallon 1990; Miller 1991; Raby 1999; Whitehead 1980). Twenty‐four studies reported departmental or other sources of funding, which we assumed to be independent (Ali 2015; Bayliff 1999; Kao 2017; Kawaguchi 2010; Liu 1986; Magnusson 1986; Menigaux 2002; Neary 2006; POBBLE 2005; Wajima 2011; Wallace 1998; White 2003; Yang 2006), or that they received no funding (Alkaya 2014; Bhattacharjee 2016; El‐Shmaa 2016; Gupta 2011; Horikoshi 2017; Jangra 2016; PRESAGE 2016; Sharma 2018; Singh 2010; Ugur 2007; Verma 2018). Three studies declared support from both pharmaceutical and independent sources (DIPOM 2006; Helfman 1991; POISE 2008). We excluded 239 articles following assessment of full texts (Figure 1). It was not practical to report the details of all 239 excluded articles in the review, and therefore, we report the details of only seven studies, which we consider to be most closely related to our review criteria (Chae 1990; Ryder 1973; Sezai 2015; Taenaka 2013; Tan 2002; Vucevic 1992; Zmora 2016). In summary, the reasons for exclusion were because studies did not measure or report review outcomes (152 studies); articles were the wrong design (such as commentaries or editorials) or were not RCTs (50 studies); studies had an ineligible population (13 studies), this included studies in which surgical procedures were not conducted under general anaesthesia and one study in which participants were as young as 10 years of age (Ryder 1973); studies had an ineligible intervention or comparison (23 studies), this includes studies that did not have a placebo or standard care group, participants in the intervention group were given an additional agent that was not a beta‐blocker (Chae 1990; Tan 2002; Zmora 2016), all participants were given beta‐blockers after surgery during the study follow‐up period (Sezai 2015), and administration was started up to seven days after surgery (Taenaka 2013). One old study did not report the number of participants in each group, and contact author detail was insufficient (Vucevic 1992). In addition, we re‐evaluated studies included in the previous version of the review (Blessberger 2018), and excluded three of these: in one study, participants were undergoing a non‐surgical procedure (Sandler 1990); in one study, the control group included an additional intervention (dobutamine echocardiography), which was not equivalent to standard care practices in other control groups (Marwick 2009); and in one study, a change to the review outcomes meant that one study no longer measured or reported review outcomes (Coleman 1980). See Characteristics of excluded studies for the 10 cited studies. This review does not include studies that were previously excluded; details of previous exclusions can be found elsewhere (Blessberger 2014; Blessberger 2018). We found 18 studies awaiting classification (ACTRN12605000639628; ACTRN12615000889; Birbicer 2007; Boussofara 2001; Boussofara 2004; Gong 1999; Hornamand 2017; Inada 2002; Itani 2013; Joo 2010; Kajiura 2013; Kawano 2005; NCT02466542; Tangoku 2016; UMIN000024040; Wang 1994; Wang 1999; Yuan 1994). Three studies were published only as abstracts, with insufficient detail to assess eligibility (Itani 2013; Kajiura 2013; Tangoku 2016), and we were unable to access the full text of seven studies (Birbicer 2007; Boussofara 2001; Boussofara 2004; Gong 1999; Inada 2002; Wang 1994; Yuan 1994). Four studies are awaiting translation in order to assess study eligibility (Hornamand 2017; Joo 2010; Kawano 2005; Wang 1999). Four studies were described as completed in a clinical trials register but results are not yet published (ACTRN12605000639628; ACTRN12615000889; NCT02466542; UMIN000024040). See Characteristics of studies awaiting classification. We found three ongoing studies (EUCTR2010‐021844‐17; NCT01555554; NCT03138603). One study compares esmolol with a placebo in people undergoing arterial vascular surgery (EUCTR2010‐021844‐17), one compares propranolol with a placebo given to veterans with post‐traumatic stress disorder (PTSD), who are scheduled for any type of surgical procedure under general anaesthesia (NCT01555554), and one compares metoprolol with a placebo in participants who have or are at risk of coronary artery disease and are scheduled for major non‐cardiac surgery (NCT03138603). See Characteristics of ongoing studies. See Characteristics of included studies, Figure 2 and Figure 3.
Overall, we found 26.5% studies were at high risk of bias in at least one domain. For random sequence generation, we found 36 studies that reported sufficient methods to randomize participants to groups, and we judged these studies to be at low risk of selection bias (Aoyama 2016; Bhattacharjee 2016; Ceker 2015; DIPOM 2006; El‐Shmaa 2016; Gupta 2011; Helfman 1991; Horikoshi 2017; Jangra 2016; Kawaguchi 2010; Kindler 1996; Lee 2017; Mallon 1990; Menigaux 2002; Miller 1990; Miller 1991; Miyazaki 2009; Neary 2006; Ojima 2017; Park 2009; POBBLE 2005; POISE 2008; PRESAGE 2016; Raby 1999; Shailaja 2013; Sharma 2018; Singh 1995; Singh 2010; Srivastava 2015; Tendulkar 2017; Ugur 2007; Van den Berg 1998; Verma 2018; Wajima 2011; Wallace 1998; Yang 2006). We judged the risk of selection bias for random sequence generation in the remaining studies to be unclear. For allocation concealment, we judged 28 studies to be at low risk of bias (Aoyama 2016; Bhattacharjee 2016; Ceker 2015; DIPOM 2006; El‐Shmaa 2016; Gupta 2011; Horikoshi 2017; Jangra 2016; Lee 2017; Mallon 1990; Meftahuzzaman 2014; Miller 1990; Miller 1991; Moon 2011; Neary 2006; Ojima 2017; Park 2009; POBBLE 2005; POISE 2008; PRESAGE 2016; Raby 1999; Shailaja 2013; Sharma 2018; Singh 1995; Singh 2010; Van den Berg 1998; Wallace 1998; Yang 2006). We judged the risk of selection bias for allocation concealment in the remaining studies to be unclear, because of inadequate reporting. Some of the studies in this review compared a beta‐blocker with standard care, and because it was not feasible to blind personnel to the intervention, we judged all studies with a standard care control group to be at high risk of performance bias because of this open‐label study design (Ali 2015; Kawaguchi 2010; Lai 2006; Liu 2006; Ohri 1999; PRESAGE 2016; Stone 1988; Tendulkar 2017; Yang 2008; Zaugg 1999). In addition, we judged four placebo‐controlled trials to be at high risk of performance bias because personnel were aware, or appeared to be aware, of group allocation (El‐Shmaa 2016; Neary 2006; POBBLE 2005; Raby 1999). In 13 studies, study authors reported insufficient information to ascertain whether anaesthetists or clinicians were blinded and we judged risk of performance bias in these studies to be unclear (Bhattacharjee 2016; Do 2012; Goyagi 2005; Harasawa 2006; Kao 2017; Lee 1994; Lee 2010; Lim 2000; Safwat 1984; Shrestha 2011; Unal 2008; Yamazaki 2005; Yoshida 2017). We judged the remaining studies to be at low risk of bias. Twenty‐nine studies reported that outcome assessors were blinded and we judged these studies to be at low risk of detection bias (Apipan 2010; Burns 1988; Ceker 2015; DIPOM 2006; El‐Shmaa 2016; Gupta 2011; Kindler 1996; Jakobsen 1986; Lee 2015; Lee 2017; Lim 2000; Menigaux 2002; Miyazaki 2009; Moon 2011; Neary 2006; Ojima 2017; Oxorn 1990; Park 2009; POBBLE 2005; POISE 2008; Raby 1999; Safwat 1984; Shailaja 2013; Sharma 2018; Sugiura 2007; Ugur 2007; Wajima 2011; Ward‐Booth 1983; Whitehead 1980). We judged studies with a standard care control group to be at high risk of detection bias if study authors did not describe whether outcome assessors were blinded (Ali 2015; Kawaguchi 2010; Lai 2006; Liu 2006; Ohri 1999; PRESAGE 2016; Stone 1988; Tendulkar 2017; Yang 2008; Zaugg 1999). Study authors did not describe whether outcome assessors were blinded in the remaining studies and we judged the risk of bias to be unclear. Most studies did not report losses, and therefore we assumed that there were no losses and we used the numbers of randomized participants to inform the assumed number of analysed participants. In addition, most studies did not report whether they used intention‐to‐treat (ITT) analysis and when losses were reported and study authors did not state use of ITT analysis, we assumed that study investigators used per‐protocol analysis. We judged five studies to have a high risk of attrition bias (Ali 2015; Cucchiara 1986; DIPOM 2006; Kao 2017; Yang 2006); these studies lost more than 10% of participants, or loss of participants was imbalanced between groups. We judged the remaining studies to have a low risk of attrition bias because study authors reported no losses or losses were fewer than 10% and we did not expect the loss to influence outcome data. Only two studies were prospectively registered with a clinical trials register (Lee 2017; Ojima 2017). We judged Ojima 2017 to be at low risk of reporting bias but, because the clinical trials register documents listed primary and secondary outcomes that were not consistent with the published study report, we judged the risk of reporting bias in Lee 2017 to be unclear. Seven studies reported clinical trial registration that was retrospective, and therefore, we could not feasibly assess risk of bias from these registration documents (DIPOM 2006; Horikoshi 2017; Kawaguchi 2010; Lee 2015; POBBLE 2005; POISE 2008; PRESAGE 2016). We judged these, and all other studies, to have an unclear risk of reporting bias because we could not assess this domain without access to published protocols or prospectively registered clinical trial documents. Six studies reported use of beta‐blockers as rescue therapy in the control group (Gibson 1988; Louizos 2007; Ojima 2017; Raby 1999; Whitehead 1980; Yang 2006). We believed that this introduced considerable bias to the data and we judged all these studies to be at high risk of bias. Three studies had limited information in the report and it was also not feasible to effectively assess risks of other bias in these studies (Burns 1988; Urias 2016; Ward‐Booth 1983). We used the information reported in the English abstract and tables in one study written in Korean and, similarly, we were unable to effectively assess other risks of bias from this report (Lee 1994). We noted a difference in the time of administration of two different blockers in Singh 2010, which were either before or after tracheal intubation, and we were uncertain whether this difference in study design might have influenced outcome data. Study authors in Jangra 2016 highlighted differences between groups and we were similarly uncertain whether this difference might influence outcome data. We judged these six studies to be at unclear risk of bias. We did not identify any other sources of bias in the remaining study reports. See: Summary of findings for the main comparison Perioperative beta‐blockers compared to placebo or standard care for preventing surgery‐related mortality and morbidity in adults undergoing non‐cardiac surgery Sixteen studies reported mortality data (Ali 2015; Aoyama 2016; Bayliff 1999; DIPOM 2006; Kawaguchi 2010; Lai 2006; Miller 1990; Miller 1991; Neary 2006; Ojima 2017; POBBLE 2005; POISE 2008; PRESAGE 2016; Wallace 1998; Yang 2006; Yang 2008). We noted that six of these studies had no deaths in either group (Aoyama 2016; Lai 2006; Miller 1990; Miller 1991; Ojima 2017; PRESAGE 2016). Based on the risk of death in the control group of 25 per 1000, the effect with beta‐agonists was between 2 fewer and 13 more deaths per 1000 (risk ratio (RR) 1.17, 95% confidence interval (CI) 0.89 to 1.54; I2 = 5%; 16 studies, 11,446 participants; low‐certainty evidence; Analysis 1.1). See Figure 4.
We used GRADE to downgrade the certainty of the evidence for all‐cause mortality by two levels: one for study limitations because we assessed 10 studies to be at high risk of bias in at least one domain, and the effect estimate was not robust when we excluded studies with poorer methodological standards in sensitivity analysis; and one for inconsistency because from visual inspection of the data, we noted that one large international study represented 63.2% of the weighting and showed an increase in mortality when beta‐blockers were used, which was not replicated in the remaining smaller studies. See summary of findings Table for the main comparison. From visual inspection of a funnel plot, we note the possibility of publication bias for this outcome; however, we did not explore this observation further (Figure 5).
Five studies measured long‐term mortality (DIPOM 2006; Kawaguchi 2010; Neary 2006; Wallace 1998; Yang 2006). Time points of measurement were at three months (Kawaguchi 2010), at six months (DIPOM 2006; Yang 2006), and at two years (Neary 2006; Wallace 1998). We were unable to include Yang 2006 in analysis because data were not clearly reported. We found little or no difference in long‐term mortality according to whether perioperative beta‐blockers were administered (RR 0.82, 95% CI 0.58 to 1.17; I2 = 20%; 5 studies, 1215 participants; low‐certainty evidence; Analysis 1.2). We used GRADE to downgrade the certainty of the evidence by two levels: one level for study limitations because we assessed three studies to be at high risk of bias in at least one domain; and one level for imprecision because, for this outcome, evidence was from few studies with few participants. Three studies reported how many participants died because of cardiac causes (POISE 2008; Wallace 1998; Yang 2006). We found little or no difference in the number of deaths according to whether perioperative beta‐blockers were administered (RR 1.25, 95% CI 0.90 to 1.75; I2 = 0%; 3 studies, 9047 participants; low‐certainty evidence; Analysis 1.3). We used GRADE to downgrade the certainty of the evidence by two levels; one level for study limitation because we assessed one of the three studies to be at high risk of bias in one domain; and one level for imprecision because studies were few, despite a large sample size owing to one large international study (97.1% weighting). Twelve studies reported whether participants had a myocardial infarction (DIPOM 2006; Jakobsen 1997; Lai 2006; Miller 1990; POBBLE 2005; POISE 2008; Raby 1999; Stone 1988; Wallace 1998; Yang 2006; Yang 2008; Zaugg 1999). Three of these studies reported no events in either group (Lai 2006; Miller 1990; Stone 1988). Beta‐blockers may reduce the occurrence of myocardial infarctions (RR 0.72, 95% CI 0.60 to 0.87; I2 = 0%; 12 studies, 10,520 participants; low‐certainty evidence; Analysis 1.4); the number needed to treat for an additional beneficial outcome (NNTB) was 74 (95% CI 52 to 160). We used GRADE to downgrade the certainty of the evidence by two levels: one level for study limitations because we assessed eight studies to be at high risk of bias in at least one domain; and one level for inconsistency because from visual inspection of the data, we noted that one large international study represented 87.9% weighting in the primary analysis and demonstrated a protective effect of beta‐blockers, which was not replicated in the remaining smaller studies. See summary of findings Table for the main comparison. Six studies reported cerebrovascular events (POBBLE 2005; POISE 2008; PRESAGE 2016; Wallace 1998; Yang 2006; Yang 2008). We found no evidence of a difference in cerebrovascular events when beta‐blockers were used (RR 1.65, 95% CI 0.97 to 2.81; I2 = 0%; 6 studies, 9460 participants; low‐certainty evidence; Analysis 1.5). We used GRADE to downgrade the certainty of the evidence by one level for inconsistency. From visual inspection of the data, we noted that one large international study showed an increase in cerebrovascular events when beta‐blockers were used, but this effect was not replicated in the smaller studies and we found too few studies in order to explore this difference effectively through subgroup analysis. We also downgraded the evidence by one level for study limitations; we found that the effect was not robust when we excluded studies with poorer methodological standards during sensitivity analyses. See summary of findings Table for the main comparison. Five studies reported ventricular arrhythmias (Bayliff 1999; Jakobsen 1997; Mallon 1990; POBBLE 2005; Wallace 1998). We found no evidence of a difference to ventricular arrhythmias when beta‐blockers were used (RR 0.72, 95% CI 0.35 to 1.47; I2 = 35%; 5 studies, 476 participants; very low‐certainty evidence; Analysis 1.6). We used GRADE to downgrade the certainty of the evidence by three levels: one level for imprecision because the evidence was from too few participants and few single‐centre studies; one level for inconsistency because we noted a moderate level of statistical heterogeneity that we were unable to explain, and one level for study limitations because we judged several studies to be at high or unclear risk of bias. See summary of findings Table for the main comparison. Nine studies reported atrial fibrillation or atrial flutter (Aoyama 2016; Bayliff 1999; Horikoshi 2017; Lai 2006; Ojima 2017; POISE 2008; PRESAGE 2016; Urias 2016; Yoshida 2017). Beta‐blockers may reduce atrial fibrillation (RR 0.41, 95% CI 0.21 to 0.79; I2 = 67%; 9 studies, 9080 participants; low‐certainty evidence; Analysis 1.7); the NNTB was 39 (95% CI 29 to 108). We used GRADE to downgrade the certainty of the evidence by one level owing to inconsistency; we were unable to effectively assess or explain the substantial statistical heterogeneity that we noted for this outcome. We also downgraded one level for study limitations; the effect estimate was not robust when we excluded studies with poorer methodological standards in sensitivity analysis. See summary of findings Table for the main comparison. Sixty‐five studies reported bradycardia (Ali 2015; Alkaya 2014; Apipan 2010; Bayliff 1999; Bhattacharjee 2016; Burns 1988; Ceker 2015; Chung 1992; Cucchiara 1986; Do 2012; El‐Shmaa 2016; Gibson 1988; Goyagi 2005; Gupta 2011; Harasawa 2006; Helfman 1991; Horikoshi 2017; Inada 1989; Inoue 2010; Jakobsen 1986; Jakobsen 1992; Jakobsen 1997; Kao 2017; Kawaguchi 2010; Kindler 1996; Lee 1994; Lee 2015; Lee 2017; Lim 2000; Liu 1986; Liu 2006; Louizos 2007; Magnusson 1986; Mallon 1990; Meftahuzzaman 2014; Menigaux 2002; Mikawa 1991; Miller 1991; Miyazaki 2009; Moon 2011; Neary 2006; Ojima 2017; Oxorn 1990; Park 2009; POBBLE 2005; POISE 2008; Safwat 1984; Shailaja 2013; Sharma 2018; Shrestha 2011; Shukla 2010; Singh 2010; Srivastava 2015; Stone 1988; Tendulkar 2017; Ugur 2007; Van den Berg 1998; Verma 2018; Wajima 2011; Wallace 1998; Ward‐Booth 1983; Whitehead 1980; Yamazaki 2005; Yang 2006; Yoshida 2017). Of these, four studies reported data unclearly (Alkaya 2014; Jakobsen 1986; Oxorn 1990; Van den Berg 1998), and 12 studies did not clearly report whether bradycardia was measured in all groups (Chung 1992; Do 2012; Gibson 1988; Helfman 1991; Kindler 1996; Neary 2006; Shrestha 2011; Singh 2010; Tendulkar 2017; Verma 2018; Wajima 2011; Yoshida 2017); we did not include these 16 studies in analysis. We note that 21 studies reported no events in either group (Ceker 2015; El‐Shmaa 2016; Harasawa 2006; Horikoshi 2017; Inada 1989; Inoue 2010; Jakobsen 1997; Kao 2017; Lee 1994; Lee 2015; Lee 2017; Louizos 2007; Menigaux 2002; Mikawa 1991; Moon 2011; Ojima 2017; Safwat 1984; Sharma 2018; Srivastava 2015; Ugur 2007; Yamazaki 2005). Perioperative beta‐blockers probably increase incidences of bradycardia (RR 2.49, 95% CI 1.74 to 3.56; I2 = 68%; 49 studies, 12,239 participants; low‐certainty evidence; Analysis 1.8); the number needed to treat for an additional harmful outcome (NNTH) was 18 (95% CI 11 to 37). We used GRADE to downgrade the certainty of the evidence for this outcome by one level for inconsistency owing to substantial statistical heterogeneity which we were unable to explain from subgroup analyses. We also downgraded by one level for study limitations because we assessed several studies to be at high or unclear risk of bias. See summary of findings Table for the main comparison. Sixty‐two studies reported hypotension (Ali 2015; Alkaya 2014; Bayliff 1999; Bhattacharjee 2016; Ceker 2015; Chung 1992; Cucchiara 1986; Do 2012; El‐Shmaa 2016; Gibson 1988; Goyagi 2005; Gupta 2011; Harasawa 2006; Helfman 1991; Horikoshi 2017; Inada 1989; Inoue 2010; Jakobsen 1992; Jakobsen 1997; Jangra 2016; Kao 2017; Kawaguchi 2010; Kindler 1996; Lee 1994; Lee 2015; Lee 2017; Liu 1986; Louizos 2007; Magnusson 1986; Mallon 1990; Menigaux 2002; Mikawa 1991; Miller 1990; Miller 1991; Moon 2011; Neary 2006; Ohri 1999; Ojima 2017; Oxorn 1990; Park 2009; POBBLE 2005; POISE 2008; PRESAGE 2016; Safwat 1984; Shailaja 2013; Sharma 1996; Sharma 2018; Shrestha 2011; Shukla 2010; Singh 1995; Singh 2012; Stone 1988; Sugiura 2007; Tendulkar 2017; Unal 2008; Verma 2018; Wajima 2011; Wallace 1998; Whitehead 1980; Yamazaki 2005; Yang 2006; Yoshida 2017). Of these, two studies reported data unclearly (Alkaya 2014; Oxorn 1990), and 11 studies did not clearly report whether hypotension was measured in all groups (Chung 1992; Do 2012; Gibson 1988; Helfman 1991; Jangra 2016; Kindler 1996; Neary 2006; Sharma 1996; Verma 2018; Wajima 2011; Yoshida 2017); we did not include these 13 studies in analysis. We note that 22 studies reported no events in either group (Goyagi 2005; Gupta 2011; El‐Shmaa 2016; Harasawa 2006; Horikoshi 2017; Inada 1989; Inoue 2010; Kao 2017; Lee 1994; Lee 2015; Lee 2017; Louizos 2007; Menigaux 2002; Mikawa 1991; Moon 2011; Ojima 2017; Safwat 1984; Sharma 2018; Shrestha 2011; Singh 2012; Tendulkar 2017; Yamazaki 2005). Perioperative beta‐blockers probably increase incidences of hypotension (RR 1.40, 95% CI 1.29 to 1.51; I2 = 0%; 49 studies, 12,304 participants; moderate‐certainty evidence; Analysis 1.9); the NNTH was 23 (95% CI 18 to 31). See Figure 6.
We used GRADE to downgrade the certainty of the evidence for hypotension by one level for study limitations because we assessed several studies to be at high or unclear risk of bias. See summary of findings Table for the main comparison. Six studies reported congestive heart failure (Bayliff 1999; Horikoshi 2017; Magnusson 1986; POISE 2008; Wallace 1998; Yang 2006). One study reported no events in either group (Horikoshi 2017). We found little or no difference in the number of people who had congestive heart failure according to whether perioperative beta‐blockers were administered (RR 1.18, 95% CI 0.94 to 1.48; I2 = 0%; 6 studies, 9212 participants; low‐certainty evidence; Analysis 1.10). We used GRADE to downgrade the certainty of the evidence by one level owing to study limitations because we assessed some studies to be at high risk of bias in at least one domain, and one level for imprecision because studies were few, despite a large sample size owing to one large international study (86.9% weighting). Seven studies reported length of hospital stay (Bayliff 1999; Horikoshi 2017; Lee 2010; POBBLE 2005; POISE 2008; PRESAGE 2016; White 2003). We did not include three studies in analysis because they did not report data using equivalent values (Bayliff 1999; POISE 2008; White 2003); study authors reported little or no difference between groups. In the remaining studies, we found little or no difference in hospital length of stay according to whether perioperative beta‐blockers were administered (mean difference (MD) −1.21 days, 95% CI −2.75 to 0.33; I2 = 66%; 404 participants; very low‐certainty evidence; Analysis 1.11). We used GRADE to downgrade the certainty of the evidence by three levels: one level for study limitation because we assessed some studies to be at high risk of bias in at least one domain; one level for imprecision because the evidence was from few studies with few participants, and one level for inconsistency because we noted a moderate level of statistical heterogeneity. No studies reported outcome data for quality of life. We did not complete subgroup analysis for cerebrovascular events, ventricular arrhythmias, and atrial fibrillation or flutter because we found fewer than 10 studies for these outcomes. Two studies were multi‐arm studies with more than one type of beta‐blocker (Inoue 2010; Stone 1988); in subgroup analysis, we divided the control group data evenly in order to avoid a unit of analysis error.
One study included two intervention groups in which beta‐blockers were given pre‐ and postoperatively, and intraoperatively (Zaugg 1999); for subgroup analysis, we included only outcome data for the group in which beta‐blockers were given intraoperatively.
We found 83 studies in which beta‐blockers were given during the perioperative period to adults undergoing non‐cardiac surgery. We also identified 18 studies awaiting classification (three were reported only as abstracts; we were unable to access the full text of seven; four are completed trials without full‐text reports; and four require translation), and three ongoing studies. The evidence for early all‐cause mortality was uncertain; based on the risk of death in the control group of 25 per 1000, the effect of beta‐blockers was between two fewer and 13 more per 1000. We found no evidence of a difference in cerebrovascular events (low‐certainty evidence), and ventricular arrhythmias (very low‐certainty evidence). We found low‐certainty evidence that beta‐blockers may reduce the incidence of myocardial infarction and atrial fibrillation or flutter. However, we also found low‐certainty evidence that beta‐blockers may increase bradycardia, and moderate‐certainty evidence that beta‐blockers probably increase hypotension. We found little or no difference in long‐term mortality, death due to cardiac causes, and in congestive heart failure, depending on whether beta‐blockers were given; we assessed the certainty of the evidence for these outcomes to be low. We found no evidence of a difference in length of hospital stay, but we did not explore reasons for moderate statistical heterogeneity in this effect, and we assessed the certainty of the evidence to be very low. No studies assessed quality of life. We identified 83 studies with 14,967 participants, of which more than half were identified in the most recent search. All studies included adult participants undergoing non‐cardiac surgery. Only one study assessed beta‐blockers for emergency surgery (Neary 2006); the remaining surgeries were elective. We anticipated that studies for this review would include a wide range of types of surgery, with different degrees of risk. The evidence included studies of surgery that were: thoracic; vascular; neuro; burns; abdominal; oral and maxillofacial; ear, nose and throat; gynaecological; orthopaedic; mixed surgeries; and some elective surgeries that were not specified by study authors. We categorized surgical risk according to Kristensen 2014, and found 26.5% studies included high‐risk surgeries and 13.3% included low‐risk surgeries. Although we categorized most studies as medium risk, we included in this category studies in which the type of surgery was not specified. We used subgroup analysis to explore whether beta‐blockers had a different effect according to the surgical risk. Whilst we found no evidence of a difference between subgroups overall, we noted the effect of POISE 2008 in the data for all‐cause mortality. For this subgroup analysis, we noted that studies of low‐ or medium‐risk surgeries showed an increase in mortality with beta‐blockers, but there was no evidence of a difference in effect in high‐risk surgeries; POISE 2008 recruited participants undergoing various surgeries to include vascular, intraperitoneal, and orthopaedic surgery and we categorized this as medium risk. We included in the review several studies published prior to 2000, and these studies may have used clinical management strategies that differ from current practice. We explored this in sensitivity analysis, and found no changes to the interpretation of the effect for the main review outcomes. We used GRADE to downgrade the certainty of the evidence. In our risk of bias assessments, we judged 26.5% of studies to be at high risk of bias in at least one domain, and we downgraded evidence for all outcomes owing to study limitations. In sensitivity analyses, we explored the effect of including only studies with more robust study designs. We excluded open‐label studies in which comparisons had been made with standard care, and studies with risks of selection bias or attrition bias. We noted some differences in effect during some sensitivity analyses and we, therefore, also considered these sensitivity analyses, downgrading the certainty of the evidence owing to study limitations for all‐cause mortality, cerebrovascular events, and atrial fibrillation or flutter. We noted that one large, international, and methodologically robust study did not report results that were consistent with smaller studies (POISE 2008). We downgraded the evidence for all‐cause mortality and cerebrovascular events because POISE 2008 showed an increase in events when beta‐blockers were used, whilst the remaining smaller studies showed evidence of neither an increase or decrease in events. Similarly, we downgraded the evidence for myocardial infarction because POISE 2008 showed a reduction in events when beta‐blockers were used, whilst the remaining smaller studies showed evidence of neither an increase or decrease in events. Evidence for ventricular arrhythmias was from few studies with few participants, and we downgraded this outcome owing to imprecision. Overall, we graded the evidence for perioperative beta‐blockers as low certainty, very low certainty, and moderate certainty, which means there is the possibility that the true estimate may be substantially different Our previous review assessed beta‐blockers in both cardiac and non‐cardiac surgery (Blessberger 2018); we split this review into two reviews according to type of surgery. This version incorporates data from studies of non‐cardiac surgery and includes an updated and refined search strategy. We conducted a thorough search in the update and used two review authors to assess study eligibility, extract data, and assess risk of bias in included studies; therefore, we reduced potential bias in the review process. During the updating process, we made changes to the review to meet current Cochrane standards. This included minor clarifications to the inclusion criteria of the review, and changes to wording of the sections of the methods within Data collection and analysis. In particular, this version of the review includes fewer outcomes. Data for myocardial ischaemia, supraventricular arrhythmias (except atrial fibrillation), bronchospasm and cost of care are reported only in the previous version (Blessberger 2018). In reducing the number of outcomes, our intention was to improve the usability of the review; therefore, we selected outcomes that we considered to be most important to users of the review. We used the work of Myles and colleagues to support this decision making process (Myles 2016), and sought advice from a Cochrane Editor before agreeing to a reduction in outcomes. In addition, we did not include meta‐regression as a method to explore heterogeneity in this review. We explored heterogeneity only through subgroup analysis, which was pre‐specified in the review protocol. We conducted analyses using a random‐effects model, rather than choosing the effects‐model based on statistical heterogeneity (Borenstein 2010); we evaluated in sensitivity analysis the effect of using Peto odds ratio for rare events, which uses a fixed‐effect model. Again, we made these decisions following advice from the Cochrane Editorial team. We included only trials that investigated at least one outcome specified in the Methods section. Because of the vast scope of the literature search and the large number of search results and included trials, this approach was justified in order to improve the management of the review. Whilst this may have introduced a source of bias, we judged the impact of potentially missing studies as small because the available set of data was derived from a large number of trials. As explained in previous versions of the review, we included studies in which some participants may not have received general anaesthesia, which differed from our original protocol. Whilst we attempted to limit the number of participants not undergoing general anaesthesia, we could not be certain of the effect of including a mixed population. Furthermore, we could not contact all study authors to gather further information about trial design or data analysis (e.g. performance of an ITT analysis) because of the large number of studies. However, it is likely that study authors, if contacted, may over‐report trial design quality criteria. In the previous version of the review we found no clear evidence of a difference in all‐cause mortality (Blessberger 2018). When analysis was restricted to placebo‐controlled studies, we continued to find an increase in all‐cause mortality with the use of beta‐blockers. However, our subgroup analysis continued to be led by one large multi‐centre, and well‐conducted, study in which metoprolol given to participants undergoing medium risk surgeries showed an increase in mortality with beta‐blocker use (POISE 2008). The effect of POISE 2008 was also evident in the analyses for myocardial infarction (POISE 2008 showed a reduction in myocardial infarction) and cerebrovascular events (POISE 2008 showed an increase in stroke), and these results were also inconsistent with the remaining smaller studies. The review by Bouri and colleagues endeavoured to re‐evaluate the evidence for perioperative beta‐blockers without studies that were at risk of including fictitious data. Bouri 2014 assessed only preoperative beta‐blocker use, therefore including POISE 2008. Their analyses showed an increase in mortality and stroke with beta‐blocker use, alongside a reduction in myocardial infarction. It should be noted that our primary analyses include evidence from a variety of different beta‐blockers, and doses of beta‐blockers, which were started before, during or after surgery of different risk statuses. We cannot ignore the results from POISE 2008, which indicate an increased risk of mortality and stroke; however, results of POISE 2008 may be influenced specifically by the use of a relatively high dose of metoprolol administered if the pre‐operative heart rate was above 50 beats per minute (not titrated to a specific heart rate), and was given just before surgery (two to four hours). We have allowed for this inconsistency when downgrading the certainty of the evidence in this review. Our findings for all‐cause mortality and myocardial infarction were consistent with an early systematic review, which included POISE 2008 (Bangalore 2008); although we noted that Bangalore 2008 found an increase in cerebrovascular events with beta‐blockers. Hajibandeh 2017 found little or no difference in effect for all‐cause mortality and myocardial infarction. The systematic review included both observational studies and RCTs and was limited to vascular and endovascular surgery; its findings were consistent with our subgroup analyses results for high‐risk surgery.
Study flow diagram for updated search on 28 June 2019 'Risk of bias' graph: review authors' judgements about each 'Risk of bias' item presented as percentages across all included studies 'Risk of bias' summary: review authors' judgements about each 'Risk of bias' item for each included study Forest plot of comparison 1. Beta‐blockers vs control, outcome: 1.1 Early all‐cause mortality Funnel plot of comparison 1. Beta‐blockers vs control, outcome: 1.1 Early all‐cause mortality Forest plot of comparison: 1 Beta‐blockers vs control, outcome: 1.9 Hypotension Comparison 1 Beta‐blockers vs control, Outcome 1 Early all‐cause mortality. Comparison 1 Beta‐blockers vs control, Outcome 2 Long‐term mortality. Comparison 1 Beta‐blockers vs control, Outcome 3 Death due to cardiac causes. Comparison 1 Beta‐blockers vs control, Outcome 4 Acute myocardial infarction. Comparison 1 Beta‐blockers vs control, Outcome 5 Cerebrovascular events. Comparison 1 Beta‐blockers vs control, Outcome 6 Ventricular arrhythmias. Comparison 1 Beta‐blockers vs control, Outcome 7 Atrial fibrillation and flutter. Comparison 1 Beta‐blockers vs control, Outcome 8 Bradycardia. Comparison 1 Beta‐blockers vs control, Outcome 9 Hypotension. Comparison 1 Beta‐blockers vs control, Outcome 10 Congestive heart failure. Comparison 1 Beta‐blockers vs control, Outcome 11 Length of stay (in days). Comparison 2 Beta‐blockers vs control: subgroup by type of beta‐blocker, Outcome 1 Early all‐cause mortality. Comparison 2 Beta‐blockers vs control: subgroup by type of beta‐blocker, Outcome 2 Acute myocardial infarction. Comparison 2 Beta‐blockers vs control: subgroup by type of beta‐blocker, Outcome 3 Bradycardia. Comparison 2 Beta‐blockers vs control: subgroup by type of beta‐blocker, Outcome 4 Hypotension. Comparison 3 Beta‐blockers vs control: subgroup by start of beta‐blocker therapy, Outcome 1 Early all‐cause mortality. Comparison 3 Beta‐blockers vs control: subgroup by start of beta‐blocker therapy, Outcome 2 Acute myocardial infarction. Comparison 3 Beta‐blockers vs control: subgroup by start of beta‐blocker therapy, Outcome 3 Bradycardia. Comparison 3 Beta‐blockers vs control: subgroup by start of beta‐blocker therapy, Outcome 4 Hypotension. Comparison 4 Beta‐blockers vs control: subgroup by risk status, Outcome 1 Early all‐cause mortality. Comparison 4 Beta‐blockers vs control: subgroup by risk status, Outcome 2 Acute myocardial infarction. Comparison 4 Beta‐blockers vs control: subgroup by risk status, Outcome 3 Bradycardia. Comparison 4 Beta‐blockers vs control: subgroup by risk status, Outcome 4 Hypotension.
Hide thumbnails zoom in zoom out |