Cluster randomised controlled trial and economic and process evaluation to determine the effectiveness and cost effectiveness of a novel intervention [Healthy Lifestyles Programme (HeLP)] to prevent obesity in school children

Article history

The research reported in this issue of the journal was funded by the PHR programme as project number 10/3010/01. The contractual start date was in March 2012. The final report began editorial review in November 2016 and was accepted for publication in February 2017. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The PHR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Stuart Logan reports grants from the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health, Research and Care. He also reports a grant from the NIHR Public Health Research programme during the conduct of the study. Colin Green served as a member of the funding panel for the NIHR Programme Grants for Applied Research programme from 2009 to 2013. Colin Green (2007–13) and Siobhan Creanor (2013–present) served as members of the NIHR Research Funding Committee for the South West Region of the Research for Patient Benefit Programme. Intervention costs for this study were paid for by the Peninsula College of Medicine and Dentistry. Stuart Logan (NF-SI-0515–10062) and Rod Taylor (NF-SI-0514–10155) are NIHR senior investigators. This study was undertaken in collaboration with the Peninsula Clinical Trials Unit (CTU), a UK Clinical Research Collaboration-registered CTU in receipt of NIHR CTU support funding. None of the funders had any involvement in the Trial Steering Committee, data analysis, data interpretation, data collection or writing of the report.

Copyright statement

© Queen’s Printer and Controller of HMSO 2018. This work was produced by Wyatt et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2018 Queen’s Printer and Controller of HMSO

Scientific background

Obesity is considered to be one of the greatest challenges facing public health in the 21st century. Currently, one-fifth of boys and girls in England start school overweight or obese, and one-third of children leave primary school (aged 11 years) overweight or obese. 1 Childhood obesity is strongly associated with socioeconomic status, with children from the least affluent decile being twice as likely to be obese as children from the most affluent decile. 2 Childhood obesity tracks into adulthood; a recent systematic review3 reported that obese children are five times more likely than non-obese children to be obese as adults, and that 80% of obese adolescents were still obese in adulthood.

School-based obesity preventative interventions have the potential to reach a large number of children and families across the socioeconomic spectrum, and schools provide the organisational, social and communication structures to educate children and parents about healthy lifestyles. Systematic reviews of school-based interventions to prevent obesity and/or increase physical activity and reduce sedentary behaviours show, at best, moderate evidence of effectiveness, with the majority of studies conducted in the USA. 4 Several methodological shortcomings have also been identified, such as differential loss to follow-up, not having sufficient statistical power to detect clinically meaningful differences between groups, particularly on adiposity outcomes, and short-term follow-up. 5,6

Evidence for the effectiveness of school-based obesity prevention programmes

In 2012, Khambalia et al. 7 examined the quality of evidence and findings from existing systematic reviews and meta-analyses of school-based programmes in the control and prevention of childhood obesity published between 1990 and 2010. Eight reviews and 112 studies were examined in total, with the most recent literature search conducted in 2008. All eight reviews acknowledged that the studies they included had heterogeneous designs, outcomes and ages of participants. The reviews were judged for their quality, and five were considered to be high quality according to their prespecified criteria.

These five reviews included three meta-analyses: Cook-Cottone et al. ,8 which examined study factors and effect sizes in obesity prevention programmes in 6- to 12-year-olds; Gonzalez-Suarez et al. ,9 which focused on the management of obesity in 11- to 18-year-olds; and Katz,10 which determined the effectiveness of school-based strategies for both obesity prevention and obesity management in 3- to 18-year-olds. The other two reviews were qualitative systematic reviews: Brown and Summerbell,11 which assessed the effectiveness of prevention programmes focusing on changing dietary intake and/or physical activity levels in 5- to 18-year-olds; and Kropski et al. ,12 which provided a focused evaluation of the quality and results of long-term obesity prevention programmes in 2- to 19-year-olds.

Despite the heterogeneity of studies included in these reviews, the authors concluded that there was evidence to recommend that school-based obesity prevention programmes include both diet and physical activity components, involve the child’s family and be of longer duration. However, they did suggest that, given no one single intervention will fit all school populations, further high-quality research needs to focus on identifying specific programme characteristics and approaches predictive of success.

Wang et al. 4 is the most recent systematic review and meta-analysis of childhood obesity prevention programmes. Of the 139 intervention studies identified, 115 (83%) were located in the primary school, of which 37 were school-based only. These studies assessed the effect of a combined diet and physical activity intervention on weight-related outcomes. The approaches included intensive classroom physical activity lessons led by trained teachers, moderate to vigorous physical activity sessions, nutrition and education materials, and promoting and providing a healthy diet. There was weak evidence that these approaches were effective at reducing body mass index (BMI), BMI standard deviation score (SDS), prevalence of obesity and overweight, percentage body fat, waist circumference and skinfold thickness. Intervention studies that reported a significant effect tended to be of long duration (between 52 and 156 weeks), with the longer-term programmes having the greatest effect. A further 28 studies were school-based with a home component (e.g. family homework lessons, training and information sheets for parents, nutrition programme for children and family). These studies provided moderate evidence of effectiveness, with half reporting statistically significant beneficial intervention effects. However, in all studies, a range of adiposity measures was used and there was high study heterogeneity in terms of setting, design, sample size, characteristics, intervention approach and length of follow-up, which makes cross-comparisons challenging. The review was unable to identify specific programme characteristics and approaches predictive of success or to explore the comparative effectiveness of specific intervention approaches (e.g. educational interventions vs. environmental intervention).

Since the publication of the above reviews, a number of additional evaluations of school-based interventions13–16 have been published involving children of a similar age to those in the Health Lifestyles Programme (HeLP) study. However, evidence of effectiveness in changing behaviours and/or weight status of children continues to be inconsistent and the content of the intervention varies greatly between studies.

In 2010, The Healthy Study Group13 published its findings from a 3-year cluster randomised controlled trial (RCT) of a multicomponent programme addressing risk factors for diabetes among American children whose race or ethnic group and socioeconomic status placed them at high risk of obesity and type 2 diabetes mellitus (T2DM). The intervention consisted of four integrated components: (1) nutrition, (2) physical activity, (3) behavioural knowledge and skills, and (4) communications and social marketing. No significant group differences were observed in the prevalence of overweight and obesity (primary outcome), but children in the intervention schools had a greater reduction in the secondary outcome of BMI SDS (–0.01 vs. –0.05; p = 0.04) at 3-year follow-up.

In 2011, Jansen et al. 15 published their findings from an 8-month cluster RCT of a school-based intervention to reduce overweight and improve fitness in 6- to 12-year-old primary school children in a multiethnic, low-income, inner-city neighbourhood in Rotterdam, Netherlands. This was a multicomponent programme based on behavioural and ecological models, involving physical education sessions delivered by a specialist physical education teacher, additional sport and play activities outside school hours, and an educational programme. Significant positive intervention effects were found in terms of the percentage of overweight children, waist circumference and 20-minute shuttle run among those aged 6–9 years. The prevalence of overweight in 6 to 8-year-olds increased by 4.3% in the control group and by 1.3% in the intervention group. No significant effects were found for BMI for 9- to 12-year-olds.

In 2014, the results of the Active for Life-year 5 (AFLY5)16 study, a UK primary school-based cluster RCT aimed at increasing physical activity, reducing sedentary behaviours and increasing fruit and vegetable consumption in 9- to 10-year-old children, were published. No differences were observed between the intervention and control groups in the three primary outcomes outlined above. The intervention was effective for three out of nine secondary outcomes after multiple testing was taken into account: self-reported time spent in screen viewing at the weekend, self-reported servings of snacks per day and servings of high-energy drinks per day were all reduced. The intervention was an adaptation of the American school-based intervention Planet Health,17 consisting of 16 lessons (delivered by the class teacher) and 10 pieces of homework in which the children were encouraged to work with their parents.

Grydeland et al. 14 assessed the effect of a 20-month cluster RCT of a school-based intervention trial on the BMI of 11-year-old children in Norway. The intervention took a multilevel approach including collaboration with school principals and teachers, school health services and parent committees. School teachers were the key in delivering the intervention components, which included lessons, fact sheets for parents, meetings and awareness raising of leisure time activities. There was no significant effect of the intervention overall, although significant effects were found for both BMI (p = 0.024) and BMI SDS (p = 0.003) in girls. Furthermore, children of higher-educated parents seemed to benefit more from the intervention, highlighting the need to develop interventions that do not widen the health inequalities that already exist between children from different socioeconomic groups.

Rationale for HeLP

We began work to design, pilot and then fully evaluate a school-based obesity prevention programme in primary schools in 2006. Our aim was to create a novel programme of activities using an intervention mapping approach18 and to follow the UK Medical Research Council’s (MRC) framework for the development and evaluation of complex interventions. 19 The intervention mapping approach combines theoretical and stakeholder perspectives, empirical evidence and new research to assess and develop a potential set of possible solutions for a particular problem, rather than defining a practice or research agenda around a specific theory. We therefore searched the literature for appropriate systematic reviews of school-based obesity prevention programmes and carried out extensive stakeholder consultation with practitioners (teachers, head teachers and drama specialists), local policy-makers (director of public health and the public health policy lead), public health commissioners and the local healthy schools team, in order to understand the school, education and public health context at that time. The overarching aim of the research was to develop a school-based intervention, assess its feasibility and complete pilot work before we sought funding to run the full-scale effectiveness trial, were that deemed appropriate. 20

Around the time of our initial research, the Brown and Summerbell review11 concluded that interventions that increase activity and reduce sedentary behaviours may help children to maintain a healthy weight, although the results were short term and inconsistent. Their conclusions were that a combined approach was likely to be more effective in preventing children becoming overweight in the long term.

We wanted to incorporate these recommendations into our programme and address the methodological limitations reported in other trials, such as insufficient statistical power; biased recruitment from, and retention of, more affluent schools, children and families; short follow-up periods; and high levels of attrition. Thus, a crucial focus in developing HeLP was to promote the engagement of schools, children and their families throughout the intervention, as we believed this to be essential for behaviour change to occur. In line with the World Health Organization’s Health Promoting Schools framework,21 we aimed to develop activities that were compatible with the existing school curriculum and promote messages in a manner that could have an impact on the wider school culture, as well as specific behaviours of children and their families. First and foremost, we sought to build supportive and trusting relationships with teachers, children and their families by employing people with specific skills and competencies to deliver the intervention.

We developed an initial phase of the intervention to create a receptive context for the programme and utilised engaging delivery methods to try to increase school, child and family engagement with the programme. Previous research into preventing childhood obesity has found that it can be difficult to engage parents in order to affect change within the family;22,23 thus, we believed that the delivery methods needed to be sufficiently dynamic, creative and empowering to motivate the children to talk about the activities at home with their parents and encourage parents to come into the school to attend key events. This thinking led us to explore the use of interactive drama, as it had shown promise in promoting positive attitudes towards a number of health behaviours24 and was a means of delivering a range of behaviour change techniques in manner that could engage the children.

The development of the intervention was guided by the information, motivation and behavioural skills (IMB) model25 and intervention activities were ordered to enable support and sustain behaviour change in accordance with the health action process model. 26

Between 2006 and 2008, we worked with 8- to 11-year-old children in one local primary school to assess a range of delivery methods for key objectives identified in the intervention mapping process. We then went on to refine the newly developed intervention based on feedback from teachers, children and parents, and carried out a feasibility study (examining before-and-after intervention changes in the same children) in a second primary school from a more economically deprived ward of Exeter, Devon. In the feasibility study we focused solely on children aged 9–10 years, as our early work demonstrated that this age group was more receptive to the messages and, crucially, engaged their families to the greatest extent. 27 Teachers also informed us that the Year 5 curriculum had the flexibility to deliver the healthy lifestyles week, whereas in Year 6 there was pressure to focus on Standard Assessment Tests (SATs). Based on qualitative data from teachers and head teachers in the piloting stages, the intervention was extended to add activities at the beginning of Year 6. Teachers felt that this would not have an impact SATs teaching and was important in further motivating the children to change behaviours. Minor refinements were also made to the education lessons and the drama scripts to enhance delivery and continuity. In 2008 we carried out an exploratory RCT [funded by the National Institute for Health Research (NIHR) Research for Patient Benefit programme] with 202 children aged 9–10 years in four schools to assess ‘proof of concept’ and the feasibility of running a cluster RCT of HeLP. The trial design and the intervention were feasible and acceptable to schools, children and their families. At 18 months, compared with children in the control schools, children in the intervention schools consumed fewer energy-dense snacks and more healthy snacks, had less negative food markers and more positive food markers, had lower mean television/screen time and spent more time doing moderate to vigorous physical activity every day. 28 Children in the intervention group had lower average anthropometric measures at 18 and 24 months than children in the control group, with larger differences at 24 months than at 18 months for all measures. 28

Funding for the definitive trial was awarded in March 2012 and the trial began in September 2012. None of the schools that were involved in the feasibility and pilot work were included in the definitive trial presented in this monograph.

Rationale for our study design

The HeLP study was designed to address some of the methodological weaknesses in previous school-based RCTs to prevent obesity. Schools have been highlighted as providing a captive audience with the potential to access children across the socioeconomic spectrum; however, few studies took a relational approach to both the intervention and the trial design, whereby the building of relationships with all participants to enhance engagement with the programme and the trial was a key focus in recruitment, data collection, and intervention design and delivery.

Specifically, the design of the HeLP study ensured that (1) baseline measurements were collected before schools (and children) were allocated to the intervention or the control group; (2) a standardised objective measure with a significant follow-up time was utilised to assess obesity prevention (BMI SDS at 24 months); (3) an objective measure of physical activity, with evidence of high compliance, was used29 (GeneActiv activity monitor; www.activinsights.com); (4) half of the schools had ≥ 19% of children eligible for free school meals (the national average at the time of the trial) to ensure that the results were generalisable across the socioeconomic spectrum; and (5) loss to follow-up was minimised by creating detailed standard operating procedures for the collection of all measures at each time point, with a focus on enhancing positive relationships with teachers, children and parents.

During intervention delivery we ensured that (1) delivery personnel had the necessary skills and competencies to build relationships; (2) there was one key contact person per school who co-ordinated the research hand delivery (the HeLP co-ordinator); (3) the delivery methods used were suitably engaging and dynamic for the target group, so that the children would be motivated to take the messages home and initiate discussion with their family; (4) the intervention used strategies to enhance identification with, and ownership of, the healthy lifestyle messages and enabled children to explore possible solutions and propose changes in preparation for real-life situations; (5) information and guidance on promoting healthy eating and physical activity went home to parents throughout the intervention period; and (6) parents were involved in setting goals with their child.

The intervention aimed to change behaviours both inside and outside the school environment; these measurements were concerned with behaviours across the week, including weekends. We aimed to capture physical activity data across the whole week for children at baseline and at 18-month follow-up. To support this aim, we used wrist-worn activity monitors (the GeneActiv accelerometer), as feasibility studies had demonstrated high rates of compliance with this monitor in this age group. 30 We took account of the clustered nature of the trial design in the sample size calculation and analysis, and examined the effect of the intervention 6 months (18 months post baseline) and 12 months (24 months post baseline) after its completion in order to examine whether or not any observed effects of the intervention were sustained (see the published trial protocol31).

Aim and objectives

The aim of this cluster RCT was to determine the effectiveness and cost-effectiveness of HeLP in preventing overweight and obesity in children.

Our objectives were:

1. to assess the effectiveness of HeLP in children aged 9–10 years by comparing between intervention and control schools:

   1. BMI SDS at 24 months (primary outcome)

   2. BMI SDS at 18 months

   3. waist circumference SDS at 18 and 24 months

   4. percentage body fat SDS at 18 and 24 months

   5. proportion of children classified as underweight, overweight and obese at 18 and 24 months

   6. physical activity (average time spent per day in sedentary, light, moderate, moderate to vigorous and vigorous activity) and average weekly volume of physical activity in milligravity (mg) units at 18 months

   7. food intake (number of energy-dense snacks, healthy snacks, negative and positive food markers) at 18 months.

2. to estimate the costs associated with the delivery of the HeLP intervention and its cost-effectiveness versus usual practice

3. to conduct a mixed-methods process evaluation and mediational analysis to explore the way the programme worked (i.e. how it was delivered, taken up and experienced, and the possible mediators of change).

Study design

We carried out a pragmatic, superiority, cluster RCT with blinded outcome assessment, allocating schools (1 : 1) to either the HeLP intervention or usual school practice. As the intervention was designed to be delivered in schools, a cluster design was used. Individual child measurements were collected at baseline, 12 months [My Lifestyle Questionnaire (MLQ) only, which assessed potential mediating variables], 18 months and 24 months (anthropometric only) post baseline. Alongside the evaluation of effectiveness, we carried out a parallel economic and process evaluation.

Following the exploratory trial28 and after assessing delivery requirements, we decided to run the trial in two cohorts, each with the same number of intervention and control schools. All schools were recruited in spring 2012 and then allocated to cohort 1 (commencing the trial in September 2012) or cohort 2 (commencing the trial in September 2013).

Randomisation, allocation concealment and blinding

Randomisation was by school. All schools were initially randomly allocated to intervention or control by a computer-generated sequence that was stratified by (1) the proportion of children eligible for free school meals (< 19% or ≥ 19%) and (2) school size (one Year 5 class or ≥ 2 Year 5 classes). For practical reasons, half of the schools commenced the study in 2012 (cohort 1) and the other half commenced it in 2013 (cohort 2). Randomisation was performed by a statistician in the UK Clinical Research Collaboration-registered Peninsula Clinical Trials Unit immediately after all schools had been recruited (i.e. in 2012) but each school’s allocated group (intervention or control) was not communicated to the schools, parents or researchers until after the baseline measures had been taken in each cohort (2012 for cohort 1 and 2013 for cohort 2). The Peninsula Clinical Trials Unit ensured that there were equal numbers of control and intervention schools in both cohorts in order to facilitate trial delivery.

Figure 1 shows the timeline cluster33 for the HeLP study and Table 1 provides the key to the figure.

FIGURE 1.

Timeline cluster diagram for HeLP. Reproduced from Lloyd et al. 32 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

Reproduced from Lloyd et al. 32 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

TABLE 1 - Key to Figure 1

FIQ, Food Intake Questionnaire.

Reproduced from Lloyd et al. 32 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

Item	Description
1	Identification of schools Trial team attended Devon primary head teachers’ meetings and network events
2	Recruitment of schools All Devon state primary and junior schools with children in at least one single Year 5 group of ≥ 20 children eligible to participate. Schools that expressed interest in participating were purposely sampled to represent a range of number of Year 5 classes (1–3 Year 5 classes), locations (urban and rural) and deprivation (< 19% and ≥ 19% of children eligible for free school meals). The remaining schools were placed on a waiting list for cohort 2 (two schools allocated to cohort 2 dropped out before commencing the trial and were replaced by a school from the waiting list)
3	Identification of children All children in Year 5 in each recruited school were eligible to participate
4	Recruitment of children Information sent to Year 5 child’s parent/carer, who had the opportunity to opt out child from trial
5	Randomisation Allocation of schools to intervention or control group was done using a computer-generated sequence, stratified on school size (one Year 5 class vs. ≥ 2 Year 5 classes), and percentage of pupils eligible for free school meals (< 19% vs. ≥ 19%) as an indicator of school-level socioeconomic status. After randomisation to intervention or control, schools were allocated to cohort 1 or cohort 2 by a statistician from the Clinical Trials Unit, with equal numbers of intervention and control schools in both cohorts. Allocation remained concealed from all delivery personnel, schools and children until after baseline measures were captured in each cohort
6	Baseline measures Performed by trained assessors, who were blinded to schools’ allocated groups
7a	Intervention delivery HeLP intervention was delivered to schools and children who were allocated to the intervention group, with no blinding for schools, HeLP co-ordinators, children or parents/carers
7b	Usual care Children in schools allocated to control group received usual care, with no blinding for schools, HeLP co-ordinators, children or parents/carers
8	12-month measure Self-completed MLQ
9a	18-month measures Anthropometric measures collected by trained independent assessors blinded to group allocation
9b	18-month measures Self-completed FIQ while children were still at primary school, 18 months post baseline. Physical activity data from a subset of children while they were still at primary school, 18 months post baseline, were objectively assessed
10	24-month measures Anthropometric measures were collected by trained independent assessors, blinded to group allocation, after children had moved to secondary school (secondary schools had a mix of children from intervention and control schools), 24 months post baseline (12 months post intervention)

Statistical analysis

A detailed analysis plan was developed by the Trial Management Group and approved by the TSC in November 2015. 42 Three minor amendments were made following the TSC meeting in July 2016 (see Appendix 5). These amendments were approved by the TSC chairperson in September 2016. The analysis plan and a summary of the amendments are published on the NIHR Public Health Research programme website, along with the study protocol. 43

Participant numbers through the HeLP study

Figure 2 shows the flow diagram of the progress of schools and children through the phases of the trial. The target of recruiting 32 schools, to ensure that a minimum of 28 schools completed the trial, was met, with all 32 completing the trial. The number of children in each school was greater than we had anticipated before the recruitment of schools started (the prior assumption was a mean of 35 Year 5 children per school). Therefore, the total number of eligible children was greater than expected. All 1371 children in the 32 participating schools were eligible for the trial: 34 were opted out by their parent/carer and 13 left the school before baseline data collection (and randomisation). All of the remaining 1324 children, irrespective of whether or not they had (all) measurements collected, were included in the trial, and so the minimum requirement of 952 children recruited was far exceeded.

FIGURE 2.

Trial profile.

Up to four visits were made to each school in an attempt to obtain measures for children who had been absent on the original day of data collection, with only a small number of children absent at all subsequent visits. As described in Chapter 2, one class per school was randomly sampled to complete the physical activity data collection. At 24 months, there was a slightly greater loss to follow-up of children in the intervention group than in the control group (6.1% vs. 3.7%). The number of children included in the primary analysis of BMI SDS at 24 months was 1244, 63% greater than the required 762 children from the a priori sample size calculation.

Figure 2 shows the trial profile. ‘N’ refers to the number of schools (clusters) and ‘n’ refers to the number of children (individual participants). Two schools that had been allocated to cohort 2 withdrew while they were waiting to commence the trial and so these were subsequently replaced with two of the four schools on the waiting list before cohort 2 commenced. All schools that started the trial remained in it, and so all of the randomised clusters are present at baseline and at each follow-up point. The percentage given in brackets for the proportion of children with data at both baseline and follow-up is calculated from the total number of recruited children in the schools at baseline. Not all children with a follow-up measure necessarily had a corresponding baseline measure (or vice versa) owing to different children being absent on the day of the main and additional assessments for each of the time points and/or owing to children leaving or moving schools. In all of the analyses, children were analysed in the group (intervention or control) to which the school they were enrolled in at baseline was randomised.

Completeness of outcome measures through the HeLP study

Figure 3 shows the completeness of the outcome measures at each of the time points. A small number of children at each time point did not assent to one or more of the anthropometric measures. In summary, between 93% and 99% of children provided data on BMI, waist circumference, percentage body fat and FIQ. The proportion of children who had usable physical activity data files was 96% at baseline and > 88% at 18 months; similarly, the proportion of children with valid physical activity data was 94% and 84% at baseline and at 18 months, respectively, following the application of the minimum wear requirements (3 weekdays and 1 weekend day, each with a minimum of 10 hours of wear time per day). There was no evidence of differences between allocated groups in terms of the completeness of outcome measures throughout the trial, with the exception of the small difference at 24 months for BMI SDS.

FIGURE 3.

Completeness of measures across time points. PA, physical activity.

All anthropometric measures were taken blind to group allocation. A further assessment of the success of allocation concealment was made at the 24-month data collection point, when there was a mixt of children from intervention and control primary schools in each secondary school, with the independent (blind) assessors requested to indicate whether or not the child had revealed their allocated group; no child had reported their allocated group.

Figure 3 shows the completeness of the measures across the time points. ‘N’ refers to the number of schools (clusters) and ‘n’ refers to the number of children (individual participants). The percentage given in brackets for the proportion of children with data at baseline and follow-up is of the total number of recruited children in the schools at baseline. Not all children with a follow-up measure necessarily had a corresponding baseline measure (or vice versa) owing to different children being absent on the day of the main and additional assessments for each of the time points and/or owing to children leaving or moving schools. In all of the analyses, children were analysed in the group (intervention or control) to which the school they were enrolled in at baseline was randomised. There were no differences in missing anthropometric data by allocated group at each time point.

Baseline characteristics

Primary intention-to-treat analysis of body mass index standard deviation score at 24 months

The unadjusted mean BMI SDS at 24 months was slightly higher in the intervention group, at 0.35, than in the control group, at 0.22, with SDs of 1.25 and 1.22, respectively (Table 9). The effect of the intervention was estimated allowing for clustering within schools, modelled as a school-centred random effect, and other prespecified variables. The results showed no evidence of an intervention effect, with an estimated between-group mean difference (intervention minus control) of –0.02 (95% CI –0.09 to 0.05) (see Table 9), and an estimated ICC of 0.014 (95% CI 0.003 to 0.069). Without adjustment for the baseline BMI scores and gender of each child, cohort and the stratification factors, but after allowing for the school-level clustering, the mean BMI SDS score was 0.11 (95% CI –0.11 to 0.33) greater in the intervention group than in the control group, but this difference was not statistically significant (see Table 9).

Secondary analysis of body mass index standard deviation score at 24 months

All of the secondary analyses that were undertaken produced results that were consistent with the primary analysis. An alternative computational approach using generalised estimating equations yielded an adjusted between-group mean difference of –0.02 (95% CI –0.08 to 0.05), allowing for clustering around schools and adjusting for the same variables as in the random-effects model in the primary analysis (see Table 9). The results from the two approaches are, thus, consistent with each other, with no evidence of an intervention effect.

The sensitivity analysis of the fully adjusted random-effects model based on the worst-case scenario for missing observations at 24 months estimated an increase in mean BMI SDS of 0.07 (95% CI –0.02 to 0.16) for the intervention group relative to the control group (see Table 9). Based on the best-case scenario for the same missing observations, the direction and size of the effect of the intervention changed to –0.02 (95% CI –0.09 to 0.05). Therefore, neither scenario provided any tentative evidence of a significant or meaningful intervention effect (see Table 9).

Longitudinal analysis of body mass index standard deviation score

The analysis was based on fitting a longitudinal model with observations at fixed time points (baseline, and 18 months and 24 months post baseline). The complexity of the between-time-point covariance for each child was reduced to an autoregressive pattern of order one. As the outcome of this longitudinal model comprised all BMI SDS measures, including those at baseline, the effect of the intervention was determined from the interaction between the allocated group and the time point.

According to this model, there was no significant difference between the two groups at baseline (intervention BMI SDS greater than controls by 0.117; p = 0.170). The BMI SDS increased significantly in both groups at each subsequent time point relative to baseline, by an average of 0.04 (95% CI 0.01 to 0.06) at 18 months and by 0.05 (95% CI 0.01 to 0.09) at 24 months, although, relative to the control group, the increase in the intervention group was attenuated, not significantly, by –0.03 (95% CI –0.08 to 0.01) at 18 months and by –0.02 (95% CI –0.07 to 0.04) at 24 months.

Figure 4 shows the predicted marginal BMI SDS by allocated group across the time points of the trial.

FIGURE 4.

Predicted marginal BMI SDS with 95% CIs in the intervention group (black with dashed connecting lines) and control group (green with solid connecting lines) across time points. According to the longitudinal model, allowing for hierarchical clustering by child within each school, modelling the within-child covariance between fixed time points as an autoregressive pattern of order one. Reproduced from Lloyd et al. 32 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

Reproduced from Lloyd et al. 32 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

Primary analyses of secondary outcomes at 24 months (anthropometric measures only)

Table 11 shows the summary data of the remaining anthropometric measures at 24 months post baseline. Among the children in the intervention group, slightly more were classified as being overweight or obese, according to the Public Health England definition of categorisation of BMI scores at 24 months, relative to the control group (30.8% vs. 26.6%). However, analysing the weight status categories under the assumptions of an ordinal logistic model with school-level random effects, there was no evidence of an intervention effect in the fully adjusted model (p = 0.472). The odds of being in the higher weight categories were greater among the children in the intervention group than among those in the control group, although this change in odds was not statistically significant [odds ratio (OR) 1.15, 95% CI 0.79 to 1.67]. The intervention effect size was similar in both the unadjusted and adjusted models. Relaxing the assumption of proportional odds, the estimated intervention effect size from the generalised ordinal model, with a robust error variance clustered on schools, was comparable with that from the random-effects ordinal logistic model (results not shown). According to the generalised ordinal model (i.e. the model without assuming proportional odds across the weight status categories), the odds of being classified as overweight or obese relative to healthy or underweight were 1.14 (95% CI 0.77 to 1.68) in the intervention group relative to the control group. The odds were 1.17 (95% CI 0.74 to 1.84) for the risk of being obese relative to the lower weight status categories in the intervention group relative to the control group. The test of proportional odds under the Peterson–Harrell parameterisation indicated that there was no evidence of a change in the odds between binary comparisons of levels of weight status category (p = 0.503), indicating that the assumption of proportional odds for intervention effect had not been violated.

Consistent with the primary outcome measure, the average anthropometric measurements were generally slightly larger for the children in the intervention group at 24 months, as they were at baseline, relative to those in the control group (see Table 11). An exception was found in the standardised body fat measurements before the exclusion of the extreme values (implausible absolute SD scores of ≥ 5). The unadjusted mean differences (intervention minus control) were positive for all anthropometric outcomes except the standardised body fat score that included extreme values. However, there was no evidence of an intervention effect from the unadjusted random-effects model, allowing for clustering on schools. After adjusting for the stratification variables, cohort, gender and the corresponding baseline measurements, all mean differences were negative, indicating lower, albeit non-significant, measurements in the intervention group at 24 months than in the control group.

Primary analyses of outcomes at 18 months

Tables 12–14 show the summarised data of the primary and secondary outcomes at 18 months post baseline. Overall, fewer children were classified as overweight or obese than at 24 months, but more children in the intervention group than in the control group were already overweight or obese at 18 months (28.9% vs. 25.4%). Although not statistically significant, this difference was reflected in both the adjusted and unadjusted analyses when the weight categories were modelled as an ordinal outcome. No violation of the proportional odds assumption was detected, according to the test of the Peterson–Harrell parameters from the generalised ordinal model (results not shown).

The anthropometric outcomes at 18 months displayed a similar pattern to those at 24 months, with larger mean values for children in the intervention group than for children in the control group, and positive mean between-group differences (intervention minus control) from the unadjusted analyses, except for the standardised measurement of percentage body fat with extreme values included (extreme values defined as absolute standard scores of ≥ 5) (see Table 12). After adjustment for all of the prespecified variables, the mean between-group differences in anthropometric measurements were all negative, indicating lower values in the intervention group at 18 months than in the control group. However, there was no evidence of a significant intervention effect for any of the anthropometric measures at 18 months.

There was no significant effect of the intervention on any of the physical activity outcomes (see Table 13). Measures of physical activity, average weekly volume and average time per day in each intensity category were all fractionally larger in the intervention group from a direct comparison of the mean values of the two groups and from both the unadjusted and adjusted analyses. Conversely, the accelerometry data classified as sedentary behaviour (between 6 a.m. and 10 p.m.) were, on average, lower in the intervention group. However, none of the observed differences was statistically significant.

The mean scores of the FIQ outcomes averaged over the week were higher in the intervention group for healthy snacks and positive food markers and lower for the energy-dense snacks and negative food markers (see Table 14). These between-group differences were statistically significant in the unadjusted analysis for the healthy snacks and positive food markers. However, once the analysis was adjusted for baseline score, stratification variables and other potential confounders, these differences were no longer statistically significant, while the between-group differences for energy-dense snacks and negative food marker scores were statistically significant in the adjusted analyses, with the strongest evidence of an intervention effect observed for energy-dense snacks (p = 0.017). According to the adjusted analysis, the mean average weekly scores for the intervention group were 0.37 (95% CI –0.66 to –0.07) lower than for the control group for the energy-dense snacks score and 0.47 (95% CI –0.91 to –0.02) lower for the negative food markers score.

The pattern of differences observed for the weekly averaged FIQ outcomes was repeated for the weekday scores. There was weak evidence of between-group differences from the unadjusted analyses for the energy-dense snacks and negative food marker scores, as well as for the healthy snacks score. After adjustment for the prespecified variables there was evidence of statistically significant differences for the energy-dense snacks (p = 0.013) and negative food markers (p = 0.020). The intervention effect sizes were of a similar magnitude and direction for the two unhealthy food scores, with the mean average weekday energy-dense snacks score being 0.47 (95% CI –0.84 to –0.11) lower in the intervention group than in the control group, and 0.64 (95% CI –1.17 to –0.11) lower for the negative food markers score.

At the weekend, the pattern of the differences between the two groups was the same, with higher mean scores for healthy snacks and positive food markers and lower mean scores for energy-dense snacks and negative food markers for the intervention group than for the control group (see Table 14). The between-group differences for the healthy snacks and positive food markers scores were statistically significant in the unadjusted analyses (p = 0.035 and 0.030, respectively); however, after adjustment for the prespecified variables, neither between-difference was statistically significant (p = 0.086 and 0.145, respectively).

Intraclass correlation coefficients

Table 16 shows the ICC and 95% CIs for each of the outcomes from the random-effects models. The estimated ICC for the primary outcome was 0.014 (95% CI 0.003 to 0.069), slightly lower than the estimate of 0.02 used in the sample size calculation. The estimated ICC for weight status when dichotomised (overweight and obese vs. healthy and underweight) was extremely small (estimated as 0.000). This was due to the estimated between-cluster variance for this outcome being particularly small relative to the variance of the standard logistic distribution, assumed to be that of the latent continuous distribution that determines the weight of status of individuals. The wide CI for this ICC is an artefact of the estimated parameter being so close to the boundary of the parameter space, and the logit function,56 which yielded a lower bound very close to zero and an upper bound at one. The estimated ICCs for waist circumference and waist circumference SDS were higher than for the other anthropometric measures.

Complier average causal effect analysis

The original analysis plan included a complier average causal effect (CACE) analysis to estimate the CACE of treatment, as a potentially unbiased estimate of receiving HeLP, with non-compliers defined as children who did not receive at least four sessions of drama and the one-to-one goal-setting session. The TSC subsequently deemed the planned analysis unnecessary, given the small number of children who ‘switched’ between allocated treatment groups: there were only 24 out of 675 children allocated to the intervention group who were categorised as non-compliers (3.6%) and 4 of these 24 did not provide primary outcome data.

Introduction

The aims of the economic analyses were to estimate the expected resource use and related costs associated with delivery of HeLP (i.e. the HeLP intervention) in a policy-relevant setting, and to estimate the cost-effectiveness of the HeLP intervention where it is shown to be an effective intervention compared with usual practice alone.

Given the effectiveness profile for the HeLP intervention, presented in Chapter 3, and the absence of statistically significant differences on the primary outcome measure of BMI SDS, we begin here with a summary statement on the cost-effectiveness of HeLP versus usual practice. We find that there is no expectation, using the framework for economic evaluation developed/described below, of improvements in the likely incidence of weight-related health events [e.g. coronary heart disease (CHD)], or related cost savings to the health and social care system, from the HeLP intervention. As described in the following sections, HeLP incurs additional costs for delivery of the intervention with no identified or expected health gains, using the analytical framework developed and presented here, and therefore usual practice ‘dominates’ HeLP in a resource allocation context (i.e. HeLP is not cost-effective compared with usual practice).

Although we highlight the results of the cost-effectiveness analysis at the outset in this chapter, for completeness, and to inform the reader and to inform future research, we provide detail in this chapter on the methods used, and the results of the cost analysis estimating the cost for delivery of HeLP. We also provide a narrative and detail on the development of a decision-analytic modelling framework developed to assess the cost-effectiveness of HeLP, using this to provide illustrative and exploratory cost-effectiveness analyses.

Below we summarise the methods used to estimate resource use and costs for delivery of HeLP, and the methods used to develop a decision-analytic modelling framework. We then present results for these two areas of research. When describing development of a modelling framework, we present summary details of a literature review of model-based economic evaluations that model childhood obesity interventions through adult years, and we describe the development of a modelling framework in this context, and the areas of evidence synthesis required to populate the model.

Methods

Results

A framework for the economic evaluation of HeLP

Here we describe the development of the Exeter Obesity Model, a decision-analytic model-based framework for assessment of the cost-effectiveness of the HeLP school-based intervention versus usual practice in children aged 9–10 years. The development of the model was based on a review of model-based economic evaluations of interventions related to childhood obesity, with a focus on models tracking from childhood through to adult years, and a summary of this systematic literature search is presented below. Thereafter we document the conceptual structure and the operational development of the model, describing data sources and the populating of the model to address the primary research question of whether or not the HeLP intervention could be considered cost-effective compared with usual practice.

Review of decision-analytic models for use in cost-effectiveness analyses on obesity

A systematic review of model-based economic evaluations of interventions related to childhood obesity, as a means of preventing obesity in adults, was conducted in order to inform the structure and development of a modelling framework suitable for assessing the cost-effectiveness of the HeLP intervention.

Methods

Search strategy

Systematic searches were conducted in several databases (including MEDLINE, EMBASE, PsycINFO, NHS Economic Evaluation Database and Database of Abstracts of Reviews of Effects). The keywords and medical subject headings (MeSH) for obesity were combined with keywords for cost and economic analysis to capture relevant economic evaluations in all databases. The search strategy is detailed in Appendix 8. In addition, references within the included studies were reviewed to identify potentially overlooked studies. The searches were date limited 1990–2012 (original search). Language restrictions were not applied. Update searches were conducted in 2015.

Eligibility criteria

Titles and abstracts for all articles identified by the literature search were screened. Studies were eligible for inclusion if they were economic evaluations (cost–benefit, cost–utility, cost-effectiveness and cost–consequence analyses) of interventions related to child/adolescent overweight/obesity as a means of preventing obesity in adults. Next, full texts of all articles deemed eligible from the previous stage were reviewed to further confirm their eligibility. At this stage, the exclusion criteria were burden-of-illness studies, cost analyses, study types other than economic evaluations, review articles, editorials or reports, and articles not written in English. Studies published as abstracts or conference presentations were included only if sufficient details were presented to allow both an appraisal of the methodology and an assessment of the results to be undertaken.

Data extraction and synthesis

For each study included in the final review, data were extracted into a table. The information obtained included authors, country, study year, comparators, population characteristics, study design, time horizon, perspective, data sources, effectiveness, cost, sensitivity analysis and study conclusions. The identified modelling studies are summarised, and the approaches available to model disease progression over time are discussed in a summary descriptive review.

Results

Studies identified

A total of seven models (reported in 11 publications) were included in the final review. 62–72 The study identification process is summarised in Figure 5.

FIGURE 5.

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram.

The study characteristics for the seven included models are summarised in Table 26.

TABLE 26 - Summary characteristics of the identified models

Active PE, active physical education; APPLE, Alberta Project Promoting active Living and healthy Eating (in Schools); CATCH, Coordinated Approach to Child Health; CG, clinical guidelines; CRC, colorectal cancer; DALY, disability adjusted life-year; ECE, early care and education policy change; EFS, Epidemiologic Follow up Study; EU, European Union; HW, healthy weight; IASO, International Association of Obesity; ICER, incremental cost-effectiveness ratio; NHANES, National Health and Nutrition Examination Survey; OB, obese; OW, overweight; PE, Project Energise; RoW, rest of world; RR, relative risk; SE, standard error; SSB, sugar-sweetened beverage; TV AD, television advertising; WS, weight status.

The first year is a transitional state in which participants remain in for one cycle in this state, after which they can move to a follow-up state or die.

Medical costs estimated were those associated with events of fatal and non-fatal CHD, hypertension, diabetes, symptomatic gallstones, and osteoarthritis expected to occur during a 25-year period in each weight cohort.

Obesity-related conditions included chronic heart failure and T2DM.

DALY outcomes only reported for the SSB outcome as for the other three interventions participants will be aged < 30 years at follow-up and RRs of obesity-related diseases are 1.0 at the age of < 35 years.

QALY outcomes not reported for the ECE or Active PE interventions as few participants would be adults after 10 years.

Calculated for SSB, TV-AD and ECE interventions only.

First author, year	Country setting	Intervention	Outcomes	Analytical approach, WS	Health events	Intervention effect	Time horizon/cycle length	Perspective
Wang et al., 200362	USA	Planet Health	Cases of adulthood OW prevented; costs; QALYs; ICER	Extrapolation from childhood WS to adult (21–29 years) WS using Whitaker et al.73 and from 21 to 29 years using NHANES EFS	Estimated 25-year direct health-care and medication costsb in women aged 40 years who maintain an OW status to age 65 years	Intervention effect expressed as probability of OW and at-risk OW at age 11 years	25 years/NA	Societal perspective (health-care sector costs and costs related to productivity loss)
Brown et al., 200763	USA	CATCH	Cases of adulthood OW prevented; costs; QALYs; ICER	Based on – same as – Wang et al.62	Estimated 25-year direct health-care and medication costsb in women aged 40 years who maintain an OW status to age 65 years	Intervention effect expressed as probability of OB and non-OB at age 14 years	25 years/NA	Societal perspective (health-care sector costs and costs related to productivity loss)
Hollingworth et al., 201264	UK	Lifestyle interventions	Costs; incidence of weight-related events; cost per life-year gained	Extrapolation over time estimated using a BMI growth equation simulated using Monte Carlo techniques	Events include CHD, stroke, diabetes, breast cancer, CRC	Intervention effect modelled as reduction in BMI SDS	Lifetime/1 year	Third-party payer (UK NHS)
Pil et al., 201465	Six EU countries (Belgium, Bulgaria, Germany, Greece, Poland and Spain)	ToyBox	Costs; incidence of weight-related events; QALYs; ICER	Extrapolation from childhood WS to adult (30–34 years) WS using Venn et al.;75 projection of the incidence of events (IASO) and mortality from age 30–34 years onwards (Markov model)	Event-free; CHD;a T2DM; stroke;a CRC;a breast cancer;a death Suffering from > 1 condition not possible in the model	Intervention effect expressed as a relative risk reduction applied to country-specific probabilities of the weight categories in the adult general population	70 years/1 year	Societal perspective (health-care sector costs and costs related to productivity loss)
Rush et al., 201466	New Zealand	Project Energise	Costs; QALYs; ICER	Population (aged from 2 to 75 years categorised (HE, OW, OB) by fitting a log-normal distribution to mean BMI and SE reported in population survey data	Risk of 14 obesity-related conditionsc using life tables, and data on RRs of disease incidence and mortality conditional on BMI	Intervention effect expressed as median difference in BMI between children receiving PE and historical control (not receiving PE)	Lifetime/NA	Third-party payer (health-care budget)
Tran et al., 201467	Canada	APPLE	Costs; OW and OB cases prevented	Applied parameters from growth curve models to OW and OB prevalence rates of Alberta (APPLE vs. no APPLE)	Health events not measured. Estimated by multiplying the total direct health-care cost for OB by the proportion of OW/OB cases prevented by the intervention	Intervention effect expressed as the prevalence of obesity among children attending APPLE schools vs. those not attending APPLE schools	Lifetime/NA	Health-care perspective
Gortmaker et al., 201569 (multiple publications: Wright et al., 2015,72 Sonneville et al., 2015,71 Long et al., 2015,70 Barrett et al., 201568)	USA	SSB; TV AD; ECE; active PE	Cost per BMI change; costs; DALYs;d QALYs;e bet savings per dollar spent on an interventionf	Logic models link the intervention to behaviour change or change in energy levels and thus change in BMI	Excess health-care costs associated with obesity among children and adults	Evidence review; studies linking the intervention to behavioural change or shift in energy balance	10 years or lifetime	Societal perspective

Summary of the literature identified on the modelling of the cost-effectiveness of interventions to address obesity in children

The results from the literature review indicated that seven studies (reported in 11 publications) used model-based economic evaluations to assess the cost-effectiveness of interventions aimed at reducing obesity in children. Most of the models identified in the review were from North America,62,63,67–72 with two European studies64,65 (one being a multisite study across six countries in Europe65) and one from New Zealand. 66 The studies varied in terms of the modelling approach used, the range of health states (including weight-related events) included in the models, the techniques used to predict weight status in adulthood as a function of childhood weight status and the outcomes included in the models. Nine of the studies64–72 were published in 2012 or later, and of note in the current report is that the initial searching of the literature in 2012 was used as the basis for the conceptual modelling framework for HeLP, as set out below, and it was, therefore, based on a limited literature (as described below and in Table 26). However, a number of the more recent studies adopt the methodological framework used in the earlier literature, which has influenced the conceptual framework used for the HeLP-specific research.

Before 2012 the evidence base for the assessment of the cost-effectiveness of interventions aimed at reducing obesity in children was small. Two models were identified. 62,63 Five models (reported in nine publications) were subsequently identified that were published between 2012 and 2015,64–67,69 with the majority of these published in the last 2 years.

The earlier models62,63 adopted a two-stage approach. The first stage used a decision tree structure and affixed a probability estimate that an individual (child) may change weight status between childhood and adult years. These predictions for adult weight status were based on longitudinal data from either Whitaker et al. 73 or NHANES I (National Health and Nutrition Examination Survey) (epidemiological study data). 74 The second stage of the model used a Markov-type approach, allocating the cohort in adult years to states by weight status (overweight, not overweight) and followed people up over time (through annual cycles) to compare a treatment versus control group in order to estimate the number of cases that were overweight in each group, and the medical care costs averted and number of QALYs saved per case of adulthood overweight prevented. Two models published in recent years65,66 also adopted a similar two-stage approach. In one model,65 the first stage takes a similar approach to that in Wang et al. 62 and Brown et al. 63 (i.e. the long-term extrapolation of the intervention effect on childhood weight status to adult age based on longitudinal data from Venn et al. 75). The later models65,66 adopted a different approach in the second stage. Rather than estimating the medical care costs averted per case of adulthood overweight prevented, both studies projected the incidence of chronic diseases associated with obesity and mortality in adulthood. A core set of weight-related events were included in the models: CHD, stroke, T2DM, breast cancer and colorectal cancer (CRC). The remaining three models64,67,69 used epidemiological models, statistical64,69 and prevalence profile67 approaches. These models present as potentially useful modelling frameworks for the assessment of interventions for childhood obesity, but at the present time the detail reported for these approaches was not sufficient to fully understand the specific detail and methods employed, and, importantly, the detail available did not allow us to replicate the methods used.

Model development (cost-effectiveness framework)

Introduction/context

The effectiveness outcomes in the definitive trial of HeLP versus usual practice (e.g. BMI SDS) do not provide a strong basis for the direct estimation and presentation of trial-based cost-effectiveness analysis, with decision-makers requiring estimates of cost per QALY and or cost per life-year saved (i.e. longer-term outcomes) to consider and to compare results of cost-effectiveness analyses across a wide range of health-care interventions. Therefore, a modelling framework was developed to assess the cost-effectiveness of the HeLP intervention versus usual practice, in a UK school/education setting, in preventing childhood overweight and obesity. The model development process was informed and guided by the need to extrapolate from effectiveness data from the HeLP RCT (at 24-month follow-up) over a longer-term and policy-relevant time horizon. That is, the modelling aimed to predict the expected impact of a change in childhood BMI SDS (primary outcome), and any resulting change in the distribution of children by weight status (intervention vs. control) over time.

At the outset the HeLP intervention was expected to be a relatively low-cost (per child) intervention, and the objective was to develop a simple model to estimate the cost-effectiveness of HeLP. At the time of developing the model, the literature was sparse and there was no readily available modelling framework publicly available. We sought to develop a simple decision-analytic model, building on the published literature in this area and using weight-related health events of primary importance (when there was robust evidence of the relationship between incidence of events and weight status, applicable to a general population analysis), to assess the cost-effectiveness of the HeLP intervention.

Model structure (summary)

The model development process is described in further detail below. In summary, the final model is a two-stage model (Figure 6). In stage 1 it predicts adult weight status (healthy weight, overweight and obese) from childhood weight status (BMI SD centile), and in stage 2 it predicts against a set of weight-related health events in adult years. A simplifying assumption is the use of only three weight status categories in adults: (1) healthy weight, (2) overweight and (3) obese. The use of three states, and the terminology used here (i.e. healthy weight), is consistent with the identified published model-based economic evaluations and also the available longitudinal data on the tracking of obesity from childhood to adult years. We acknowledge, and note here, that a small percentage of people will be underweight; however, for simplicity they are part of the group referred to as healthy weight and are treated in the same way as others in the healthy weight status category.

FIGURE 6.

Model overview. HW, healthy weight; OB, obese; OW, overweight; WS, weight status.

In stage 1 of the model, a decision tree structure is used to predict adult weight status as a function of childhood weight status. The model starts with a cohort of children distributed by weight status per the UK90 growth reference standards51 using clinical assessment classifications (≤ 91st centile; > 91st to ≤ 95th centile; > 95th to ≤ 98th centile; and > 98th centile), and predicts the expected distribution of the cohort by weight status in adulthood. Data from the UK longitudinal study reported by Power et al. ,76 which followed up participants from age 7 to age 33 years [see detail in Model development (stage 1): predicting adult weight status from child weight status], are used to predict adult weight status. Based on data reported by Power et al. ,76 the probability of children from a particular weight category remaining or changing weight status at the age of 33 years was estimated. These probabilities were applied to the starting cohort of children distributed by weight status in order to predict the likely distribution of people in adulthood, by weight status category. The impact of interventions, in this instance HeLP, can be considered/integrated in this first stage of the model via the specification of the starting distribution of children by weight status (e.g. an effective intervention will have fewer children in BMI SD centile categories associated with overweight and/or obese weight status). The model structure uses (starts with) inputs for a base cohort distribution of children by weight status categories and applies effectiveness data (e.g. hazard rates/relative risks) to proportions of children in categories of weight status associated with overweight and obesity in order to simulate the reported treatment effectiveness in a comparison of the same cohort of people both with and without the intervention (effectiveness).

In stage 2, a Markov model is used to model weight-related health events over time by adult weight status (Figure 7). Weight-related health events included in the model are T2DM, CHD, stroke and CRC [see detail in Model development (stage 2): predicting the impact of overweight and obesity on incidence of weight-related health events]. The model cycle length is 1 year and has a 30-year time horizon. The time horizon is chosen to reflect a longer-term adult time horizon, over which adult weight-related health events may be predicted (i.e. over age 33–62 years). Adults enter the model in one of three event-free health states depending on their weight status (healthy weight, overweight or obese). In the base-case model, adult overweight and obese weight status categories are BMI of 25.0–30.0 kg/m² and BMI of > 30.0 kg/m², respectively, with healthy weight being BMI of < 25 kg/m². Each year (1-year cycle) adults are at risk of remaining in the event-free health state, developing a weight-related event, or death. The model assumes that adults who experience an event will remain in that health state until death and are not exposed to any other events; this is a simplifying assumption, consistent with previously published model-based evaluations. People in the event-free health state are exposed to general population all-cause mortality risks. People in states defined by weight-related events are exposed to condition-specific mortality risks [see Model development (stage 2): predicting the impact of overweight and obesity on incidence of weight-related health events]. The costs and outcomes associated with weight-related events are estimated based on a review of the published literature; costs are presented as 2015 UK pounds sterling (£). Data on mortality are from national data sources (all-cause mortality) and from published literature on event-related mortality. Future costs and benefits are discounted at 3.5% per annum in the base-case analysis, with the option to vary this in sensitivity analyses, and the perspective of the analysis, consistent with the context of the decision problem, is that of the UK NHS and Personal Social Services (third-party payer perspective). The decision-analytic context is guided by the UK NHS decision-making context, using methods guidance for conduct of economic evaluations issued by NICE. 77

FIGURE 7.

Markov model structure. Costs by event/state: costCHD, costT2DM, costCRC and costStroke. Health state values (QALY weights) by event/state: hsvGenPop, hsvCHD, hsvT2DM, hsvCRC and hsvStroke. HW, healthy weight; OW, overweight.

Model development (stage 1): predicting adult weight status from child weight status

Longitudinal studies were used to inform the prediction of the proportion of children that will be healthy weight, overweight or obese in adulthood based on their childhood weight status.

A literature search was undertaken to identify systematic reviews that reviewed the literature concerning the tracking of childhood weight status into adulthood. A total of three systematic reviews were identified. 3,78,79 Owing to methodological differences between the reviews, the bibliographies of included studies for all three reviews were scrutinised for cohort studies reporting cohort data tracking childhood weight status into adulthood. A total of eight studies were identified. 75,76,80–85 Of these, four75,76,80,85 presented data projecting weight status from childhood into adulthood and included classifications by weight status (healthy weight, overweight), and were therefore selected for consideration to populate the model. The remaining four studies were excluded, either because they provided no usable data81,82,84 or because the data were for a child age range that was inappropriate for the current research. 83 The study characteristics for the included studies are presented in Table 27, and a summary description is provided below.

TABLE 27 - Study characteristics of the included weight status tracking studies

CDC, Centre for Disease Control; IOTF, International Obesity Task Force; NHANES, National Health and Nutrition Examination Survey; OW, overweight.

The centile indicates the relative position of the child’s BMI among children of the same sex and age. The growth charts show the weight status categories used with children and teens: underweight, healthy weight, overweight and obese.

Also included a very obese group defined as having a BMI of ≥ 35 kg/m².

Author, year	Country	Reference standard	Age in years (assessment time points)	BMI centile groupa	Weight status (BMI in kg/m2)
					Underweight	Healthy weight	Overweight	Obesea
Power et al., 199776	UK	UK90 Cole et al. 51	11–33	≤ 91st > 91st to ≤ 95th > 95th to ≤ 98th > 98th	< 20	20–24.9	25.0–30.0	> 30.0
Wright et al., 200180	UK	UK90 Cole et al. 51	11–50	< 25th 25–74th 75–90th > 90th	–	–	> 24	> 30
Laitinen et al., 200185	Finland	NHANES I and II Whitaker et al. 73 Frisancho86 (cited in Whitaker)	14–31	< 85th 85th to < 95th (obese) ≥ 95th (very obese)	< 18.5	18.5–24.9	25.0–29.9	≥ 30
Venn et al., 200775	Australia	IOTF Cole et al. 87	15–34	≥ 5th to 90th (normal) ≥ 90th to 97th (at risk OW) > 97th (OW)	Not defined as a separate category	< 25.0	25.0–29.9	≥ 30b

Power et al. 1997 (1958 UK birth cohort)

Power et al. 76 followed up a cohort of children who were born in Britain in March 1958 (the 1958 UK birth cohort). In this longitudinal study, height and weight measurements were taken at various points between the ages of 7 and 33 years. The weight status of children was defined by the UK90 growth reference standards. 51 At the end of the data collection, measurements for analysis were available for 11,212 adults. Adult BMI was categorised as healthy weight (20.0–24.9 kg/m²), overweight (25.0–30.0 kg/m²) and obese (> 30.0 kg/m²). From the study, the authors reported the weight status of an adult aged 33 years as a function of child weight status at the age of 11 years.

Wright et al. 2001

Wright et al. 80 reported data on a UK cohort of children who were born in Newcastle upon Tyne between May and June 1947. Height and weight measurements were taken at 9, 13 and 50 years. Weight status of children was defined by UK90 growth reference standards. The analyses were based on the proportion of adults who were overweight or obese at the age of 50 years as a result of weight status at 9 and 13 years, respectively.

Laitinen et al. 2001

Laitinen et al. 85 followed a cohort of children from birth year 1966 in northern Finland. Height and weight measurements were taken at birth, and at 1, 14, and 31 years of age. Data at 14 years were obtained from a questionnaire that was sent to participants, and follow-up data at 31 years (1997–8) were obtained from a questionnaire, a clinical examination or both. Adolescence (age 14 years) overweight was defined as a BMI ≥ 85th to < 95th percentile and obesity as a BMI ≥ 95th percentile. Adult BMI at 31 years was classified as follows: underweight (< 18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25.0–29.9 kg/m²) and obese (≥ 30.0 kg/m²). The analysis was restricted to individuals for whom BMI data were available for all measurement points (males, n = 2876; females, n = 3404).

Venn et al. 2007

Venn et al. 75 examined overweight and obesity in Australian children followed through to adulthood. A cohort study of 8498 children aged 7–15 years participated in the 1985 Australian Schools Health and Fitness Survey. Of these, 4571 completed a follow-up questionnaire at age 24–34 years in 2001–5. Height and weight were measured in 1985, and self-reported at follow-up. The accuracy of self-reported data was checked for 1185 participants. Overweight and obesity in childhood were defined according to international standard definitions for BMI, and, in adulthood, as a BMI of 25–29.9 kg/m² and ≥ 30 kg/m², respectively, after correcting for self-report error.

Summary

The longitudinal studies identified here presented data on tracking of weight status from childhood into adulthood. The results presented for each of the identified studies indicate that childhood overweight/obesity persists into adulthood, endorsing the broader evidence base which supports the use of childhood weight status as a strong predictor of adult weight status in a policy context (e.g. Whitaker et al. 73). There were methodological differences between the included studies, for example the number of participants included in the studies, the ages at which authors predicted weight status to and from, prevalence of overweight and obese across the studies, the thresholds used to define weight status as a child, the centile categories used to describe the distribution of weight status, and the growth reference population. All studies have limitations; however, these are considered to be high-quality longitudinal studies.

Data from each of these four longitudinal studies have been used to predict adult weight status by childhood weight status, represented in the form of transition probabilities across the defined child to adult weight status categories (Table 28). For stage 1 of the Exeter Obesity Model transition probabilities derived from the longitudinal data reported by Power et al. ,76 for the 1958 British birth cohort, are used in the base-case analysis (model structure), predicting the proportions of children (aged 11–12 years) expected to be healthy weight, overweight or obese at 33 years. These probabilities were applied to the distribution of children at the start of the model, expected to be informed by the 24-month follow-up data in the HeLP study, in order to predict the expected distribution of the cohort by adult weight status at the age of 33 years.

TABLE 28 - Predicted probabilities of adult weight status by child BMI centile category

HW, healthy weight; OB, obese; OW, overweight.

Data from Power et al. 76

Data from Wright et al. 80

Data from Laitinen et al. 85

Data from Venn et al. 75

BMI centile category	Weight status
	HW	OW	OB
Age 11 yearsa	Weight status at age 33 years
≤ 91st	0.59	0.32	0.08
> 91st to ≤ 95th	0.25	0.39	0.36
> 95th to ≤ 98th	0.24	0.37	0.39
> 98th	0.10	0.34	0.56
Age 13 yearsb	Weight status at age 50 years
< 25th	0.57	0.38	0.05
≥ 25th to ≤ 74th	0.32	0.54	0.14
> 74th to ≤ 90th	0.23	0.55	0.23
> 90th	0.13	0.42	0.45
Age 14 yearsc	Weight status at age 31 years
< 85th	0.69	0.27	0.04
≥ 85th to < 95th	0.28	0.48	0.23
≥ 95th	0.15	0.33	0.51
Age 11 yearsd	Weight status at age 29 years
≥ 5th to < 90th	0.62	0.29	0.09
≥ 90th to < 97th	0.21	0.37	0.42
≥ 97th	0.14	0.29	0.57

Data from Wright et al. 80 (British cohort data to age 50 years) and Laitinen et al. 85 (Finnish cohort data to age 31 years) were set out for potential use in sensitivity analysis to assess structural uncertainty (see Table 28). Although not specifically appropriate for sensitivity analyses, the estimate transit probabilities using data reported by Venn et al. 75 (Australian cohort data to age 29 years) are presented to provide a comparison between a relatively recent set of data and the data from other cohort studies76,80 that were collected in earlier years.

Model development (stage 2): predicting the impact of overweight and obesity on incidence of weight-related health events

Weight-related events

Stage 2 of the model starts with a cohort of people with a given distribution by weight status (healthy weight, overweight or obese), as described in stage 1. All persons start stage 2 as disease free (i.e. free of the major weight-related health events modelled in stage 2). Stage 2 uses CHD, stroke, T2DM and CRC as the weight-related health events of interest. The selection and inclusion of these four events is considered a conservative assumption (i.e. it is expected that the incidence of a greater number of health events will be influenced by weight status). The selection and inclusion of these four events is based on the review of the previously published model-based economic evaluations, discussed above, which typically included a small core set of weight-related health events similar to those selected here, and consideration of the data on incidence of events by weight status (see below), plus discussion of the conceptual modelling framework across the HeLP stakeholder group(s). Of importance when identifying health events for inclusion in the model was the simplifying assumption in the model that once a person experienced a condition they were then not exposed further to additional events. This led to a focus on the main primary health conditions, rather than events such as hypertension, which, although relatively frequent (and serious), was considered to be analogous to an intermediate outcome increasing the future risk of events such as CHD, stroke and T2DM.

Incidence (probability) of the weight-related health events was informed using a targeted keyword literature search (in MEDLINE) to identify studies that estimated the incidence of comorbidities related to obesity and overweight. The search strategy was structured using the following terms: (obesity OR adipose tissue OR body mass index OR body composition) AND (comorbidit$OR multimorbidit$OR chronic disease). No date limits or language restrictions were applied to the search. Eligible studies were systematic reviews of cohort studies, reporting risk estimates based on the incidence of weight-related events including T2DM, cardiovascular disease (including hypertension, coronary artery disease, congestive heart failure, pulmonary embolism, stroke and dislipidaemia) and CRC. The searches retrieved 126 titles and abstracts. Ten reviews were considered at full-text stage. Of these, only one was considered eligible for inclusion. 88

Guh et al. 88 conducted a literature search to provide an estimate of the incidence of each comorbidity related to obesity and overweight using a meta-analysis. Twenty comorbidities were initially included in this analysis: cancer (kidney, colorectal, prostate, ovarian, uterine/endometrial, esophageal, pancreatic and postmenopausal breast), T2DM, cardiovascular disease (hypertension, coronary artery disease, congestive heart failure, pulmonary embolism, stroke and dislipidaemia), gallbladder disease, chronic back pain, osteoarthritis, asthma and sleep apnoea.

For each of the weight-related events evaluated in Guh et al. ,88 we sourced the studies included in the review and extracted the data on number of incident cases and person-years from the primary papers contributing to the reported pooled relative risk estimate. We used these data to estimate the incidence of the events in the reference group (i.e. healthy weight), assuming a male-to-female gender ratio of 1 : 1. Using these raw data on incidence in the reference group, we then applied the risk ratios reported by Guh et al. 88 to estimate the incidence of the weight-related events in people with weight status at overweight and obese (Table 29).

TABLE 29 - Estimated incidence rate (probability) of adult weight-related events by weight status: cases per 1000 person-years

HE, healthy; OB, obese; OW, overweight.

Data source: Guh et al. 88

Weight-related event	Weight status
	HE	OW	OB
CRC	0.00087	0.00129	0.00159
T2DM	0.00165	0.00456	0.01334
Stroke	0.00141	0.00188	0.00234
CHD	0.00237	0.00329	0.00482

Mortality data

People are exposed to an annual mortality risk in each cycle of the model in stage 2. People in the disease-free state (no weight-related event) are subject to all-cause mortality risks based on age-specific data from the UK Office for National Statistics,89 using data reported for 2012–14, assuming a cohort of people with a gender ratio of 1 : 1.

Data on mortality risks by health condition, whereby people are in health states defined by the weight-related health events, were informed by a targeted literature search of the published literature. We searched the literature using terms for the population of interest (i.e. T2DM, CHD, stroke and CRC) with terms for mortality rate using the MEDLINE database in December 2015. Table 30 presents the data that we have used to populate the model on mortality risks for the health events. For stroke and CRC, we have used a mortality input for year 1 of the event, together with a separate mortality parameter for subsequent years in that condition/health state. This structure on mortality data is consistent with the findings from the literature search.

TABLE 30 - Mortality rates by condition of interest

Health state	Annual mortality rate	Source
CHD	0.170	Sigurdsson et al., 199590
Diabetes (T2DM)	0.056	Roper et al., 200291
Stroke: year 1	0.370	Hankey et al., 200092
Stroke: subsequent years	0.100	Hankey et al., 200092
CRC: year 1	0.243	Cancer Research UK, 201293
CRC: subsequent years	0.080	Cancer Research UK, 201293

Health state values (QALY weights)

Data used in the model on health state values to derive QALYs are reported in Table 31. The health state value used for the ‘event free’ state (0.86) is based on UK EuroQol-5 Dimensions (EQ-5D) data, consistent with methods guidance from NICE,61 and is the general population mean health state value reported by Kind et al. 94 Data inputs for other model states were based on a literature review to identify published systematic reviews reporting health state values (utilities) associated with the health states (events, conditions) used in the model. The search method involved targeted keyword searching in MEDLINE (December 2015) for CHD, T2DM, stroke and CRC. Terms for health state values (‘health state value$’ and ‘EQ-5D’ and ‘EQ5D’) were combined with terms for the specified conditions of interest. A filter was applied to the searches to identify systematic reviews (review or ‘systematic review’ or ‘literature review’ or meta$). In addition, we used the data reported in the model-based economic evaluations for obesity identified earlier (see Table 26). In considering the appropriateness of the identified health state values, the following attributes were taken into account: the population describing the health state (e.g. age, sex, disease severity), the utility value elicitation technique (e.g. time trade-off, standard gamble, visual analogue score), the sample size and the country. Further detail on the literature identified is presented below. The data used in the model for health state values are broadly in line with published estimates used in other public health modelling contexts (e.g. see studies reported in Table 27).

TABLE 31 - Health state values

NR, not reported.

Health state	Parameter value, mean (SD)	Source
Event-free	0.86 (0.23)	Kind et al., 199994
T2DM	0.78 (0.77)	Clarke et al., 200295
CHD	0.67 (0.34)	Matza et al., 201596
Stroke	0.52 (0.38)	Matza et al., 201596
CRC	0.76 (NR)	Djalalov et al., 201497

Summary of literature identified, and selection of health state values as model inputs

Type 2 diabetes mellitus

One systematic review was identified98 that identified a set of health state/utility values for one or more complications in people with diabetes. The methodology of identified studies was considered against the requirements of the UK NICE reference case (2013) (i.e. criteria included use of EQ-5D questionnaire value set or other established measure of preference, health states should be determined by patients, and valuation performed using a societal valuation algorithm [preferences of general public]). The authors created a preferred input set of utility values. Of 16,574 records identified, they report on 61 studies, of which we considered 19 to be consistent with the NICE criteria.

The studies included in the Beaudet et al. 98 review were scrutinised and baseline T2DM utility values were extracted. For the base case for the current Exeter Obesity Model, the utility value of 0.78 reported in the study by Clarke et al. 95 was considered to be the most methodologically robust, given its large sample size, T2DM-specific population and use of the EQ-5D in a UK population. This estimate is also broadly consistent with other published estimates for T2DM without complications (e.g. NICE99).

Coronary heart disease

One systematic review was identified100 that reviewed utility data based on EQ-5D in cardiovascular disease. Sixty articles were identified that reported EQ-5D scores (health state values). There was wide variation in terms of cardiovascular disease subgroups, disease stage, age distribution and other methodological aspects. The majority of studies reported EQ-5D index scores using the UK-based algorithm. The majority of identified studies reported data for ischaemic heart disease and cerebrovascular disease. Stratification by disease severity was possible for ischaemic heart disease and heart failure participants.

Supplementary searches in PubMed identified one additional study96 reporting health state values for cardiovascular conditions, also distinguishing between acute impact including the cardiovascular event and the chronic post-event impact. Three cardiovascular conditions were described: stroke, acute coronary syndrome and heart failure. One-year acute health states represented the event and its immediate impact, and post-event health states represented chronic impact. Values were from a sample (n = 200) of the UK general population. The mean time trade-off health state value for acute coronary syndrome was 0.67 (SD 0.34) and for chronic post event health states was 0.82 (0.17).

For the base case in the Exeter Obesity Model, the time trade-off value of 0.67 (SD 0.34) reported by Matza et al. 96 was considered the most reflective of the health state in the model. This is in line with estimates used in other models of this type. 101–103

Stroke

One systematic review was identified104 that reviewed evidence on health state values for stroke. Mean utilities of major stroke (Rankin scale 4–5) and minor stroke (Rankin scale 2–3) were presented, stratified by study design and elicitation method. A total of 23 studies were included in the review. Study populations in the included studies were healthy people, people at risk of stroke or stroke survivors. The studies included in the review reporting data from general public (non-patient, healthy participants) were considered most relevant to the health state used in the Exeter Obesity Model. 105–108 However, two of these studies105,108 used only the visual analogue scale to elicit utilities, and the remaining studies used mixed samples and had relatively small sample sizes. The supplementary searches carried out to identify utility values for CHD identified one paper96 that also reported health state values for stroke. In that study, a number of health state values are reported, including a mean (SD) value of 0.52 (0.38) for chronic stroke. We also reviewed previous health technology appraisals and clinical guidelines reported by NICE for stroke, identifying data presented for stroke in the appraisal of apixaban for the prevention of stroke and systemic embolism in people with non-valvular atrial fibrillation. 103,109

Considering the varied evidence available, we use the data reported by Matza et al. 96 in the base case for the model presented here. This estimate is consistent with health state values used for stroke in prior public health modelling contexts, such as exercise referral schemes. 102,110

Colorectal cancer

One systematic review was identified97 that reviewed evidence on utility weights for CRC health states. Twenty-six articles met the inclusion criteria for the review. The authors report a mean EQ-5D health state value of 0.76 (based on 33 reported utilities), and this is used in the current model framework. This estimate is broadly consistent with data used to inform the NICE technology appraisal of drugs CRC,111 aligned to first-line and second-line treatment health states. 111

Cost data

As stated previously, the perspective applied for the economic analyses is that of the UK NHS and Personal Social Services, and cost estimates included for model inputs are consistent with that perspective (i.e. third-party payer).

Intervention costs (HeLP)

The additional cost included for the delivery of the HeLP intervention, in the base-case analysis, is the estimated mean per participant cost of £214.14 (see Table 22).

Costs for disease states (weight-related events)

An annual estimate of costs associated with model states by condition (weight-related event) is applied in the model when people are in these states. The estimates of annual costs used are presented in Table 32. The estimates used are informed by a literature search to identify published systematic reviews of the literature in each of the disease areas.

TABLE 32 - Treatment costs per health state

2010/11 costs.

Reflects the cost of a rapid recovery case, applied as a one-off treatment cost (in year 1 of stroke).

Health state/condition	Parameter value/costs (£)	Source
T2DM	3717a per year	Kanavos et al., 2012112
CHD	2852 per year	NHS Scotland, 2007113
Stroke	2800b	Saka et al., 2009114
CRC	8441 per year	Cancer Research UK, 2014115

Targeted literature searches in MEDLINE and web searches combining terms for the population of interest (i.e. T2DM, CHD, stroke and CRC) with terms for costs (e.g. cost, annual cost, per patient cost, UK cost) were conducted in December 2015. Cost estimates used in the model were inflated/uprated to 2014/15 prices, when appropriate, using inflation indices from Curtis and Burns. 57

Type 2 diabetes mellitus

Kanavos et al. 112 report costs for diabetes (direct and indirect) across France, Germany, Italy, Spain and the UK. Estimates include inpatient, outpatient, primary and community costs, and medication-related costs. Annual UK costs for T2DM are estimated at £3717 mean per-person costs. These costs are based on high-quality data from large retrospective cohort studies for primary care and outpatient activity116 and inpatient costs. 117

Coronary heart disease

A report from NHS Scotland113 presents estimates of the resource use and costs associated with cardiovascular disease-related diagnoses. The cost per patient for CHD was estimated as £2852. This estimate was based on a comprehensive costing methodology, assessing the average cost per patient for the following principal diagnoses: unstable angina, £1760; acute myocardial infarction, £3990; subsequent myocardial infarction, £4240; chronic ischaemic heart disease unspecified, £2765; and stable angina, £1500.

Stroke

Saka et al. 114 reported detailed data on the resource use and cost associated with stroke, estimating societal and direct costs of care. Direct care costs include diagnosis, inpatient care and outpatient care and are based on data from a population-based register. The data are predominantly presented as aggregate burden/cost of illness results; however, when considering an estimate of cost per case/per stroke patient (applying assumptions/methods from Bosanquet and Franks118), the authors present estimates of £2800 and £17,500 for costs associated with a rapid recovery case and a case with disability but discharged into the community. As a conservative assumption here we use the cost estimate of £2800 for the treatment of stroke events arising in the model presented here. This cost is applied as a one-off treatment cost (model payoff), rather than an annual treatment cost.

Colorectal cancer

A report from Cancer Research UK115 presents mean estimated costs for CRC of £8441, covering diagnosis and treatment costs, across disease stages 1–4. Cost estimates are based on a mapping of national treatment guidelines, national data sets and clinical audit data. This cost estimate is similar to the data presented by Trueman et al. 119 in a report commissioned by the UK Department of Health (mean cost approximately £8800).

Modelling summary

The model structure has been described above, and the parameter inputs to populate the model have been presented together with a description of methods and rationale for data inputs and assumptions used. Appendix 9 presents a summary of the model parameters and inputs used.

The Exeter Obesity Model has been developed as part of the research funded alongside the NIHR-funded clinical trial (RCT) on HeLP versus usual practice. The model is set out here as a parsimonious modelling framework appropriate for estimating cost-effectiveness for HeLP versus usual practice, given the expectation of a relatively low-cost intervention being used in a public health context, with benefits accruing over the longer term (i.e. in adult years, 20–50 years after the intervention is delivered). The specific focus of the model development has been the use of key weight-related events as outcomes of interest in the model, the use of published data to populate the model, and the presentation of cost-effectiveness analyses using cost per QALY, and cost per life-year saved, in order to inform health policy and decision-making in a UK context. The conceptual model and the simple model design are based on the need to answer the a priori research question on whether or not HeLP is cost-effective versus usual practice, that is, whether it represents value for money to a UK third-party payer (such as the NHS) when assessed against commonly used estimates of willingness to pay for health benefits (i.e. cost-per-QALY thresholds). 77

However, as referred to above, the HeLP intervention is not effective compared with usual practice, and therefore the modelling framework is presented here as a potentially useful generic contribution to the methods available to model the effectiveness and cost-effectiveness of interventions for childhood obesity, where it is important to extrapolate from relatively short-term effectiveness outcomes to longer-term impacts of interventions, into and over adult years.

Cost-effectiveness analyses (HeLP vs. usual practice)

The above sections have presented results of the estimates of the costs associated with delivery of the HeLP intervention, and have described the modelling framework set out to estimate the cost-effectiveness of the HeLP intervention versus usual practice.

Effectiveness of HeLP versus usual practice

Data presented in Chapter 3 have reported that there is no evidence that HeLP is more effective than usual practice. Table 9 reports no statistically significant differences between HeLP participants and controls at 24-month follow up on the primary outcome measure of BMI SDS. The Exeter Obesity Model estimates cost-effectiveness with a starting premise that an intervention has shown a difference in effectiveness that can be translated to a reduction in the relative risk associated with being in BMI SDS weight status categories associated with overweight and/or obesity, using the BMI centile categories applied in the base data drawn from Power et al. 120 (see Table 28). This model structure allows a comparison of a cohort of children over time, with and without the intervention. The results from the HeLP RCT do not show reductions in the relative risk for being in states aligned to overweight and/or obesity, finding no statistically significant difference between the intervention group and the control group (see Table 11).

Cost-effectiveness of HeLP versus usual practice

Given the effectiveness profile for the HeLP intervention, there is no expectation, using the Exeter Obesity Model framework, that there will be any improvements in the likely incidence of the weight-related health events (CHD, stroke, T2DM and CRC), or cost savings to the health and social care system associated with weight-related events. This profile, together with a certain expectation that the introduction of the HeLP intervention is associated with additional resource use and costs, in the short term, leads to the clear conclusion that HeLP is not considered to be a cost-effective alternative to usual practice.

HeLP incurs additional costs with no identified or expected health gains, using the analytical framework developed and presented here, and therefore usual practice ‘dominates’ HeLP in a resource allocation context (Table 33). HeLP does not offer value for money to the UK health-care system, compared with usual practice.

TABLE 33 - Summary of differences in outcomes and costs (HeLP treatment vs. control)

NA, not applicable.

See Table 9.

See Table 7.

See Table 11.

See Table 14.

See Table 22.

Outcome	Group		Adjusted difference (intervention vs. control)
	Intervention (HeLP)	Control (usual practice)
Mean BMI SDS (SD)a (n = 650)	0.35 (1.25) (n = 630)	0.22 (1.22) (n = 620)	–0.02, 95% CI –0.082 to 0.045; p = 0.568 (control = lower); no statistically significant difference
Proportion (%) overweight at 24-month follow-upb (baseline)c	14.1 (12.1)	13.5 (10.7)	NA
Proportion obese at 24-month follow-upb (baseline)c	16.7 (14.6)	13.1 (12.6)	NA
Proportion (%) overweight or obese at 24-month follow-upb (baseline)c	30.8 (26.8)	26.6 (23.3)	OR (adjusted analyses) 1.15, 95% CI 0.79 to 1.67b (Control lower odds than intervention) No statistically significant difference
Physical activity outcomes, 18-month follow-up,d average daily total mean (SD)	199.71 (43.9) (n = 359)	198.05 (40.2) (n = 386)	1.26, 95% CI –6.84 to 9.36; no statistically significant difference
Mean weekly FIQ (18-month follow-up)d
Energy-dense snacks, mean number (SD)	3.72 (1.86)	4.06 (2.07)	–0.37, 95% CI –0.066 to –0.07; statistically significant improvement with intervention
Healthy snacks, mean number (SD)	3.61 (1.63)	3.30 (1.5)	0.22, 95% CI –0.04 to 0.47; no statistically significant difference
Negative food markers, mean number (SD)	5.90 (2.73)	6.38 (3.0)	–0.47, 95% CI –0.91 to –0.02; statistically significant improvement with intervention
Positive food markers, mean number (SD)	6.20 (2.36)	5.77 (2.31)	0.26, –0.12 to 0.64; no statistically significant difference
Additional treatment costs (mean, per participant)e	£214.13	None	£214.13; intervention more costly than control

As this is a clear and unambiguous outcome, consistent with agreement with the TSC, no further detailed cost-effectiveness analysis is presented here. We have presented detailed cost analysis and results on the estimated expected mean incremental cost per child for delivery of the HeLP intervention, together with estimates of expected additional costs per school (per class).

Exploratory cost-effectiveness analyses using the Exeter Obesity Model

Predicted results for control participants

In the HeLP RCT the distribution of children starting the trial and at 24-month follow-up by weight status categories, as centile categories used by Power et al. ,120 are reported in Table 34, together with their predicted weight status profile as adults aged 33 years (using standard adult BMI categories).

TABLE 34 - Summary of weight status profile by group at childhood and by adult predicted profile

NA, not applicable.

Source: HeLP RCT, intention-to-treat participants.

Predictions using data from Power et al. 120

Weight status category	Group
	HeLP		Control (usual practice)
	Baseline	24 months	Baseline	24 months
BMI centile category (Power et al.120)a
≤ 91st	80.42%	76.51%	82.27%	80.32%
> 91st to ≤ 95th	4.93%	6.83%	5.13%	6.61%
> 95th to ≤ 98th	6.28%	7.46%	6.38%	4.84%
> 98th	8.37%	9.21%	6.22%	8.23%
Adult weight statusb
Healthy weight	NA	49.87%	NA	51.37%
Overweight	NA	33.31%	NA	33.14%
Obese	NA	16.82%	NA	15.48%

When modelling the experiences of a cohort (n = 1000) of children defined using the weight status data for the control participants in the HeLP study, we predict that at age 33 years the cohort will result in an adult cohort with 49% overweight or obese adults and 51% healthy weight adults. We predict that 33% will be overweight and 15.5% will be obese, and this is consistent with the data reported by age in the Health Survey for England (2009 data). 121 Over the 30-year follow-up period, adult-years, there are 252 weight-related events (CHD, stroke, T2DM and CRC). These 252 events comprise 76 cases of CHD, 105 cases T2DM, 28 cases of CRC and 42 cases of stroke; 675 participants remain event free at age 62–63 years. The mean per participant cost associated with those events is estimated at £1140 (£6821 undiscounted; £3145 discounted at 1.5% per annum). Over the 30-year time horizon (30 adult-years) where events, costs and event related outcomes are estimated, in addition to all-cause mortality, the mean number of life-years per participant is 26.75, prior to discounting (4.47 when discounted at 3.5% per annum; 12.34 at 1.5% per annum), and the mean number of QALYs is 22.78 prior to discounting (3.8 when discounted at 3.5% per annum; 10.5 at 1.5% per annum). Of note, overall we are modelling life expectancy over a 50-year time horizon, with specific payoffs (future costs and outcomes for health events) captured in the latter 30 years.

The modelling framework developed to assess the cost-effectiveness of interventions such as HeLP is dependent on identifying a beneficial (effective) intervention, with effectiveness reflected through an improved weight status profile in childhood. That mechanism of effect provides an opportunity to consider the future potential impact of a change in childhood weight status, into adult years, and the estimation of that effect over adult years through a reduction in future incidence of a core set of weight-related health events. When applying the data from the HeLP exploratory trial (crude unadjusted relative risks, by BMI SDS category), the modelling framework indicates that HeLP has the potential to be a cost-effective intervention (see scenario analyses E in Table 35). This is primarily due to the reduced number of participants in the HeLP exploratory trial intervention arm, compared with the number in the control arm, who ended up in the childhood weight status category associated with obesity. However, the sample size in the exploratory trial was very small, and the distributions by weights status in the intervention and control arms were not balanced at baseline, given that the study design was for an exploratory trial, and feasibility research.

TABLE 35 - Scenario analyses: controls (HeLP RCT) vs. hypothetical comparison scenarios for effectiveness

HW, healthy weight; LYG, life-year gained; OB, obese; OW, overweight; RR, relative risk; WS, weight status.

Events comprising CHD, stroke, T2DM and CRC.

Cost per life-year and QALY at base-case discount rate of 3.5% per annum, with figure in brackets the estimated cost per life-year and QALY when using a discount rate of 1.5% per annum.

Lloyd et al. 28

Note

Baseline/starting distribution (standardised to cohort, n = 1000): 82.3% BMI SDS centile ≤ 91st, 5.1% BMI SDS centile > 91st to ≤ 95th, 6.3% BMI SDS centile > 95th to ≤ 98th, 6.2% BMI SDS centile > 98th.

	BMI SD centile category (%)				Adult WS distributions (%)			Eventsa (n)	Mean costs (£)	Mean life-years	Mean QALYs	Cost per LYG (£)a	Cost per QALY (£)b
	≤ 91st	> 91st to ≤ 95th	> 95th to ≤ 98th	> 98th	HW	OW	OB
Control, 24 months	80.3	6.6	4.8	8.2	51.4	33.1	15.5	252	1140	4.47	3.81	N/A	N/A
A: vs. control, 25% reduction in OB (RR 0.75)	82.3	6.6	4.8	6.2	52.4	33.1	14.5	249	1335	4.476	3.812	44,723 (13,540)	46,654 (14,124)
B: vs. control 50% reduction in OB (RR 0.50)	84.4	6.6	4.8	4.1	53.4	33.1	13.5	246	1318	4.480	3.816	20,264 (4673)	21,139 (4875)
C: 25% reduction in OB and OW (RRs 0.75)	83.6	6.6	3.6	6.2	52.83	33.04	14.13	248	1328	4.478	3.813	30,850 (8513)	32,183 (8881)
D: 50% reduction in OW and OB (RRs 0.50)	86.9	6.6	2.4	4.1	54.29	32.93	12.78	243	1303	4.484	3.819	13,330 (2162)	13,906 (2255)
E: Applying data from HeLP exploratory RCTc (crude RRs)	87.6	5.2	4.7	2.6	54.7	32.9	12.4	242	1295	4.486	3.821	10,996 (1302)	11,439 (1361)

Here, to illustrate the use of the modelling framework, and to illustrate the potential scenarios in which an intervention such as HeLP may be considered cost-effective, we present a series of scenario analyses in Table 35 describing alternative weight status profiles for a potential hypothetical treatment group, compared with the data reported for the HeLP control participants (standardised to a cohort of n = 1000).

In these exploratory analyses we have used the inputs on expected additional costs for the HeLP intervention, and the data on the weight status of the control participants in HeLP at 24-month follow-up, to explore the cost-effectiveness of the HeLP intervention at hypothetical scenarios on intervention effectiveness [i.e. relative risk(s) of being overweight and/or obese]. We use hypothetical estimates of the relative risk, between controls and intervention, and we present data on predicted distribution by adult weight status, and summary data on weight related events, cost per life-year saved and cost per QALY gained.

The exploratory results demonstrate that when the relative risks are ≥ 0.75 and above (≤ 25% reduction in proportion in BMI SDS centile category) the cost per QALY is above the commonly referred to UK NHS willingness to pay for a QALY, of between £20,000 and £30,000 per QALY. 61 However, of note is that the number of children in the BMI SD centiles associated with overweight and obesity is relatively small (i.e. circa 6% in each category), which therefore requires a relatively high risk reduction in order to make a meaningful impact on the actual numbers of children, and thereafter adults, in the overweight and/or obesity related weight status categories (e.g. a relative risk of 0.75 for childhood obesity represents a difference of approximately 20 cases per 1000 children). We therefore suggest that relatively modest effectiveness results may lead to a scenario in which an intervention is cost-effective, regardless of the apparent magnitude of the estimated relative risk (for small proportions of the cohort). We advise that the exploratory results presented here should not be interpreted in a way that indicates that interventions may be cost-effective only when dramatic effects are seen. The exploratory results also illustrate that when considering the impacts of such a public health intervention over time, the mean incremental costs and outcomes are very small, and the cost per life-year gained, and cost per QALY gained, estimates are very sensitive to small changes in the mean incremental inputs to the cost-effectiveness ratios. Furthermore, the exploratory results highlight that the results are sensitive to the parameter used to discount future costs and outcomes, with cost-effectiveness estimates markedly different (more attractive) when using a discount rate of 1.5% per annum (vs. the base case 3.5% per annum).

Discussion

We have estimated the additional cost for delivery of the HeLP intervention to be £214 per participant, based on delivery to a cohort of schools (classes) similar to that in the HeLP study. The estimated cost is based on good-quality data collected within the trial, describing the staff inputs and resource use required to deliver the intervention, alongside realistic estimates of the costs associated with the resource inputs. The drama component, integral to the HeLP intervention, is the primary area of resource use and cost, and, together with the costs associated with the HeLP co-ordinator role, accounts for over three-quarters of the cost for delivery of HeLP. There are economies of scale when delivering HeLP in larger schools, given that some components are delivered once at the school level, with costs spread over a larger number of children receiving the intervention. We would anticipate that interventions such as HeLP would be commissioned (funded) across a cohort of schools, and in that setting we report an estimated mean cost per class of approximately £5350. This is a significant additional cost per class, at the level of the school budget, but we have shown in the illustrative cost-effectiveness analyses that a relatively small reduction in the predicted number of adverse weight-related health events (e.g. T2DM and CHD) in future adult years demonstrates that this investment represents value for money using a health and social care sector perspective.

However, in the trial we found no evidence that the HeLP intervention was effective in terms of a difference in mean BMI SDS and in preventing overweight or obesity at 24-month follow-up. Given the scenario of additional costs, together with an absence of demonstrated benefits in weight status, the result of the cost-effectiveness analyses is unambiguous, with usual care dominating the HeLP intervention.

As part of the research process, we have described the development of a framework for assessing the cost-effectiveness of the HeLP intervention (or similar interventions). We have described the Exeter Obesity Model, which builds on the published evidence base for methods surrounding the model-based economic evaluation of interventions for childhood obesity, in a public health context. When seeking to estimate the cost-effectiveness of interventions such as HeLP, there is no readily available decision-analytic modelling framework to utilise. We have therefore set down such a framework and described it in detail to support any future research in a similar context. The modelling framework is simple, using good-quality longitudinal data, and limiting its scope to a small number of health events where there is robust evidence that the incidence of disease has a strong association with weight status, and where there is a published meta-analysis reporting relative risks by weight status categories. However, the modelling framework makes a number of simplifying assumptions, which are deserving of further sensitivity analyses and research. We have not presented a full detailed set of cost-effectiveness results owing to the effectiveness scenario for the HeLP intervention. We had anticipated presenting cost-effectiveness analyses (results) when addressing structural uncertainty, and applying alternative estimates of the predictions of adult weight status as a function of childhood weight status (BMI SD centiles) (i.e. alternatives to the data use in the base-case analysis120). Such sensitivity analyses have not been undertaken, but we do report the estimated transition probabilities for alternative sources of data to inform future modelling developments. Based on our preliminary modelling, when using data from Wright et al. 80 or Laitinen et al. 85 (see Table 27) we would expect the outcome for estimates of cost-effectiveness (i.e. binary outcome where intervention is either considered cost-effective or not) to be similar to that when using the base case data from Power et al. 120

Modelling the cost-effectiveness of public health interventions typically results in the use of a long-term time horizon, often over 30–40 years into the future, and given the expectation that interventions are population based with only a small number of identifiable beneficiaries (in terms of hard outcomes), the mean incremental differences in costs and outcomes are relatively small (averaged across a large target group for the intervention). This leads to difficulties in articulating the expected benefits at a population level, and to difficulties associated with sensitivity in cost- effectiveness ratios (e.g. cost per QALY estimates) to very small changes to the estimated incremental costs and effects. We see these scenarios played out here in the exploratory cost-effectiveness analyses and cost-per-life-year and cost-per-QALY estimates. For example, in scenario B (see Table 35), we report a gain of 10 life-years (approximately 6.5 QALYs) per 1000 children exposed to the HeLP intervention, owing to the prevention of approximately six adverse health events (including preventing five cases of T2DM). At a geographical region, such as in Devon, we would see this multiplied using the target population of 9- to 10-year-olds, approximately 10,000 children, and benefits would be in the region of 100 life-years saved per year. This estimate is adjusted using a discount rate, but an unadjusted (non discounted) estimate would be in the region of 520 life-years saved per 10,000 children receiving an intervention (or 240 life-years saved, using a discount rate of 1.5% per year). These illustrations are speculative and hypothetical, and scenario B assumes an effectiveness scenario with a relative risk of 0.5 for obesity, so we must urge caution here when we seek to illustrate a potential analytical framework (although the HeLP exploratory RCT suggested an effect that was much higher than that in this scenario).

A related characteristic of the modelling framework, although more specific to estimates of effectiveness, is that the magnitude of the relative risk estimates is likely to be quite sensitive to changes in weight status for relatively small numbers of children. That is, given the small numbers of children considered obese in the starting control cohort (e.g. approximately 60–80 per 1000 children), and the large numbers in other weight status categories, the relative risk statistic can be in the region of 0.5 when there are differences of only 20 children in differing directions of weight status (when comparing two similarly distributed cohorts). The exploratory data presented in Table 35 give some indication of this prevailing characteristic of the effectiveness inputs (albeit in a hypothetical setting), although further consideration is recommended.

Specific challenges in modelling the impacts of childhood obesity interventions, using the framework suggested here, include the profiles for the incidence of health events by weight status. The differences in the distributions by weight status are small in both child and adult populations, with only around one-fifth of the population likely to be overweight or obese. This, combined with the data available on the difference in rates of events by weight status, presents an analytical challenge. For example, the incidence of T2DM and CHD is approximately 13.3 cases and 4.8 cases, respectively, per 1000 person-years in obese states, compared with estimates of 1.7 cases and 2.4 cases per 1000 person-years in healthy weight status. Of note here is the 8.5-fold risk of T2DM for obese versus healthy weight status, compared with the twofold risk of CHD for obese versus healthy weight status. We see this reflected in the modelling results, with the vast majority of the differences in predicted events due to cases of T2DM, with very little difference in the incidence of the other events when comparing a potential treatment cohort with controls. Such results provide a rationale for modelling only those health events that exhibit a potentially meaningful difference in incidence by weight status (i.e. meaningful in a mathematical modelling context).

The model presented here has simplifying assumptions, which in many instances are due to the parsimonious structure of the modelling framework, although the absence of data in some areas is also a mitigating factor. We suggest that the model is likely to be sufficient to answer the research question on the cost-effectiveness of an intervention versus usual practice (or other), when the intervention is shown to be effective on weight status outcomes, but we do not propose this modelling framework for more detailed examination of the consequences of overweight and obesity.

The HeLP intervention has been shown to have positive outcomes in areas of secondary importance, and the overall feedback and satisfaction with HeLP from the school setting is very positive. That there is no specific policy-relevant evidence to indicate it is cost-effective, when considering weight status, points commissioners to the dominance of usual practice. However, given that there is a wide range of other interventions being used across the school setting, in the absence of evidence of their effectiveness and cost-effectiveness, it may be that commissioners consider the potential use of HeLP, or HeLP-like interventions, in a school-based setting, and if so here we have provided strong evidence on the resource use and cost associated with delivery of HeLP, and a framework that may be used in any further future assessment of HeLP or similar interventions.

Introduction

The overarching aim of the process evaluation for the cluster RCT of HeLP was to contextualise the trial effectiveness and cost-effectiveness results with respect to how the intervention may or may not have worked.

Research questions

To address the aims of the process evaluation, we devised the following research questions.

Aim 1: to assess uptake and fidelity of the HeLP intervention:

1. How much of HeLP did children and families receive?

2. Was the programme delivered as designed?

Aim 2: to assess whether or not the intervention worked in the way it was expected to in terms of the intervention logic model (Figure 8):

1. Did schools, children and parents engage with HeLP?

2. How were the attempts to change behaviours experienced by the children?

3. Were there reported changes to children’s eating and physical activity behaviours?

4. Do statistical models combining potential cognitive and behavioural changes mediate observed between group differences in outcomes?

FIGURE 8.

The logic model for HeLP.

Section 1: process data collected from the intervention arm of the trial

Section 2: mediation analyses

Methods

Sampling and recruitment

Data were collected from all participating children in both intervention and control schools (see Chapter 2) completing the MLQ at baseline and at 12 months post baseline (i.e. immediately after the end of the intervention).

Data collection

The HeLP co-ordinators collected the MLQ data during a dedicated lesson and read the questions from the front of the class, with the children completing the questionnaire at the same time and in silence. Clarifications were given for specific questions, and a poster was produced to demonstrate percentages so that the children were able to understand a question relating to proportions (question 4 in the knowledge section; see Appendix 3). Children who required additional support completed the questionnaire in a smaller group outside the classroom with an additional researcher.

Analysis

Factor analyses to create composite My Lifestyle Questionnaire variables

An exploratory factor analysis (EFA) was conducted on half of the baseline MLQ data set, followed by a confirmatory factor analysis (CFA) (in the remaining data set) to generate composite variables that loaded together and could be taken into the subsequent mediation analyses. We adopted a parsimonious approach to the construction of these variables, summarising across theoretically derived and HeLP-specific MLQ items (see Appendix 18).

Table 43 shows the five statistically generated composite variables from the factor analyses, the MLQ constructs that they came from, the questions that fed into the constructs and the corresponding score ranges. The variable referred to as ‘knowledge’ was originally designed as the sum of five questions, with each question having several examples for the children to answer. We have called the four composite variables derived from the factor analysis ‘confidence and motivation’, ‘peer norms’, ‘family approval/behaviours and child attitudes’ and ‘behaviours and strategies’.

TABLE 43 - Derivation of MLQ composite variables

Composite variable (score range)	MLQ construct	Number of items	Question number in the MLQ	Minimum–maximum score
Knowledge (0–20)	Healthy snack/drink alternatives	1	1	0–6
	Food group proportions	1	2	0–5
	Lifestyle physical activity	1	3	0–4
	Energy balance	1	4	0–2
	Strategies for change	1	5	0–3
Confidence and motivation (9–36)	Self-efficacy to make healthy eating and activity choices	3	6–8	3–12
	Intentions to make healthy eating and activity choices	6	9–14	6–24
Peer norms (8–32)	Peer norms	3	15–17	3–12
	Peer approval	5	18–22	5–20
Family approval/behaviours and child attitudes (9–36)	Family approval	3	23–25	3–12
	Attitudes towards restrictions on behaviour	3	26–28	3–12
	Parental provision and rules around food and physical activity	3	29,30,32	3–12
Behaviours and strategies (18–72)	Goal setting	6	33–38	6–24
	Self-monitoring	4	39–42	4–16
	Discussion about healthy lifestyles with parents/family	3	43–45	3–12
	Encouraging the family to be more healthy	2	46–47	2–8
	Helping parents to choose healthy alternatives when shopping	1	48	1–4
	Helping with cooking at home	1	49	1–4
	Trying new, healthy food and drinks	1	50	1–4

Structural equation modelling

Structural equation modelling was used for the mediation analyses. SEM allows the assessment of the relationships among a set of variables using multi-item scales, multiple variables and multiple outcomes. 129,130 The variables included in the mediational analyses are baseline and 12-month BMI SDS, gender, school IMD score, class size and baseline and 18-month energy-dense snacks and negative food markers, as well as the five composite variables from the MLQ (see Table 43). The independent variable is the allocation group: intervention versus control.

There are two dependent variables: the number of weekday energy-dense snacks consumed per day at 18 months and the number of weekday unhealthy foods consumed per day (negative food markers) at 18 months. These were the secondary outcome variables from the main trial that had statistically significant between-group differences. We took a parsimonious approach to the modelling exercise and hence did not select the two other statistically significant secondary outcomes (the average number of energy-dense snacks and negative food markers foods consumed across the whole week at 18 months). Furthermore, we did not select healthy snacks at the weekend as the adjusted between-group difference from the linear random-effects regression model was not statistically significant (mean difference 0.23, 95% CI –0.04 to 0.50) [see Chapter 3, Primary analyses of secondary outcomes at 24 months (anthropometric measures only)].

Eleven variables were controlled at baseline. These included baseline scores for the outcome variables and the potential composite mediating variables, as well as four demographic measures – gender, BMI SDS, school size and school-level deprivation – that could have influenced (enhanced or diminished) scores on the dependent outcomes.

Variables in the path model to be tested using SEM are shown in Figure 9.

FIGURE 9.

Pathway analysis conceptual model. B&S, behaviours and strategies; C&M, confidence and motivation; EDS, energy-dense snacks; FAB&CA, family approval/behaviours and child attitudes; NFM, negative food markers; PN, peer norms; SES, socioeconomic status.

Preparation of the data included examination of, and possible replacement of, any missing values. For the ‘knowledge’ items, if a child did not answer an item it was assumed that they did not know the answer and were scored zero for that item. For the composite variables that contained Likert scales (see Table 43), the following rules were used to replace missing values: if up to 75–80% of items were available for a particular variable, this was averaged, and the rounded average was used to replace the 20–25% of missing values, thereby retaining the ordinal nature of the individual items.

The model shown in Figure 9 was tested twice: once for the outcome variable weekday energy-dense snacks at 18 months and once for weekday negative food markers at 18 months. According to our proposed model, the intervention may have a direct effect on the outcome variable at 18 months but also indirect effects through changes in the five composite mediating variables at 12 months.

Structural equation modelling127 was used to test this model, specifically whether the mediating variables function as full or partial mediators between the intervention and the outcome variables. Evidence for full mediation would be that a previously significant effect of the intervention on the outcome variable is no longer present when the composite mediating variables have been added to the model. On the other hand, if a significant association still remains between the intervention and the outcome variables after adding the composite mediating variables, this would indicate partial mediation, at best.

Software and type of factor analysis and goodness-of-fit indices used

Structural equation modelling was conducted with the software package AMOS version 23 (IBM Corporation, Armonk, NY, USA) using maximum likelihood estimation. As chi-squared is known to be inflated with sample size,131 it was not used as an indicator of goodness of fit (and misfit means that there is major deviation between what the model specifies and what exists in the actual data). Instead, a set of indicators was used to provide a fair assessment of goodness of fit. These indicators are known to be sensitive to different aspects of model misfits. The following criteria were used:

• root-mean-square error of approximation (RMSEA) of < 0.06132

• comparative fit index (CFI) of > 0.95132

• standardised root-mean-square residual (SRMR) of < 0.08132

• relative/normed chi-squared [χ²/degrees of freedom (df)] of < 5 (although this criterion is commonly used and well known, it is equally sensitive to sample size133).

Results

Of the 1324 children, 105 had missing data that did not meet the rules for replacing missing values. A conservative approach was taken by discarding these participants, which reduced the sample size from 1324 to 1219.

A descriptive summary of data for the five composite variables at baseline and at 12 months is provided in Tables 44 and 45. None of the variables demonstrated elevated levels of kurtosis and skewness. Cronbach’s alpha was good for all of the composite variables except peer norms, which approached the consensual cut-off value of 0.70 for acceptable internal consistency.

TABLE 44 - Descriptive summary of the five composite variables at baseline

NA, not applicable.

Variable (number of included items)	Score range	Mean	SD	Kurtosis	Skewness	Inter-item correlation range (mean)	Cronbach’s alpha
Knowledge (5)	1–19	8.47	2.69	–0.11	0.09	NA	NA
Confidence and motivation (9)	9–36	23.95	6.38	–0.56	–0.28	0.18–0.59 (0.37)	0.84
Peer norms (9)	5–32	19.63	4.49	–0.22	–0.12	0.07–0.48 (0.22)	0.69
Family approval/behaviours and child attitudes (8)	9–36	26.72	5.92	–0.19	–0.56	0.17–0.55 (0.33)	0.82
Behaviours and strategies (18)	3–72	41.19	12.79	–0.54	0.22	0.18–0.61 (0.40)	0.92

TABLE 45 - Descriptive summary of the five composite variables at 12 months

NA, not applicable.

Variable (number of included items)	Score range	Mean	SD	Kurtosis	Skewness	Inter-item correlation range (mean)	Cronbach’s alpha
Knowledge (5)	2–20	10.49	3.18	–0.26	–0.05	NA	NA
Confidence and motivation (9)	9–36	24.21	5.79	–0.49	–0.29	0.21–0.65 (0.35)	0.83
Peer norms (9)	7–30	19.55	4.25	–0.32	–0.23	0.00–0.55 (0.21)	0.70
Family approval/behaviours and child attitudes (8)	9–36	27.22	5.80	0.04	–0.70	0.17–0.66 (0.35)	0.83
Behaviours and strategies (18)	18–72	42.33	11.78	–0.55	0.16	0.18–0.68 (0.39)	0.92

Tables 46 and 47 show the correlation coefficients for the five mediating variables at baseline and at 12 months, respectively. Knowledge had very low correlations, and in one case no correlation, with the other variables. The correlations among the confidence and motivation, peer norms, family approval/behaviours and child attitudes and behaviours and strategies variables were small to medium.

TABLE 46 - Pearson correlation coefficients of the composite variables at baseline

*p < 0.05, **p < 0.01.

Variable	Knowledge	Confidence and motivation	Peer norms	Family approval/behaviours and child attitudes
Confidence and motivation	0.06*	–
Peer norms	0.09**	0.37**	–
Family approval/behaviours and child attitudes	0.22**	0.49**	0.41**	–
Behaviours and strategies	–0.02	0.61**	0.34**	0.40**

TABLE 47 - Pearson correlation coefficients of the composite variables at 12 months

*p < 0.05, **p < 0.01.

Variable	Knowledge	Confidence and motivation	Peer norms	Family approval/behaviours and child attitudes
Confidence and motivation	0.16**	–
Peer norms	0.15**	0.36**	–
Family approval/behaviours and child attitudes	0.22**	0.53**	0.38**	–
Behaviours and strategies	0.10**	0.69**	0.33**	0.51**

Preliminary analyses indicated that the conditions for mediation analysis126 had been met: (1) the intervention had a significant effect on each of the five 12-month composite variables, namely knowledge [t(1217) = 15.27; p < 0.01], confidence and motivation [t(1217) = 4.00; p < 0.01], peer norms [t(1217) = 4.01; p < 0.01], family approval/behaviours and child attitudes [t(1217) = 3.40; p < 0.01] and behaviours and strategies [t(1217) = 5.66; p < 0.01]; and (2) the composite variable scores at 12 months were significantly (p < 0.01) correlated with the two outcome variables: weekday energy-dense snacks and negative food markers (Table 48). As the outcome variable negative food markers at 18 months as well as the negative food markers score at baseline revealed slightly elevated kurtosis (1.86 and 2.18, respectively), the scores were logarithmically transformed before the SEM was conducted.

TABLE 48 - Summary of Pearson correlation coefficient between the outcome variables energy-dense snacks and negative food makers with the mediating variables at 12 months

**p < 0.01.

Variable	Energy-dense snacks	Negative food markers
Knowledge	–0.16**	–0.16**
Confidence and motivation	–0.19**	–0.20**
Peer norms	–0.12**	–0.10**
Family approval/behaviours and child attitudes	–0.28**	–0.27**
Behaviours and strategies	–0.13**	–0.12**

Our longitudinal path analysis contained data from three time points: (1) at baseline, including the demographic variables gender, baseline BMI SDS, school size and school-level deprivation, with direct paths to the outcome variable as well as the baseline scores of the outcome and demographic variables; (2) the 12-month mediating variables; and (3) the 18-month outcome data. The baseline score of the outcome variable (energy-dense snack or negative food marker) as well as the child’s allocated group (independent variable) also had direct paths to the outcome variable. The model shows that baseline composite variables predicted scores on these same variables at 12 months, which in turn predicted the outcome variable at 18 months. The intervention versus control variable also predicted the scores on the five composite variables at 12 months. Several of the composite variables were correlated at baseline (guided by modification indices of > 10). The error variances of the mediating variables at 12 months were correlated in an identical pattern to those pairs of mediating variables at baseline.

Table 49 shows the goodness-of-fit indices for the two longitudinal path analysis models (with energy-dense snacks and negative food markers as outcome variables). Overall, the model fits were acceptable for both outcome variables, and although the CFI results were slightly below the 0.95 criterion for an excellent fit, the other indices met the criteria.

TABLE 49 - Goodness-of-fit indices for the longitudinal path analysis models with energy-dense snacks and negative food markers at 18 months as outcome variables

Goodness-of-fit indices	Energy-dense snacks	Negative food markers
Satorra–Bentler scaled χ2/df	4.303	4.336
RMSEA	0.052	0.052
CFI	0.943	0.940
SRMR	0.057	0.061

Path analysis for weekday energy-dense snacks (18 months)

Figure 10 presents a diagram of the path analysis for energy-dense snacks. Not shown are the correlations of the variables at baseline, error covariances and non-significant regression weights (dashed lines). The following composite variables were significantly (p < 0.05) correlated at baseline: knowledge with family approval/behaviours and child attitudes and confidence and motivation; family approval/behaviours and child attitudes with confidence and motivation, peer norms and behaviours and strategies; confidence and motivation with peer norms and behaviours and strategies; and peer norms with behaviours and strategies. All paths shown in the diagram were significant (p < 0.05), except for the associations between gender and energy-dense snacks (18 months), between school-level deprivation and energy-dense snacks (18 months), between peer norms and energy-dense snacks (18 months) and between behaviours and strategies and energy-dense snacks (18 months). The direct effect of the intervention on energy-dense snacks (18 months) was significant (p = 0.041), indicating partial mediation by knowledge, family approval/behaviours and child attitudes and confidence and motivation. Of these three mediating variables, the effect of family approval/behaviours and child attitudes was the strongest, with a standardised regression weight of –0.18, followed by confidence and motivation (β = –0.09) and knowledge (β = –0.07).

FIGURE 10.

Partial mediation model with significant standardised regression weights of pathways for weekday energy-dense snacks at 18 months. B&S, behaviours and strategies; C&M, confidence and motivation; EDS, energy-dense snacks; FAB&CA, family approval/behaviours and child attitudes; NFM, negative food markers; PN, peer norms; SES, socioeconomic status. *p < 0.05, **p < 0.01, ***p < 0.001.

Path analysis for weekday negative food markers (18 months)

Figure 11 shows the full mediation model with significant standardised regression weights of pathways for weekday negative food markers at 18 months. Not shown are the correlations of the variables at baseline, error covariances and non-significant regression weights (dashed lines). The results for negative food markers were generally similar, with the same variables correlated at baseline and at 12 months. Similar to the results for energy-dense snacks, there were no significant associations between gender and negative food markers (18 months), between school-level deprivation and negative food markers (18 months) and between peer norms and negative food markers (18 months). However, the association between behaviours and strategies and negative food markers (18 months) was significant (p < 0.05), and there was no significant association between school size and negative food markers (18 months). Additionally, the direct effect of the intervention on negative food markers (18 months) was not significant (p = 0.059) indicating full mediation by knowledge, family approval/behaviours and child attitudes, confidence and motivation and behaviours and strategies. Of these four variables, the effect of family approval/behaviours and child attitudes was the strongest, with a standardised regression weight of –0.18, followed by confidence and motivation (β = –0.10), behaviours and strategies (β = 0.08) and knowledge (β = –0.06).

FIGURE 11.

Full mediation model with significant standardised regression weights of pathways for weekday negative food markers at 18 months. B&S, behaviours and strategies; C&M, confidence and motivation; EDS, energy-dense snacks; FAB&CA, Family approval/behaviours and child attitudes; NFM, negative food markers; PN, peer norms; SES, socioeconomic status.*p < 0.05, **p < 0.01, ***p < 0.001.

Overall, the effects of the composite mediating variables on the outcome variables were fairly small. The largest association was between family approval/behaviours and child attitudes at 12 months and the outcome variables. For both paths (between family approval/behaviours and child attitudes and energy-dense snacks and between family approval/behaviours and child attitudes and negative food markers), β was –0.18, which means that an increase of 1 SD of the family approval/behaviours and child attitudes variable was associated with a reduction of 0.18 in energy-dense snacks and negative food markers. Additionally, the results of the behaviours and strategies variable need to be interpreted with caution. As outlined in Appendix 18, this variable was the least psychometrically robust of MLQ variables. The path between behaviours and strategies at 12 months and energy-dense snacks was not significant, but the path between behaviours and strategies at 12 months and negative food markers resulted in a significant change in the opposite direction to those of the other mediating variables. This counterintuitive result is most likely a result of collinearity; as shown in Table 48, behaviours and strategies was negatively correlated with negative food markers. However, in the context of the other MLQ variables, this association became positive (β was 0.08; see Figure 11). Such seemingly paradoxical cases have been described in the literature134 and are most likely due to collinearity with other predictor variables or the operation of suppressor variables. Table 47 shows that behaviours and strategies and confidence and motivation were correlated to an extent (r = 0.69) that could question the statistical independence of the variables and further multiple-linear regression analyses confirmed that the coefficient of behaviours and strategies does change direction when confidence and motivation is added as a second predictor. Further exploration of these composite variables could clarify which items are responsible for these effects.

Conclusions from the process evaluation

Data from the process evaluation show that HeLP was delivered as designed in all 16 intervention schools, with high uptake and engagement from schools, children and their families across the socioeconomic spectrum. Qualitative data suggest that HeLP worked in accordance with the underpinning IMB model and the logic model by changing children’s self-reported levels of knowledge, cognitions and behaviours as intended; however, these changes were insufficient to affect physical activity behaviours at 18 months or BMI SDS at either 18 or 24 months post baseline. The mediation analyses show that the intervention effects on the consumption of weekday energy-dense snacks were mediated by knowledge and two composite variables, namely family approval/behaviours and child attitudes and confidence and motivation, whereas the intervention effect on weekday consumption of unhealthy foods (negative food markers) was mediated by the same variables as well as the composite variable behaviours and strategies.

Summary of findings

We evaluated the effectiveness and cost-effectiveness of HeLP, a novel school-based obesity prevention programme, in a cluster RCT involving 32 schools and > 1300 children. We found no evidence of a difference in BMI SDS at 24 months or that participating in HeLP reduced the likelihood that children would be overweight or obese compared to children not receiving the intervention. Similarly, no differences between the intervention and control groups were observed in either anthropometric measures or physical activity objectively assessed using accelerometers at 18 months post baseline. Self-reported weekly average consumption of different types of energy-dense snacks was lower in those attending intervention schools (0.4 mean difference in score), as was the number of different types of negative food markers (0.5 mean difference), small differences that were statistically significant. These differences were largely accounted for by reported differences in weekday consumption.

The cost of implementing HeLP was estimated at approximately £210 per child. Assumptions are reported regarding the proportions of children needing to move weight category for cost-effectiveness to be achieved using NICE cost-per-QALY methodology.

Comparison with other studies

Wang et al. 4 is the most recent systematic review and meta-analysis of childhood obesity prevention programmes targeting both diet and physical activity. The review identified 139 intervention studies that had weight-related outcomes, of which 115 were located in the primary school. The 37 studies that were purely school-based and did not have a family component showed a low strength of evidence for reducing BMI, BMI SDS, prevalence of obesity and overweight, percentage body fat, waist circumference and skinfold thickness. However, studies that also included a family component provided moderate evidence of effectiveness, with half reporting statistically significant beneficial intervention effects. Other systematic reviews and meta-analyses also suggest that school-based obesity prevention interventions can have a modest effect on BMI SDS and it is unclear whether such effect sizes (typically < 0.1) are clinically meaningful at either a population or an individual level. 7,136

We are not aware of any recent, well-conducted, school-based obesity prevention RCTs, using objective outcome measures, for this age group, that have shown a clinically relevant effect on adiposity measures at 2-year follow-up. A very recent school-based trial (Active for Life-year 5)16 involving 60 schools and > 2000 children (aged 9–10 years), which aimed to increase physical activity, reduce sedentary behaviour and increase fruit and vegetable consumption at 2-year follow-up, found no effect of the intervention on any of these primary outcomes or on weight status.

The exploratory trial of HeLP demonstrated the feasibility and acceptability of the programme, trial processes and provided evidence of ‘proof of concept’ for HeLP. Furthermore, the exploratory trial showed changes in diet and physical activity behaviours and weight status; however, these were not replicated in the main trial. 28 No changes were made to the programme or trial design between the exploratory trial and the definitive RCT, with the manuals and standard operating procedures, developed in the exploratory trial, used for delivery and assessment in the main trial. In addition, HeLP was delivered as designed in all intervention schools with very high levels of engagement, as was also seen in the exploratory trial. The findings from the process evaluation allow us to be confident that the difference in results between the exploratory and the definitive trial are not due to scale-up issues of delivery. The only difference in trial design was that we had to run the main trial in two cohorts; however, there was no evidence of a differential effect of the intervention between the two cohorts on the primary outcome, indicating that this logistical requirement did not affect the overall findings, and our follow-up rates at 18 and 24 months were similar across both trials.

Understanding the lack of effectiveness

Conducting health promotion interventions within schools has the obvious potential advantage of being able to reach virtually all children. The behaviours that underlie the development of obesity and overweight in children and adolescents result from a complex interaction of individual, family and social factors. This is particularly relevant for children of primary school age, as their ability to influence their diet and activity is directly limited by decisions made by their parents/carers, as well as being affected by wider social influences. We therefore aimed to develop an intervention that would influence not only the children themselves, but also their parents and the school environment, as we felt that this would have a higher likelihood of being effective.

The design of the intervention followed the MRC’s guidance for the development and evaluation of complex interventions19 and was based on a review of research evidence and on the use of intervention mapping18 to draw on evidence regarding potentially effective methods of achieving behaviour change. Our review of existing evidence suggested that multifaceted interventions were more likely to be effective when they addressed both diet and exercise, were of significant duration and involved the family, although the strength of these conclusions was limited owing to the paucity of existing high-quality studies. We were also aware that a common reason for failure in health promotion programmes is a failure to persuade the target group to participate and to stay involved, so strategies to achieve engagement by children, parents and schools were fundamental in the design of both the intervention and the trial. In addition, for public health interventions to have an impact, they need to be deliverable without disrupting normal activities and at an affordable cost. In particular, in the context of crowded school curricula, interventions need to reduce rather than increase teachers’ workloads and, ideally, help to achieve educational targets if schools are to embrace them. We therefore worked closely with children from the target age group, parents, teachers and education advisors at all stages of the design.

We assessed the extent to which children, parents and teachers actually engaged with the programme using prespecified criteria for engagement, as well as conducting focus groups with children and interviews with parents and teachers. The results suggested very high levels of engagement across all socioeconomic groups and considerable enthusiasm for the programme. We succeeded in achieving extremely high levels of participation: only 34 out of 1371 children opted out of the trial and only 80 out of 1324 who started the study were lost to follow-up, and hence > 94% of children provided anthropometric data at 2 years and we have accelerometry data at 18 months post baseline on 84% of children (3 weekdays and 1 weekend day).

The lack of any effect of the intervention on our primary outcome measure is particularly disappointing given the high levels of engagement and that the programme was developed with substantial stakeholder involvement and reflected the current best evidence regarding techniques to change behaviour. There is a number of potential explanations: we recognise the importance of wider family and social factors in driving health behaviours, which may limit the potential effects of interventions that are delivered primarily at the level of the individual. We are also aware that although the increase in overweight among children is perceived by policy-makers as constituting a major threat to health, it is less clear whether or not parents share this view. It has been repeatedly reported that a large proportion of parents of overweight children perceive their children as being of normal weight. 137 Stakeholder consultation in the development of HeLP clearly showed that focusing on weight as a health issue would not be acceptable to head teachers, teachers or parents, and thus the programme was careful to avoid discussing obesity or weight in any context. Furthermore, we believe that our very high levels of engagement support the programme’s focus on healthy lifestyle behaviours for all. However, it may be the case that the HeLP messages regarding diet and physical activity behaviours were not really seen by parents as referring to a problem that required significant and sustained behaviour change.

We chose to target children aged 9–10 years for the intervention on the basis of our early pilot work. Broadly speaking, our initial development work showed that younger children were keen to engage but did not appear to successfully absorb the messages such that they engaged their parents with them, while older children were less easy to engage. However, it is the case that, at this age, the ability of children to actually influence their own diet and activity may be limited, as, inevitably, most decisions will be made by their parents. Parental involvement in obesity prevention programmes has been frequently cited as one of the key characteristics associated with behaviour change, and throughout the programme there were activities specifically aimed at engaging parents across the socioeconomic spectrum.

Our data showed that > 50% of parents attended at least one parental engagement event, and three-quarters either attended an event or signed support for their child on the goal-setting sheet. Although it is possible to engage children (and assess their levels of engagement) in school-based programmes, the lack of direct contact with parents means that parental engagement is more difficult to achieve and assess. Many parents reported (in questionnaire responses and interview) that they had made changes at a family level and gave examples of how they were supporting their child, but this was only a subset of all of the parents involved. Given the children’s comments about the need for parental support, it is likely that the programme did not enable parents or families to make the types of sustained behaviour changes necessary to affect the weight status of their child. We consider it unlikely that school-based obesity prevention programmes will ever be of sufficient intensity to affect the family environment, and more direct approaches would be required.

We were mindful from the outset that the programme should complement educational activities (by meeting national curriculum objectives) and not be a burden to teachers. It is possible that, although the external delivery of the majority of programme activities (by sport and dance groups, actors and the HeLP co-ordinators) was welcomed by teachers and children, this delivery mode meant that the programme was probably not embedded across the school and hence could not systematically affect the school environment.

Our development and early piloting work suggested that a more intensive intervention with greater parental involvement in the programme would be less acceptable or unfeasible for schools, suggesting that those planning future obesity prevention interventions may need to consider new approaches and settings if they are seeking to affect family behaviours and the home environment.

Trial strengths and limitations

There are well-recognised potential sources of bias associated with cluster RCTs,33 including recruitment bias, performance bias and detection bias. We tried to reduce the likelihood of these biases by recruiting schools and children and collecting baseline measures before the known allocation of schools to intervention or control (to reduce differential take-up); by capturing evidence of changes in school policies around food or physical activity through school completion of a checklist at the beginning of the trial and at 18 months (see Appendix 16); and by using assessors blinded to allocation to measure the anthropometric outcomes. The primary outcome for the trial was based on measures taken when the children had transferred to secondary school, and hence secondary schools included children from both the intervention and the control groups. No child revealed their group allocation to the blinded assessor at 24-month follow-up. However, it should be noted that although the MLQ and FIQ were completed before the group allocation at baseline and at 12 and 18 months, respectively, the children were aware of their group allocation, as shown in Figure 1.

Sample size calculations, which allowed for the anticipated level of clustering as estimated from the exploratory trial and NCMP data,28,138 suggested that we needed measures from approximately 760 children at 24 months to detect a 0.25 difference in BMI SDS. We estimated that we needed 28 schools, each with approximately 35–40 children, to complete the study and so, allowing for a 20% loss to follow-up at 24 months, we sought to recruit 32 schools to allow for both school-level and child-level dropouts. No schools dropped out of the study. Few eligible children (34/1371) opted out of the study and we achieved follow-up rates of between 84% and 96% for all outcome measures, and hence there is a very low risk of attrition bias.

Reviews and recent studies of school-based obesity prevention and management trials in children have highlighted low participation, differential dropout and loss to follow-up rates. 4,5,9,139 For example, completeness of anthropometric data in school-based obesity prevention programmes has ranged between 70% and 80% for follow-up of ≥ 24 months13,140,141 and the percentage of children providing a representative pattern of their physical activity levels across the entire week (at least 3 weekdays and 1 weekend day of 10 hours’ wear time) tends to be much lower (between 40% and 60%). 38 In the HeLP trial, 84% of children met the minimum wear time criteria of 3 weekdays and 1 weekend day, and 79% provided data on all 7 days at 18 months. The HeLP trial, therefore, has one of the most complete follow-up rates and physical activity compliance of recent obesity prevention trials for children in this age group. We attribute this to the relational approach that we took in the trial delivery, as well as to the stakeholder involvement we had in developing the intervention and the trial design. 27

We weighted our recruitment of schools to achieve a similar proportion of pupils receiving free school meals as the national average in 2012 (19%), which is higher than the average for Devon (12.7%), and the schools in the HeLP trial were larger than the average Devon primary school. In other respects, the schools that participated in the trial are representative of Devon, and the anthropometric data from the children are broadly similar to the Devon NCMP Year 6 data. 142 This gives us confidence that the results are mostly applicable to a wider population. The one exception is that, although the included schools reflected the proportion of children with English as an additional language typical for Devon (2.6%), this is substantially lower than the national figure of 16.8% of pupils with English as an additional language in England.

A further strength of this trial was its robust process evaluation involving qualitative and quantitative data and its prespecified logic model and mediating variables questionnaire to understand whether or not HeLP was working in the way it was intended, with the qualitative data being analysed prior to the outcome result being known. We also attempted to assess engagement (school, child and parent); whereas school and child engagement was assessed using different methods (observation of intervention sessions and response to the programme), parent engagement was assessed as attending one of six possible parent events and/or providing a signature on the child’s goal-setting sheet. We acknowledge that this is a weak assessment of engagement and does not capture engagement with the messages of the programme. The low response rate (25%) from parents to the questionnaire about HeLP, which formed part of the process evaluation, limits the strength of these data and also suggests that the sample interviewed (who were recruited from responders to the questionnaire) is unlikely to be representative of all parents. This suggests that the qualitative data from parents cannot be considered as representative of all parents receiving the programme. However, the children recruited for the process evaluation were representative of the overall sample, and all Year 5 teachers were interviewed.

Research recommendations

The disappointing results of this study and other recent well-designed, UK-based trials suggest to us that it is unlikely that affordable, school-based interventions that are inherently time-limited can achieve clinically meaningful change in weight status/anthropometric outcomes in a single targeted age group. This view is reinforced by other research using data from the NCMP that suggested that there was no consistent school effect on childhood obesity. 143 Future studies need to develop interventions that aim to specifically affect the family environment, perhaps employing a more widely co-ordinated whole-systems approach, although it is unlikely that such an approach would be feasible to deliver within a school setting.

We recognise that one reason for the lack of effectiveness of this intervention may be that, despite engaging and affecting knowledge and motivation, children of this age have limited ability to affect their own diet and levels of physical activity. This suggests that researchers should consider whether or not other age groups offer greater possibilities for affecting behaviour. It is possible that older children, who have a greater degree of autonomy, if engaged, may be more able to make changes in their behaviour. It has also been suggested that pregnant women and parents of very young children may be more amenable to adopting health messages. Examples include successful campaigns to promote smoking cessation, the avoidance of soft cheeses during pregnancy and the ‘Back to Sleep’ campaign. Programmes addressing this group (e.g. the Henry Programme: www.henry.org.uk/homepage/) have claimed impact, but rigorous evaluations are lacking and are urgently needed.

Conclusions

This cluster randomised trial provides rigorous evidence about the effects of HeLP, a novel school-based obesity prevention programme aimed at 9- to 10-year-old children, in a sample of children broadly representative of the UK population. The programme achieved high levels of engagement with children, schools and parents, but did not affect their BMI or their likelihood of being overweight or obese 2 years later compared with children attending schools in which the intervention was not delivered.

Trial registration and ethics approval

The trial was registered with the International Standard Randomised Controlled Trial Number 15811706. Ethics approval for this study was given by the Peninsula College of Medicine and Dentistry Research Ethics Committee (reference number 12/03/140) in March 2012.

Funding

This project was funded by the NIHR Public Health Research programme (project number 10/3010/01). The trial was conducted with the support of the Peninsula Clinical Trials Unit, a UK Clinical Research Collaboration-registered clinical trials unit (registration number 31) in receipt of NIHR Clinical Trials Unit support funding.

Contributions of authors

Katrina Wyatt (Associate Professor of Health Research, University of Exeter Medical School) led the study, co-designed the HeLP intervention and co-led the process evaluation. She was involved in all stages of the HeLP trial, including conception, design, interpretation of data, drafting and critical revision of the report for important intellectual content and approval of the final version, and co-ordinated responses from other authors.

Jenny Lloyd (Senior Research Fellow Child Health, University of Exeter Medical School) co-led the project, co-designed the HeLP intervention, led the process evaluation and managed the delivery of the trial. She was involved in all stages of the HeLP trial, including conception, design, interpretation of data, drafting and critical revision of the report for important intellectual content and approval of the final version.

Siobhan Creanor [Associate Professor (Reader) in Clinical Trials & Medical Statistics, Plymouth University Peninsula Schools of Medicine and Dentistry] co-led the project, led the statistical analyses and was involved in all stages of the HeLP trial, including conception, design, interpretation of data, drafting and critical revision of the report for important intellectual content and approving of the final version.

Colin Green (Professor of Health Economics, University of Exeter Medical School) co-led the project, led the economic evaluation and was involved in all stages of the HeLP trial, including conception, design, interpretation of data, drafting and critical revision of the report for important intellectual content and approval of the final version.

Sarah G Dean (Associate Professor of Psychology Applied to Rehabilitation and Health, University of Exeter Medical School) co-led the project, led the psychometric evaluation of the MLQ and was involved in all stages of the HeLP trial, including conception, design, interpretation of data, drafting and critical revision of the report for important intellectual content and approval of the final version.

Melvyn Hillsdon (Associate Professor of Sport and Health Sciences, University of Exeter) co-led the project, led the physical activity analyses and was involved in the interpretation of data, drafting and critical revision of the report for important intellectual content and approval of the final version.

Charles Abraham (Professor of Psychology Applied to Health, University of Exeter Medical School) co-led the project, and advised on the theoretical design of the trial and evaluation and of the psychological and behavioural measures of change. He supported the drafting and critical revision of the report for important intellectual content and approval of the final version.

Richard Tomlinson (Consultant Paediatrician, Royal Devon and Exeter NHS Trust) co-led the project and provided input and understanding regarding child health from a NHS perspective. He supported the drafting and critical revision of the report for important intellectual content and approval of the final version.

Virginia Pearson (Director of Public Health Devon) co-led the project, provided input into understanding the public health context and policy implications, and supported the recruitment of schools and the attainment of school data.

Rod S Taylor (Professor of Health Services Research, University of Exeter Medical School) co-led the project, provided input into the statistical analysis plan and was involved in all stages of the HeLP trial, including conception, design, interpretation of data, drafting and critical revision of the report for important intellectual content and approval of the final version.

Emma Ryan (Special Education Support Assistant, ISCA Academy) co-led the project, supported the wider stakeholder involvement, supported the recruitment of schools, children and HeLP co-ordinators and provided input into understanding the school context.

Adam Streeter (Research Fellow in Medical Statistics, Plymouth University Peninsula Schools of Medicine and Dentistry) undertook the majority of the statistical programming and analyses and contributed to the mediational analysis. He co-wrote Chapters 2 and 3 and supported the drafting and critical revision of the report for important intellectual content and approval of the final version.

Camilla McHugh (Associate Research Fellow Child Health, University of Exeter Medical School) was a HeLP co-ordinator for the trial. She contributed to the collection and analysis of the qualitative data for the process evaluation, as well as supporting the writing of the process evaluation section of the monograph.

Alison Hurst (Data Manager, Associate Research Fellow Child Health, University of Exeter Medical School) was the data manager for the trial and contributed to the analysis of the process evaluation data for the monograph.

Lisa Price (Lecturer in Sport and Health Sciences, University of Exeter) was responsible for the collection, analysis and interpretation of the physical activity data. She was also involved in baseline data collection for anthropometric and questionnaire data, alongside data entry.

Louise Crathorne (Honorary Research Associate Health Economics, University of Exeter Medical School) contributed to the economic analysis and co-wrote Chapter 4 of the report.

Chris Krägeloh (Associate Professor Health Sciences, Auckland University of Technology) co-led the psychometric evaluation of the MLQ and carried out the meditational analyses.

Richard Siegert (Professor of Psychology and Rehabilitation, Auckland University of Technology) co-led the psychometric evaluation of the MLQ and carried out the meditational analyses.

Stuart Logan (Director of Institute for Health Research, University of Exeter Medical School) co-led the project, co-designed the intervention and was involved in all stages of the HeLP trial, including conception, design, interpretation of data, drafting and critical revision of the report for important intellectual content and approval of the final version.

All authors made critical revisions to the monograph.

Publications

Wyatt KM, Lloyd JJ, Abraham C, Creanor S, Dean S, Densham E, et al. The Healthy Lifestyles Programme (HeLP), a novel school-based intervention to prevent obesity in school children: study protocol for a randomised controlled trial. BMC Trials 2013;14:95.

Lloyd JJ, Wyatt KM. Qualitative findings from an exploratory trial of the Healthy Lifestyles Programme (HeLP) and their implications for the process evaluation in the definitive trial. BMC Public Health 2014;14:578.

Lloyd J, Wyatt K. The Healthy Lifestyles Programme (HeLP) – an overview of and recommendations arising from the conceptualisation and development of an innovative approach to promoting healthy lifestyles for children and their families. IntJ Environ Res Public Health 2015;12:1003–19.

Lloyd J, Wyatt K. Uptake, retention and engagement of children participating in the cluster randomised controlled trial of the Healthy Lifestyles Programmes (HeLP). Educ Health 2015;33:88–95.

Creanor S, Lloyd J, Hillsdon M, Dean S, Green C, Taylor RS, et al. Detailed statistical analysis plan for a cluster randomised controlled trial of the Healthy Lifestyles Programme (HeLP), a novel school-based intervention to prevent obesity in school children. Trials 2016;17:599.

Lloyd J, Creanor S, Logan S, Green C, Dean SG, Hillsdon M, et al. Effectiveness of the Healthy Lifestyles Programme (HeLP) to prevent obesity in UK primary-school children: a cluster randomised controlled trial [published online ahead of print November 28 2017]. Lancet Child Adolesc Health 2017.

Data sharing statement

We are keen for these data to be used widely by the scientific community. Details of the study can be found on the website (http://medicine.exeter.ac.uk/help) and anyone interested in using these data should contact the corresponding author in the first instance. We do not have funds for data extraction and putting data files together for collaborators and so may have to charge for providing this service.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the PHR programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the PHR programme or the Department of Health.

Food Intake Questionnaire: weekday

(PDF download)

Food Intake Questionnaire: weekend

(PDF download)

Food Intake Questionnaire scoring

My Lifestyle Questionnaire

(PDF download)

My Lifestyle Questionnaire scoring

(PDF download)

Estimate training costs: HeLP co-ordinators

Based on within-trial experience on training requirements, it is assumed that training requirements for HeLP co-ordinators consist of 5 full days (10 sessions) of training, with two sessions focusing on the organisational set-up required for the intervention and the remaining sessions focused on specific skills related to the delivery of the intervention. It is expected/assumed that the training of four HeLP co-ordinators will be delivered by one experienced trainer.

Costs associated with training are estimated assuming that costs are distributed over a 12-month period (for base-case cost estimates; over 27 classes). No venue costs were included in the estimated costings, and an allowance of £500 for other costs (e.g. materials, subsistence) has been included. Using these assumptions, the estimated cost for the delivery of the training component for the HeLP co-ordinators is £191 per class or approximately £8 per participant (Table 50).

Estimate training costs: drama team

Based on within-trial experience on training requirements, it is assumed that training requirements for drama delivery staff will consist of 5 full days (37.5 hours for drama facilitators, 30 hours for actors). The delivery of drama is carried out by a drama team comprising one drama facilitator and four actors. It is anticipated that the drama training, for two drama teams, will be delivered by one trainer.

Costs associated with training are estimated assuming that costs are distributed over a 12-month period (for base-case cost estimates; over 27 classes). No venue costs were included in the costing, and an allowance of £500 for other costs (e.g. materials, subsistence) has been included. Using these assumptions, the estimated cost for the delivery of the training component for the drama teams is £290 per class or £11.60 per participant (Table 51).