Comparative effectiveness of lifestyle interventions on children's body composition management: A systematic review and network meta-analysis

Highlights

• •Multi-component approaches incorporating elements such as physical activity, diet, behavioral therapy, and informational support have been shown to be more effective in reducing children's BMI and improving body composition than single-component approaches.
• •Among various single-component lifestyle interventions, physical activity has proven to be the most effective.
• •Children with a higher baseline BMI demonstrated a significant association with a larger effect size in response to lifestyle interventions.

Abstract

Purpose

This study aimed to provide comparative evidence on the effectiveness of various lifestyle interventions on body composition management for preschool and school-aged children.

Methods

PubMed (MEDLINE), Embase, CINAHL, and Web of Science were systematically searched for this network meta-analysis. Randomized controlled studies (RCTs) that included children aged 4–12 years with no physical or mental conditions; performed at least 1 type of lifestyle intervention; reported change in body mass index (BMI), BMI z-score, or body fat percentage (BFP); and were published between January 2010 and August 2023 were included.

Results

The final analysis included 91 RCTs with aggregate data for 58,649 children. All interventions were categorized into single-arm approaches (physical activity, diet, and behavioral and informational support) and combined arms approaches (bicomponent and multicomponent treatment). Multicomponent treatment showed significant effectiveness on the reduction of BMI (mean deviation (MD) − 0.49, 95% confidence interval (95%CI): –0.88 to –0.12), BMI z-score (MD = –0.11, 95%CI: –0.18 to –0.04), and BFP (MD = –1.69, 95%CI: –2.97 to –0.42) compared to the usual care condition. Bicomponent treatment also significantly reduced BMI (MD = –0.28, 95%CI: –0.54 to –0.04) and BMI z-score (MD = –0.07, 95%CI: –0.12 to –0.02) compared to usual care.

Conclusion

Interventions targeting multiple lifestyle components achieved greater reductions in children's BMI and BFP. Among single-component approaches, physical activity engagement emerged as the most effective. These findings should guide practitioners in recommending comprehensive lifestyle modifications for children. Moreover, children with higher initial BMI and body fat levels tend to exhibit more positive responses to lifestyle interventions aimed at managing obesity.

Graphical abstract

1. Introduction

The global prevalence of overweight and obesity among children and adolescents aged 5–19 is a significant concern as it exceeds 18% according to the World Health Organization (WHO).¹ The Centers for Disease Control and Prevention (CDC) reports a comparably alarming 19.7% average prevalence of obesity among children and adolescents aged 2–19 years in the USA, with increasing prevalence as age advances (12.7%, 20.7%, and 22.2% for children aged 2–5, 6–11, and 12–19 years, respectively).² Childhood obesity has been shown to persist into adolescence and adulthood,^3,4 exhibiting notable correlations with various chronic diseases and psychological issues, including cardiovascular disease, type 2 diabetes, osteoarthritis, anxiety, depression, and low self-esteem.5, 6, 7, 8 A critical period influencing subsequent adiposity development is the onset of adiposity rebound, which may manifest as early as around 5 years old.⁴ Therefore, early actions to address obesity are essential, especially during the middle childhood period, which is a crucial yet often overlooked stage in human development.⁹

Obesity treatment in children may include physical activity (PA) intervention, dietary modification, behavioral therapy, pharmaceutical treatment, and metabolic and bariatric surgery.^10,11 The American Academy of Pediatrics (AAP) recommends considering pharmacotherapy for children aged 12 years and older as an adjunct to health behavior and lifestyle treatment, while adolescents aged 13 years and older may be evaluated for surgery as a third step when deemed appropriate.¹² In a comprehensive systematic review, it was revealed that nearly half of the randomized controlled trials (RCTs) on lifestyle interventions were effective in reducing children's adiposity.¹³ Meta-analyses on childhood obesity control often group together diverse strategies, such as exercise, diet, and education, under the umbrella of lifestyle modification interventions. These analyses typically conduct pairwise comparisons against control groups and have shown improvements in weight-related outcomes across children ranging from preschool-aged to adolescents.14, 15, 16, 17, 18, 19 This is noteworthy, as it challenges the conventional belief that childhood obesity interventions are generally ineffective.¹³ However, the lifestyle interventions analyzed in meta-analyses encompass multiple types of approaches, each with varying details of intervention delivery, making it challenging to determine which approach among the lifestyle modifications is more effective. Network meta-analysis (NMA) enables the simultaneous comparison of multiple interventions using both direct and indirect evidence. This provides a robust framework for generating effectiveness rankings that inform clinical and policy decisions.^20,21 Nonetheless, previous NMAs have not yielded consistent findings regarding a universally more effective approach among all lifestyle interventions.^20,21 Bae and Lee's analysis²² suggested that interventions with a clear focus outperformed combined component interventions while Liang et al.’s²³ study concluded the opposite, suggesting that a combination of approaches demonstrated superiority over single-target interventions. Therefore, in this NMA, our objective was to incorporate more recent evidence and assess the impact of various types of lifestyle interventions on the body composition of preschool and school-aged children. Additionally, we aimed to reconcile previous discrepancies in the field and provide evidence-based recommendations for targeted treatment options in addressing childhood obesity.

2. Methods

2.1. Search strategy

This review was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses extension statement for Network Meta-Analysis (PRISMA-NMA) guidelines,²⁴ with International Prospective Register of Systematic Reviews (PROSPERO) registration (CRD42020216394) (Supplementary Table 1). PubMed (MEDLINE), Embase, CINAHL, and Web of Science were systematically searched from January 2010 to August 2023. We conducted the systematic search using a combination of relevant medical subject heading (MeSH) terms and keywords including “children”, “overweight”, “obesity”, “physical activity”, “exercise”, “diet”, “nutrition”, “behavior therapy”, “health education”, “lifestyle”, “lifestyle intervention”, and “randomized controlled trials”. Additionally, we manually examined the reference lists of previous related systematic reviews and meta-analyses to identify studies not captured in the initial search. Further details about our search strategy can be found in Supplementary Table 2.

2.2. Eligibility criteria

The eligibility criteria were defined following the participants, intervention, comparator, outcome, timing, and setting (PICOTS) framework. A detailed criteria table is provided in Supplementary Table 3. The participants included children with a mean age between 4 and 12 years old with no preexisting physical or mental health conditions. This age range was chosen to mitigate the potential confounding effect of puberty on weight gain and body composition.²⁵ To ensure adequate effectiveness, only RCTs with interventions lasting longer than 6 weeks were included in this analysis. Studies using face-to-face-delivered lifestyle modification intervention, either in comparison to usual care or an active control group, were included where the “lifestyle” intervention was defined to exclude any pharmaceutical or surgical components. Both prevention and treatment intervention programs were included in this analysis. The reported outcome measures must include body mass index (BMI), BMI z-score, or body fat percentage (BFP); otherwise, studies on obesity interventions solely reporting other outcomes were excluded. The primary outcome for this analysis was BMI, which was chosen due to its extensive reporting in research studies, both as demographic information and intervention outcome. Furthermore, its simplicity of measurement—requiring minimal equipment and financial resources—makes it a practical selection. Despite not directly measuring body fat, BMI is widely regarded as a reliable and commonly used proxy for body composition.26, 27, 28, 29, 30 Alongside BMI, BMI z-score, and BFP were investigated as secondary outcomes in this study. Trials involving children with morbid obesity, those conducted in specialty care settings, and those published in languages other than English were excluded from this analysis.

2.3. Screening and data extraction

Studies were initially screened by title and abstract by 2 independent reviewers (XS and MAH). If the studies met the predetermined criteria, a full-text screening was performed to assess their eligibility. Any discrepancies were discussed between the 2 reviewers (XS and MAH), and a consensus was achieved with the involvement of a third reviewer (ZG). Data from eligible studies were independently extracted using Microsoft Excel (Version 16.44 for Mac; Microsoft, Redmond, WA, USA) by 2 reviewers, with discrepancies resolved through consultation with the third reviewer (ZG). Extracted data included: last name of first author, year of publication, country of experiment, sample size of each group, gender distribution and age (mean ± standard deviation (SD)) of participants, type and duration of intervention, and outcome measurements (Supplementary Table 4).

Mean change (SD_change) from baseline to post-intervention were the outcomes of interest and were extracted when reported. In cases where these values were not reported, baseline and post-intervention mean (± SD) data were extracted. Mean change was computed by subtracting baseline from the post-intervention value, and SD_change was determined by pooling the variances of the baseline and post-intervention SDs. When unreported, SD was calculated from standard errors, 95% confidence intervals (95%CIs), p values, or t statistics.

In cases where multiple follow-up time points were documented, we selectively extracted the initial post-intervention value. This decision was made to focus our analysis on the immediate effects of the intervention rather than the long-term outcomes. Intent-to-treat (ITT) analysis was prioritized when reported with per-protocol analyses to reflect the real-world scenario where not all participants adhere perfectly to the treatment.

2.4. Intervention categories

Lifestyle modification approaches were categorized into single-arm approaches (which included PA engagement, behavioral and educational intervention, and dietary intervention) and combined arms approaches, including bicomponent and multicomponent treatment.

2.5. Risk of bias (RoB)

Two reviewers (XS and MAH) evaluated the RoB using Version 2 of the Cochrane risk-of-bias tool (RoB 2).³¹ The assessment covered domains including random sequence generation, allocation concealment, blinding of participants, providers and outcome assessment, incomplete outcome data, and selective outcome reporting. Every domain was determined to be of high, moderate, or low RoB (Supplementary Table 5). Any disagreement was resolved by consensus.

2.6. Statistical analysis

We separately performed pairwise meta-analyses for each of the primary outcome measures using Comprehensive Meta-Analysis software (Version 4; Englewood, NJ, USA). Weighted mean difference (WMD) was used to establish the effect sizes of each lifestyle modification intervention compared to the usual care control. Cochrane Q-test and Higgins and Green's I² test were used to test heterogeneity. A p-value less than 0.05 for Cochrane Q-test or an I² statistic higher than 50% would indicate the presence of significant heterogeneity. Random-effect analysis would be applied when substantial heterogeneity existed among included studies.³² Trim and fill techniques³³ and Egger's regression test³⁴ were used to assess publication bias by evaluating funnel plot asymmetry. Copas selection model was applied to address publication bias and provide an adjusted estimate of the effect size.^35,36 Subgroup analyses were conducted to explore the influence of various factors on intervention effectiveness, including RCT design (cluster RCT vs. standard RCT), participants’ age group (preschooler vs. school-aged children), weight status (healthy, overweight, or obese), and intervention purpose (prevention vs. management). Participants under 6 years were considered preschool-age children, while those 6 years and older were categorized as school-aged children. Children's weight status was determined using the Centers for Disease Control and Prevention (CDC) BMI-for-age Growth Chart,³⁷ considering their reported baseline age and BMI. The subgroup analysis for RCT design was pre-specified to examine the impact of randomization types and delivery methods on intervention effectiveness,³⁸ while all other subgroup analyses were conducted post hoc. Additionally, meta-regressions were performed to assess the effects of baseline values, mean age, and intervention duration on each outcome variable.

We conducted NMAs to make further comparisons for lifestyle modification approaches that have not been compared directly in RCTs and to include RCTs with active controls. NMA was performed in RStudio (Version 4.1.2; Posit PBC, Boston, MA, USA) using the JAGS-based package “BUGSnet”.³⁹ Network plots were graphed using direct evidence present for each outcome. Bayesian NMA were implemented using the Markov chain Monte Carlo algorithm with 3 chains and 50,000 interactions per chain (burn-in period of 20,000). Trace plots were used for convergence assessment. Deviance information criterion (DIC) and residual deviance were used to determine model choice between fixed and random effects modeling (Supplementary Fig. 1). To assess inconsistency, we used the inconsistency model method and compared DICs and leverage plots between consistency models vs. those built assuming inconsistency. Furthermore, we compared each data point's posterior mean deviance contributions in both consistency and inconsistency models (Supplementary Fig. 2).⁴⁰ Separate NMAs were run by BMI, BMI z-score, and BFP. Descriptions of the network of studies were presented in a graphical format for each outcome. Surface under the cumulative ranking curve (SUCRA) plot, league tables, and forest plots were generated to show ranking probabilities and relative effects Fig. 1.

Fig. 1.

Preferred reporting items for systematic reviews and meta-analyses (PRISMA) flow chart. RCT = randomized controlled trial.

2.7. Equity, diversity, and inclusion statement

Our study included all identified RCTs of lifestyle modification interventions for the interested age-range of children, inclusive of all genders, races/ethnicities, and socioeconomic levels. Our author team included various genders and disciplines. Members of our author group are from different countries and are in various career stages.

3. Results

The final analysis included 91 RCTs, with aggregate data for 58,649 children. The mean age of the children included was 8.92 years, and 47.2% were girls. More specifics about the characteristics of all included RCTs can be found in Supplementary Table 4. Fig. 1 shows the PRISMA flow diagram for systematic reviews.⁴¹ For RoB assessment of all included studies, 32% had low risk, 39.2% had moderate risk, and 28.9% had high risk. Details of the assessment can be found in Supplementary Table 5. For inconsistency evaluation, we observed smaller DIC values for consistency models compared to models assuming inconsistency across all 3 outcomes in their respective leverage plots. In summary, our analysis reveals no compelling evidence of inconsistency within the network (Supplementary Fig. 2).

BMI emerged as the most frequently reported outcome, with evidence from 65 studies, 30 of which were conducted under a clustered RCT design. BMI z-score and BFP change were reported in 57 and 41 studies, respectively, with less than half utilizing clustered RCT designs for each outcome. The results of the network meta-analysis for the 3 outcomes were presented in a comparative effectiveness table (Table 1), featuring the mean deviation (MD) values with 95%CI for each treatment comparison. SUCRA ranking plots were employed to display the ranking probability of each treatment and forest plots were used to illustrate the MD of each treatment relative to the usual care condition.

Table 1.

Comparative analysis for the effectiveness of lifestyle intervention on BMI, BMI z-score, and BFP in MD (95%CI).

Variable	Summary results
	1.MC	2.Bi	3.PA	4.Inf	5.UC	6.N
BMI
1.MC	MC	0.20 (–0.21 to 0.62)	0.37 (–0.10 to 0.85)	0.37 (–0.06 to 0.83)	0.49 (0.12 to 0.88)*	0.59 (0.04 to 1.17)*
2.Bi	–0.20 (–0.62 to 0.21)	Bi	0.16 (–0.21 to 0.55)	0.17 (–0.22 to 0.57)	0.28 (0.04 to 0.54)*	0.38 (–0.06 to 0.87)
3.PA	–0.37 (–0.85 to 0.10)	–0.16 (–0.55 to 0.21)	PA	0.01 (–0.45 to 0.47)	0.12 (–0.20 to 0.44)	0.22 (–0.28 to 0.75)
4.Inf	–0.37 (–0.83 to 0.06)	–0.17 (–0.57 to 0.22)	–0.01 (–0.47 to 0.45)	Inf	0.11 (–0.28 to 0.52)	0.21(–0.34 to 0.80)
5.UC	–0.49 (–0.88 to –0.12)*	–0.28 (–0.54 to –0.04)*	–0.12 (–0.44 to 0.20)	–0.11 (–0.52 to 0.28)	UC	0.10 (–0.32 to 0.54)
6.N	–0.59 (–1.17 to –0.04)*	–0.38 (–0.87 to 0.06)	–0.22 (–0.75 to 0.28)	–0.21 (–0.80 to 0.34)	–0.10 (–0.54 to 0.32)	N
BMI z-score
	1.MC	2.Bi	6.N	3.PA	4.Inf.	5.UC
1.MC	MC	0.04 (–0.04 to 0.12)	0.06 (–0.07 to 0.20)	0.07 (–0.07 to 0.20)	0.09 (–0.01 to 0.18)	0.11 (0.04 to 0.18)*
2.Bi	–0.04 (–0.12 to 0.04)	Bi	0.02 (–0.10 to 0.14)	0.03 (–0.09 to 0.15)	0.05 (–0.03 to 0.12)	0.07 (0.02 to 0.12)*
6.N	–0.06 (–0.20 to 0.07)	–0.02 (–0.14 to 0.10)	N	0.01 (–0.16 to 0.18)	0.03 (–0.11 to 0.17)	0.05 (–0.06 to 0.17)
3.PA	–0.07 (–0.20 to 0.07)	–0.03 (–0.15 to 0.09)	–0.01 (–0.18 to 0.16)	PA	0.02 (–0.10 to 0.13)	0.04 (–0.08 to 0.17)
4.Inf	–0.09 (–0.18 to 0.01)	–0.05 (–0.12 to 0.03)	–0.03 (–0.17 to 0.11)	–0.02 (–0.13 to 0.10)	Inf	0.02 (–0.06 to 0.11)
5.UC	–0.11 (–0.18 to –0.04)*	–0.07 (–0.12 to –0.02)*	–0.05 (–0.17 to 0.06)	–0.04 (–0.17 to 0.08)	–0.02 (–0.11 to 0.06)	UC
BFP
	1.MC	2.Bi	4.Inf	3.PA	5.UC	6.N
1.MC	MC	1.03 (–0.46 to 2.46)	1.18 (–0.62 to 2.97)	1.46 (–0.44 to 3.29)	1.69 (0.42 to 2.97)*	2.30 (0.47 to 4.13)*
2.Bi	–1.03 (–2.46 to 0.46)	Bi	0.16 (–1.64 to 1.99)	0.43 (–1.14 to 1.99)	0.67 (–0.20 to 1.56)	1.27 (–0.07 to 2.67)
4.Inf	–1.18 (–2.97 to 0.62)	–0.16 (–1.99 to 1.64)	Inf	0.27 (–1.75 to 2.25)	0.51 (–1.22 to 2.24)	1.12 (–1.02 to 3.28)
3.PA	–1.46 (–3.29 to 0.44)	–0.43 (–1.99 to 1.14)	–0.27 (–2.25 to 1.75)	PA	0.24 (–1.20 to 1.72)	0.85 (–1.07 to 2.81)
5.UC	–1.69 (–2.97 to –0.42)*	–0.67 (–1.56 to 0.20)	–0.51 (–2.24 to 1.22)	–0.24 (–1.72 to 1.20)	UC	0.60 (–0.75 to 1.99)
6.N	–2.30 (–4.13 to –0.47)*	–1.27 (–2.67 to 0.07)	–1.12 (–3.28 to 1.02)	–0.85 (–2.81 to 1.07)	–0.60 (–1.99 to 0.75)	N

Notes: All references of included studies are provided in Supplementary File 1.

*Statistically significant at the 5% level, with a 95%CI excluding zero.

Abbreviations: 95%CI = 95% confidence interval; BFP, body fat percentage; Bi = Bicomponent intervention; BMI = body mass index; Inf = informational and behavioral intervention; MC = multiple component treatment; MD = mean deviation; N = nutrition and diet; PA = physical activity; UC = usual care.

3.1. Pairwise analysis

All outcomes, including BMI, BMI z-score, and BFP, showed significant reductions following lifestyle modifications in the pairwise meta-analysis (Supplementary Fig. 3). The reductions in BMI were notable, with an overall reduction of 0.12 (WMD = 0.120, 95%CI: 0.071–0.169, standard error (SE) = 0.025, p < 0.001, I² = 91.47%). Similarly, reductions were observed in BMI z-score (WMD = 0.081, 95%CI: 0.052–0.110, SE = 0.015, p < 0.001, I² = 95.37%) and BFP (WMD = 0.208, 95%CI: 0.064–0.352, SE = 0.073, p = 0.005, I² = 96.69%) overall.

The trim and fill technique applied to the funnel plots revealed minimal risk of publication bias for BMI and BMI z-score, whereas Egger's regression test indicated significant publication bias for these 2 outcomes (BMI: β = 1.362, p = 0.001; BMI z-score: β = 2.580, p = 0.001). Meanwhile, no evidence of publication bias was identified for BFP using either assessment approach (β = 1.709; p = 0.059) (Supplementary Fig. 4). The Copas selection model was applied to estimate the effect sizes for BMI and BMI z-score, accounting for potential publication bias. The results showed that the adjusted effect sizes were slightly smaller but still marginally significant for both variables (BMI: WMD = 0.100, 95%CI: 0.001‒0.199, SE = 0.051; BMI z-score: WMD = 0.080, 95%CI: 0.050‒0.111, SE = 0.016). High levels of heterogeneity were observed across individual studies for all outcomes. Meta-regression and subgroup analyses were carried out to further explore the heterogeneity.

3.2. Subgroup analyses

For our primary outcome, BMI, subgroup analyses were conducted based on RCT design, participants’ age group, weight status, and intervention purpose (Table 2). The subgroup analyses highlighted significant differences between groups within each category, contributing to an understanding of the high heterogeneity in the overall BMI dataset. However, certain subgroups, like school-age children, continued to exhibit unexplained high heterogeneity. To address this, we conducted meta-regression for mean age, the results of which will be outlined in the subsequent results section.

Table 2.

Subgroup analyses results for BMI.

Subgroup	WMD	SE	95%CI	p	Q-stat	pQ	I2	pbetween group
RCT design
Standard RCT	0.478	0.016	0.234 to 0.723	<0.001	149.540	<0.001	75.40
Cluster RCT	0.065	0.026	0.014 to 0.116	0.012	550.806		94.83	0.001
Age group
Preschooler	0.021	0.044	–0.066 to 0.108	0.640	7.385	0.689	0.000
School-age	0.147	0.029	0.091 to 0.204	<0.001	737.228	<0.001	92.95	0.017
Body weight status
Healthy	0.073	0.046	–0.017 to 0.163	0.110	191.495	<0.001	85.90
Obesity	0.783	0.183	0.424 to 1.143	<0.001	110.564	<0.001	80.10
Overweight	–0.020	0.003	–0.026 to –0.013	<0.001	11.654	0.474	0.000	<0.001
Intervention purpose
Management	0.506	0.110	0.290 to 0.722	<0.001	184.539	<0.001	81.03
Prevention	0.073	0.046	–0.017 to 0.163	0.110	191.495	<0.001	85.90	<0.001

Abbreviations: 95%CI = 95% confidence interval; BMI = body mass index; p_Q = p value for the Q statistic; Q-stat = Q statistic value; RCT = randomized controlled trial; SE = standard error; WMD = weighted mean difference.

Subgroup differences in BMI z-score and BFP were analyzed for RCT design. Standard RCTs exhibited greater reductions (BMI z-score: WMD = 0.103, SE = 0.016, 95%CI: 0.071‒0.134, p < 0.001, I² = 69.46%; BFP: WMD = 1.318, SE = 0.339, 95%CI: 0.653‒1.982, p < 0.001, I² = 88.22%) compared to clustered RCTs (BMI z-score: WMD = 0.063; SE = 0.020; 95%CI: 0.023–0.103; p = 0.002, I² = 98.09%; BFP: WMD = –0.007; SE = 0.084; 95%CI = –0.171 to 0.157; p = 0.933, I² = 98.72%). Overall, standard RCTs demonstrated larger effect sizes and lower heterogeneity across all outcomes (Supplementary Fig. 5). Subgroup differences in BFP were analyzed based on age group. However, BMI z-score was not analyzed in this context as it had already been adjusted for children's age. The subgroup analysis for different age groups did not yield any significant conclusions regarding BFP (Supplementary Table 6).

3.3. Meta-regression

We conducted meta-regression analyses to explore whether baseline value, mean age, and intervention duration were associated with the effectiveness of lifestyle interventions across all 3 outcomes (Supplementary Fig. 6). In the case of BMI, higher baseline BMI and mean age showed significant associations with a larger effect size of lifestyle intervention (p < 0.001, p = 0.044, respectively) (Fig. 2). No significant relationship was found between intervention duration and effectiveness for BMI reduction (Table 3). Higher baseline values were also found to have a significantly positive relationship with intervention effects for BFP (p < 0.001). None of the other regression analyses revealed significant relationships for BMI z-score and BFP (Supplementary Table 7). Meta-regression results indicated that children with a higher baseline BMI and BFP may exhibit a greater response to lifestyle modifications for obesity management.

Fig. 2.

Meta-regression plot for (A) baseline BMI values and (B) mean age. BMI = body mass index.

Table 3.

Meta-regression analysis results for BMI.

Covariate	Coefficient	SE	95%CI	z	p
Baseline BMI
Intercept	–0.972	0.183	–1.331 to –0.612	–5.30	<0.001
Baseline value	0.059	0.010	0.040 to 0.078	6.08	<0.001
Mean age
Intercept	–0.069	0.117	–0.298 to 0.160	–0.59	0.555
Age	0.027	0.013	0.001 to 0.053	2.01	0.044
Intervention duration
Intercept	0.115	0.067	–0.016 to 0.246	1.72	0.085
Duration	0.004	0.006	–0.007 to 0.015	0.68	0.450

Abbreviations: 95%CI = 95% confidence interval; BMI = body mass index; SE = standard error.

3.4. NMA

In Fig. 3, the dimensions of nodes and edges in the network graph are proportionally adjusted to represent the number of studies investigating a specific treatment and the number of direct comparisons between any 2 interventions, respectively. Across all 3 outcomes, the treatment arms most frequently represented were usual care and bicomponent lifestyle modification, demonstrating the most extensive direct comparison evidence between these 2 interventions as well. Among the 3 outcomes, BMI yielded the most intricate network diagram. None of the outcomes featured a direct comparison between multicomponent intervention and dietary intervention from any of the trials included in this analysis.

Fig. 3.

Network diagram for (A) BMI, (B) BMI z-score, and (C) BFP. BFP = body fat percentage; Bi = bicomponent intervention; BMI = body mass index; Inf = informational and behavioral intervention; MC = multiple component treatment; N = nutrition and diet; PA = physical activity; UC = usual care.

The SUCRA plot showed the probability of the ranking of each treatment (Fig. 4). The multicomponent intervention curve was consistently above the other curves for all 3 outcomes, suggesting that it is the most effective treatment among the 6 interventions based on the outcomes investigated. The bicomponent intervention demonstrated the second-highest effectiveness across the outcomes. In the context of single-arm treatment, PA attained the highest average ranking when considering performance across all 3 measures. This is attributed to the distinctiveness of the most effective single-arm intervention for each outcome. Specifically, PA emerged slightly superior to behavioral intervention as the most effective single-arm intervention for BMI. Meanwhile, for BMI z-score and BFP, dietary intervention and behavioral intervention proved to be the most effective single-arm interventions, respectively.

Fig. 4.

SUCRA plot for (A) BMI, (B) BMI z-score, and (C) BFP. X-axis: ranking of treatment; Y-axis: probability of a given treatment to be ranked 1st to last. BFP = body fat percentage; Bi = bicomponent intervention; BMI = body mass index; Inf = informational and behavioral intervention; MC = multiple component treatment; N = nutrition and diet; PA = physical activity; SUCRA = surface under the cumulative ranking curve; UC = usual care.

Table 1 displayed the effect estimates and 95%CIs derived from the NMA. The results, presented as MD with 95%CI, highlighted significant differences indicated by darker backgrounds. The difference between the multicomponent intervention and the usual care conditions were all statistically significant at the 95% level for all outcomes. Meanwhile, the bicomponent intervention exhibited a significant reduction in BMI and BMI z-score. Additionally, the multicomponent intervention demonstrated significance of dietary intervention for BMI and BFP, which relied on indirect evidence given the absence of direct comparisons between the 2 approaches as shown in the network diagram (Fig. 3). The forest plot (Fig. 5) visually represented the comparative effectiveness of usual care in relation to all other treatments. The plot highlighted the significant reductions achieved by both the multicomponent and bicomponent interventions.

Fig. 5.

Forest plots for (A) BMI, (B) BMI z-score, and (C) BFP. X-axis: mean differences relative to UC; Y-axis: Treatment (e.g., lifestyle intervention). 95%CI = 95% confidence interval; BFP = body fat percentage; Bi = Bicomponent intervention; BMI = body mass index; Inf = informational and behavioral intervention; MC = multiple component treatment; N = nutrition and diet; PA = physical activity; UC = usual care.

4. Discussion

This study aimed to compare the effectiveness of various lifestyle interventions on body composition in preschool and school-aged children using the most recent data published within the past decade. Our findings indicated that interventions incorporating multiple components were more effective in reducing children's BMI, BMI z-score, and BFP compared to single-arm interventions. Among single-component approaches, PA engagement emerged as most effective for BMI reduction and second-best for reducing BMI z-score and BFP. These findings diverged from a previously published analysis suggesting that interventions with a clear focus outperformed combined component interventions.²² In the analysis by Bae and Lee,²² which involved 24 RCTs, exercise (without parental involvement) was identified as the most effective approach, followed by diet intervention (with parental involvement), and a combination of exercise and diet was ranked third. Discrepancies in findings were likely due to differences in search strategy, inclusion criteria, and the analysis framework. Compared to the frequentist NMA employed in their analysis, Bayesian NMA offers greater flexibility in addressing detected heterogeneity,⁴² which was present in both the current study and theirs. On the other hand, our findings aligned with those of a previous NMA with an expanded evidence base, which similarly found that combined approaches were more effective than single-target interventions.²³ Specifically, their study concluded that face-to-face-delivered PA and dietary combined intervention ranked highest, followed by multi-lifestyle intervention delivered through face-to-face and mobile health methods. The slight variations in the rankings of bicomponent and multicomponent interventions may be attributed to differences in the characteristics of the study populations and intervention variability. Notably, their study included adolescents, whereas ours specifically focused on preschool-aged children and middle childhood before puberty onset. The primary aim of their study was to compare different delivery methods, and their findings indicated that face-to-face interventions yielded superior effects compared to those delivered via mobile health platforms. Despite the growing accessibility of telehealth, the quality of intervention delivery was observed to be less effective than in-person delivery. Consequently, our analysis exclusively focused on face-to-face interventions to emphasize intervention impacts, while acknowledging their study's broader comparison of delivery methods. We consider these results promising and robust, with significant potential to have clinically meaningful implications. Essentially, this discovery offers valuable insights into the management of children's body composition. If feasible, addressing various facets of children's lifestyle patterns and living environment to enhance awareness of healthy lifestyle, encourage PA, and promote healthy eating may yield a substantial impact. However, if simultaneous implementation of multiple intervention components proves challenging, prioritizing the promotion of PA engagement potentially remains a highly effective approach.

4.1. Strengths and limitations

This study contributes valuable evidence to the application of NMA in the context of pediatric PA and health promotion by showcasing its feasibility and acceptability. Notably, previous NMA studies in the realm of obesity treatment primarily concentrated on pharmaceutical therapies, comparing the effectiveness of various drugs in reducing weight; they also involved different, predominantly adult, populations.43, 44, 45, 46, 47 We advocate for a shift in focus towards early-stage interventions and lifestyle modifications in childhood, emphasizing the importance of such approaches over medical treatments. The strengths of our study lie in its substantial sample size, rigorous objective data analysis, and adept use of indirect evidence in the absence of direct comparisons. However, it is important to acknowledge a few limitations in this study and discuss some directions and suggestions for future studies to advance research in the field: (a) While our exclusion criteria ensured a minimum intervention duration of 6 weeks to allow for adequate intervention time, this analysis was constrained by the absence of other pertinent intervention-related factors. Details such as the combined effects of intervention frequency and duration, the allocated time for each intervention session, and participants’ engagement levels during the intervention were not thoroughly quantified. These aspects are pivotal for accurately evaluating intervention impact delivery. However, the diverse array of intervention designs, particularly those employing multi-component approaches, poses challenges to devising a standardized method for quantifying specific intervention intensities. Furthermore, certain studies lacked information necessary for retrieving additional details regarding intervention design and procedures.⁴⁸ It is worth noting that children's engagement levels were rarely measured and reported in the included studies. Moving forward, it's essential for future clinical studies to incorporate this parameter as a crucial measurement of intervention efficacy and participant involvement. (b) Our concentration on short-term effects stems from our belief that this initial assessment is crucial for evaluating different intervention approaches. Limited short-term effectiveness suggests reduced potential for long-term impact,⁴⁹ emphasizing the importance of identifying significant short-term effects to inform future investigations. Incorporating follow-up results was further complicated by the varied assessment intervals and the challenge of establishing a comparable cut-off for defining “long-term” across diverse datasets.50, 51, 52, 53 However, future studies that prioritize long-term effects will undoubtedly contribute meaningfully to the field. (c) Due to the considerable variability between studies, we observed significant heterogeneity in most of our analyses. Furthermore, we detected significant publication bias for overall BMI and BMI z-score. Our assessment, illustrated through funnel plots adjusted using the trim and fill method, along with results from the Copas selection model, suggested that the effect size for these outcomes may be skewed, potentially leading to an overestimation of the true intervention effectiveness. (d) Our analysis did not differentiate specific components for multicomponent and bicomponent interventions. Consequently, while these interventions have shown effectiveness, determining the most influential component contributing to the final effect becomes challenging. Future studies could address this limitation by specifically targeting multi-component interventions and bicomponent interventions, further categorizing them into distinct treatment groups (e.g., PA + diet, diet + behavioral, PA + diet + behavioral, etc.). This approach would enable a more nuanced understanding of their effects through different combinations. (e) We chose to include preschool-aged children to facilitate early intervention in obesity prevention and management. A critical period in obesity development is adiposity rebound, typically occurring between 5–7 years of age.⁵⁴ Early adiposity rebound has been identified as a predictive marker of obesity in later childhood, adolescence, and adulthood.^55,56 Although the moderating effect of adiposity rebound may complicate the interpretation of intervention effects, we believe it strengthens the external validity of the interventions. This is because accurately determining the timing of adiposity rebound for each child is challenging without sufficient longitudinal data with frequent measurements.⁵⁷ This scenario is prevalent in most studies included in our analysis focusing on this age group. (f) While numerous studies have demonstrated correlations between BMI and body fat, as well as future health risks for children and adolescents, it remains a proxy for body composition rather than a direct measurement. Future research could enhance the comprehensiveness of body composition assessment by incorporating additional parameters, such as fat mass index and fat-free mass index, to evaluate adiposity levels in children more accurately.⁵⁸ (g) Lastly, it is essential to consider the categorization definition of interventions when comparing results across different NMA studies, as this can be a relatively subjective process. For example, in our analysis, dietary intervention encompassed interventions providing low-calorie food or unsweetened drinks as well as programs focused on educating participants about nutrition and promoting healthy eating habits and cooking practices. It is noteworthy that other analyses may adopt different definitions for diet interventions (e.g., only including specific meal plans or substitutes¹³); therefore, we need to take caution when interpreting results based on treatment category.

4.2. Future directions and recommendations

To enhance the convenience of conducting NMA in the future, studies can adopt certain practices to streamline the process. Firstly, researchers can contribute to the ease of NMA by reporting data in a standardized format, including both adjusted and unadjusted original data for pre- and post-intervention (e.g., mean value and SD) along with calculated data within and between groups (e.g., mean difference and SE). This standardized reporting approach will significantly facilitate data pre-processing for subsequent secondary analyses, including NMA. Secondly, current meta-analyses in lifestyle modification for obesity management primarily rely on measures like BMI, BMI z-score, and BFP for comparability. It would be beneficial to establish standardized quantifications for other health-related outcomes, especially regarding moderate-to-vigorous intensity PA. This is crucial because interventions may not always demonstrate significance for BMI or BFP but could show a notable increase in PA levels.⁵⁹ PA intensity is measured through various methods including oxygen consumption, metabolic equivalent of task (MET), step counts, and activity trackers. However, there's a lack of consensus on cut points for each measurement to determine intensity levels across studies.^60,61 Compared to other parameters, METs offer relatively stable thresholds for intensity level categorization.⁶² The Youth Compendium of Physical Activities provides comprehensive MET values for approximately 200 activities across 4 age groups, offering a standardized method for scoring and interpreting PA data in children and adolescents.⁶³ While METs can help quantify PA intensity comparably among youth, it is essential to acknowledge their limitations. These include potential misclassification due to factors such as fitness level, body weight, and sex not being considered when using absolute thresholds.⁶¹ Lastly, while usual care served as the most common control condition in all studies, intentionally incorporating active control groups or alternatives to usual care where applicable and with sufficient participants would contribute to more comprehensive results. This is beneficial not only for individual studies, but it also provides more head-to-head comparisons for future NMAs.^64,65 Given the rapid growth of NMA studies in recent years,^21,66 there is reason to believe that this approach will continue to be promising, showing significance in advancing practical analysis methods across various health-related fields, including the realm of PA and health.

5. Conclusion

Our findings highlighted the effectiveness of multi-component approaches integrating PA, diet, behavioral therapy, and/or informational support over single-component approaches for lifestyle interventions. PA engagement emerges as particularly effective among single-arm approaches. Moreover, children with higher baseline BMI and BFP responded more significantly to lifestyle modifications for obesity management, which means these approaches offer promising prospects for this population.

Authors' contributions

ZG conceived of the presented idea and supervised the project; XS contributed to the conception of the idea, performed the database search, screened articles, extracted data, assessed evidence, contributed to model building, and analyzed the data; MAH assisted with the database search, article screening, data extraction, and evidence assessment; HK contributed to model building and data analysis. All authors have read and approved the final version of the manuscript, and agree with the order of presentation of the authors.

Competing interests

The authors declare that they have no competing interests.