The Impact of Modifying Food Service Practices in Secondary Schools Providing a Routine Meal Service on Student’s Food Behaviours, Health and Dining Experience: A Systematic Review and Meta-Analysis

The education sector is recognised as an ideal platform to promote good nutrition and decision making around food and eating. Examining adolescents in this setting is important because of the unique features of adolescence compared to younger childhood. This systematic review and meta-analysis examine interventions in secondary schools that provide a routine meal service and the impact on adolescents’ food behaviours, health and dining experience in this setting. The review was guided by Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Checklist and Cochrane Handbook recommendations. Studies published in English searched in four databases and a hand search yielded 42 interventions in 35 studies. Risk of bias was assessed independently by two reviewers. Interventions were classified using the NOURISHING framework, and their impact analysed using meta-analysis, vote-counting synthesis or narrative summary. The meta-analysis showed an improvement in students selecting vegetables (odds ratio (OR): 1.39; 1.12 to 1.23; p = 0.002), fruit serves selected (mean difference (MD): 0.09; 0.09 to 0.09; p < 0.001) and consumed (MD: 0.10; 0.04 to 0.15; p < 0.001), and vegetable serves consumed (MD: 0.06; 0.01 to 0.10; p = 0.024). Vote-counting showed a positive impact for most interventions that measured selection (15 of 25; 41% to 77%; p = 0.002) and consumption (14 of 24; 39% to 76%; p = 0.013) of a meal component. Interventions that integrate improving menu quality, assess palatability, accessibility of healthier options, and student engagement can enhance success. These results should be interpreted with caution as most studies were not methodologically strong and at higher risk of bias. There is a need for higher quality pragmatic trials, strategies to build and measure sustained change, and evaluation of end-user attitudes and perceptions towards intervention components and implementation for greater insight into intervention success and future directions (PROSPERO registration: CRD42020167133).


Introduction
Globally in 2017, 11 million deaths and 255 million disability adjusted life years were attributable to dietary risk factors, in particular diets high in sodium, and low in fruit, wholegrains, nuts and seeds, vegetables, and seafood omega-3 fatty acids [1]. Eating behaviours track from childhood and adolescence to adulthood and across generations [2][3][4][5][6][7][8][9], and a higher intake of fruit and vegetables is protective against burden of disease [10].

Eligibility Criteria
The PICOS model was used to develop and tailor inclusion and exclusion criteria (Table 1). Randomised and non-randomised experimental trials with or without a control group (includes no intervention or comparing an intervention), and single group beforeafter studies were considered for this review. Setting-based public health interventions are often evaluated by clusters (i.e., groups such as schools), and we used Schmidt (2017) to categorise study designs for cluster-level interventions as either cluster randomised trial (C-RT), cluster non-randomised trial (C-NRT), controlled before-after study (CBA), and a before-after study without control (BA) [66]. In addition, we used the Cochrane Handbook to classify other studies as a non-randomised trial (NRT) when groups being compared were allocated based on methods outside the control of investigators, such as the allocation of groups according to the natural course of people's choice [67]. C-NRT, CBA, BA and NRT studies were included to better address the review questions and PICOS criteria as only a small number of randomised trials are available, or likely to be available in such setting-based interventions [67]. Table 1. Eligibility criteria using the PICOS model.

Inclusion Criteria Exclusion Criteria
Population Secondary (i.e., middle or high) schools that provide a routine main meal service (≥1 main meal/day) to most students (≥50%) on most days; students aged 10-19 years; generally well and independent of activities of daily living; upper-middle and high-income countries Primary (i.e., elementary) schools; before or after school care; schools that only provide optional purchases that may supplement a meal provided from home or elsewhere; people aged <10 or >19 years; high-needs populations who are acutely or chronically unwell; selection of participants based on special nutritional needs (athletes, dance groups, high or at-risk of nutrient deficiency), specific disease state or weight status Intervention Single or multi-strategy nutrition-related interventions that target and modify the practices of the routine meal service; includes nudging strategies, policy implementation, menu changes, staff training; may vary in method, duration, or mode of delivery Interventions that focus on components outside the routine meal service, e.g., introduce a new routine meal service, or target the total school food environment without specific routine meal service strategies

Inclusion Criteria Exclusion Criteria
Comparison Experimental studies with control or comparison groups (both classified as 'controlled studies' throughout review), not limited to parallel controls; single group experiments with comparison of before and after measurements Experimental studies without control or comparison data; studies with comparative data but without an intervention (e.g., menu comparison across schools)

Outcomes
Objective or subjective measures of students' food behaviours and dining experience that reflect a change in practice within the routine meal service; includes selection or consumption of a meal component (a food item, food group or nutrient), qualitative feedback, attitudes or satisfaction scores, knowledge, school meal program participation rates Measurements that do not reflect student outcomes (e.g., menu assessment) or the impact of strategies targeting the routine meal service (e.g., dietary intake from total diet, anthropometric measures for interventions that include physical activity or classroom education unrelated to the routine meal service)

Study design
Randomised and non-randomised experimental trials, single group before-after studies; peer-reviewed publications; may be a pilot study Non-peer-reviewed publications, reviews, observational studies, commentaries, editorials, conference proceedings, reports, PhD dissertations PICOS, Population Intervention Comparison Outcome Study design; PhD, Doctor of Philosophy.
Interventions were required to be implemented in a school-based setting and focused on modifying the practices of the schools' routine meal service. For example, schools that provide a daily meal program to students, or boarding schools rather than canteen purchases over the counter. For this review, the meals needed to be provided to most students (≥50%), rather than optionally to target groups, to capture those students who are repeatedly exposed to the intervention. At <50% participation our assumption is that most students are consuming their dietary intake outside of that routine meal service, and therefore changes to the food service would have less impact. To meet the population criteria, a study had to clearly indicate students' level of participation (i.e., text indicating most students participated or a given participation rate of ≥50% of enrolled students). If not clearly indicated, assumptions were made for relevant interventions in countries where school meal programs are known to be provided to most students: (1) Finland, where all students attending primary or secondary school are entitled to a free daily school meal [37], (2) Sweden, where a free daily school meal is offered to all students aged 7-16 years and to most aged 16-19 years [68], (3) France, where 64% of middle and high school students eat a school lunch at least three times per week [34], and (4) US, where 95% of all schools and 58% of enrolled students participate in the NSLP [31,69]. Studies were restricted to those undertaken in upper-middle-and high-income countries as defined by the World Bank Group [70] to reduce heterogeneity and increase generalisability among schools in these countries with similar nutrition governance and investment in meal services at school. Future reviews should address the needs of low-and lower-middle income countries due to the increasing prevalence of school feeding programs [43].

Data Extraction and Management
A data collection form was adapted from Cochrane Handbook recommendations [71] and the following extracted by the first author (E.M.) and checked by at least one other author (A.H., M.H., E.D.): study design, setting, duration and purpose, student details, inclusion and exclusion criteria, intervention details and duration, analytical methods, range of outcomes examined related to a meal component (this included a food item, food group or nutrient) or school meal program participation rate or other student measures related to the meal service, measurement tools and tool scoring, key findings and limitations. If a study included both relevant and irrelevant data components, according to the eligibility criteria, only relevant data was extracted for analysis and reporting. For example, where a study included both elementary and middle schools, and outcomes reported separately by school type, only middle school data was extracted for analysis. Study data were tabulated and managed in Microsoft Excel version 2206 (Microsoft Corporation, Redmond, WA, USA); data for the meta-analysis were managed using REDCap version 12.5.5 (Research Electronic Data Capture; Vanderbilt University, Nashville, TN, USA) tools [72,73] hosted at Hunter Medical Research Institute, Newcastle, NSW, Australia.

Risk of Bias and Quality Criteria
Studies included in this review were assessed for risk of bias and quality by two authors independently (E.M., A.H., M.H., E.D., K.P.) using the Quality Criteria Checklist (QCC) for primary research according to the Academy of Nutrition and Dietetics Evidence Analysis Manual [74]. This critical appraisal tool allows assessment of multiple study designs and identifies sub-questions that are the most important quality considerations for each study design. Each relevant study was rated on validity (10 questions) for the scientific soundness of the investigation, assigning each question as 'yes', 'no', 'unclear' or not applicable (NA). The most important sub-questions by study design were prioritised for each validity question according to guidelines outlined in the Evidence Analysis Manual. Overall, a study was rated as 'positive' when ≥6 questions (including 4 designated priority questions) were answered 'yes'. If all 4 priority questions were not answered as 'yes' but ≥6 questions overall were 'yes' the study was rated as 'neutral'. If ≥6 questions were answered 'no' the study was rated as 'negative' and excluded from the review.

Data Analysis
A narrative synthesis was adopted to classify intervention strategies and identify and group outcomes measured. To assess the impact of interventions, meta-analysis, votecounting synthesis based on the direction of effect, or a narrative summary were performed.

Classification of Intervention Strategies
One author (E.M.) used the NOURISHING framework [75] to classify intervention strategies according to the frameworks' suite of three key domains: (1) food environment, (2) food system, and (3) behaviour change communication; and accompanying ten action areas. The framework applies a socio-ecological and comprehensive approach to capture both environmental and behavioural strategies to promote healthier eating, improve dietary behaviours and optimise health [75,76], and has been used in recent reviews to classify intervention strategies [77,78]. Within this classification, and where relevant, behavioural economics theory [79] was also applied to categorise 'nudging' strategies to influence food decisions towards healthier choices (similar to other reviews [46,50]) according to concepts of: (1) acceptability: to address palatability and taste expectations, (2) accessibility: addresses the placement and convenience of healthier options, (3) availability: providing adequate variety of healthy options and limiting less healthy items, (4) presentation: improvements to the dining room and display of food, and (5) promotion: includes marketing strategies, activities and material.

Grouping of Outcomes Measured
Interventions could contribute multiple outcomes, and each eligible nutrition-related outcome was categorised according to pre-specified outcome domains from a post hoc review of included studies: (1) student selection of a meal component, (2) student consumption of a meal component, (3) health status, (4) knowledge, (5) meal program participation rate, or (6) attitudes and perceptions related to changes to the meal service. Measurements of knowledge were included when education or promotion related to modifications to the meal service, thereby indirectly contributing to food behaviours in the dining room. Meal program participation rates reflect student acceptability of a meal service at the population level. Measurements assessing attitudes and perceptions allowed for additional insight into the dining experience including evaluation of the cafeteria environment, meal quality and palatability, qualitative feedback, sensory attributes or student satisfaction. Acceptable methods from Cochrane Handbook were applied to conduct all metaanalyses [80][81][82]. A meta-analysis was performed where possible to pool post-intervention scores of parallel arm controlled trials (randomised or non-randomised) with change-frombaseline scores (where baseline scores act as comparator) from BA studies (i.e., single group) and from the intervention arm of CBA studies (i.e., rather than a comparison of post-intervention scores between groups) [80,82]. Manipulation of data and analyses were conducted in Stata version 15 (StataCorp LLC, College Station, TX, USA) [83]. To be eligible for inclusion in a meta-analysis, at least two studies were required to report a pre-specified outcome with a common scale of measurement [80] for a meal component that was reported separately as either fruit, vegetables, milk or entrée (the primary component of a NSLP meal that contains grains, meat, vegetables and/or fruit; e.g., chicken salad sandwich, tacos or spaghetti [84]). For this analysis, pre-specified outcomes included: (1) percent of students selecting a meal component, (2) percent of serve consumed of a meal component by students, (3) mean number of serves selected per student per day, or (4) mean number of serves consumed per student per day. A 'serve' reflects a standardised portion; for example, a piece of whole fruit or prespecified weight or volume of fruit or vegetables. Each outcome per meal component was analysed using a separate meta-analysis model, and interventions could contribute multiple outcomes. For example, one intervention may contribute three outcomes such as percent of students selecting fruit, mean number of fruit serves selected by students per day, percent of fruit serve consumed when selected by students. For studies with multiple intervention arms, each intervention arm was treated separately for analysis using change-from-baseline scores. We hypothesise a random distribution of estimate effects for each outcome, because while outcome estimates for studies are related, there are noted differences in population characteristics, study designs, and the way outcomes are measured for each paper. Statistically, to estimate a random effect, a large enough number of studies must be combined; there is no universal recommendation for the minimum number of studies needed to perform a random effects meta-analysis [80], so we have used a cutoff of 5 or more studies. For those outcomes combining less than 5 studies, a fixed effects model was performed (which assumes a common estimate effect), however the results should be interpreted cautiously as they are not then generalisable to similar other studies. Combined estimates are presented as odds ratios (OR) and 95% confidence intervals (CI), and estimates of heterogeneity (I squared) are presented.
For continuous outcomes, where a standard deviation (SD) was not reported, alternative statistics (standard error, CI, t-statistic or p-value) were used to calculate a SD [82]. Where studies measured the percent of students selecting a meal component, dichotomous variables were determined for number of students that 'selected' or 'did-not-select' meal component for analysis. For both continuous and dichotomous outcomes, sample sizes were based on number of observations per study. We were not able to identify individual student level data because aggregate group level data were provided, and the number of observations during a data collection time point may represent multiple measurements per student. The meta-analysis pooled data from school cafeteria records and/or researcher direct observation of lunch trays depending on data collection methods; the former presenting a much larger sample size for analysis. In some studies, where limited sample sizes were provided, an estimate sample size was calculated where possible using reported data (e.g., from school cafeteria records: number of data collection days per arm, number of schools per arm, mean student enrolment per school, and proportion of students selecting a school lunch). Where there were sufficient clusters per arm (more than five), a design effect was applied to estimates assuming an intra-cluster correlation coefficient (ICC) of 0.05 [81]. We have chosen to ignore a clustering effect when there were five or fewer schools per arm because there are not enough clusters to account for within-cluster correlation. Where studies measured more than one change-from-baseline time point following inter- vention implementation, the meta-analysis prioritised mean values measured during the intervention period rather than follow-up measurements after the intervention ceased.

Vote-Counting Based on the Direction of Effect
Because numerous studies could not be included in the meta-analysis modelling (due to limited information about effect estimates, sample size, or without one of the pre-specified outcomes for meta-analysis), vote-counting based on the direction of effect method was used to synthesise results of all included studies that provided information on direction of effect [85]. Manipulation of data and analyses were conducted in Microsoft Excel version 2206 (Microsoft Corporation, Redmond, WA, USA). Vote-counting based on direction of effect and preparation of the effect direction plot followed Cochrane Handbook guidelines and methods according to Boon and Thomson (2021) [85,86]. The direction of effect (where provided) of each eligible outcome was recorded as either a positive impact (favouring intervention), negative impact (favouring comparator) or no change. Similar outcomes were combined into pre-specified outcome domains. For studies with multiple outcomes within a given outcome domain, a direction of effect was reported where a clear majority (≥70%) of outcomes reported similar direction (i.e., either positive or negative). If <70%, direction of effect was reported as no change or mixed effects [86]. For studies with multiple outcomes that included whole food items alongside measurements of their macroand micronutrient parts, a sensitivity analysis was conducted to exclude measures that would overinflate the vote count. A sign test was performed for each outcome domain using the count of positive and negative effects (excluding no change/mixed effects) to determine any evidence of effect along with a 95% CI estimation for binomial proportions using the Wilson interval method [85][86][87]. For studies that incorporated change from baseline scores (controlled or uncontrolled), within-group results (i.e., before and after measurements) were prioritised for inclusion in analysis, otherwise post-intervention scores were used for parallel arm or crossover trials. To assess the robustness of synthesised results, a post hoc sensitivity analysis was conducted for outcome domains apportioning variables for study quality, study design, intervention duration, number of NOURISHING framework domains or action areas, student engagement, and behaviour change communication strategies that include promotional activities or student and/or staff training.

Narrative Summary
A narrative summary of results is provided for studies or specific outcomes in studies that were not eligible for inclusion in the meta-analysis or vote counting synthesis. Due to the range of intervention strategies in the school dining room, a post hoc analysis of intervention components warranting closer examination is also provided.

Results
Our results highlight a range of study designs and intervention strategies that were implemented in the school dining room setting. While the selection and consumption of meal components were the most frequently measured outcomes, measurement of attitudes and perceptions related to the changes to the meal service provide useful insight into student experiences and intervention success. To assess the impact of interventions, the meta-analysis, vote-counting and narrative synthesis found no trend associated with study design or quality. However, the assessment did highlight the importance of the school food environment as an ideal platform to improve nutrition by uncovering trends associated with certain intervention components according to the NOURISHING frameworks' suite of domains and action areas.  school food environment as an ideal platform to improve nutrition by uncovering trends associated with certain intervention components according to the NOURISHING frameworks' suite of domains and action areas.   Table 2 presents study characteristics and intervention components classified according to the NOURISHING framework's domains, and corresponding details of strategies implemented. Included studies were conducted predominately in the US (31 of 35, 88.6%). Two studies were conducted in the United Kingdom, one in Sweden and one in France. Study designs included C-RTs (n = 4, 11%), C-NRTs (n = 3, 9%), NRTs (n = 5, 14%), CBA (n = 7, 20%) and BA studies (n = 16, 46%). Over one-third of the studies were pilot studies (13 of 35, 37%; 2 of these were randomised). All studies were pragmatic experimental trials conducted in real-life routine practice conditions within the dining room of secondary  Table 2 presents study characteristics and intervention components classified according to the NOURISHING framework's domains, and corresponding details of strategies implemented. Included studies were conducted predominately in the US (31 of 35, 88.6%). Two studies were conducted in the United Kingdom, one in Sweden and one in France. Study designs included C-RTs (n = 4, 11%), C-NRTs (n = 3, 9%), NRTs (n = 5, 14%), CBA (n = 7, 20%) and BA studies (n = 16, 46%). Over one-third of the studies were pilot studies (13 of 35, 37%; 2 of these were randomised). All studies were pragmatic experimental trials conducted in real-life routine practice conditions within the dining room of secondary schools (includes middle and high schools; students aged 10-19 years). Publication dates ranged from 1985 to 2021. Duration of interventions ranged from one day to two years.

Risk of Bias and Quality Assessment
Nine studies (26%) were rated as 'positive' indicating the studies adequately addressed issues of inclusion/exclusion, bias, generalisability, data collection and analysis. The remaining 26 studies (74.3%) were rated as 'neutral' indicating they are neither exceptionally strong nor exceptionally weak; no studies were rated as 'negative'. Inadequate description of handling withdrawals, lack of blinding, and missing data were the main risks of bias in those studies rated as neutral. Full quality assessments are provided (Supplementary  Materials Table S1).

Intervention Strategies
A total of 42 interventions were implemented across 35 studies (7 studies included 2 intervention arms). Table 3 presents the classification of intervention strategies from included studies using the NOURISHING framework. Table 4 summarises the number of domains and action areas targeted per intervention. All interventions targeted at least one action area from the food environment domain; mostly the provision of healthy food in some form (i.e., policy implementation, reformulation of recipes or menus, increased availability of healthy options or reduced availability of less healthy options) and/or choice architecture to nudge students towards healthier food choices. Twenty-five interventions targeted the food system domain (60%) which included engagement with stakeholders across the food service (students and staff) and procurement of healthier ingredients, and 29 interventions targeted at least one action area from the behaviour change communication domain (69%). Overall, 22 of 42 interventions (52%) included components across all 3 domains, 10 interventions (24%) across 2 domains, and 10 interventions (24%) across 1 domain only.

Outcomes Measured
A detailed assessment of outcomes measured, measurement tools and scoring, results, major findings and limitations are provided (Supplementary Materials Table S2). Most interventions contributed multiple outcomes that were eligible for inclusion. Selection and/or consumption of a meal component/s were the most frequently measured outcomes (in 35 of 42 interventions, 83%). Fifteen interventions measured selection and consumption, 11 measured selection only, and 9 measured consumption only. Outcomes (categorised according to pre-specified outcome domains) included:

3.
Health status: Blood pressure (BP) was measured in n = 1 intervention to assess the impact of reduced sodium in school meals. Body mass index (BMI) was measured in n = 1 intervention to assess the impact of interactive kiosks to guide student lunch choices.

4.
Knowledge: One study (n = 2 intervention arms) measured knowledge about fish before and after an intervention that aimed to increase students' intake of fish at school and included classroom education about fish preparation in the school kitchen.

5.
Meal program participation rate: assessed in n = 5 studies and represents the proportion of enrolled students that participated in the school meal program pre-and post-intervention, reflecting population level selection/acceptance of the school meal program without separating components of the meal program or reflect consumption. 6.
Attitudes and perceptions related to changes to the meal service: assessed in n = 15 interventions (n = 13 with before and after measurements) to assess students' attitude toward school lunch and the cafeteria, acceptability of modified or new menu items, or feedback on intervention components.

3.
Mean number of serves of a meal component selected per student per day (Supplementary Materials Figure S3a): Two separate meta-analyses were prepared for fruit (n = 4 studies) and vegetables (n = 4 studies). The pooled effect showed interventions increased the number of fruit serves selected per student per day (MD: 0.09; 95% CI: 0.09, 0.09; p < 0.001), with no change in vegetable serves selected (p = 0.977). The pooled estimate for both fruit and vegetable serves selected per student per day is not a good representation due to the large sample size for one study (Bogart et al., 2014 [88]; n = 102,262) that highly influenced the pooled estimate (weighting > 99%).

Vote Counting Based on the Direction of Effect
Forty one of 42 interventions and four outcome domains (selection and consumption of a meal component, meal program participation rate, attitudes and perceptions) were eligible for inclusion in vote counting based on direction of effect analysis; nine were judged to be at low risk of bias with either positive or mixed effects across outcome domains. Two outcome domains (health status and knowledge) were excluded because they only included one study, or outcomes within the domain were not suitable to combine. Figure 2 presents the effect direction plot for eligible outcome domains and includes intervention duration and components according to the NOURISHING framework's domains and action areas. There was evidence that interventions that modified the routine meal service had an impact on (1) student selection of a meal component, with 15 of 25 interventions reporting a positive impact (60%; 95% CI 41% to 77%, p = 0.002), 2 negative, and 8 mixed effect (32%), and (2) student consumption of a meal component, with 14 of 24 interventions reporting a positive impact (58%; 95% CI: 39% to 76%, p = 0.013), 3 negative, and 7 mixed effect (29%), and (3) meal program participation rate, with 3 of 5 interventions reporting a positive impact (60%; 95% CI: 23% to 88%) and 2 negative, and (4) attitudes and perceptions related to changes to the meal service, with 9 of 13 interventions reporting a positive impact (69%; 95% CI: 42% to 87%, p = 0.267) and 4 negative. Nine interventions were judged to be at low risk of bias; 5 of 8 (63%) favoured the intervention for selection of a meal component, 3 of 3 (100%) showed mixed effects for consumption of a meal component, and 3 of 3 (100%) favoured the intervention for student attitudes and perceptions. NOURISHING framework action areas: there was evidence that interventions targeting more action areas (≥3) had a greater impact on selection and consumption of a meal component compared to interventions that targeted less (≤2): selection, 11 of 16 with more action areas favoured the intervention (69%; 95% CI: 44% to 86%, p = 0.006), compared to 4 of 9 with less (44%; 95% CI: 19% to 73%, p = 0.375); consumption, 11 of 17 with more action areas favoured the intervention (65%; 95% CI: 41% to 83%, p = 0.022), compared to 3 of 7 with fewer (43%; 95% CI: 16% to 75%, p = 0.625).

Narrative Summary
The health status outcome domain included two outcomes from two studies that were not similar to combine for analysis. Ellison et al. [100] measured BP (in addition to sodium intake). The intervention showed a significant improvement in systolic (−1.7 mmHg; 95% CI: −0.6, −0.29; p = 0.003) and diastolic BP (−1.5 mmHg; 95% CI: −0.6, −2.5; p = 0.002). The knowledge outcome domain included one study with two intervention arms [101]; both measured and significantly increased pre-to post-intervention student knowledge about fish (p < 0.001). Bean et al. [102] was excluded from all previous analyses due to limited data and direction of effect not reported. The impact of food service staff training on implementing be-havioural economics strategies showed no change in student sales of fruit (p = 0.150), vegetables (p = 0.245), salad bar (p = 0.525), milk (p = 0.245) or water (p = 0.986).
All interventions targeted at least one action area from the NOURISHING frameworks' food environment domain, and components warrant closer examination in respect to four outcome domains (selection and consumption of a meal component, meal program participation rate, attitudes and perceptions). Notable findings include the benefits associated with school-food policy implementation, increasing the availability and accessibility of healthy options, and reduced availability of less healthy options. For example, Cullen et al. [103] mandated restrictions on less healthy options and increased consumption of vegetables, fibre, vitamin A, vitamin C, calcium, sodium (all p < 0.025) and the percentage of energy of the lunch meal consumed from fruit, vegetables and entrée (p < 0.002) [104]. Schwartz et al. [92] implemented updated NSLP nutrition standards and increased vegetable and entrée consumption (p < 0.05). Bhatia et al. [44] improved meal program participation at all participating schools (statistical significance not assessed) after removing competitive offerings and expanding and promoting NSLP options, Boehm et al. [96] removed competitive foods and increased number of entrees served daily compared to control (p < 0.05), and Madden et al. [105] placed restrictions on less healthy options and increased consumption of fruit and vegetables (p < 0.001). The introduction of a fast service lane offering pre-plated healthy options increased service speed (p < 0.01) and students were satisfied with the service speed and meal quality [106]. Installation of water jets near the lunch line improved students' water-drinking behaviours [107]. Strategic placement of healthier options [91,[96][97][98]108] and pre-sliced fruit [88,91,96,98,[108][109][110][111] were mostly effective in increasing selection or consumption of meal components. A novel intervention that integrated technology in the dining room for students to visualise and select a balanced meal increased the proportion of students selecting fruit and vegetables (p < 0.05; consumption not assessed) [112].
To improve the nutritional quality of meals, eight studies engaged a professional chef or dietitian to guide recipe or menu reformulation, staff training, and/or facilitate promotional events for students [89,93,94,98,105,[113][114][115]. Three of these studies included student taste-testing of modified recipes; spices and herbs were added to NSLP vegetable dishes with mixed effects on consumption [113], a chef modified pizza and burger recipes and increased vegetable consumption (p < 0.005) [93], and new vegetable-focused entrée recipes increased selection (p < 0.001) [94]. Three studies included staff training to modify recipes resulting in increased consumption of vegetables (p < 0.01) [89], reduced consumption of sodium (p < 0.001) and saturated fat (statistical significance not assessed) [115,116], and improved nutritional quality of the lunch meal and increased overall fruit and vegetable consumption (p < 0.001) [105]. The remaining two studies engaged a dietitian to implement menu change goals and improved fruit and vegetable selection (statistical significance not assessed) [114], and as part of a team to support implementation of choice architecture strategies which had mixed results across meal components but increased consumption of fruit excluding juice (p < 0.05) [98]. New point-of-service system, additional staff for line control, a la carte line removed and re-purposed for NSLP b.
A la carte options removed, expanded NSLP options, add salad bars and refrigerators, student taste testing, installation of student-designed mural, designed and posted new menus c.
Students engaged for taste testing, surveys and mural design; staff consultation for design and implementation of initiatives d.
Branded and marketed former a la carte locations; student taste testing e.
Training on NSLP rules        Professional chef engaged to use ingredients available in school cafeteria to develop 3 types of pizza (meat taco, bean taco, garlic spinach) and a ranch flavoured burger; new chefs lunch items available in cafeteria on 1 d c.
Engaging students during after-school event d.
After-school event for students to taste the chef's lunch on offer the following day, meet the chef, talk about her profession and new recipes created Apples offered to all students for free in addition to the regular lunch meal as (1) phase 1, whole apples, (2) phase 2, pre-sliced apples, and (3) phase 3, whole and pre-sliced apples; for all interventions fruit placed at the end of lunch line for students to take as many whole and/or pre-sliced apples as they wanted  Grade 6 students consulted for development of promotional posters; grade 6-8 students voted for best posters for display in school cafeteria c.
Teachers implemented a standards-based curriculum on sustainable food systems d.
Posters promoting waste reduction displayed in school cafeteria   • Changes to food service operations such as modifications to the point of service or service lines • Availability-increased variety or expand healthy options (includes offering pre-sliced fruit in addition to whole fruit); reduced availability of less healthy food and beverages

Figure 2.
Effect direction plot summarising direction of student food behaviours in the dining room from studies that modified the practices of the routine meal service at secondary schools. LEGEND: Study design: C-RT, cluster randomised trial; C-NRT, cluster non-randomised trial; CBA, controlled before after study; BA, before after; NRT, non-randomised trial; studies include pre-post scores for single or multiple arm trials unless indicated as parallel arm or crossover beside study design. Study quality according to the Academy of Nutrition and Dietetics Quality Criteria Checklist [74]: denoted by row colour: green = positive rating; amber = neutral rating. Effect direction: upward arrow = positive impact, downward arrow = negative impact, sideways arrow = no change or mixed effects for multiple outcomes. Subscript numbers: Number of outcomes within each category synthesis is 1 unless indicated in subscript beside effect direction. Sign test: excludes studies with mixed effects direction as they cannot be said to represent either a positive or a negative effect direction [86]. 95% CI (confidence interval): estimation for binomial proportions using the Wilson interval method [85]. y: year/s; m: month/s; w: week/s; d: day/s. SLHE: school lunch plus home economics intervention; SL: school lunch intervention; +: indicates the intervention has included components from the nominated NOURISHING framework domain [44,[88][89][90][91][92][93][94][95][96][97][98][99]101,.  LEGEND: Study quality: variables apportioned per risk of bias assessment results as either, (1) positive rating, or (2) neutral rating; Study design: variables apportioned according to measurement scores used for analysis as either, (1) pre-post measurements = intervention arm before and after scores, or (2) parallel arm or crossover = comparison of post-intervention scores; Intervention duration: variables apportioned according to duration of intervention implementation as either, (1) ≤2 months, or (2) 3+ months; NOURISHING domains: variables apportioned according to number of NOURISHING framework domains as either, (1) interventions targeting 3 domains, or (2) interventions targeting 1-2 domains. NOURISHING action areas: variables apportioned according to number of NOURISHING framework action areas as either, (1) interventions targeting 3-6 action areas, or (2) interventions targeting 1-2 action areas; Stakeholder engagement: variables apportioned according to presence of stakeholder engagement during intervention development and/or implementation as either, (1) with students, or (2) without students; Behaviour change communication: variables apportioned for interventions as either, (1) including promotion and/or training, or (2) without promotion and/or training.

Discussion
Our systematic review included 42 interventions across 35 studies that focused on modifying the school's routine meal service. Results from our meta-analysis indicate significant improvements in student's fruit and vegetable selection and consumption. The vote-counting synthesis found more than half of the interventions had a positive impact on selection and consumption of a meal component, program participation rate, and attitudes and perceptions related to changes. There were only a few studies that assessed health outcomes and service speed, all of which showed promising benefit. These results support existing evidence that nutrition interventions targeting the school food environment can improve students' food behaviours, health and dining experience [39,41,49,58].

Interpretation of Results
Utilising the NOURISHING framework to unpack each intervention strategy allowed an examination of their component parts across environmental and behavioural contexts. In other words, each strategy was not limited to one domain and one action area. For example, Hanks et al. [97] introduced and promoted a healthy convenience line, improved presentation and accessibility of fruit and vegetables, and implemented staff prompts (3 domains, 4 action areas). This examination was useful to identify trends in intervention impact according to the component parts, as previous literature has recognised the challenge of identifying the 'active' ingredient within multi-strategy interventions [58,126,127]. In particular, we found interventions that showed most impact incorporated student and/or staff engagement, targeted more NOURISHING framework domains or action areas, increased accessibility of healthier options, or reduced the availability of less healthy options. This highlights opportunities to scale-up future nutrition interventions in this setting by utilising the NOURISHING framework in program design and development. Importantly, the interventions showing less impact were those that excluded collaboration with key stakeholders; primarily the staff who prepare the food, and students who con-sume the food. Overall, the range of intervention strategies for examination was diverse which is similar to other reviews examining the school food environment [39,49,58,60,63]. There were novel interventions (for example, interactive kiosks to encourage healthy lunch choices [112]), simple interventions (for example, installation of water jets near the lunch line to increase water consumption [107]), and multi-strategy interventions (for example, integrating behavioural economics, staff training and promotional activities).
Selection and consumption of a meal component (together or separately) were the predominate outcomes measured in our review, which is consistent with recent reviews [39,[46][47][48][49]. While selection does not accurately reflect or guarantee consumption, food cannot be consumed until it's selected. Therefore, both measures can provide important insight for decision making around food choices or dietary intake and food waste. For example, two high quality and controlled studies warrant comparison. Firstly, Cohen et al. [89,119] engaged a professional chef to improve menu quality and modify cooking techniques to enhance palatability. Increasing selection was effective for only one of six meal components (wholegrains), but students consumed a greater proportion of the vegetables they selected. While change in selection was limited, the results highlight good news for palatability and diet quality (increased fibre, vitamin A, and vitamin C, and reduced saturated fat), and reduced food waste. Secondly, Cullen et al. [90] investigated changes in student food selection and consumption in response to the updated United States Department of Agriculture (USDA) School Meal Standards that allowed increased servings of fruit and vegetables. Students selected more fruit and vegetables; however, they did not consume a greater proportion what they selected, resulting in more food waste. This comparison reinforces the importance of addressing palatability. The best interventions will increase consumption without increased waste through additional selection that is only to be discarded.
In other studies (not included in this review), Cohen and colleagues advocate for engagement with professional chefs following their examination of a chef-initiative to enhance palatability of school meals, on student consumption in elementary and middle schools [128,129]. The studies included both short-and longer-term exposure to chefenhanced meals, and choice architecture components were incorporated in combination and separately to examine their impact. Overall results (not separated by school type) showed that longer-term exposure to the chef initiative increased consumption of both fruit and vegetables. Interestingly, there was no effect on consumption during the short-term exposure to the chef initiative or the 'choice architecture only' intervention. The missing link that may further entice students to select and consume reformulated meals in the short-and longer-term is student sampling of new recipes. Other reviews highlight, and advocate for, the potential synergistic impact of engaging professional chefs and dietitians, improving staff skills and active involvement with students through food preparation and sampling to address palatability and improve student dietary intake [49,59].
Measurements of attitudes and perceptions contributed useful insight into students dining experience. For example, Fritts et al. [120] found with repeat exposure, students were willing to eat vegetables again that had added herbs and spices, Sharma et al. [106] found students were satisfied with service speed and the quality of healthier meals made available from a fast service lane, and Koch et al. [124] found students had a positive attitude toward changes to the dining room. Furthermore, several authors commented on aspects of intervention success in their discussion. While clear evidence was not provided, these comments add insight into stakeholder and end-user experience within the school dining room of pragmatic trials. For example, Chu et al. [118] commented that food service staff were not provided with instructions for preparing alternative lunch products, and Just et al. [93] highlighted the taste-testing event was an integral part of the overall experience for students. Previous studies have recognised a multitude of other factors that influence the success (or not) of school-based interventions such as acceptability, implementation fidelity, organisational and staff readiness to participate, adequate training and collaboration between stakeholders [130,131]. Additionally, this can contribute to future directions. The school mealtime environment and dining experience is important to foster positive attitudes and values, build skills, promote health and socialise with peers [132]. Therefore, measurements of stakeholder attitudes and perceptions need to be considered in this setting. In particular, to further explore adolescents' views on what can and should be done to improve nutrition and the food culture within a school dining room.
The findings from the meta-analysis did not indicate any clear pattern of effect according to study quality, study design or intervention duration. Eligible studies for the meta-analysis were limited and varied in intervention duration and types of strategies implemented. The significant improvements across meta-analyses were small on each occasion. We suggest all of these factors contributed to the high heterogeneity, and therefore limit the confidence in pooled estimates. Further examination using the NOURISHING framework in conjunction with sensitivity analyses of intervention effect direction, identified trends in study methodology or intervention strategies. Shorter interventions were more beneficial than longer interventions which may suggest a novelty effect. This concurs with other reviews of school-based interventions that found longer interventions are not necessarily more effective [58,63]. Our results support recommendations to explore strategies that maintain momentum to build sustained change [63]. Again, we suggest evaluation of stakeholder satisfaction and implementation fidelity in the school dining room where most students have daily exposure to the intervention. This could provide useful insight into sustained engagement, proper processes and a feedback loop (potentially to modify design and re-implementation) to maintain momentum over longer-term periods. Another notable trend showing more benefit included interventions incorporating student engagement. Of note, three studies evaluated a community-based participatory research program in middle schools via a pilot, efficacy and dissemination test [88,109,110]. On each occasion, peer advocacy was a key component, initially adopted in the pilot study, and guided by the 'diffusion of innovation' theory that suggests ≥15% of the target population be trained as advocates [109]. These findings contribute to evidence that advocate for stakeholder engagement, in particular the adolescents themselves and food service staff, throughout the intervention process for enhanced success [58,[133][134][135]. The clinical significance associated with 'active involvement' and collaboration between stakeholders was recognised in several included studies [93,102,117,125].
We note interventions that targeted more NOURISHING framework domains or action areas were more beneficial than those targeting fewer which is consistent with the frameworks' recommendations for a comprehensive response to effectively promote healthy eating [76]. However, we cannot ignore the success of some interventions that targeted fewer. For example, implementing updated school nutrition standards is not a simple strategy, and such mandated change can improve targeted food behaviours in the school dining room as evidenced in three included studies [90,92,103]. While these studies varied in levels of success and outcomes measured, barriers and enablers were not evident. Previous reviews have highlighted enablers to successful school food policy implementation include adequate support from schools, positive staff attitudes, training, collaboration, communication, tailoring to local contexts and adequate planning [136,137]. This reinforces the concept that multiple factors can impact intervention success in the school-based setting. We suggest incorporating qualitative work that could provide important insight to better understand the barriers and enablers to intervention success.
Other notable findings include the benefits evident from increasing accessibility of healthier options. This review exposed a variety of strategies, illustrating how accessibility of healthier options can take many forms and are important in a school setting where the lunch break time is limited. The placement and convenience of healthier options can encourage selection and reduce time in the lunch line. This allows more time to sit and eat lunch, which has been shown to improve consumption of fruit, vegetables, entrees and milk, and therefore reduced waste [138,139]. Qualitative studies exploring adolescent perceptions of school food indicate lengthy queues are a barrier to a positive dining experience, access to fresh food a facilitator, and availability of less healthy options detract from healthier choices [140][141][142]. Most studies in the current review were conducted in the US where competitive foods (often less healthy options from alternative menu offerings, school stores, snack bars, or vending machines that compete with school meal program participation) have a significant presence in schools. In our review removing competitive foods or reducing less healthy options had beneficial results [44,103]. This supports other evidence that such strategies can improve student participation in meal programs [143,144], and selection and consumption of healthier food [145], therefore improving diet quality. Interestingly, adolescents have identified what they were being taught at school conflicts with the availability and accessibility of unhealthy food offered at school, sending mixed messages to students and giving rise to unsupportive food environments [141,146]. This highlights the importance of including adolescents as key stakeholders who can offer valuable insights into where the individual issues lie, and they should be engaged in co-designing the solutions.
Promotion strategies included events for student sampling, awareness campaigns, nutrition labelling, dining room signage, peer advocates, staff verbal prompts, school announcements. While it is difficult to understand exactly how effective these strategies are without direct feedback from staff or students, they contribute visual and strong marketing appeal, and potential synergistic impact alongside other strategies, to steer decision making or thoughts about food and eating. An Australian study of 12,188 secondary students aged 12-17 found that cumulative exposure to food marketing was positively linked to adolescents' food choices and eating behaviours [147], and another study reframed food marketing to reject junk food in favour of healthier alternatives, and found positive and sustained change in dietary attitudes and food choices [148]. Knowing the social ecology of adolescents lives has changed rapidly with the emergence and globalisation of all forms of media and communication [149], which now forms part of their daily lives, this may be a promising area for closer examination within a schools dining room.

Limitations
Most studies in this review were repeat cross-sectional by design, a large proportion were before-after studies or non-randomised trials, and many were pilot studies. These study designs are not methodologically strong, and generally at higher risk of bias, placing limitations on the representativeness of results and conclusions about why and how changes occurred. However, they do provide an estimate of the likely impact of interventions, and in our review, the benefit (or not) of their component parts using the NOURISHING framework. Future work can build the evidence with larger-scale pragmatic trials under real-world conditions for measures of effectiveness driven by a sample that includes more schools along with the use of validated measurement tools and adequate statistical analyses. Low-and lower-middle income countries were excluded from our review to increase generalisability among schools in countries with similar nutrition governance. Feeding programs in these regions typically focus on micro-nutrient deficiencies and undernutrition. However, the coverage of school feeding programs in low-and lower-middle income countries is increasing and should be considered for future reviews. Interventions were required to focus on modifications to the meal service at school, and eligible outcomes for the selection or consumption of a meal component were restricted to measurements within this setting rather than habitual intake. This means some studies in secondary schools that modified food service practices alongside physical activity strategies and/or classroom nutrition education and measured dietary intake from total diet were excluded.
The short duration of some interventions limits reliability and strength of results. Without longer-term implementation and follow-up, the results may indicate a novelty rather than any sustained effect. For example, Wansink et al. [111] implemented an effective single-strategy sliced-apple intervention over one month in intervention schools with two days data collection (one week apart) to assess student selection and consumption. While as a stand-alone study these results have limitations, the findings contribute to the formative research (interviews with elementary and middle school students) and pilot study indicating student preference for pre-sliced fruit. Furthermore, a recent review found the majority of studies offering pre-sliced fruit (6 of 8) had a positive association with fruit consumption [39]. Another example included in this review by the same group of researchers examined whether promoting and incorporating school garden produce into school salads impacted student selection and consumption. This pilot study was conducted over one day only and resulted in an increase in salad selection but not consumption [95].
Most studies were not eligible for meta-analysis due to missing data such as sample size, inadequate statistical analysis (e.g., statistical significance often not assessed for outcome measures, meal program participation rates or percentage of students selecting food items) or a unique outcome measure or unit of measurement not common across other studies. High levels of heterogeneity were evident. Contributing factors include the range of study designs, type and duration of interventions. There were limitations associated with data synthesis using vote-counting based on direction-of-effect which is less powerful than methods combining statistical significance. The binary measure of effect as either 'positive' or 'not' does not account for varying sample sizes or provide information on the magnitude of effect. However, this synthesis method was appropriate for this review due to the inconsistency of available data across all studies [85].
Some studies that included primary and secondary schools disaggregated some, but not all outcome data by school type or age group. Therefore, some study's results were excluded from this review [102,107,118]. This limitation has been identified in other reviews that highlight the importance of disaggregating data to recognise differences in food behaviours according to life stage [49], in this case the differences between childhood and adolescence characterised by unique biological change, increased autonomy, peer influence and social development that can influence their decisions around food and eating [149].
Most studies were conducted in the US, similar to other reviews that have assessed nutrition-related interventions at school [46][47][48][49]62]. There is a need for studies across a wider variety of settings including boarding schools and school meals outside the US that target the adolescent population. Well-designed and higher quality pragmatic trials are required. There is an opportunity for enhanced impact through engagement with end users (including the students and staff) in the design of new and novel interventions. For example, one of the most successful trials in this review used an interactive kiosk and a points system similar to a computer game to select a balanced lunch meal [112]. Further trials that take a novel approach and integrate technology and end user involvement in the school dining room are needed, rather than implementing the same old strategies with moderate to no effectiveness for improved nutrition. There is a need to further evaluate changes implemented within a meal service such as assessment of implementation fidelity, stakeholder feedback via survey and/or qualitative methods. A mixed-methods approach incorporating qualitative research methods can provide an insight into stakeholders' perception of an intervention that strengthens our understanding of adolescents' food behaviours and their dining experience; why they eat the way they do, their attitudes toward food and eating, and what influences food selection and consumption [150].

Implications
Adolescents who participate in a school meal program or attend a boarding school consume a significant proportion of their daily dietary intake while at school. Therefore, access to nutritious, palatable and appealing meals is paramount and should be mandated through policy. Targeting school menus by increasing wholefoods and decreasing the availability and affordability of less healthy options is actionable. Importantly, this translates to an increase in diet quality (macro-and micronutrients) as evidenced in this review among studies that measured nutrient intake [89,[103][104][105]115,116,119,125]. The ultimate objective is to reverse the trend of poor diet quality during adolescence and influence their habitual food behaviours as they transition to adulthood, to reduce their risk and burden of disease. Our review has exposed the following opportunities, across environmental and behavioural contexts, for interventions within a schools' routine meal service to improve adolescent food behaviours, health and dining experience:

•
Engage the stakeholders who prepare the food (food service staff) and consume the food (adolescent students) through formative research, program development and/or implementation; recruit peer advocates to act as change agents; • Explore novel approaches in the school dining room such as integrating technology which now forms part of adolescents daily lives; • Ensure nutritional quality of school menus alongside assessment of palatability; they must go hand in hand to increase consumption, reduce waste, and improve students' diet quality. Allow students to sample modified foods, and if feasible, engage experts in the field of food and nutrition (dietitians, school nutrition specialists or professional chefs) to inform recipe or menu reformulation; • Healthy options must be accessible (front and centre), visually appealing (showcase them), and fast to access because time allowed for lunch at school is limited; • Restrict the availability and portion size of less healthy options. Students can only make decisions based on the options placed in front of them; • Include marketing strategies and positive health messaging to engage adolescents and promote positive changes to the meal service; • Use short-and longer-term evaluations to monitor progress and build sustained change; • Measure selection and consumption of meal components to assess intake and waste.

Conclusions
The impact of a routine school meal service on adolescents' food behaviours, health and dining experience is affected by multiple factors including the food environment, food service practices and skills, nutritional quality and palatability of menus, schools' engagement and attitude towards health promotion, competing food offerings, and varying student preferences. This review has identified a range of opportunities available to target these factors and supports the view that the education sector is a key domain to reach adolescents and improve nutrition and decision making around food and eating. A comprehensive approach that integrates environmental and behavioural strategies that engage adolescents in the development and/or implementation of initiatives, and evaluation can enhance success. Novel approaches such as integrating technology and student sampling of meals warrant further exploration. Higher quality pragmatic trials are required with longer term implementation and evaluation to build sustained change that will improve diet quality and ultimately benefit adolescents physical and mental health and wellbeing.