Changes in Food-Group and Processing-Level Composition of Lunches Consumed Following Implementation of a Universal School Lunch Program: A Quasi-Experimental Study in Saskatchewan, Canada

Abstract

Background/Objectives: Universal school food programs may reshape lunch composition by changing foods available at school and lunch sources. We assessed whether food-group and processing-level composition changed differently in intervention and control schools following implementation of a universal school lunch program. Methods: Four elementary schools in Saskatoon, Saskatchewan, Canada, were included; two implemented a universal free lunch program and two served as controls. Lunch consumption was measured in May–June 2021 and May–June 2023 using photography-assisted weighed plate waste. Foods were classified using Canada’s Food Guide and NOVA. Outcomes were expressed as grams and percentage of total quantified lunch grams and analyzed using an equal-school-weighted difference-in-differences approach. Results: The analysis included 980 student-period observations. Total quantified lunch grams declined in both groups, with a smaller reduction in intervention schools (DiD = 32.62 g). Vegetable and fruit intake declined less in intervention schools (DiD = 3.52 g), although proportional change did not favour intervention schools (DiD = −3.75 percentage points). Whole-grain foods declined more in intervention schools in grams (DiD = −18.71 g) and proportional contribution (DiD = −8.07 percentage points), a pattern consistent across sensitivity analyses. Ultra-processed foods declined in both groups, with little differential proportional change (DiD = −0.17 percentage points). Conclusions: Implementation was not accompanied by consistent differential improvements in lunch composition. Whole-grain foods showed the most consistent between-group pattern, whereas reductions in ultra-processed foods possibly reflected broader changes across schools. These findings highlight the importance of distinguishing the foods available through school food programs from those students actually consume.

1. Introduction

Canada is currently expanding its investment in school food. The 2024 National School Food Policy established a federal framework for broader school food provision, making evidence of actual student consumption increasingly relevant [1].

School food programs can change what children eat during the school day by supplementing or replacing lunches brought from home. Evaluations have often focused on energy and nutrient intake, nutrient adequacy, or overall diet quality. Although these measures provide important information about nutritional intake, they do not identify the specific dietary components responsible for observed changes.

Food-group composition can show whether implementation changes consumption of dietary categories emphasized in national guidance. Canada’s Food Guide recommends regular consumption of vegetables and fruits, whole-grain foods, and protein foods while advising that highly processed products be limited [2]. Canadian studies have reported low intake of whole fruit and whole grains during school-day lunches, as well as higher diet quality among students accessing school food programs [3,4,5]. More broadly, reviews of universal school meals and Canadian school nutrition initiatives suggest potential benefits for diet quality, but effects vary considerably according to program design and setting [6,7]. A systematic review of school food environment interventions, mostly conducted in high-income countries, found that nutrition standards, changes in how foods are presented and positioned, and fruit and vegetable provision may have modest benefits for the consumption of healthy foods and beverages [8].

Evidence from Brazil’s National School Feeding Program (PNAE) also suggests dietary benefits. Cross-sectional studies have associated school-meal consumption with higher fruit and vegetable intake among children living in areas of greater social vulnerability and better diet quality among adolescents. The availability of PNAE meals has also been associated with lower consumption of ultra-processed salty foods and soft drinks [9,10,11].

A recent systematic review of universal free school meals in the United States identified six longitudinal studies covering more than 11,000 schools, but none reported student dietary intake or diet quality outcomes [12]. Despite the available observational evidence, studies examining changes in actual food consumption following the implementation of universal school food programs remain limited.

Processing level captures a related but distinct dimension of lunch composition. The NOVA system classifies foods and beverages into four groups—unprocessed or minimally processed foods, processed culinary ingredients, processed foods, and ultra-processed foods—according to the nature, extent, and purpose of industrial processing [13,14]. This framework complements Canada’s Food Guide because processing characteristics are not equivalent to food-group classification or nutrient content. In a natural experiment in England and Scotland, implementation of universal infant free school meals increased school-meal uptake and reduced the proportion of lunch grams derived from ultra-processed foods [15]. Using both Canada’s Food Guide and NOVA therefore allows changes in recommended dietary components and processing-level composition to be examined separately.

Universal availability of a school lunch does not necessarily mean that all students participate or that those who do consume the entire meal offered. Assessing all foods actually consumed during the lunch period therefore provides a population-level evaluation of dietary change following program implementation rather than an assessment of the nutritional quality of the planned menu alone.

Previous work from this cohort characterized the NOVA composition of lunches at baseline and found substantial differences according to lunch provenance, with a lower contribution of ultra-processed foods in school-provided than in home-packed lunches [16]. However, that analysis was cross-sectional and conducted before implementation of the universal lunch program.

The Good Food for Learning project was a universal, free-of-charge school lunch intervention implemented in two elementary schools in Saskatoon, Saskatchewan. A related analysis from the same evaluation examined changes in lunch-period energy intake, nutrient intake, and nutrient density using the same study periods [17]. The present study, which represents the food-based component of this evaluation, examined changes in lunch composition using Canada’s Food Guide and NOVA frameworks. Specifically, we assessed whether baseline-to-follow-up changes in the absolute and proportional intake of vegetables and fruits, whole-grain foods, protein foods, and each of the four NOVA processing categories differed between intervention and control schools following implementation of the universal school lunch program.

2. Materials and Methods

2.1. Study Design and Setting

This study used dietary data collected as part of the Good Food for Learning evaluation. We examined lunch composition at baseline (May–June 2021) and near the end of the intervention (May–June 2023). Four public elementary schools in Saskatoon, Saskatchewan, participated in the evaluation, with two implementing the universal lunch program and two serving as controls. The main analysis compared repeated cross-sectional school populations at the two periods; a separate sensitivity analysis examined the subset of students observed at both time points.

Details of the study design and school selection are available in the published protocol [18]. Participating schools were located in Saskatoon neighbourhoods with lower mean household incomes than the city average and had existing school food programs. Intervention schools were purposively selected in collaboration with the school division based on their capacity to expand lunch provision, including available infrastructure and staff support. Control schools were selected based on similar school and neighbourhood characteristics, including neighbourhood median income [18]. Intervention status was not randomly assigned.

Reporting was guided by the STROBE statement and its extension for nutritional epidemiology (STROBE-nut) [19]. The University of Saskatchewan Behavioural Research Ethics Board approved this study (BEH-509). Parents and caregivers received written study information and could decline their child’s participation; students provided verbal assent before dietary data collection, and permission to conduct this study was obtained from Saskatoon Public Schools.

2.2. Participants

Students enrolled in Kindergarten through Grade 6 were eligible on each data collection occasion. The repeated cross-sectional analysis included every student with at least one usable lunch observation at baseline, follow-up, or both. Age and reported gender were obtained from school records. Gender records coded as U were classified as unidentified.

Observations were summarized at the student-period level. When two lunches were observed for a student within the same study period, intake was averaged across those days; otherwise, the single usable lunch represented that period. Thus, the baseline and follow-up samples were not restricted to the same students. All eligible observations from the four participating schools were retained, and no separate sample-size calculation was undertaken. Participant identifiers were used to identify the subset contributing data at both periods for the longitudinal sensitivity analysis described in Section 2.7.

2.3. Intervention and Control Context

School food was already available in all four schools at baseline, but provision was limited, inconsistently funded, often dependent on donations, and mainly intended for students requiring additional food support. Common offerings included breakfasts, sandwiches, pre-packaged snacks, and other simple meals. This general pattern was observed in both intervention and control schools, although the number of students reached and portion sizes were not systematically documented.

Beginning in October 2021, the Good Food for Learning program expanded existing provision in the intervention schools so that a free lunch was available to every enrolled student each school day. Participation was optional, and students could consume school-provided food, food brought from home, or both. The control schools did not adopt the Good Food for Learning model. By follow-up, food provision in the control schools had become more frequent and included a wider range of foods, although the extent of these changes was not systematically documented. Further implementation details are reported in the accompanying process evaluation [20].

A six-week rotating menu provided the initial framework for meal planning and was informed by Canada’s Dietary Guidelines [2]. Meals were generally planned around a hot main dish accompanied by vegetables and/or fruit and milk or water, with whole-grain and protein foods incorporated where feasible. Students could select individual meal components and receive additional servings.

The educational component included six curriculum-linked lesson plans for each grade from Kindergarten through Grade 6 and hands-on activities such as cooking and gardening [18,20]. However, these activities were delivered intermittently, and their frequency, duration, and student participation were not systematically recorded [20]. The present analysis evaluates the intervention as a whole and does not estimate the independent contribution of the educational component.

The menu functioned as a guide rather than a fixed schedule. During routine delivery, schools adapted recipes and meal components in response to ingredient availability, food donations, staffing and equipment, preparation requirements, dietary needs, and student uptake. Consequently, the foods available and consumed varied across observation days. This variation is particularly relevant to the present analysis because food-group and processing-level outcomes reflect what students actually consumed under routine program delivery, rather than adherence to the originally planned menu.

Both intervention schools maintained daily lunch provision, although staffing continuity and meal-service coverage differed between them. During the second intervention year, recorded meal servings represented approximately 67% of potential enrolled-student meal opportunities (57% and 87% at the two schools). These estimates describe meal-service volume rather than individual student participation [20].

2.4. Dietary Assessment and Lunch Provenance

Lunch intake was assessed using photography-assisted weighed plate waste. The protocol targeted two lunch observations per student within each study period. Observation days were coordinated in advance with participating schools, without selecting dates based on the foods or menus being served. Classrooms were generally confirmed shortly before data collection. Baseline and follow-up assessments were conducted in May–June 2021 and May–June 2023, respectively, to align data collection by season. Dietary data were collected over four school days per school at baseline and generally six to eight days per school at follow-up. When only one usable observation was obtained, that lunch was retained for analysis. Some students had only one usable observation due to absences, eating elsewhere, school activities, or non-participation. Observation dates were not disclosed to students or families in advance. Each student was assigned a unique identifier linking dietary observations with demographic information.

On each observation day, trained research staff weighed foods and measurable beverages immediately before consumption and reweighed remaining items after lunch using calibrated digital scales. Intake was calculated as the difference between the initial and remaining weights. Photographs and field notes were used during coding to confirm food identity, portions, and circumstances affecting measurement. Full details of the dietary assessment procedures have been published previously [21]. Plain water intake was not quantified.

Using direct observation and food-source records, each lunch was classified as school-provided if all items came from school food programming, home-packed if all items were brought from home or obtained elsewhere, or mixed if foods came from both sources. Student-period provenance was determined using the available lunch observations. For students with only one observation, the classification of that lunch was retained. For students with two observations, a school-provided or home-packed classification required both lunches to belong to that category; all other combinations were classified as mixed.

2.5. Food-Group and Processing-Level Classification

All quantified foods and beverages were classified using two complementary food-based frameworks: categories aligned with the 2019 Canada’s Food Guide (CFG) and the NOVA food-processing classification. The final coding rules were applied consistently across intervention and control schools and at baseline and follow-up. Classification used the most detailed available information, including food descriptions, recipes, ingredient lists, product information, photographs, and field notes. Detailed operational classification rules and examples for the CFG and NOVA frameworks are provided in Supplementary Tables S1 and S2, respectively.

2.5.1. Canada’s Food Guide Categories

The CFG analysis focused on three categories emphasized in Canada’s Food Guide: vegetables and fruits, whole-grain foods, and protein foods [2]. Food-category definitions and examples from the Canadian Food Intake Screener were used as an additional coding reference for CFG-aligned categories; the screener itself was not administered or scored in this child sample [22].

Single-component foods were classified directly. Mixed dishes were disaggregated into component ingredients when recipes or sufficiently detailed ingredient information were available, and each component was classified separately. Aggregate records representing dishes that had been disaggregated were excluded from quantitative calculations to prevent double counting.

Vegetables and fruits included fresh, frozen, dried, and canned vegetables and fruits and unsweetened fruit purées meeting the study coding criteria. One hundred percent fruit juice and fried vegetable products were not classified as vegetables and fruits. Whole-grain foods included grain products for which the grain component was identified as predominantly whole grain or whole wheat based on food descriptions, recipes, or available ingredient information. Protein foods included legumes, nuts and seeds, tofu and other soy products, eggs, fish and seafood, poultry, unprocessed meat, dairy products, and fortified plant-based alternatives. Processed meats were not classified as protein foods in the primary analysis and were retained under Other foods.

Foods not meeting the definitions of these three categories were classified as “Other foods”. Other foods contributed to total quantified lunch grams but were not analyzed as a separate CFG outcome.

2.5.2. NOVA Classification

All quantified foods and beverages were also assigned to one of four NOVA categories: unprocessed or minimally processed foods (NOVA 1), processed culinary ingredients (NOVA 2), processed foods (NOVA 3), and ultra-processed foods (NOVA 4) [14].

Classification was based on recorded food names, photographs, packaging or ingredient information when available, field notes, and recipes for school-provided foods. Composite dishes were disaggregated into ingredients when sufficient information was available, and individual ingredients were classified according to NOVA.

Consistent with previously published Canadian applications of NOVA, burgers, pizzas, sandwiches, and similar mixed dishes obtained from quick-service restaurants were classified as NOVA 4 [23]. Commercial purchase or ready-to-eat status alone was not sufficient for NOVA 4 classification; standalone foods such as whole fruit and plain milk retained their ingredient-specific classifications.

The same classification rules were applied to school-provided, home-packed, and mixed lunches and at both study periods. Two researchers reviewed the NOVA classifications and resolved differences by consensus; formal inter-rater reliability was not calculated.

2.6. Derivation of Food-Composition Outcomes

For each student-period, grams consumed were summed across foods assigned to each CFG and NOVA category. For students with two observed lunches within a study period, category-specific and total quantified grams were averaged across the two observation days to obtain average grams consumed per lunch; students with one observation contributed the value from that lunch.

Each CFG and NOVA category was examined both as an absolute amount consumed (g/lunch) and as its percentage contribution to total quantified lunch grams. Percentage contribution was calculated as category-specific grams divided by total quantified lunch grams and multiplied by 100, using the sums across all available lunch observations within each student-period. Weight-based measures were used because consumed grams could be assigned to individual food categories, including ingredients from mixed dishes, whereas energy estimates for these preparations were available at the whole-dish level rather than for individual ingredients.

Plain water intake was not quantified and therefore was not included in total quantified lunch grams or percentage denominators. The three reported CFG categories were not exhaustive and consequently were not expected to sum to 100%; foods classified as Other remained in the denominator. In contrast, all quantified foods included in the NOVA analysis were assigned to one of the four NOVA categories; so, their percentage contributions summed to approximately 100%, subject to rounding.

Total quantified lunch grams were also examined as a separate outcome to distinguish changes in the amount consumed from changes in the composition of lunch intake.

2.7. Statistical Analysis

Descriptive statistics were calculated by study group and period. Continuous variables are presented as means and standard deviations, and categorical variables as counts and percentages.

We used a difference-in-differences (DiD) approach to compare baseline-to-follow-up changes in food-group and processing-level outcomes between intervention and control schools [24,25]. All quantified foods and beverages consumed during the observed lunch period were included, whether school-provided, home-packed, or mixed. The estimate therefore represents change in lunch-period food composition across the participating school populations rather than the effect of consuming a school-provided lunch among program participants only.

For each outcome, student-period values were averaged within school and study period, producing eight school-period means. Each school-period entered the model once; so, schools contributed equally regardless of the number of student-period observations. A linear regression including study period, study group, and their interaction was fitted:

Y_st = β₀ + β₁Post_t + β₂Intervention_s + β₃(Post_t × Intervention_s) + ε_st

where Y_st is the mean outcome for school s at period t; Post_t indicates follow-up versus baseline; Intervention_s indicates intervention versus control school; and ε_st is the error term. The period-by-group interaction coefficient (β₃) was the DiD estimate. Separate models were fitted for total quantified lunch grams and for each absolute and proportional Canada’s Food Guide (CFG) and NOVA outcome. Given the small number of schools, primary DiD estimates were presented descriptively without conventional confidence intervals or confirmatory hypothesis tests. For NOVA outcomes, baseline values in the present study may differ from those reported previously for this cohort because the current analysis gives equal weight to each school, whereas the earlier baseline analysis pooled participant-level observations [16].

To examine sensitivity to measured sample composition, DiD models were also fitted using student-period data while retaining equal school-period influence through inverse school-period sample-size weights. The first model adjusted for age at the corresponding study period and reported gender with school fixed effects; a second model additionally adjusted for the number of lunch observations contributing to each student-period value (one versus two). The two student-period observations with reported gender recorded as unidentified (U) were excluded from both adjusted models but retained in the primary analysis. Both adjusted models included 978 student-period observations for every outcome. Weights were recalculated within each school-period after these exclusions. The study-group main effect was absorbed by the school fixed effects because intervention status did not vary within school. Lunch provenance was not adjusted for because it could change as a consequence of intervention implementation.

The primary school-level DiD analysis was also repeated separately for student-periods represented by one versus two observed lunches. These strata were examined descriptively, and no formal comparison between them was performed.

Within-student change was examined among students observed at both periods who remained in the same participating school. Repeated student-period observations were analyzed using linear mixed-effects models fitted with the lme() function in the nlme package, with a participant-level random intercept to account for within-student correlation. Models were fitted separately for each outcome using restricted maximum likelihood. The longitudinal models did not apply equal-school weighting. The initial model included study period, study group, and their interaction. Additional models sequentially adjusted for age and reported gender recorded at baseline, the number of lunch observations contributing to each student-period, and school fixed effects. In the model containing school fixed effects, the study-group main effect was omitted because intervention status was fixed within school and therefore collinear with the school indicators. Students whose reported gender was classified as unidentified (U) at either study period were excluded so that all longitudinal models used the same analytic sample. Model-based 95% confidence intervals and p-values are reported as sensitivity results rather than as confirmatory inference about the school-level intervention effect. These models accounted for repeated observations within students but did not account for residual correlation between different students in the same school.

An additional sensitivity analysis assessed the influence of classifying processed meats as protein foods rather than Other foods. The primary equal-school-weighted DiD analysis was repeated for protein-food outcomes using this alternative classification.

No statistical imputation was performed. All analyses were conducted using R version 4.5.0 (R Foundation for Statistical Computing, Vienna, Austria) [26].

3. Results

3.1. Sample Characteristics and Lunch Provenance

The analysis included 980 student-period observations from four schools (379 at baseline and 601 at follow-up; Table 1). Mean age was 8.5 years at baseline and 8.3 years at follow-up in control schools, compared with 8.5 and 8.4 years, respectively, in intervention schools. Age-group and reported-gender distributions were also similar across periods in both groups. Two lunch observations were available for 627 student-period observations (64.0%).

Table 1. Sample characteristics and lunch provenance by study group and period, Saskatoon, Saskatchewan, Canada, 2021 and 2023.

Characteristic	Category	Control Baseline, n (%)	Control Follow-Up, n (%)	Intervention Baseline, n (%)	Intervention Follow-Up, n (%)
Student-period observations	Total	193 (100.0)	310 (100.0)	186 (100.0)	291 (100.0)
Reported gender	Female	89 (46.1)	157 (50.6)	92 (49.5)	135 (46.4)
	Male	103 (53.4)	153 (49.4)	93 (50.0)	156 (53.6)
	Unidentified	1 (0.5)	0 (0.0)	1 (0.5)	0 (0.0)
Age group	5–8 years	99 (51.3)	168 (54.2)	93 (50.0)	156 (53.6)
	9–13 years	94 (48.7)	142 (45.8)	93 (50.0)	135 (46.4)
Lunch provenance	Home-packed	84 (43.5)	159 (51.3)	40 (21.5)	63 (21.6)
	Mixed	45 (23.3)	45 (14.5)	66 (35.5)	48 (16.5)
	School-provided	64 (33.2)	106 (34.2)	80 (43.0)	180 (61.9)
Number of lunch observations	One	56 (29.0)	135 (43.5)	65 (34.9)	97 (33.3)
	Two	137 (71.0)	175 (56.5)	121 (65.1)	194 (66.7)

Note: Values are student-period observations. Percentages are column percentages within each study group and period. Lunch provenance was classified as home-packed, mixed, or school-provided using all available lunch observations within the corresponding student-period.

Of the 379 baseline student-period observations, 373 had participant identifiers usable for linkage across study periods. Among these students, 141 (37.8%) were also observed at follow-up. Of the 141 linked students, 137 remained in the same participating school. One student with unidentified reported gender was excluded from the longitudinal analysis, resulting in a final longitudinal sample of 136 students contributing 272 student-period observations.

Lunch provenance shifted over time. In intervention schools, the proportion of student-period observations classified as school-provided increased from 43.0% at baseline to 61.9% at follow-up, while mixed lunches decreased from 35.5% to 16.5%. In control schools, school-provided lunches remained relatively stable, increasing from 33.2% to 34.2%, while home-packed lunches increased from 43.5% to 51.3%.

3.2. Total Quantified Lunch Intake and Canada’s Food Guide Outcomes

Total quantified lunch intake decreased between baseline and follow-up in both groups, with a smaller reduction in intervention schools than in control schools (DiD = 32.62 g; Table 2).

Table 2. Total quantified lunch intake and Canada’s Food Guide food-group outcomes by study group and period, Saskatoon, Saskatchewan, Canada, 2021 and 2023.

Outcome	Control Baseline, Mean (SD)	Control Follow-Up, Mean (SD)	Intervention Baseline, Mean (SD)	Intervention Follow-Up, Mean (SD)	DiD Estimate
Total quantified lunch intake, g	309.9 (0.8)	193.6 (20.9)	327.0 (14.9)	243.3 (29.6)	32.62
Vegetables and fruits, g	71.1 (9.5)	49.6 (0.6)	85.2 (10.9)	67.2 (11.7)	3.52
Vegetables and fruits, % of total quantified lunch grams	22.1 (2.6)	27.2 (5.2)	29.1 (2.2)	30.4 (3.4)	−3.75
Whole-grain foods, g	10.7 (1.9)	6.1 (2.2)	24.2 (5.8)	0.9 (0.3)	−18.71
Whole-grain foods, % of total quantified lunch grams	3.9 (1.1)	4.3 (1.7)	8.0 (2.6)	0.4 (0.1)	−8.07
Protein foods, g	73.4 (60.3)	32.5 (12.1)	96.4 (15.8)	67.7 (3.0)	12.13
Protein foods, % of total quantified lunch grams	23.2 (17.0)	16.2 (6.0)	28.8 (3.9)	25.0 (2.9)	3.26

Note: Values are unweighted means of the two school-level means within each study group and period, with SDs calculated across the two schools. Each school contributed equally regardless of the number of student-period observations. The DiD estimate represents the mean baseline-to-follow-up change in intervention schools minus the mean change in control schools. DiD estimates for percentage outcomes are expressed in percentage points. Plain water was not quantified and was not included in total quantified lunch grams. Foods classified as Other were included in the CFG percentage denominator but not reported as a separate outcome; therefore, the three reported CFG categories did not sum to 100%. Primary DiD estimates are descriptive and were not accompanied by confidence intervals or confirmatory hypothesis tests. DiD, difference-in-differences; SD, standard deviation.

Vegetable and fruit intake also decreased in absolute grams in both groups, although the reduction was smaller in intervention schools (DiD = 3.52 g). In contrast, vegetables and fruits increased as a proportion of total quantified lunch grams in both groups, with a larger increase in control schools, resulting in a DiD estimate of −3.75 percentage points.

Whole-grain foods showed the largest negative differential change among the Canada’s Food Guide outcomes. Absolute whole-grain intake decreased from 24.2 to 0.9 g in intervention schools, compared with a decrease from 10.7 to 6.1 g in control schools, yielding a DiD estimate of −18.71 g. The proportional contribution of whole-grain foods similarly decreased from 8.0% to 0.4% in intervention schools while remaining relatively stable in control schools, resulting in a DiD estimate of −8.07 percentage points.

Protein-food intake decreased in both groups, but the reduction was smaller in intervention schools (DiD = 12.13 g). Protein foods also represented a smaller proportion of total quantified lunch grams at follow-up in both groups, although the proportional decline was smaller in intervention schools (DiD = 3.26 percentage points). In the sensitivity analysis including processed meats as protein foods, the DiD estimates decreased to 8.98 g and 2.16 percentage points, respectively, without changing their direction (Supplementary Table S3).

3.3. NOVA Processing-Level Outcomes

Unprocessed or minimally processed food intake decreased in absolute grams in both groups, with a smaller reduction in intervention schools (DiD = 14.44 g; Table 3). However, its proportional contribution to total quantified lunch grams changed little between groups (DiD = −1.54 percentage points).

Table 3. NOVA processing-level outcomes by study group and period, Saskatoon, Saskatchewan, Canada, 2021 and 2023.

Outcome	Control Baseline, Mean (SD)	Control Follow-Up, Mean (SD)	Intervention Baseline, Mean (SD)	Intervention Follow-Up, Mean (SD)	DiD Estimate
NOVA 1: Unprocessed/minimally processed foods, g	126.0 (66.7)	75.9 (8.5)	159.7 (29.0)	124.1 (12.6)	14.44
NOVA 1: Unprocessed/minimally processed foods, % of total quantified lunch grams	39.4 (20.3)	40.7 (8.8)	51.6 (6.5)	51.3 (3.6)	−1.54
NOVA 2: Processed culinary ingredients, g	0.6 (0.7)	1.1 (1.1)	1.6 (1.8)	1.2 (1.1)	−0.92
NOVA 2: Processed culinary ingredients, % of total quantified lunch grams	0.2 (0.3)	0.6 (0.6)	0.5 (0.5)	0.5 (0.5)	−0.37
NOVA 3: Processed foods, g	29.3 (28.6)	31.6 (2.1)	21.7 (5.2)	36.7 (5.9)	12.70
NOVA 3: Processed foods, % of total quantified lunch grams	8.1 (6.6)	13.9 (0.6)	5.6 (1.7)	13.5 (0.1)	2.09
NOVA 4: Ultra-processed foods, g	154.0 (39.7)	84.9 (26.2)	143.9 (17.5)	81.3 (10.0)	6.41
NOVA 4: Ultra-processed foods, % of total quantified lunch grams	52.3 (14.0)	44.8 (8.7)	42.3 (7.7)	34.6 (4.1)	−0.17

Note: Values are unweighted means of the two school-level means within each study group and period, with SDs calculated across the two schools. Each school contributed equally regardless of the number of student-period observations. The DiD estimate represents the mean baseline-to-follow-up change in intervention schools minus the mean change in control schools. DiD estimates for percentage outcomes are expressed in percentage points. The four NOVA percentage categories sum to approximately 100% of total quantified lunch grams, subject to rounding. Plain water was not quantified. Primary DiD estimates are descriptive and were not accompanied by confidence intervals or confirmatory hypothesis tests. DiD, difference-in-differences; NOVA, food-processing classification system; SD, standard deviation.

Processed culinary ingredients contributed very little to lunch intake across study groups and periods, with small DiD estimates for both absolute intake (−0.92 g) and proportional contribution (−0.37 percentage points).

Processed food intake increased in both groups, with a larger increase in intervention schools (DiD = 12.70 g). Its proportional contribution also increased in both groups, corresponding to a DiD estimate of 2.09 percentage points.

Ultra-processed food intake decreased substantially between baseline and follow-up in both intervention and control schools. The reduction in absolute intake was slightly smaller in intervention schools (DiD = 6.41 g). Ultra-processed foods also decreased as a proportion of total quantified lunch grams in both groups, from 52.3% to 44.8% in control schools and from 42.3% to 34.6% in intervention schools. The corresponding DiD estimate was −0.17 percentage points, indicating little differential change between groups in the proportional contribution of ultra-processed foods.

Supplementary Table S4 presents descriptive food-group and NOVA composition according to lunch provenance. Across both study groups and periods, school-provided lunches generally contained higher proportions of vegetables and fruits and unprocessed or minimally processed foods and lower proportions of ultra-processed foods than home-packed lunches. In intervention schools, the proportion of ultra-processed foods in school-provided lunches decreased from 27.7% at baseline to 19.5% at follow-up, while increasing from 66.2% to 74.1% in home-packed lunches. Whole-grain foods in school-provided intervention lunches decreased from 12.3% to 0.0% of total quantified lunch grams.

3.4. Covariate-Adjusted Repeated Cross-Sectional Sensitivity Analysis

Both adjusted models included 978 student-period observations. Covariate adjustment did not materially change the primary findings. Fourteen of the 15 outcomes retained the same direction as the primary analysis in both adjusted models. The exception was the percentage contribution of ultra-processed foods, which had a small negative DiD estimate in the primary analysis (−0.17 percentage points) and small positive estimates after adjustment (+0.05 and +0.33 percentage points). The negative whole-grain estimates remained similar to the primary analysis. Full estimates are presented in Supplementary Table S5.

3.5. Sensitivity Analysis by Number of Lunch Observations

Estimates were more variable when analyses were stratified according to whether student-period values were based on one or two observed lunches. Estimates were in the same direction as the primary analysis for 11 of 15 outcomes among student-periods based on one lunch and for 11 of 15 outcomes among those based on two lunches. Whole-grain intake showed the most consistent pattern, with negative differential changes in both absolute and proportional intake in both strata. In contrast, protein foods and several NOVA outcomes differed in direction across the one- and two-lunch analyses. Full estimates are presented in Supplementary Table S6.

3.6. Longitudinal Sensitivity Analysis

In the longitudinal analysis, 11 of the 15 outcomes had estimates in the same direction as the primary analysis in the unadjusted and age- and gender-adjusted models. This decreased to 10 of 15 in the models that additionally adjusted for the number of lunch observations. Whole-grain estimates remained negative across all four models. The proportional estimates for vegetables and fruits, protein foods, and unprocessed or minimally processed foods, as well as the estimate for ultra-processed food grams, differed in direction from the primary analysis. Protein-food grams also became slightly negative after adjustment for the number of lunch observations. Full model estimates, 95% confidence intervals, and p-values are presented in Supplementary Table S7.

4. Discussion

This study examined whether the food-group and processing-level composition of lunches consumed changed differently in intervention and control schools following implementation of a universal school lunch program. All foods consumed during the lunch period were included, whether school-provided, home-packed, or mixed; so, the analysis assessed dietary change across the participating school populations rather than the nutritional composition of school-provided meals alone. Overall, the pattern of differential change varied across outcomes. Total quantified lunch intake declined substantially in both groups, vegetable and fruit findings differed depending on whether intake was expressed in absolute or proportional terms, and whole-grain foods showed the clearest negative differential pattern. Ultra-processed foods declined in both groups with little differential proportional change. At the same time, lunch provenance shifted substantially in intervention schools, with a greater share of observations classified as school-provided at follow-up. Covariate adjustment had little influence on the primary estimates, whereas observation-stratified and longitudinal analyses were more variable.

The distinction between absolute and proportional outcomes is important for interpreting these findings. Vegetable and fruit intake, for example, declined in grams in both groups, although the decline was smaller in intervention schools. At the same time, vegetables and fruits increased as a proportion of total quantified lunch grams in both groups, with a somewhat larger proportional increase in control schools. Thus, the higher proportional contribution at follow-up did not represent increased absolute vegetable and fruit consumption; rather, this category declined less than total quantified lunch intake. Similar differences between absolute and proportional estimates occurred for other food categories. Reporting both measures therefore helps distinguish changes in the amount consumed from changes in the composition of the lunch.

Whole-grain foods showed the clearest differential pattern. Intervention schools began with higher whole-grain intake than control schools, but intake declined from 24.2 g to less than 1 g per lunch at follow-up, whereas the decline in control schools was substantially smaller. The proportional contribution showed the same pattern, and negative estimates persisted across adjusted, observation-stratified, and longitudinal analyses. Although this consistency makes the finding notable, the small number of schools and non-randomized design preclude interpreting it as definitive evidence of an intervention effect.

The whole-grain finding is practically important because whole grains are already a limited component of Canadian school-day lunches [4]. In Good Food for Learning, procurement, ingredient availability, food donations, staffing, equipment, preparation demands, and student response all contributed to menu and recipe adaptation [20]. The decline in whole-grain consumption was observed in both intervention schools (Supplementary Table S8) and was also evident among school-provided lunches (Supplementary Table S4). However, changes in whole-grain availability and specific recipe substitutions were not systematically documented; so, we could not determine whether the decline reflected changes in the foods offered, student selection, or both. In U.S. national survey data, implementation of whole-grain-rich school meal requirements was followed by substantial increases in students’ whole-grain consumption from school foods [27]. Taken together, these findings support monitoring whole-grain provision, selection, and consumption directly rather than assuming that general alignment with dietary guidance ensures consistent exposure.

Ultra-processed food intake decreased substantially from baseline to follow-up in both intervention and control schools, but the proportional declines were similar, resulting in little differential change. This differs from the natural experiment in England and Scotland, where universal infant free school meals increased school-meal uptake and reduced the proportion of lunch weight derived from ultra-processed foods [15]. UK cross-sectional evidence has also reported lower ultra-processed food content in school meals than in packed lunches [28].

Our baseline analysis of this cohort similarly found a substantially lower participant-level contribution of ultra-processed foods in school-provided than in home-packed lunches [16]. The additional descriptive analysis showed a similar pattern at follow-up, with ultra-processed foods accounting for 19.5% of quantified lunch grams in school-provided lunches compared with 74.1% in home-packed lunches in intervention schools (Supplementary Table S4). These findings are not necessarily inconsistent with the present results. The analyses by lunch provenance describe differences between lunch types, whereas the primary DiD analysis examined whether lunch composition changed differently over time across the intervention and control school populations. Because home-packed and mixed lunches remained common at follow-up and food provision also evolved in control schools, a lower ultra-processed food contribution in school-provided lunches would not necessarily translate into a large intervention-level DiD estimate.

The concurrent decline in total quantified lunch grams is also relevant. Total lunch intake decreased substantially in both intervention and control schools, suggesting that this pattern was not specific to the intervention. The reasons for this decline are unclear. Changes in food provision, lunch composition, and student participation may have contributed, although these factors were not assessed as explanations for the change in total intake. Both unprocessed or minimally processed foods and ultra-processed foods declined substantially in absolute amounts, while processed foods increased in both groups. These concurrent changes indicate that the observed shifts in NOVA categories occurred alongside lower overall quantified lunch intake and should not be interpreted as simple substitution between ultra-processed and minimally processed foods.

The findings contrast somewhat with broader observational and review evidence on school food programs. Systematic reviews have generally reported improvements in dietary outcomes associated with universal school meals, although effects vary substantially by program design and setting [6,7]. Canadian observational studies have also reported higher diet quality among students accessing school food programs [3,5]. These studies, however, address a different question from population-level change following implementation. A recent systematic review of universal free school meals identified no eligible longitudinal studies reporting dietary intake or diet quality outcomes [12], highlighting the limited evidence directly comparable with the present design. Here, change was evaluated across the full populations observed in intervention and control schools regardless of the source of an individual lunch.

Lunch provenance illustrates why that distinction matters. In intervention schools, school-provided lunches increased from 43.0% at baseline to 61.9% at follow-up, but more than one-third of student-period observations at follow-up remained home-packed or mixed. Students consuming a school-provided lunch may also have consumed only selected components. Changes in the foods offered through the program would therefore not necessarily translate into equivalent changes in food consumption across the entire student population. Universal availability and universal exposure are not the same, and evaluations based only on menus, meals served, or participation counts may overestimate the extent to which intended food components are actually consumed.

The control-school context is also relevant. The control schools were not food-program-free at baseline, and the nature and extent of food provision changed during the study period. Lunch provenance also shifted in the control group, including an increase in home-packed lunches. The accompanying process evaluation documented changes in school food provision and evolving implementation conditions in both intervention and control settings [20]. These concurrent changes reduced the contrast between a universal program and a static no-program counterfactual and may have contributed to similar temporal patterns across groups.

Covariate adjustment had little influence on the primary estimates, whereas the observation-stratified and longitudinal analyses were more variable. Whole-grain estimates were comparatively stable across approaches, while several other food-group and NOVA outcomes were more sensitive to the analytic sample. For ultra-processed foods, the DiD estimates differed in direction between the one- and two-lunch strata, indicating that this finding was sensitive to the analytic sample and should be interpreted cautiously. The repeated cross-sectional analysis addressed changes across the observed school populations, whereas the longitudinal analysis was restricted to students observed at both periods. The longitudinal models also did not apply equal-school weighting; so, differences from the primary estimates may reflect both the restricted sample and differences in school contributions. The longitudinal analyses were therefore interpreted as complementary to the primary analysis.

For implementation and monitoring, the findings point to a need to distinguish what is planned, what is prepared or served, what students select, and what they ultimately consume. A systematic review of the U.S. National School Lunch Program found that strategies such as offering vegetables first, changing how vegetables are prepared, and procuring local produce may increase fruit and vegetable consumption [29]. Procurement, staffing, equipment, recipe consistency, and component-level uptake can influence movement between these stages. This is especially relevant as school food programming expands in Canada [1] and current Saskatchewan guidance encourages nutritious school food while reducing the availability of highly processed foods [30].

Several limitations should also be considered. Only four schools were included, with two intervention and two control schools, and intervention status was not randomly assigned. Although this study included 980 student-period observations, the intervention was implemented at the school level, leaving only four independent intervention-assignment units. Primary DiD estimates were therefore interpreted descriptively rather than as confirmatory evidence of program effectiveness. Only one pre-intervention measurement was available; so, the parallel-trends assumption could not be empirically assessed. With only two schools per group, school-specific characteristics and differences in intervention implementation may also have influenced the DiD estimates. The primary analysis used repeated cross-sectional samples and therefore represents changes in the student populations observed at each period rather than within-student change. The longitudinal sensitivity analysis provided a complementary within-student assessment but included only a subset of the baseline sample and should not be interpreted as an alternative primary analysis.

Dietary assessment was also based on one or two observed lunches per student-period and may not fully represent usual lunch intake or variation across the six-week menu cycle. Food classification required judgement, particularly for mixed dishes, foods with incomplete ingredient information, and foods prepared outside the schools. Recipes and component information were used when available and consistent coding rules were applied across groups and periods, but some misclassification remains possible. Percentage outcomes were based on food weight rather than energy intake. Because foods differ in water content and energy density, their proportional contribution to total lunch grams may not reflect their contribution to total dietary energy. Plain water was not quantified, and this study assessed intake only during the lunch period rather than total daily dietary intake. Finally, both intervention delivery and control-school conditions evolved over time, limiting causal attribution and generalizability beyond the participating schools.

This study has several strengths. Dietary intake was assessed objectively using a photography-assisted weighed plate-waste protocol rather than participant self-report. All foods and measurable beverages consumed during lunch were included regardless of provenance, allowing the analysis to capture changes across the participating school populations rather than only among students consuming school-provided meals. Applying both Canada’s Food Guide and NOVA provided complementary food-based measures, while reporting outcomes in both grams and percentage of total quantified lunch grams distinguished absolute consumption from lunch composition. Data collection was conducted during the same months at baseline and follow-up, reducing seasonal differences in measurement. In addition, covariate-adjusted, observation-number, and longitudinal sensitivity analyses allowed the stability of the primary estimates to be examined across different assumptions and analytic samples.

5. Conclusions

Implementation of universal school lunch provision substantially changed lunch provenance, but was not accompanied by consistent improvements in food-group or processing-level composition relative to control schools. The most consistent between-group finding was the decline in whole-grain foods in intervention schools, while ultra-processed foods decreased in both groups with little differential proportional change. These findings should be interpreted cautiously given the small number of schools and changes in food provision in both study groups.

As school food provision expands in Canada [1], evaluation should extend beyond access and meals served to the foods children actually select and consume. Monitoring meal quality, implementation fidelity, procurement and preparation capacity, student participation, and component-level consumption may help identify where universal provision is translating into improved lunch composition and where additional implementation support is needed.