Systematic review and meta-analysis of school-based interventions to reduce body mass index

Background Childhood obesity predisposes to adult obesity and increases the risk of many diseases. Schools provide a vehicle to deliver public health interventions to all children. Methods Medline and Embase were used to undertake a systematic review of published studies of school-based interventions aimed at reducing the body mass index (BMI) of children (cid:2) 18 years. Preferred reporting items for systematic reviews and meta-analyses guidelines were followed, and eligible studies subjected to a random effects meta-analysis. Results Between 1991 and 2010, 43 published studies provided 60 measurements of effect. The pooled effect was a 0.17 (95% CI: 0.08, 0.26, P , 0.001) reduction in BMI. Heterogeneity was high ( I 2 ¼ 93.4%) but there was no signiﬁcant small study bias (Egger’s test, P ¼ 0.422) nor signiﬁcant variation by length of follow-up. The intervention comprised physical activity only in 11 (26%) studies, education only in three (7%), and combinations of these and improved nutrition in the remaining 29 (67%). On stratiﬁed analysis, physical activity used in isolation ( 2 0.13, 95% CI: 2 0.22, 2 0.04, P ¼ 0.001) or combined with improved nutrition ( 2 0.17, 95% CI: 2 0.29, 2 0.06, P , 0.001) was associated with signiﬁcant improvements in BMI. Interventions targeted at overweight/obese children reduced their BMI by 0.35 (95% CI: 0.12, 0.58, P ¼ 0.003). Those delivered to all children reduced it by 0.16 (95% CI: 0.06, 0.25, P ¼ 0.002). to reduce BMI in children.


Introduction
The increasing prevalence of childhood obesity poses a major threat to public health. In the USA, the prevalence of severe [defined as body mass index (BMI) ! 99th centile] childhood obesity has tripled in the last 25 years. 1 Obesity increases the risk of many conditions, including type II diabetes, hypertension, cardiovascular disease and musculoskeletal disease. 2 Lifestyle behaviours developed in childhood tend to perpetuate into adulthood. Hence, obese children are more likely to become obese adults. 3 The WHO 4 has acknowledged that childhood interventions are required to combat adult obesity effectively. In the UK, education is free to all children between 3 and 18 years of age. Therefore, schools provide an ideal vehicle for delivering public health interventions to all children, 5 including those from the most socio-economically deprived communities who are most at risk, 6 -8 and hardest to reach. According to the primary prevention strategy first mooted by Rose,9 small population shifts in BMI may be more effective at a population level than simply reducing the prevalence of obesity. The most recent meta-analysis of school-based interventions included studies published up to 2007. 10 It demonstrated a significant reduction in the prevalence of obesity but, at that time, there was no evidence of a significant overall reduction in BMI. Because of the increasing public health importance of childhood obesity, many studies have been published more recently. Therefore, we conducted an up-to-date meta-analysis of published studies that evaluated the impact of schoolbased interventions on the body mass index (BMI) of pupils.

Systematic review
A literature review was conducted in parallel by H.V.L. and J.P.P. in accordance with the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines. 11 The search was undertaken using the Medline and Embase databases, applying the following search terms and Boolean connectors to titles, abstracts and subject headings: (child OR children OR childhood OR toddler* OR school-age* OR schoolage* OR infant* OR pediatric* OR paediatric*) AND (school* OR kindergarten* OR creche OR nursery OR nurseries OR afterschool) AND ( prevent* OR intervention*) AND [(obes* OR overweight OR (weight adj1 gain)) OR ((increase* OR gain* OR change*) adj2 (BMI OR body mass index))]. The electronic search was limited to studies conducted on human subjects of 18 years of age or younger that were published in or translated into English. No restrictions were placed on publication date so as to include as many studies as possible. Duplicate articles were excluded. The last date on which the electronic search was run was 21 February 2011. The identified articles were reviewed manually and the following inclusion criteria applied: † children aged 18 years of age or younger, † any intervention delivered in a school setting and aimed at decreasing BMI or weight, † effect reported as the mean change in BMI or this could be calculated from the pre-and post-intervention data provided and † inclusion of a control group which received no intervention beyond normal school-based activities and for which change in BMI was also reported or able to be calculated.
Non-randomized intervention studies were not excluded. Where more than one article was found relating to the same study, the most recent publication was used. The reference lists of both review articles and eligible original articles were searched manually to identify any additional eligible articles not found as a result of the electronic search.

Meta-analysis
Study characteristics were extracted and recorded: publication year, country, age and sex of participants, study size and design, selection criteria, nature, timing and duration of the intervention, and length of follow-up. Where follow-up results were recorded at different time points, the longest follow-up measure was used and, where available, sub-group results were used in favour of overall results. For studies that did not report the confidence interval or standard deviation (SD) for the mean change in BMI, this was imputed from the studies that did. Correlation coefficients were derived for the intervention and control groups using the formula: Corr ¼ (SD 2 baseline þ SD 2 final 2 SD 2 change )/(2 Â SD baseline Â SD final ). The SD of the change in BMI was then calculated using the formula: p (SD 2 baseline þ SD 2 final 2 (2 Â Corr Â SD baseline Â SD final )). A random effects meta-analysis was conducted on the full dataset and then repeated stratified by sex and then intervention type. I 2 was calculated as a measure of heterogeneity between studies. Bias was assessed both subjectively, using a funnel plot, and formally, using Egger's regression asymmetry test for small study bias. The influence of individual studies on the overall effect size was assessed using a meta-influence plot and a cumulative meta-analysis was performed to determine whether the pooled effect size changed over time as new studies were published. Univariate and multivariate meta-regression analyses were used to determine the effect of specific study characteristics on the overall effect size and, therefore, potential sources of between-study heterogeneity. Meta-regression analyses were subjected to 20 000 permutations to adjust for multiple testing, and therefore reduce the chance of type 1 errors. The adjusted R 2 and residual I 2 values were calculated in order to determine how much of the effect size was accounted for by the study characteristics recorded and how much heterogeneity remained after taking account of these. A bubble plot was produced to determine whether there was any relationship between length of follow-up and effect size. All analyses were performed using Stata version 11.1.

Main findings of the study
There is accumulating evidence that school-based interventions can significantly reduce children's BMI, especially if they include a physical exercise component. The evidence is reasonably consistent, in that a relatively large number of studies have now demonstrated a benefit. The effect size did not vary by length of follow-up suggesting that the benefits may be maintained over time, but only one study has followed-up participants for more than 4 years. Evidence of significant benefit is currently lacking for interventions that do not include a physical activity component. The absolute reduction in BMI was greater for interventions targeted at overweight and obese children, but studies delivered to all children nonetheless produced a significant reduction in overall BMI.
What is already known on this topic?
The prevalence of childhood obesity in developed countries is high and increasing, focusing attention on the urgent need to identify effective interventions. 1 Interventions can be targeted at individuals, families, the whole population or our obesogenic environment and all play a role. In a recent survey, 65% of American citizens believed that schools have a major role to play in tackling the obesity epidemic and only 7% believe that the school had no role to play at all. 55 The current NICE 5 guidelines also recommend schoolbased interventions. Earlier meta-analyses demonstrated conflicting results. 10,56 -58 The most recent, published in 2009, included 19 studies published up to 2007. 10 Since then, an additional 21 eligible studies have been published. Therefore, an updated review is timely.

What this study adds?
Previous meta-analyses had already demonstrated the potential of school-based interventions to reduce the prevalence of obesity. Inclusion of more recent studies enabled us to demonstrate that a statistically significant reduction in overall BMI is also achievable. The absolute benefit was a 0.17 kg/ m 2 reduction in BMI. This was statistically significant but is unlikely to be clinically significant at an individual level. It may, nonetheless, produce tangible health benefits at a population level. As first described by Rose, a small shift in population distribution can be an effective primary preventative strategy because more events occur among the large number of individuals at moderate risk than the small number at high risk. 10   Our study included data collected on 36 579 pupils, and was conducted in accordance with PRISMA guidelines. Random effects meta-analysis was chosen over fixed-effect because of differences in inclusion criteria and the nature of the intervention. Published studies have used a number of different anthropometric measures. By necessity, the meta-analysis needed to be based on studies using the same measure. Following a preliminary review, we chose BMI change as it was the most commonly used measure, enabling us to include the maximum number of studies. The need to exclude similar studies that used other measures, such as individual percentage fat mass or the overall percentage of pupils that were overweight, is an obvious limitation. Use of BMI z-score would have been preferable to the use of absolute BMI. However, only a minority of studies reported their results in terms of a change in z-score, therefore this was not possible. Furthermore, BMI may not be the best measure of childhood adiposity nor the best predictor of adult adiposity. 59 Where studies reported change in BMI adjusted for potential confounders, the adjusted result was used in the meta-analysis. However, some studies reported only unadjusted results which may be subject to bias. Some studies did not report SDs or confidence intervals for their results. In order to include as many studies as possible, we derived SDs for 16 studies that did not report this information. Since, the correlation coefficients for intervention and control groups were within one decimal place of each other (0.802 for intervention groups and 0.891 for control groups), this approach is unlikely to have introduced a large or systematic error and enabled us to include the maximum number of studies. Because of the small number of studies conducted in this area, we included non-randomized intervention studies. This is likely to have added to the heterogeneity of the results. In the future, as more studies become available, it would be useful to repeat the meta-analysis excluding non-randomized studies.

Conclusions
School-based interventions can reduce the BMI of pupils. The meta-analysis identified several areas where further research would be useful. The interventions examined to date appear to be less effective in boys than girls and further work is required to explore the reasons and whether they require modifications to the school-based interventions or an alternative approach. Existing studies suggest a benefit up to 6 years follow-up. Further research is required to determine whether it is maintained thereafter. Benefit has been demonstrated for a number of different types of intervention. Further research is required to determine the ideal type of intervention, taking cognisance of cost-effectiveness as well as clinical effectiveness.