Study design and participants
UK Biobank is a prospective cohort study. Between 2007 and 2010, UK Biobank recruited over 500,000 participants from the general population. Participants attended one of 22 assessment centres across England, Scotland, and Wales where they completed a self-administered, touch-screen questionnaire and face-to-face interview, and trained staff took a series of phenotypic measurement and biological samples. Participants who withdrew from the UK Biobank, including those who have left the UK (n=1099), were excluded from this study.
Exposure
Participants’ ethnicity was self-reported as baseline in a touch-screen questionnaire and was categorised as White, Black, and South Asian.
Outcomes
Outcomes were ascertained through individual-level record linkage of the UK Biobank cohort to routine administrative databases. Date and cause of death were obtained from death certificates held by the National Health Service Information Centre (England and Wales) and the National Health Service Central Register Scotland (Scotland). Date and cause of hospital admissions were obtained through record linkage to Health Episode Statistics (England and Wales) and Scottish Morbidity Records (Scotland). Detailed information about the linkage procedures can be found at http://content.digital.nhs.uk/services. At the time of analysis, hospital admission data were available up to 31 March 2017 and mortality data up to 30 April 2020. We defined CVD deaths using the European Society of Cardiology SCORE definition [9] (ICD-10 [International Classification of Diseases, 10th revision] codes I10-25, I47-50, I60-69, and I70-73) and incident CVD cases as CVD deaths or hospital admissions for myocardial infarction (I20, I21, I25), heart failure (I50), or stroke (I60-64).
Covariates
We developed models based on modifiable risk factors, including lifestyle, anthropometric parameters, and biomarkers. Television viewing time, dietary intake of food items, smoking status, and alcohol consumption were self-reported. Townsend area deprivation index was derived from the postcode of residence using aggregated data on unemployment, car and home ownership, and household overcrowding [10]. Systolic blood pressure (SBP) was measured by a trained nurse. Physical activity was self-reported using the validated International Physical Activity Questionnaire [11]. Grip strength was measured to the nearest 0.1 kg using a Jamar J00105 hydraulic hand dynamometer and the mean value from both hands used in the analyses. Height was measured to the nearest centimetre, using a Seca 202 stadiometer, and body weight to the nearest 0.1 kg, using a Tanita BC-418 body composition analyser. Body mass index (BMI) was calculated as weight/height2 and the World Health Organisation’s criteria were used to classify BMI into underweight (<18.5 kg/m2), normal weight (18.5 to 24.9 kg/m2), overweight (25 to 29.9 kg/m2), and obese (≥30.0 kg/m2). Central obesity was defined as waist-hip ratio (WHR) >0.85 for women and >0.90 for men. Biomarkers were measured at a dedicated central laboratory between 2014 and 2017. Our analyses included low-density lipoprotein cholesterol (LDL-c), triglycerides, glycated haemoglobin (HbA1c), cystatin C, and gamma-glutamyltransferase (GGT) as potential mediators. LDL-c and triglycerides are related to CVD and lifestyle factors such as smoking and obesity; HbA1c is a marker for diabetes and related to obesity; cystatin C is a marker of kidney function and related to diet, smoking, and body weight; and GGT is a marker of liver function and related to alcohol drinking and fatty liver disease. These measures were externally validated with stringent quality control [12]. High-density lipoprotein (HDL) cholesterol and C-reactive protein, two important CVD risk markers, were not included because those were not likely to be causal factors of CVD [13, 14].
Statistical analyses
Four analyses were conducted. Firstly, Cox proportional hazard models were used to analyse the association between risk factors and CVD for each ethnic group, with the results reported as hazard ratios (HRs) and 95% confidence intervals (CIs). Risk factors were grouped as deprivation, lifestyle (physical activity, TV viewing, diet, smoking, alcohol drinking), adiposity (BMI and WHR), physical factors (grip strength and SBP), and serum-based biomarkers (LDL-c, triglycerides, HbA1c, cystatin C, GGT). All risk factors were adjusted for age at baseline assessment and sex (model 1), lifestyle risk factors additionally for deprivation (model 2), adiposity additionally for lifestyle factors (model 3), and physical factors and biomarkers additionally for adiposity (model 4). We chose these models so that we only adjusted for potential confounders, and upstream causes, not for downstream mediators on causal pathways. For example, because the factors in models 2–4 are likely mediators between deprivation and CVD, they were not adjusted to avoid overadjustment. Continuous variables were standardised to their SDs for comparison between risk factors.
Secondly, we explored if any of the risk factors could explain the association between ethnic group and CVD risk. A baseline Cox model (model 1, adjusted for age and sex only) was used to assess the total association between ethnicity and CVD. Groups of risk factors as described above were adjusted for sequentially. The HR for ethnic group was then used to examine whether, and to what extent, the associations between ethnic group and CVD were attenuated. Percentage risk reductions (%RD) after adjusting for these factors were calculated as \(\% RD=\frac{HR_{M_1}-{HR}_{M_i}}{HR_{M_1}-1}\) to illustrate the magnitude of mediation [15]. This method is used instead of the counterfactual-based methods because it is hard to disentangle the interrelationship among risk factors in this study. This mediation analysis serves as an exploration of the relative importance of mediators, rather than an attempt to quantify the proportion mediated.
Thirdly, moderation analysis was conducted to quantify whether the effect estimate for a risk factor was significantly stronger or weaker in an ethnic minority group compared to White participants. The associations between risk factors and CVD were studied in each of the ethnic group. Then, the interaction terms between risk factors and ethnicity were included in Cox models and can be interpreted as the ratio of hazard ratios (HRBlack:HRWhite and HRSouth Asian:HRWhite). In a sensitivity analysis, the associations of SBP, LDL-c, triglycerides, and HbA1c were also studied with the adjusted of relevant diagnosis and medications (anti-hypertensive and cholesterol-lowering medications) as those could be confounders.
Lastly, the population attributable fractions of risk factors were calculated for each of the three ethnic groups in stratified analyses. These indicate the relative contributions of risk factors to CVD cases or death within each ethnic group, taking account of both the prevalence and hazard ratios of the risk factor [16, 17]. Prevalence was estimated from the included participants. Continuous variables (grip strength and serum biomarkers) were categorised by quintiles because assuming linear association between the risk factor and CVD in PAF analysis would imply all individuals could be modified to the lowest level of that risk factor which is unrealistic. The prevalence for these was calculated using the quintile associated with the highest risk, i.e. the 1st quintile for grip strength and the 5th quintile for serum biomarkers. The remaining quintiles (2nd–5th for grip strength; 1st–4th for other biomarkers) were combined and used as the reference group in estimating HRs.
To avoid inflated type I error due to multiple testing, all P-values presented were adjusted using Holm’s procedure and a corrected P-value <0.05 is considered statistically significant [18]. Proportional hazard assumptions were checked using Schoenfeld residuals, which revealed no significant violation (all P-values 0.05). Non-cardiovascular mortality was considered a competing event, and follow-up was censored at death [19]. All analyses were conducted using R version 4.0.2 with packages survival, and AF.