Effects of policies or interventions that influence the school food environment on children’s health and nonhealth outcomes: a systematic review

Abstract

Context

Globally, 1 in 3 children under 5 years is undernourished or overweight, and 1 in 2 suffers from hidden hunger due to nutrient deficiencies. As children spend a considerable time at school, school-based policies that aim to improve children’s dietary intake may help address this double burden of malnutrition.

Objective

This systematic review aimed to assess the effects of implementing policies or interventions that influence the school food environment on children’s health and nonhealth outcomes.

Data sources, extraction, and analysis

Eleven databases were searched up to April 2020 and the World Health Organization (WHO) released a call for data due in June 2020. Records were screened against the eligibility criteria, and data extraction and risk-of-bias assessment were conducted by 1 reviewer and checked by another. The synthesis was based on effect direction, and certainty of evidence was assessed using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach.

Conclusions

Seventy-four studies reporting 10 different comparisons were included. The body of evidence indicates that interventions addressing the school food environment may have modest beneficial effects on certain key outcomes. Nutrition standards for healthy foods and beverages at schools, interventions that change how food is presented and positioned, and fruit and vegetable provision may have a beneficial effect on the consumption of healthy foods and beverages. Regarding effects on the consumption of discretionary foods and beverages, nutrition standards may have beneficial effects. Nutrition standards for foods and beverages, changes to portion size served, and the implementation of multiple nudging strategies may have beneficial effects on energy intake. Regarding effects of purchasing or selecting healthier foods, changes to how food is presented and positioned may be beneficial. This review was commissioned and supported by the WHO (registration 2020/1001698-0). WHO reviewed and approved the protocol for the systematic review and reviewed the initial report of the completed systematic review.

Systematic Review Registration

PROSPERO registration no: CRD42020186265.

INTRODUCTION

Globally, 1 in 3 children under 5 years is undernourished or overweight, and 1 in 2 suffers from hidden hunger due to nutrient deficiencies, and among school-age children 11% were wasted and 18% were overweight or obese in 2016.¹ While there is a reported reduction in stunted children worldwide, this is not the case in poorer settings, such as in Africa. On the other hand, the prevalence of overweight and obesity is increasing globally.¹

Malnutrition undermines the ability of children to grow and achieve their full potential. Children who are overweight or obese are at increased risk for adult obesity, early-onset type 2 diabetes and other noncommunicable diseases, stigmatization, and depression.^1–3 These, in turn, have serious socioeconomic consequences.⁴ Undernutrition poses a risk to mortality as well as for physical and cognitive developmental difficulties.^5,6 Malnutrition in children is largely due to poor diets, especially among those who are socioeconomically disadvantaged. Diets of malnourished children are generally low in fruits and vegetables (FV), whole grains, and nuts and fiber, and high in processed foods, salt, sugars, sodium, and calories.^7–10 School-based policies or interventions that aim to improve children’s dietary intake may help address this double burden of malnutrition.

The World Health Organization (WHO) Nutrition-Friendly Schools Initiative provides a framework for integrated school-based programs addressing all forms of malnutrition: policy options for improving the health and nutrition of young people. One of the Nutrition-Friendly Schools Initiative components concerns supportive school food environments. The accessibility and cost of the foods available in school environments and regulations regarding food marketing influence children’s food choices. Additionally, schools are important settings to promote healthy diets and instill lifelong lifestyle habits, as much of children’s time during the formative years is spent at school, where they can consume up to 50% of their daily calories.^11–15 The Second International Conference on Nutrition (ICN2) Framework for Action calls for healthy diets to be fostered in schools and includes recommendations for school food and nutrition standards, food programs aligned with national food-based dietary guidelines, and nutrition education as part of school curricula, among others.¹⁵ Therefore, policies or interventions that influence the school food and nutrition environment are important.^16,17

To ensure efficient use of resources, it is important to assess which policies or interventions are effective and can thus be recommended for implementation. Recent reviews have addressed this question. One of these had a few limitations, such as including uncontrolled studies, combining different study designs in meta-analyses, or not carrying out duplicate screening and data extraction,¹⁸ all of which reduce the rigor and validity of the review. Others focused on specific interventions, such as school lunches in the United States alone,¹⁹ or assessed policies or broad interventions addressing multiple risk factors, such as physical activity and tobacco and alcohol consumption.²⁰ To inform a WHO guideline on school food and nutrition policies, a systematic review was conducted to assess the effects of implementing policies or interventions that influence the school food environment on children’s health and nonhealth outcomes.

METHODS

This systematic review was registered with PROSPERO (registration no. CRD42020186265), was conducted according to Cochrane methods,²¹ and is reported according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement.²² The eligibility criteria were determined by the WHO Nutrition Guidance Expert Advisory Group Subgroup on Policy Actions, which are detailed in Table 1.

Search methods for identification of studies

Eleven databases were searched from their inception to April/May 2020: PubMed, Scopus, Web of Science, Open Grey, Cochrane Library, ERIC, CINAHL, WHO Global Index Medicus, Epistemonikos, and ClinicalTrials.gov (search strategies are shown in S1 File; please see the Supporting Information online). Further studies were retrieved through the WHO call for data (submissions deadline was 20 June 2020: https://www.who.int/news-room/articles-detail/call-for-data-on-nutrition-school-food-environment-2020).

Study selection process

All titles and abstracts were screened independently and in duplicate against the eligibility criteria. Full-texts of potentially eligible records were screened by 1 reviewer; all excluded records were checked by a second reviewer. Any disagreements regarding eligibility were resolved through discussion.

Data extraction and risk-of-bias assessment

One reviewer extracted data, and a second reviewer checked it. Information was extracted about the author, study location and funding, study design and methods, participants (including PROGRESS-Plus characteristics that can be used to differentiate intervention effects across social categories; eg, place of residence, race/ethnicity/culture/language, etc),²³ interventions, and outcome measures. One author assessed the risk of bias using the Cochrane “Risk of Bias” tool, as modified by Cochrane Effective Practice and Organization of Care,²⁴ and a second author checked it. For each criterion in the tool, each study was assessed as being at low, unclear, or high risk of bias. A study’s overall risk of bias was based on the risk for 2 prespecified domains: selection and attrition bias.

Data synthesis

To identify the comparisons for the synthesis, each study’s intervention characteristics were first tabulated. For the nudging interventions category, comparisons were organized according to the TIPPME (typology of interventions in proximal physical micro-environments) intervention typology categories: those that (1) influence how food is presented (eg, fun stickers on healthier options, (2) influence how food is positioned (eg, less healthy foods placed in a less accessible location), (3) influence size (eg, serving larger portion sizes), (4) influence the functionality of food (eg, allowing easier opening or pouring), and (5) implement multiple nudging strategies.²⁵

Data for each outcome were synthesized separately by study design. Due to substantial heterogeneity and lack of data for some studies, our main synthesis approach was based on vote counting based on effect direction (ED).²⁶ Five categories of ED were used: (1) clear effect favoring the intervention (effect estimate favors the intervention; 95% CI excludes the null, ▲), (2) unclear effect potentially favoring the intervention (effect estimate favors intervention; 95% CI includes the null and is wide, △), (3) no difference in effect (95% CI crosses the null and is narrow, □□), (4) unclear effect potentially favoring the control (effect estimate favors the control; 95% CI includes the null and is wide, ▽), and (5) clear effect favoring the control (effect estimate favors the control; 95% CI excludes the null, ▼). Where a 95% CI was not reported and could not be calculated, the P value was used to determine whether the ED was clear or unclear. If effect estimates were not reported but the authors reported the ED in the text, this information was used to guide decisions. Effects on key outcomes were illustrated graphically using harvest plots.²⁷

Where possible, meta-analyses were carried out for subsets of studies of the same design reporting the same comparison and outcome measure, using random-effects meta-analyses in Review Manager (The Cochrane Collaboration).²⁸ Results from crossover trials were included, with results from parallel trials in meta-analyses using the generic inverse variance method.²⁹

For dichotomous outcomes percentage points, odds ratios (ORs) or risk ratios were reported, with a 95% CI if this was reported or could be calculated. For continuous outcomes, mean differences (MDs) were reported if studies measured the outcomes in the same way, and the standardized MD if not, along with a 95% CI. Standardized MDs were interpreted as follows: 0.2 = small effect, 0.5 = moderate effect, and 0.8 = large effect.³⁰

Adjustment for clustering in cluster randomized controlled trials (RCTs) was checked. If studies had more than 1 intervention arm, relevant arms were combined to conduct pairwise comparisons. When outcome data were available for multiple time points, all that were measured during and immediately after the intervention were reported. Where multiple outcome measures of the same outcome were reported, 1 measure to include in the synthesis was selected to avoid double counting; the measure judged to be the most comprehensive, validly measured, and relevant for decision making (eg, body mass index [BMI] instead of weight and height) was selected. To ensure transparency, the results of all included studies and all measures reported are presented in S2 File (please see the Supporting Information online), per comparison and study design.

Heterogeneity was assessed in relation to PICO (Population, Intervention, Comparison and Outcome) elements, by visually inspecting the CI overlap in forest plots and using the χ² and I² statistics where meta-analyses were carried out. If a meta-analysis had substantial heterogeneity (ie, if the χ² test had a low P value [P < 0.1] or if the I² statistic was ≥50%),³¹ data were not pooled.

The planned subgroup analysis could not be carried out as there were insufficient data per outcome and comparison. The planned sensitivity analysis could not be carried out as meta-analyses were not the main synthesis approach and the meta-analysis conducted did not include 5 studies or more. However, the synthesis was presented separately for different study designs and the risk of bias was reported in the tables of individual study results and in the text. No outcome measures were reported in 10 or more studies and thus it was not possible to investigate the risk of publication bias.

Two reviewers assessed the overall certainty of evidence for all critical outcomes and one important outcome (anthropometric measures) using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach.³² The evidence profile tables were based on evidence from randomized studies, unless no randomized study reported a key outcome that was reported by a nonrandomized study.

RESULTS

The titles and abstracts of 17 706 de-duplicated records and then 1471 full texts were screened (Figure 1). Of the latter, 1285 records were excluded (S3 File; please see the Supporting Information online), 80 are awaiting classification as their full text could not be accessed (S4 File; please see the Supporting Information online), 7 are ongoing studies (S5 File; please see the Supporting Information online), and 99 records pertaining to 74 unique studies were included.

Characteristics of included studies

Table 2^33–103 presents an overview of included studies; details about each are available in the S6 File (please see the Supporting Information online). Thirty-two prospective controlled studies (PCSs), 34 cluster RCTs (10 crossover trials), 3 RCTs, and 5 uncontrolled interrupted time-series (ITS) studies were included. Most studies allocated the intervention at the school level (n = 53), included primary school children, and compared the intervention with no intervention. The sample sizes ranged from 23 to 24 291 participants and 1 to 235 schools, with follow-up ranging from 1 day to 8 years. Most studies (61/74) were conducted in high-income countries (HICs). The following intervention categories were included: (1) nutrition standards (n = 9), (2) nudging interventions (n = 22), and (3) direct food provision (n = 43). No study assessed marketing restrictions of unhealthy food and nonalcoholic beverages (FNABs) or pricing policies to promote healthier alternatives. All critical outcomes were reported, except for nutrient and calorie content of available foods or exposure to marketing of FNABs. All the important outcomes were reported, except for the prices of available foods and portion sizes. Most studies (n = 54) were funded by governmental or intergovernmental agencies or nonprofit organizations, and 7 studies were funded by the food industry. Most studies did not report on potential conflicts of interest (n = 37) or declared that none of the authors had any potential conflict of interest (n = 36).

Risk of bias in included studies

The summary of risk-of-bias assessments is presented in Figure 2. For included studies with a comparison group (n = 69), most RCTs (n = 22/37) did not report enough information about how the randomization sequence was generated and were thus at unclear risk of selection bias, and most reported that allocation was done centrally or at a cluster level and were thus at low risk of selection bias. All PCSs (n = 32) were at high risk of selection bias as there was no randomization. Most studies (n = 40/69) were at unclear risk of selection bias due to imbalances of baseline characteristics, whereas regarding the balance of baseline outcomes, studies were mostly at unclear risk of bias (31/69). Almost all studies (n = 64/69) were at low risk of performance bias due to blinding; blinding was mostly unfeasible, but lack of blinding was judged as unlikely to influence participant and personnel behavior in ways beyond what the intervention intended. Most studies were at low risk of detection bias (n = 42/69) and 22 were at high risk. Most studies (n = 52/69) had no evidence of contamination and were thus at low risk of performance bias. The studies varied in terms of missing outcome data, with 24 at unclear, 23 at high, and 22 at low risk of attrition bias. Most studies did not have a protocol and were thus at unclear risk of reporting bias. Most studies were at low risk of other types of bias (n = 36).

All of the ITS studies (n = 5) were classified as low risk of performance bias as it was unlikely that the participants were aware of the intervention or the data were objective. Four studies had all or almost all data available for analysis, and one was classified as unclear risk of attrition bias as it was unclear whether the meals data available were complete. All studies were classified as unclear risk of reporting bias as the protocol was not available. Four studies were judged as low risk of other bias and one was classified as unclear risk. Four ITS studies were classified as high risk of performance bias, as other changes that might have impacted the outcome occurred at the same time as the intervention, and 1 study was classified as unclear risk. All studies were at low risk of detection bias as trends were analyzed before and after the intervention, and the methods of data collection were the same before and after the intervention implementation. A more detailed risk-of-bias assessment is presented in S7 File (please see the Supporting Information online).

Effects of interventions

The effects on key outcomes for each comparison across included intervention categories are reported below; the effects on all other outcomes are presented in the S8 File (please see the Supporting Information online). No key outcomes were reported for comparison 6; the results for this comparison are available in S8 File. The individual results for all outcomes in each included study across all comparisons are available in S2 File; the shaded rows in these tables indicate the outcomes selected for the synthesis.

Intervention category: Nutrition standards/rules that determine the quality of food served or sold in and around schools

Comparison 1: Nutrition standards increasing the availability of healthy foods in school vs no intervention

Eight studies assessed this intervention, which are briefly described in Table 3.^33–39 The harvest plot in Figure 3 illustrates the effects on key outcomes.

Consumption of healthy food and nonalcoholic beverages

A PCS³⁸ reported that implementing nutrition guidelines to increase the availability of healthy foods sold at secondary schools may increase the mean consumption score of salty snacks meeting nutrition standards (including baked chips, popcorn, and pretzels; with higher scores indicating better results) at 1 year (coefficient: 0.29 points; P < 0.05; n = 996 children; low-certainty evidence) (Table 4^{33–35,37,38}). Similar results were observed for the consumption of beverages and sweet snacks meeting nutrition standards.

Consumption of discretionary food and nonalcoholic beverages

One PCS³⁸ reported that the implementation of nutrition guidelines may decrease the mean consumption score of salty snacks excluded by nutrition standards (including regular chips; with a higher score indicating poorer results; coefficient: –0.3; P < 0.05; n = 996 children; low-certainty evidence). Similar results were reported for the consumption of beverages excluded by nutrition standards and “no effect” was reported for sweet snacks excluded by nutrition standards.

Diet

An uncontrolled ITS study³⁷ reported that standards increasing the availability of whole grains, fruits, and vegetables at secondary schools in the United States may reduce energy intake slightly at 1.3 years (mean monthly energy density calculated as the available energy divided by the weight of foods served; coefficient = –0.46; P < 0.001; n = 6 schools; low-certainty evidence).

Purchasing behavior or sales data

Evidence from 2 RCTs indicates that nutrition standards that increase the availability of healthy food at school may make little to no difference to the proportion of favorable products (<100 kcal or low-fat entrees) sold (2 cluster RCTs; n = 30 schools; low-certainty evidence). In one of these studies³³ primary schools offered low-fat entrées compared to no-fat-content restrictions and the other³⁴ increased the availability of lower-calorie products (favorable: <100 kcal; moderately favorable: 100–170 kcal) in vending machines at secondary schools. One ITS study⁶⁸ assessing the removal of unhealthy food from secondary schools’ snack bars and cafeterias reported an unclear effect potentially favoring the control on the percentage of weekly sales for fruit, juice, and vegetables (percentage points [pp]: –12; 95% CI: –6.65 to 6.41; n = 3 schools).

Anthropometric outcomes

The evidence is very uncertain regarding the effects of a 10% reduction in the energy content of pre-primary school meals on obesity (BMI z score ≥ 2 SDs) trends over 2.8 to 4.8 years (3 uncontrolled ITS studies; n = 67 841 children; very-low-certainty evidence). Of the 3 uncontrolled ITS studies, all reported in 1 paper,³⁵ one reported a clear effect favoring the intervention at 4 years and 8 months and the other 2 studies reported no effect. The same studies reported no difference on the trends in stunting (height-for-age z score < –2 SDs).

Comparison 2: Nutrition standards increasing the availability of healthy beverages in school vs no intervention

One PCS was included in this comparison³⁹; it assessed eliminating or reducing sugar-sweetened beverages (SSBs) and diet soda availability in secondary schools in the United States (Table 3).

Consumption of discretionary food and nonalcoholic beverages

Nutrition standards to increase the availability of healthy beverages in schools may make little to no difference on the consumption of SSBs (1 study, 456 children; low-certainty evidence; Table 5³⁹). It reported no difference in the mean servings per day of SSBs for boys (MD: –0.01; 95% CI: –0.30 to 0.28; n = 183) or girls (MD: –0.01; 95% CI: –0.20 to 0.18; n = 273).

Intervention category: Nudging interventions promoting healthy food behavior in the school environment

Comparison 3. Changes to food presentation vs no changes

Seven studies, all conducted in the United States, reported on the effects of changing how food is presented; they are briefly described in Table 6.^40–61

Consumption of healthy food and nonalcoholic beverages

Nudging interventions that change how food is presented may slightly increase consumption of healthy FNABs, including FV and white milk (6 studies, n = 596 children; low-certainty evidence; Table 7).^40–46 Of the 6 trials, 3 reported a clear effect favoring the intervention^41,42,45 and 3 reported an unclear effect potentially favoring the intervention^40,43,44 (seeFigure S1 in the Supporting Information online).

Diet

Two trials^44,46 reported no difference in the effect of nudging interventions that change how food is presented on energy intake, but the evidence is very uncertain (2 trials, n = 37 children; very-low-certainty evidence).

Purchasing behavior or sales data

Nudging interventions that change how food is presented may increase healthy food purchasing behavior (3 trials, n = 1134 children; low-certainty evidence). Of the 3 trials, 2 reported clear effects favoring the intervention on the selection of FV^41,45 and 1 trial reported an unclear effect potentially favoring the intervention on the selection of white over chocolate milk.⁴⁰

Comparison 4. Changes to how food is positioned vs no changes to food position

Two cluster trials^40,47 and 3 PCSs^48–50 assessed the effects of changing how healthy food/beverages are positioned so that they are more easily accessible to children (Table 6).

Consumption of healthy food and nonalcoholic beverages

Nudging interventions that change how food is positioned may increase the consumption of healthy FNABs (3 RCTs, n = 14 schools; low-certainty evidence; Table 8).^40,47–49 One study assessing the positioning of water so that it is more accessible reported a clear effect favoring the intervention⁴⁹ and 1 study assessing the placement of water coolers in cafeterias reported an unclear effect potentially favoring the intervention⁴⁸ (seeFigure S2 in the Supporting Information online). The third⁴⁰ assessed changes to the position of white and chocolate milk so that the former was more accessible; it reported a decrease in the proportion of white milk consumed in the intervention group (50% to 47%) and an increase in the control group (54% to 56%); no effect measure was reported (seeFigure S2 in the Supporting Information online).

Purchasing behavior or sales data

Nudging interventions that change how food is positioned may increase the selection of healthy target foods or beverages (2 trials, n = 2462 children; low-certainty evidence). Of the 2 trials, 1 reported a clear effect favoring the intervention (OR: 3.81; 95% CI: 2.4 to 6.1)⁴⁰ and the other an unclear effect potentially favoring the intervention (OR: 1.136; 95% CI: 0.791 to 1.632).⁴⁷ Two PCSs also reported this outcome, both assessing the placing of water coolers in school canteens. One reported an unclear effect potentially favoring the intervention on sales of SSBs⁴⁸ and the other reported that the changes in SSB sales were similar in intervention and control schools but did not report effect measures.⁵⁰

Comparison 5. Serving larger portion sizes vs reference portion sizes

Five cluster crossover trials assessed changes to portion sizes served on diet intake measures (Table 6).

Diet

Serving smaller portion sizes may reduce energy intake slightly (4 trials, n = 346 children; low-certainty evidence; Table 9).^51,53–55 Two trials reported a clear reduction in energy intake with the reference portion size,^54,55 1 study reported an unclear effect potentially favoring the smaller child-sized dishware (ie, lower energy intake),⁵¹ and 1 study reported no difference between the groups⁵³ (seeFigure S3 in the Supporting Information online).

Comparison 7. Multiple nudging strategies

Three cluster trials^57–59 and 2 PCSs^60,61 assessed these multiple nudging strategies (Table 6). The harvest plot in Figure 4 illustrates effects on key outcomes.

Consumption of healthy food and nonalcoholic beverages

Evidence from 2 trials^58,59 shows that multiple nudging interventions result in little to no difference in vegetable intake (2 trials, n = 176 children/7752 trays; moderate-certainty evidence; Table 10,^57–59,Figure 4). One PCS reported no difference in effect on the proportion of students consuming any vegetable.⁶¹

Diet

Multiple nudging interventions promoting FV consumption probably reduce energy intake (2 studies, n = 2547 children; moderate-certainty evidence).^57,58

Purchasing behavior or sales data

Implementing multiple nudging strategies to promote fruit consumption may result in little to no difference on the number of vegetable units selected in an average tray (1 trial, percentage points = 24.6; P = 0.074, n = 2108 students; low-certainty evidence). ⁵⁹ Two PCSs^60,61 reported an unclear effect potentially favoring multiple nudging strategies to promote selection of fruits, vegetables, and low-fat milk, on the proportion of students selecting FV or salads at 1 year (MD = 0.03servings, P = 0.3, n = 7 752 trays).

Intervention category: Direct food provision to students in schools (n = 43)

Comparison 8. Provision of fruits/vegetables at school vs no intervention

Sixteen studies assessed this intervention; they are briefly described in Table 11.^62–77 The harvest plot in Figure 5 illustrates effects on key outcomes.

Consumption of healthy food and nonalcoholic beverages

Evidence from 5 trials ^62–66 indicates that FV provision may result in little or no difference in the consumption of FV (5 trials, n = 8256 children; low-certainty evidence; Table 12).^62–66 These trials reported a range of effects (Figure 5). In one of these studies fruits could be purchased from a fruit tuck shop at minimal cost. Eight PCSs reported on FV intake, with effects ranging from clearly favoring FV provision in 6 studies,^{67,69–72,77} to an unclear effect potentially favoring FV provision in 1 study, ⁷⁴ and no difference in effect in 1 study that assessed a paid subscription for receiving FV at school.⁷⁵

Consumption of discretionary food and nonalcoholic beverages

Direct FV provision in schools may result in little to no difference in the consumption of sweets and cakes (2 trials, n = 6308 children; low-certainty evidence). One trial, assessing the implementation of a fruit tuck shop at a primary school, reported a very small unclear effect potentially favoring the intervention⁶⁶ on the consumption of cookies, cakes, and brownies in calories, and the other, assessing funding for schools to provide FV, reported no difference in effect on the consumption of sweets, chocolates, and biscuits.⁶⁵

Diet

Evidence from 1 trial⁶⁶ indicates that FV provision may make little to no difference to energy intake (MD = 47; 95% CI: –21.6 to 115.6; 1 study, n = 4696 children; low-certainty evidence).

Anthropometric outcomes

Evidence from 1 trial⁶³ in primary school classrooms in Australia, aiming to reduce overweight/obesity, indicates that direct provision of fruit in schools may make little or no difference on BMI (kg/m²) (1 study, n = 267 children; very-low-certainty evidence). One PCS⁷³ assessing the inclusion of vegetables from the school garden in primary school meals in South Africa, with the aim to reduce undernutrition, reported an unclear effect potentially favoring the control on BMI z scores.

Comparison 9. Provision of school meals vs no intervention

Twenty-five studies assessed school meal provision (Table 13).^78–101 Two of these assessed a slightly different comparison: free universal access to school meals vs income-based access.^86,101

Consumption of healthy food and nonalcoholic beverages

Evidence from 1 trial⁷⁹ indicates that the breakfast provision in primary schools likely results in little to no difference in the consumption of healthy items at breakfast (including fruit, bread, cereal, and milk products; 1 RCT; MD = 0.16; 95% CI: 0.02 to 0.3; P < 0.05; n = 3068 children; moderate-certainty evidence; Table 14,^{79–81,84,88,}Figure 5). Breakfast provided in this study included 1 item from 4 food groups—milk-based drinks or products, non–sugar-coated cereal, fruits, and breads. One PCS at high overall risk of bias⁹⁷ assessing the provision of free primary school meals in Norway reported no effect on the number of FV portions consumed per day at 5 months (MD = –0.10; 95% CI: –0.66 to 0.46; n = 159 children). One PCS¹⁰¹ assessing the provision of universal free school meals in primary and secondary schools in the United Kingdom compared with income-based access reported a clear effect favoring the intervention on the consumption of vegetables at lunchtime (MD: 26 pp; P < 0.01) and a clear effect favoring the control on the consumption of whole fruits at lunchtime (MD: –19 pp; P < 0.01) after 2 years. However, the study reported that the overall consumption of FV of both groups did not differ.

Consumption of discretionary food and nonalcoholic beverages

Evidence from 2 trials indicates that the provision of school meals likely results in little to no difference in the consumption of unhealthy items consumed at breakfast or of sugar or candy (2 studies, n = 3768 children; low-certainty evidence).^79,88 One PCS,¹⁰¹ which assessed universal vs income-based access to school meals in the United Kingdom, reported a clear effect favoring the intervention on the percentage of students consuming soft drinks at lunchtime (MD = –16 pp; P < 0.05) and an unclear effect potentially favoring the control on the percentage of students consuming cakes and other types of sweets at lunchtime (MD = 2 pp; P > 0.05) after 2 years.

Diet

The evidence is very uncertain about the effect of direct provision of school meals on energy intake at 1 to 3 months (3 trials, n = 1643; very-low-certainty evidence). Two trials aimed to reduce undernutrition. Both reported an increase in energy intake; however, the effect was clear for 1 study assessing a breakfast program (4 cookies and an instant drink) in primary schools in Peru ⁸¹ and unclear for the other assessing the provision of a stew in primary schools in Kenya.⁹⁰ The third trial aimed to prevent obesity through the provision of a morning and afternoon snack and lunch following the Nordic nutrition standards⁸⁸; it reported an unclear effect potentially favoring the intervention. One cluster RCT, at unclear overall risk of bias, assessing free breakfast compared with an income-based breakfast program reported a small decrease in energy intake/day with the intervention (MD = −32 kcal; 95% CI: –54.17 to –9.83; n = 3347 children).⁸⁶

Anthropometric outcomes

Of 5 trials reporting on BMI measures (kg/m² or BMI z scores), 2 aimed to prevent obesity^80,88 and evidence from these trials indicates that the provision of school meals likely results in little to no difference to BMI (2 trials, n = 798 children; moderate-certainty evidence). Of 7 PCSs reporting BMI or BMI z scores, 6 aimed to prevent obesity,⁹¹ with 2 reporting clear effects favoring the control,^91,94 1 reporting unclear effects potentially favoring the control,⁹⁰ and 3 studies reporting an unclear effect potentially favoring the intervention.¹⁰⁰ A pooled analysis of 2 of these studies^90,94 reporting on BMI (kg/m²) indicates that providing healthy school meals slightly increases BMI at 4 to 8 months (seeFigure S4 in the Supporting Information online). A pooled analysis of the other 2 of these studies reporting on BMI z scores^91,97 shows that providing school meals results in a small increase in BMI z scores (MD = 0.13; 95% CI: 0.02 to 0.24; n = 1990 children; I² = 0%) (seeFigure S5 in the Supporting Information online).

Comparison 10. Provision of milk at school vs no intervention

Two trials assessed the daily provision of milk, in primary schools in the United Kingdom over 21 months¹⁰³ and in preschools in Sweden over 3 months.¹⁰² Baker et al¹⁰³ was at unclear overall risk of bias and Koivisto et al¹⁰² was at high overall risk of bias.

Purchasing behavior/sales data

Milk provision likely increased the selection of milk with the school lunch instead of water (1 study, n = 132 children; moderate-certainty evidence; Table 15).¹⁰²

Anthropometric outcomes

One trial¹⁰³ reported an unclear effect potentially favoring the intervention on growth in height (MD = 0.28 cm; 95% CI: –0.01 to 0.57) and in weight (MD = 0.13; 95% CI: –0.23 to 0.49) at 21 months.

DISCUSSION

Summary of main results and comprehensiveness of the evidence

This review included 74 studies evaluating the effects of nutrition standards, nudging interventions, and direct food provision, across 10 comparisons.

Nutrition standards to increase availability of healthy food and beverages at school may increase the consumption of healthy foods and beverages and may reduce the consumption of unhealthy foods and beverages (low-certainty evidence), may reduce energy intake slightly (low-certainty evidence), may result in little to no difference to the proportion of children selecting target healthy foods (low-certainty evidence), and effects on obesity trends are very uncertain (very-low-certainty evidence). Nutrition standards to increase the availability of healthy beverages at school may make little to no difference in the consumption of SSBs (low-certainty evidence).

Nudging interventions that influence food presentation may increase intake of healthy foods (low-certainty evidence), may increase the selection of target foods (low-certainty evidence), and their effects on energy intake is very uncertain (very-low-certainty evidence). Nudging interventions that influence food positioning may increase the consumption of water (low-certainty evidence) and may increase the selection of target foods (low-certainty evidence). Nudging interventions that reduce portion sizes served may reduce energy intake slightly (low-certainty evidence). Implementing multiple nudging strategies likely makes little to no difference in vegetable intake (moderate-certainty evidence), probably reduces energy intake (moderate-certainty evidence), and may make little to no difference in selecting vegetables (low-certainty evidence).

Fruit or vegetable provision at school may increase their intake slightly (low-certainty evidence), likely makes little to no difference in the consumption of unhealthy foods or in energy intake (moderate-certainty evidence), and may increase BMI slightly (low-certainty evidence). Meal provision at school likely makes little to no difference in consuming healthy items at breakfast, in consuming unhealthy items, or in BMI (moderate-certainty evidence), and the effects on energy intake are very uncertain (very-low-certainty evidence). Milk provision at school likely increases selection of milk (moderate-certainty evidence).

No eligible studies assessing marketing restrictions of unhealthy food and beverages or pricing policies to promote healthier alternatives were identified. Some of the studies screened addressed these interventions but had ineligible study designs. Very few studies were conducted in low-income settings, which may affect the applicability of these interventions in these settings. Included studies did not report on 2 of the critical outcomes: nutrient and calorie content of available foods and exposure to marketing of FNABs. The latter was only relevant for 1 category of interventions for which no studies were included, which explains why it was not reported. For each included comparison, many critical and important outcomes were not reported in included studies. All of these gaps represent opportunities for future research.

Quality of the evidence

Across all comparisons, most of the evidence was of low certainty. Reasons for downgrading were related to high or unclear overall risk of bias and imprecision due to small sample sizes and large CIs that include both appreciable harm and benefit. Although most of the evidence was from HICs, which may not be representative of all relevant settings, the WHO guideline group opted not to downgrade due to indirectness.

Comparison with similar studies

Existing reviews in this field, some of which had limitations in the rigor of conduct, and which have included other types of study designs, have reported similar results to our review overall. A recent review assessing the impact of school food environment policies included 91 RCTs and quasi-experimental studies, all in HICs and with high heterogeneity¹⁸; we also included mostly studies from HICs and found the evidence base to be highly heterogeneous. This study reported that FV provision in schools slightly increased consumption of fruits or vegetables and no impact on calorie intake; our review also found little to no effect on energy intake, but we found little to no difference in FV intake. In terms of nutrition quality standards, they reported that these increased the consumption of healthy snacks and beverages and reduced consumption of unhealthy snacks and beverages, which our review also found. They reported no effect on energy intake, whereas our review found a slight reduction in energy intake with the intervention Another review assessing the effects of school wellness policies that increased access to healthy foods at schools, such as the US federally reimbursable meal programs, reported that 14 of 19 studies showed improved selection, intake, and sales of healthy foods, and decreased plate waste.¹⁹ This contrasts with our review, which found that school meal provision makes little to no difference in the consumption of healthy or unhealthy foods. A review of environmental interventions to reduce SSB consumption reported low-certainty evidence that reducing the availability of SSBs in schools was associated with reduced SSB consumption.¹⁰⁴ Our review also found that nutrition standards for food and beverages in schools may reduce the consumption of SSBs. A review of population-level interventions for obesity prevention included studies in the school setting; some of these that evaluated physical activity interventions (3/6), food/beverage environment (8/14), and all 4 evaluating both types of interventions reported beneficial effects on body weight.¹⁰⁵ We found that the evidence of effects on body weight was very uncertain for nutrition standards and FV provision, or that for meal provision there was little to no difference in effect. The obesity-prevention review also reported reduced SSB intake with state- or school-wide policies to decrease access to SSBs and a beneficial effect on FV intake with competitive food policies. These findings are similar to ours, as described above. Of 2 recent reviews evaluating nudging interventions in schools, 1 reported that results of the 12 studies were inconclusive,¹⁰⁶ and the other, which included 25 studies assessing a variety of strategies, reported that most studies showed improved selection and intake of target foods.⁵⁸ Our review found that implementing multiple nudging strategies makes little difference in vegetable intake or selection but that it probably reduces energy intake. Both studies reported problems with the quality of the evidence base, which we also report.

Strengths and limitations

Although Cochrane Review methodology was followed, potential biases could have been introduced due to pragmatic decisions we had to make during the review process.

Although we intended to search African Journals Online (AJOL), access to the database could not be secured; Africa Wide Info was searched instead. There had also been an intention to search WHO’s International Clinical Trials Registry Platform (ICTRP); however, due to COVID-19 demands this database was available only to WHO staff at the time the searches were conducted. The reference lists of included studies or systematic reviews identified were not searched, as initially intended. Given the extensive search results, it is unlikely relevant studies were missed.

Where multiple outcome measures for the same outcome were reported in the same study, measures were selected for the synthesis, which could have introduced selective reporting bias. To ensure transparency, all relevant measures reported in included studies are presented in S2 File.

When assessing risk of bias due to missing outcome data, an approach based on a threshold was used, which is no longer recommended. This could have caused the classification of more studies at risk of attrition bias.

As the main synthesis approach was based on ED, it was not possible to draw conclusions on mean effect size for many outcomes.

CONCLUSION

The body of evidence included in this review indicates that interventions that address the school food environment may have modest beneficial effects on certain key outcomes. Implementation of these interventions should consider the local context and factors that would enable or limit implementation. Most of the evidence currently included in the review is from HICs and it would therefore be important to take this into consideration when making decisions about implementation of these interventions in lower-income settings.

Acknowledgments

The authors thank Joy Oliver (Cochrane SA, SA Medical Research Council) for developing and implementing the search strategy; Anel Schoonees (Centre for Evidence-based Health Care, Stellenbosch University) for helping conduct the searches on some databases; Kerry Dwan (Cochrane Methods Support Unit) for methods advice regarding the inclusion of crossover trials; Yusentha Balakrishna and Charl Janse Van Rensburg (Biostatistics Unit, SA Medical Research Council) for support with statistical queries during data extraction; Selvan Naidoo (Cochrane SA) for support with screening titles and abstracts; Ntombi Blose (University of Cape Town) for help with drafting the characteristics of included studies document/matrix; and Elie Akl (American University of Beirut) for peer reviewing the report of this review before it was presented to the Guideline Development Group.

Funding. This review was commissioned and supported by the WHO (registration 2020/1001698–0). The WHO reviewed and approved the protocol for the systematic review and reviewed the initial report of the completed systematic review. S.D., T.K., M.W., and A.G. are partly supported by the Research, Evidence, and Development Initiative (READ-It; https://www.evidence4health.org/). READ-It (project number 300342–104) is funded by UK aid from the UK government; however, the views expressed do not necessarily reflect the UK government’s official policies. The authors are staff of Cochrane South Africa, which was commissioned to do the review by the WHO.

Declaration of interest. The authors have no relevant interests to declare.

Abbreviations: BMI, body mass index; ED, effect direction; FNAB, food and nonalcoholic beverage; FV, fruits and vegetables; HIC, high-income country; ITS, interrupted time-series; MD, mean difference; OR, odds ratio; PCS, prospective controlled study; RCT, randomized controlled trial; SSB, sugar-sweetened beverage; WHO, World Health Organization; pp: percentage points.

Supporting Information

The following Supporting Information is available through the online version of this article at the publisher’s website.

S1 File Search strategies

S2 File Results of individual studies

S3 File List of excluded studies

S4 File Summary of studies awaiting classification

S5 File Summary of ongoing studies

S6 File Characteristics of included studies

S7 File Detailed risk-of-bias assessment

S8 File Effects of interventions on outcomes other than key outcomes

S9 File Supplementary figures

S10 FilePRISMA checklist

Figure S1 Harvest plot for comparison 3

Figure S2 Harvest plot for comparison 4

Figure S3 Harvest plot for comparison 5

Figure S4 Forest plot for comparison 9: school meals provision vs no intervention. Outcome: BMI (kg/m²; prospective controlled studies)

Figure S5 Forest plot for comparison 9: provision of school meals vs no intervention. Outcome: BMI z scores (prospective controlled studies)

References