Journal of Public Health | pp. 1–10 | doi:10.

1093/pubmed/fdaa095

Greater risk of severe COVID-19 in Black, Asian and Minority

Downloaded from [Link] by PYEONGTAEK UNIVERSITY user on 27 July 2020

Ethnic populations is not explained by cardiometabolic,
socioeconomic or behavioural factors, or by 25(OH)-vitamin D
status: study of 1326 cases from the UK Biobank
Zahra Raisi-Estabragh1,2 , Celeste McCracken1 , Mae S. Bethell3 , Jackie Cooper1 ,
Cyrus Cooper4,5,6 , Mark J. Caulfield1 , Patricia B. Munroe1 , Nicholas C. Harvey4,5 , and
Steffen E. Petersen1,2
1
William Harvey Research Institute, NIHR Barts Biomedical Research Centre, Queen Mary University of London, London, UK
2
Barts Heart Centre, St Bartholomew’s Hospital, Barts Health NHS Trust, London, UK
3
North West Anglia NHS Foundation Trust, Hinchingbrooke Hospital, Huntingdon, UK
4
MRC Lifecourse Epidemiology Unit, University of Southampton, Southampton, UK
5
NIHR Southampton Biomedical Research Centre, University of Southampton and University Hospital Southampton NHS Foundation Trust, Southampton, UK
6
NIHR Biomedical Research Centre, University of Oxford, Oxford, UK
Address correspondence to Zahra Raisi-Estabragh, E-mail: zahraraisi@[Link].

ABSTRACT

Background We examined whether the greater severity of coronavirus disease 2019 (COVID-19) amongst men and Black, Asian and Minority
Ethnic (BAME) individuals is explained by cardiometabolic, socio-economic or behavioural factors.

Methods We studied 4510 UK Biobank participants tested for COVID-19 (positive, n = 1326). Multivariate logistic regression models including
age, sex and ethnicity were used to test whether addition of (1) cardiometabolic factors [diabetes, hypertension, high cholesterol, prior
myocardial infarction, smoking and body mass index (BMI)]; (2) 25(OH)-vitamin D; (3) poor diet; (4) Townsend deprivation score; (5) housing
(home type, overcrowding) or (6) behavioural factors (sociability, risk taking) attenuated sex/ethnicity associations with COVID-19 status.

Results There was over-representation of men and BAME ethnicities in the COVID-19 positive group. BAME individuals had, on average,
poorer cardiometabolic profile, lower 25(OH)-vitamin D, greater material deprivation, and were more likely to live in larger households and in
flats/apartments. Male sex, BAME ethnicity, higher BMI, higher Townsend deprivation score and household overcrowding were independently
associated with significantly greater odds of COVID-19. The pattern of association was consistent for men and women; cardiometabolic,
socio-demographic and behavioural factors did not attenuate sex/ethnicity associations.

Conclusions In this study, sex and ethnicity differential pattern of COVID-19 was not adequately explained by variations in cardiometabolic
factors, 25(OH)-vitamin D levels or socio-economic factors. Factors which underlie ethnic differences in COVID-19 may not be easily captured,
and so investigation of alternative biological and genetic susceptibilities as well as more comprehensive assessment of the complex economic,
social and behavioural differences should be prioritised.
Keywords communicable diseases, epidemiology, public health

Introduction Zahra Raisi-Estabragh, Dr. Cardiology registrar and BHF clinical research fellow
Celeste McCracken, Ms. Statistician and data scientist
The coronavirus disease 2019 (COVID-19) pandemic has to Mae S. Bethell, Dr. junior doctor (Foundation programme)
date resulted in over 6 million cases and 376 000 deaths world- Jackie Cooper, Ms. Senior Statistician
wide1 . Growing reports highlight men and Black, Asian and Cyrus Cooper, Director, Professor
Minority Ethnic (BAME) cohorts as at higher risk of adverse Mark J. Caulfield, Prof/Sir Co-Director

COVID-19 outcomes2,3 . Variations in cardiometabolic Patricia B. Munroe, Professor

disease burden4 , oestrogen pathway activity5 , vitamin D Nicholas C. Harvey, Professor

Steffen E. Petersen, Professor
levels6 and angiotensin-converting enzyme (ACE) 2 receptor

© The Author(s) 2020. Published by Oxford University Press on behalf of Faculty of Public Health.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License ([Link] which permits unrestricted reuse, 1
distribution, and reproduction in any medium, provided the original work is properly cited.
2 JOURNAL OF PUBLIC HEALTH

expression7 have been proposed as potential explanations ethnicity as White and BAME. For the latter we report
for the differential pattern of disease severity. Furthermore, breakdown of ethnicities as per existing UKB categories:
disparities in socio-economic standards, housing conditions, Black (Caribbean, African, any other Black background),

Downloaded from [Link] by PYEONGTAEK UNIVERSITY user on 27 July 2020

socialization habits and risk perception have potential impli- Asian (Indian, Pakistani, Bangladeshi, any other Asian
cations for risk of exposure and transmission. Understanding background), Chinese, Mixed (White and Black Caribbean,
the significance of these factors is urgently needed to inform White and Black African, White and Asian, any other mixed
public health and research efforts. background) and ‘other’. Townsend deprivation score is
We therefore investigated, in the UK Biobank (UKB) reported by the UKB as a measure of material deprivation
cohort, whether differential patterns of COVID-19 incidence calculated at baseline: zero, positive and negative scores
and severity by sex and ethnicity might be explained by correspond to average, higher and lower levels of deprivation,
cardiometabolic, socio-economic, lifestyle and behavioural respectively, relative to national averages12 . We used type of
exposures. housing as a binary variable comprising communal living
spaces (flat, apartment, sheltered accommodation) versus
stand-alone housing (house, bungalow). We considered
Methods household overcrowding based on self-report of household
size and intergenerational cohabitation. Socialization habits
Setting and study population
were defined per self-reports of frequency of family/friend
UKB is a prospective cohort study of over half a million
visits and participation in regular leisure activities outside
men and women from across the UK covering a range of
the home. Attitude to risk was assessed using self-report of
urban and rural settings. Recruitment was between 2006
tendency ‘to take risks’. Body mass index (BMI) was calculated
and 2010 through postal invite of individuals aged 40–
from height and weight recorded at baseline. Smoking
69 years old identified through National Health Service
status was based on self-report. Hypertension, diabetes and
(NHS) registers. All individuals living within 10 miles of one
hypercholesterolaemia were defined through cross-checking
of 22 UKB assessment centres were invited to participate.
across self-report and HES data. A list of ICD codes used
Individuals who were unable to consent were not recruited.
is available in Supplementary Table 2. Prior MI was obtained
Baseline assessment included detailed characterization of
from UKB algorithmically defined health outcomes. We used
socio-demographics, lifestyle, health, a series of physical
serum 25(OH)-vitamin D levels measured at baseline [Clinical
measures and blood biochemistry. The protocol is publicly
Laboratory Improvement Amendments (CLIA) analysis on
available8 . Data linkage with Hospital Episode Statistics
a DiaSorin Ltd. LIASON XL], limiting to results between
(HES) enables prospective tracking of health outcomes for all
10 and 375 nmol/L based on the manufacturer’s analytic
participants with conditions recorded according to interna-
range13 . We adjusted for seasonality by regressing vitamin D
tional classification of disease (ICD). Incidence of key events,
on month of sampling as a factor; this allowed derivation of
such as myocardial infarction (MI), is algorithmically defined
vitamin D adjusted to the same month for each participant.
by cross-checking over multiple data sources9 . Linkage with
There were differences in vitamin D levels and degree
Public Health England has enabled rapid release of linked
of seasonal variation by ethnicity (Fig. 1D). We therefore
COVID-19 test results of UKB participants to researchers10 .
performed seasonality adjustment separately for White and
The latest data release (29 May 2020) included test results
BAME populations and added the intercept to the adjusted
from 16 March 2020 to 18 May 2020. As UK testing during
variables to maintain the difference between the two groups.
this period was almost entirely restricted to hospitalized
We considered processed meat intake as a marker of poor diet
patients, researchers have been advised that COVID-19
quality. We converted self-reported weekly intake frequencies
positive status can be taken as surrogate for severe disease11 .
into probabilities of daily intake and multiplied by portion
size to derive a continuous measure of daily consumption in
Exposures grams, as previously published using this dataset14,15 .
We considered relevant demographic (age, sex, ethnicity),
biological (cardiometabolic, 25(OH)-vitamin D status), socio-
economic (material deprivation, type of home, household Ethics
overcrowding, poor diet quality) and behavioural (sociability, This study was covered by the ethics approval for UKB studies
attitude to risk) exposures (Supplementary Table 1). from the NHS National Research Ethics Service on 17 June
We used age and sex as recorded at baseline. For 2011 (Ref 11/NW/0382) and extended on 10 May 2016 (Ref
consistency with wider UK classification, we document 16/NW/0274).
 SEX AND ETHNICITY DIFFERENTIAL PATTERN OF SEVERE COVID-19 3

Downloaded from [Link] by PYEONGTAEK UNIVERSITY user on 27 July 2020

Fig. 1 Baseline participant characteristics. Panel A: Male: Female split by COVID-19 status; Panel B: Percentage of participants from different BAME ethnicities
by COVID-19 status; Panel C: Townsend deprivation score by ethnicity and COVID-19 status; Panel D: Vitamin D levels by month of measurement stratified
by sex and ethnicity; Panel E: Cardiometabolic profile stratified by ethnicity; Panel F: Cardiometabolic profile stratified by sex.

Statistical analysis simultaneously. We defined a final model using variables noted

Statistical analysis was performed using R Version 3.6.2 to be important from previous model permutations. We tested
[R Core Team (2019). R: A language and environment for for multicollinearity setting a variance inflation factor (VIF)
statistical computing. R Foundation for Statistical Computing, cut-off of 2.5. We present odds ratio (OR) for each exposure
Vienna, Austria. URL [Link] and with the corresponding 95% confidence interval (CI) and
RStudio Version 1.2.5019 [RStudio Team (2015). RStudio: P-value.
Integrated Development for R. RStudio, Inc., Boston, MA
URL [Link] RESULTS
UKB participants were grouped according to COVID-19
status: test positive, test negative and untested. In the analysis Population characteristics
of an earlier data release, we demonstrated similar associa- Sex and ethnicity
tions when comparing the untested cohort with both the test Test results for 4510 participants were available (positive,
negatives and test positives, suggesting that comparison with n = 1326; negative, n = 3184). Baseline characteristics are sum-
the whole cohort reveals associations with general hospitaliza- marized in Table 1. Comparisons with the untested cohort
tion rather than specifically with COVID-1916 . Therefore, to (n = 497 996) and characteristics by sex and ethnicity are
avoid bias relating to hospitalization, in the present study, we summarized in Supplementary Tables 3, 4 and 5. There was
limited to modelling within the tested cohort. We performed over-representation of men and BAME ethnicities in the
analyses in the whole tested sample, and separately in men and test positive cohort (Fig. 1A and B). Individuals of Black and
women. Logistic regression models were first used to examine Asian ethnicity were most disproportionately affected with
univariate associations. We then undertook individual multi- Black ethnicities contributing over 3.5× the number of pos-
variate models for each hypothesis to minimize loss of par- itive cases than their representation in the untested cohort
ticipants due to missingness from adding multiple variables (Supplementary Table 3, Fig. 1B).
4 JOURNAL OF PUBLIC HEALTH

Table 1 Baseline demographics by COVID-19 status

Test positive (n = 1326) Test negative (n = 3184)

Downloaded from [Link] by PYEONGTAEK UNIVERSITY user on 27 July 2020

Men 696 (52.5%) 1,505 (47.3%)
Age 68.11 (± 9.23) 68.91 (± 8.72)
White ethnicity 1,141 (86.0%) 2,927 (91.9%)
BAME ethnicities (total) 174 (13.1%) 241 (7.6%)
Black ethnicity 76 (5.7%) 91 (2.9%)
Asian ethnicity 60 (4.5%) 78 (2.4%)
Chinese ethnicity 6 (0.5%) 3 (0.1%)
Mixed ethnicity 9 (0.7%) 24 (0.8%)
Other ethnicity∗ 34 (2.6%) 61 (1.9%)
Smoking (current or previous) 683 (51.5%) 1,653 (51.9%)
Processed meat intake (g/day) 17.08 (± 15.67) 16.33 (± 15.00)
BMI (kg/m2 ) 28.04 [± 6.47] 27.41 [± 6.37]
Diabetes 217 (16.4%) 449 (14.1%)
Hypertension 624 (47.1%) 1,457 (45.8%)
High cholesterol 437 (33.0%) 1,034 (32.5%)
Prior MI 96 (7.2%) 242 (7.6%)
Vitamin D (nmol/L)∗∗ 33.88 [± 27.01] 35.45 [± 26.78]
Townsend deprivation score −0.91 [± 5.34] −1.55 [± 5.00]
Home type (flat/apartment) 191 (14.4%) 455 (14.3%)
Household size 2.50 (± 1.31) 2.32 (± 1.22)
Number of generations in household 1.41 (± 0.52) 1.35 (± 0.50)
Family/friend visits 975 (73.5%) 2,438 (76.6%)
Regular leisure activity 897 (67.6%) 2,124 (66.7%)
Tendency to take risks 404 (30.5%) 916 (28.8%)

Results are number (percentage) for categorical and mean (standard deviation) or median [interquartile range] for continuous variables.
∗ Ethnicity was missing for n = 11 test positive and n = 16 test negative participants, these participants are included as part of ‘other ethnicity’ in this table
but have been excluded from subsequent modelling.
∗∗ Vitamin D has been adjusted for seasonality.

Cardiometabolic factors and vitamin D tendency to risk-taking behaviour in the test positive cohort,
Men and BAME ethnicities had overall greater burden of which was greater in men versus women and in BAME versus
cardiometabolic morbidities compared to women and White White ethnicities.
cohorts, respectively (Fig. 1E and F). Serum 25(OH)-vitamin
D levels were, on average, higher in White ethnicities than Univariate associations of exposures with
BAME cohorts (Fig. 1D). COVID-19 positive status
We tested the univariate association of all defined expo-
sures with COVID-19 positive status within the tested
Socio-demographic and behavioural factors cohort (Supplementary Table 6). Male sex, BAME ethnicity,
In comparison to the test negatives, those with a positive higher BMI, greater material deprivation and greater house-
test had greater levels of material deprivation and were more hold overcrowding (household size, generations in house-
likely to live in crowded households (Fig. 1C). BAME indi- hold) were associated with increased odds of COVID-19
viduals had, on average, higher levels of material deprivation positive test. More frequent visits from family/friends were
by Townsend score compared to those of White ethnicity associated with lower risk of COVID-19 hospitalization,
(Supplementary Table 4). The frequency of family/friend vis- perhaps reflecting the role of social support in enabling
its and leisure activities outside the home was similar between individuals to remain at home when ill (given that a positive
the test positive and test negative groups. There was greater test implied hospital attendance). There was a negative
 SEX AND ETHNICITY DIFFERENTIAL PATTERN OF SEVERE COVID-19 5

Table 2 Multivariate logistic regression models testing the role of cardiometabolic factors (Model 1), vitamin D (Model 2) and poor diet (Model 3) in
determining risk of COVID-19

Downloaded from [Link] by PYEONGTAEK UNIVERSITY user on 27 July 2020

Exposures Whole tested sample Men n = 2201 Women n = 2309
n = 4510

Model 1: sex, age, ethnicity, Male sex 1.28∗ [1.12, 1.46] – –

smoking, BMI, diabetes, 4.05×10−4 – –
hypertension, high Age 0.99∗ [0.98, 1.00] 1.00 [0.98, 1.01] 0.99∗ [0.97, 1.00]
cholesterol, prior MI 0.0157 0.5128 0.0097
BAME ethnicity 1.78∗ [1.43, 2.20] 2.07∗ [1.50, 2.84] 1.55∗ [1.15, 2.09]
1.88×10−7 7.90×10−6 0.0040
Smoking (previous/current) 1.02 [0.89, 1.16] 1.12 [0.92, 1.36] 0.91 [0.75, 1.10]
0.7961 0.2533 0.3352
BMI (kg/m2 ) 1.02∗ [1.01, 1.03] 1.03∗ [1.01, 1.05] 1.02 [1.00, 1.03]
0.0015 0.0051 0.0537
Diabetes 1.08 [0.88, 1.32] 1.06 [0.82, 1.38] 1.08 [0.77, 1.49]
0.4781 0.6529 0.6665
Hypertension 1.01 [0.86, 1.18] 0.93 [0.74, 1.16] 1.11 [0.89, 1.40]
0.8875 0.5004 0.3563
High cholesterol 0.97 [0.82, 1.15] 1.04 [0.83, 1.31] 0.89 [0.68, 1.15]
0.7479 0.7108 0.3690
Prior MI 0.89 [0.68, 1.16] 0.85 [0.62, 1.15] 0.97 [0.55, 1.65]
0.4041 0.2961 0.8990
Model 2: sex, age, ethnicity, Male sex 1.31∗ [1.14, 1.50] – –
vitamin D 1.85×10−4 – –
Age 0.99∗ [0.98, 1.00] 1.00 [0.99, 1.01] 0.99∗ [0.97, 1.00]
0.0166 0.5500 0.0073
BAME ethnicity 1.77∗ [1.41, 2.22] 2.02∗ [1.45, 2.82] 1.60∗ [1.16, 2.18]
9.27×10−7 3.51×10−5 0.0038
Vitamin D 1.00 [1.00, 1.00] 1.00 [1.00, 1.01] 1.00 [1.00, 1.01]
0.7185 0.7464 0.9288
Model 3: sex, age, ethnicity, Male sex 1.26∗ [1.10, 1.44]
processed meat 8.55×10−4
Age 0.99∗ [0.98, 1.00] 1.00 [0.99, 1.01] 0.99∗ [0.98, 1.00]
0.0144 0.4993 0.0082
BAME ethnicity 1.81∗ [1.46, 2.24] 2.08∗ [1.52, 2.85] 1.62∗ [1.21, 2.17]
4.18×10−8 4.95×10−6 0.0011
Processed meat intake (100 1.26 [0.81, 1.94] 1.01 [0.57, 1.77] 1.83 [0.91, 3.66]
grams/day) 0.3032 0.9742 0.0871

Results are ORs, 95% CI and P-values for each exposure from three separate models (1, 2 and 3). Exposures are mutually adjusted.

association between age and COVID-19 positivity, which Independent associations of specific exposures
may reflect the narrow range and distribution of ages in with COVID-19 status
the sample. Testing separately in men, BAME ethnicity, Cardiometabolic factors
greater material deprivation and higher BMI were the only We undertook multivariate logistic regression models incor-
statistically significant exposures. For women, additionally, porating sex, age, ethnicity, smoking, BMI, diabetes, hyperten-
lower 25(OH)-vitamin D status, greater household over- sion, high cholesterol and prior MI (Table 2, Model 1). Male
crowding (household size, generations in household) and sex and BAME ethnicity were associated with greater odds
greater risk-taking behaviour were associated with COVID-19 of COVID-19 positive status with OR 1.28 (1.12, 1.46) and
positivity. 1.78 (1.43, 2.20), respectively. Every 1 kg/m2 of BMI was
6 JOURNAL OF PUBLIC HEALTH

Table 3 Multivariate logistic regression models testing the role of material deprivation (Model 4), housing conditions (Model 5) and final model (Model 6)
in determining risk of COVID-19

Downloaded from [Link] by PYEONGTAEK UNIVERSITY user on 27 July 2020

Exposures Whole tested sample Men n = 2201 Women n = 2309
n = 4510

Model 4: sex, age, ethnicity, Male sex 1.27∗ [1.11, 1.45] – –

Townsend deprivation score 3.87×10−4 – –
Age 0.99∗ [0.98, 1.00] 1.00 [0.99, 1.01] 0.99∗ [0.98, 1.00]
0.0222 0.6323 0.0089
BAME ethnicity 1.67∗ [1.34, 2.07] 1.92∗ [1.39, 2.64] 1.49∗ [1.11, 2.01]
3.94×10−6 6.15×10−5 0.0084
Townsend deprivation score 1.03∗ [1.01, 1.05] 1.03∗ [1.00, 1.06] 1.03∗ [1.00, 1.06]
0.0024 0.0402 0.0232
Model 5: sex, age, ethnicity, Male sex 1.24∗ [1.09, 1.42] – –
home type, household size∗ 0.0016 – –
Age 1.00 [0.99, 1.01] 1.00 [0.99, 1.01] 0.99 [0.98, 1.00]
0.3827 0.8207 0.1655
BAME ethnicity 1.73∗ [1.39, 2.17] 1.86∗ [1.33, 2.59] 1.66∗ [1.22, 2.24]
1.36×10−6 2.60×10−4 0.0011
Home type 0.98 [0.80, 1.20] 1.05 [0.80, 1.38] 0.90 [0.66, 1.22]
0.8650 0.7044 0.4918
Household size 1.08∗ [1.02, 1.14] 1.08 [0.99, 1.18] 1.07 [0.99, 1.16]
0.0140 0.0764 0.0941
Model 6 ‘final model’: sex, Male sex 1.23∗ [1.08, 1.41] – –
age, ethnicity, BMI, 0.0021 – –
Townsend deprivation score, Age 1.00 [0.99, 1.00] 1.00 [0.99, 1.01] 0.99 [0.98, 1.00]
household size 0.3648 0.8297 0.1674
BAME ethnicity 1.59∗ [1.26, 1.99] 1.74∗ [1.24, 2.45] 1.50∗ [1.10, 2.04]
7.85×10−5 0.0015 0.0105
BMI (kg/m2 ) 1.02∗ [1.01, 1.03] 1.03∗ [1.01, 1.05] 1.02∗ [1.00, 1.03]
9.71×10−4 0.0036 0.0476
Townsend deprivation score 1.03∗ [1.01, 1.06] 1.04∗ [1.01, 1.07] 1.03∗ [1.00, 1.06]
0.0011 0.0133 0.0319
Household size 1.09∗ [1.03, 1.16] 1.09 [1.00, 1.18] 1.10∗ [1.01, 1.19]
0.0022 0.0529 0.0203

Results are ORs, 95% CI and P-values for each exposure from three separate models (4, 5 and 6). Exposures are mutually adjusted.
∗ Initial analyses additionally included number of generations in household, however, we observed significant multicollinearity between this variable and
household size with higher VIF in the latter, hence it was removed from the final model.

associated with 1.03 (1.01, 1.04) greater odds of COVID- positivity (Table 2, Model 2). Similarly, in a separate model,
19 positivity. There was a borderline negative association with adjustment for sex, age and ethnicity demonstrated no sta-
age 0.99 [0.98, 1.00], which remained significant for women in tistically significant association between processed meat con-
sex-stratified analysis. There was no evidence of attenuation sumption and COVID-19 status (Table 2, Model 3). In both
(compared with the crude models) in the associations with models, male sex and BAME ethnicity were associated with
BAME ethnicity and higher BMI, consistent across men and higher odds of COVID-19 positive test across men and
women. women, with no evidence of attenuation.

25(OH)-vitamin D status and poor diet quality Material deprivation

In multivariate logistic regression models incorporating sex, We tested the effect of material deprivation in multivariate
age and ethnicity, there was no significant association between models with mutual adjustment for sex, age and ethnicity
season-adjusted 25(OH)-vitamin D status and COVID-19 (Table 3, Model 4). There was a small, but statistically sig-
 SEX AND ETHNICITY DIFFERENTIAL PATTERN OF SEVERE COVID-19 7

nificant association between greater material deprivation and not significantly associated with COVID-19 positivity and did
higher odds of COVID-19 positivity [OR 1.03 (1.01, 1.05)]. not explain the strong association with ethnicity. The pattern
There remained strong and significant associations with of associations did not vary between men and women.

Downloaded from [Link] by PYEONGTAEK UNIVERSITY user on 27 July 2020

male sex [OR 1.27 (1.11, 1.45)] and BAME ethnicity [OR
1.67 (1.34, 2.07)]. What is already known on this topic
Mounting evidence suggests disproportionate adverse effects
Housing conditions of COVID-19 in BAME populations2 . UK national audit
We considered the effect of housing conditions in multivariate data demonstrate that up to one-third of COVID-19 patients
logistic regression models including sex, age, ethnicity, home requiring intensive care are from BAME backgrounds, a rate
type and household size. In the whole sample, male sex, far greater than their representation in the general popula-
BAME ethnicity and greater household size were associated tion17 . An analysis of COVID-19 deaths amongst NHS staff,
with greater odds of COVID-19 positivity (Table 3, Model found that 64% of deaths were in BAME cohorts, markedly
5). Testing separately in men and women, BAME ethnicity disproportionate to their 20% contribution to the NHS work-
was the only exposure which remained significantly asso- force18 . The latest report from the Office of National Statis-
ciated with COVID-19 status. Attenuation of associations tics (ONS) also demonstrates greater risk of COVID-19
with household size is likely due to the small effect size and mortality in BAME groups19 ; individuals of Black ethnicity
limited heterogeneity of the exposure in each of the sexes had over 3.5× greater risk of COVID-19 death compared to
individually. Whites, followed by Asian ethnicities19 . Similarly, in the USA,
there has been growing concern over the disproportionate
number of COVID-19 deaths amongst African Americans20 .
Socialization habits and attitudes to risk
These patterns are echoed across Europe, with Nordic coun-
We undertook separate multivariate logistic regression models tries reporting as much as 10× greater risk of COVID-19
testing for associations between COVID-19 status, socializa- in Somali populations21 . We had previously documented this
tion habits and risk-taking attitude (Supplementary Table 7) preponderance of cases amongst BAME individuals in our
while adjusting for age, sex and ethnicity. Statistically signif- analysis of the initial UKB data release16 ; here, we have con-
icant associations were observed with male sex and BAME firmed the observation in this larger dataset, and importantly
ethnicity which were not attenuated from crude models by demonstrated a non-uniform impact across different BAME
adjustment for socialization or risk-taking attitude, which did groups with highest rates amongst Black followed by Asian
not show significant associations. ethnicities.
The greater cardiometabolic burden in both BAME and
Final model male cohorts has been proposed as potentially important
We built a final multivariate logistic regression model, with in driving adverse COVID-19 outcomes. In our analysis,
covariates selected based on previous model permutations cardiometabolic morbidities were not significantly associated
including sex, age, ethnicity, BMI, Townsend score and house- with COVID-19 status in multivariate models and did not
hold size (Table 3, Model 6). Male sex and BAME ethnicity attenuate sex and ethnicity associations. This suggests that
were associated with greater odds of COVID-19 positivity: the greater cardiometabolic burden in BAME individuals does
OR 1.23 (1.08, 1.41) and 1.59 (1.26, 1.99), respectively. Every not account for the adverse COVID-19 outcomes in this
1 kg/m2 increase in BMI was associated with 1.02 (1.01, 1.03) group.
greater odds of COVID-19 positivity and for every additional Consistent with our findings, data from the UK and the
person living in the same household the odds increased by USA highlight obesity as a marker of poor COVID-19 out-
1.09 (1.03, 1.16). comes, such as requirement for intensive care22 . There are
suggestions of a possible pathophysiological link between
adiposity and COVID-19 severity. Wide expression of ACE2
Discussion
receptors within adipose tissue is thought to promote binding
Main finding of this study and cellular entry of severe acute respiratory syndrome coro-
In 4510 UKB participants tested for COVID-19 in a hospital navirus 2 (SARS-CoV-2)23 . It has been suggested that adipose
setting, male sex, BAME ethnicity, higher BMI and greater tissue may act as a ‘viral reservoir’ thereby contributing to
household size were associated with significantly greater a more prolonged and severe illness23 . In addition, adipose
odds of a positive result. Despite variation in burden of tissue is a known source of inflammatory cytokines, such
cardiometabolic morbidities, 25(OH)-vitamin D levels and as Interleukin 624 . This is hypothesized to be linked to the
material deprivation by sex and ethnicity, these factors were association of adiposity with greater likelihood of cytokine
8 JOURNAL OF PUBLIC HEALTH

storms and the consequent risk of severe respiratory com- augmented risk in BAME populations is non-uniform and
plications in COVID-19. Indeed, studies have demonstrated disproportionately affects Black and Asian ethnicities. Higher
association of higher Interleukin 6 levels with respiratory BMI, greater material deprivation and household overcrowd-

Downloaded from [Link] by PYEONGTAEK UNIVERSITY user on 27 July 2020

failure and requirement for mechanical ventilation in COVID- ing are independent risk factors for COVID-19. The sex
19 patients25 . Greater adiposity, as well as BAME ethnicity, and ethnicity differential pattern of COVID-19 is not ade-
is associated with lower 25(OH)-vitamin D status. Although quately explained by variations in cardiometabolic factors,
the active 1,25(OH)2 -vitamin D form has immune system 25(OH)-vitamin D levels, socio-economic or behavioural fac-
functions26 , evidence linking low 25(OH)-vitamin D [the cir- tors. However, factors which underlie ethnic differences in
culating storage form, and poorly correlated with 1,25(OH)2 - COVID-19 may not be easily captured. Investigation of alter-
vitamin D] with COVID-19 disease have been contradic- native biological and genetic susceptibilities as well as more
tory27 . In our study, we found no independent associations comprehensive assessment of the complex economic, social
between 25(OH)-vitamin D status and COVID-19 disease, and behavioural differences is warranted.
suggesting that the relationship is confounded by ethnicity
and BMI. Interestingly, the BMI association was retained in
Limitations of this study
multivariate models, suggesting a possible independent role
Given the observational nature of the study, we cannot dis-
for adiposity, which clearly deserves further investigation.
cern causal relationships, and although we controlled for a
Socio-economic deprivation is associated with poorer
wide range of covariates, the possibility of residual confound-
global health outcomes28 . It has been suggested that ethnic
ing should be considered. The vitamin D levels used in this
differences in COVID-19 severity may relate to clustering
analysis are based on measurements taken at the UKB base-
of material deprivation with BAME status29 . In the UKB,
line visit; therefore, we cannot account for possible changes
material deprivation is reported using the Townsend score,
that may have occurred since this measurement was taken.
which is based on four factors—employment, car ownership,
However, there is evidence that vitamin D status tends to
home ownership and household overcrowding. Consistent
track with time, particularly after adjustment for season of
with national reports, we found higher material deprivation in
blood draw33, 34 (as we present in the current paper) and there
BAME individuals participants. In multivariate models includ-
is no reason to expect population level shifts in vitamin D
ing age, sex, ethnicity and Townsend score, there were signifi-
levels in this time period. Studies with more recent vitamin D
cantly greater odds of COVID-19 with greater material depri-
measures would be of interest. The limited age range in this
vation, while the association with ethnicity appeared strong
dataset precludes widely generalizable conclusions about the
and significant. Testing separately for the effect of household
effects of age, and there are clearly wider social, economic and
overcrowding, this exposure appeared significant indepen-
behavioural factors beyond those which we were able to study
dent of sex, ethnicity, age and home type. This suggests that it
in UK Biobank. Occupational factors may have relevance
may not be global economic deprivation, but specific aspects
in determining risk of exposure and viral transmission; this
relating to household overcrowding that has relevance to
topic requires detailed dedicated study. Aggregating all BAME
COVID-19. Consistent with these observations, a survey of
populations may overlook important differences between eth-
COVID-19 cases from New York reports the highest number
nicities; studies in samples with greater ethnic diversity are
of cases occurring in areas with the largest average household
needed.
size30 . Furthermore, analysis of UK cases by the ONS also
demonstrates that material deprivation does not adequately
explain the ethnic disparities in COVID-19 outcomes19 .
Funding statement
Behavioural factors, in particular attitudes that may com-
promise adherence to lockdown measures, have been pro- Z.R.E. is supported by a British Heart Foundation Clinical
posed as potentially important in determining risk of expo- Research Training Fellowship (FS/17/81/33318). S.E.P.,
sure to SARS-CoV-231,32 . In our analysis, we did not find P.B.M. and M.J.C. acknowledge support from the Barts
socialization habits and attitude to risk to be significantly Biomedical Research Centre funded by the National Institute
important in conferring COVID-19 positive status. for Health Research (NIHR). N.C.H. and C.C. acknowledge
support from the UK Medical Research Council, NIHR
Southampton Biomedical Research Centre, University of
What this study adds Southampton and University Hospital Southampton and
This study is consistent with growing reports of higher risk NIHR Oxford Biomedical Research Centre, University of
of severe COVID-19 in men and BAME populations. The Oxford.
 SEX AND ETHNICITY DIFFERENTIAL PATTERN OF SEVERE COVID-19 9

Acknowledgements 15 Anderson JJ, Darwis NDM, Mackay DF et al. Red and processed meat
consumption and breast cancer: UK Biobank cohort study and meta-
This study was undertaken using the UK Biobank resource, analysis. Eur J Cancer 2018;90:73–82.
Access Application 2964.

Downloaded from [Link] by PYEONGTAEK UNIVERSITY user on 27 July 2020

16 Raisi-Estabragh Z, McCracken C, Ardissino M et al. Non-white eth-
nicity, male sex, and higher body mass index, but not medications
acting on the renin-angiotensin system are associated with coronavirus
Conflicts of interest disease 2019 (COVID-19) hospitalisation: review of the first 669 cases
from the UK Biobank. medRxiv 2020. doi: 2020.05.10.20096925.
None.
17 ICNARC. ICNARC COVID-19 Study Case Mix Program. https://
[Link]/Our-Audit/Audits/Cmp/Reports (26 May 2020,
date last accessed).

References 18 Cook T, Kursumovic E, Lennane S. Exclusive: Deaths of NHS Staff

from COVID-19 Analysed . [Link]
1 World Health Organization (WHO). Coronavirus Disease (COVID-19) f-nhs-staff-from-covid-19-analysed/[Link] (26 May 2020,
Situation Report-134. [Link] date last accessed).
naviruse/situation-reports/[Link] (26
19 Office for National Statistics. Coronavirus (COVID-19) Related Deaths
May 2020, date last accessed).
by Ethnic Group, England and Wales: 2 March 2020 to 10 April 2020.
2 Kirby T. Evidence mounts on the disproportionate effect of [Link]
COVID-19 on ethnic minorities. Lancet Respir Med 2020;S2213-2600: thsandmarriages/deaths/articles/coronavirusrelateddeathsbyethni
30228–9. cgroupenglandandwales/2march2020to10april2020 (26 May 2020,
3 Wenham C, Smith J, Morgan R. COVID-19: the gendered impacts of date last accessed).
the outbreak. Lancet 2020;395:846–8. 20 City of Chicago. Latest Data. [Link]
4 Chaturvedi N. Ethnic differences in cardiovascular disease. Heart s/covid-19/home/latest-data/[Link] (24 May 2020, date
2003;89:681–6. last accessed).
5 Channappanavar R, Fett C, Mack M et al. Sex-based differences in sus- 21 UK Reuters. COVID-19 Takes Unequal Toll on Immigrants in Nordic
ceptibility to severe acute respiratory syndrome coronavirus infection. Region. [Link]
J Immunol 2017;198:4046–53. y-immigrants/covid-19-takes-unequal-toll-on-immigrants-in-nordi
6 Herrick KA, Storandt RJ, Afful J et al. Vitamin D status in the United c-region-idUKKCN2260Y2 (24 May 2020, date last accessed).
States, 2011–2014. Am J Clin Nutr 2019;110:150–7. 22 Kass DA, Duggal P, Cingolani O. Obesity could shift severe
7 Cao Y, Li L, Feng Z et al. Comparative genetic analysis of the novel COVID-19 disease to younger ages. Lancet 2020;395:1544–5.
coronavirus (2019-nCoV/SARS-CoV-2) receptor ACE2 in different 23 Kruglikov IL, Scherer PE. The role of adipocytes and adipocyte-
populations. Cell Discov 2020;6:1–4. like cells in the severity of COVID-19 infections. Obesity 2020. doi:
8 UK Biobank. Protocol for a Large-Scale Prospective Epidemiological Resource. 10.1002/oby.22856.
[Link] 24 Michalakis K, Ilias I. SARS-CoV-2 infection and obesity: com-
[Link] (13 May 2020, date last accessed). mon inflammatory and metabolic aspects. Diabetes Matab Syndr
9 Schnier C, Bush K, Nolan J, et al . Definitions of Acute Myocar- 2020;14:469–71.
dial Infarction and Main Myocardial Infarction Pathological Types UK 25 Herold T, Jurinovic V, Arnreich C et al. Level of IL-6 predicts respira-
Biobank Phase 1 Outcomes Adjudication Documentation on Behalf of tory failure in hospitalized symptomatic COVID-19 patients. medRxiv
UK Biobank Outcome Adjudication Group Definitions of Acute Myocar- 2020. doi: 2020.04.01.20047381.
dial Infarction. [Link] 26 Aranow C. Vitamin D and the immune system. J Investig Med
cs/alg_outcome_mi.pdf (26 May 2020, date last accessed). 2011;59:881–6.
10 Bugbank Homepage. Bugbank. [Link] 27 Lanham-New SA, Webb AR, Cashman KD et al. Vitamin D and
(14 May 2020, date last accessed). SARS-CoV-2 virus/COVID-19 disease. BMJ Nutr Prev Heal BMJ
11 UK Biobank Data Showcase. Records of COVID-19 Test Results 2020;bmjnph-2020-000089.
(data-field 40100). [Link] 28 Office for National Statistics. Health State Life Expectancies by National
d=40100 (14 May 2020, date last accessed). Deprivation Deciles, England and Wales: 2016 to 2018. [Link]
12 Townsend P, Phillimore P, Beattie A. Health and deprivation: inequal- [Link]/releases/healthstatelifeexpectanciesbynationaldeprivationdeci
ity and the north. Nurs Stand 1988;2:34–4. lesenglandandwales2016to2018 (26 May 2020, date last accessed).
13 Fry D, Almond R, Moffat S, et al . UK Biobank Biomarker Project Com- 29 [Link] Ethnicity Facts and Figures. People Living in Deprived
panion Document to Accompany Serum Biomarker Data. [Link] Neighbourhoods. [Link]
[Link]/uk-biobank-biomarker-panel/ (14 May 2020, date last k-population-by-ethnicity (24 May 2020, date last accessed).
accessed). 30 Wadhera RK, Wadhera P, Gaba P et al. Variation in COVID-19
14 Schenker S. Portion Sizes Food Fact Sheet. [Link]/foodfacts (25 hospitalizations and deaths across New York City boroughs. JAMA
April 2020, date last accessed). 2020. doi: 10.1001/jama.2020.7197.
10 JOURNAL OF PUBLIC HEALTH

31 Pawlowski B, Atwal R, Dunbar RIM. Evolutionary psychology sex 33 Jorde R, Sneve M, Hutchinson M et al. Tracking of serum 25-
differences in everyday risk-taking behavior in humans. Evol Psychol Hydroxyvitamin D levels during 14 years in a population-based study
2008;6:29–42. and during 12 months in an intervention study. Am J Epidemiol

Downloaded from [Link] by PYEONGTAEK UNIVERSITY user on 27 July 2020

32 Kivimäki M, Batty GD, Pentti J et al. Association between socioe- 2010;171(8):903–8.
conomic status and the development of mental and physical health 34 Moon RJ, Crozier SR, Dennison EM et al. Tracking of 25-
conditions in adulthood: a multi-cohort study. Lancet Public Heal hydroxyVitamin D status during pregnancy: the importance of vitamin
2020;5:e140–9. D supplementation. Am J Clin Nutr 2015;102(5):1081–7.