Study design
The Transform-Us! 2 × 2 factorial design cluster randomised controlled trial including three different interventions (targeting increases in physical activity, reductions in sedentary time, or both) and a usual practice control group was delivered in Melbourne, Victoria, Australia between Feb-Jun 2010 and Nov-Dec 2012. Study details can be found in the previously published study protocol [10] and outcomes papers ([11, 12], Salmon J, Arundell L, Cerin E, Ridgers ND, Hesketh KD, Daly RM, et al: The Transform-Us! cluster RCT: 18- and 30-month effects on children’s physical activity, sedentary time and cardiometabolic risk markers, unpublished). The trial is registered with the International Standard Randomized Controlled Trial (ISRCTN83725066) and the Australian New Zealand Clinical Trials Registry (ACTRN12609000715279). Ethical approval was obtained by the Deakin University Human Research Ethics Committee (EC141–2009), the Victorian Department of Education and Early Childhood Development (2009–000344), and the Victorian Catholic Education Office (1545).
Recruitment and randomization
Primary schools within 50 km of the Melbourne Central Business District with ≥300 students and at least 2 year 3 (aged 8–9 years) classes were eligible to receive an invitation for the study (n = 219 schools). Schools were identified as being in low, mid and high socioeconomic status (SES) areas based on the first, third and fifth quintile of the Australian Bureau of Statistics Socio-Economic Index for Areas [13]. Schools in low (n = 74), mid (n = 74) and high (n = 71) SES areas were then randomly ordered with probabilistic weighting according to the number of students enrolled [10, 14]. In total, 127 primary schools were approached of which 20 (low: 8; mid: 11; high: 1) agreed to participate and were randomly allocated to one of four groups: 1) usual practice control group (C); 2) intervention group targeting increases in physical activity (PA-I); 3) intervention group targeting reductions in sedentary time (SB-I); and 4) intervention group targeting both movement behaviours (PA + SB-I). Fig. S1 in Additional file 1 presents a flow diagram with recruitment and randomisation participant numbers. Detailed randomisation and masking procedures have been described previously [10, 15]. While all children in the intervention schools (PA-I, SB-I and PA + SB-I) received the program, only those with written parental consent were included in the evaluation. Parents could elect for their child to complete any combination of the behavioural, demographics and health assessments described below.
Intervention
The intervention program was delivered by accredited Victorian Institute of Teaching classroom teachers and targeted physical activity and/or sedentary behaviours in the school and home settings [10]. The intervention used educational, pedagogical, behavioural, social and environmental strategies [10] and was based on cognitive theory [16], behavioural choice theory [17], and ecological systems theory [18]. In the first year of the intervention, all Year 3 teachers in the intervention groups were provided with detailed lesson plans and asked to deliver nine key learning messages, focusing on sedentary behaviour and/or physical activity depending on the intervention allocation (PA-I, SB-I or PA + SB-I). The intervention messages for students through the health lessons included for example ‘making active choices’, ‘importance of being active’, and ‘self-monitoring activity levels’. Homework tasks involved either adapting the child’s existing homework (e.g., to complete it while standing at a bench) or incorporating a component/task that involved reducing sedentary time or increasing physical activity (e.g., switch off screens and/or create an active game to perform with their parent). In the second year of the intervention, Year 4 teachers were asked to deliver an additional nine key learning messages which expanded/built on the previous year. Nine newsletters were sent to parents via the school each year (18 in total) that reinforced the sedentary behaviour and/or physical activity key messages (depending on the intervention group) and promoted family involvement (e.g., go for a walk with parents and count letterboxes in their street).
In addition to the key learning messages, the PA-I grade 3 classes were provided with pedometers, physical activity and novel circus equipment in classroom tubs (e.g., juggling balls), and the school was provided with asphalt line markings and signage promoting physical activity [10]. Teachers were asked to encourage and support children’s physical activity during recess and lunchtime [10]. In contrast, the SB-I teachers were asked to interrupt children’s seated class time once a day using a 30-minute active curriculum (e.g., ‘active mathematics’), to break up children’s sitting with a 2-minute ‘activity breaks’ approximately every 30 minutes, and to adapt homework tasks to break up sitting and incorporate standing [10].
The PA + SB-I schools were provided with all of the above physical activity and sedentary behaviour strategies. The usual practice control schools received the intervention information and supporting materials once the study was completed. Further intervention details have been previously described ([10,11,12], Salmon J, Arundell L, Cerin E, Ridgers ND, Hesketh KD, Daly RM, et al: The Transform-Us! cluster RCT: 18- and 30-month effects on children’s physical activity, sedentary time and cardiometabolic risk markers, unpublished).
Data collection and measures
Data were collected by trained research staff who were blinded to intervention group allocation at baseline [10]. The present study used data from two timepoints, namely Feb-May 2010 (baseline) and Nov-Dec 2011 (18 months; post-intervention). Data procedures at these time-points were the same for all measures.
Accelerometry
Participants were provided with an ActiGraph GT3X (ActiGraph, Pensacola, FL, USA) accelerometer and asked to wear it on their right hip for eight consecutive waking days, except during water-based activities [10]. The normal frequency filter was selected and data were collected and stored in 15-second epochs [19]. These data were then downloaded using the ActiLife software (ActiGraph, Pensacola, FL, USA) and reduced to daily values using a customised Excel macro. Non-wear time was defined as ≥20 minutes of consecutive zeros [19] and valid days were set at a minimum of 8 hours on weekdays or 7 hours on weekend days [20]. To optimize the sample size, participants had to have ≥3 valid days at both time points to be considered for inclusion in the study. This, in combination with the minimum valid wear time per day, gives reasonable reliability and power for assessing children’s habitual movement behaviours [20].
To determine the total duration of time in LPA, moderate- (MPA), and vigorous-intensity (VPA) physical activity, the age-specific cut-points developed by Freedson and colleagues [21] were adapted to 15-s epoch cut-points. Sedentary time was defined as ≤25 counts per 15-s [19]. Time in each intensity was then divided into sporadic time versus uninterrupted bouts of a specific intensity, using ≥5 min for sedentary bouts, and ≥ 1 min for physical activity bouts (including LPA, MPA and VPA), respectively, based on previous work using this dataset [22]. No tolerance (i.e., interruption in intensity) was allowed in defining the bouts, based on previous recommendations for sedentary bouts [15] and in the absence of recommendations for physical activity bouts. Mean values for all accelerometry variables were calculated over all valid days.
Participant characteristics
Child age and sex were self-reported at baseline or, if missing, collected from parental proxy-reports. School SES was determined using the postal code in the national 2006 Socio-Economic Indexes for Areas Index of Relative Socioeconomic Advantage and Disadvantage of the Socioeconomic Indexes Areas and classified as low (i.e., first quintile), mid (i.e., third quintile), and high (i.e., fifth quintile) SES [13]. Height, weight, and waist circumference (WC) were measured at school by trained research staff using standardized procedures with portable stadiometers (SECA 220, Los Angeles, CA, USA) and digital scales (Wederburn Tanita, Melbourne, Vic, Australia) [10, 23]. Body Mass Index (kg/m2) was calculated and converted to age- and sex- standardized z-values (zBMI) using World Health Organization (WHO) child growth charts [24].
Statistical analysis
Analyses were performed in STATA version 16 (STATACorp, College Station, TX, USA) and RStudio version 1.4.453 (R version 3.6.3; R Foundation for Statistical Computing, Vienna, Austria), using the compositions (acomp framework [23]), zCompositions (multLN zero replacement [14]) and lme4 (lmer function [25]) packages.
An analytical sample was created consisting of participants that had valid accelerometry at both timepoints (as defined above) and covariates (age, sex and SES) data at baseline. Participant characteristics (mean ± standard deviation [SD]) were compared using t-tests (continuous variables) and chi-square tests (categorical variables) between the analytical sample and those participants excluded for incomplete data. Baseline characteristics (mean ± SD) for the four groups (usual practice, PA-I, SB-I and PA + SB-I) were presented separately, but not compared using inferential tests as per the Consolidated Standards of Reporting Trials recommendations [26].
Accelerometry variables were checked for zeros. This was warranted as transforming values to isometric log ratio (ilr) pivot coordinates cannot be done for 0 as dividing by 0 or taking the logarithm of 0 are undefined mathematical operations [9]. Most accelerometry-based variables did not contain zeros, except for the VPA bouts component at both baseline (n = 1; < 1%) and post-intervention (n = 6; 2%). Consequently, these were replaced using the multiplicative lognormal imputation (multLN) replacement method of the zCompositions package with the observed minimal value at baseline and post-intervention (0.16 min [14]).
A baseline and post-intervention waking movement behaviour accumulation composition of eight components, including sporadic and bouts of sedentary time, LPA, MPA and VPA, was created using compositional data analysis (acomp) [22]. The sequential binary partition with 7 (N components in the waking compositions – 1) ilr coordinates was set up so that it enabled comparisons of a) total time spent in one intensity versus others (e.g., total time in sporadic and bouts of sedentary time versus higher intensities) and, b) time spent in bouts (e.g., ≥5-min sedentary bouts) versus sporadically accumulated time of a specific intensity. A visual overview of the sequential binary partition is displayed in Fig. 1.
As the changes between baseline and post-intervention are paired data, a “change composition” was calculated with Aitchison’s perturbation method [27,28,29]. Bar plots of geometric means of the baseline, post-intervention and “change compositions” were used to illustrate the accumulation patterns and changes in these patterns for each intervention group [30]. Linear mixed models were then fitted to assess the associations between intervention groups (as a categorical exposure variable) with the post-intervention composition (multivariate outcome). The initial partially adjusted model accounted for the random effect of individual within each school using default unstructured variance-covariance structure [31, 32] and adjusted for baseline compositions. The fully adjusted model additionally included baseline age (continuous), sex (categorical) and school SES (categorical) as fixed effects. The significance of the explanatory variables was examined with the car::Anova() function [33], which uses Wald Chi squared to calculate Type II tests according to the principle of marginality, testing each covariate after all others [32, 34]. A p-value of ≤0.05 was set as the level of statistical significance.