Background

Overweight and obese children persist as a serious health problem and a public challenge of the twenty-first century. Obesity among children and adolescents is a leading cause of health and contributes to cardiovascular disease, cerebrovascular disease, and metabolic diseases [1]. Nearly one in five children and adolescents are overweight or obese [2], and the growing prevalence of obesity in youth has led to an alarming increase of 18.5% in children and adolescents between the ages of 2–19 years [3]. Obese children are at greater risk of obesity in adulthood; a recent study of 200,777 participants showed that 80% of teens with obesity remained obese in adulthood and this continued with a prevalence of 70% past the age of 30 [4]. According to a recent study in the United States comparing the cost–benefit of prevention versus treatment interventions in youth, preventive interventions in the early stages of life were found to be more beneficial than in adulthood, and addressing childhood obesity as early as possible is an effective strategy against obesity in later ages [5]. Although the underlying reasons of genetics and individual behavior for being overweight in adults and young people are almost the same [6], obesity prevention policies in the younger age group are different from those adopted in adulthood. Developing and implementing effective strategies to prevent childhood obesity is difficult at the population level. The National Academy of Sciences recommended that more attention should be paid to providing opportunities to choose healthy foods in society [7]. Obesity prevention is a public health priority around the world. The effectiveness of childhood obesity prevention programs has been shown by previous Cochrane reviews [8]. Some previous systematic reviews have focused on childhood obesity prevention programs that were not at national, governmental or macro-population level policies or that focused on some specific interventional approaches, including changes in physical activity (PA), diet and education [9,10,11,12,13]. Although there is evidence to support the beneficial effects of increased PA and diet as a basic and early strategy at any time and for any age against obesity [14, 15], no summarized study is available to critically evaluate the effectiveness of different policies with different interventional approaches in prevention of childhood obesity considering the role of setting, age, geographical distribution, and intervention type or strategy. Therefore, the aim of the current study was to systematically search controlled trials that evaluated the effectiveness of pediatric obesity prevention policies among children and adolescents and to analyze the effectiveness of these policies on the study outcomes of body mass index (BMI) and BMI-Zscore (BMI-Z) measurements while considering a possible dose–response association with preventive tools.

Methods and materials

The current systematic review and meta-analysis was prepared according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement for reporting systematic reviews and meta-analyses [16] (checklist is provided in Additional file 1: Table S1). The study protocol was registered in PROSPERO (identifier: CRD42019138359) and was approved by the Research Undersecretary of the Tabriz University of Medical Sciences as the Ph.D. thesis of SHT (Registration number: IR.TBZMED.REC.1398.840).

Data sources and search strategy

Searches were conducted using SCOPUS, PubMed and Embase. All articles were considered eligible, if published before April 6, 2020. Additional file 1: Table S2 shows the full search strategy in PubMed. Four concept groups were organized according to the search terms: (a) Population (pediatric, children, or adolescents); (b) Health problem under consideration (obesity, pediatric obesity); (c) Intervention (policy, program, strategy); and (d) Relevant outcomes of interest (BMI, BMI-Zscore). The reference lists of all related and available articles were reviewed to reduce the possibility of missing articles. The selection criteria for this review were independently verified by two researchers (SHT, MAF).

Study selection

Relevant studies conducting a community approach that evaluated policies to prevent obesity in children and adolescents aged 0–18 years were included in the current review. Studies were excluded if they were aimed to treat childhood overweight/obesity), were performed in children with other diseases, or if their full text was not available. Detailed exclusion and inclusion criteria are shown in Table 1.

Table 1 Inclusion and exclusion criteria for study selection

Quality assessment and data extraction

Study quality was assessed using the Effective Public Health Practice Project Quality Assessment Tool for Quantitative Studies, a useful tool for quality assessment of randomized and non-randomized intervention trials [17, 18]. This tool is comprised of six components that include selection bias, study design, confounders, blinding, data collection methods (validity/reliability), and withdrawals and dropouts. The overall quality rating and the components are scored as strong, moderate and weak according to the tool’s instructions. Individual component quality rankings are shown in Additional file 1: Table S3. General study characteristics (author, year of publication, country, sample size, number of intervention and control, type of study (randomized or non-randomized), duration of intervention, follow-up from baseline, follow-up from end of intervention, participant characteristics, outcomes (BMI, BMI-Zscore), and policy characteristics were extracted for included studies. Effect size was defined as changes in BMI and BMI-Zscore compared with control group. Two researchers (SHT, MAF) independently extracted the data from all studies.

Statistical analysis

The data were analyzed using STATA version 15 (STATA Corp, College Station, TX, USA), and p-values of less than 0.05 were considered statistically significant.

Two-class meta-analysis of continuous variable

The studies that reported BMI and BMI-Zscore as primary or secondary outcomes in intervention and control groups were included for two-class meta-analysis synthesis. The means and standard deviations (SD) of variables were used to compute standardized mean differences as effect size computed by pooled estimate of weighted mean difference (WMD) at a 95% confidence interval (CI). Subgroup analyses were conducted to explore sources of heterogeneity. Due to high heterogeneity values (i.e., above 50%), the random effects model was used. Between-study heterogeneity was identified using Cochran's Q and I-squared tests as follows: I2 < 25%, no heterogeneity; I2 25% to 50%, moderate heterogeneity; I2 > 50%, large heterogeneity [19]. Studies that reported separate results for both sexes, in different age categories, or at different time periods of follow-up were included as individual studies. Publication bias was examined using Begg’s funnel plots, followed by Egger's regression asymmetry test and Begg's rank correlation for formal statistical assessment of funnel plot asymmetry. For missing SDs, the method described by Walter and Yao was used to calculate SD [20]. Studies were excluded from the analysis if they (a) were not controlled trials or (b) did not report sufficient data of outcome variables.

Dose–response meta-analysis of continuous variables

For dose–response meta-analysis of variables, variables of duration of intervention and PA time and training sessions (as education time) were included. The mean difference of variables in each study was also identified. A dose–response meta-analysis of BMI and BMI-Zscore was performed using fractional polynomial modeling [21] to explore nonlinear potential effects of duration of intervention (year), PA and education time and study-specific parameters.

Results

Literature search and study characteristics

A search of electronic data bases retrieved 30,719 records. After removing duplicates, 20,686 items were screened by title/abstract (Fig. 1) and selected according the criteria identified above. The remaining 224 full text articles were screened and 49 publications were selected in a qualitative synthesis; finally, 38 publications were included in a quantitative synthesis, which contained outcomes for 64 individual studies as described above.

Fig. 1
figure 1

Flow chart of study selection

Grey literature searches identified no published results for policies in scope. Study, participant, and program characteristics of the quantitative synthesis (meta-analysis) are presented in Table 2 with additional information including the full name of the studies shown in Additional file 1: Table S4. Studies were performed in various settings of school (n = 16) [22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37], community and school (n = 10) [38,39,40,41,42,43,44,45,46,47], school and home (n = 1) [48], community, school, and home (n = 2) [49, 50], community, school, home, and primary care clinic (n = 5) [51,52,53,54,55], community and home (n = 2) [56, 57], primary care clinic (n = 1) [58], and cyberspace/online (n = 1) [59]. In all, 64 individual studies were obtained from 38 publications included in the quantitative synthesis. Twelve studies were performed as combinations of different follow-up times, age groups, genders, or different durations or populations; therefore each was included as two [23,24,25, 31, 35, 36, 42, 44, 48, 50, 54,55,56, 59], three [41, 49, 52], or four individual studies [30, 51]. The rationale for extracting several studies from these publications and additional information about the policies are shown in Table 2 and Additional file 1: Table S5). Characteristics of studies that were not included in the meta-analysis with the exclusion reasons are shown in Additional file 1: Table S6.

Table 2 The general characteristics of the studies included in the meta-analysis of the association between childhood obesity prevention policies and Body mass index (BMI) and BMI-Zscore

Approximately 35% of programs were carried out in the United States (n = 13) [29, 31,32,33,34, 37, 40, 42, 49,50,51,52, 57], and 31% (n = 12) studies in Australia [24,25,26,27,28, 38, 43, 44, 47, 53, 54, 59]. Other studies took place in China (n = 1) [22], Brazil (n = 1) [23], New Zealand (n = 3) [30, 35, 55], Spain (n = 2) [36, 39], the United Kingdom (n = 1) [41], Fiji (n = 1) [45], Tonga (n = 1) [46], France (n = 1) [48], Sweden (n = 1) [58], and one study which was conducted in eight European countries (Belgium, Cyprus, Estonia, Germany, Hungary, Italy, Spain and Sweden) [56].

Thirty studies reported BMI [22,23,24,25,26,27,28,29,30,31, 33,34,35,36, 38, 39, 43,44,45,46,47,48, 50, 52,53,54,55, 57,58,59] and 27 studies reported BMI-Zscore [22,23,24,25, 27, 28, 32, 33, 35, 37, 39,40,41,42,43,44,45,46,47, 49, 51,52,53,54,55,56,57]. The total number of participants in the systematic reviews was 200,255; 178,017 participants were included in the meta-analysis, ranging from 86 [59] to 35,157 [54], with an average sample size of 2849. Nine studies were carried out among girls, [23, 27, 30, 31, 36, 42, 48, 51, 56], eight studies among boys [24, 26, 30, 36, 42, 48, 51, 56] and 21 studies were performed with both genders. The majority of policies (n = 33) examined combined diet and PA interventions, with five studies that consisted of only PA [22, 26, 30, 34, 36] and no study focused only on diet. The majority of studies (n = 31) were conducted as randomized controlled trials (81.5%), and seven [35, 47, 48, 51, 55, 56, 58] were non-randomized controlled trials (18.4%). BMI or BMI-Zscore as outcomes were reported at the end of the intervention in 31 studies [22,23,24,25,26,27, 29,30,31,32,33,34,35,36,37,38,39,40, 43,44,45,46,47,48,49,50,51,52,53, 55, 57], and 14 programs had follow-up periods after the end of the intervention [23, 24, 26, 28, 31, 35, 41, 42, 49, 52, 54, 56, 58, 59]. The length of follow-up ranged from 6 weeks [52] to 3 years [54].

Dose–response meta-analysis of the association between education time, PA, duration of intervention and BMI or BMI-Zscore

The non-linear dose–response association between the study outcomes of BMI or BMI-Zscore and education time, PA, and duration of intervention was performed using fractional polynomial (FP) modelling. Thirteen studies were assessed for a dose–response association between BMI and education time [23,24,25,26,27, 29,30,31, 33, 37, 39, 52, 57], and 12 studies for BMI-Zscore and education time [23,24,25, 27, 33, 37, 39, 41, 49, 51, 52, 57] (Figs. 2a, 3a). There was no evidence for nonlinear association between BMI (P- for nonlinearity = 0.163) or BMI-Zscore (P- for nonlinearity = 0.270) with education time. Ten studies were assessed for a dose–response association between BMI and PA [24,25,26,27, 30, 31, 33, 34, 36, 52] and 8 studies for BMI-Zscore [24, 25, 27, 33, 41, 42, 51, 52] (Figs. 2b, 3b). No evidence of nonlinearity association was observed between BMI (P- for nonlinearity = 0.254) or BMI-Zscore (P- for nonlinearity = 0.452) and PA. All 30 studies of BMI and 27 studies of BMI-Zscore were included for calculating the dose–response association between changes in BMI or BMI-Zscore with duration of intervention, respectively (Figs. 2c, 3c). There was evidence of a nonlinear association between the duration of intervention and BMI (P- for nonlinearity < 0.001) as well as BMI-Zscore (P- for nonlinearity = 0.023).

Fig. 2
figure 2

Dose–response association between duration of intervention, PA, education time and body mass index (BMI). Linear relation (solid line) and 95% confidence interval (CI) (gray area) of mean difference in BMI. This figure indicates the association between mean difference of BMI and a education time, b PA, c duration of intervention

Fig. 3
figure 3

Dose–response association between duration of intervention, PA, education time and BMI-Zscore. Linear relation (solid line) and 95% confidence interval (CI) (gray area) of mean difference in BMI-Z. This figure indicates the association between mean difference of BMI-Z and a education time, b PA, c duration of intervention

Details of the dose–response association between duration of intervention, PA, education time and BMI and BMI-Zscore are shown in Table 3.

Table 3 Details of non-linear association between BMI and BMI-Zscore with study specific parameters

Two-class meta-analysis of the comparison of effectiveness of childhood obesity prevention policies on BMI and BMI-Zscore

A total of 38 publications [22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59] were included in the two-class meta-analysis of the effects of obesity prevention policies on BMI (Fig. 4) and BMI-Zscore (Fig. 5).

Fig. 4
figure 4

The forest plot showing the weighted mean difference (WMD) of the effect of childhood obesity prevention policies on body mass index (BMI) [weighted mean difference (WMD): − 0.127; confidence interval (CI) − 0.198, − 0.056; P < 0.001]

Fig. 5
figure 5

The forest plot showing the weighted mean difference (WMD) of the effect of childhood obesity prevention policies on body mass index Z score (BMI-Zscore) [weighted mean difference (WMD): − 0.020; confidence interval (CI) − 0.061, − 0.021; P = 0.340]

The results showed that obesity-prevention policies had a significant effect in reducing BMI (WMD: − 0.127; CI − 0.198, − 0.056; P < 0.001; I2 = 99.7%; P-heterogeneity < 0.001) and a non-significant reduction in BMI-Zscore (WMD, − 0.020; CI − 0.061, − 0.021; P = 0.340; I2 = 99.8). A subgrouping meta-analysis (shown in Tables 4 and 5) and a meta-regression (Table 6) were also performed to assess the source of heterogeneity for the included studies. According to the subgroup meta-analysis, school-based policies in children aged 5–10 years, in relatively short period of time (less or equal to 2 years), with approaches to practical changes in diet and PA (i.e., not consisting of education only) and the policies that were performed in combination with both genders seemed to be more effective in reducing BMI and BMI-Zscore with more favorable changes. Subgrouping also revealed that the heterogeneity level for BMI was reduced in subgrouping according to target group (e.g., for the parent group it was reduced from 99.7 to 49.8%), type of intervention (e.g., for only education it was reduced from 99.7 to 30.9%), study focus (e.g., for PA it was reduced from 99.7 to 35.7%), and frequency of intervention (e.g., for monthly it was reduced from 99.7 to 13.4%). In examining setting, the setting of community, school, and home and school, home and cyberspace and continent as US, the frequency of intervention as weekly, baseline BMI as a range of 22–25 and ≥ 25 kg/m2, and gender as male, heterogeneity disappeared. For BMI-Zscore, the target group, the continent, the gender, and the setting were the primary sources of heterogeneity.

Table 4 Results of subgroup analysis for the effects of childhood obesity policies on childhood BMI
Table 5 Results of subgroup analyses for the effects of childhood obesity policies on childhood BMI-Zscore
Table 6 Meta regression analysis for in obesity prevention policies on BMI and BMI-Zscore

Quality assessment of included studies

The Effective Public Health Practice Project Quality Assessment Tool for Quantitative Studies was used for quality assessment of the studies. Study quality [17, 18] was evaluated as “weak” for 15 studies [22, 24, 27,28,29,30,31, 34, 42, 43, 48, 52, 54,55,56], “moderate” for 10 studies [25, 26, 36, 37, 40, 44, 45, 47, 50, 51], and “strong” for 13 studies [23, 32, 33, 35, 38, 39, 41, 46, 49, 53, 57,58,59]. Quality assessment results also showed that the average change in BMI or BMI-Zscore in the follow-up compared to baseline was 0.5401 and − 0.0054 in the intervention groups and 0.7291 and 0.5401 in the control groups (Additional file 1: Table S3).

Publication bias

Publication bias was determined using the funnel plot of BMI and BMI-Zscore (Additional file 1: Figure S1). Begg's and Egger's regression tests were used to further illustrate publication bias (Additional file 1: Table S7). No evidence of publication bias was seen for BMI in Begg's (P = 0.08) or Egger's regression tests (P = 0.54) or for BMI-Zscore in Begg's (P = 0.89) or Egger's regression test (P = 0.65).

Sensitivity analysis

A sensitivity analysis was performed to obtain the effects of individual studies on the BMI-Zscore results and the results of the sensitivity analysis is presented as a plot in Additional file 1: Figure S2. By removing the studies of Kremer et al. [45] and de Silva-Sanigorsk et al. [54] a significant change in the results occurred (WMD: − 0.036; CI − 0.068, − 0.005; P = 0.025; I2 = 72.4; P < 0.005). When Sadeghi et al. [42] among boys was also removed, the changes were even more pronounced (WMD: − 0.042; CI − 0.073, − 0.010; P = 0.009; I2 = 71.5; P < 0.001).

Discussion

This systematic review and meta-analysis is the first, to our knowledge, to evaluate the quantitative effects of various childhood obesity prevention policies on children's BMI and BMI-Zscore in an interventional design. There are many systematic reviews or meta-analysis studies that have been performed in specific settings such as schools only [12, 13, 60] or were performed for single-axis interventions such as physical activity only [10, 61], diet only [13] or with limited duration of intervention [62] or follow-up [63, 64] and different age ranges [9, 10, 60, 64]. The current comprehensive meta-analysis evaluated the isolated effects of settings, intervention materials, duration and length of follow up, with a focus on the adiposity-related outcome of BMI or BMI-Zscore. The key findings of the current study were as follows. First, obesity prevention policies were associated with 0.127 kg/m2 reduction in BMI but with no significant change in BMI-Zscore. Second, there was a nonlinear dose–response association between duration of intervention and reduction in BMI and BMI-Zscore in studies with duration of intervention of ≤ 2 years.

In a meta-analysis by Stice et al. [65], no statistically significant effects on prevention or treatment of obesity were reported in a large percentage of studies (79%). In the current meta-analysis childhood obesity prevention policies were associated with 0.127 kg/m2 decrease in BMI. This BMI reduction due to weight control programs in the present study was similar to Peirson et al. [63], who assessed 76 studies for normal, overweight and obese children. In contrast in a study by Harris et al., in a systematic review of 18 interventions studies, no significant effects on BMI were found [61]. Another finding in the current study was a small but non-significant change in BMI-Zscore in intervention groups (e.g., 0.0054 units’ reduction of BMI-Zscore in the intervention vs 0.5401 units’ increase in the control). On the other hand, Peirson et al. [63] found a significant reduction in BMI-Zscore in their study. These inconsistencies might be due to differences in inclusion criteria. A nonlinear dose–response association between the duration of intervention (less than 2 years) and decrease in BMI and BMI-Zscore indicated long-term duration of intervention reduces the efficacy of weight management policies. As shown in Fig. 2c, for interventions longer than 2 years, the increase in intervention time reduced the mean change in BMI between the intervention and control groups. Consistent with our findings, Stice et al. also found that the weight reducing effects of weight management programs disappeared after a 3-year follow-up, suggesting that short-term obesity prevention programs are more effective than long-term ones in obesity management [65]. These findings were not similar for adults; for example, in a study of adults with an intervention duration that ranged from 6 weeks to 2 years, it was reported that obesity prevention programs could be effective for more than 4 months [66]. Some studies have found no association between the duration of the intervention and weight change [63]. These differences could be due to different populations, age groups, or settings. Stone et al. in a study conducted in Italy to evaluate the effectiveness of the recommended activities in schools, with at least 20 min’ physical activity in a day, reported that less than half of children (49%) took part in the physical activity, while after 7 years follow-up none of the children were engaged in physical activity schedules of more than 20 min [67]. Although we did not show the minimum possible time for the interventions to be effective in this study, the theory of Prochaska and DiClemente [68], recommended that 6 months is the minimum time for stabilizing behavior change involving PA practice. We were not able to assess the long-term sustainability of obesity prevention policies, because there was a limited number of studies that included long-term follow-up after the end of the intervention [54, 69, 70]. From the perspective of the frequency of intervention, optimal frequencies seemed to be daily or weekly schedules, with little effectiveness seen at monthly intervals. It has been established that integration of obesity prevention interventions in the classroom is difficult to achieve [65] and their long-term effectiveness is negligible [67]. Another finding of this study was that school-based programs had the most favorable results in prevention of obesity, which was consistent with the results of some previous studies [64] supporting Centers for Disease Control and Prevention (CDC) [71] and World Health Organization (WHO) [72] recommendations that schools are the best place for obesity prevention in children and adolescents. Wang et al. found that multi-setting trials had beneficial and significant effects compared to single-setting interventions against pediatric obesity [9]. Since most studies of the studies in pediatrics are conducted in schools, further investigations in other settings are indicated to elucidate their effectiveness in pediatric obesity prevention. In our finding, the integration of education alongside changes in the school environment had more favorable results compared with education only. Similarly, Sbruzzi et al. [73] reported that education-only interventions are effective the obesity treatment but not prevention. The heterogeneity of educational materials that are provided in different studies make it difficult to achieve a unique finding about their effectiveness [74]. Most studies (65%) were carried out in either Australia or the United States. Wang et al., in a meta-analysis across high-income countries, found similar results [9]. In subgrouping according to age, reductions in BMI and BMI-Zscore were observed in children aged 5–10 years old; similarly, in one study conducted by Peirson et al. in 2013 [63] among 0–18 years old children, beneficial results were observed in the same age range. Richards et al. showed that the strongest effect of PA intervention was found in the youngest children (grade 3 learners compared to the grade 4–6 learners). This was interpreted to be because the intervention promoted PA in the form of playing may have been more attractive and suitable for the younger children [75], or maybe it is because of the ease of interventions in this age groups [76]. On the other hand, high schools and middle schools were more likely to sell competitive foods than were elementary schools [77], which can have a negative impact on the implementation of obesity prevention policies. Finkelstein et al. in their study demonstrated that the consumption of unhealthy foods were high in the high schools children than in elementary school children [78], which is probably due to the fact that the behavior of buying fast food and soft drinks is not fully formed at this age group of children. Finally, most of the childhood obesity prevention studies used diet and physical activity combined as an intervention strategy. The result of the current study showed that diet and physical activity-based policies were more effective regarding BMI and BMI-Zscore reduction while studies with physical activity-only interventions were not effective. The results of studies by Katz et al. [79], Peirson et al. [63] and Wang et al. [9] found that a combination of diet and physical activity compared to diet-only or physical activity-only interventions had the most favorable results in pediatric obesity prevention. Our sensitivity analysis showed that by removing the studies of Kremer et al. [45], de Silva-Sanigorsk et al. [54] and Sadeghi et al. [42], a significant reduction in BMI-Zscore was observed. One of the most important features that these three studies had in common was poor management of selection bias in the quality assessment. As shown by Munafò et al., selection bias can considerably influence observed associations in large-scale cross-sectional studies [80].

Strengths and limitations

The principal strength of the current study is a comprehensive assessment of obesity prevention policies with an emphasis on different settings, age ranges, and interventional materials and content with BMI and BMI-Zscore as target outcomes. We also considered the possible role of the intervention duration, follow-up time and the amount of physical activity by including both randomized and non-randomized controlled clinical trials. Some of the limitations of the current meta-analysis should also be noted; for example, we were not able to obtain the education time and physical activity from all included articles because some of the articles did not specify these. Physical activity and nutrition education interventions are complex and, in each study, different approaches and theories may be used, which in all studies didn’t mention the approach and method of them, therefore, different approaches in educational methods and physical activities made it difficult to classify.

Conclusion

In conclusion, childhood obesity prevention (a) in school-based policies (b) between the ages of 5–10 years old children, (c) in short-term periods (less than 2 years) at more frequent intervals, (d) with a dual approach of diet and physical activity, can be effective in preventing childhood obesity. These findings can be used by health policymakers and policy providers to apply more effective strategies for obesity prevention in this age group.