Deconstructing Complex Interventions: Piloting a Framework of Delivery Features and Intervention Strategies for the Eating Disorders in Weight-Related Therapy (EDIT) Collaboration

(1) Background: weight-management interventions vary in their delivery features and intervention strategies. We aimed to establish a protocol to identify these intervention components. (2) Methods: a framework was developed through literature searches and stakeholder consultation. Six studies were independently coded by two reviewers. Consensus included recording conflict resolutions and framework changes. (3) Results: more conflicts occurred for intervention strategies compared to delivery features; both required the updating of definitions. The average coding times were 78 min (SD: 48) for delivery features and 54 min (SD: 29) for intervention strategies. (4) Conclusions: this study developed a detailed framework and highlights the complexities in objectively mapping weight-management trials.


Introduction
Behavioural interventions including diet, physical activity, and psychological components are first-line treatments for obesity [1][2][3][4]. However, there are ongoing concerns that behavioural weight management, or some included strategies or methods of delivery, may induce or exacerbate eating disorders [5][6][7]. The eating disorders in weight-related therapy (EDIT) Collaboration aims to identify individual changes in eating disorder risk during weight management; further details are available elsewhere [8,9]. We hypothesise that specific components of interventions may influence eating disorder risk. For example, dieting predicts eating-disorder development in community samples [10], and dieting at a "severe" level is considered risky in terms of the likelihood of triggering binge-eating episodes [11]. Hence, there is a need to examine interventional evidence to determine whether specific intervention components may increase or decrease eating-disorder risk.
Understanding how intervention components may increase or decrease the risk of eating disorders would enable future interventions to be optimised. To date, no systematic examination of weight-management intervention components relevant to eating disorder risk has been conducted. Evidence shows that multicomponent (dieting, physical activity and behavioural) interventions are, for most people, effective for weight loss in the short term [12,13]. Further, systematic reviews have demonstrated that some components of these complex interventions may be more effective than others. For example, certain behaviourchange techniques, encouragement for positive health-related behaviours and contact with a dietitian are associated with intervention success [14][15][16]. Such research provides a useful insight into the tailoring of weight-management interventions to improve effectiveness.
However, no such evidence synthesis has examined the risk of weight-management interventions. Detailed mapping of weight-management intervention components is an important step in understanding how weight-management trials may increase or decrease eating disorder risk.
This study aimed to establish a detailed, objective coding framework of intervention components (i.e., delivery features and intervention strategies) used in weight-management interventions for the EDIT Collaboration.

Development of a Coding Framework
The coding framework was developed through an iterative consultation process. An initial list of intervention components was drafted (HJ) and refined (NBL and BJJ). The initial codes were then expanded and refined through consultation with experts in the field via the EDIT Collaboration Scientific Advisory Panel and Stakeholder Advisory Panel including clinicians, researchers and people with lived experience of obesity and/or eating disorders [8]. The revised coding framework was further refined via a stakeholder consultation survey, with input from researchers, clinicians and those with lived experience of obesity and/or eating disorders internationally [17]. Stakeholders were asked to rate intervention strategies for likelihood to increase or decrease eating disorder risk within the context of weight management, and to identify any additional strategies which may be relevant to eating disorder risk [17]. A detailed guidebook was developed which included a descriptor for each unique code.
Delivery features are defined as "a broad number of intervention characteristics that relate to how an intervention is delivered" [18]. Delivery features were developed based on the Template for Intervention Description and Replication (TIDieR) checklist [19], including the overarching goal, target population, materials provided, procedures used, who delivered the intervention, delivery mode, intervention setting and dose, as well as any tailoring, modifications and fidelity measures. We also summarised the number and range of different outcome assessment procedures, as these may unintentionally deliver important messages about the aim and intended outcome of the intervention in addition to the planned intervention content. Categories under each delivery feature item/cluster were developed for this project drawing on relevant examples from child obesity prevention [18] and the Human Behaviour Change Project ontologies [20,21].
Intervention strategies is the broad term used to describe the behaviour change content of interventions grouped under key categories (i.e., highest level grouping) relevant to weight-management interventions. Clusters (i.e., mid-level grouping) of intervention strategies were captured under the following categories: intervention intent, framing and outcomes, dietary strategies, eating behaviours/disorder eating, movement and sleep related strategies, and psychological health-related strategies. There were 86 unique intervention strategies (i.e., lowest-level grouping) across these five clusters.

Eligibility Criteria for the Pilot
Trials eligible for inclusion in the EDIT Collaboration were randomised controlled trials of behavioural weight-management interventions recruiting adolescents (aged 10 to <19 years at baseline) or adults (aged ≥18 years at baseline) with overweight or obesity defined as body mass index (BMI) z-score > 1 in adolescents and BMI ≥ 25 kg/m 2 in adults [22]. Trials must measure eating disorder risk at baseline and post-intervention or follow-up using a validated assessment tool.
Purposeful sampling methods were used to select studies for piloting. Eligible studies (n = 73 as at May 2022) were grouped by decade of publication, with each group weighted by the total number of studies to determine how many studies should be selected from each decade. Both random and purposeful approaches guided the final selection of studies for piloting, ensuring diversity of target population (adolescents, adults), availability of a published protocol and study country. For pragmatic reasons, any studies identified that declined to join the EDIT Collaboration were replaced with a study of a similar profile.

Pilot Process
Published intervention descriptions, from trial registries, protocol and main results publications, were used to code intervention components using a standardised procedure following a brief training session. The training session involved familiarisation with the coding framework and practising coding to assist with consistency. Each unique intervention arm was coded by two independent coders (RK and SP, with a background in dietetics, and psychology, respectively), conflicts were identified and resolved through discussion (all authors). Duration of coding time was recorded.
Following initial coding and consensus of all studies, all authors critically discussed and reviewed the coding framework to ensure adequate coverage and clarity of delivery features and intervention strategies. Existing codes and descriptors were refined and additional codes included from discussion. Studies were then recoded using the updated codes.

Synthesis of Results
Descriptive statistics were used to summarise coding conflicts, number of modifications to the codebook, and to calculate the average and standard deviation (SD) of the time required to code components of each study. Results were synthesised separately for delivery features and intervention strategies, and examined by cluster.

Results
The characteristics of the selected studies are available in Supplementary Table S1. Six studies consisting of a total of 14 active intervention arms were selected. The included studies originated from the United States of America (n = 3), Australia (n = 1), Brazil (n = 1), and the United Kingdom (n = 1). The average time to code each study was 78 min (SD: 48) for delivery features and 54 min (SD: 29) for intervention strategies.
There was a greater number of coding conflicts in the intervention strategies (n = 237, 19.7%) compared to the delivery features (n = 156, 13.9%) ( Table 1). The number of conflicts for each intervention arm ranged between 7 and 16 (8.9-20.5%) for delivery features and 2 and 34 (2.3-39.5%) for intervention strategies. When coding the delivery features, unclear definitions were the most common reason for conflicts and updating these definitions was required to achieve consensus. For instance, "duration of contact" was redefined to duration in minutes rather than brief, moderate or extended contact, which was subjective and difficult to code. In contrast, the consensus discussions for the intervention strategies revealed varying interpretations of definitions due to differing coder backgrounds. For example, the cluster "delivery of dietary intervention" had the greatest proportion of conflicts due to one coder having a comprehensive knowledge of clinical dietetic interventions. A total of 43 code definitions were modified, 15 new variables were added, and 7 were removed following the consensus meetings (Supplementary Table S2).
Consensus procedures for the new code resulted in 28 conflicts (13.3%) which were resolved through discussion and no further updates to variable definitions were required. The final revised coding framework included 86 delivery features ( Table 2) and 88 intervention strategies (Table 3).

Discussion and Conclusions
The EDIT Collaboration aims to identify intervention components that increase or decrease eating disorder risk as part of weight-management interventions [9]. We developed a framework for identifying intervention components, by deconstructing complex interventions into well-defined delivery features and intervention strategies. Using an established coding framework reduces the subjectivity of intervention deconstruction. In addition, this pilot study demonstrates the resourcing required to conduct a comprehensive deconstruction of complex interventions.
Our study highlights the need for utilising established and tested definitions to appropriately deconstruct complex behavioural interventions. The taxonomic deconstruction of interventions is useful for examining which components of an intervention may be driving outcomes. However, previous studies of deconstruction behavioural weight-management interventions have predominately focussed on weight outcomes [14][15][16]18] or other measures of effectiveness [23][24][25]. Future research should consider whether interventions, or components of interventions produce unintended effects, such as increasing eatingdisorder risk [5].
The strengths of this study include the use of a systematic coding process to validate the codes and definitions within the framework developed through extensive stakeholder consultation involving expertise from both the obesity and eating disorder fields. The framework was tested on a range of studies, varying in population (adolescents, adults), countries, and publication date. Coders were from differing disciplines (dietetics and psychology). The limitations include reliance on published or publicly available intervention descriptions, which may provide insufficient detail in the reporting of intervention components [26]. Further, we did not quantify the frequency or intensity of the intervention strategies.
Our coding framework will be implemented for all trials included in the EDIT Collaboration to examine eating-disorder risk during weight management [27]. Moreover, this framework can provide insight into a broad range of weight-management interventions and can be transferred or adapted to examine other safety or effectiveness outcomes (e.g., weight regain, health-related QOL, depression, etc). Coding frameworks, such as the one developed in this study, can assist in the transparent and systematic coding of existing interventions to enhance our understanding of the components of complex interventions.