Study design and participants
The UK Biobank is a national prospective cohort involving a sample of 502,536 participants aged 37–73 years who were recruited between 2006 and 2010. Sociodemographic information, lifestyle factors, medical history, physical and functional measures, and biological samples were collected from participants at baseline or follow-up assessments across 22 assessment centres in England, Scotland, or Wales [9].
Study measures
Sociodemographic, physical, and biological assessments
Face-to-face interviews and self-administered, touchscreen questionnaires conducted at the baseline assessment centre were used to collect information on age, sex, ethnicity, education, smoking status, alcohol consumption, and menopause status for women. Townsend area deprivation index was derived from participant postal codes of residence using aggregated data on unemployment, car and home ownership, and household overcrowding [10]. Physical activity levels over a typical week were self-reported using the validated international physical activity questionnaire, from which the total metabolic equivalent of task (MET) hours per week were calculated [11]. Trained staff took a series of physical measurements, including height, weight, and blood pressure. Additional File 1: Supplementary Table S1 contains details on the collection and analysis of biomarkers of interest, including markers of adiposity and central adiposity, glucose and lipid metabolism, blood pressure, liver enzymatic activity, and inflammation.
Dietary assessment
Dietary data were derived from the Oxford WebQ (to estimate average daily intakes of saturated fats, free sugars, and fibre) and a brief touchscreen questionnaire (to estimate daily servings of fruits and vegetables). All UK Biobank participants completed the touchscreen questionnaire on a computer at their initial assessment centre visit. We used the frequency of consumption for (i) fruit (fresh or dried) and (ii) vegetables (excluding potatoes), namely never, less than once per week, once per week, 2–4 times per week, 5–6 times per week, or once or more daily.
The Oxford WebQ is a web-based self-administered dietary assessment tool. It collects information on foods and beverages (up to 206 food items and 32 types of drinks commonly consumed in the UK over the previous 24 h) [12]. Between 2009 and 2010, 70,724 participants completed the Oxford WebQ as part of their baseline assessment. Between 2011 and 2012, all participants with valid email addresses (n = 331,013) were invited to complete the Oxford WebQ on four separate occasions every 3–4 months, from which approximately 53% of participants (n=176,012) contacted by email completed at least one assessment. A total of 211,050 participants completed at least one WebQ assessment [13].
For this study, average daily intakes for nutrients and energy were calculated from at least two WebQ assessments in order to estimate usual intakes. Total energy and nutrient intakes were generated by multiplying the number of portions consumed by the set quantity of each food portion size and its nutrient composition obtained from the UK Nutrient Databank Food Composition Tables from survey year 6 (2012–2013 and 2013–2014) [14,15,16]. Dietary fibre (non-starch polysaccharides) was calculated using the Englyst method [17].
In order to exclude dietary under/overreporters, individual estimated energy requirements (EERs) were calculated as basal metabolic rates with the use of the Schofield Equation from the 1985 FAO/WHO/UNU Expert Consultation Report on Human Energy Requirements; implausible energy intakes were assessed by taking the ratio of reported energy intake (EI) to EER (EI:EER) [18].
Exposure ascertainment
The main exposure was adherence to dietary recommendations based on WHO criteria [2, 19]: ≤ 10% saturated fats, ≤ 10% free sugars, ≥ 25 g/day fibre, ≥ 5 servings of fruits and vegetables/day (including fresh fruit intake, raw vegetable intake, and cooked vegetable intake).
Outcome ascertainment
All-cause mortality, total CVD, and fatal CVD were obtained through hospital admissions and death registries linked to the UK Biobank. Total CVD was defined as a hospital admission or death using ICD-10 (International Classification of Diseases, 10th revision) codes, including coronary heart disease (I20–I25), congestive heart failure or cardiomyopathy (I50, I50.1, 150.9, I11.0, I13.0, I13.2, I42, I43.1), and total stroke (I60–I64). Hospital admission data were available until June 30th, 2020, in England; October 31st, 2016, in Scotland; and February 29th, 2016, in Wales. Death registries included date of deaths if occurred before July 30th, 2020, in England, Wales, and Scotland.
We selected available biomarkers of the metabolic syndrome, including adiposity and central adiposity (body mass index [BMI], total body fat percentage, waist circumference), glucose metabolism (fasting blood glucose, glycated haemoglobin), lipid metabolism biomarkers (blood lipid fractions [triglycerides, LDL cholesterol, high-density lipoprotein cholesterol, lipoprotein A, and apolipoprotein B]), blood pressure (systolic and diastolic), liver and other organ enzymes (alkaline phosphatase, alanine aminotransferase, aspartate aminotransferase, and gamma glutamyltransferase), and inflammation (high-sensitivity C-reactive protein [hsCRP]).
Statistical analyses
Prospective associations between adherence to dietary recommendations and all-cause mortality, total CVD, and fatal CVD
The primary outcome analysis was based on the number of dietary recommendations met out of a total of 4, using 0 as the reference category. In secondary analyses, we examined associations based on adherence to individual dietary recommendations by mutual adjustment.
We examined longitudinal associations between adherence to dietary recommendations and all-cause mortality, total CVD, and fatal CVD. We used multivariable Cox proportional hazards models with age as the underlying timescale to estimate hazard ratios (HRs) with 95% confidence intervals (CIs) for each analysis of categorical and individual recommendations met. We also used the floated absolute risk method, which relies on group-specific variances to calculate CIs around the estimate of risk in each group (including the reference) and allows comparisons across exposure groups [20]. The proportional hazards assumption was based on Schoenfeld residuals and was not violated for the variables of interest in the adjusted model (P > 0.05). To calculate the time to follow-up, we used age at completion of the last dietary assessment as the start date until age of occurrence of the first event (death or CVD) or censoring date, whichever came first. This analysis was stratified by sex, adjusted for ethnicity (whites, others, unknown), region (England, Scotland, Wales), Townsend index of deprivation (quintiles 1–5 or unknown, with lower scores representing greater affluence), education group (vocational qualifications [NVQ, HND, HNC], any school degree [A-level, AS-level, O-level, GCSE, CSE], higher degree [college, university, of professional degree/qualification], none of the above, unknown), smoking status (never, previous, current, unknown), physical activity (continuous, total MET-hours/week), alcohol consumption (none, occasional < 1 unit/week, moderate 1–14 units/week, heavy > 14 units/week, unknown), menopausal status (yes, no, not applicable [men]), and log-transformed total daily energy.
Sensitivity analyses were conducted for the prospective analysis to exclude participants who had a CVD event within 2 years after completing their last 24-h online dietary assessment to account for reverse causality. A post hoc sensitivity analysis was conducted including people who completed 3+, 4+, and 5+ dietary questionnaires as more dietary questionnaires would reflect usual intakes more accurately. A post hoc exploratory subgroup analysis was also conducted by sex, age at recruitment (< 60 years or ≥ 60 years), smoking status, BMI group, and presence of risk factors (hypertension, diabetes, and high cholesterol).
Cross-sectional associations between adherence to dietary recommendations and cardiometabolic markers
We used multivariable linear regression to estimate geometric means (95% CIs) of metabolic syndrome biomarkers according to total adherence to dietary recommendations, adjusted for age, sex, and the same covariates specified above. This analysis included a sub-sample of participants who had cardiometabolic biomarkers measured at baseline, in addition to at least 2 dietary assessments (the first one at baseline and the subsequent ones during the follow-up).
All statistical analyses were performed using Stata (version 14.0; StataCorp LP) statistical software; a two-sided p value was set at < 0.05 for statistical significance.