Study population
The UK Biobank study is a prospective cohort of more than 500,000 adults aged 37–73 years recruited from 22 assessment centers across the UK between 2006 and 2010 (instance 0) [36]. At enrollment, participants completed a questionnaire, a verbal interview, and a physical assessment for anthropomorphic data and vital signs. The blood specimens for laboratory analysis were also provided. Between 2012 and 2013 (instance 1), approximately 20,000 participants returned for the first repeat of the entire baseline assessment. A subset of UK Biobank participants was invited for multiparametric magnetic resonance imaging (MRI) between 2014 and 2016 (instance 2). In the present study, a total of 327,387 participants at instance 0 were included in the cross-sectional analyses after applying the following exclusions: 70,383 individuals with missing information for baseline lifestyle evaluation, 35,558 with any missing data related to MAFLD definition, and 97,888 participants with prevalent cancer (Figure S1). Meanwhile, participants in the prospective analysis were a subgroup of 15,721 participants in instance 1 after applying the following exclusions: 2,360 individuals with missing information for baseline lifestyle evaluation, 2,536 with any missing data related to MAFLD definition, 90,159 participants with prevalent cancer, and 4,582 with MAFLD at baseline (Figure S1).
The UK Biobank study was approved by the National Information Governance Board for Health and Social Care and North West Multi-Center Research Ethical Committee (11/NW/0382). All participants provided informed consents.
Data collection
The questionnaires and verbal interviews collected detailed information on sociodemographic data, lifestyle factors, medical history, and medication use at baseline. Townsend deprivation index, a composite measure of socioeconomic deprivation based on employment, home and car ownership, and household crowding, was assigned based on the postal code of residence [37]. Nurses conducted physical measurements and collected data on height, weight, waist circumference, and blood pressure. Body mass index (BMI) was calculated as weight in kilograms divided by the square of height in meters. Measurements of triglycerides, gamma-glutamyl transferase (GGT), high-density lipoprotein-cholesterol, glycated hemoglobin, and C-reactive protein were completed according to a standardized procedure [38]. The abdominal MRI scan in the multimodal imaging study was completed using the LiverMultiScan© protocol (Perspectum Diagnostics) [39]. The date and cause of hospital admissions were obtained from record linkage to Health Episode Statistics (England and Wales) and the Scottish Morbidity Records (Scotland) (see https://content.digital.nhs.uk/services).
Assessment of healthy lifestyles
The combined overall healthy lifestyle score in our study was composed of six components (diet [21,22,23, 40], alcohol consumption [41,42,43], physical activity [12, 13, 44], sedentary behavior [12,13,14,15], sleep [16, 17], and smoking [18,19,20]) based on a priori knowledge of NAFLD risk factors from published literatures and public health recommendations, such as those from the World Health Organization [45] and the European Association for the Study of the Liver-European Association for the Study of Diabetes-European Association for the Study of Obesity (Figure S2) [8].
The information on lifestyle was obtained through a self-completed touch-screen questionnaire at baseline, and the data was used in both cross-sectional and prospective analysis. Participants got one additional point if they met one of the following six criteria respectively (Table S1). A healthy diet was assessed by food frequency questionnaires and defined as having adequate intakes for ≥ 5 of 10 food components according to a previous paper in UK Biobank [32], which was considered as a healthy dietary pattern for cardiovascular disease, T2D, and obesity [32]. It included high intakes of fruits, vegetables, whole grains, fish or shellfish, dairy products, and vegetable oils, and low consumptions of refined grains, processed meats, unprocessed meats, and sugar-sweetened beverages. Moderate alcohol consumption was defined as drinking ≤ 20 g/day for women and ≤ 30 g/day for men, or drinking not regularly (“never drinking”, “drinking on special occasions only” or “drinking one to three times a month”) [8]. Sufficient physical activity was defined as moderate activity ≥ 150 min/week, vigorous activity ≥ 75 min/week or any combination of moderate-intensity and vigorous-intensity activities achieving ≥ 600 metabolic equivalent task min/week [45]. The low risk for sedentary behavior was ≤ 3 h/day of TV sitting which has been shown to better reflect overall sedentary behavior than other variables [10]. Adequate sleep was having ≥ 3 of the following sleep-related behaviors: (1) early chronotype (“morning” person or “morning” than “evening” person), (2) sleeping for 7 to 8 h/day, (3) never, rarely or sometimes experiencing insomnia symptoms, (4) no snoring, and (5) no excessive daytime sleepiness (“never/rarely” or “sometimes”) [46]. Finally, the low risk for smoking was defined as never smoking [18]. An overall healthy lifestyle score was calculated by summing points for all six factors above, with a range from 0 (least healthy) to 6 points (most healthy). Details for lifestyle definition were shown in Table S1.
Assessment of MAFLD
Because there was no sufficient UK Biobank data to define MASLD, we defined MAFLD as hepatic steatosis according to the international expert consensus statement, with at least one of the following criteria: (1) overweight or obesity; (2) lean/normal weight and presence of at least two metabolic abnormalities; or (3) T2D as described in Table S2 [1]. Fatty liver index (FLI) was used to define hepatic steatosis, where an FLI < 30 rules out and an FLI ≥ 60 rules in hepatic steatosis [1, 47]. FLI was calculated as (e0.953ln (triglycerides) +0.139BMI+0.718ln (GGT) +0.053waist circumference−15.745) / (1 + e0.953ln (triglycerides) +0.139BMI+0.718ln (GGT) +0.053waist circumference−15.745) *100, in which triglycerides were measured in mg/dL, GGT in IU/L, waist circumference in cm, and BMI in kg/m2 [47]. We used self-reported diagnoses, blood biochemistry measurement, medication, and hospital episode statistics data to define T2D [48] and liver diseases. According to the diagnostic criteria above, patients with MAFLD were assigned into (1) MAFLD-overweight/obesity subtype (MAFLD-O); (2) MAFLD-lean/normal weight and metabolic dysfunction subtype (MAFLD-L); and/or (3) MAFLD- type 2 diabetes mellitus subtype (MAFLD-T2D) [1]. MAFLD with any other liver disease was defined as dual (or more) etiology fatty liver disease (MAFLD-dual), while those without any other liver disease were defined as single etiology MAFLD (MAFLD-single) [1]. Accordingly, 122,269 participants were defined as MAFLD in the cross-sectional analysis, out of which 119,962 were MAFLD-O, 2,295 were MAFLD-L, 12,782 were MAFLD-T2D, 120,739 were MAFLD-single, and 1,530 were MAFLD-dual (Figure S1). Meanwhile, 5,543 participants were defined as MAFLD in the prospective analysis, in which 5,389 were MAFLD-O, 132 were MAFLD-L, 661 were MAFLD-T2D, 5,444 were MAFLD-single, and 78 were MAFLD-dual (Figure S2).
Statistical analysis
Participants were divided into 5 lifestyle score groups (0–2, 3, 4, 5, and 6 points) by pooling extreme groups with limited cases. None of the participants had missing values for the covariate information due to the participant exclusion mentioned above.
In light of the high MAFLD prevalence (37.3%) and incidence (35.1%) in the study population, we used Poisson regression with robust variance [49, 50] to estimate prevalence ratios (PRs) in the cross-sectional analysis and risk ratios (RRs) in the prospective analysis. The initial model (Model 1) was adjusted for age (years), sex (female and male), ethnicity (Caucasian, Asian or other), education (college/university degree or other), and Townsend deprivation index (quintiles). For the analysis of a single lifestyle, we provided a model (Model 2) that mutually adjusted the other five lifestyle factors. The linear trend for the association of the overall healthy lifestyle score with MAFLD was estimated by modelling the median values of each lifestyle score category as a continuous variable. We further analyzed the associations of each single lifestyle and overall healthy lifestyle scores with MAFLD subtypes. In addition, we assessed the statistical heterogeneity between MAFLD subgroups through Cochran’s Q test and Higgins’s I2 statistics [51].
We used linear regression to assess the associations of single and overall healthy lifestyle score with indicators of liver fat content, including FLI and proton density fat fraction (PDFF). The analysis on the associations of lifestyle with FLI was performed after further excluding participants with other liver diseases (n = 5,431 at instance 0 and 178 at instance 1, except NAFLD) from the primary analysis. The association of lifestyle with PDFF was estimated using data from instance 2. Among 16,307 participants with PDFF data, we excluded those with prevalent cancer (n = 5,791) and other liver diseases (n = 719, except NAFLD), leaving 12,102 participants for analysis. The linear regression analyses were adjusted for the same covariates as before.
We also performed stratified analyses according to age at enrollment (≥ 58 years old or not), sex (female or male), and ethnicity (Caucasian, Asian or other). Several sensitivity analyses were conducted to test the robustness of our results. First, we repeated the association of lifestyle with MAFLD among participants who were in the cross-sectional analysis but not in the prospective analysis. Second, we excluded participants with cardiovascular disease when examining the association between lifestyle and MAFLD to account for potential reverse causation. Third, as MAFLD might also present in participants with FLI between 30 and 59 [47], we redefined MAFLD as FLI ≥ 30 with one or more of the three criteria mentioned above. Fourth, we modelled a lifestyle weighted by the beta coefficient of each lifestyle factor estimated in the multivariate model, since the effect size between each lifestyle factor and MAFLD risk varied. Fifth, we replaced watching TV ≤ 3 h per day as a fourth subgoal in the sufficient physical activity category to avoid too much focus on the activity of the participants. Sixth, as the definition of MAFLD-single and MAFLD-dual already considered the association between MAFLD and alcoholic fatty liver disease, we excluded alcohol consumption from the overall healthy lifestyle when estimating the association of lifestyle with MAFLD-single and MAFLD-dual.
All P-values were two-tailed, and P < 0.05 was considered statistically significant. The analyses were performed with R version 3.6.3 (R Foundation for Statistical Computing, Vienna, Austria).