Schools are considered the central focus of activities for the prevention of paediatric obesity(Reference Hoelscher, Kirk and Ritchie1). In a meta-analysis of 139 studies of childhood obesity prevention conducted in high-income countries, 83 % were conducted in schools and the strength of evidence was higher for: (1) physical activity interventions delivered in schools, but with home involvement, or (2) combined diet–physical activity interventions delivered in schools with both home and community components(Reference Wang, Cai and Wu2). A previous systematic review based on findings from eight reviews, three meta-analyses and five systematic reviews of school-based programmes to prevent and control obesity did not find evidence for both prevention and reduction of obesity(Reference Khambalia, Dickinson and Hardy3). In a specific review of twenty-two studies conducted in low- and middle-income countries on dietary behaviour and physical activity for obesity prevention, most studies had positive behaviour effect, but the mean BMI had a small reduction only in eight studies(Reference Verstraeten, Lachat and Kolsteren4), but the review was concluded by the effectiveness of the school-based approach for obesity prevention.
By June 2020, a total of twenty systematic reviews and meta-analyses(Reference Wang, Cai and Wu2–Reference Oosterhoff, Joore and Ferreira22) were published, including two systematic review of systematic reviews(Reference Militello, Kelly and Melnyk8,Reference Peirson, Fitzpatrick-Lewis and Morrison20) . Studies have reported small, but statistically significant results in the treatment of obesity and less conclusive effect on prevention. In line with these findings, the editorial of the journal Childhood Obesity stated: (1) obesity prevention trials emphasising diet and physical activity/sedentary behaviour have had small or no effects on obesity and (2) quality of the studies has been relatively low(Reference Baranowski and Taveras23). Also, a review with pooled analysis of the reduction of sugar-sweetened beverages, one of the main goals in most dietary behaviour interventions for obesity, found only modest effectiveness of educational interventions(Reference Rahman, Jomaa and Kahale24). Alternatively, comparisons of earlier studies with more recent ones showed that more recent school-based interventions are at least mildly effective in reducing BMI in children, possibly because these newer studies tended to be longer, more comprehensive and included parental support(Reference Sobol-Goldberg, Rabinowitz and Gross18).
For low- and middle-income countries, the number of studies is still small and there is no evidence of reduction of obesity even in Mexico, a country with the highest prevalence of obesity among school-age children. A 2-year controlled intervention in schools in Mexico improved children’s food intake and activity, but BMI and obesity prevalence did not change(Reference Safdie, Cargo and Richard25). In contrast, a review of ten Latin American studies found evidence to recommend school-based interventions to prevent obesity among youth, although only five studies, three prevention and two treatment interventions, found improvements in obesity-related outcomes(Reference Lobelo, Garcia de Quevedo and Holub26).
Two major limitations in school-based studies are an underestimation of sample size due to the cluster randomised design and overestimation of changes in BMI. Also, only a sub-sample of participants may respond favourably to the intervention(Reference Baranowski27) requiring large studies or pooled analysis.
This study pooled five randomised behavioural school-based interventions, all conducted in the metropolitan area of Rio de Janeiro, Brazil, that have shown small non-significant changes in BMI after one school year of positive changes in dietary behaviours associated with obesity. This analysis with about 6000 participants may overcome the sample size limitation of the individual studies. Also, subgroup analysis was conducted for BMI status.
Materials and methods
A total of 5926 participants from public schools were included in the pooled analysis. Detailed information of participants and interventions are found in the publications, which references are in Table 1. All five studies analysed were carried out by our research group between the years 2010 and 2017 and had their raw data made available by the authors themselves. Complete follow-up varied from 87·8 % in the study number 2 to 79·0 % in the study number 5 (Table 1). In short, classroom activities were delivered by research assistants in all studies, except for study number 4, where activities were implemented by the regular teachers, after training. Culinary classes aimed at developing cooking skills to increase healthy eating choices were conducted by nutritionists. Three studies (numbers 3, 4 and 5) also encouraged physical activities and reduction of sedentarism beyond changes in dietary behaviour. Facilities for physical activities free of charge in the neighbourhood were indicated to the students and parents in two studies (4 and 5). Sedentarism approach stimulated reduction of 1 h of computer games and television; standing or walk during the interval of television programmes or in the game phase shift. In study number 4, those participants with overweight or obesity as diagnosed at school were also followed at the household monthly by healthy agents. Folders explaining the intervention programme and suggesting the participation of the family were delivered in studies 1, 2 and 4.
Characteristics of the studies included in the present pooled analysis(Numbers and percentages)
The allocation concealment strategy was the use of opaque envelopes with the names of the participant schools. Randomisation of the schools was conducted by professionals who were not related to the project. In all studies, baseline average BMI indicates a balance of the outcome. Blindness of outcome measure and food intake was not possible because there are many clues in the intervention schools. However, blindness has low chance to bias the results because the measurements are objective, and they were entered in the computer during the measurements by field researchers.
In all studies, school randomisation was implemented using opaque envelopes. The sample size calculation in the studies 1 and 2 was based on a difference in BMI comparing intervention and control of 1 unit and in the study 4 on 0·4 units of BMI. Study 3 was a feasibility one for primary combined with secondary activity among those overweight or with obesity through classes of dance and soccer. In study 5, the sample size was calculated to detect an average difference of 10 min in daily time spent on physical activity and a sd of 47·4.
The students completed a self-reported questionnaire with questions on food intake, sedentary behaviour, sociodemographic and the practice of physical activity. Changes in food and beverage intake were assessed by a 24-h recall and by a FFQ. Skin colour was self-defined as white, brown and black. Weight and height were measured at school, and the body composition was estimated by bioelectrical impedance at the beginning and the end of the school year by trained field workers using the same protocol in all studies. The main outcome was the change in BMI (BMI = weight (kg)/height2 (m2)) calculated in Z-score, according to WHO curves(Reference de Onis, Onyango and Borghi28), using the WHO AnthroPlus programme, version 1.0. The WHO BMI classification was used. For longitudinal analysis, the main outcome was the change in the BMI because it better evaluates the change in adiposity in growing children and adolescents compared with the Z-score of the BMI(Reference Cole, Faith and Pietrobelli29,Reference Berkey and Colditz30) .
The main analysis was an intention-to-treat performed through mixed models considering the cluster effect of schools. Subgroup analysis evaluated the effect of intervention by BMI classification (overweight and obesity) and stratified by sex. Age at each measurement was the time variable, allowing to correct for the increase of BMI with age. The main effect was estimated by the interaction between age and intervention, meaning that the change in BMI and percentage body fat over time is modified by the intervention. Analyses were conducted using the software Statistical Analysis System (version 9.4; SAS Institute).
Results
There were no differences between the control and intervention groups at baseline for age, skin colour and sex. Of the students, 46 % defined themselves as Brown (Table 2).
Characteristics of the students at baseline according to group allocation(Numbers and percentages)
The mean age of participants increased from the first to the last study. Since age is the time variable, all longitudinal analyses were adjusted for age. The prevalence of overweight at baseline increased from 15·3 % in 2005 to 17·1 % in 2017 and the obesity from 10·2 to 11·3 %. There is a small imbalance of BMI in the intervention and control groups of the 2010 study, but at the pooled analysis, the two groups are balanced at baseline (Table 3).
Sample size, prevalence of overweight and obesity and BMI at baseline according to group allocation(Numbers and percentages; mean values and standard deviations)
Losses to follow-up were not related to weight status, both in the intervention and in the control group. Prevalence of overweight and obesity at baseline, among those lost to follow-up in the control group, was, respectively, 15·4 and 12·5 %; these values in the intervention group were 16·3 and 11·5 %, respectively. Both sets of values were close to the prevalence in the overall study: 16·6 and 11·5 %. Losses were also unrelated to sex. The percentage of boys in pooled analysis was 52·6 % and among those lost to follow-up was 54·8 %. However, those lost to follow-up were younger. Group 9 to 11 years represented 56·8 % of the overall participants and 49·8 % among those lost to follow-up.
BMI variation over time was linear in both control and intervention groups in all studies (Fig. 1). Overall, there was no intervention effect on BMI, also for the subgroups. The subgroup analysis by BMI classification showed the greatest increase in BMI in the intervention group compared with the control for those with obesity at baseline. The regression coefficient of the change in BMI was 0·15 for obesity; for overweight, the coefficient was 0·03, and among normal weight 0·05; none of these changes was statistically significant (respective P values of 0·30, 0·52, 0·10). However, among boys, the percentage of body fat showed a greater reduction in the control compared with the intervention group (P < 0·01), while, in the girls, there was an increase in both groups without statistical significance (Fig. 2).
Estimated* mean BMI of overall studies and individual studies.
Baseline and follow-up predicted BMI and percentage body fat. Overall data and according to BMI status and sex. 
, Intervention group; 
, control group.
Discussion
The pooled analysis showed a lack of change in BMI associated with the intervention, and the small change in body fat among boys was of a greater reduction in the control compared with intervention group.
As expected, there was an increase in the prevalence of overweight and obesity over the years with obesity in our study changing from 9·2 to 11·8 % from 2010 to 2016. These data are in line with what is being shown in major national surveys. In the National Survey of Schools in Brazil (PeNSE), in 2009, the percentage of adolescents with overweight was 23·0 and 7·3 % were classified with obesity. These percentages, in 2015, went up to 23·7 and 7·8 %, respectively(31,32) . Our study participants are from public schools, and the prevalence of obesity is also greater among students from public schools in the PeNSE survey.
Although our negative results for obesity prevention, school level activities may be still attractive because: (1) children and adolescents can be reached at school; (2) changes in behaviour have been observed in our and most studies and (3) adolescents tend to respond better when treated as a group. However, the results of the present study, in line with most studies, had no impact on BMI(Reference Khambalia, Dickinson and Hardy3,Reference Verstraeten, Lachat and Kolsteren4,Reference Baranowski and Taveras23) .
A possible explanation for the observed results is the short time between interventions and the evaluation of BMI. However, the large sample size increases the power of showing small changes, and the only observed change was against the hypothesis. Another possible explanation for the results in this and many related studies with positive changes in behaviours and lifestyle habits, without BMI change, is the lack of reduction of energy intake. Reduction of energy intake is never a target in the interventions.
Messages for reducing overall energy intake, the most important factor to be changed in obesity, are not easy to implement since many students may be experiencing growth spurts which increase energy needs, whereas others could be susceptible to developing eating disorders(Reference Kenney, Wintner and Lee33). For these reasons, primary prevention strategies have concentrated on the quality of the diet rather than energy restriction. Messages of healthy eating may not change energy intake, as documented in our first study when sodas were replaced for juices with added sugar(Reference Sichieri, Paula Trotte and De Souza34). Increasing intake of fruits also has no power to dislocate the consumption of other high-energy items(Reference Kaiser, Brown and Brown35).
Also, actions to improve physical activity at school often attract those who are more fit compared with those with excessive weight, and school prevention strategies should not focus solely on obesity due to the increased risk of stigmatisation(Reference Kenney, Wintner and Lee33).
The cumulative experience from the five studies is of minimal participation of the families. All studies were conducted at schools attending families of low socio-economic level from metropolitan areas that spend many hours commuting to work. Lack of family participation may have hampered behaviour changes leading to a reduction of excessive weight gain. Also, students attending the public schools in the metropolitan area of Rio de Janeiro are in greater percentage Black or Brown, and studies conducted with African Americans also found that effects on weight-related behaviours and weight change were generally promising but often non-significant(Reference Barr-Anderson, Adams-Wynn and Disantis16). The percentage of Whites in the pooling analysis was 27·8 % compared with 47·7 % in the Rio de Janeiro population (Census 2010).
The main limitation of the study is the possibility of a social desirability bias in the reported behaviour changes, with girls showing a greater frequency of this bias compared with boys(Reference Hebert, Ma and Clemow36); however, the behaviour changes were not the issue in this pooled analysis. Girls with overweight and obesity could avoid having weight and height measured, but losses to follow-up were unrelated to sex and to weight status both in the control and in the intervention group.
The strengths of this pooled analysis are to add information to the very few studies conducted in low- and middle-income countries, the large sample size and high follow-up participation. The positive side of all analysed studies is the possibility of behaviour change with all school-based interventions as have been shown in the individual analysis.
Acknowledgements
We would like to thank everyone involved in the intervention studies conducted over the years by our research group (Nucleus of Epidemiology and Nutrition Biology – Nebin). We also thank all the co-authors who contributed to various phases of the project, helping to improve this manuscript
This research received no specific grant from any funding agency, commercial or not-for-profit sectors.
R. S. is the principal investigator of this study. R. S., R. P., E. Y. and D. C. conceived and designed the study. R. S., B. H., M. S., B. S. and R. R. contributed to the analysis of the data. R. R., B. H. and R. S. wrote the paper. All authors contributed to revising the manuscript and all read and approved the final manuscript.
There are no conflicts of interest.
