Data
Data came from three surveys representative of schools in 13 states of Mexico (out of 32) carried out at three implementation stages: June 2012, April 2013 and April 2015. The first survey included public and private elementary and high schools, while the last two included public elementary schools only. The present study focuses on morning and afternoon shifts of public elementary schools and their students. Sampling was done in two stages. First, clusters of schools were selected within strata (a combination of state, public/private and educational stage i.e. elementary or secondary), then children were randomly selected within schools (using a numbered list of all children in the target grades and selecting random numbers using random.org). A core group of schools were followed-up over time with a different sample of children drawn in each wave (N = 35 schools three time-points; N = 10 schools two time-points). On average four children per school were selected (min = 1, max = 5). Figure 1 illustrates the procedure followed in each stage, target and achieved samples with complete information at school and individual level for elementary schools. The total sample for the present study included 645 children from 99 different schools distributed as follows: 123 children from 38 schools in stage 2; 357 children from 96 schools in stage 3 and 165 children from 44 schools in stage 4. A complete case analysis was conducted.
Data collection in each survey consisted of a series of questionnaires and direct observations. Trained teams of interviewers (five-person teams) visited the schools and recorded the number of functioning drinking water sources. Then, the number of plates, packages or pieces of every single food and beverage available at the school food stores was recorded including their weight or volume. For packaged foods, interviewers recorded the number of portions contained in each package. For sampled children in each school, interviewers recorded basic demographic information, a list of foods consumed during school hours and whether the foods were brought from home or purchased in school. Interviewers were trained at the National Institute of Public Health in Cuernavaca. Questionnaires were tested for face validity in a pilot carried out in Cuernavaca.
Study variables
Outcome variable
Healthy snack was a binary variable where one denotes a snack which contained at least one fruit or vegetable and did not contain sugar sweetened beverages. Information to construct this variable was obtained from direct observation of children’s lunchboxes and/or children’s purchases in school food stores. Sugar sweetened beverages included soda, industrialized juices, energy drinks and flavoured water which is prepared with added sugar and a small amount of fruit.
Exposure variables
School compliance with the standards was operationalized as the proportion of food items sold in school food stores that complied with the nutrition criteria in the different stages of the standards (Table 1). The numerator was the sum of foods and beverages sold in school that complied, and the denominator was the sum of all foods sold in the school. To calculate the numerator, the energy and macronutrient content of each food and beverage sold in schools and regulated by the standards was calculated using the nutrient composition tables compiled by the National Institute of Public Health [17]. Foods and beverages were grouped according to the food categories established in the guideline (see Table 1). Then according to the food group, we assessed compliance with the nutrition criteria. If the food item in question complied with n-1 (to allow for a few missing values) of the criteria established for its food group, we concluded that it complied with the guideline. Fruits, vegetables and plain water always complied.
Time was operationalized as stage and coded 0 for stage 2 (2012) 1 for stage 3 (2013) and 2 for stage 4 (2015).
Effect modifiers
Snack origin, whether the child’s snack was brought from home or purchased in the school (home = 0, school = 1).
Change in compliance, whether school’s compliance with the standards improved remained stable or declined over time, was constructed for the group of schools which were followed up for at least two timepoints. First, tertiles of the variable school compliance with the standards (described above) were created in each stage. This produced a measure of relative compliance useful for comparison across stages. Then, using the tertiles at the beginning and end of each school’s follow-up period, the variable change in compliance was created (0 = decline in compliance; 1 = stable; 2 = increase).
Covariates
Models were adjusted for individual and school-level confounders selected a priori. Individual level variables included: shift (1 = morning, 2 = afternoon), school grade (3rd, 4th, 5th 6th) and sex. School-level variables were availability of free drinking water, defined as at least one functioning water fountain or other communal water source in the school (1 = yes; 0 = no), municipal-level extreme poverty (continuous) and municipal-level education (continuous). Municipal-level extreme poverty was based on the proportion of households living in extreme poverty at the municipality where schools were located. Households in extreme poverty were defined as those having an income below the basic food basket value, calculated from the National Survey on Income and Expenditure 2014 [18]. Similarly, the proportion of the population over 24-years old at municipality level with completed high school or more was computed from the Intercensal Survey 2015 [19].
Statistical analysis
First, we tested hypothesis 1, that better school compliance with the standards would be directly associated with a healthy snack, in particular, if the snacks were purchased in the school. Data from the three survey waves were pooled and observations where part of the snack was purchased in school and part was brought from home were excluded. Before combining the data, the time variable was created to identify observations according to stage of the standards. We used logistic regression with adjustment for clustering at school level. The outcome variable was a healthy snack and the main exposure variable was school compliance with the standards. To assess effect modification, an interaction term between school compliance with standards and snack origin was included. Unadjusted and fully adjusted models were fitted. Predicted probabilities and probability differences for the effect of school compliance with standards on consumption of a healthy snack by whether the snack was brought from home or purchased in school were computed.
Hypothesis 2 was tested next, that the probability of a healthy snack would increase over time in schools in which compliance with the standards improved but not in those where compliance remained stable or worsened. Thus, we fitted a model for repeated surveys aggregated at the school level, keeping only schools that had information on compliance for at least two time-points for this analysis (N = 45). Further, observations where the snack or part of it was brought from home were excluded, since results for hypothesis 1 suggested that snacks brought from home were not affected by schools’ compliance with the standards. We used logistic regression with adjustment for clustering at school level. The outcome variable was a healthy snack and the main exposure variable was time. An interaction term was included between time and change in compliance. Unadjusted and a fully adjusted models were fitted, using the same covariates as the model used for hypothesis 1. Data management and all analyses were performed using Stata 12.