Healthy together Victoria and childhood obesity study: effects of a large scale, community-based cluster randomised trial of a systems thinking approach for the prevention of childhood obesity among secondary school students 2014–2016

The HTV and Child Obesity study was a CRT involved 23 communities within a repeat-cross-sectional evaluation design [20]. The Victorian Department of Health selected the 23 communities that would participate in the trial, organised them in 11 strata (one with 3 communities) based on socio-demographic indices (Socioeconomic Index for Areas [SEIFA] [21]) and prevalence of unhealthy weight in adults. Adult BMI data was used for stratification because the HTV initiative targeted all ages. Communities were randomly allocated to intervention condition (n = 12) or comparison (n = 11).

A list of randomly selected schools was produced for each of the 23 participating communities. Written invitations and follow-up phone calls, emails, and/or in-person visits were held with school principals of the first three schools in the list to ascertain school participation. If all three secondary schools declined participation, then the next three schools were invited to participate until at-least one school per community participated, all schools within the community had been invited, and/or the data collection window expired. In 2016, all participating schools in 2014 were re-invited to participate, with a booster random sample of 3 schools for each community. Within each school, all Grade 8 (typically aged 13–14 years old), and Grade 10 (typically aged 15–16 years old) students were invited to participate via a 5–10 minute in-school presentation from the research team and an item about the study placed in the school newsletter. Each student received a written plain language statement and opt-out form. All students were enrolled in the study unless they returned an opt-out form signed by their parent/guardian or verbally opted out on the day. On the day of testing, students were re-informed that participation was voluntary and that they could participate as much as they like (e.g., only have their height measured, only complete the questionnaire) or opt-out at any stage. Consent was implied if children did not opt out. Data collection occurred across July–September 2014 and 2016.

The original power calculation was based on a sample of students from both primary and secondary schools, opt-in consent, and a four-year intervention (2012 to 2014) [20]. As previously published, (19) in order to detect a BMIz score change between intervention and control communities of 0.18 with a power of 0.8 and alpha of 0.05. A sample size of ≥25 children in each of the 4 -year levels (Grade 4, Grade 6, Grade 8 and Grade 10) is to be recruited in each of the 12 intervention and 12 control clusters (12*25*4 = 1200 intervention) and (12*25*4 = 1200 control). We substantially exceeded the required sample size due to the approval of an opt-out approach and have met the minimum 1200 intervention and 1200 control to examine the data separately for Grade 8 and Grade 10 [19]. Due to extremely low recruitment rates using an opt-in approach in 2013, we obtained ethics approval to use an opt-out approach in 2014 and 2016. We present here the analysis of the outcomes collected using opt-out approach for secondary school students in the years 2014 and 2016. No a priori power calculation was undertaken for this analysis.

HTV sought to take a whole of population approach to addressing rising obesity rates and preventing obesity-related chronic conditions, and involved the implementation of multiple strategies, policies, and initiatives at both the state and local levels [18, 19]. The HTV intervention targeted early childhood services, primary and secondary schools, and medium to large workplaces. Actions taken included: political leadership for health promotion; providing quality improvement frameworks for early childhood services, schools, and workplaces; supporting policy development/implementation and targeted strategies within settings (e.g. the multicomponent health and wellbeing Achievement Program within early childhood services, schools, and workplaces [22]); working with food growers, producers, and sellers to increase access to healthy food, especially for disadvantaged populations; working with urban planning to support active transport; working with sport and recreation centres to increase healthy food access; building leadership for prevention; and social marketing of health promotion messaging [18]. Due to the broad scope and the adaptive nature of HTV, providing detailed information is beyond the scope of this paper. A brief video on HTV is available at https://​www.​youtube.​com/​watch?​v=​pZU8MYGqm2s and a summary of one community’s experience (Mildura) has been published [23].

Students reported date of birth, gender, residential postcode or town/suburb, Aboriginal and/or Torres Strait Islander background, language spoken at home, country of birth, and ancestry. The Index of Community Socio-Educational Advantage (ICSEA) was used to examine socioeconomic background at the school-level [24]. ICSEA is a measure of the average educational advantage of a school’s student population compared with other schools based on child characteristics (parent/guardian occupation and education) and school factors (geographic location and proportion of Indigenous students) [24]. ICSEA has a median value of 1000, with higher values indicative of greater socioeconomic advantage. School-level ICSEA scores were obtained from the Australian Curriculum, Assessment and Reporting Authority’s My School website [25].

We conducted a repeat cross-sectional analysis on data collected from 35 secondary schools. All analyses were conducted using an intention-to-treat approach (i.e., independently of whether the intervention communities had received any level of intervention and whether the comparison communities had implemented other strategies to improve healthy behaviours or prevent obesity). We estimated the change between 2014 and 2016 on continuous outcomes (BMIz, waist circumference, abdominal obesity, health-related quality of life, and depressive symptoms) for each group and the difference in change between groups using linear mixed models. The same estimates for binary variables (overweight and obese; physical activity, sedentary behaviour, and dietary behaviours; and potential depressive disorder) were obtained using a generalised estimating equations (GEE) approach (logit link and binomial distribution). All models included fixed effects: group (intervention/comparison), wave (2014, 2016), interaction group × wave; and two school level factors, ICSEA (≤999, ≥1000), and location of the school (major cities, inner regional, outer regional). The school level factors were incorporated to adjust for imbalances in the socioeconomic level of schools participating at different waves. Clustering induced by school was accounted for by including school as random effect in linear mixed models and considering a working covariance matrix with compound symmetry in GEE models. The same models were fitted for each gender separately. No adjustment for multiplicity of outcomes was considered. All analyses were performed using SAS (version 9.4).

Strengths of this study include the embedding of the CRT within a natural experiment, scale, opt-out approach to the recruitment of adolescents, and measurement of waist circumference. Undertaking a CRT is relatively rare in evaluating systems thinking approaches for obesity prevention (WHOSTOPS is another example [16]), with non-randomised controlled trials, mixed methods evaluations, and case studies more common in the field [9]. The method is also rarely used in natural experiments like HTV and enabled the evaluation, in real time, of a government initiative. The trial evaluation was large in scale, with approximately 4000 adolescents from 23 communities participating at each of the two time points. The opt-out recruitment approach central to achieving high adolescent participation rates in 2014 (85.5%) and 2016 (90%) and exceeding the sample-size calculations of 1200 intervention and 1200 control; enabling analyses by secondary school only. These high participation rates compare favourably with student participation rates of 30–60% when opt-in procedures are used [44]. The use of waist circumference, together with BMIz, as a primary outcome recognises its strong correlation with abdominal subcutaneous adipose tissue, in particular, and visceral adipose tissue [45].

First, the very low student participation rates among adolescents at baseline constrained our ability to collect sufficient data in 2013 (390/3789 invited adolescents, 11%). Traditional (i.e., opt-in) data collection approaches, like we attempted to use in 2013, are problematic for this type of research, because they are known to introduce participant bias [46]. Therefore, we support mandatory monitoring of height and weight (e.g., as part of government child monitoring programs) to enable the effectiveness of obesity prevention initiatives and policies to be accurately assessed [47, 48]. Our use of 2014 data in the present analysis does not represent a true baseline. Given the scale of what was being attempted in HTV, however, we observed that momentum for HTV was just emerging in 2014 within intervention schools. Second, the intervention was terminated suddenly – operating for only 4 of the 6 years planned – due to an abrupt cessation of the National Partnership Agreement on Preventative Health that had provided funding for HTV. If the trial had continued, the momentum of the intervention may have produced more pronounced changes in the primary and or secondary outcomes. Third, the first data collection point for this analysis (2014) occurred 1 year after implementation had begun. Although many of the schools had not heard of HTV at the time these data were collected, it is possible that HTV affected the 2014 data. Fourth, data collection on risks for obesity (e.g., physical activity, sedentary behaviour, and diet quality) relied upon self-report measures. These types of instruments are subject to various measurement errors (e.g., respondent bias, respondent memory lapses [49], incorrect estimation of portion sizes, and complexity in estimating mixed dishes [50]) and the results should be considered in this light. Fifth, obesity prevention initiatives may have been initiated or ongoing in comparison communities at the time of the trial. Further, there is a possibility that HTV-related activities may have spread beyond the intervention communities. If they were in place, obesity prevention initiatives in control communities could have attenuated the observed effects of HTV. Sixth, we have not controlled for multiple comparisons in our analysis and, therefore, the possibility of false positives cannot be ruled out. Finally, baseline BMIz scores and prevalence of overweight/obesity were lower in children from comparison communities which may partially explain the positive change in both outcomes observed in the comparison communities.