Understanding, comparing and learning from the four EPOCH early childhood obesity prevention interventions: A multi-methods study

Novel frameworks, taxonomies and experience from the Early Prevention of Obesity in CHildren (EPOCH) trials were applied to unpack interventions. Objectives: Deconstruct interventions into their components (target behaviours, delivery features and behaviour change techniques [BCTs]). Identify lessons learned and future recommendations for intervention planning, delivery, evaluation and implementation. Methods: This multi-methods study deconstructed the four EPOCH interventions into target behaviours, delivery features and BCTs from unpublished and published materials using systematic frameworks. Additionally, semi-structured interviews were conducted with intervention facilitators and principal investigators. Results: Each trial targeted between 10 and 14 obesity-related behaviours. Key variations in delivery features related to intensity, delivery mode and tailoring. BCTs consistently used across trials included goal-setting, social support, shaping knowledge, role-modelling and credible source. Recommendations from interview analyses include the importance of stakeholder collaboration and consideration of implementation throughout the study process. Conclusions: The combination of frameworks, methodologies and interviews used in this study is a major step towards understanding complex early obesity prevention interventions. Future work will link systematic intervention deconstruction with quantitative models to identify which intervention components are most effective and for whom.


| INTRODUCTION
Childhood obesity is a major public health and equity issue that sets children on a lifelong negative health trajectory. [1][2][3] Globally, 41 million children under the age of five had overweight or obesity in 2016. 3 Since children with obesity are five times more likely to have obesity as adults, a high number of those affected in the early years will be living with obesity for most of their lives. 4 Childhood obesity has been linked to many adverse chronic health conditions and poorer mental health. 5 Obesity prevention should start early when biology is most amenable to change, and before obesogenic behavioural patterns are established. 6 Previous systematic reviews have explored the effectiveness of early obesity prevention interventions conducted in a range of settings such as homes and primary care, highlighting the growing number of trials in the area in the last decade. [7][8][9] The Early Prevention of Obesity in CHildren (EPOCH) Collaboration 10 includes four landmark randomised controlled trials (RCT) to target very early childhood obesity prevention. All of the trials started within the first 6 months of life, and two began in pregnancy. Whilst all four interventions aimed to prevent early childhood obesity through behavioural and lifestyle components, they differed in design, intervention content, mode and intensity.
Through the EPOCH Collaboration, a core set of outcomes was identified to enable a prospective meta-analysis to be undertaken. 11 On average, the EPOCH interventions were effective in reducing body mass index (BMI) z-scores at 18 to 24 months of age by 0.12 standard deviations (95% confidence interval 0.02 to 0.22, P = .17), which translates into a 2% reduction in obesity prevalence. 12 Individually, the trials had less statistical power (all <0. 35) to detect an effect of this size on BMI z-score, compared to the combined analysis which had a power of 0.83 (as each trial was powered to detect differences in their respective primary outcomes). Only the Healthy Beginnings trial 13 showed a significant reduction in BMI z-score; however, all trials were successful in reducing some obesity-related behaviours such as reducing television times and improving feeding practices. [14][15][16] Further work is needed to unpack the similarities and differences across the interventions. This information will aid replication and translation of interventions as they are adapted for different populations and contexts or delivered at scale.
Standardised taxonomies and frameworks have been proposed to help develop, understand, compare and replicate complex intervention components. The multidisciplinary taxonomy of Behaviour Change Techniques (BCTs) 17 offers an internationally recognised system to classify the smallest, measurable and reproducible behaviour change components ("active ingredients") of interventions. The Template for Intervention Description and Replication (TIDieR) 18 checklist offers a clear, standardised approach for reporting and understanding intervention features. Applying these taxonomies and frameworks provides a consistent language to synthesise and compare characteristics and lessons across multiple interventions.
The practicalities of conducting complex early childhood obesity interventions can be challenging. 19 These interventions target parent behaviours to influence or change child outcomes. In this age group, most of the child's life takes place within the family environment, so the interventions need to be focused on the family. Interventions are delivered at an age where development happens very rapidly, and it is key to deliver appropriate intervention content at the right time point.
In addition, a number of stakeholders and theoretical frameworks need to be taken into account, and recruiting and engaging parents to such Previous studies have attempted to apply these taxonomies and frameworks to enhance understanding of complex interventions. [21][22][23][24][25][26] However, these studies were either reviews of the published literature, which were limited by the scarcity of information from published records, or were focussed on single interventions, thereby only offering one perspective. The EPOCH Collaboration offers a unique opportunity to obtain in-depth information on four different interventions, taking into account the perspectives and experiences of their investigators and facilitators.
In the current study, we aimed to capitalise on the recently developed approaches to understanding interventions presented above and the experience and resources from the EPOCH collaboration to look into the "black box" of complex early childhood obesity interventions.
The objectives of this study were to: (a) deconstructing intervention and (b) qualitative interviewing of key trial personnel.

| Deconstructing interventions into components
Intervention content was deconstructed into target behaviours, delivery features and behaviour change techniques using systematic frameworks to provide a consistent language for comparison across trials. 17,18 Published materials (protocols, outcome and other publications resulting from the trials) were used for analysis, along with unpublished intervention materials sourced directly from trial investigators, including protocol documents, facilitator handbooks, slide shows, videos and parent resources.
The behavioural targets of the interventions were extracted from the study protocols and tabulated by SM. The TIDieR checklist was applied by four independent coders (ALS, KEH, BJJ and CEM) to describe intervention delivery features, 18 including intervention materials and procedures, intervention facilitators (who delivered the intervention), how and where the intervention was delivered, when the intervention occurred, the number and frequency of sessions (duration and intensity), how consistently the intervention was delivered (fidelity) and whether any tailoring (planned personalisation, titration or adaptation) or modification of interventions occurred. Behaviour Change Techniques (BCTs) were coded using the BCT Taxonomy version 1. 17 Four trained coders 31 (ALS, KEH, BJJ and CEM) independently coded the interventions. Each intervention was coded twice with discrepancies discussed until consensus was reached. Coding followed a standardised process, with only BCTs that were clearly supported by evidence in published and/or unpublished content being coded as present. 17 Coder agreement was assessed by prevalenceadjusted, bias-adjusted Kappa. 32 Results were tabulated and summary statistics were calculated.

| Qualitative interviews
Semi-structured interviews were conducted between March and April 2019. One principal investigator and one intervention facilitator from each of the four trials were invited to participate. Each was provided with the study protocol, participant information sheet, interview guide and consent form. Participants were known to authors, due to their involvement in the EPOCH Collaboration, and all were invited to be co-authors on this manuscript. Interviews were conducted using the video conference software Zoom (http://www.ZOOM.us), and prospective verbal consent was provided. Ethics approval was granted by the University of Sydney Human Research Ethics Committee (2018/986).
The interviews explored the planning, delivery, evaluation and implementation of the complex interventions and contextual factors that enabled or hindered their delivery. The interview guide (see Table S1) was informed by the UK Medical Research Council development-evaluation-implementation process framework, 20 as well as by consultation with experts in the area of qualitative research in childhood obesity prevention (including ST) and relevant literature. 25,33 Particular focus was placed on exploring key lessons and recommendations for researchers and policy makers developing, evaluating and implementing interventions. Interview questions were pilot-tested with an experienced researcher (SM) and modified for flow. All interviews were conducted by ME, a doctoral student who is trained and experienced in qualitative research methods. Interviews lasted, on average, 53 minutes, were digitally recorded, transcribed verbatim and transcripts were cross-checked against recordings to ensure accuracy.
Inductive thematic analysis was used to analyse the interviews. 34 Two researchers (ALS and KEH) independently conducted the coding and analysis in consultation with a senior qualitative researcher (ST) using NVIVO software (version 12). This process firstly involved familiarisation with all of the interview transcripts. Each researcher then independently reviewed two transcripts to identify initial codes, and these codes were then combined in a preliminary coding framework through a consensus process. The researchers then independently coded four transcripts each using this framework, adding additional codes where necessary. This iterative process involved regular coding meetings (ALS, KEH and ST) to develop and define the themes until a final framework was achieved, and applied to all interviews.

| Behaviour change techniques
Across the four trials, 35 of the possible 93 unique BCTs were coded (Table 2). Only five to ten BCTs per trial were coded from published trial content, with an additional eight to 17 BCTs (63% increase in BCTs) coded using unpublished intervention materials.
Nine Examples of applications of the most commonly used techniques are presented in Table 3.
There was variation in the use of the remaining coded BCTs with six additional BCTs used in two trials and an additional 13 BCTs used in only one trial. The BCTs used less frequently were those related to reviewing and comparing goals and behaviour A key to our success has been involving government from the start. We have never been researchers on our own. We've always been co-designing every aspect of the programme. Ensuring that staff were adequately trained in what we were delivering and that they were giving consistent messages across the board.-We found that we employed some great people, but they also liked to do things their own way, and so they put their own spin on things (…) and it wasn't always correct.

Repetition of key messages
Information retention -re-emphasise key messages Simple and snappy messages We always needed to be reemphasizing and re-going over our key messages. There needed to be a lot of repetition because, as I said, a lot of the mums missed things the first time around.
EVALUATION -Themes to do with evaluating the intervention, such as feedback from participants and facilitators, and the evaluation process

| Qualitative interviews
Of the eight eligible participants, all agreed to be interviewed. Analysis of the transcripts resulted in the generation of four main themes: Planning, Delivery, Evaluation and Implementation/Scalability. Several

T A B L E 4 (Continued)
Themes Sub themes Illustrative quotes

Sleep -lactation/breastfeeding How to play with your baby
They were happy that we had hit the target with the things that we thought were important that they also thought were important (…) certainly breastfeeding, introduction to solids, when can my child eat with the family and eat family food. And a lot of emphasis on healthy food not take away. And with physical activity, definitely a lot of anxious mums around Tummy Times. So very happy with lots of the support around Tummy Time that we were able to provide.

Reach
Considerations how to reach groups from diverse backgrounds, for example, low SES, non-English speakers, regional and rural areas.
Tailor to more diverse, vulnerable populations, adjust to factors such as low literacy.
Importance of a large recruitment pool.

Awareness of cultural sensitivities
We still struggled to get to the harder-to-reach ones. It would have been better to have a different system for that.
With the non-English speaking groups, I think we tried to cater for them as much as we could but I don't think we were prepared.
I think group sessions are very difficult to get people to… they tend to be the worried well, the ones who are going to come to you anyway I think group sessions are very difficult to get people to… they tend to be the worried well, the ones who are going to come to you anyway I think it's very, very important that you look at having your resources so that they can be modified and very applicable across socioeconomic groups, across different cultures.

Ongoing process evaluations
Adapt intervention using ongoing process evaluations Meet with all of the facilitators and head researchers before writing the session and then we would meet after running the session (…) and review it and then upgrade it, change it, modify it based on all of the facilitator and researchers' feedback.

Minimum effective intervention
Focus on successful content, find minimal intervention to make a difference I'd like to go back and look at that data again and just say, well, if those mediators didn't affect the outcome, maybe they shouldn't be in the model. You know, maybe they shouldn't be part of the intervention. The intervention (…) had too much in it.
What's the sort of minimal intervention that I can do and still make a difference?

Sustainability
Importance of intervention fitting with existing services Political factors Cost and feasibility: reflect on resources available for scaling at planning stage, for example, time, delivery mode, intervention intensity, staff, etc. Consider cheaper delivery modes, such as phones, online modes, combination of face-to-face and technology Discontinuation of intervention (fade-out effect):

External influences can dilute intervention effect over time
It has to be embedded in service (…) It certainly can't be run by universities.
How do you support local government areas to adopt a programme (…) that is going to be scalable over time (…) sitting within funding frameworks that are sustainable Just providing it online isn't going to be the answer (…) it needs to be a mix. I think mother's need a choice.
I was very aware of the relationships with whole of government if you wanted to get a programme up and running. I was always very mindful of the potential for scaling up.
You embed a lot of really good knowledge into that family environment, but then a lot of other influences come into a kid's life when they start child care or grandparents look after them or lots of other things happen (…) that learning might get dissolved sub-themes were identified across each of the main themes. The themes Planning, Delivery and Evaluation mainly focus on the participants' experiences with each phase of the trial, whilst the Implementation/Scalability theme focuses on the participants' reflections, recommendations and improvements to scale up the interventions. Table 4 provides a summary of the main findings.

| Planning
Participants described factors during the planning phase of the intervention, which they felt influenced the uptake and success of the trial.
Across all four trials, participants emphasised the importance of collaboration with stakeholders throughout all intervention stages, from planning to implementation/scaling. In particular, they highlighted that researchers should seek input from a variety of sources to incorporate diverse perspectives into the development and implementation of the intervention to ensure it aligns with existing services and current practice. The mentioned sources of input included other researchers, clinicians, health services, government bodies, parents and the community.
"A key to our success has been involving government from the start.
We have never been researchers on our own. We've always been codesigning every aspect of the programme." Most of the interviewees pointed out the necessity to gain an indepth understanding of the target groups and target behaviours when planning the intervention, to ensure appropriate, efficient and effective design.

Most participants mentioned that intervention design should be
"You need to understand what determines the behaviours that you're targeting (…) So if it's a behaviour that requires skill as opposed to knowledge, your intervention will have a different spin." A recurring theme was the need to consider implementation and scalability issues, such as cost-effectiveness, feasibility and sustainability, early on during the planning stages (rather than at the completion of the RCT). In particular, some respondents emphasised the importance of being cognisant of real-life constraints that may limit implementation and translation into practice.
"Could we actually afford for maternal child health nurses to deliver this over time?"

| Delivery
Themes related to the process of delivering interventions differed based on the mode of delivery within individual trials. For group-based interventions, interviewees noted the advantages of interaction, social support and peer-modelling, in addition to efficiency and cost-effectiveness. One respondent mentioned the ability to magnify learning opportunities by having multiple voices in a room and sharing experiences.
"We felt that by role modelling with each other and sharing information there would be a lot more support to actually pick up some of our key messages." Delivering an intervention using a group-based model was "So what are the core components, how do we make sure they're being delivered?"

| Evaluation
Interviewees positively evaluated a variety of content that was covered in the interventions, for example, feeding and parenting style, introducing solids, sleep, tummy time and breastfeeding. Many noted the importance of developing self-efficacy and empowerment amongst parents. For example, the "Parent provide, child decide" message was described as "enlightening." Most interventions aimed to provide practical and low-cost solutions that parents could easily practice at home. Many interviewees raised the importance of ongoing process evaluations, so that the intervention could be continually enhanced based on feedback from facilitators, researchers and parents. For example, one trial removed a module on house safety due to lack of interest amongst participants when it moved into the small-scale implementation phase. Reach and accessibility of the intervention were key themes raised in many interviews. Respondents often reported difficulties in recruiting participants from diverse and vulnerable populations, such as low socio-economic position groups and those living in regional and rural areas. It was commonly recommended that how best to reach these groups should be carefully considered, and intervention materials should be adjusted to cater for factors such as low literacy and cultural sensitivities.
"It's very, very important that you look at having your resources so that they can be modified and very applicable across socioeconomic groups, across different cultures."

| Implementation/scalability
Two important and interrelated themes were those of a minimum effective intervention and sustainability. The concept of a minimum effective intervention refers to focussing on successful content and discarding ineffective or inefficient components to deliver a more targeted, effective and streamlined intervention.
"I'd like to go back and look at that data again and just say, well, if those mediators didn't affect the outcome, maybe they shouldn't be in the model. You know, maybe they shouldn't be part of the intervention.

The intervention (…) had too much in it."
The concept of sustainability refers to implementing an intervention program that can be sustained over time, and that has long-term help to maintain progress during and after the intervention to mitigate fade-out. 35 These BCTs could be particularly important given improved opportunities for e-health, allowing tailored messaging and goal setting after the main face-to-face intervention has been completed. 36 Importantly, minimum effective interventions may vary across different populations.
In addition to the systematic coding of BCTs, our study provides detailed description and comparison of the intervention delivery features of the four trials. 19 Interventions commonly targeted dietary, physical activity and parenting behaviours, through face-to-face delivery by a trained health professional, often combining written and visual materials.
Interventions delivered in a group setting enabled better monitoring of fidelity, whilst interventions delivered in an individual setting allowed the integration of additional elements of tailoring. Intervention procedures (what) and intensity (when and how much) varied between trials. A reason for this may be that there is no one best approach, but instead, future studies and implementation efforts may choose the approach most suitable to their context and population.
Applying the recommendations resulting from the current study (Box 1) may help to refine the key delivery features and BCTs for the local context to deliver the minimum effective intervention for each target group. As a next step, we recommend that investigators seek to develop streamlined or minimum effective interventions, which would maintain effect size when translated from research into practice and when delivered at scale, by determining which of the identified BCTs and intervention characteristics are associated with effectiveness. The newly formed TOPCHILD Collaboration will address the questions stemming from this current study (www.topchildcollaboration.org).

| Adaptation to different target groups and populations
A key consideration raised during interviews with trial principal investigators and facilitators was the need to adapt current intervention As a first step, the needs and context of the target group should be analysed, and the minimum effective intervention should be adapted

Box 1 Key recommendations for future very early childhood obesity prevention interventions
Key recommendations for future trials

Collaboration
• In addition to theory and evidence-informed planning, collaborate with existing stakeholders throughout (including other researchers/ experts, health services and the government, parents, and clinicians)from the planning stage to the implementation and scaling stageto ensure scalable and valuable interventions. • Ensure the intervention fits with existing services and aligns with current priorities and background care.
Feasible at scale/in health care setting • Be aware of scalability and real-life constraints from the start when planning a trial, for example, cost-effectiveness, feasibility, flexible and complementary delivery modes • Consider methods to develop sustainable, long-term interventions, for example, co-design, embedding within existing service structures • Strive for seamless provision of services across age groups and content areas (to avoid fade-out effect)

Minimal effective intervention
• Find out what worked, and use this content to develop effective minimal interventions. Trialists should consider whether commonly targeted behaviours relating to early feeding, diet quality, activity and sedentary behaviour/screen time are appropriate for their trial. • Consider commonly used behaviour change techniques (such as Goals and Planning, Social Support, and Shaping Knowledge) that can be leveraged when planning intervention content. Note that less frequently used BCTS may also be promising and should be considered as possible candidates.

Delivery lessons
• Repeat and re-emphasise key messages throughout the intervention.
• Ensure facilitators are properly trained and have experience working with children and families • Consider opportunities to tailor intervention elements, for example, parent-led interactive sessions, flexible timings and locations.

Adaptation to different target groups
• Think about who your target groups are, for example, most vulnerable populations, and design the intervention to meet their needs.
• Cater for diverse groups, for example, offer interventions in multiple languages/simplified formats, consider cultural sensitivities.

Transparent reporting
• Be transparent in reporting intervention content, and where possible make all intervention materials accessible to other researchers, endusers and other stakeholders. • Clearly report which behaviours were targeted in an intervention, and whether the intervention is targeting a parent or child behaviour. to best suit their requirements. For this purpose, it is crucial for researchers to work with their communities to design effective and well-received interventions. Yet, in a competitive funding environment, this can be difficult to achieve. The expectation to present a fully planned randomised controlled trial to receive research funding limits the ability for process evaluation and piloting of different intervention components.

| Feasibility of interventions at scale
The present results add to the growing body of evidence examining factors that need to be considered when designing an intervention to be delivered at scale. Specifically, our findings align with previous research by emphasising the need to ensure that facilitators are appropriately trained, key messages are repeated throughout a program and delivery occurs across childhood to avoid fade-out effects of health gains. 37,38 A key message from the interviews was the need to seamlessly align the intervention with other efforts to prevent childhood obesity. A common recommendation was to offer novel interventions through existing health services, which can ensure alignment of interventions and also reduce cost and organisational strain, thereby improving the scalability of interventions. Additionally, aligning new early childhood obesity prevention interventions with existing interventions in later childhood, adolescence and adulthood in a variety of settings (eg, home, primary care and broader environment) can prevent fade-out effects and may also improve overall effectiveness by providing continuity.

| Moving the field forward through collaboration and co-design
The importance of co-design for effective health interventions has been documented previously, 38,39 and our findings support this. Collaboration and co-design with existing stakeholders, health services and governments were pointed out throughout the interviews as key to the success of the EPOCH interventions and their translation efforts. In each interview, collaboration with a variety of stakeholders was extensively discussed, and communication and co-design with policy makers were highlighted as a key factor for successful implementation efforts. In trials where implementation was unsuccessful, frequently cited reasons were lack of interest from the relevant policy makers and political factors such as government changes or limited funding for prevention efforts.
The EPOCH prospective meta-analysis is a prime example of a successful collaboration of researchers to build a program of research and move their field forward. Collaborating in a prospective metaanalysis enabled successful synthesis of study results upon completion to detect that very early interventions are successful in reducing childhood obesity, at least in the short term. In this current study, the collaboration enabled a deep understanding of interventions from a variety of perspectives and the derivation of recommendations for future interventions. Previous reviews 23 have deconstructed the EPOCH interventions using only published material (protocol and results papers). Access to unpublished intervention materials enabled us to identify more than double the number of BCTs compared with these previous reviews, thereby allowing a new level of insight to the interventions. Direct access to investigators and facilitators enabled us to conduct interviews and compare and contrast lessons from the four trials. Prospective meta-analysis is an emerging collaborative approach, 11 and this study is an important example of how this approach can enable new insights that progress the research agenda and implementation of interventions.

| Improving intervention transparency
Research into the "black box" of complex interventions depends on transparent reporting of intervention content and components. The introduction of standards such as TIDieR and CONSORT has improved the quality of research reporting. [40][41][42] However, guidance for the detailed reporting of intervention content remains limited.
Future studies should use systematic frameworks such as those outlined in the Behaviour Change Wheel to guide comprehensive reporting. 43 The development of a repository of intervention materials would also enable transparent reporting and avoidance of research waste (eg, no repeat of testing strategies less likely to work). This study uses systematic frameworks to describe intervention features and behaviour change content. 17 Increased transparency through the use of frameworks, reporting guidelines and taxonomies in the development and reporting of new interventions could improve the ability to understand and replicate interventions, and would enable quantitative modelling of effective components across trials, thereby enabling researchers to work together to find the most effective interventions.

| Strengths and limitations
Our study provides important insight into trial components, and reflections by experienced investigators on what did and did not work, across multiple family-focused childhood obesity prevention trials that were independently planned and conducted. A major strength of this study was the coding of intervention features and behaviour change content using both published and unpublished intervention resources, and the direct access to intervention facilitators and principal investigators.
Inclusion of unpublished intervention resources allowed coding of an additional 12 to 19 BCTs per trial compared with those coded by Matvienko-Sikar and colleagues using only published manuscripts. 23 The combination of different frameworks and methodologies used in this study was a key step towards enhancing our understanding of complex interventions in this field. These methodologies will be important for deconstructing interventions used in future trials and ultimately to quantitatively determine the effectiveness of each intervention component.
There are several limitations. Due to reporting and overlap of intervention content, we only coded whether BCTs were applied in a trial, but not which target behaviour a BCT was coded to, nor could we determine the dose or effectiveness of BCTs. For instance, we did not differentiate whether goal setting was applied to the target behaviour physical activity or diet, and whether it was applied only once or multiple times. Instead, we only coded whether goal setting was present in a trial. Furthermore, we did not code BCTs that parents were encouraged to expose their children to, but that were not actually delivered as part of the intervention. For instance, we only coded the BCT behavioural practice if physical activity sessions were directly offered as part of the intervention, but not, if parents were simply told that they could improve their child's fitness through physical activity sessions. In addition, the trials were planned and implemented several years ago; therefore, interviews may be subject to recall bias.

| Conclusion
Our study provides rich insights to inform future very early childhood obesity prevention initiatives that can be delivered cost-effectively at scale. The deconstruction of multiple early obesity prevention trials into their components, highlights both similarities and differences in intervention characteristics and behaviour change content. Behaviour change content that was consistent between studies included goal setting, social support, how and why to change a behaviour, being a role model and persuasion by a credible source. Interviewing trialists and facilitators across multiple interventions provides understanding of the challenges of and recommendations for intervention planning, delivery, evaluation and implementation. We recommend that researchers, policy makers and health service delivery practitioners collaborate in the planning and implementation of minimum effective interventions to ensure both scalability and sustainability.
Future research will explore the development of quantitative models to answer important questions about how these complex interventions work, for whom, and under what circumstances. For this purpose, a global trials registry is being established through the EPOCH and TOPCHILD Collaborations (www.topchildcollaboration.org) to quantitatively explore the key components of these complex interventions that are associated with effectiveness.

CONFLICTS OF INTEREST
All authors have completed the ICMJE uniform disclosure form at www.