Design
A pre-post non-controlled design was used to report the outcomes of the Go4Fun program implemented between 2009 and 2012. In total, 293 programs were delivered by 15 Local Health Districts (LHD) across NSW. Government funding for the program is based on the proportion of overweight/obese children and geographical spread of each LHD. Written consent by a parent was a requirement for participation. The study was approved by the University of Sydney Human Research Ethics Committee.
Participants
Participants were either self-referrals (via a toll-free phone number, a text message or online registration) or referred by health professionals, relevant organisations and community members. Whilst the Ministry of Health centrally manages the program, each LHD undertakes recruitment within their area (via local media, health services, schools, councils and non-government organisations) and participates in local partnerships and promotions to target communities at social disadvantage. Children were eligible for the program if they were aged 7–13 years, however a small number of children whose birthdays were close to the age criteria were included in the program and the analysis (age 6.2-6.9 years n = 70 and 14.0-14.8 years n = 19). Children had to be overweight or obese (body mass index [BMI] ≥85th percentile based on the CDC growth references) [9], with no co-morbidities, and have a parent/adult carer to accompany them to each session. Eligibility was assessed at the time of referral/contact with LHDs and based on anthropometric measures and a medical history questionnaire completed by a parent/carer.
Program description
Go4Fun is a replication of the MEND program and the details of the program format and components have been published [10, 11]. Briefly, MEND Australia Pty Ltd is responsible for providing the Go4Fun program to the NSW Ministry of Health, including delivery of program training, resources, tools, infrastructure to monitor and track participants through the program, and program fidelity. The program comprises 2-h sessions twice a week for 10 consecutive weeks (i.e., 20 sessions), after-school during school terms. The sessions address key components for individual-level behavioural change including education, skills training, and motivational enhancement [10]. At each LDH the sessions were conducted by the same facilitators who received two days of face-to-face and online training facilitated by MEND Australia.
Data on all children enrolled in 293 programs across the 15 LHDs between 27th July 2009 and 22nd October 2012 were examined. At enrolment, parents provided socio-demographic information including their child’s sex, date of birth, postcode of usual residence, and Aboriginal status. Postcode of residence was used as a proxy for socioeconomic status (SES), based on the Australian Bureau of Statistics’ Socio-Economic Indexes for Areas (SEIFA) [12] and the Accessibility-Remoteness Index of Australia Plus (ARIA+) which determines geographical remoteness [13]. SES was based on SEIFA quintiles and ARIA+ measures were categorised as major city and other (i.e., inner regional, outer regional, remote/very remote).
Outcome measures
One trained facilitator at each LHD collected outcome measures at the first and last session of the program. Measures included the child’s height (m), weight (kg) and waist circumference (cm). BMI (kg/m2) was calculated and BMI z-scores determined from CDC reference data [9]. Waist-to-height ratios (WtHtr) were derived and categorised as <0.5 and ≥0.5 [14].
The UK MEND questionnaires and measures were used to assess differences between outcomes derived from a controlled trial situation and real world implementation. Parents completed the questionnaires on their child’s weight-related behaviours including the number of days a week their child spent engaging in one hour or more of moderate-intensity physical activity, and their perception of how active their child was compared with same-aged peers. Cardiovascular fitness was assessed by heart rate recovery one minute after completing a height-adjusted 3-min step test [15]. Screen-time was determined from the question “how many hours per week does your child spend watching TV/DVDs/video or playing on the computer/video games?”
Indicators of dietary habits and patterns included the frequency of eating breakfast (Response categories rarely; a few times a month; a few times a week; most days of the week and; everyday), daily serves of fruit and vegetables (open ended), consumption of sugar-sweetened beverages, potato chips, lollies/chocolate, and takeaway foods (response categories rarely; once a week; a few times a week; most days of the week and; everyday). For the analysis, the responses were dichotomised as ‘not frequent’ (rarely once a week) and ‘frequent’ (a few times a week; most days of the week and; everyday). An unhealthy food score was also derived as an index of the frequency of eating unhealthy foods (i.e., sugar-sweetened beverages, potato chips, lollies/chocolate, and takeaway foods), with higher scores indicating a higher frequency of eating unhealthy foods. Parents also reported on their comprehension of nutrition labels, how often they cooked “fresh food from scratch at home”, and how often the family ate meals together at the table. Children completed the adapted Rosenberg Self-Esteem Scale, which comprises 10 items on a four-point Likert scale, with higher values (score range: 0–30) indicating higher levels of self-esteem [16].
Analysis
Data analysis was undertaken using SAS software (version 9.3, SAS system for Windows). The purpose of this study was to report change in real-world, uncontrolled circumstances so missing values were not imputed. However we did assess imputation of baseline values and the direction and order of magnitude of coefficients were similar in those estimated using complete case data. Additionally, to provide policy-makers with evidence based on real world compliance where the threat of not achieving 100 % participation in community-based programs must be considered we made a pragmatic decision which would be considered useful for policy decisions and defined ‘completers’ as children who attended ≥75 % of intended program sessions and non-completers as children who attended <75 % of the program.
For continuous variables mixed models were used to estimate the mean values and 95 % confidence intervals (CIs) at each time point (before and after the program) controlling for age, sex, season, area level of socio-economic disadvantage (SEIFA, quintiles), area level of remoteness (ARIA+), and the number of sessions attended. Participant was modelled as a random factor to account for the paired before and after values on the same participant. Programme ID was also modelled as a random factor to account for the cluster design of the study. The mean change in outcome between before program attendance and after the program was assessed using a linear mixed model with change as the dependent variable and controlling for baseline as well as the factors above but with only Programme ID as a random factor. Effect sizes were determined using Cohen's d [17] and calculated using mean differences and standard errors from the mixed model where 0–0.2 is small, 0.2-0.4 is medium and 0.4+ is a large effect size.
For categorical dietary variables, generalized linear mixed models were used to calculate the prevalence of being in the frequent consumption group before the program and after the program. Again the model was controlled for the relevant fixed and random effects. The prevalences were compared using odds ratios and their 95 % confidence intervals to compare the odds of being in the frequent consumption group after the program as compared with before.
Further modeling was done to compare the outcomes of completers and non-completers. The models were analogous to those previously described but instead of number of sessions included as in independent variable in the model, a binary variable of completers versus non completers was included. In the continuous case, change from baseline scores were modeled as the dependent variable and the mean change in each group of completers and non-completers was calculated and then compared. In the binary outcome case, the probability of being in a high consumption group was modeled as the dependent variable through a logit transformation. A completer versus time interaction was added to the models to calculate both the odds ratios for the odds of being in the high consumption group after the program compared with before for the completers and non-completers groups separately. The overall odds ratios comparing completers and non-completers over both time points were also calculated. The linear mixed models were implemented in SAS v9.3 using proc mixed for the continuous outcomes and PROC GLIMMIX for the binary outcomes. Model parameters were estimated using residual maximum likelihood and residual pseudo-likelihood methods respectively.