Identification of predictive factors interacting with heart rate reduction for potential beneficial clinical outcomes in chronic heart failure: A systematic literature review and meta-analysis

Highlights • We used Bayesian statistics to illustrate factors that affect HR reduction therapy.• HR-reducing therapy was associated with significant reductions in risk.• Presence of comorbid T2DM significantly affected mortality risk.


Introduction
Chronic heart failure (HF) is a life-threatening clinical syndrome characterized not only by cardiac dysfunction but also by various other systemic disorders and comorbidities [1][2][3]. In 2021, the burden of HF across 195 countries was reported with data showing that the global number of cases increased by 91.9 % from 33.5 million in 1990 to 64.3 million in 2017 [4]. However, the international prospective REPORT-HF registry suggests there are regional differences in treatment and medical management [5]. Hospitalization for and management of acute HF must take into consideration the etiology and precipitants of HF, various relevant patient-and disease-related characteristics, and the presence of Abbreviations: AF, atrial fibrillation; CrI, credible interval; CV, cardiovascular; HF, heart failure; HR, heart rate; HFrEF, HF with reduced ejection fraction; LVEF, left ventricular ejection fraction; MI, myocardial infarction; NYHA, New York Heart Association; PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses; RR, risk ratio; T2DM, type 2 diabetes mellitus; WHF, worsening heart failure.
comorbidities. Although the objective of therapy is primarily to treat the symptoms of HF that are affecting a patient's functional capacity and quality of life, therapy should also improve a patient's clinical status by suppressing pathological cardiac remodeling (i.e., neurohormonal activation) to improve prognosis [6].
HF leads to increased cardiac contractility and heart rate (HR) by decreasing the carotid baroreceptor response and subsequently increasing sympathetic nervous activity [7]. Many studies have shown an association between HR and clinical outcomes in patients with HF. In patients with chronic HF, an increase in HR of 10 beats per minute (bpm) was associated with an 8 % increase in the risk of cardiovascular death or HF hospitalization regardless of their ejection fraction [8,9]. In addition, a previous meta-analysis reported a significant association between the magnitude of HR reduction (a decrease in HR of 5 bpm) and survival benefit of beta-blockers (18 % reduction in risk of death) in patients with HF [10]. Another more recent meta-analysis study reported a reduction in mortality in patients that have HF with reduced ejection fraction (HFrEF) in sinus rhythm who were treated with beta-blockers, regardless of their pre-treatment HR, and that achieving a lower HR was associated with a better prognosis [11]. Therefore, given HR is a factor that affects prognosis, it is more clinically meaningful to identify which patient-and disease-related factors interact with HR-reducing treatment to modify the clinical outcomes in patients with HFrEF who are receiving standard HF therapy. In the SHIFT trial it was suggested that baseline factors (i.e., baseline HR, background therapy, cardiac parameters, and medical history) may interact with HR-reducing treatment [12]. However, the relationship between a patient's background characteristics and the effect of HR reduction on prognosis has yet to be sufficiently evaluated. It was hypothesized that several patient and disease-related factors including demographic characteristics, comorbidities, and biomarker levels may interact with therapy-induced HR reduction to predict clinical outcome in patients with chronic HFrEF The overall objective of this systematic literature review and metaanalysis is to investigate the interaction between predictive factors (e.g., age, sex, comorbidities, causes and complications of HF, concomitant treatment, and other baseline factors) with HR-reducing treatment and how it influences the clinical outcome in HFrEF patients. The primary study objective is to evaluate how different predictive factors modify the efficacy of HR-reducing treatment, regardless of a drug's mechanism of action, in HFrEF patients. The secondary study objective includes evaluating how different predictive factors modify the efficacy of HRreducing treatment in subgroups stratified by a HR reduction threshold of 10 bpm.

Search strategy
The methodology and results of this study are reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Text A.1) [13]. Two independent reviewers (from Edanz) conducted the search, screened all studies for eligibility, performed data extraction, and assessed the risk of bias for each included study. All disagreements were resolved by consensus or by consultation with a third reviewer if necessary. PubMed, EMBASE, and Cochrane CENTRAL databases were searched, and the search string used in this systematic review was as follows: (chronic heart failure) AND (heart rate); AND (left ventricular contractile dysfunction) OR (reduced ejection fraction) OR (heart failure with reduced ejection fraction); NOT (acute) OR (acute decompensated) OR (acute decompensated heart failure) OR (preserved) OR (heart failure with preserved ejection fraction).

Study eligibility
The inclusion criteria were as follows: studies that were randomized and placebo-controlled clinical trials; studies involving symptomatic HFrEF patients aged ≥ 18 years; studies investigating the effect of HF therapies on change in HR and/or other clinical outcomes; studies published in English and with full text available; and studies published between database inception and December 2020. Studies that fulfilled the following criteria were excluded from the literature review: casecontrol studies, observational studies, studies not investigating HRreducing therapy, studies with no quantitative data or measurable outcomes, studies with incomplete or qualitative data alone, and reviews and collections of conference abstracts.

Quality assessment
The Cochrane Risk of Bias 2.0 tool was used to assess the methodological quality of the studies to be included [14].

Data extraction
The following data, if available, were extracted: study ID, study title, year of publication, study design, sample size, study duration, patient demographic characteristics (age, sex, body mass index, race, and New York Heart Association [NYHA] class), prior HF therapy, baseline characteristics (HR, left ventricular ejection fraction [LVEF], blood pressure, brain natriuretic peptide, serum creatinine, and estimated glomerular filtration rate), HF characteristics (atrial fibrillation [AF], ischemia, prior myocardial infarction [MI], hypertension-induced, valvular disease, coronary artery disease, and dilated cardiomyopathy), comorbid disorders (type 2 diabetes mellitus [T2DM], hypertension, chronic obstructive pulmonary disease, and prior stroke), and outcome measures (all-cause mortality, cardiovascular [CV]-related mortality, hospitalization due to worsening HF [WHF], incidence of major adverse cardiovascular events [classically defined as a composite of nonfatal stroke, nonfatal MI, and CV death], and any composite outcomes). Post-hoc and secondary analyses of the studies identified in our literature search were also searched for any missing data.

Statistical methods
The data were synthesized using an empirical Bayesian random effect meta-analysis to estimate the overall effect on a clinical outcome (all-cause mortality, CV-related mortality, and rehospitalization due to WHF). A restricted maximum likelihood using a Bayesian framework was used for estimating heterogeneity and posterior distributions in this meta-analysis as it allowed for the approximation of posterior distributions using priors derived from the data. All eligible studies were combined to estimate the risk ratio (RR), log(RR), and 95 % credible interval (CrI) on each of the outcomes studied. Heterogeneity between studies was assessed using the I 2 metric (with the alpha level set at 0.05). Subgroup analysis was conducted based on studies in which HR was reduced by ≥ 10 bpm and those in which HR was reduced by < 10 bpm [8]. For this subgroup analysis, we used the same restricted maximum likelihood principle as in the main analysis.
An empirical Bayes random effects meta-regression was used to evaluate the predictive factors of HR-reducing therapies on clinical outcomes using a maximum marginal likelihood method. Here, a metaregression was conducted for all eligible studies where at least 80 % of the predictive factor data were available in 10 or more studies [15]. Prior distributions were calculated from the data and included to model posterior estimates using an empirical Bayes estimator (without any predefined priors) and inputting priors that were derived from the data.
The strength and direction of various associations between the various risk factors and clinical outcomes were described as follows: log (RR) for meta-regression models represents a change in log(RR) when a predictor increases by a 1 %-unit rather than a categorical presence or absence. The only exception was LVEF, which is a 1 % increase in LVEF; a positive log(RR) implies that increasing the percent of that factor will increase the risk of an outcome (i.e., it reduces the risk-lowering effect of HR-reducing therapy), whereas a negative log(RR) implies that increasing the percent of that factor will decrease the risk of an outcome (i.e., it increases the risk-lowering effect of HR-reducing therapy). Bayesian posterior probability p-values were estimated using the  maximum probability of effect method [16]. P values < 0.05 were considered statistically significant. When constructing meta-regression plots, a P value of < 0.1 was considered a factor of interest for evaluating the robustness of the slope profile. Sensitivity analyses were performed to check the primary pooled analysis. In addition, sensitivity analyses were conducted with subjective priors to determine the impact of individual studies on the pooled result. A leave-one-out cross-validation was performed to cross-validate the results and further assess the sensitivity of the meta-analysis and the risk of bias in individual studies. Funnel plots and the Egger's test were used to assess publication bias in conjunction with the leave-one-out cross-validation data. The alpha level for plot asymmetry was set at 0.05. R version 4.2 (R Foundation for Statistical Computing, Vienna, Austria) and RevMan 5.3 (The Cochrane Collaboration, London, UK) were used for statistical analysis. Forest plots were made in the metafor package in R [17]. Bayesian posterior probability p values were estimated using the bayestestR package in R [16].

Bayesian random effect meta-analysis
In this analysis, we included 23,564 patients from 20 studies, and based on the available data, the HF characteristics analyzed included AF, ischemia (grouped as patients with a prior MI or diagnosed with either ischemia or coronary artery disease), non-ischemia, and a NYHA classification of II, III, and IV. We also analyzed a HR reduction ≥ 10 bpm as well as the comorbid presence of T2DM and hypertension.
The data were pooled using a Bayesian random effect meta-analysis (empirical Bayes) to estimate the overall effect on all-cause mortality, CV-related mortality, and rehospitalization due to WHF. The empirical Bayes model showed that HR reduction therapy was associated with a 16.7 % reduction in the risk of all-cause mortality, relative to placebo (RR 0.833 [95 % CrI 0.776, 0.890]) (Fig. 1a). HR reduction therapy was associated with a 16.4 % reduction in the risk of CV mortality in the pooled analysis, relative to placebo (RR 0.836; [95 % CrI 0.769, 0.903]) (Fig. 1b). HR reduction therapy was associated with a 21.1 % reduction in the risk of rehospitalization due to WHF in the pooled analysis, relative to placebo (RR 0.789 [95 % CrI 0.729, 0.849]) (Fig. 1c).
A leave-one-out analysis was performed to cross-validate the results from the Bayesian random effects model and to further assess the sensitivity and the risk of bias in individual studies (Tables A.2-4). The point estimate RRs remained consistent in terms of both magnitude and significance (all p < 0.01), regardless of which study was omitted, and no individual study significantly affected the overall result.

Subgroup analysis stratified by a HR reduction of < 10 bpm
When estimating the overall effect of therapy when stratified by a HR reduction of < 10 bpm or ≥ 10 bpm, we show a significant 15.0 % and 22.4 % reduction in the risk of all-cause mortality in the therapy group, relative to placebo (RR 0.850 [95 % CrI 0.772, 0.929], p < 0.0001 and RR 0.776 [95 % CrI 0.656, 0.896], p < 0.0001) (Fig. 3a). Furthermore, we also observed that there were significant reductions (16.9 % and 25.1 %, respectively) in the risk of CV-related mortality (RR 0.831 [95 %  Metaregression analysis of HR reduction showed non-significant trends in the slope profile for all-cause mortality and CV-related mortality risk per bpm (p = 0.180 and p = 0.224, respectively) (Fig. 3b, c). However, for rehospitalization due to WHF we observed a significant reduction in risk per bpm reduced (p = 0.004) (Fig. 3d). There were insufficient studies to perform a meta-regression analysis when stratifying by a HR reduction threshold of 10 bpm.

Publication bias
Funnel plots and the Egger's test were used to assess publication bias (Fig. A.2a-c). All studies lie within the geometric threshold for plot asymmetry, which indicates that there was no plot asymmetry or publication bias. Egger's test p values were all > 0.05, which also indicates that there was no publication bias in any of the meta-analyses.

Discussion
We conducted the present meta-analysis using a Bayesian approach to evaluate patient and disease-related factors that interact with HRreducing therapy on the clinical outcomes in patients with HFrEF. In this meta-analysis, we showed that HR-reducing therapy was associated with significant reductions in the risk of all-cause mortality, CV-related mortality, and rehospitalization due to WHF. Furthermore, we evaluated nine potential predictors (atrial fibrillation, T2DM, hypertension, LVEF, ischemia, NYHA class [I-III], and HR reduction) and showed that the presence of T2DM significantly modifies the effect of HR-reducing therapy on the risk of all-cause mortality and CV-related mortality.
In this study, the presence of T2DM was associated with an increase in the relative risk of all-cause mortality and CV-related mortality. Patients with T2DM not only have a higher risk of developing HF, but also their CV outcomes, hospitalization rates, and prognoses are substantially worse than those without T2DM [37,38]. Although previous meta-analyses support the benefit of HR reduction in HF patients with T2DM, it has been reported that the magnitude of benefit may be somewhat reduced in chronic HF patients with T2DM [39]. Interestingly, the study by Haas et al used data from CIBIS-II, BEST, ANZ, US-CHF, COPERNI-CUS, and MERIT-HF, which is a small subset of the studies included in the present analysis [39]. There is a complex and interrelated pathophysiology of HF in T2DM due to the dysregulation of several mechanisms; for instance, research shows that HF development in T2DM patients is strongly influenced by hyperglycemia [40] and obesity [41]. Furthermore, T2DM also directly impacts the myocardium leading to progressive structural and functional changes (i.e., diabetic cardiomyopathy) [42]. Overall, the cardiac changes observed in T2DM include increased interstitial fibrosis, increased LV wall thickness, functional myocardial impairments, and chronological impairment due to cardiovascular autonomic neuropathy [43], which together are likely to reduce the benefits achieved by HR reduction.
The meta-regression plot for T2DM's influence on mortality suggests that our observations are well represented by the linear trend in the data. When comparing the all-cause mortality meta-regression plots for T2DM and hypertension, we can see that for T2DM each study included is near the straight line, whereas for hypertension there is a greater degree of heterogeneity, suggesting this slope profile may not be as robust. Furthermore, when comparing the angle of these slope profiles, we can determine that T2DM has a stronger moderating effect on the relative risk for all-cause mortality compared with hypertension. Although this meta-regression plot showed that the percentage of patients was generally greater and there was a non-significant trend, it is not clear whether the log(RR) for hypertension is indicative of an increasing effect on intervention efficacy. Certainly, high blood pressure Fig. 3. Forest plot for HR subgroups (a) and meta-regression plots for HR reduction and all-cause mortality (b), CV-related mortality (c), and rehospitalization due to WHF (d). bpm: beats per minute; CrI: credible interval; CV: cardiovascular; HR: heart rate; RR: risk ratio; WHF: worsening heart failure. An Omnibus test was used for comparing outcome risk for HR reduction < 10 bpm versus HR reduction ≥ 10 bpm.
is an important risk factor for CV disease and mortality, and a reduction in cardiac burden, as elicited by HR reduction therapy, should be more beneficial in this patient subgroup [44]. A previous meta-analysis did identify that a beta-blocker-associated reduction in HR increased the risk of mortality in patients with hypertension compared with patients with HF [45]. However, the meta-analysis by Bangalore et al primarily included studies that administered atenolol and as such this observation may not be a class effect but instead a specific effect of the soluble beta blocker, atenolol. Overall, the relationship between HR reduction and comorbid hypertension is complicated in this patient population and although there was a trend, our results do not support any significant interaction.
Finally, we also showed that HR reduction ≥ 10 bpm conferred a non-significant greater reduction in risk for all-cause mortality, CVrelated mortality, and rehospitalization due to WHF in comparison with a HR reduction of < 10 bpm. The magnitude of HR reduction has previously been shown to be significantly associated with survival benefit, whereas the dose of beta-blockers was not associated with survival benefit [10]. Therefore, our results support previous observations that a greater reduction in HR is associated with a greater reduction in mortality. When performing a meta-regression, the regression slope profile was indicative of a reduction in risk for each outcome, but this was only significant for rehospitalization due to WHF. Here, our data show that as mean HR is progressively reduced there is a significant association with greater reductions in the risk of rehospitalization due to WHF.

Study limitations
Although our meta-analysis included all available randomized controlled trials, there were insufficient data to conduct any subgroup sensitivity analyses. Second, this was a publication-based meta-regression and as such our observations might be influenced by ecological bias [46]. Third, because of the limited number of studies, this meta-regression was not a multivariable regression and thus the association of each possible predictor might be confounded by other prognostic factors. Finally, although we assessed the risk of bias as acceptable, there were a substantial number of biases graded as unclear. However, when exploring the potential risk of bias across the studies, publication bias was excluded using Egger's test.

Conclusions
We observed that the presence of comorbid T2DM in HFrEF patients significantly reduces the benefit of HR reduction, while for comorbid hypertension, there was a non-significant trend for a reduced benefit from HR reduction. As such, the patient populations that would benefit the most from HR reducing therapy are HFrEF patients without comorbid diabetes or hypertension, although only the former was significant. The absence of more factors exhibiting significance is potentially indicative of HR reduction being the most important treatment modality in HFrEF patients. However, we are also aware that our study did not include individual patient data and as such more research is required to further elucidate the predictive nature of the factors evaluated in our study to affect clinical outcomes in patients with HFrEF.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments
This study was supported by Ono Pharmaceutical who funded the medical writing support and paid the article processing charges. We thank Michelle Belanger, MD, and James Graham, PhD, of Edanz (www. edanz.com) for providing medical writing support, which was funded by Ono Pharmaceutical. We also thank Darko Medin and Tim Spelman, PhD, of Edanz for statistical analysis support.

Data Sharing Statement
Qualified researchers may request Ono Pharma to disclose individual patient-level data from clinical studies through the following website: https://www.clinicalstudydatarequest.com/. For more information on Ono Pharma's Policy for the Disclosure of Clinical Study Data, please see the following website: https://www.ono.co.jp/eng/rd/policy.html.

Author Contributions
All authors conceptualized and designed the research, wrote the manuscript, and reviewed the final draft. M.N. and H.F. contributed to project administration. A.Y. and K.O. provided supervision.

Conflicts of interest
A.Y. has nothing to disclose. M.N. and H.F. are employees of Ono Pharmaceutical. K.O. reports honoraria from Daiichi Sankyo and participation in a data safety monitoring board for Ono Pharmaceutical.