Globally, the number of children and adolescents (5–19 years) with obesity is estimated at 124 million, with a prevalence of 5·6 % in girls and 7·8 % in boys in 2016(Reference Abarca-Gómez, Abdeen and Hamid1). Childhood obesity, typically defined by the BMI cut-off points specific for age and gender(Reference Cole and Lobstein2), increases the risks of being an adult with obesity(Reference Ward, Long and Resch3), developing type 2 diabetes, hypertension, dyslipidaemia(Reference Juonala, Magnussen and Berenson4) and CHD(Reference Li, Chen and Srinivasan5–Reference Baker, Olsen and Sørensen7). BMI reduction amongst individuals with childhood obesity reduces the risk of developing type 2 diabetes(Reference Juonala, Magnussen and Berenson4), hypertension and dyslipidaemia(Reference Li, Chen and Srinivasan5) in adulthood to a level comparable to those with normal weight. Hence, interventions for the treatment of childhood obesity are necessary.
Despite the extensive literature on childhood obesity treatments in healthcare settings and community-based interventions in schools, primary care and community sites(Reference Al-Khudairy, Loveman and Colquitt8–Reference Sargent, Pilotto and Baur11), the evidence is primarily limited to well-funded clinical trials and poses limitations in scalability in practice(12). These limitations include the lack of resources when scaling out those interventions and present potential difficulties when delivered to a diverse population(Reference Aarons, Sklar and Mustanski13,Reference Dietz, Baur and Hall14) . Compared with randomised controlled trials (RCT) which determine the efficacy of an intervention given in a specific condition, cluster-RCT evaluate the effectiveness of an intervention in a real-world setting(Reference Cook, Delong and Murray15). Therefore, evidence gathered on cluster-RCT allows a better generalisation of results which could guide policy decisions in the translation of knowledge into practice(Reference Freemantle and Strack16).
To combat childhood obesity, the treatment should be an element of universal health coverage and integrated into existing systems where children spend most of their time to achieve maximum effectiveness(17). Based on Cochrane reviews of the treatment of childhood obesity, the use of diet, physical activity and behavioural interventions can improve BMI outcomes with no difference observed between treatments conducted in schools and other settings(Reference Al-Khudairy, Loveman and Colquitt8,Reference Mead, Brown and Rees9) . School-based interventions for the treatment of childhood obesity allow access to children and adolescents who spend a considerable amount of time in school(Reference Patton, Sawyer and Santelli18,Reference Gortmaker, Swinburn and Levy19) , including those from lower socio-economic status who may not seek healthcare(Reference Safron, Cislak and Gaspar20). As many children and adolescents have meals in school which are equipped with facilities for physical activity, school-based interventions provide an avenue to increase the quality of their diet and physical activity duration(Reference De Bourdeaudhuij, Van Cauwenberghe and Spittaels21). Parental involvement could be made possible by leveraging existing communication channels with parents to sustain behavioural changes at home(Reference Smith, Wohlstetter and Kuzin22). Though children and adolescents with overweight and obesity may face poorer self-perception(Reference Ottova, Erhart and Rajmil23) and discrimination from peers in school(Reference Analitis, Velderman and Ravens-Sieberer24), school-based obesity interventions were found to improve the self-esteem of children and adolescents more than interventions delivered in a clinical setting(Reference King, Jebeile and Garnett25).
Three systematic reviews and meta-analyses studied the effectiveness of school-based intervention for the treatment of childhood obesity(Reference Gonzalez-Suarez, Worley and Grimmer-Somers26–Reference Schroeder, Travers and Smaldone28), and two of the reviews found school-based interventions to be effective in reducing BMI outcomes(Reference Sbruzzi, Eibel and Barbiero27,Reference Schroeder, Travers and Smaldone28) . The reviews identified different components of parental involvement(Reference Schroeder, Travers and Smaldone28), educational interventions(Reference Sbruzzi, Eibel and Barbiero27) and the combination of physical activity and educational interventions(Reference Gonzalez-Suarez, Worley and Grimmer-Somers26) which contributed to the reduction in BMI outcomes. Two reviews had differing findings on the duration required for significant changes in BMI of 6 months(Reference Sbruzzi, Eibel and Barbiero27) and 1 year(Reference Gonzalez-Suarez, Worley and Grimmer-Somers26). Statistical heterogeneity was encountered in the reviews(Reference Gonzalez-Suarez, Worley and Grimmer-Somers26–Reference Schroeder, Travers and Smaldone28), and their analyses were limited to sensitivity analyses rather than exploring the effectiveness of interventional components through subgroup analyses. Based on the existing systematic reviews, firm conclusions could not be drawn on school-based interventions for the treatment of childhood obesity due to the following limitations. First, the reviews included obesity prevention trials which consisted of participants with normal weight(Reference Gonzalez-Suarez, Worley and Grimmer-Somers26–Reference Schroeder, Travers and Smaldone28). Second, the reviews had a specific focus on educational interventions(Reference Sbruzzi, Eibel and Barbiero27) and interventions led by nurses(Reference Schroeder, Travers and Smaldone28) instead of a multicomponent approach delivered by various professionals. Third, separate analyses were conducted for trials which reported BMI and BMI z-scores, respectively, which led to inconclusive findings regarding the effectiveness of school-based interventions(Reference Sbruzzi, Eibel and Barbiero27,Reference Schroeder, Travers and Smaldone28) .
Therefore, this review aimed to systematically locate, appraise and review the evidence of the effectiveness of school-based interventions for the treatment of childhood obesity and to identify variations in the effect size attributed to the interventional characteristics of theoretical framework, type of professional support, duration of intervention, components and parental support. The results of this review can contribute to guiding the design of effective school-based interventions for the treatment of childhood obesity.
Method
The systematic review and meta-analysis are reported in accordance with the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) (refer to PRISMA Checklist)(Reference Moher, Liberati and Tetzlaff29). This review is registered in the International Prospective Register of Systematic Reviews (CRD42020160735).
The eligible trials were (i) cluster-RCT, (ii) included children and adolescents (aged 6–18 years) defined as overweight or obese using the BMI-for-age percentiles based on a reference population, (iii) interventions included which aimed towards the treatment of overweight and obesity, (iv) conducted in the school setting or carried out in the school vicinity and (v) measured primary outcomes of BMI and BMI z-score (refer to online supplementary material, Supplemental Table S1). Secondary outcomes were cardiovascular, metabolic, nutrition, physical activity, sedentary behaviour and psychosocial-related outcomes. Trials were excluded if the participants were diagnosed with any medical or psychiatric condition, received medication or underwent surgical intervention that may influence their weight. Obesity prevention studies and publications in languages other than English were excluded.
The review adopted the recommendations of the Cochrane Handbook for Systematic Reviews (2019)(Reference Higgins, Thomas and Chandler30). First, the authors searched through eight databases (PubMed, Embase, CINAHL, Cochrane Library, PsycINFO, ProQuest, Scopus and Web of Science) from inception until 30 May 2020 according to the individual database’s subject terms and syntax rules (refer to online supplementary material, Supplemental Table S2). Second, the authors searched for ongoing and unpublished trials in various clinical trial registries. Third, the authors searched the reference list of relevant systematic reviews, selected trials, grey literature and targeted journals.
The study selection was conducted in four phases, which are graphically illustrated using the PRISMA flow diagram(Reference Moher, Liberati and Tetzlaff29). First, all identified records from the eight databases were collated in EndNote version X9 (Thomson Reuters), and duplicates were removed. Second, two authors (T.J.H.H. and L.J.C.) independently screened the titles and abstracts to remove irrelevant records. Third, the two authors assessed the full-text articles independently based on the eligibility criteria. Studies were collated when multiple reports of the same study were identified. Last, the two authors compared their findings to verify if any articles were overlooked. The two authors resolved incongruity through discussion or with the involvement of a third author (L.Y.).
The revised Cochrane risk-of-bias (RoB 2.0) tool was used to identify the risk of bias among the selected trials(Reference Sterne, Savović and Page31). RoB 2.0 assessed individual randomised trials based on five domains of bias: bias arising from the randomisation process, bias due to deviations from intended intervention, bias due to missing outcome data, bias in the measurement of the outcome and bias in the selection of reported result. As this review only considered cluster-RCT, additional considerations were taken for the domain of randomisation that could arise from the timing of identification and recruitment of participants(Reference Eldridge, Campbell and Campbell32). For the domain of bias due to deviations from intended interventions, the review used signalling questions based on the effect of the assignment to intervention. The RoB 2.0 algorithmic tool proposed the risk of bias as (1) low risk of bias, (2) some concerns or (3) high risk of bias for each domain based on the responses to the signalling questions. Two authors independently responded to each of the signalling questions with (1) yes, (2) probably yes, (3) probably no, (4) no or (5) no information using the RoB 2.0 tool.
The Grading Recommendations, Assessment, Development and Evaluation (GRADE) system was used to assess the overall quality of the evidence(Reference Guyatt, Oxman and Schünemann33). Each author rated the quality of evidence independently by considering five factors: (1) methodological limitation, (2) inconsistency, (3) indirectness, (4) imprecision and (5) publication bias. Any discrepancy between the results of the two authors for the risk of bias or the quality of evidence was resolved through discussion or by consultation with the third author (L.Y.).
The authors (T.J.H.H. and L.J.C.) independently extracted data from the included trials using a standardised data extraction form based on the Cochrane Handbook for Systematic Reviews of Interventions. The extracted data from the included trials included authors, year of publication, setting, inclusion criteria, sample, details of the intervention, control, outcomes, attrition rate, protocol, trial registration and grant support. The authors of the studies were contacted if information was missing or insufficient. The authors (T.J.H.H. and L.J.C.) compared their findings and verified them with the original publication to check for inconsistencies in the data extracted.
Meta-analysis and meta-regression were performed using Stata 16 software’s(34), Meta command procedures(35). The primary outcome was the change in BMI and BMI z-score between the intervention and control groups. Preference was given to the BMI outcome in studies that reported both BMI and BMI z-score, as BMI is superior in assessing the changes in adiposity(Reference Cole, Faith and Pietrobelli36). If the change in BMI or BMI z-score was not provided, the mean difference was calculated using the baseline(T1) and post-intervention’s(T2) mean and standard deviation using the formulas provided by the Cochrane Handbook for Systematic Reviews (2019). The sd of the mean difference was calculated as sd =
$\sqrt {{\rm{SD}}_{\left( {T1} \right)}^2 + {\rm{\;SD}}_{\left( {T2} \right)}^2{\rm{\;}}-{\rm{\;}}2{\rm{\;}} \times {\rm{\;r\;}} \times {\rm{\;S}}{{\rm{D}}_{\left( {T1} \right)}}\; \times {\rm{\;}}} {\rm{S}}{{\rm{D}}_{\left( {T2} \right)}}$
, where
${\rm{S}}{{\rm{D}}_{\left( {T1} \right)}}$
represents sd at baseline and
${\rm{S}}{{\rm{D}}_{\left( {T2} \right)}}$
represents sd at post-intervention. The BMI correlation coefficients of 0·78, 0·84 and 0·83 were used to represent the 1-year duration of follow-up correlation for children (under 10 years), pubescents (10–14 years) and adolescents (above 14 to under 18 years), respectively(Reference Bayer, Krüger and Von Kries37). The standardised mean difference with inverse variance method was used to synthesise the outcomes of the BMI and BMI z-score. The review measured the magnitude of the intervention effect between the intervention and control groups using effect size expressed as Hedges (g), which are defined as very small (0·1), small (0·2), medium (0·5), large (0·8), very large (1·2) and huge (2·0)(Reference Hedges, Olkin, Hedges and Olkin38). The review used the random effects model, which assumes that the distribution of effect varies across studies, and the variation may arise from the intervention implementation and participants(Reference Borenstein, Hedges and Higgins39). Previous meta-analyses of children and adolescent weight management(Reference Gonzalez-Suarez, Worley and Grimmer-Somers26–Reference Schroeder, Travers and Smaldone28,Reference Guerra, Nobre and da Silveira40–Reference Lavelle, Mackay and Pell43) reported a substantial statistical heterogeneity. Therefore, the Hedges’ random effects estimator was chosen because of its better performance compared with other estimators in the presence of between-study variance(Reference Hedges and Olkin44,Reference Veroniki, Jackson and Viechtbauer45) . A cumulative meta-analysis was conducted based on the year of publication to examine the accumulation of evidence over time by performing separate meta-analyses with the addition of one trial at each time(Reference Lau, Schmid and Chalmers46). When two or more trials reported secondary outcomes using the same measures, the continuous data were analysed with mean difference, whereas continuous data were measured with a variety of measures that were analysed in standardised mean difference.
A test for statistical heterogeneity was conducted using Cochran’s Q (χ 2) with statistical significance of P < 0·1 because of its low statistical power in detecting true heterogeneity(Reference Higgins, Thompson and Deeks47).
${I^2}$
statistics was used to evaluate inconsistency and was interpreted as not important (0–40 %), moderate heterogeneity (30–60 %), substantial heterogeneity (50–90 %) and considerable heterogeneity (75–100 %). Overlapping
${I^2}$
values were classified according to the direction and magnitude of treatment effect and the P-value of the χ 2 test. Sensitivity analysis was conducted when heterogeneity was statistically significant to maintain homogeneity amongst the pooled trials(Reference Higgins and Green48). Subgroup analyses were conducted to identify the sources of heterogeneity related to trial characteristic, the classification of countries’ economies provided by the World Bank(49), trial duration, professional background delivering the intervention, number of sessions delivered and components of physical activity, nutrition education and parental involvement. In addition, Hedges random effects univariate meta-regression was performed to explore the potential sources of heterogeneity on the effect size of the change of BMI outcomes with variables similar to subgroup analyses and additional trial characteristics for year of publication, risk of bias, total sample size, total number of clusters and percentage of females recruited. Multivariate meta-regression analysis was conducted using a forward stepwise procedure of adding variables with P < 0·1, starting with the smallest P-value into the model(Reference Ranganathan, Pramesh and Aggarwal50).
Results
The study selection process is illustrated in Fig. 1. A total of 17 252 records were identified from eight databases and screened to remove duplicates (n 9557). The remaining 7695 records were screened to remove the records with irrelevant titles (n 6305) and abstracts (n 1128). Full-text articles (n 262) were downloaded and assessed against the eligibility criteria of this review. The authors excluded 249 records based on the reasons outlined in Fig. 1. Supplemental Table S3 outlines the reasons for the exclusion of each record. The systematic review and meta-analysis included thirteen articles from twelve trials(Reference Amini, Djazayery and Majdzadeh51–Reference Trost, Sundal and Foster63), in which one trial reported physical activity(Reference Bagherniya, Darani and Sharma52) and nutrition outcomes(Reference Bagherniya, Sharma and Mostafavi53) in separate publications.
Fig. 1 Selection of trials for inclusion in the systematic and meta-analysis
Table 1 outlines the characteristics of the twelve trials conducted in seven different countries amongst 1755 participants across the USA (n 5), Asia (n 5) and Europe (n 2). The trials were published between 2004 and 2017, and the sample size ranged from 41(Reference Grey, Berry and Davidson56) to 334(Reference Amini, Djazayery and Majdzadeh51). The twelve cluster-RCT included a total of eighty-four clusters, with each comprising two clusters(Reference Grey, Berry and Davidson56,Reference Mayurachat, Warunee and Jutamas58,Reference Toruner and Savaser62) to twelve clusters(Reference Amini, Djazayery and Majdzadeh51). All of the trials included participants with overweight and obesity status, except three trials, which targeted participants with overweight and a family history of diabetes (n 1)(Reference Freira, Lemos and Fonseca54) and participants with overweight and obesity and a family history of diabetes (n 2)(Reference Grey, Berry and Davidson56,Reference Grey, Jaser and Holl57) . The mean age of the participants ranged from 8·4 years(Reference Graf, Koch and Bjarnason-Wehrens55) to 15·9 years(Reference Freira, Lemos and Fonseca54). Only one trial with two publications exclusively recruited female participants(Reference Bagherniya, Darani and Sharma52,Reference Bagherniya, Sharma and Mostafavi53) .
Table 1 Characteristics of selected cluster randomised controlled trials for school-based weight management
AC, abdominal circumference; AMA, American Medical Association; CATCH, Child and Adolescent Trial for Cardiovascular Health; CDC, Centers for Disease Control and Prevention; CDI, Children’s Depression Inventory; HC, hip circumference; HOMA, Homoeostasis Model Assessment of Insulin Resistance; HPLP, Health Promoting Lifestyle Profile; N, No; NA, not applicable; non-part, non-participant; NR, not reported; OGTT, Oral Glucose Tolerance Test; PA, Physical Activity; RGSAS, Revised Godin-Shepard Activity; SB, Sedentary Behaviour; SFT, Skin Fold Thickness; T2DM, Type 2 Diabetes; WC, waist circumference; Y, Yes; YAQ, Youth/Adolescent Questionnaire; YRBS, Youth Risk Behaviour Survey.
Half of the trials (n 6, 50 %) were rated to have a high risk of overall bias, and the other half (n 6, 50 %) were rated as having some concerns (Fig. 2). In trials rated as having a high risk of overall bias, this was attributed to deviations from the intended interventions (n 5, 41·2 %), lack of intention-to-treat analysis (n 5, 41·2 %) and missing outcome data (n 4, 33·3 %) due to loss to follow-up, which differed between intervention and control groups. In trials rated as having some concerns for overall bias, this was mainly due to lack of information on the randomisation process (n 55, 41·2 %), as well as the lack of registration of a trial protocol (n 4, 33·3 %).
Fig. 2 Risk of bias summary. 
, Low risk; 
, some concerns; 
, high risk
The quality of evidence evaluated using the GRADE system was graded low based on the domains of risk of bias and inconsistency (see online supplementary material, Supplemental Table S4). The risk of bias was downgraded as 50 % of the trials were rated high risk in the overall bias. Heterogeneity was considered significant with Cochran’s Q, P < 0·1 with substantial inconsistency of I 2 = 88·9 %. Other domains of indirectness and imprecision were rated as insignificant. Egger’s regression asymmetry test (P = 0·68) was statistically insignificant, and the qualitative examination revealed a symmetrical funnel plot (see online supplementary material, Supplemental Figure S1), which suggested no evidence of small-study effects, including publication bias.
All trials’ clusters were conducted in schools with the exception of one trial, where clusters consisted of schools and the Young Men’s Christian Association(Reference Trost, Sundal and Foster63). The school-based interventions included components of counselling (n 2)(Reference Freira, Lemos and Fonseca54,Reference Pbert, Druker and Gapinski61,Reference Trost, Sundal and Foster63) , nutrition education (n 1)(Reference Mayurachat, Warunee and Jutamas58), counselling and physical activity sessions (n 1)(Reference Pbert, Druker and Barton60), nutrition education and physical activity sessions (n 4)(Reference Amini, Djazayery and Majdzadeh51–Reference Bagherniya, Sharma and Mostafavi53,Reference Graf, Koch and Bjarnason-Wehrens55,Reference Nayak and Bhat59) and a combination of nutrition education, physical activity sessions and counselling (n 4)(Reference Grey, Berry and Davidson56,Reference Grey, Jaser and Holl57,Reference Toruner and Savaser62,Reference Trost, Sundal and Foster63) (see online supplementary material, Supplemental Table S5). Five trials had varying degrees of parental involvement, which included sessions conducted solely for parents (n 3)(Reference Amini, Djazayery and Majdzadeh51,Reference Graf, Koch and Bjarnason-Wehrens55,Reference Nayak and Bhat59) , to parents participating in sessions with their child (n 2)(Reference Bagherniya, Darani and Sharma52,Reference Bagherniya, Sharma and Mostafavi53,Reference Trost, Sundal and Foster63) .
A meta-analysis was conducted to assess the effectiveness of school-based interventions for the treatment of childhood obesity compared with controls for all twelve trials (n 1755) on the measure of mean change in BMI outcomes from pre-intervention to post-intervention. The meta-analysis revealed a medium effect (g = 0·52) favouring school-based interventions (z = –3·43; P < 0·001; Fig. 3). Substantial statistical heterogeneity was noted and explored through sensitivity analysis and meta-regression (Q = 108·20; P < 0·1; I 2 = 88·9 %). Sensitivity analyses were conducted and did not have a remarkable impact on the effect size for BMI outcomes and statistical heterogeneity. Cumulative analysis by publication year revealed a trend favouring school-based interventions (see online supplementary material, Supplemental Figure S2).
Fig. 3 Forest plot of effect size change in BMI outcomes for school-based interventions and control groups
The exploration of intervention characteristics was conducted using subgroup analyses (Table 2). Significant subgroup differences (Q = 24·0, P = < 0·01) were found between the types of economies where the trial was conducted with large effects for lower-middle to upper-middle economies (g = 0·84) and a small effect for high-income economies (g = 0·27). The subgroup analysis of intervention length revealed a subgroup difference based on intervention duration (Q = 24·0, P = < 0·05) with a medium effect for trials of 6 months or less (g = 0·79) and a small effect for trials of more than 6 months (g = 0·24). School-based interventions with parental involvement (g = 0·59), nutrition education (g = 0·59) and focus on individual participants yielded a larger effect size compared with their counterparts despite the lack of a substantial subgroup difference.
Table 2 Subgroup analyses of school-based weight management interventions
* P < 0·1.
† P < 0·05.
‡ P < 0·01.
Univariate analysis observed an increase in BMI outcomes (
$\beta $
= 0·09; 95 % CI 0·00, 0·19; P = 0·046) for every month of increase in trial duration. The univariate analysis suggested a decrease in BMI outcomes (
$\beta $
= 0·55; 95 % CI –1·08, –0·03; P = 0·039) for weight management interventions delivered in lower-middle to upper-middle economies compared with those delivered in high-income economies (Table 3). The two covariates were incorporated in the multivariate meta-regression model and explained almost half of the heterogeneity (R 2 = 41·9 %, P = 0·03).
Table 3 Results of univariate and multivariate meta-regression
Ref, Reference.
* Explained variance, R 2 = 41·20 %, P = 0·018;
$\beta $
-coefficient.
† P < 0·05.
Seven selected trials assessed thirteen secondary outcomes (see online supplementary material, Supplemental Table S6). Under nutrition outcomes, a significantly lower energy intake of 50·09 kcal (95 % CI –79·3, –20·9), favouring school-based interventions, was observed compared with controls. A very large effect on increasing fruit and vegetable intake (g = 2·55) and a large effect on decreasing junk food intake (g = 1·05) were observed, favouring school-based interventions compared with controls. A large effect was observed favouring school-based interventions for dietary self-efficacy (g = 0·86) under the category of psychosocial-related outcomes.
Discussion
This meta-analysis quantitatively synthesised the body of evidence of twelve cluster-RCT examining the effects of school-based interventions for the treatment of childhood obesity. The result of the meta-analysis suggests that school-based interventions are possible approaches for the treatment for childhood obesity with effectiveness comparable to other meta-analytic reviews involving behavioural interventions(Reference Peirson, Fitzpatrick-Lewis and Morrison64) and lifestyle interventions(Reference Kitzmann, Dalton Iii and Stanley65). Subgroup analyses showed greater effectiveness of brief school-based interventions and interventions conducted in lower-middle to upper-middle economies. Meta-analyses of the secondary outcomes found school-based interventions resulted in positive behavioural changes of lower energetic consumption, reduced junk food intake and a greater intake of fruits and vegetables.
Consistent with other meta-analytic reviews on the treatment of paediatric obesity(Reference Mead, Brown and Rees9,Reference Peirson, Fitzpatrick-Lewis and Morrison64) and school-based interventions(Reference Sbruzzi, Eibel and Barbiero27), the subgroup analysis of this review indicated that brief school-based interventions of 6 months or shorter observed a greater reduction in BMI outcomes. In addition, this review found a positive association between trial duration and BMI outcomes, which suggests the potential effectiveness of school-based obesity treatment during the initial phase. The lower reduction in BMI outcomes for obesity treatment trials beyond 6 months may not be attributed to the failure of the intervention, but rather to the weight regain following weight loss due to physiological adaptations(Reference Greenway66), and the seasonal increases in BMI during school holidays(Reference Baranowski, O’Connor and Johnston67). This stresses the need for interventions to focus on the maintenance of weight reduction(Reference Mead, Brown and Rees9).
Despite school-based interventions being placed in a unique position to allow children and adolescents to be monitored for any possible relapse throughout the schooling years, the degree of attrition varied from 0 % to 66·9 % between trials and three trials reported the longest trial duration of only 1 year(Reference Grey, Berry and Davidson56,Reference Grey, Jaser and Holl57,Reference Toruner and Savaser62) . Additionally, trials reported challenges in implementing school-based interventions due to the lack of prioritisation of attending intervention sessions over supplementary lessons after school(Reference Grey, Jaser and Holl57,Reference Nayak and Bhat59) , and a perception that participants lacked interest in addressing obesity(Reference Graf, Koch and Bjarnason-Wehrens55). Trials which reported no attrition integrated the intervention into existing school hours and conducted sessions during non-essential classes(Reference Pbert, Druker and Barton60,Reference Pbert, Druker and Gapinski61) , and after-school care settings through active video games to sustain participants’ interest(Reference Trost, Sundal and Foster63). Partnership between healthcare providers and stakeholders in education is integral to ensure interventions for the treatment of childhood obesity are sustained throughout the schooling years(Reference Dietz, Baur and Hall14).
To a certain extent, the duration of the intervention and the type of economies where the trial was conducted explained the substantial statistical heterogeneity amongst the pooled trials. Trials in lower-middle to upper-middle income economies had a larger effect where the control group received no intervention or utilised a wait-list control(Reference Amini, Djazayery and Majdzadeh51–Reference Bagherniya, Sharma and Mostafavi53,Reference Mayurachat, Warunee and Jutamas58,Reference Nayak and Bhat59,Reference Trost, Sundal and Foster63) . This finding was consistent with the previous Cochrane review, where the subgroup analysis found larger effects when no interventions were delivered to the control group and a smaller effect when the control group received an active intervention(Reference Al-Khudairy, Loveman and Colquitt8). The findings from this review of selecting trials using a cluster-RCT design likely mirrored the standard of care of public health programmes available in schools to address childhood obesity in the respective economies. With the exception of one trial(Reference Graf, Koch and Bjarnason-Wehrens55), the control group of trials in high-income economies received dietary and physical activity sessions(Reference Grey, Berry and Davidson56,Reference Grey, Jaser and Holl57,Reference Trost, Sundal and Foster63) , or leveraged on existing healthcare services in schools for weight management(Reference Freira, Lemos and Fonseca54,Reference Pbert, Druker and Barton60,Reference Pbert, Druker and Gapinski61) which resulted in a smaller effect size. The observed differences in the standard of usual care suggest separate immediate approaches to address childhood obesity. For lower- to middle-income economies, resources would be needed to establish evidence-based interventions using approaches adapted to the existing school systems to ensure programme sustainability, as part of the universal healthcare coverage(12). High-income economies should make alterations to strengthen existing school-based interventions and ensure up-to-date practices in reducing childhood obesity.
School-based interventions for the treatment of childhood obesity utilised a combination of nutrition education, physical activity, counselling and parental involvement. Determining the effectiveness of a particular component was not possible considering the clinical heterogeneity amongst the trials. Consistent with other meta-analytic reviews on school-based interventions(Reference Gonzalez-Suarez, Worley and Grimmer-Somers26–Reference Schroeder, Travers and Smaldone28,Reference Guerra, Nobre and da Silveira40–Reference Lavelle, Mackay and Pell43) and treatment on childhood obesity(Reference Al-Khudairy, Loveman and Colquitt8,Reference Mead, Brown and Rees9,Reference Peirson, Fitzpatrick-Lewis and Morrison64) , this review encountered statistical heterogeneity among the pooled trials. Attempts were made to explore the heterogeneity, but to a large extent, the variation remains unexplained and requires further investigation.
Interestingly, the subgroup analysis revealed that the use of educators and trained staff had a large effect on the reduction of BMI outcomes. Although more studies are needed to ascertain the effectiveness of school-based interventions conducted by educators and trained staff, this approach has the potential for scalability to provide universal healthcare coverage for the treatment of childhood obesity(17), given the shortage of healthcare professionals(Reference Tsolekile, Abrahams-Gessel and Puoane68). Nonetheless, professionals in medical and health science can play a pivotal role in weight management because of their in-depth understanding of healthcare knowledge and, more importantly, the understanding of the social determinants of health(Reference Schroeder, McCormick and Perez69). The trials that used professionals in medical and health science were able to provide individual problem-solving sessions and tailor the healthcare interventions to ensure participation in weight management(Reference Freira, Lemos and Fonseca54,Reference Pbert, Druker and Barton60,Reference Pbert, Druker and Gapinski61) .
The strength of this review includes the use of a meta-analytic approach to synthesise the evidence for the effectiveness of school-based intervention for the treatment of childhood obesity from eight databases using a comprehensive and rigorous search strategy. However, this review has limitations. First, only trials published in English were included with evidence limited to USA, Europe and Asia, and lower-middle to high-income economies. Second, the quality of evidence is low, and substantial statistical heterogeneity was found amongst the included trials, with a great portion of the heterogeneity remaining unexplained. Third, the results of this meta-analysis are subject to ecological fallacy and/or Simpson’s paradox as with any meta-analysis(Reference Cooper and Patall70). Hence, the results need to be interpreted with caution.
Given the low quality of evidence for school-based intervention for the treatment of childhood obesity, this review has essential implications for future research. More high-quality cluster-RCT with longer follow-up are needed to determine the optimal duration of childhood obesity treatment and maintenance, and the appropriate combination of components. Special attention needs to be given to the randomisation process for cluster-RCT with an emphasis on missing data management, including the use of an intention-to-treat analysis according to the recommendations of the Consolidated Standards of Reporting Trials (CONSORT) for cluster-RCT(Reference Campbell, Piaggio and Elbourne71). The review highlights the need to conduct trials using the cluster-RCT design to allow better generalisation of results in the real-world setting. These cluster-RCT should have multiple clusters to allow a comparison of the interventions delivered by professionals in medical and health science against those delivered by educators and trained staff and to identify the maximum number of participants for each staff to deliver treatment successfully. A cost-effectiveness analysis would provide essential knowledge for scaled-up interventions, especially in developing countries. Such knowledge would allow the achievement of universal healthcare coverage, one of the UN’ Sustainable Developmental Goals(Reference Desa72).
The findings of this review highlight the potential for school-based interventions in treating childhood obesity. This review provides valuable information for policymakers regarding contextual factors designing school-based interventions, with consideration on the contextual appropriateness to ensure the sustainability of the programme. Ideally, the school-based intervention should be incorporated into existing systems by conducting it during school curriculum hours or during an after-school programme, to reduce the attrition of participants, and involving professionals in medical and health sciences in the delivery of the programme. Where resources are limited, involving educators and staff trained to deliver treatment of childhood obesity may be a possibility. For practitioners, delivering school-based interventions should utilise a combination of nutrition, physical activity, counselling and parental involvement components. Follow-up is required to monitor for weight regain and to engage children during school holidays.
In conclusion, school-based interventions for the treatment of childhood obesity can potentially reduce BMI and BMI z-score outcomes through the use of a variety of components, such as physical activity, nutrition education, counselling and parental involvement. Findings from this review suggest that the duration of the intervention is negatively associated with the effectiveness of the intervention and that the type of economy where the trial is conducted contributes to the heterogeneity of the meta-analysis. Given the low quality of evidence, high-quality trials are needed with longer periods of follow-up to better compare the cost-effectiveness between the types of professions delivering school-based interventions. School-based intervention is a possible approach to provide universal healthcare to end childhood obesity.
Acknowledgements
Acknowledgements: None. Financial support: This research did not receive any specific grant from funding agencies in the public, commercial or not-for-profit sectors. Conflict of interest: The authors declare no competing interests. Authorship: T.J.H.H. and L.Y. conceptualised and designed the study. T.J.H.H. and L.J.C. conducted the systematic literature search. T.J.H.H., L.J.C. and L.Y. performed the title and abstract screening, data extraction and data analysis. T.J.H.H., L.J.C. and L.Y. participated in the reference editing. T.J.H.H. wrote and formatted the article, and all authors read and approved the final version of the article. Ethics of human subject participation: Not applicable.
Supplementary material
For supplementary material accompanying this paper visit https://doi.org/10.1017/S1368980021001117
