What exercise prescription is optimal to improve body composition and cardiorespiratory fitness in adults living with obesity? A network meta‐analysis

Summary Current international guidelines recommend people living with obesity should be prescribed a minimum of 300 min of moderately intense activity per week for weight loss. However, the most efficacious exercise prescription to improve anthropometry, cardiorespiratory fitness (CRF) and metabolic health in this population remains unknown. Thus, this network meta‐analysis was conducted to assess and rank comparative efficacy of different exercise interventions on anthropometry, CRF and other metabolic risk factors. Five electronic databases were searched for randomized controlled trials (RCTs) that compared different exercise modalities to improve anthropometry, CRF and/or metabolic health in adults living with obesity. RCTs were evaluated using the Cochrane risk of bias tool. A random effects network meta‐analysis was performed within a frequentist framework. Of the 6663 articles retrieved, 45 studies with a total 3566 participants were included. Results reveal that while any type of exercise intervention is more effective than control, weight loss induced is modest. Interventions that combine high‐intensity aerobic and high‐load resistance training exert beneficial effects that are superior to any other exercise modality at decreasing abdominal adiposity, improving lean body mass and increasing CRF. Clinicians should consider this evidence when prescribing exercise for adults living with obesity, to ensure optimal effectiveness.


Summary
Current international guidelines recommend people living with obesity should be prescribed a minimum of 300 min of moderately intense activity per week for weight loss. However, the most efficacious exercise prescription to improve anthropometry, cardiorespiratory fitness (CRF) and metabolic health in this population remains unknown. Thus, this network meta-analysis was conducted to assess and rank comparative efficacy of different exercise interventions on anthropometry, CRF and other metabolic risk factors. Five electronic databases were searched for randomized controlled trials (RCTs) that compared different exercise modalities to improve anthropometry, CRF and/or metabolic health in adults living with obesity. RCTs were evaluated using the Cochrane risk of bias tool. A random effects network metaanalysis was performed within a frequentist framework. Of the 6663 articles retrieved, 45 studies with a total 3566 participants were included. Results reveal that while any type of exercise intervention is more effective than control, weight loss induced is modest. Interventions that combine high-intensity aerobic and highload resistance training exert beneficial effects that are superior to any other exercise modality at decreasing abdominal adiposity, improving lean body mass and increasing CRF. Clinicians should consider this evidence when prescribing exercise for adults living with obesity, to ensure optimal effectiveness. HDL, high-density lipoprotein; HIIT, high-intensity interval training; NMA, network meta-analysis; RCT, randomized controlled trial; T2DM, type 2 diabetes mellitus; TG, triglycerides; VO2max, maximal oxygen uptake; WC, waist circumference.

| BACKGROUND
The prevalence of obesity has tripled over the past 35 years, 1 and it is estimated that it will affect over one billion people worldwide by 2030. [2][3][4] Obesity has far reaching negative effects on health, significantly increasing the risk of cardiovascular disease (CVD), metabolic disease and certain cancers, 1 primarily driven by comorbidities such as type 2 diabetes mellitus (T2DM), dyslipidaemia and hypertension. 2 Diet, exercise and behaviour modification remain the cornerstones of obesity management. 1 However, sole focus on weight loss is not the optimal, as it fails to consider cardiorespiratory fitness (CRF), which can, at medium to high fitness levels, attenuate the adverse consequences of obesity on health, irrespective of body mass index (BMI). [5][6][7] Growing evidence suggests that improved CRF largely neutralizes the adverse effects of increased adiposity as well as other traditional CVD risk factors. 8 Current guidelines recommend that exercise programmes for weight loss in obesity prioritize continuous moderate intensity aerobic exercise, and supplement this approach, where possible, with resistance training. 9,10 However, multiple exercise modalities, with varying intensities, feature in the obesity literature and data identifying the relative effect of the different exercise interventions on CRF, body composition and metabolic health are somewhat inconsistent. 11 Several randomized controlled trials (RCTs) and systematic reviews report improvements in maximal oxygen uptake (VO 2max) , waist circumference (WC) and BMI with aerobic and/or combined training, [11][12][13][14][15][16][17][18][19] whereas improvements in VO 2max and in some cases decreased BMI and blood pressure are similarly reported by applying resistance training alone. 16,[18][19][20][21][22][23] It is difficult in this context to determine the superiority of the different exercise interventions using only individual RCTs or even pairwise meta-analysis, as often these studies were designed to compare one or more exercise interventions with data from nonexercising control groups and therefore cannot discriminate between exercise modalities, making it difficult to draw firm conclusions.
Also called mixed treatments comparison or multiple treatments comparison meta-analysis, network meta-analysis (NMA) expands the scope of a conventional pairwise analysis by analysing simultaneously both the direct and the indirect evidence from different studies, allowing for estimation of the relative effectiveness among all interventions and rank ordering of the interventions, even where two interventions comparisons are lacking. 24 To date, no systematic review has pooled the effects of different training modalities on outcomes of anthropometry, CRF and cardiometabolic risk factors, focusing exclusively on adults living with obesity and including only RCTs where exercise is the only intervention being investigated. Therefore, this study aimed to conduct a NMA of RCTs to (i) assess the comparative efficacy of different exercise types and their intensities on anthropometry, fitness and other metabolic markers in adults living with obesity and (ii) establish a hierarchy of these exercise interventions.

| Registration
This systematic review and NMA is reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. 25 The study protocol was registered (registration number: CRD4201811373) with the International Prospective Register of Systematic Reviews (PROSPERO).  (Table S1). In addition to the databases, the reference lists of included articles were scanned for articles that met the inclusion criteria.

| Eligibility criteria
RCTs published in scientific peer reviewed papers, written in English and from the past 30 years (January 1998 to November 2019) were included (conference abstracts, reports and theses were excluded).
Adult-based studies were defined by a study population aged between ≥18 years and <65 years and obesity was defined as a BMI > 30 kg m −2 or body fat > 30% in women and >25% in men. Articles where the population was reported to be taking medication directed at excess weight or had a diagnosed pathology (e.g., type 2 diabetes) were excluded. Outcomes of interest included anthropom-

| Study selection
Endnote X8 literature management software was used to manage the literature search records. The selection process consisted of three phases. In the initial phase, three reviewers independently screened the yielded articles based on title. In the case of doubt, the articles were included in the abstract review phase. In Phase 2, all articles selected from the initial phase were reviewed by abstract and assessed for eligibility by two independent reviewers. Any disagreements were resolved by discussion between reviewers and consultation with a third party from the review team. In the final phase, the remaining articles were fully reviewed by the same two independent reviewers that reviewed abstracts, using the predetermined inclusion criteria. Any disagreements between reviewers in this phase were again resolved by discussion within the wider team.

| Data extraction
A nine-item, standardized and prepiloted data extraction form was used to record data from the included studies under the following headings: (i) author, (ii) year of publication, (iii) country, (iv) study period, (v) sample size, (vi) mean age, (vii) mean baseline, (viii) followup BW, BMI, %BF, WC, BP, HDL, TG, FBG and VO 2max and (ix) details of the exercise intervention. Data were recorded for each exercise intervention using the F.I.T.T.

| Risk of bias of individual studies
Two authors (G.O. and C.C.) independently assessed the risk of bias (ROB), in accordance with the Cochrane Handbook version 5.1.0 tool for assessing ROB in RCTs. 27 The following seven domains were considered: (i) randomized sequence generation, (ii) treatment allocation concealment, blinding of (iii) participants and (iv) personnel,

| Data analysis
First, we qualitatively described included trials, their exercise intervention characteristics and their relative contributions to the overall body of evidence available. We converted outcomes to standard units and calculated mean difference, subtracting the mean at the end of the intervention versus baseline. End of the intervention was always at the end of exercise participation to avoid any potential wash out. If the standard deviation (SD) difference was missing, we calculated using the SD formula of the difference between two means using the following formula: Omitted data (e.g., missing SDs or only p values reported), was dealt with by using the metaeff command procedures in Stata to calculate SMDs and 95% confidence intervals (CIs) from available data. 28 When a study had multiple intervention arms and where we defined two arms as constituting the same exercise intervention (e.g., 30 min of continuous moderately intense aerobic exercise versus 3 × 10-min bouts of moderately intense aerobic exercise), the data from the intervention groups were pooled.
Where studies used the same outcome measure, we pooled mean, SD and sample size under each of the predefined exercise categories. Regarding VO 2max , when it was reported in relative terms (ml kg −1 min −1 ), absolute VO 2max was calculated using the mean BWs for the appropriate time points. For each exercise intervention employed in a study, the absolute change in VO 2max (L min −1 ) was calculated, whereas the pooled SD (SD P ) was calculated using the sample size (n) and SDs from before and after in each study or study group. 29 As interventions are by definition heterogeneous and pairwise meta-analytic estimates are usually reported in addition to the network estimates, 30 random effects pairwise meta-analyses were first used to obtain pooled SMDs (with associated CIs) for weight loss, BMI, WC, %BF and fitness, respectively, with the I 2 statistic used to quantify heterogeneity. Transitivity is a key assumption of NMA and refers to the belief that indirect comparison is a valid estimate of the unobserved direct comparison 31 and that all studies have homogenous distribution of effect modifiers. 32 Because duration of the exercise intervention (number of weeks) had been signalled as a potential modifier for all outcomes, we explored by pairwise meta-analyses with duration as a covariate as to whether it modified the magnitude and direction of the estimates. As the intervention duration did not modify the estimates for each of the outcomes (Table S2), we proceeded to NMA.
We used STATA software (Version 15.1) command 'mvmeta' to perform a multivariate NMA within a frequentist framework 33 in accordance with current PRISMA NMA guidelines. 25 To allow for between-study heterogeneity, a random effects NMA was performed to calculate pooled estimates and 95% CIs. The mean difference was used as the effect estimate to analyse the results.
In addition to the pairwise meta-analysis, transitivity assumptions were also assessed through assessment of individual studies inclusion criteria, whether all participants in the network could have been randomized to any intervention, logical inference and using consistency models. 34  Intervention hierarchy was summarized and reported as a P score. 35 The P score is considered as a frequentist analogue to surface under the cumulative ranking curve (SUCRA) values and measures the extent of certainty that a treatment is better than another treatment, averaged over all competing treatments. The P score ranges from 0 to 1, where 1 indicates best treatment with no uncertainty and 0 indicates worst treatment with no uncertainty.
While the P score or SUCRA can be usefully re-expressed as the percentage of effectiveness or acceptability of the exercise interventions, such scores should be interpreted cautiously unless there are actual clinically meaningful differences between interventions. 36 To check for the presence of bias due to small-scale studies, which may lead to publication bias in NMA, a network funnel plot was generated ( Figure S8) and visually inspected using the criterion of symmetry. 37 3 | RESULTS

| Literature selection
A total of 6663 studies were initially identified. Following review by title and abstract, 174 studies progressed to full manuscript review. Of these, 129 were excluded as they did not fulfil our inclusion criteria. The remaining 45 studies were included in this review. [13][14][15]17,21, The detailed process is illustrated in

| Characteristics of the included studies
The characteristics of the included studies are presented in Table 2.

| Body mass index
Twenty-seven studies, including 1543 participants and all six exercise interventions categories, contributed to this analysis. The majority of the data analysed came from aerobic interventions (39.3%).
Resistance and combined training contributed 12.7% and 10.0%, respectively (Table S5) Table 3 illustrates the complete matrix. COM-HI was the best intervention in the network comparison for decreasing BMI (P score = .85) ( Table 4).  Table 2). COM-LM (P score = 0.82) was the best exercise intervention in the network comparison for reducing WC (Tables 3 and 4).

| Percentage body fat
Twenty studies, including 1480 participants, reported %BF and were included in the NMA. Over 40% of the data included in the NMA was

| Cardiorespiratory fitness
Twenty-one studies with 1689 participants and all six intervention categories contributed to the NMA assessing CRF. Aerobic exercise interventions contributed the majority of data to this NMA (  Hence, dietary intake is key to weight loss and exercise should be combined with diet to optimize its effectiveness for the management of obesity. 80 Several studies have reported %BF to be more responsive to exercise than BW, and because body fat is the most metabolically harmful tissue type, it may be a more meaningful measure of health change for evaluating exercise interventions. 81 82 However, the effect size reported for both outcomes was small, and as exercise was reported as a single modality irrespective of the mode, frequency or intensity, it is difficult to decipher which specific exercise prescription/s resulted in these changes. 82 Our NMA goes one step further by providing detailed breakdown by exercise intervention type. We found that the optimal exercise intervention for decreasing %BF is a COM-HI training programme. This intervention resulted in a much greater reduction in % BF (−2.15 [CI = −4.0%, −0.3%]) than previously reported by Kim et al. 82 and was shown to be significantly more effective than either isolated aerobic or resistance training.
Beyond total body fat, abdominal adiposity is independently associated with all-cause mortality and is recognized as a better predictor of obesity-related conditions than either BW or BMI. [83][84][85] T A B L E 4 Ranking of exercise interventions in order of effectiveness Abbreviations: AE-M, aerobic moderate intensity; AE-V, aerobic vigorous intensity; COM-HI, combination of high-intensity aerobic and high-load resistance exercise; COM-LM, combination of low-moderate intensity aerobic and low-moderate load resistance exercise; R-HI, resistance high load; R-LM, resistance low-moderate load. a P score ranges from 0 to 1, where 1 indicates best treatment with no uncertainty and 0 indicates worst treatment with no uncertainty.
WC is often used as a surrogate marker of abdominal fat mass, as changes in WC are associated with changes in intra-abdominal fat and in turn changes in health risk profile. 84 Findings from a large meta-regression analysis exploring WC as a predictor of CVD reported an increase of 1 cm in WC resulted in a 2% increase in CVD risk. 84 Using these findings in reverse and the fact that even modest reductions in WC (<2 cm) confer improvements in all metabolic risk factors, 86 it is evident from this NMA that COM-LM exercise was most effective when it came to reducing WC and this is associated with a 5.5% decrease in CVD risk, followed by AE-V and AE-M with a 4.5% and 4% decrease, respectively. Furthermore, COM-LM was more than twice as effective in changing CVD risk compared with either low or moderate load resistance training, a finding that both consistent with 87 and contradictory 16 to previously published meta-analyses.
Until recently, CRF has been overlooked as a potential modifier of the inverse association between obesity and mortality. Despite a growing body of evidence that demonstrates that CRF exerts more influence on morbidity and mortality than %BF and its distribution, 8,88 even the most recent meta-analyses investigating the effectiveness of different exercise interventions for adults living with obesity continue to exclusively focus on anthropometric measures to establish efficacy or effectiveness 18,89 and do not include CRF in their analyses. Others that have included CRF as a primary outcome are not specific to populations with obesity, do not distinguish between exercise types and report fitness in relative terms.
Reporting a change in relative (ml kg −1 min −1 ) as opposed to absolute (L min −1 ) VO 2max has been shown to hamper interpretation of the data in studies of people with increased body mass. 90 It can result in overestimated level of fitness as any loss in body mass automatically increases fitness expressed in relative terms, often without the desired underlying metabolic changes.
Although our findings in relation to fitness were statistically insignificant, all exercise interventions brought about an increase in absolute VO 2max of between 7% and 15% (Table S4). Previous research has shown an improvement in fitness of between 8% and 10% is associated with a 12% decrease in mortality 6 and is comparable with a 7-cm reduction in WC. 7 When applying this finding to our study, we find that while aerobic exercise (AE-V = 12.9%; AE-M = 9.2%) outperformed both types of resistance training (R-HI = 7.4%; R-LM = 7.2%) as a single modality, combined low high (COM-HI = 15%) intensity training resulted in the greatest increase in fitness and consequently should be considered as the preferential exercise intervention to improve CRF in this population. 6,7

| STRENGTHS AND LIMITATIONS
This study has several strengths and weaknesses. The review was systematic and exhaustive. A considerable sample size of adults living with obesity (n = 3566) were included, thus providing the power to detect statistically significant mean differences. Only RCTs, the gold standard for evaluating the effectiveness of an intervention, were included. Incorporating CRF as an outcome measure in the NMA also strengthened this study as it is regularly omitted in obesity studies that evaluate exercise interventions, despite its importance as a predictor and correlate of all-cause mortality.
Our review shares a number of limitations with the studies on which it is based. Although we attempted to limit heterogeneity by using strict inclusion and exclusion criteria, study populations Although the majority of the included trials reported metabolic outcome measures as baseline data (BP, TG, HDL and FBG), they did not report them in postintervention results. Exercise is considered as a cornerstone in the prevention and management of the metabolic syndrome; hence, future trials need to consider these metabolic markers as primary rather than secondary or tertiary outcomes when designing studies.
This NMA identified missing evidence in relation to a number of the exercise categories we included. Aerobic exercise, was, by far, the commonly prescribed intervention, contributing more than half (60-68%) of the intervention data to all five of the NMA's (Table S6). In comparison, resistance training and combined only contributed between 12-18% and 10-18%, respectively. Consequently, due to limited data for head-to-head comparisons for some interventions (