Small portions of meat are the biggest contributor to reducing meat consumption in the UK

This study did not require ethical or regulatory approval as it used publicly available anonymised data from the UK National Data Service. The NDNS rolling program adheres to the Declaration of Helsinki and operates under the UK Health Research Authority’s Research Ethics Committee; approval references: #07/H0604/113 (Years 1–5) and 13/EE/0016 (Years 6–11)²⁹. Data collection for the NDNS was carried out in accordance with ethical regulations, as described in the NDNS documentation^29,30. Participants received up to £50 compensation for their time and participation in the survey: £20 for providing a blood sample during the nurse’s visit and £30 for completing three of four days of a food diary³⁰. Informed consent was obtained from all participants — or their guardians, as appropriate — as part of the original NDNS data collection process²⁹.

Data source and example

The NDNS rolling program is an ongoing cross-sectional survey, which collects detailed data on dietary intake and nutritional status information from the UK population aged 1.5 years and over, living in private households.^28,29,30. The NDNS is funded by Public Health England and the Food Standards Agency in the UK. It aims to monitor the diet and nutrition of the UK population, providing evidence of adherence to public health nutrition targets, ensuring continued government support and resources to collect and analyze nutritional data. Additional information on the NDNS methodology, including survey design and weighting, has already been described³⁰. In summary, the NDNS is designed to be nationally representative of the UK population and adjusts for age and sex population distributions through survey weighting. The sample was drawn from Postcode Address Files, which were grouped into Principal Sampling Units (PSUs) based on postcode sectors. From each PSU, a mailing list was randomly selected, and the interviewer randomly selected up to one adult and one child to participate from each household. This study included survey participants from Years 1–11 (2008–2009 to 2018–2019) of the rolling program.

Dietary intake data

Dietary data were collected using food diaries for four consecutive days, and the survey design ensured a fair representation of all days of the week. Detailed methodology for NDNS data collection has been described elsewhere²⁸. Briefly, participants were instructed to record all food and drinks they consumed over the assigned four-day period within a paper diary. For children ≤11 years of age, a parent or carer was asked to complete the four-day food diary with input from the child as appropriate. Children aged ≥12 years were asked to complete the diary themselves, confirming details with others where necessary²⁸. Participants estimated portion sizes using household measures (for example, a tablespoon) or reporting the weights on food labels. As this study explored the frequency of days when meat was eaten, participants were not addedn= 323, 2%).

Categorization of meat

We explored the consumption of whole meat and processed meat, red meat and white meat separately. We did not include fish consumption in this analysis. Estimates of meat intake were based on disaggregated data in which all non-meat ingredients of composite dishes were excluded (that is, grams of beef were only estimated in beef lasagne).^28,31. Meat items were disaggregated into previously existing categories within the NDNS. For dishes containing more than one type of meat, each type of meat was disaggregated separately, into one of 11 mutually exclusive categories. We grouped these categories into processed, red and white meat, consistent with the approach used in previous trend analysis of meat consumption in the NDNS.⁷:

1. 1.

   Processed meat – processed red meat, processed poultry, sausages and burgers.

2. 2.

   Red meat – beef, lamb, pork, other red meat and offal.

3. 3.

   White meat – poultry and game including duck.

Meat consumption behaviour

We assessed how meat intake has changed over time, and specifically explored change in four distinct meat consumption behaviors: (1) proportion of the population that eats meat, (2) frequency of meat-eating days for meat-eaters; (3) daily meat eating occasions for days on which the meat is eaten and (4) amount (in grams) of meat within meat eating occasions. For dishes containing more than one meat subtype (ie, processed, red and white meat), consumption frequency and portion size of each subtype were established separately. For example, ‘chicken, bacon and cream of mushroom’ was considered both a white meat and processed meat item. As a result, portion sizes of both white and processed meat were estimated separately and the frequency of consumption for each subtype was counted independently.

The percentage of the population that eats meat was calculated by a weighted ratio from a survey between meat consumers (>0 g) and non-consumers (0 g).

Meat-eating days were defined as the number of days on which meat consumers consumed any quantity of meat (>0 g) over the four-day period of the food diaries. We also explored the number of days on which no meat was eaten and investigated the distribution of individuals who ate meat on days 0, 1, 2, 3 and 4 over the four-day period.

Eating occasions were defined as an intake of ≥50 kcal (from all food and drink) recorded with an interval of >30 minutes between meals²⁶. We defined a meat-eating occasion as an intake of ≥50 kcal (from all food and drink) and >0 g of recorded meat with an interval of >30 min between meals. Mean daily meat eating occasions were calculated as the participants’ average daily number of meat eating occasions, over each day of a meat eating food diary.

We determined the amount of meat (in grams) consumed during each meat-eating occasion. Mean meat portion size was calculated by averaging the grams of meat consumed over each meat eating occasion. In addition, we investigated the amount of meat consumed according to typical meal times: breakfast (6:00–11:00 am), lunch (12:00–3:00 pm) and dinner (4:00–11 :00 pm), the same. time periods for previous mealtime analysis in the NDNS²⁶.

Socio-demographic characteristics

Sociodemographic variables included self-reported age, sex and tertile of equivalent family income. Regarding their age, participants were asked to provide their date of birth, or age at last birthday if not known; using interviewer estimates if participants were unable or unwilling to provide this information. We categorized participants into the age groups of children (<18 years) and adults (≥18 years). Participants were asked to identify themselves as male or female, and in undisclosed cases, the interviewer reported sex. Participants reported their total household income from the previous 12 months, before deductions and tax, including housing benefit and child allowance. Within the NDNS data files, these data were equalized, accounting for household size and composition and divided into boundaries.³².

Statistical analysis

To address clustering within the sample and to minimize potential selection and non-response bias at the household and individual level, our analyzes included weights of surveys and PSUs published in the NDNS dataset and additional clustering at the household level .

We report the percentage of meat consumers, the average number of days eating meat, the average number of daily meat-eating occasions per person and the average portion of meat within a meat-eating occasion in each survey year. We also report average daily meat consumption per capita. We explored trends over time (2008–2009 to 2018–2019) using Poisson regression models in analyzes of count data (frequency of meat-eating days) and generalized linear regression models for continuous data (percentage of meat consumers, occasions daily meat eaters). , portion size and average consumption per capita). In addition, we performed separate univariate analyzes for each population subgroup, taking into account factors such as gender, age group and tertile of equivalent family income. In those models, confidence intervals for the coefficients were calculated using the profile likelihood method, implemented by the confint() function in R base and illustrated for interpretability where applicable.

To estimate the percentage of responsibility for each meat consumption behavior relative to the overall reduction in consumption⁷we used a decomposition analysis, based on the following equation applied by Alexander et al.³³.

Here c_i,t it is the total average amount of meat consumed per person i in time t (where t₁ is a baseline and t₂ the following year survey), and XShows (separately): the proportion of meat consumers, the average number of days eating meat, the average number of daily meat-eating occasions and the average portion of meat (g) within a meat-eating occasion. We also ran the disaggregated analysis by sex, age group and income equivalent.

All analyzes were performed in R version 4.2, using the ‘survey’ and ‘srvyr’ packages to account for survey weighting in the demographic and regression analyses. pThe criterion for statistical significance in trend analyzes was < 0.05 and p< 0.1 for subgroup interactions.

Reporting summary

More information on research design is available in the Nature Portfolio Reporting Summary linked to this article.