Using a variation of the policy Delphi technique (Franklin and Hart 2007), a multidisciplinary panel with professional expertise in public health, nutrition and obesity was consulted on implementation of the meal deals legislation (Public Health Wales 2025). This technique is particularly helpful where multiple models for implementation exist, providing an opportunity for experts to reflect on peer feedback and strengthen evidence for the final approach (Franklin and Hart 2007; Cubelo et al. 2024). It involved two key components. First, three iterations of an online survey were used to identify the acceptability of different nutritional thresholds and reference guidelines for an NP model. Second, an online panel meeting was used to confirm agreement with recommendations, including opportunities for feedback and a final vote. Figure 1 provides an overview of the Delphi process. Full surveys are available on the Open Science Framework (OSF) repository (https://osf.io/5u6gk/), and reporting of this study aligns with best practice recommendations for the Delphi method (Franklin and Hart 2007).

Overview of the Delphi process

A targeted, purposive approach to recruitment was used to identify stakeholders working in areas aligned to diet, nutrition and obesity in Wales. To facilitate this approach, the lead Public Health Consultant (IJ) invited contacts from existing organisation networks with broad reach across public health, healthcare, academia and third sector. Sample size guidelines for Delphi studies recommend recruiting between 10 and 15 stakeholders where expertise and professional backgrounds are relatively homogeneous (Taylor 2020). To account for variability between sectors, 38 participants were initially contacted to join the panel via a personalised email invitation, and a smaller list of additional contacts (N = 10) was collated via wider professional networks to combat participant attrition in follow-up rounds. This meant that a total of 48 stakeholders were invited to take part across rounds.

Abiding by national policy for research ethics (Health Research Authority), a protocol for the study was submitted internally to the Public Health Wales Research & Development Division, which advised that National Health Service (NHS) Research Ethics approval was not required. All participants completed the study on a voluntary basis and responded to recommendations in a professional stakeholder capacity. Information about the study was included at the beginning of each survey round and meeting. All participants were aware of the study aims from the outset and were clearly informed of their right to withdraw. All participants subsequently consented to taking part by continuing with the study and submitting a response.

A series of online surveys were created using Microsoft Forms (https://forms.office.com/) and circulated to stakeholders at three different time points between May and June 2023. Though the term ‘meal deal’ was not specifically defined in the first round, stakeholders were asked to respond to ‘lunch’ and ‘dinner’ as two distinct promotions, with explanations of both the bundle type (i.e. item combinations) and number of intended servings (i.e. single, couple or family of four) included in subsequent rounds. Though all contacts were invited to complete the second-round survey (regardless of participation in the first round), only those completing the second-round survey were invited to the final-round survey. Each survey was completed in approximately 10–20 min, and all participants had the option to submit responses in Welsh or English.

For the first-round survey, stakeholders were presented with three NP models relating to: (1) limiting the total energy per person for a purchased meal, (2) mandating the proportion of daily nutrients (energy, fat, sugar and salt) for a purchased meal, and (3) scoring individual items for HFSS levels (relative to fruit, vegetable and nut content, fibre and protein), excluding those with an overall ‘maximum’ (higher) score classified as ‘less healthy’ (Department of Health 2011). Stakeholders were asked to indicate their level of agreement with each model using a five-point Likert scale (from ‘Strongly disagree’ to ‘Strongly agree’). They were then asked to select the most appropriate reference intake values for the first two models as a single-option response (British Nutrition Foundation 2021). For dinner, this included the recommended limit for energy per person for a meal (ranging from 600 to 1100 kcal), and the recommended proportional limit per adult and child for daily nutrients provided by a purchased meal (ranging from 30% to 50%), identifying the most appropriate child age to use in an open-text field. For lunch, this included the recommended limit for energy per person for a meal (ranging from 400 to 900 kcal), and the recommended proportional limit per adult for daily nutrients provided by a purchased meal (ranging from 20% to 45%). Where relevant, stakeholders had the option to select ‘Other/Don’t know’, and could provide additional recommended models, reference values or criteria in an open-text field.

For the second-round survey, stakeholders responded to the same models identified in the first round. However, they were asked to indicate their level of agreement with each model and associated reference values using a three-point Likert scale (‘Disagree’ – ‘Neutral’ – ‘Strongly agree’). Where appropriate, additional models suggested by stakeholders were included for consideration, and the range of reference values presented were reduced in line with panel consensus. They could also select ‘Other/Don’t know’, and could continue to provide additional recommended models, reference values or criteria in an open-text field.

For the final-round survey, stakeholders were presented with a visual summary of previous round responses, including current group-level agreement for each model and reference value expressed as a cumulative percentage. They were asked to either ‘Agree’ or ‘Disagree’ with each item as their final survey response, or otherwise refrain from responding by selecting ‘Prefer not to comment’. Where relevant, any final suggestions or feedback about their participation in the study were provided in an open-text field.

A final panel meeting was held in December 2023 to discuss survey recommendations. With the addition of one new stakeholder at this stage, all invitees to survey rounds (N = 47) were contacted to attend a group meeting or one-to-one discussion with a member of the research team (IJ) hosted via Microsoft Teams (https://teams.microsoft.com). Prior to discussion, all stakeholders received a summary of survey results via email, including the level of consensus reached for each recommendation across survey rounds. This allowed stakeholders to engage in an open discussion of recommendations, including reasons for agreement and disagreement shared among peers. All stakeholders were invited to respond to a final vote for draft recommendations, selecting ‘Agree’ or ‘Disagree’ for each statement when submitting a response.

Across surveys and the final vote, frequency of agreement (disagreement) was calculated for recommended models and reported as a percentage across relative values (i.e. ‘Strongly agree’ and ‘Agree’, vs ‘Strongly disagree’ and ‘Disagree’). In line with similar Delphi approaches for public health (Dedewanou et al. 2023; Embling et al. 2025), the target level of agreement for each recommendation was set to 70% of stakeholders (excluding neutral/abstain responses) to revise statements and measure consensus. Where appropriate, reference values associated with models were removed from consideration in round 1 if these were not selected by any stakeholder. Otherwise, item changes were minimal across rounds. As additional measures of consistency between panellists, intra-class correlation coefficients (ICC) and Fleiss’ kappa were used to indicate inter-rater reliability within and between stakeholders for each survey round, and Cronbach’s alpha was used as a measure of internal consistency across statements (excluding one item as variance was equal to 0). Higher values indicate greater consistency within and between stakeholders for each survey round, reaching ‘moderate’ levels where values exceed 0.60 (McHugh 2012).