The role of cardiac rehabilitation using exercise to decrease natriuretic peptide levels in non-surgical patients: a systematic review

The protocol was registered with PROSPERO (CRD42017051468). The Preferred Reporting Items for Systematic reviews and Meta-Analysis (PRISMA) guidelines were adhered to (Moher et al. 2009). This protocol included a meta-analysis included on the effect of medical therapy on NP levels, which is presented in the accompanying paper (Alphonsus et al. 2019).

The inclusion criteria for this systematic review have been described in the systematic review on natriuretic peptide-directed medical therapy which included exercise therapy trials.

In this systematic review, we report prospective randomised clinical trials of adult medical patients who were randomised to exercise as part of cardiac rehabilitation, where the subsequent changes in natriuretic peptide levels are reported. We excluded (i) trials that monitored natriuretic peptides for prognostic or diagnostic purposes, without a strategy to lower natriuretic peptide levels; (ii) reviews of natriuretic peptide or biomarker physiology; and (iii) trials reporting natriuretic peptides in patients with acute myocardial infarction, pulmonary hypertension, cardiac resynchronisation therapy, and left ventricular assist devices.

Three searches were conducted using search terms ‘brain natriuretic peptide’ AND ‘treatment’, ‘brain natriuretic peptide’ AND ‘heart failure’, and ‘brain natriuretic peptide’ AND ‘exercise’. The following databases were accessed: MEDLINE (PubMed), CINAHL (EBSCO host), EMBASE (EBSCO host), ProQuest, Web of Science, and Cochrane database. There were no filters used for year of publication or language. Non-English titles were not excluded. An example of the search is shown in Additional file 1. The initial search was conducted on 22 December 2016 and updated on 4 March 2018.

Titles were screened for potential inclusion by CA and PG. Abstracts of potential papers identified through the title search were then screened using inclusion and exclusion criteria by CA and PG. The full texts of potential trials were then extracted for full text review and analysis. Reference lists were searched for additional papers that could be included in this review. Data extraction was done by one author (CA) and then checked by a co-author (BB). When required data was not presented in the publication, the authors were contacted for these data.

We extracted data on the NP reduction at the end of the exercise trials. Data on the patient characteristics, the type of exercise intervention, the physical activity in the control group, and the mortality and morbidity in the trials was also extracted.

The primary outcomes for this review were to determine (i) whether exercise therapy was effective in reducing NP levels as compared to control group, (ii) the shortest time period required to reduce NP levels after exercise therapy, and (iii) whether reducing NP levels decreased morbidity and mortality. The secondary outcome was to determine which specific exercise regimens were more effective in reducing NP levels.

Assessment of bias in the studies was conducted by CA and verified by BB following discussion. The Cochrane Collaboration risk of bias tool was used and assessed selection bias, concealment bias, performance bias, detection bias, attrition bias, and other biases. Studies were assessed as having low, unclear, or high risk of bias.

Statistical analyses were conducted using Review Manager Version 5.3 (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014). For NP reduction, we tabulated the absolute NP change. NP levels which were reported as median and interquartile range (IQR) were converted to mean and SD (Wan et al. 2014). As the included trials used either BNP or NT-proBNP to monitor therapeutic response, we made use of standardised mean difference (SMD) for our meta-analysis. SMD addresses the difference in the effect size for an intervention when the units of measurement differ between trials, e.g. use of BNP or NT-proBNP. The SMD is the difference between groups in mean end point divided by the SD of the control group (or pooled SD of the treatment and control groups) (Guyatt et al. 2015). These data are presented as a forest plot. Random effects models were used where the I² statistic > 25% (representing significant heterogeneity); otherwise, a fixed effects model was used.

Risk of publication bias across studies was assessed with funnel plots for NP reduction.

After the initial search, 64 articles were reviewed for potential inclusion. Twenty-six trials (27 publications) were selected and 8 trials added from references, of which 18 were trials of medical therapy interventions and 16 trials were of an exercise intervention (Fig. 1).

We evaluated 2 previous systematic reviews using the AMSTAR format (Additional file 2).

Cardiac rehabilitation exercise trials were conducted in adult patients 18 years and older, in an outpatient setting (Table 1). The included cardiac rehabilitation exercise trials all included some form of aerobic exercise, either walking, bicycle, or treadmill. Trials that involved interval training were considered as a separate subgroup for analysis (Aksoy et al. 2015; Nilsson et al. 2010; Stevens et al. 2015). Three trials included resistance training in addition to aerobic training in the intervention group (Gary et al. 2011; Jonsdottir et al. 2006; Malfatto et al. 2009; Norman et al. 2012). Most trials ran for up to 12 weeks, one trial for 20 weeks (Jonsdottir et al. 2006), and another for 24 weeks (Norman et al. 2012). Most trials were small, with a maximum of 40 patients in each arm, with the exception of 1 large trial (HF-ACTION) which included 477 patients (Ahmad et al. 2014). Most trials included patients with an ejection fraction (EF) < 40%, two trials had patients with EF 40–49% (Guazzi et al. 2012; Parrinello et al. 2010), and three trials had patients with mixed categories of heart failure (Aksoy et al. 2015; Jonsdottir et al. 2006; Nilsson et al. 2010). The exercise intervention protocols were not individualised to the NP levels, but the NP response to the exercise intervention was reported in all the trials.

The exercise intervention group received supervised exercise training in all the trials, except two where the exercises were home-based after participants were given instructions (Eleuteri et al. 2013; Parrinello et al. 2010). The control group were given exercise information (except Brubaker (Brubaker et al. 2009)) but did not receive supervised exercise training.

The risk of bias of the included trials is shown in the Additional file 3: Figure S1 and Additional file 4: Figure S2. The random sequence generation was acceptable in three trials (Ahmad et al. 2014; Guazzi et al. 2012; Maria Sarullo et al. 2006). By virtue of the intervention (supervised exercise versus exercise recommendation), blinding of patients was impossible, and investigator blinding was poor. Outcome assessors were only blinded in two trials (Ahmad et al. 2014; Brubaker et al. 2009). The funnel plots for SMD (Fig. 2) did not suggest publication bias.

All trials presented data on NP levels at the end of the intervention period.