The UK Biobank is a large-scale prospective cohort study that recruited over 500,000 participants aged 37 to 73 years at baseline from diverse regions across the United Kingdom. From 2006 to 2010, participants attended one of 22 assessment centers, where they underwent comprehensive physical measurements, provided biological samples, and completed detailed touchscreen questionnaires. Ethical approval for the UK Biobank study was granted by the North West Multi-centre Research Ethics Committee [26].

For this analysis, the UK Biobank dataset included 502,406 participants. We excluded individuals who subsequently withdrew, those diagnosed with MAFLD or other liver diseases prior to study entry, participants with a history of substance abuse or alcoholism at baseline, and those lacking data on meat intake. A total of 484,875 participants remained eligible for the analysis of associations between meat consumption and MAFLD incidence. For the gene-diet interaction analysis, we further restricted the sample to individuals of white British descent with available genetic data, yielding a final sample of 364,619 participants for this analysis (Fig. 1).

Flow of participants in current UK biobank study

The intake of various types of meat, including unprocessed red meat, processed meat, oily fish, non-oily fish, beef, lamb, pork, and poultry, was assessed through a touchscreen-based short food frequency questionnaire (FFQ) covering the past 12 months. Meat intake frequencies were categorized as never, < 1 time/week, 1 time/week, 2–4 times/week, 5–6 times/week, or ≥ 7 times/week. Total meat consumption was calculated by summing the intake frequencies of unprocessed red meat (beef, lamb, and pork), unprocessed poultry, processed meat (bacon, ham, sausages, meat pies, kebabs, burgers, and chicken nuggets), oily fish, and non-oily fish. The short FFQ has been validated against 24-hour dietary recalls. Diet quality was assessed using the Alternate Mediterranean Diet (AMED) score, which ranged from 0 to 9, with higher scores indicating better diet quality, as previously described [27]. Healthy diet score includes 10 foods predictive of cardiometabolic disease risk, emphasizing higher intake of vegetables, fruits, fish, dairy, whole grains, and vegetable oils and lower intake of refined grains, processed meats, unprocessed red meats, and sugar-sweetened beverages [28]. Each dietary component was scored from 0 (unhealthiest) to 10 (healthiest) points, with intermediate values scored proportionally. The total diet quality score was the sum of all the diet component scores and ranged from 0 to 100, with a higher score representing a higher overall diet quality. The UK Biobank also utilized the Oxford WebQ, a web-based 24-hour recall questionnaire administered on five occasions, to collect dietary data between 2009 and 2012. The average dietary intake was calculated using all available assessments to represent long-term dietary intake.

Potential confounding factors were collected via touchscreen questionnaire, including age, sex, ethnicity, weight, height, income, education level, smoking and drinking habits, physical activity, Townsend deprivation index [29], and medical history. The metabolic equivalent of task (MET) was calculated using the short form of the International Physical Activity Questionnaire [30].

Genotyping considerations, quality control, and genetic imputation details have been described previously [26]. We selected five single nucleotide polymorphisms (SNPs)—rs738409, rs58542926, rs641738, rs1260326, and rs72613567 (Table S1)—associated with MAFLD risk, based on prior MAFLD cohort analyses. The genetic risk score (GRS) was calculated using these SNPs, with corresponding β coefficients applied in the following formula: GRS = (β₁ × SNP₁ + β₂ × SNP₂ +. + βi × SNPi) × (i / sum of the β coefficients) [31], where β represents the coefficient for each SNP, and i indicates the number of risk alleles for each SNP. A higher GRS reflects greater genetic susceptibility to MAFLD.

In this study, severe MAFLD was defined as hospitalization or death due to MAFLD or non-alcoholic steatohepatitis (NASH), based on linked hospitalization and mortality databases. Hospitalization data, including dates and diagnoses, were obtained from hospital episode statistics, covering participants in England and Wales until September 30, 2021, and in Scotland until September 24, 2021. Severe MAFLD was identified using the International Classification of Diseases, 10th Revision (ICD-10) codes K76.0 (fatty liver, not elsewhere classified) and K75.8 (NASH, other specified inflammatory liver diseases) [32]. The follow-up period was calculated from the date of recruitment to the earliest of the following events: first diagnosis of severe MAFLD, death, loss to follow-up, or the end of the study on November 12, 2021.

Continuous variables are presented as mean ± standard deviation, and categorical variables are presented as percentages. To examine the association between meat consumption (assessed by short FFQ) and the risk of severe MAFLD, we used Cox proportional hazards models, categorizing meat intake into groups: never, 0–3 times/week, 3–5 times/week, 5–7 times/week, or ≥ 7 times/week (C1 to C5), with the lowest category as the reference group. The proportional hazards assumption was validated using Schoenfeld residuals, and results were expressed as hazard ratios (HR) with 95% confidence intervals (CI). Sequential models were adjusted for confounders identified in previous studies: Model 1 included adjustments for age (continuous) and sex (men or women); Model 2 further adjusted for ethnicity (white, Asian, black, mixed, or other ethnic group), assessment center location, BMI (in kg/m2; <18.5, 18.5 to 25, 25 to 30, ≥ 30, or missing), education level (college or university degree, vocational qualifications, optional national exams at ages 17–18 years, national exams at age 16 years, others, or missing), household income (<£18,000, £18,000-£30,999, £31,000-£51,999, £52,000-£100,000, >£100,000, or missing), smoking status (never, former, current, or missing), alcohol consumption (never or special occasions only, 1 to 3 times/month, 1 or 2 times/week, 3 or 4 times/week, or daily/almost daily), physical activity (quartiles), and the Townsend deprivation index (quartiles); Model 3 additionally controlled for intake of other meat types (unprocessed red meat, processed meat, unprocessed poultry, oily fish, and non-oily fish; categorical), vegetables (0 to 1 times/day, 1 to 3 times/day, more than 3 times/day), fruits (0 to 2 times/day, 2 to 4 times/day, more than 4 times/day), and total energy intake (quartiles); Model 4 included adjustments in Model 2 as well as additional controls for other meat types (5 categories), total energy intake (quartiles), and AMED score (total score minus the component for meat; quartiles). Missing data were managed by creating a missing indicator category where applicable. Additionally, to enhance the accuracy of dietary assessment and complement the analysis based on frequency, we incorporated food weight, which provides a more precise quantification of intake and reduces potential misclassification. Using variables from 24-hour dietary recalls reflecting mean intake in grams per day (g/d), we further examined the association between meat consumption and the risk of MAFLD based on food weight (n = 207,465).

To assess the interaction between meat intake and MAFLD GRS, we included a multiplicative interaction term in the Cox models. Further categorical analyses were conducted to determine whether HRs for a 1-standard deviation increase in meat consumption varied across tertiles of GRS. Subgroup analyses stratified by baseline characteristics were performed to evaluate the impact of covariate variations on associations. Sensitivity analyses included additional adjustments for lipid-lowering medication use and replacement of the AMED score with a healthy diet score, and exclusion of incident MAFLD cases diagnosed within the first five years of follow-up to minimize reverse causality.

All analyses were conducted using SAS 9.4 (SAS Institute, Cary, NC, USA), and a p-value of less than 0.05 was considered statistically significant.