Comparative effectiveness of school‐based interventions targeting physical activity, physical fitness or sedentary behaviour on obesity prevention in 6‐ to 12‐year‐old children: A systematic review and meta‐analysis

A systematic search of the literature was performed to compare the effects of interventions that targeted sedentary behaviours or physical activity (PA) or physical fitness on primary prevention of obesity in 6‐ to 12‐year‐old children. The search identified 146 reports that provided relevant data for meta‐analysis. Point estimates in % body fat were higher for fitness interventions compared with PA interventions (standardized mean difference = −0.11%; 95% CI = −0.26 to 0.04, and −0.04%; 95% CI = −0.15 to 0.06, respectively). Including sedentary behaviour to a PA‐ or fitness‐oriented intervention was not accompanied by an increase in intervention effectiveness, as the point estimates were slightly smaller compared with those for PA‐ or fitness‐only interventions. Overall, the effects tended to be larger in girls than in boys, especially for PA + sedentary behaviour interventions. There was some evidence for inequality, as the effects on body mass index were seen when interventions were delivered in the general population (standardized mean difference = −0.05, 95% CI = −0.07 to −0.02), but not in groups of disadvantaged children (standardized mean difference = −0.01, 95% CI = −0.29 to 0.19). In conclusion, school‐based PA interventions appear to be an effective strategy in the primary prevention of childhood obesity among 6‐ to 12‐year‐old children, but targeting sedentary behaviour in addition to PA or fitness does not increase the effectiveness of the intervention.


| INTRODUCTION
Noncommunicable diseases remain the leading cause of death in most parts of the world, and a large part of this mortality is ascribed to insufficient physical activity (PA) and obesity. 1 Specifically, physical inactivity is the fourth, and obesity has been ranked as the fifth leading risk for global mortality. 1 At the same time, the prevalence of overweight and obesity is rising worldwide among all age groups, with the epidemic being especially marked among children and adolescents. 2 In this age group, obesity has increased dramatically during the last few decades of the 20th century, especially in the most developed countries. 2 Interestingly, it seems that this increase has been much larger in 5-to 19-year-old children as compared with younger children. 3 Obesity in children has been linked to both short-4 and long-term adverse health outcomes. 5 Furthermore, childhood obesity frequently persists in adulthood, which is accompanied by many well-known detrimental effects on health. 6 PA, alongside unhealthy dietary habits, is proposed as one of the major contributors to childhood obesity. 7 In addition, PA in childhood has been linked to many other favourable health outcomes as well as to improved academic performance. 8 Although there remains little doubt that PA is beneficial for health, many posit that physical fitness is an even more powerful marker of health. 9,10 On the other hand, sedentary time has also been associated with several adverse health outcomes, although evidence for a specific link with obesity is weak. 11 Currently, there is only little or no evidence that a relationship between sedentary time and adiposity in children and adolescents is causal. 12 Indeed, a recent study that collated data from 14 accelerometer investigations in children and used iso-temporal substitution to model the effects of reduced sedentary time on health estimated that replacing 1 h of persistent sedentary time with nonsedentary pursuits would lead to only a mild reduction of body mass index (BMI). 13 On the other hand, the same study found that replacing 1 h of sedentary time by moderateto-vigorous PA increases the estimated decrease in BMI by more than seven times. 13 Hence, PA interventions might exhibit larger effects on obesity-related outcomes than interventions aimed at reducing sedentary behaviours. However, this remains to be confirmed in clinical trials.
Obesity, PA and sedentary pursuits are complex phenomena that require population-based solutions. For children, schools are frequently identified as an ideal setting for introducing lifestyle change and the prevention of weight gain. In most countries, school is obligatory, at least by midadolescence; hence, all children can be reached, which makes schools a perfect setting to reduce health inequalities. In addition, children spend a significant portion of the day in school.
Because academic activities are mostly sedentary, ample opportunities for PA should be provided in order to increase energy expenditure and introduce the well-known benefits of PA on health and academic performance. Indeed, several previous systematic reviews that examined the effects of obesity prevention interventions have shown that school-based interventions are most effective when a PA component is included. [14][15][16] However, the characteristics of successful PA intervention are less understood.
We aimed to bridge this gap by assessing what types of PA interventions in schools are the most effective in improving obesityrelated outcomes. To this end, we compared the effects of three groups of interventions: (1) programmes that aimed to reduce sedentary behaviour, (2) interventions that intended to increase PA and (3) interventions that were designed to improve physical fitness. We identified several systematic reviews published in the last 10 years that assessed the effectiveness of PA interventions on obesity prevention. [14][15][16][17][18][19][20][21] However, none of these studies attempted to document and analyse specific elements of PA programmes. Moreover, several of these analyses might have missed large studies as they were restricted to randomized designs. 18,21 Others were restricted to high-income countries 15 or to a single obesity-related outcome only. 19,20 Thus, in order to cover a complete spectrum of PA interventions, we included all school-based interventions that targeted energy expenditure, regardless of the type or duration of the intervention.
The wide range of included interventions will serve to identify features that enhance the effectiveness of these programmes in primary prevention of obesity, with special focus on the type of energy expenditure component targeted.

| METHODS
The protocol for this review was registered with Prospective Register of Systematic Reviews (PROSPERO 2019 CRD42019129295), and the methods are briefly described in the following sections. intervention-PA, fitness and sedentary behaviour; setting-school; outcome-adiposity); and adapted this strategy to individual databases (see List S1 for MEDLINE search strategy). We did not limit our search to any specific geographical region; however, we included only studies written in European languages. The search strategy was validated by conducting sensitivity analysis in MEDLINE with a test set of 10 key papers selected as exemplary papers answering our research question.

| Literature search and data extraction
Adjustments to the search strategy finished when all 10 key papers were identified by the search. All database search results were extracted and imported into the web-based reference manager: Rayyan. After duplicates were removed, results were screened initially by abstract and title. The first 500 results were screened independently by two reviewers (H.P. and J.K.). Given that >95% agreement between reviewers in included studies was recorded, each of the two reviewers screened half of the remaining results. Ambiguities on study eligibility were resolved through discussion with a third reviewer (M.So.). In addition to this, we checked reference lists of key systematic reviews in the same area for eligible studies. 14,[16][17][18]20,22 Lastly, we searched reference lists of all included reports.
Inclusion criteria were as follows: (1) randomized or nonrandomized control trial, controlled before and after study or natural experiment; (2) control group; and (3) participants aged 6 to 12 years (mean age at the start of the study = 5.5 to 12.49). Namely, school-going youth are typically divided into two age groups: children (6-12 years) and adolescents (13-18 years). Considering that the same strategies would probably not work for both children and adolescents and because of large heterogeneity in high school curricula that preclude one-size-fits-all policies, we decided to limit this review to children aged 6 to 12 years; (4) interventions of any duration that have aimed to either (a) increase PA and/or physical fitness or (b) reduce sedentary behaviour; (5) intervention that was performed primarily in school setting; (6) follow-up of at least 12 weeks from the start of the intervention; and (7) any obesityrelated outcome was measured (e.g. BMI, BMI z-score, BMI percentile, prevalence or incidence of overweight or obesity, percentage of body fat [%BF], skinfold thicknesses, waist circumference, waist circumference percentile, and waist-to-height ratio). Studies were excluded if (1) no obesity-related outcome was reported or the data came from self-report; (2) they included exclusively children with overweight or obesity or only special populations (e.g. children with a specific illness, blind, and physically disabled); and (3) full text was not available (i.e. only conference abstract).
After study selection, an extraction template was created (M.So.), and study characteristics were extracted by two reviewers working independently (Ž.L.P. and P.J.). Papers reporting on the results of the same study were collated so each study is the unit of analysis rather than each paper. Values at the longest available follow-up were taken for quantitative analyses.
The details on the intervention content were extracted from the main papers, the intervention protocols and the related web resources. Two reviewers independently extracted half of the data (Ž.L.P. and P.J.), and about 10% of the extracted data were double checked by the third reviewer (M.So.). Extracted items included authors, year, period of the study, number of clusters and participants, demographic characteristics, details on intervention type and content, duration of intervention and follow-up, and adverse outcomes.
Obesity-related study outcomes were extracted by two reviewers (JK and P.J.), working independently on half of the data, and entered in a predesigned excel template. The third reviewer (M.So.) verified 10% of the extracted results, and any discrepancies were resolved through discussion.

| Risk of bias assessment
A single reviewer (M.So.) assessed the risk of bias of all studies that met our inclusion criteria using Cochrane 'risk of bias' assessment tool for randomized studies 23 and modified Newcastle-Ottawa scale for nonrandomized study designs. 24 For individual randomized controlled trial (RCT), the assessment contained the following domains: (1) random sequence generation, For nonrandomized study design, risk of bias assessment was performed using modified Newcastle-Ottawa scale for cohort studies.
This scale originally includes eight domains, but one domain (i.e. demonstration that the outcome of interest was not present at the start of the study) was deemed not to be applicable for studies included in this review; hence, it was omitted. The domains assessed included the following: (1) representativeness of the intervention cohort (were participants representative for the community?), (2) selection of the nonintervention cohort (were controls drawn from the same community as the participants of the intervention?), (3) ascertainment of intervention (was the intervention implemented according to the plan?), (4) comparability of cohorts on the basis of the design or analysis (were analyses adjusted for age, gender and other important features, such as clustering and baseline values for the outcome of interest?), (5) assessment of outcome (was the outcome measured with an objective method?), (6) follow-up longer than 6 months from the start of the intervention (was the follow-up long enough for outcomes to occur?) and (7) adequacy of follow-up of cohort (subjects lost to follow-up unlikely to introduce bias owing to low or balanced attrition). According to the standard scoring protocol, 24 we awarded one star for Domains 1, 2, 3, 5, 6, and 7 and a maximum of two stars for comparability of cohorts domain. Studies that totalled at least six stars were classified as having an overall low risk of bias.

| Data analysis
We combine, in all cases, mean differences, calculated as follows: Mean difference = Differences in the intervention group − Differences in the control group, where the mean differences in the intervention and control groups denote the differences between values at follow-up and baseline in each of the groups. The units of measurement were kg m −2 , units of the standardized normal and % for BMI, BMI z-score and %BF, respectively. Next, when comparing five groups of interventions in both the main analyses and subgroup analysis (described in the next section), we standardized these raw differences using the standard deviations of the differences (with the exception of the BMI z-score, because it is already standardized).
We computed the uncertainty parameter (I 2 ) representing the percentage of total variance in the observed results explained by heterogeneity and assessed heterogeneity using the Q test. 25 We performed the meta-analysis using a random-effects model, which takes into account both within-and between-study heterogeneity. 26 Finally, we assessed publication bias for both overall results and main subgroup analyses with Egger's test for asymmetry. 27 The level of significance (alpha) was set at 5%.

| Subgroup analyses
To We further stratified all analyses by gender (both genders, i.e. studies that did not distinguish gender; boys and girls). Lastly, as a very limited number of studies that directly compared the effects in children of varying socio-economic status (SES) were available, for examining the equality aspect of interventions studied here, we compared the effects found in studies that focused on economically deprived children with interventions that included general population of children.

| Sensitivity analyses
For sensitivity analysis, we stratified the analyses, separately for PA, fitness and PA + sedentary and fitness + sedentary behaviour interventions, by study design (RCT vs. other designs), risk of bias (low risk of bias vs. moderate and high risk of bias), study period (<2009 vs. ≥2009) and mean age of participants (6-9 vs. 10-12 years).

| RESULTS
The search strategy retrieved 18 239 studies from eight databases.
After duplicates were removed, 17 014 records were screened by title and abstract. In the next step, 1091 were selected for screening of the full-text paper, and 242 were found to conform to our inclusion criteria. Searching the reference list of seven systematic reviews

| Characteristics of the included studies
An overview of the characteristics of the included studies is given in Table 1, and the details on individual studies are presented in

| Risk of bias
Risk of bias across domains for randomized and nonrandomized studies is shown in Figures 2 and 3, respectively, whereas risk of bias assessment across individual studies is presented in Tables S2 and S3.
We considered almost all trials to have low bias in blinding of outcome assessors domain as the outcomes were objectively assessed and as only several outcome measures were subject to observer bias.  Yet it has to be noted that indices of heterogeneity were large for all outcomes and ranged from I 2 = 82% to I 2 = 92%.

| Results by intervention type
When only studies that provided effects by gender are examined (n = 22G and 21B for BMI, n = 11G and 12B for BMI z-score, n = 13G and 13G for %BF), it becomes evident that gender is a significant  One study had two experimental groups: one that included PA and the other exposed to a combined intervention that additionally included sedentary behaviour component.    Table S5. The number of studies included in these analyses was fairly small (n = 3-7 for PA and n = 2-10 for fitness), hence smaller power and large confidence intervals. Still, point estimates for interventions that aimed to increase PA were generally larger in girls for most outcomes, whereas for interventions that were designed to improve physical fitness, the opposite was true.
Finally, the findings from sensitivity analyses (see Table S6) showed that the results were fairly robust, except when considering study design for PA interventions and age group for fitness interventions, and only for BMI as the outcome. Specifically, the effects of PA interventions were lower in RCTs compared with studies that employed other designs, whereas the effects of fitness interventions were larger in the younger age group.

| Effectiveness of interventions in vulnerable groups of children
Our search identified 26 studies that included predominantly economically deprived children and reported data appropriate for a meta-analysis. In general, the effects on BMI were not seen when interventions were delivered to vulnerable groups of children Comparisons of the effects on body fat were impeded by too few studies that focused on low SES and included body fat as an outcome.
Still, estimates from the few studies available show no effect in F I G U R E 6 Forest plot of standardized mean differences in change in percentage body fat between the intervention group and the control group for physical activity (PA), fitness, PA + sedentary behaviour and fitness + sedentary behaviour interventions disadvantaged children (standardized mean difference = −0.01, 95% CI = −0.13 to 0.12), and a trend to reduction in %BF in the general population (standardized mean difference = −0.04, 95% CI = −0.10 to 0.02).

| DISCUSSION
In this systematic review, we compared the effects of interventions that intended to increase PA with interventions that were designed to improve physical fitness and with interventions that aimed to reduce sedentary behaviour on obesity-related outcomes in 6-to 12-year-old children. The main results of our study include the following: In terms of characteristics that moderate the effectiveness of these type of interventions, the World Health Organization has recommended that obesity prevention programmes should span over at least 1 year, include both PA and a diet component, and involve parents, if possibly extending also to the home and community settings. 29 Our findings supplement these guidelines by indicating that interventions should be designed to improve fitness in order to maximize the effects on obesity prevention in 6-to 12-year-old children. However, this finding needs to be corroborated in future studies, as there was considerable overlap in confidence intervals of the effects of PA and fitness interventions studied here. Next, when analysing a smaller number of studies that reported effects by gender, we found evidence that fitness-oriented interventions are more effective than the ones directed to PA only in boys but not among girls. Therefore, more evidence is needed that this applies to both genders. Still, epidemiological studies support evidence from trials described here by reporting stronger cross-sectional associations with cardiometabolic risk factors for fitness than for PA. 10 Similarly, physical fitness has been identified as a moderator of the relationship between PA and cardiometabolic risk in children. More specifically, PA was associated with cardiometabolic risk factors in low-fit children but not in their fit peers. 30 The finding that the interventions that encompass several behaviours are not superior to programmes that focus on just one behaviour has already been reported for combination of PA with a diet component. Although evidence on this is not unequivocal, it was previously shown that diet + PA interventions in a variety of settings are not superior to programmes that target a single behaviour 28 and that these kinds of combined interventions have an even smaller impact on obesity-related outcomes than single-component programmes when set in schools. 15 Similarly, a meta-analysis of mostly nonschoolbased programmes showed that interventions targeting sedentary behaviour and PA simultaneously were not more effective in BMI reduction than interventions that focused exclusively on sedentary behaviour. 31 Our search strategy allowed for only two studies that focused only on reducing sedentary behaviour to be included in the quantitative synthesis. This precluded us from estimating reliable pooled effect sizes for any of the obesity-related outcomes assessed.
However, we were able to estimate the impact of adding sedentary behaviour component to PA or fitness intervention programmes on the primary prevention of obesity and found no added value of including sedentary behaviour component. Prior studies that included a variety of settings and a wider age range also failed to show the effectiveness of these types of intervention in obesity prevention. 31,32 This is hardly surprising given the low intensity of these kinds of programmes, strong reliance on educational content only and the high reinforcement of media use in today's cultures. Although it has been reported that these types of programmes can produce a significant decrease in sedentary behaviours, the effect size is too small to have an impact on weight regulation. 32 Nevertheless, given the unprecedented increase in exposure to screens faced by contemporary children, 33  Overall, mean pooled effects of interventions for primary prevention of obesity analysed in this review were larger in girls than in boys, especially for interventions aimed at both PA or fitness increase and a reduction in sedentary behaviours, although it has to be emphasized that the confidence intervals did not overlap only for BMI as an outcome measure. It is well known that school-aged girls are less physically active than boys. 34 To that end, the amount of PA typically used in intervention studies probably contributes more to the overall daily PA of girls. This, in turn, might lead to larger effects on energy expenditure and weight regulation.
The increasing burden of obesity and inactivity across SES has been well documented. 35 We found evidence that interventions that were delivered to economically deprived children analysed here were less able to induce favourable effects on each of the obesity-related outcomes studied than interventions in general population of children. shown that interventions targeting individual-level behaviour change may be less successful in disadvantaged children and that structural changes to the environment might be a better approach in reducing inequalities. 22 In addition, addressing social determinants of health outside the school setting is mandatory to ensure a sustainable reduction in the socio-economic disparities in children's health.
Finally, although only a handful of analysed studies provided data on adverse outcomes, we found no evidence for changes in body satisfaction, eating behaviours or underweight prevalence. In addition, the incidence of injuries was very low, even in studies with large volume of PA. Hence, school-based PA programmes can be considered very safe, regardless of the components used or PA volume implemented.

| Strengths and limitations
Our review has many strengths. First, we did not rely on search strategies set by prior reviews. Instead, we searched eight databases, including grey literature sources. Second, unlike most of the previous similar reviews, we did not limit our search to English language, thus increasing the probability of detecting evidence from low-to middleincome countries. Third, we accepted different study designs instead of constraining to RCTs while insisting on the control group to minimize bias. Fourth, we gathered very detailed data on the content of interventions, with a special reference to the frequency, intensity, duration and type of PA. Fifth, we included measures of body composition instead of relying only on BMI, which is regularly critiqued as an imperfect measure of adiposity. Furthermore, BMI can be affected by PA through an increase in lean body mass, which then typically leads to underestimation of intervention effects on adiposity.
Several limitations of this review are also worth noting. First, although unlike prior reviews we extended our search beyond English language, we could not include non-European languages, so we might have missed studies from Asia or Africa. Second, large variability in intervention characteristics led to statistical heterogeneity, which warrants caution when interpreting the results of meta-analysis. Third, over one third of studies that met inclusion criteria failed to provide all the data needed for a meta-analysis. Given the large number of studies included in the quantitative synthesis, we did not perform a qualitative synthesis of these studies. Hence, we cannot infer that the results of this additional qualitative evidence synthesis would agree with our conclusions. Fourth, a large number of studies describing PA interventions failed to document the exact duration of PA, and even fewer studies have quantified the intensity of implemented activities.
This limitation precluded us from describing the dose-response relationship. Fifth, although we restrained from predefining specific obesity-related outcomes, a sufficient number of studies for a metaanalysis were found only for BMI, BMI z-score and %BF. Both BMI and BMI z-score have often been criticized for inadequately assessing change in adiposity. 37 Abundant evidence has emerged that supports replacing BMI or BMI z-score with alternative metrics that can better capture longitudinal changes in obesity (e.g. Percent Over BMI-BMI50 and BMI85). 38