Meta-analysis: -Blocker Dose, Heart Rate Reduction, and Death in Patients With Heart Failure

Over the past decade, several randomized clinical trials have established that -blockers are beneficial in patients with heart failure (1, 2). Although resting heart rate is known to be a prognostic factor in patients with heart failure (3, 4), it remains unclear whether the benefits of -blockade in patients with heart failure are related to the degree of reduction in heart rate or the dosage of -blocker administered. This question is important because the adverse effects of -blockers are dose-related (5). Although heart failure guidelines recommend up-titration of -blockers to the target doses used in -blocker trials, outcome studies reveal that only some patients achieve these doses outside of specialized heart failure clinics (68). Secondary analyses of the CIBIS (Cardiac Insufficiency Bisoprolol Study) (9), COMET (Carvedilol Or Metoprolol European Trial) (10), and CIBIS II (11) data suggested that the magnitude of heart rate reduction with -blockers was an important mediator of -blocker effect. In addition, a small clinical trial of 49 pacemaker-dependent patients with left ventricular systolic dysfunction demonstrated that patients who were paced at a lower rate (60 beats/min) had improvements in left ventricular function and left ventricular dimensions compared with those paced at a higher rate (80 beats/min) (12). However, other studies have not confirmed a relationship between the magnitude of heart rate reduction and the efficacy of -blockers (1315). We designed this meta-analysis to investigate the between-study heterogeneity in heart failure -blocker trials. Specifically, we examined whether -blocker dose or magnitude of heart rate reduction could account for the differences in treatment effects among heart failure -blocker trials. Methods Identifying Relevant Studies We searched for randomized trials in MEDLINE (1966 to 2008), EMBASE (1980 to 2008), CINAHL (1982 to 2008), SIGLE (1980 to 2008), Web of Science, and the Cochrane Central Register of Controlled Trials. We did not apply language restrictions, but we restricted our searches to human studies and clinical trial or randomized, controlled trial publications. We used the keywords and Medical Subject Headings adrenergic -antagonists, heart failure, and congestive (exp). We also hand-searched bibliographies of identified studies, recent meta-analyses of -blockers in heart failure (1, 2), and heart failure guidelines. Study Selection and Data Abstraction Two authors independently reviewed the results of the search strategy and selected all studies that reported the effect of -blockers on all-cause mortality in patients with heart failure. The authors excluded studies if they were published in abstract form only, did not report death, used -blockers for 1 month or less, or enrolled fewer than 50 patients. Two authors extracted all outcome data independently, with subsequent discussion of any discrepancies. Outcomes from each study were extracted in intention-to-treat categories rather than per-protocol categories (that is, all outcomes were analyzed by randomization group to avoid bias from excluding patients who dropped out, were withdrawn, or did not adhere to treatment). We calculated the magnitude of heart rate reduction in each trial by comparing the heart rate at the end of the dose titration phase of each trial with the baseline values and subtracted the change in the placebo group from the change in the -blocker group. Statistical Analysis Because of the expected differences in patient samples and length of follow-up in these studies, we did our primary analyses by using the DerSimonianLaird random-effects model. We did analyses by using Review Manager, version 4.2 (The Cochrane Collaboration, Copenhagen, Denmark), and Stata SE, version 10 (StataCorp, College Station, Texas). Because the outcome of interest was relatively common, we calculated risk ratios (RRs) and 95% CIs. We assessed and quantified statistical heterogeneity for each outcome of interest by using the Cochran Q test and the I 2 statistic, respectively (16). The I 2 statistic quantifies the percentage of statistical heterogeneity due to between-study variability. By convention, values less than 25%, 25% to 50%, and greater than 50% are considered low, moderate, and high amounts of heterogeneity, respectively (17). To explore potential explanations for the between-trial heterogeneity, we did meta-regression analyses by using the weighted least-squares method (16). The logarithm of relative risk for death, weighted by the inverse variance of each study, was regressed against the following variables 1 at a time: sex, age, ischemic cause, left ventricular ejection fraction (LVEF), New York Heart Association (NYHA) class, atrial fibrillation, use of digoxin, heart rate at baseline, magnitude of heart rate reduction achieved with treatment, dose of -blocker reached, systolic blood pressure (SBP) at baseline, magnitude of treatment-related SBP reduction, and specific -blocking agent. We explored continuous variables in these meta-regressions both linearly and categorically by using tertiles. We reported the P values from Wald tests. In a sensitivity analysis, we investigated the stability of our meta-regression result by using a Monte Carlo simulation to explore the effect of sampling variability around the point estimates of heart rate reduction in each trial (18). We sampled 5000 data sets, and the mean heart rate reduction was varied along a normal distribution (by using the mean and SE reported by each trial); trial selection remained fixed in each Monte Carlo iteration. In addition, we ran meta-regressions incorporating various combinations of 2 or 3 of these variables to investigate the robustness of our findings (the number of trials was insufficient to run meta-regressions with more than 3 variables). Role of the Funding Source There was no specific funding for this project. Results Study Selection and Evaluation Of the 548 citations that we identified in our search, 108 were potentially eligible for inclusion, but we excluded 85 after detailed review (Figure 1). Of note, the 34 trials that we excluded because they did not report death but instead evaluated levels of biomarkers or neurohormonal measurements, hemodynamic changes, or echocardiographic measurements, were generally small (mean sample size, 43 patients). The 12 trials that reported death, but were excluded because they included fewer than 50 patients (mean sample size, 28 patients), had a total of only 15 deaths (compared with 2720 deaths in the 23 randomized trials included in our meta-analysis). Disagreement between 2 reviewers about eligibility of the studies occurred on 3 occasions, for a value of 0.92. All disagreements were resolved by consensus. Figure 1. Study flow diagram. Studies Included in the Systematic Review Table 1 shows the baseline data from 23 randomized trials (1942). Three trials [23, 28, 38] reported outcome data in -blocker dosage subgroups (each of these subgroups is reported as a separate row in Table 1) (1942). The Appendix Table outlines -blocker titration schedules, dosing, duration, and effect on SBP and heart rate. Four trials (21, 24, 26, 40) could not be included in the analyses comparing death with physiologic variables because they did not report heart rate data for trial participants. Table 1. Baseline Data for Included Trials Appendix Table. Changes in Clinical Variables During Trials Qualitative Synthesis All but 2 (32, 41) trials were restricted to patients with systolic dysfunction, and only 4% of trial participants had preserved systolic function. Two trials enrolled only patients with nonischemic heart failure; 2 trials were restricted to patients with ischemic cardiomyopathy; and in the other trials, the frequency of ischemic heart disease ranged from 27% to 90%, with a median of 59% (Table 1). In addition to standard antiheart failure therapy except -blockers, the control groups received placebo in all but 2 trials (in which the control group received an angiotensin-converting enzyme inhibitor but no -blocker) (33, 36). Use of angiotensin-converting enzyme inhibitors (median, 93% [interquartile range {IQR}, 87% to 96%]) and digoxin (median, 75% [IQR, 57% to 91%]) was high in these trials (Table 1). Mean LVEF in these trials ranged from 0.17 to 0.36 (median, 0.24), with all but 1 trial reporting mean LVEF less than 0.30 (Table 1). Few trials reported comorbid conditions, but in those that did, 12% to 35% of participants had atrial fibrillation (Table 1) and 12% to 36% had diabetes mellitus. Most patients in these trials had NYHA class III or IV symptoms at baseline (median, 54% [IQR, 50% to 66%]). Most of these trials were of relatively short durationonly 6 trials (19, 3335, 39, 40) followed patients longer than 12 months (Appendix Table). Fifteen trials did not report subgroup analyses, whereas 8 trials did: patients with ischemic versus nonischemic heart failure (22, 31, 34, 35, 39, 41, 43); results by NYHA class (22, 3436, 39), age (22, 31, 36, 41, 43), sex (22, 31, 39, 41, 43), or race (39); and results for patients with comorbid conditions, such as diabetes (22, 36, 41), hypertension (22, 36), smoking (22), or chronic kidney disease (36). Although 6 trials reported that -blocker efficacy did not statistically differ between any of the subgroups examined, most of these subgroup analyses were presented as forest plots or KaplanMeier curves with no explicit reporting of raw numbers, such that we could not pool subgroup data to examine the consistency across trials with formal interaction tests. One trial (BEST [Beta-Blocker Evaluation of Survival Trial]) reported that -blockers demonstrated a survival benefit in nonblack patients but not in black patients (P for interaction= 0.02) (39). Because no other trials reported outcomes separately for black and nonblack patients, we could not evaluate the consistency of this subgroup finding across trials. The BEST trial enrol