This study used an ECC-based cluster randomized controlled trial design, where ECC were randomly allocated to either the intervention (HSDS) or control group (usual practice). A complete description of the trial protocol was published in 2016 and is registered (ClinicalTrials.gov #NCT02375490).(14) The study protocol was implemented as planned; however, modifications were made in the method used to score fundamental movement skills as explained below. Further, as detailed in the analysis section, the amount of missing data for the outcomes forced us to modify the analysis plan from an intention-to-treat to a complete-cases analysis approach. The study received ethics approval from Health Canada, the University of Saskatchewan, and the Université de Sherbrooke.
Provincial registries of licenced ECC in Saskatchewan and New Brunswick, Canada, were used as sampling frames. ECC were excluded if they had previously received a physical activity or nutrition intervention, did not prepare and provide lunch to children, and for feasibility reasons, if they had less than 20 children enrolled full-time in a preschool program. ECC were stratified according to province, geographical location (urban/rural)(15) and their respective school division (English or French). Once stratification was completed, project coordinators randomly selected ECC using the Stata SE statistical sequence generator software. ECC were then contacted, provided information about the study, and invited to participate. ECC which agreed to participate in the study were sent a consent form, as well as parental consent forms to recruit preschoolers attending their ECC on a full-time basis. If the ECC declined, they were replaced by another randomly selected ECC from the same stratum. Once informed consent was obtained, simple randomization was used to allocate ECC to either the intervention or control group with a 1:1 ratio. Parents of all participating children provided signed, informed consent. Prior to initiating recruitment and based on pilot work, we estimated that 700 children (350 per group) would provide 80% power to detect a 10% between-group difference in outcomes, considering a within-group standard deviation of 40%, a two-sided α of 0.05, an intra-class correlation of 0.02 and an estimated multiple correlation of 0.15 between the intervention and other explanatory variables. To compensate for losses to follow-up, our target was to recruit a minimum of 735 participants (5% over the 700 calculated).
The HSDS intervention was delivered over the course of 6 to 8 months, and included a 3-hour on-site training, resources (i.e. an implementation manual, physical activity and healthy eating manuals, an active play equipment kit), and on-going on-line and telephone support and monitoring; centres were also offered a tailored 90-minute booster session at the midway point of the intervention period. ECC randomly allocated to the control group continued their usual practice and were not provided with any training, resources or support. However, once the study was completed, all childcare centres from the control group were offered the HSDS intervention.
On-site training and resources – All ECC allocated to the intervention group were provided with a 3-hour on-site training, which was offered to childcare educators, directors and cooks after work hours. This training session was delivered by trained specialists (dietitians, kinesiologists or other experts in the fields of nutrition and physical activity), and covered best practices in physical activity and healthy eating in early childhood, including topics such as the importance of physical activity and healthy eating for preschoolers, how to easily integrate physical activity and healthy eating in the ECC’s daily routine, how to introduce and encourage children to try new and healthy foods, and how to help children develop their fundamental movement skills. ECCs were also provided with the evidence-based LEAP BC™-GRANDIR CB resources which included a physical activity and healthy eating manual. In addition, a New Brunswick developed fundamental movement skills manual (Active Kids Toolkit Foundations for All©), a kit with active play equipment, an implementation manual, and other complementary resources for childcare staff and families were shared with all participating sites.
On-going support and monitoring – ECC were encouraged to identify a Healthy Star, which was a childcare staff member who was a champion for physical activity and healthy eating and who was a knowledge-sharing contact between the ECC and the HSDS coordinators. The HSDS team checked-in with the intervention ECC on a regular basis by phone or email and provided them with support and encouragement. Monthly newsletters were also sent to all intervention ECC, which included tips on how to get children moving or on how to improve healthy eating. ECC were encouraged to share these newsletters with parents.
Booster session – A 90-minute booster session was offered to all intervention ECC approximately three months after the initial training. This on-site session was personalized based on challenges identified by each individual ECC, and was offered as a staff meeting, an in-class demonstration, a parent presentation, a cooking class, or a staff mini-training.
Each participating ECC was visited by two trained research assistants over two weekdays to collect data prior to the start of the intervention period and again 9 months later. This two-day data collection period was chosen for feasibility and logistical purposes, as well as to reduce the burden on ECCs. While blinding was not possible for the ECC, parents and children were not informed about group assignment. Research assistants responsible for collecting data were not told about the ECC’s group allocation.
Physical activity was assessed using the Actical accelerometer (B and Z-series, Mini Mitter/Respironics, Oregon, USA)(16), which has been shown to be a valid tool for measuring physical activity in preschoolers.(17) The Actical was worn by children during childcare hours for five consecutive days. Educators were instructed to place the accelerometer around each participating child’s waist when they first arrived at the ECC in the morning, and to remove it before the child went home at the end of the day. Once the measurement period of five work days was completed, the accelerometers were collected and sent back to the research team.
Accelerometer data were recorded in 15 second intervals. Time spent in physical activity (PA), moderate-to-vigorous PA (MVPA), light intensity PA (LPA) and sedentary time were measured based on predetermined validated thresholds for preschoolers.(17) Counts of less than 25 per 15 seconds represented sedentary time (which included nap time)(18), counts between 25 and 714 per 15 seconds represented LPA(17,18) while counts of 715 and above defined MVPA.(19) Non-wear time was defined as any period of 60 consecutive minutes where no counts were measured. To provide the most reliable data while maximizing sample size, it was determined that children had to have worn the accelerometer for a minimum of 2 hours on at least 4 days to be included in the analyses.(19) To control for within and between participant wear time variations, accelerometer data were standardized to an 8-hour period,(20) which represents the typical number of hours children in our study attended the ECC. The SAS codes used to clean and manage raw accelerometer data for this study are available as open source.(21)
Fundamental movement skills
The Test of Gross Motor Development (TGMD-II), a valid and reliable tool used to assess fundamental movement skills among children 3-11 years of age, was used to measure children’s fundamental movement skills.(22) Children were videotaped while completing two trials of each locomotor (run, hop, gallop, leap, horizontal jump) and object control skills (catch, kick, overhand and underhand throw), using the standard TGMD-II protocol. Videos were then reviewed by trained assessors who scored each skill and calculated a total raw score for locomotor skills and object control. The TGMD-II scoring protocol uses raw locomotor and object control skill scores to calculate an age adjusted Gross Motor Quotient (GMQ). The GMQ score applies a denominator which assumes that the child has performed each skill. However, some items of the TGMD were eliminated (slide, striking a stationary ball, and stationary dribble) due to the young age of the children. As a result the GMQ could not be accurately calculated for these children and thus, the Percent of Maximum Possible (POMP) scoring system was applied to score children’s fundamental movement skills.(23) The children’s raw object control and locomotor scores were converted to POMP scores to generate the maximum possible score based on the skills, which we included. This also enabled maximizing use of data for cases where children had missing data for a particular skill. For example, if a child had missing data for the run skill (i.e. because they did not want to run at the time of testing), the score for that child would be calculated on a maximum of 40 instead of having a score out of 48 as usual. POMP scores were computed, and age-adjusted as defined by the TGMD-II protocol.
Food intake and food served
Amount of food served by educators or cooks and children’s intake of vegetables and fruit (servings), fiber (g) and sodium (mg) were measured at lunch on 2 consecutive weekdays using weighed plate waste enhanced with digital photography. The intent of capturing at least two consecutive days of usual intake was to minimize the day by day variation in order to obtain a more representative measure while being logistically feasible. The weighed plate waste method has been shown to be a precise measurement of dietary intake(24,25) and has been previously used in studies conducted among school-aged children.(26–28) This method consisted of weighing a standard serving of each food item served to the children. Digital photography was also used to document the weight of the food item sitting on the scale and its type or composition (e.g. mixed dish versus a single-ingredient item). Each child’s plate was weighed and photographed before each serving and after the child was done eating. . In the cases where children served themselves rather than being provided a pre-plated meal, each child’s individual servings of food were weighed and photographed in the same manner. If a second serving was requested by a child, the same procedure was repeated. With digital photography it was possible to estimate the quantity of individual food items first served and then left on each child’s plate. Food intake was calculated as the difference in weight between the total amount of food served and the amount of food leftover.(25) Plate waste data and recipes obtained from the childcare centres were used to assess the amount of vegetables and fruit, fibre and sodium served and consumed by each child, using the ESHA Food Processor nutritional analysis software, version 10.10.00 (Salem, Oregon). Finally, amount of vegetables and fruit (servings), fibre (g) and sodium (mg) served and children’s average intake over the 2 days of data collection were calculated.
Children’s age and sex were obtained through a questionnaire administered to parents. The number of children in each ECC was based on the total number of preschoolers attending the centre. The ECCs were categorized as having 20 preschoolers or less, between 21 and 26 preschoolers or more than 26 preschoolers. The socioeconomic status of ECC was estimated based on the median income of individuals aged 15 years and older living within the same postal code as the ECC, using data from the Canadian 2011 National Household Survey.(29) Each ECC was placed into one of four socioeconomic status categories according to if their regional median income was less than $40 000, between $40 000 and $59 999, between $60 000 and $69 999, or $70 000 and above. As for geographical location, centres were defined as urban if they were in a census metropolitan area or a census agglomeration with a strong metropolitan influenced zone (MIZ), as defined by the Community Information Database, 2006.(15) Centres were categorised as being in a rural area if they were in an area with moderate, weak or no MIZ.
Opportunities for physical activity and healthy eating were assessed using 55 items (25 items related to nutrition and 30 items related to physical activity) of the Nutrition and Physical Activity Self-Assessment for Child Care (NAP SACC)(30,31) by two trained research assistants who scored the childcare centre’s environment over the two days of data collection. Each research assistant recorded their observations independently and compared their observations at the end of the second day. Excellent inter-rater reliability was shown between the research assistants (Cohen’s kappa = 0.942, p<0.001). The mean ± SD for the scores of nutrition and physical activity components of the NAP SACC are reported separately for intervention and control groups at baseline (Table 1). The 55 items were summarized into fewer categories using principle component analysis. Given NAP SACC-derived variables were ordinal, we used the untie method (PRINQUAL procedure in SAS 9.4) to transform the data, which helps retain variance of the original data for finding correlations. The factor loadings are the correlation coefficient of the relationship between categories of the NAP SACC and the underlying factors.(32) For labelling the factors, we considered all questions with factor loadings above or below the cut off of ± 0.4. Four groupings were identified to represent environmental factors related to physical activity and nutrition in ECC (see Additional file “1”).
We used complete case analysis, such that only participants with complete outcome data were included. This represents a deviation from our original protocol, which planned for analyses to be pursued according to the intention-to-treat principle.(14) This modification was necessary as the issue of missing data largely affected outcome variables, and it is generally the norm not to use imputation for missing data among outcome variables, especially when the proportion of missing data is large.(33) To assess the effect of the intervention, measures of the outcomes of interest were fitted in mixed-effect models using time of measurement (baseline or endpoint), group (intervention or control), and an interaction between time and group as fixed effects (Models 1). Additional models (Models 2) were built on these initial models to account for potentially confounding variables identified using Directed Acyclic Graphs (DAG) for each outcome.(34) These graphs are frequently used in epidemiological studies as they help illustrate the potential sources of bias and help identify confounding variables which should be controlled in the statistical analyses.(34) DAGs help researchers visually represent their hypotheses and the relationships between the variables of interest. Specifically, 2 Models were adjusted for age, sex, size of ECC, neighbourhood income, language, province, rurality and a physical activity environment score for physical activity and gross motor skills outcomes, or a nutrition environmental score for food intake and food served outcomes. To account for clustering related to repeated measures and due to the sampling of participants in ECC, variables representing participants and ECC were included as random effects in all models. In a secondary set of analyses, we tested additional interaction terms to assess whether the intervention would have different effects across different strata (i.e., sex, province—Saskatchewan, New Brunswick, language of centre—English, French, or location of centre—urban, rural). Analyses were conducted with the MIXED procedure in SAS, version 9.4.