Introduction

Heart failure (HF) is a complex chronic condition associated with a high rate of comorbidity, significant impairment of quality of life, frequent hospitalization, progressive disability, and reduced survival [1,2,3]. Recent estimates suggest that HF affects one million Canadians, with at least 50,000 new patients diagnosed each year [1, 2]. HF patients have a shorter life expectancy than most cancer patients, with a median survival of 1.7 years for men and 3.2 years for women [4].

HF is also associated with an enormous economic burden. It is one of the most costly chronic conditions in developed countries [5], costing Canada an estimated $3 billion each year [6]. The majority of these costs are attributable to hospitalization and other healthcare services, medications, and missed days of work [6]. HF-related hospitalizations are one of the major drivers of avoidable admissions [5, 7,8,9]. Roughly 25% of patients who are hospitalized for HF are readmitted within 30 days of discharge [10]. The costs of hospital admissions related to HF alone are currently estimated to be $482 million per year; this is projected to increase to $722 million by 2030 [11].

Current Canadian guidelines recommend that HF care be provided in multidisciplinary, specialized HF clinics [12]. A number of studies have demonstrated the effectiveness of these clinics [13,14,15,16,17,18]; a systematic review and meta-analysis by Holland et al. reported significant reductions in all-cause hospital admissions, HF-related admissions, and all-cause mortality [17]. A more recent study demonstrated a significant reduction in unplanned admissions among patients treated in HF specialist clinics [18]. Although studies have demonstrated the effectiveness of HF clinic care, the optimal model for management of patients in these HF clinics is unknown [8, 19]. HF clinics are very heterogeneous in structure, population, and patient management [8, 20, 21]; this heterogeneity may considerably undermine their effectiveness [19]. Recent studies have reported variability in the effectiveness of different models of care offered in HF clinics, emphasizing the gap in knowledge and practice regarding the optimal components of HF care [22,23,24].

The objective of this systematic review and network meta-analysis is to determine which components of care provided in HF clinics are the most effective in reducing mortality, emergency department visits, and hospitalizations and improving patient quality of life.

Methods

This systematic review will be designed based on the Cochrane review methods [25] and will follow the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [26]. The protocol follows the Preferred Reporting Items for Systematic Review and Meta-analysis Protocols (PRISMA-P) and checklist (Additional file 1) [27, 28] and the PRISMA extension for network meta-analyses (PRISMA-NMA) [29]. This protocol is registered with PROSPERO as CRD42017058003.

Eligibility criteria

Participants

The population will consist of adult patients (18 years of age and older) who are receiving treatment for a diagnosis of HF. Studies with participants who were treated by a HF clinic prior to study enrollment will be excluded. Studies that included both eligible and ineligible participants will be excluded unless they report data for the subgroup of patients who are eligible for this study.

Interventions

Studies must evaluate at least one component of multidisciplinary HF clinic care (or any combination of care components) with a defined intervention [18, 30,31,32]. Based on Canadian and US guidelines [31, 32], we anticipate including the following interventional components: education and counseling, medication management and optimization, social/peer support, clinical monitoring, referral for diagnostic testing, referral for devices (e.g., implantable cardioverter defibrillator [ICD], cardiac resynchronization therapy [CRT]), and referral for cardiac rehabilitation. This is not an exhaustive list of care components, so we will include others that have been studied in the literature.

Comparator

Studies must include at least one comparator arm. Comparisons can be made between other care components or to standard care. Non-comparative studies reporting a single intervention or component of care will be excluded.

Outcomes

Studies will be included if they report on at least one of the following outcomes, assessed after a follow-up of 30 days or more: all-cause or HF-related mortality, all-cause or HF-related hospitalizations or emergency department visits, or health-related quality of life (HRQoL) using a validated measure or a composite end-point of any of these.

Study designs

This systematic review will include both randomized and non-randomized controlled trials and observational cohort studies evaluating at least one component, or any combination, of multidisciplinary HF clinic care with a defined intervention [18, 30,31,32]. Including both controlled trials and observational studies will allow us to compare findings between these study designs, comparing the efficacy of the interventions in an ideal context versus the effectiveness in the real world.

Study designs other than controlled trials or observational cohort studies will be excluded, as will other types of publications, including letters, commentaries, and editorials. Previous systematic reviews will be used to identify relevant studies. Studies that use duplicate study populations will be reviewed to ensure that they provide different information; otherwise, only the most recent study will be included.

Setting

Restrictions on the type of setting will not be imposed.

Language

Both English and non-English language articles will be considered. Non-English articles will be translated into English for the purpose of inclusion in the systematic review.

Information sources

Eligible studies will be identified through a systematic, comprehensive search of the following databases: MEDLINE and MEDLINE In-Process (Ovid), EMBASE (Ovid), CINHAL (Ebsco), Cochrane Database of Systematic Reviews (CDSR), DARE (Database of Abstracts of Reviews of Effectiveness), LILACS, International Clinical Trials Registry Platform (ICTRP), Science Citation Index Expanded (SCI-EXPANDED; Web of Science), PsycINFO, and Conference Proceedings Citation Index-Science (CPCI-S; Web of Science), as well as ProQuest Dissertations and Theses Global. HF disease management programs were launched in the 1990s; as such, a database search from 1990 onward is considered sufficient. The PROSPERO registry will also be searched for all active or completed systematic reviews. Electronic database searches will be supplemented with inspection of reference lists of relevant articles and hand searching of pertinent journals.

Search strategy

The search strategy will be designed and conducted by an information specialist experienced in systematic reviews, following the Cochrane systematic review methodology [25]. It will include MeSH and natural language terms in the following concept areas: heart failure and eligible interventions. No language restrictions will be applied. The search will be updated closer to the end of our review to ensure that the most recent eligible articles are captured. A detailed preliminary search strategy for MEDLINE (Ovid) is provided in Additional file 2. The final Medline strategy will be translated into syntax appropriate for each database used in the search.

Data management

The results of the literature search will be collected in an EndNote library, where duplicate studies will be flagged and removed. Each reviewer will receive a copy of the library for the stepwise review process. A coding scheme will be developed and used within the EndNote software to indicate eligibility status at each stage of the screening process (i.e., title/abstract, full-text screening).

Study selection and screening process

A stepwise review process will be utilized [26]. Using a study eligibility form (Additional file 3) based on the Data Collection Form for Cochrane Reviews [33], two reviewers will independently screen in duplicate the titles and abstracts of studies identified by the search strategy. All studies deemed eligible by either reviewer will be selected for full-text review. The same reviewers will then screen the full article text and apply the eligibility criteria stated above. During the full-text screening phase, both reviewers will record the reason for exclusion in a standardized data abstraction document based on the Data Collection Form for Cochrane Reviews [33]. Disagreements during this screening phase will be resolved through discussion with a senior author. For each stage of the review process, agreement will be measured through a weighted kappa statistic.

The eligibility criteria will be pilot tested on a sample of citations at both stages of screening. This will ensure that both reviewers understand the eligibility criteria and allow for refinement if needed. These sets of citations will not be included in the calculation of overall agreement at the end of full-text screening.

Data collection

A standardized data abstraction document will be developed based on the Data Collection Form for Cochrane Reviews [33]. Two independent reviewers will extract data for each included study. Any discrepancies will be resolved through discussion with the reviewers and a senior author. If necessary, corresponding authors of the eligible studies will be contacted. Where possible, data will be extracted for the intent-to-treat analyses.

All abstracted data will be collected and organized into a MS Excel spreadsheet.

Data items

For each included study, information related to the study characteristics, patient population, and outcomes of HF care assessed will be extracted. Study characteristics will include study design, component(s) of care assessed in the comparison groups, length of follow-up, setting (i.e., hospital, home, community), sample size, year of study conduct, year of publication, and country of origin. Data regarding the patient population will include age, gender, and severity of HF. In addition, the following outcomes will be extracted where applicable: all-cause and HF-related mortality, all-cause and HF-related hospitalizations, all-cause and HF-related emergency department visits, and health-related quality of life.

Risk of bias and quality assessment

We will use the modified Cochrane risk of bias tool [34] to assess the risk of bias in each included randomized controlled trial (RCT) in the following domains: random sequence generation, allocation concealment, blinding of outcome assessment, incomplete outcome data, and selective reporting. The Cochrane tool includes a domain to assess blinding of patients or participants; however, for the purpose of this study, this domain is not relevant due to the procedural nature of the intervention(s). Through the Cochrane bias tool, each reviewer assigns a judgment of “high,” “low,” or “unclear” risk for each domain item.

The Risk Of Bias In Non-randomized Studies of Interventions (ROBINS-I) tool will be used to assess bias in each included non-randomized controlled trial or observational study [35] in the following domains: confounding; participant selection; intervention classification; deviations from intended intervention; missing data, including loss to follow-up or missing data on intervention status or other important variables; outcome measurement; and selection of reported results. Risk of bias for each domain is categorized as “low,” “moderate,” “serious,” or “critical.”

Two reviewers will independently assess bias for each included study using the appropriate tool. Any disagreements will be resolved by discussion with a senior author.

Strength of the evidence assessment

Two reviewers will independently assess the quality of evidence using the Grades of Recommendation, Assessment, Development, and Evaluation (GRADE) system [36] for each included study. We will take into account the risk of bias at the study level, inconsistency, imprecision, indirectness, publication bias, gradient response, large effect, and bias against a significant association [37,38,39,40,41,42]. Confidence will be rated as high, moderate, low, or very low [43], and disagreements will be resolved by discussion with a senior author. Funnel plots will be used to assess publication bias [44, 45].

Synthesis of results

The data extracted from the included RCTs will be analyzed separately from the non-randomized controlled trials and cohort studies using network meta-analytic (NMA) techniques to allow us to contrast findings between these study designs, comparing their effectiveness in an ideal context versus in a pragmatic, real-world setting.

In an approach similar to previous work evaluating the effectiveness of different components of psychological care for coronary heart disease [46], four increasingly complex models of the effects of the core components of HF clinic care will be fit for each outcome:

  1. 1.

    Single-effect model. All HF clinic care components will be grouped together as a single treatment, estimating the overall effect of HF clinic care. This is similar to traditional meta-analytic methods that compare HF clinic care to standard care.

  2. 2.

    Additive main effects model. This assumes a separate treatment effect for each of the HF clinic care components. The effect of having two or more components is additive. Each arm of each trial will be coded based on the presence or absences of the particular care component. This decomposes the effect in any given arm into the sum of effects of the components present in that arm.

  3. 3.

    Two-way interaction model. This is an extension of the main effects model that allows for synergistic or antagonistic effects of combinations of each pair of the components of HF clinic care. For example, the effect of component X (e.g., education and counseling) and Y (e.g., clinical monitoring) is not simply the effect of X + Y but could be larger or smaller.

  4. 4.

    Full interaction model. This model assumes that each possible combination of the core components has a distinct effect that cannot be computed from their individual effects or the sums of the two-way interactions. In practice, this model assumes that each different combination of HF clinic care components is a different intervention.

We will choose the simplest model that is consistent with the data and choose between models using the deviance information criterion (DIC). The hazard ratio (HR) and its 95% credible interval (CrI) for each core HF clinic component will be estimated for the mortality, hospitalizations, and emergency department visits outcomes using Markov chain Monte Carlo (MCMC) simulations implemented in JAGS software run through R [47]. For health- related quality of life outcomes, we will estimate the median difference and its 95% CrI. We will run three parallel MCMC chains with a sufficiently large number of iterations and ensure convergence using Gelman-Rubin diagnostic trace plots. Heterogeneity and model fit of the NMA will be assessed through standard approaches [48, 49]. Posterior predictive checks will also be used to identify outlying observations (arms or trials) [50] in the models examining core components. This simulation framework will allow for the presentation of key data summaries of clinical and policy interest, such as the probability that a particular component of care is the most effective for each outcome and the estimated treatment effects for any specified combination of care components.

As we hypothesize that the effectiveness of HF clinic care may differ across a variety of factors, we will explore any potential causes of heterogeneity through the following subgroup analyses, conditional upon data availability: gender, follow-up period, HF severity, method of enrollment and referral (i.e., self-report, systematic referral, or other processes), setting, and number of sites. In addition, as the intensity of care components may vary between studies (i.e., providing one education and counseling session versus providing five sessions), we will define a subjective measure of “intensity” for each component of care, as agreed upon by the study team, and perform stratified analyses to assess the impact of component intensity on effectiveness. We will also consider the country and year of study as standard care may differ between countries and changed over time (i.e., the introduction of new medications and medication protocols in 2016 [51]). Sensitivity analyses including and excluding low-quality studies will be conducted to assess the effect of study quality on our findings.

Discussion

The global burden of HF is enormous, affecting approximately 26 million people worldwide [52] and reaching yearly global costs of $108 billion [53]. While HF clinic care has been proven beneficial for patients, they have been implemented in widely different formats [8, 20, 21]. Unfortunately, this heterogeneity may considerably undermine their effectiveness [19]. Given limited healthcare resources, identifying the parts of HF clinic care that are associated with better outcomes may allow not only improved care but also reduced costs. As such, it is crucial to understand which components of care offered in HF clinics are the most effective and central to quality patient care.

Studies have attempted to partially address this via traditional systematic review and meta-analysis; for example, a recent Cochrane systematic review and meta-analysis of 25 trials assessed the effectiveness of disease management interventions for patients with HF [30]. The authors note that they were unable to determine the optimal components of HF management programs; however, the study was not designed to address this important question. We believe that the use of network meta-analysis will allow us to identify which components of HF clinic care are associated with improved outcomes.

Care offered to patients in HF clinics is an example of a complex intervention that involves multiple components and multidisciplinary healthcare providers [54,55,56,57]. Traditional approaches for evidence synthesis, such as meta-analysis and meta-regression, are not sufficient to handle the assessment of complex health interventions such as HF clinic care [58, 59]. Traditional meta-analysis works well with simple interventions but is unable to separate the effects of individual components of complex interventions. NMA not only allows information from indirect comparisons to be integrated with head-to-head comparisons used in traditional meta-analysis, but will also allow us to model the effects of individual components of HF clinic care. Most commonly based on a framework of Bayesian statistics, NMA has been developed to assess the relative effectiveness of several interventions, synthesizing evidence across a network of randomized trials and observational studies [60,61,62,63,64,65,66]. The method simultaneously analyzes direct evidence (studies that directly compare the interventions of interest) and indirect evidence (studies that compare interventions of interest through a common comparator) [67,68,69,70], allowing comparison of interventions that have not been evaluated head-to-head. Increasingly, NMA is being used [46, 69, 71,72,73,74,75] to answer questions such as ‘Which type of intervention has the greatest possibility of being most effective?’ or ‘Which combination of components are likely to be most effective?’. As such, we feel that a NMA is a more appropriate method to address our research question.

To our knowledge, this will be the first study to evaluate the comparative effectiveness of the different components of care offered in HF clinics. The findings of this systematic review and network meta-analysis may inform HF guidelines and recommendations regarding the key components of HF clinic care. While referral to and enrollment in HF clinics are currently low, with only one-seventh of HF patients being referred [20], we anticipate growing referral rates as the aging population increases. With limited resources for healthcare, it is critical to understand which components of care yield the best outcomes.