Beta-Blockers in Congestive Heart Failure: What Can We Learn from a Bayesian Meta-Analysis?

Original Citation

Brophy JM. Joseph L, Rouleau JL. Beta-Blockers in congestive heart failure: a Bayesian meta-analysis.

Overall Study Question

Congestive heart failure (CHF) is a common syndrome characterized by exceedingly high rates of morbidity and mortality.  In Canada, there are reportedly more than 350,000 persons afflicted with the condition, and the mortality rate one year after diagnosis ranges between 25-40%. In addition, CHF is the leading single indication for admission to the hospital among elderly Canadians. Accumulated evidence from published randomized control trials demonstrates that effective treatment strategies (angiotensin-converting enzyme inhibitors (ACEI), digoxin and spironolactone) are available to improve both quantity and quality of life of patients with CHF.  Definitive evidence for the use of beta-blockers in CHF has also emerged in the last few years.  This evidence is in direct contrast to conventional medical teaching, which viewed CHF as a contraindication to beta-blocker therapy due to the propensity for this class of drugs to exert negative inotropic effects in the short-term.  Consequently, there exists a care gap between the research evidence and clinical practice for the use of beta-blockers in patients with CHF.

The present investigation was a meta-analysis of published randomized control trials that involved a comparison of beta-blockers with placebo in patients with CHF, quantifying the effect of treatment on morbidity and mortality. While no less than four meta-analyses of randomized control trials with beta-blockers in CHF have previously been published, the present study is unique for two reasons.  Firstly, the previous meta-analyses were published a few years ago and since that time, additional large trials with beta-blockers in CHF have been published.  Secondly, unlike previous meta-analyses that used a “fixed-effects model”, the present study employed a “Bayesian hierarchical random-effect model”.  This novel method of meta-analysis has several advantages, including the ability to account for between-study variation.

Are the Results of the Study Valid?

Did the overview address a focused clinical question?

Yes.  The investigators performed a meta-analysis of the cumulative experience of beta-blockers compared to placebo in patients with heart failure from randomized control trials published prior to July 2000.

Were the criteria used to select articles for inclusion appropriate?

Yes.  The investigators defined the criteria for study selection a prior and included all randomized clinical trials of beta-blockers versus placebo in chronic stable CHF that evaluated mortality as an outcome.

Trials were excluded if they involved:

1) cross-over designs;

2) beta-blockers with intrinsic sympathomimetic activity (ISA);

3) a follow-up period of <3 months, or;

4) patients admitted for acute myocardial infarction.

Twenty-two randomized controlled trials published between 1985 and 2000 fulfilled the criteria for inclusion in this meta-analysis.

Is it unlikely that important, relevant studies were missed?
Yes.  The investigators searched the MEDLINE database from 1966 to July 2000 using the keywords adrenergic beta-antagonists, congestive heart failure, and trial to identify relevant studies. This initial approach identified 105 trials, of which 17 met the inclusion criteria.  Subsequently, they examined the references cited in the four meta-analyses previously published in 1987 and 1988 to find an additional 5 trials that were appropriate for inclusion. The final step undertaken to identify pertinent trials was a search of the Cochrane and Web of Science databases.  This process did not reveal any additional trials.
Was the validity of the included studies appraised?
Yes.  Each of the 22 randomized controlled trials included in this meta-analysis followed a double-blind protocol, with only one study reported as having possible irregularities in the randomization process.  For all of the studies, the follow-up of randomly assigned patients was almost complete. The investigators commented that the only minor methodological flaw in the studies was a lack of description of the randomization process in 18 of the 22 trials included.
Were assessments of studies reproducible?
Unable to determine.  The investigators did not explicitly state how the studies were scored for inclusion in the meta-analysis, nor did they comment as to whether or not agreement between two or more of the investigators was required for a study to be included.

Were the results similar from study to study?

Unable to determine.  There was no formal test of heterogeneity performed in this meta-analysis because data analysis was conducted using a Bayesian hierarchical random-effect model. The inclusion of both the between-study variability and the usual random variability in this model reportedly allows for the simpler translation of the effect of the treatment to real clinical practice.

What were the Results?

What are the overall results of the review?

There were 110 studies identified, however only 22 studies (involving 10,135 patients) met the inclusion criteria for meta-analysis.  The observed mortality rate among placebo treated patients was 12.8% (624/4862) and 8.4% (444/5273) for patients treated with a beta-blocker.  The combined odds ratio (OR) for total mortality among all patients in the studies was 0.65 (95% credible interval, 0.53 to 0.80).  Utilizing the hierarchically determined mortality rate of 12% (credible interval, 4% to 26%) in the first year of follow-up among placebo treated patients from the 3 largest and most recent trials, the best estimate of the absolute mortality reduction during the first year of treatment was 3.8 lives saved per 100 patients treated (credible interval, 2.1 to 5.3).  From the Bayesian analysis derived probability density curve, the probability that beta-blocker therapy saves at least 2 lives per 100 patients in the first year of treatment is 99% and the probability that 3 or more lives are saved per 100 patients over this timer period is 85%.

While various beta-blockers were evaluated in the individual randomized control trials (8 trials of metoprolol, 8 of carvedilol, 3 of bucindolol, 2 of bisoprolol, and 1 of nebivolol), the mortality benefit observed with selective agents (metoprolol, bisoprolol) and non-selective agents (carvedilol, bucindolol) was similar (OR 0.67 (credible interval, 0.57 to 0.79) and OR 0.52 (credible interval, 0.28 to 0.89), respectively).

Beta-blocker therapy was also associated with a reduction in morbidity in these trials.  Seven hundred and fifty-four of 4862 (15.5%) placebo recipients were hospitalized for heart failure as compared with 540 of 5273 (10.2%) beta-blocker recipients.  The combined odds ratio for hospitalization for CHF was 0.64 (credible interval, 0.53 to 0.79).  Based on trials published since 1996, the weighted average of admissions for CHF among placebo recipients in the first year of follow-up was 14%.  This translated into a best estimate of 4.0 fewer hospitalizations per 100 patients treated with a beta-blocker (credible interval, 2.4 to 5.6).  From the Bayesian analysis derived probability density curve, the probability that beta-blocker therapy prevents at least 2 hospital admissions for CHF per 100 patients 99% and the probability that 3 or more hospital admissions for CHF are prevented per 100 patients is 90%.

How precise were the results?

The results of this meta-analysis are statistically significant in favour of beta-blocker therapy as compared to placebo for all outcomes analyzed.  The Bayesian hierarchical random-effect model utilized for this meta-analysis generated “credible intervals”, the Bayesian analogue to 95% confidence intervals, as a measure of precision.  For all of the outcomes evaluated, the “credible intervals” were provided in the text.  The combined odds ratios for total mortality and hospital admissions for heart failure were statistically significant since the credible intervals did not cross 1.0 (the point of unity).  In addition, since the credible intervals around the point estimate were narrow and the lower boundaries of these intervals represent minimal clinical important differences, these results can be considered clinically significant.

How much does allowance for uncertainty change the results?

In the sensitivity analysis, the investigators considered an extreme situation in which further trials involving 2,000 patients with mortality rates of 20% in the beta-blocker treated patients (200/1,000) and 10% in the placebo treated patients (100/1,000) were included in the meta-analysis.  Inclusion of this hypothetical negative data in the meta-analysis revealed an OR of 0.84 (credible interval, 0.74 to 0.94).  This result would still support the finding that beta-blockers reduce the odds ratio of mortality from CHF by 16%.  This sensitivity analysis demonstrates the robustness of the findings from this study and the probability that they will hold up to possibly unfavourable future or unpublished results.  The investigators also comment that the Bayesian method of meta-analysis can be easily repeated and updated to allow for inclusion of new trial data. They used the preliminary results of the Beta-blockers Evaluation Survival Trial (BEST) presented at the 72nd Scientific Sessions of the American Heart Association, as an example.  In this randomized control trial of 2,708 patients, the mortality rate after 2 years of follow-up was 33% in the placebo group and 30.2% in the bucindolol treated group.  Incorporation of these results into the meta-analysis revealed an accumulated OR of 0.72 (credible interval, 0.61 to 0.84).  Finally, an additional analysis performed for the mortality benefit observed with beta-blockers categorized according to their adrenergic receptor selectivity revealed consistent results.  The OR was 0.67 for selective agents (credible interval, 0.57 to 0.79) and the OR was 0.52 for non-selective agents (credible interval, 0.28 to 0.89).

Will the Results Help Me in Caring for My Patients?

Can the results be applied to my patient care?
CHF is a common syndrome characterized by exceedingly high rates of morbidity and mortality despite the utilization of effective pharmacotherapeutic strategies, such as ACE-I’s, digoxin and spironolactone.  This meta-analysis demonstrates that beta-blockers have a large beneficial effect on mortality (3.8 lives per 100 patients treated) and morbidity (4.0 fewer hospitalizations per 100 patients treated) in stable NYHA class II or III CHF.  The benefits of beta-blockers in CHF go beyond those provided by ACEI since the majority of patients were receiving an ACEI at the time of randomization.  In addition, most patients were also receiving diuretics and digoxin at the time they were randomized to receive treatment with a beta-blocker.

Were all clinically important outcomes considered?

Mortality is a hard endpoint and the ultimate goal of CHF pharmacotherapy is to reduce overall mortality.  Additionally, hospitalization for CHF is a clinical relevant morbidity endpoint that is targeted for reduction with current pharmacotherapy.
Are the likely treatment benefits worth the potential harms and costs?
The reductions in mortality (3.8 lives per 100 patients treated) and hospitalizations for heart failure (4.0 fewer hospitalizations per 100 patients treated) observed in this meta-analysis are clinically important.  The initiation of beta-blocker therapy is not without toxicity.  The potential for short-term deterioration in clinical status (e.g. dizziness, hypotension and worsening CHF) of patients during the initiation and titration of the beta-blocker is a frequent problem. This is confirmed by the 5% withdrawal rate observed in the run-in periods of the clinical trials.  Nevertheless, if the patients are informed of this potential and physicians carefully supervise the initiation of beta-blocker therapy in their patients, the short-term “pain” appears to result in significant long-term gain.  While the cost-effectiveness of beta-blocker therapy in the treatment of CHF was not assessed in this study (and remains to analyzed formally), the investigators do comment that the large treatment benefits and low cost of therapy indicate that it will be cost attractive.


This meta-analysis provides further support for the recommendations made in recent consensus conferences that strongly suggest the use of beta-blockers in all patients with NYHA class II-III heart failure.  This study also builds on our previous knowledge of the effect of beta-blockers in CHF from the 4 previously published meta-analyses, with the inclusion of the results from more recent large trials.  Of interest, before 1999 only 3,071 patients were randomized in the trials of beta-blockers in CHF and this number had increased to 10,135 by July 2000.  The results of the present study are reported as odds ratios (OR), which are the traditional means for reporting treatment effects in a meta-analysis.  One limitation of reporting OR is the fact that they fail to account for absolute differences.  Consequently, OR may obscure the true clinical significance of an intervention. In this study, the investigators used the Bayesian hierarchical random-effect model specifically to avoid this problem.  With Bayesian analysis, probability density distributions of the differences in survival and hospitalization at 1 year between patients receiving beta-blockers and placebo were created. Subsequently, direct probability estimates to pertinent questions were calculated by measuring the area under the curve (AUC).  Most clinicians will not necessarily be familiar with probability density distributions; however, be assured that they are “clinically friendly” since they supply direct probability estimates to multiple questions.  This method permits the calculation of probabilities that relate to any interval null or alternative hypotheses, as well as any range of clinically meaningful effects.  Consequently, this Bayesian method provides for clearer clinical interpretations of the accumulated data.

The present study demonstrates that beta-blocker therapy is associated with significant and clinically meaningful reductions in mortality (3.8 lives per 100 patients treated) and hospitalizations for heart failure (4.0 fewer hospitalizations per 100 patients treated).  These results are limited to clinically stable patients with NYHA class II or III CHF treated with either selective or non-selective beta-blockers. The role of beta-blocker therapy in patients with NYHA class IV CHF remains uncertain and cannot be extrapolated from these results since this population accounts for <5% of the patients studied to date. While large mega-trials will be required to detect whether any clinically significant difference exists between beta-blockers in CHF, this meta-analysis should provide clinicians with the impetus to use beta-blockers for the treatment of all eligible patients with CHF.  Eligible patients receiving standard triple drug therapy for CHF (diuretics, digoxin, and ACE-I) should be initiated with low doses (e.g., 6.25-12.5 mg of metoprolol, 3.125 mg of carvediolol, or 2.5 mg of bisoprolol) and titrated slowly under the close monitoring of a health professional. Judicious monitoring of patients for clinical deterioration and appropriate management of such symptoms (dizziness, hypotension and worsening CHF) will ensure that they adhere to therapy and derive the significant clinical benefits that are possible with beta-blocker therapy.

Latest Posts