Differences in the risk of cardiovascular disease across ethnic groups: UK Biobank observational study

Background and aims: To describe sociodemographic, lifestyle, environmental and traditional clinical risk factor differences between ethnic groups and to investigate the extent to which such differences confound the association between ethnic groups and the risk of cardiovascular disease (CVD) Methods and results: A total of 440,693 white European (55.9% women), 7305 South Asian (48.6%) and 7628 black African or Caribbean (57.7%) people were included from UK Biobank. Associations between ethnicity and cardiovascular outcomes (composite of non-fatal stroke, non-fatal myocardial infarction and CVD death) were explored using Cox-proportional hazard models. Models were adjusted for sociodemographic,


Introduction
Cardiovascular diseases (CVD) are the most prevalent and costly non-communicable diseases worldwide, with ischaemic heart disease and stroke both ranked within the top three causes of death globally [1]. Differences in CVD risk between ethnicities is well evidenced, with South Asian (SA) and black African or Caribbean (BAC) individuals displaying an increased risk of developing and dying from CVD, compared to white European (WE) individuals within the UK [2e4] and other European countries [5,6]. SA individuals display a particularly high risk of ischaemic heart diseases, while black Caribbean individuals report elevated risk of cerebrovascular diseases.
Understanding the reasons behind this increased risk is important for public health strategies. Lifestyle behaviours, such as smoking, physical activity, physical function/capability (measured via simple to measure behaviours such as walking pace), sedentary time and diet have been found to be important risk factors that may contribute towards CVD [7e14]. Furthermore, lifestyle behaviours have consistently demonstrated to improve cardiovascular health and reduce the risk of CVD, diabetes and mortality in all ethnicities [15e19]. Minority ethnic groups have been noted to possess different dietary [20] and sleep [21] habits compared to WE individuals, whilst engaging in less physical activity [22] and having lower cardiorespiratory fitness [23]. In addition to lifestyle behaviours, wider sociodemographic and environmental factors, such as levels of deprivation and air pollution, are also important determinants of cardiovascular and wider health [24e26], with exposures reported to differ by ethnicity [27,28]. Clinical risk factors have also been established as risk factors for CVD [29], with known differences in clinical risk factors between ethnic groups potentially contributing to (confounding) the increased risk between ethnic groups [30,31].
Although there is some evidence that differences in individual lifestyle behaviours or risk factors, such as physical activity or fitness, may explain some of the excess risk of cardiovascular disease in minority ethnic groups [32,33], the full extent to which wider differences in lifestyle, sociodemographic, and environmental factors act as confounders in the association between ethnicity and CVD risk has not yet been fully explored.
Therefore, the aim of this study was to describe sociodemographic, lifestyle, environmental and traditional clinical risk factor differences between WE, SA and BAC individuals in the UK Biobank cohort and to investigate the extent to which such differences confound the association between ethnicity and risk of fatal and non-fatal CVD events.

UK Biobank
Data from 502,539 participants from UK Biobank were included, the methods and aims of which have been described elsewhere [34]. In brief, UK Biobank is a prospective cohort of middle-aged adults designed to support health research focused on improving the prevention, diagnosis, and treatment of chronic diseases. Between March 2006 and July 2010, individuals living within 25 miles of one of the 22 study assessment centres located throughout England, Scotland and Wales were recruited and attended data collection. All participants provided written informed consent and the study was approved by the NHS National Research Ethics Service (Ref: 11/NW/ 0382).
For each variable reported in this analysis, we provide the UK Biobank Data-Field (DF) number. Each DF is linked to detailed information on measurement procedures in the UK Biobank showcase.

Study population, participant inclusion and classification of ethnicity
From the initial 502,539 participants, we included those without prevalent CVD at baseline, a valid censor date and defining themselves as white (British, Irish, white or any other white background; n Z 440,693; 96.7%), SA (Asian or Asian British, Indian, Pakistani or Bangladeshi; n Z 7,305; 1.6%) or BAC (black or black British, Caribbean, African or any other black background; n Z 7,628; 1.7%). Other ethnicities were excluded due to low numbers. From this sample, only subjects with data for each exposure variable investigated were included (Supplementary Fig. S1 and Table S1). Recruitment to UK Biobank occurred between the 2001 and 2011 national census dates.

Lifestyle, environmental, sociodemographic, medical history and traditional clinical risk factors
The following variables were included as possible factors that may be important in confounding the higher risk of CVD in minority ethnic groups. These factors were chosen based on a combination of previous evidence suggesting that traditional clinical, medical history and sociodemographic risk factors such as triglycerides, previous chronic disease and deprivation are associated with CVD, in addition to emerging risk factors such as environmental, lifestyle and sociodemographic factors like air pollution and walking pace, where there are known differences between ethnic groups. Leisure time physical activity was derived by creating a variable that combined the duration and frequency of selfreported leisure time physical activity behaviours. Walking (DF 971 and 981), strenuous sports (DF 991 and 1001), other exercises (DF 3637 and 3647) and heavy DIY (do-ityourself) (DF 2624 and 2634) were included and a metabolic equivalent (MET) was assigned to each activity based on previous literature [35].
Frequency categories for fruit and vegetable intake and red meat intake were converted into two continuous measures of consumption, as follows: a value of 0 was allocated to "never", 0.5 for "less than once a week", 1 for "once a week", 3 for "2e4 times a week, 5.5 for "5e6 times a week" and 7 for "once or more daily". Servings for the fruit and vegetable and red meat variables were separately summed to create the frequency of fruit and vegetable and red meat consumption, respectively.
The environmental exposure noise estimates were assigned to residential address at recruitment in year 2005e10. NO 2 and PM measures were modelled to participant address using land use regression (LUR) models originally developed for the European Study of Cohorts for Air Pollution Effects (ESCAPE) [36,37]. Particulate matter ESCAPE estimates were not available for addresses >400 km away from Greater London as the model evaluations were poor. Noise estimates were modelled as recommended by the European Noise Directive 2002/49/ EC and relate to 2009 [38]. As the spatial structure of air pollution and noise exposure is relatively stable over time (as road networks, land use characteristics change slowly), estimates are valid across the recruitment period.

Medical history
Number of medications/treatments taken (DF 137), number of self-reported cancer illnesses (DF 134) and number of self-report non-cancer illnesses (DF 135) were included.
Total cholesterol (mmol/l; DF 30690), LDL cholesterol (mmol/l; DF 30780) and body fat percentage (%; DF 23099) were additionally used. Participants provided blood samples at their assessment centre visit, which was taken by a trained healthcare professional. Glucose, triglycerides, HDL-cholesterol, C-reactive protein and total cholesterol serum samples were measured using the Beckman Coulter AU5800 analytical platform and analysis method was by hexokinase, GPO-POD, enzyme immuneinhibition, immunoturbidimetric -high sensitivity and CHO-POD, respectively [34]. HbA1c red blood cell samples were measured using the Bio-Rad VARIANT II Turbo analytical platform and analysis method was by HPLC.

Cardiovascular events
Three-point major adverse cardiovascular events (3P-MACE) was defined as a combined outcome of non-fatal stroke, non-fatal myocardial infarction (MI) and CVD death. UK Biobank undertook comprehensive data linkage for mortality status. Date and cause of death were obtained from the National Health Service (NHS) Information Centre for participants from England and Wales, and from the NHS Central Register, Scotland, for participants from Scotland. We defined cardiovascular disease mortality using the International Classification of Diseases edition 10 (ICD-10) codes I05-I89.9. Incidence of MI (DF 42000) and stroke (DF 42006) were quantified using a validated UK Biobank algorithm that triangulates nurse-confirmed selfreport and hospital admissions data [39]. Linkage captured fatal and non-fatal events occurring until 30 September 2021 for England and Wales and 31 October 2021 for Scotland, with these dates being used as censor dates. Individuals with prevalent CVD at baseline were removed from the cohort (Fig. S1).

Statistical analysis
Unadjusted generalised linear models and nonparametric tests were used to explore differences in sociodemographic, lifestyle, environmental and traditional clinical risk factors between ethnic groups.
The associations between ethnicity and cardiovascular outcomes were explored using Cox-proportional hazard models with years of follow-up as the time scale. Clustering by UK Biobank assessment centre was included in our models to account for potential differences by assessment centre. Analyses were performed using 3P-MACE as an outcome. The results were reported as hazard ratios (HRs). WE individuals were used as the reference group. First, we estimated the association in an unadjusted model (model 0). Models were then adjusted for key sociodemographic factors considered to be potential confounders, namely age, sex, self-reported cancer illnesses, number of self-reported non-cancer illnesses, number of medications/treatment taken, deprivation, employment status, qualifications and household income (model 1). Model 1 was subsequently considered the base model. To model 1 we added markers of physical function (handgrip strength, walking pace; model 2), lifestyle (MET score for leisure time physical activity, TV viewing, sleep duration, processed meat consumption, red meat score, fruit and veg score, milk type used, spread type used, alcohol intake, BMI, smoking status; model 3), environmental (NO 2 and PM 2.5 air pollution, LA eq24 noise; model 4) and traditional clinical risk factors (HbA1c, glucose, triglycerides, HDLcholesterol, C-reactive protein, cholesterol-HDL ratio and systolic blood pressure; model 5) considered to be potential confounders to the base model. A final model which included all factors was also considered (model 6). This analysis plan was developed to assess the degree to which the risk of cardiovascular outcomes between different ethnic groups were independent of differences in assessed sociodemographic, lifestyle, environmental and traditional clinical risk factors. This clustering of factors allowed us to evaluate the comparative impact of lifestyle vs environmental vs clinical factors as we separately added these to the same base model. The degree of attenuation upon adjustment was interpreted as the extent to which ethnic differences in the risk of cardiovascular outcomes were independent of differences in assessed sociodemographic, lifestyle, environmental and traditional clinical risk factors. A further analysis that separately adjusted for each individual lifestyle, environmental and traditional clinical risk factor, instead of grouping them together, was conducted.
In order to maximise power, a complete case sample was used for each model individually; Fig. S1 and Tables S1 and S2 report the flow of data inclusion/exclusion and numbers in each model. To account for missing data in the multivariable models, multiple imputation using chained equations were performed using mi impute in Stata for all variables with missing data, and results were obtained using Rubin rules to combine 10 imputed datasets. We undertook a further complete case sensitivity analysis where only individuals without missing data for any of the variables used across all models were included.
Analyses were performed using Stata; results are reported with 99% confidence intervals (CIs) and a two-sided p-value <0.01 was considered statistically significant to account for multiple testing in models. Descriptive data is reported as median (IQR) unless stated otherwise.

Results
Of the 502,539 participants, 455,626 from the relevant ethnic backgrounds were included. Baseline characteristics are reported in Table 1. Median (IQR) age was higher in WE individuals (58 years [50, 63]) vs. SA (52 [46,60]) and BAC individuals (50 [45,57]). Women represented a greater proportion of the WE (55.9% women) and BAC (57.7%) samples, compared to the SA cohort (48.6%). Table 1 displays the lifestyle, environmental, sociodemographic and traditional clinical risk factors for each ethnicity. BMI was higher among BAC than WE and SA individuals. SA and BAC individuals reported similar levels of physical activity, which was lower than WE individuals. SA individuals also reported the lowest levels of handgrip strength and brisk walking pace compared to both WE and BAC individuals. Whereas, WE and BAC individuals reported the highest levels of TV viewing. Conversely, SA individuals reported eating the least red and processed meat and the most fruit and vegetables, whilst also being more likely to report never drinking alcohol or being a current or previous smoker. Both minority ethnic groups had greater exposure to noise and air pollution and greater levels of deprivation compared to WE individuals, with levels tending to highest in BAC individuals. SA and BAC individuals had similar levels of HbA1c that were 2.9 mmol/mol (0.2% DCCT unit) higher than WE individuals. SA individuals also reported 0.2 and 0.7 mmol/l higher triglycerides levels and 0.3 and 0.6 higher cholesterol-HDL ratio than WE and BAC individuals, respectively, whilst additionally reporting 0.1 mmol/l higher glucose levels than BAC individuals, with WE individuals reporting similar glucose levels to SA individuals. SA individuals also displayed 0.2 mmol/l lower HDLcholesterol levels than both WE and BAC individuals. Creactive protein in SA individuals was 0.3 mg/l and 0.2 mg/ l higher than WE and BAC individuals, respectively. Systolic blood pressure was 3 mmHg lower in SA individuals compared to WE individuals, with no differences observed between WE and BAC individuals. SA and WE individuals had similar levels of body fat percentage that were 2.3e2.4% lower than BAC individuals. The number of events across each adjustment model is shown in Table S2. Ethnicity was associated with 3P-MACE across all sociodemographic, lifestyle, environmental and clinical models, with the risk being higher in SA individuals than WE (unadjusted HR: 1.28; 99% CI: 1.16, 1.43) ( Fig. 1; Table S3). Results remained largely unchanged after adjustment for sociodemographic, lifestyle, environmental and traditional clinical risk factors (Fig. 1) Fig. S2) however, the elevated risk in SA individuals was still found to be independent of all individual risk factors.

Cardiovascular events
In the unadjusted model, BAC individuals were at lower risk of 3P-MACE compared to WE individuals. Such lower risk was largely attenuated in the majority of adjusted models, except for the clinical model where the lower risk in BAC individuals was independent of traditional clinical risk factors (HR 0.76; 0.60, 0.96) ( Fig. 1; Table S3). While the risk of 3P-MACE was found to be non-significant (continued on next page) Ethnic differences in fatal and non-fatal cardiovascular events between BAC and WE ethnic groups in sociodemographic, lifestyle, environmental and combined models (Fig. 1), the general patterns of all the models were similar and generally report a lower risk for BAC individuals compared to WE individuals. Results from the multiple imputation (Table S5; Fig. S3) and complete case (Table S6; Fig. S4) sensitivity analyses were found to be similar to that of the main results, with no differences in patterns of association found between the sensitivity analyses and the main analysis.

Discussion
In a large contemporary UK population SA individuals had a higher risk of fatal and non-fatal CVD than WE individuals, which was independent of differences in a wide range of sociodemographic, lifestyle, environmental and clinical factors. BAC individuals, in contrast to SA individuals, did not appear to have a different risk profile for cardiovascular morbidity or mortality than WEs in fully adjusted models.
Our finding of an elevated risk of CVD events in a UK SA diaspora compared to WE individuals is consistent with previous research from developed countries [40,41]. However, as far as we are aware, this is the first study to examine the extent to which this elevated risk profile is independent of differences in sociodemographic, lifestyle, environmental and traditional clinical risk factors. Whilst large disparities in lifestyle behaviours and environmental risk factors were observed between ethnic groups, these differences did not meaningfully confound the elevated risk of CVD in SA individuals. This finding is in contrast to other studies that have investigated, using a similar methodology to the present study, the impact of individual measures of physical activity and fitness, which have been shown to explain a substantial proportion of the increased insulin resistance and risk of coronary heart disease mortality found in SA individuals [23,32]. However, whilst our study confirmed that levels of physical activity, function and strength were lowest in SA individuals, their dietary, alcohol and smoking choices were healthier. This is consistent with previous research [20] and may be related to dietary choices for cultural or religious reasons [42]. Interestingly, risk of fatal and non-fatal CVD was also found to be independent of BMI in SA individuals. Previous evidence has shown that adiposity is an important contributor to CVD in SA individuals [3]. However, the measurement of adiposity in our study may not fully capture the most suitable method to measure adiposity in minority ethnic groups and further examination of the relationship between measures of adiposity and CVD risk between ethnic groups is required. Therefore when considered overall, lifestyle did not confound the adverse fatal and non-fatal CVD event profile in SA individuals.
The higher risk of CVD in SA compared to WE individuals was also independent of traditional clinical markers. This is an important finding given levels of HbA1c, triglycerides, C-reactive protein and total-to-HDL cholesterol ratio were all higher in SA participants. If confirmed to be causal, these findings suggest that aggressive management of risk factors in SAs communities, whilst reducing overall risk, may not by itself fully eliminate the relatively higher risk of CVD within these populations [43,44]. Instead, a more holistic approach to CVD risk reduction may have to be considered, with a focus not only management of risk factors through pharmacological and lifestyle measures, but wider determinants and earlier intervention too.
BAC individuals, in contrast to SA individuals, did not have an elevated risk of CVD compared to WE individuals. This contrasts with some previous work that demonstrates that African or Caribbean individuals are at increased CVD and, particularly, stroke risk [2,31]. However, this ethnic group may also be at lower risk of coronary/ischaemic heart disease [31], which may partially explain our results. Recent population-level analysis of the leading causes of mortality in the UK by ethnic group found that all BAC individuals had lower rates of ischaemic heart disease (e.g. MI) than white individuals. Rates of cerebrovascular diseases (e.g. stroke) were found to be lower in black African individuals than white, while rates in black Caribbean individuals were found to be similar or higher in women and men, respectively [4]. These results may further explain our findings. Our study also found that BAC individuals report some favourable biochemical cardiometabolic risk factors compared to WE and SA individuals, predominantly in relation to lipid based risk factors, such as triglycerides and HDL-cholesterol, which corroborates previous research [31]. Growing research has suggested that environmental factors, particularly air pollution, is closely associated with cardiometabolic health [24]. This study shows that both minority ethnic groups were exposed to worse air and noise pollution levels than WE individuals, with BAC individuals reporting the worst levels of air and noise pollution. This corroborates previous research showing black and Asian individuals were exposed to more PM 2.5 , PM 10 and NO 2 in London [45] and the Netherlands after adjustment for deprivation [27]. However, this study suggests that the more detrimental environmental conditions surrounding the homes of minority ethnic groups may not, by itself, confound the elevated CVD risk.
While a diverse range of factors were measured in this study, genetic and epigenetic factors cannot be discounted as potentially contributing to cardiometabolic risk between ethnicities. While several genes that potentially contribute to overall cardiometabolic risk are found in all ethnic groups, some of the susceptibility for certain genes attributing towards cardiometabolic disease are likely to be specific to certain populations [46]. However, it has been argued that genetic factors alone are unlikely to explain an increased risk of certain cardiometabolic diseases, such as type 2 diabetes, and rather that the interaction between environmental, lifestyle and genetic factors may be important [46].

Strengths and limitations
This study has important strengths, including the large sample size and large number of SA and BAC individuals, compared to previous research into ethnic health differences. The extensively phenotyped population also allowed for comprehensive investigation into a variety of different sociodemographic, lifestyle, environmental and clinical factors in a contemporary population. Nonetheless, important limitations remain. We recognise that this sample is not fully representative of modern Britain, with the number of individuals from minority ethnic groups being lower than the general population (2011 UK Census ethnic breakdown: 86% white British or European; 5.3% SA; 3.4% BAC [47]). The ethnic composition of the population from this study more closely reflects the Census figures from 2001. Additionally, the health status of this sample has been noted to be healthier than the national average [48]. However, when compared to other nationally representative cohort studies, risk factor associations have been shown to be generalisable to the general population [49]. It is also acknowledged that the terms 'South Asian' and 'black African or Caribbean' cover a wide range of different cultures, languages and religions with genetic heterogeneity. Consequently, our results may not apply to all SA or BAC populations. The majority of the lifestyle factors were obtained by self-reported measures. Further, literature and methods for collecting data on certain variables, such as dietary factors, has moved on from the time of baseline data collection in UK Biobank (e.g. measuring individual parameters to instead focusing on dietary Model 0: unadjusted model. Model 1: age, sex, self-reported cancer illnesses, number of selfreported non-cancer illnesses, number of medications/treatment taken, social deprivation, employment status, qualifications and household income. Model 2: model 1 þ handgrip strength mean and walking pace. Model 3: model 1 þ MET score for leisure time physical activity, TV viewing, sleep duration, processed meat consumption, red meat score, fruit and vegetable score, milk type used, spread type used, alcohol intake, BMI and smoking status. Model 4: model 1 þ environmental nitrogen dioxide, particulate matter 2.5 and averaged 24-hr noise pollution. Model 5: model 1 þ HbA1c, glucose, HDL-cholesterol, cholesterol-to-HDL ratio, triglycerides, C-reactive protein and systolic blood pressure. patterns). In addition, the environmental factors were estimated at place of residence as individual monitoring of many thousand individuals is not feasible. These measurement limitations may have resulted in some residual confounding, potentially acting to dilute their relative importance in contributing to the ethnic risk of cardiovascular disease. Individual chronic morbidity data was unavailable in this analysis due to low event numbers in the minority ethnic groups, therefore the impact that individual diseases had was unavailable. Further, we recognise that other criteria than our own can be used for the identification of possible confounders and their clustering and progressive inclusion in the regression model, which will have an impact on the change in the estimates. Finally some key factors that may affect health that are unique to minority ethnic groups, such as racial discrimination, poor language skills, access to healthcare and lack of opportunity for social mobility were unavailable and may therefore have acted as further confounders of the observed differences in health status and incident CVD events.

Conclusion
To conclude, our data indicate that SA individuals have an increased risk of cardiovascular morbidity and mortality compared to WE and BAC individuals, and this higher risk was observed after accounting for a wide range of sociodemographic, lifestyle, environmental and clinical risk factors. This research therefore suggests addressing key disparities in sociodemographic, lifestyle, environmental and clinical factors between ethnic groups, whilst of vital importance for social reasons, may not by themselves fully address the inequalities seen in CVD health status. Further epidemiological and experimental research is needed to confirm these conclusions and their implications for public health policies.

Funding
This report is independent research funded by the NIHR. The research was supported by the NIHR Leicester BRC which is a partnership between University Hospitals of Leicester NHS Trust, Loughborough University and the University of Leicester, and the NIHR ARCeEM. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR, Leicester BRC or the Department of Health and Social Care.

Author contribution statement
CR completed data acquisition, analysed the data and drafted the manuscript; TY assisted with data acquisition, data analysis and interpretation of data; FZ assisted with data acquisition and data analysis. JM assisted with data acquisition. TY, KK, MJD, FZ, JM and AH provided critical revision of the manuscript for important intellectual content; all the authors approved the manuscript for publication. CR will act as the guarantor for this work.