Association between oral anticoagulants and COVID-19

related outcomes: a population-based cohort study

Contents
Supplementary Information S1. Information Governance...................................................... 2
Supplementary Information S2. Deviations from pre-specified protocol, with reasons......... 3
Table S1. Demographic and Clinical Characteristics comparing current use in people with
atrial fibrillation with general population ............................................................................... 4
Figure S1. Directed Acyclic Graph investigating COVID-19 related outcomes comparing
current use of oral anticoagulants and non-use ....................................................................... 6
Figure S2. Time to COVID-19 outcomes in adjusted cumulative incidence plots ................. 7
Table S2. Demographic and Clinical Characteristics ............................................................. 9
Table S3. Results in the DAG and fully adjusted models comparing current use of oral
anticoagulants with non-use. ................................................................................................. 12
Table S4. Additionally adjusted for ethnicity in DAG models. ............................................ 13
Table S5. Excluded people prescribed antiplatelets 4 months before study start date. ........ 14
Table S6. Limited the study cohort to people who aged 55 or above. .................................. 15
Table S7. Subgroup analysis on positive COVID-test according to care home residence in
the DAG and fully adjusted models comparing current use with non-user. ......................... 16
Table S8. Bias-analyses for DAG-adjusted hazard ratios. .................................................... 17

1
Supplementary Information S1. Information Governance

NHS England is the data controller; TPP is the data processor; and the key researchers on
OpenSAFELY are acting on behalf of NHS England. This implementation of OpenSAFELY
is hosted within the TPP environment which is accredited to the ISO 27001 information
security standard and is NHS IG Toolkit compliant [1,2]; patient data has been
pseudonymised for analysis and linkage using industry standard cryptographic hashing
techniques; all pseudonymised datasets transmitted for linkage onto OpenSAFELY are
encrypted; access to the platform is via a virtual private network (VPN) connection, restricted
to a small group of researchers, their specific machine and IP address; the researchers hold
contracts with NHS England and only access the platform to initiate database queries and
statistical models; all database activity is logged; only aggregate statistical outputs leave the
platform environment following best practice for anonymisation of results such as statistical
disclosure control for low cell counts [3]. The OpenSAFELY research platform adheres to
the data protection principles of the UK Data Protection Act 2018 and the EU General Data
Protection Regulation (GDPR) 2016. In March 2020, the Secretary of State for Health and
Social Care used powers under the UK Health Service (Control of Patient Information)
Regulations 2002 (COPI) to require organisations to process confidential patient information
for the purposes of protecting public health, providing healthcare services to the public and
monitoring and managing the COVID-19 outbreak and incidents of exposure [4]. Taken
together, these provide the legal bases to link patient datasets on the OpenSAFELY platform.
GP practices, from which the primary care data are obtained, are required to share relevant
health information to support the public health response to the pandemic, and have been
informed of the OpenSAFELY analytics platform.

Reference:

1. BETA – Data Security Standards - NHS Digital. NHS Digital.

https://digital.nhs.uk/about-nhs-digital/our-work/nhs-digital-data-and-technology-
standards/framework/beta---data-security-standards. Accessed 30 Apr 2020.
2. Data Security and Protection Toolkit - NHS Digital. NHS Digital.
https://digital.nhs.uk/data-and-information/looking-after-information/data-security-
and-information-governance/data-security-and-protection-toolkit. Accessed 30 Apr
2020.
3. ISB1523: Anonymisation Standard for Publishing Health and Social Care Data - NHS
Digital. NHS Digital. https://digital.nhs.uk/data-and-information/information-
standards/information-standards-and-data-collections-including-
extractions/publications-and-notifications/standards-and-collections/isb1523-
anonymisation-standard-for-publishing-health-and-social-care-data. Accessed 30 Apr
2020.
4. Secretary of State for Health and Social Care - UK Government. Coronavirus
(COVID-19): notification to organisations to share information. 2020.
https://web.archive.org/web/20200421171727/https://www.gov.uk/government/public
ations/coronavirus-covid-19-notification-of-data-controllers-to-share-information.

2
Supplementary Information S2. Deviations from pre-specified protocol, with reasons

We noted that people who were prescribed and not prescribed an OAC could have important

clinical differences which might bias comparisons between them. To explore this as a pre-

specified analysis, we further matched each identified current OAC user to up to 10 people

from the general population (regardless of an AF diagnosis) based on age, sex, and general

practice on study start date. The same exclusion criteria applied to the AF population were

also applied. In addition, people who had no GP visit within one year before study start date

or had any OAC prescription within 4 months before study start date were excluded.

However, after matching, we observed very large differences in baseline characteristics

between the OAC exposed group and the comparison group from general population (see

Table S1), particularly on the proportions of ischaemic stroke and venous thromboembolism

which are components of CHA₂DS₂-VASc score. This is because people with a CHA₂DS₂-

VASc of 2 would tend to have a low prevalence of VTE and stroke, since these comorbidities

would generally result in a CHA₂DS₂-VASc higher than 2, when age and other factors were

added in. Given these large, conditional differences between groups, we determined any

comparisons between them would be subject to substantial confounding bias and would

likely be uninterpretable. This analysis was therefore not considered further in this study.

3
Table S1. Demographic and Clinical Characteristics comparing current use in people
with atrial fibrillation with general population
Matched general population Current use
Total 510,931 51,834
Age as of 1st Mar2020
18-<40 1,237 (0.2) 125 (0.2)
40-<50 7,582 (1.5) 768 (1.5)
50-<60 46,243 (9.1) 4,670 (9.0)
60-<70 156,251 (30.6) 15,741 (30.4)
70-<80 223,483 (43.7) 22,468 (43.3)
80+ 76,135 (14.9) 8,062 (15.6)
Median, IQR 71 (66-75) 71 (66-75)
Sex
Female 107,313 (21.0) 10,825 (20.9)
Body mass index
<18.5 5,768 (1.1) 542 (1.0)
18.5-24.9 135,212 (26.5) 11,920 (23.0)
25-29.9 197,891 (38.7) 17,723 (34.2)
30-34.9 94,887 (18.6) 11,021 (21.3)
35-39.9 29,519 (5.8) 4,825 (9.3)
40+ 11,288 (2.2) 2,829 (5.5)
Missing 36,366 (7.1) 2,974 (5.7)
Ethnicity
White 378,636 (74.1) 38,119 (73.5)
Mixed 1,694 (0.3) 102 (0.2)
Asian/Asian British 10,989 (2.2) 509 (1.0)
Black 3,928 (0.8) 180 (0.3)
Other 3,062 (0.6) 221 (0.4)
Missing 112,622 (22.0) 12,703 (24.5)
Index of Multiple
Deprivation
1 (least deprived) 103,168 (20.2) 11,256 (21.7)
2 102,264 (20.0) 10,908 (21.0)
3 102,283 (20.0) 10,375 (20.0)
4 101,552 (19.9) 9,838 (19.0)
5 (most deprived) 101,664 (19.9) 9,457 (18.2)
Smoking status
Never 182,630 (35.7) 18,824 (36.3)
Former 272,961 (53.4) 29,077 (56.1)
Current 54,162 (10.6) 3,840 (7.4)
Missing 1,178 (0.2) 93 (0.2)
Hazardous alcohol use 55,833 (10.9) 6,428 (12.4)
Comorbidities
Hypertension 249,617 (48.9) 21,676 (41.8)
Heart Failure 15,683 (3.1) 5,603 (10.8)

4
 Myocardial infarction 32,497 (6.4) 1,616 (3.1)
Peripheral arterial disease 14,499 (2.8) 418 (0.8)
Stroke/transient ischaemic
attack 36,110 (7.1) 1,030 (2.0)
Venous thromboembolism 12,724 (2.5) 541 (1.0)
Diabetes
Controlled
(HbA1c < 58 mmols/mol) 77,670 (15.2) 4,266 (8.2)
Uncontrolled
(HbA1c ≥ 58 mmols/mol) 32,365 (6.3) 2,079 (4.0)
HbA1c not measured 1,344 (0.3) 70 (0.1)
COPD 44,554 (8.7) 5,162 (10.0)
Other respiratory diseases 16,656 (3.3) 2,066 (4.0)
Cancer 71,014 (13.9) 7,849 (15.1)
Immunosuppression 3,443 (0.7) 436 (0.8)
Chronic kidney disease 64,764 (12.7) 8,266 (15.9)
Primary care consultations
Median, IQR 6 (3-11) 10 (5-17)
Min, Max 1, 367 0, 198
A&E attendance
Median, IQR 0 (0-0) 0 (0-1)
Min, Max 0, 70 0, 54
Vaccination
Flu 335,715 (65.7) 39,123 (75.5)
Medications
Oestrogen/ oestrogen-like
drugs 3,607 (0.7) 307 (0.6)
Antiplatelets 78,663 (15.4) 1,658 (3.2)

5
Figure S1. Directed Acyclic Graph investigating COVID-19 related outcomes
comparing current use of oral anticoagulants and non-use

*During peer-review process, it was suggested that dementia could be one of the
confounders and should be introduced into DAG. Therefore, dementia was also included to
the DAG-adjusted model in addition to the prespecified covariates shown above.

6
Figure S2. Time to COVID-19 outcomes in adjusted cumulative incidence plots

(a) Time to being tested for SARS-CoV-2

(b) Time to testing positive for SARS-CoV-2

7
 (c) Time to COVID-19 related hospital admission

(d) Time to COVID-19 related deaths

These figures present cumulative incidence and mortality predicted from a Royston-Parmar model including all covariates from the fully-adjusted Cox model, with the baseline
hazard parametrized as a 2-degrees-of-freedom cubic spline for COVID-19 related hospital admissions, 3-degrees-of-freedom cubic spline for being tested for SARS-CoV-2,
and testing positive for SARS-CoV-2, and COVID-19 related deaths; predictions standardized to the covariate distribution of the exposure group.

8
Table S2. Demographic and Clinical Characteristics
Non-use of oral anticoagulants Current use of oral anticoagulants

Total 18,271 52,832

Age as of 1st Mar2020

18-<40 169 (0.9) 126 (0.2)

40-<50 625 (3.4) 777 (1.5)

50-<60 2,406 (13.2) 4,736 (9.0)

60-<70 6,231 (34.1) 15,983 (30.3)

70-<80 6,295 (34.5) 22,886 (43.3)

80+ 2,545 (13.9) 8,324 (15.8)

Median, IQR 69 (63-74) 71 (66-75)

Sex

Female 5,406 (29.6) 10,962 (20.7)

Body mass index

<18.5 275 (1.5) 556 (1.1)

18.5-24.9 4,818 (26.4) 12,147 (23.0)

25-29.9 6,002 (32.8) 18,100 (34.3)

30-34.9 3,349 (18.3) 11,223 (21.2)

35-39.9 1,441 (7.9) 4,903 (9.3)

40+ 773 (4.2) 2,876 (5.4)

Missing 1,613 (8.8) 3,027 (5.7)

Ethnicity

9
 White
13,258 (72.6) 38,921 (73.7)

Mixed 51 (0.3) 102 (0.2)

Asian/Asian British 329 (1.8) 515 (1.0)

Black 127 (0.7) 181 (0.3)

Other 92 (0.5) 221 (0.4)

Missing 4,414 (24.2) 12,892 (24.4)

Index of Multiple Deprivation

1 (least deprived) 3,764 (20.6) 11,436 (21.6)

2 3,762 (20.6) 11,116 (21.0)

3 3,672 (20.1) 10,631 (20.1)

4 3,599 (19.7) 10,041 (19.0)

5 (most deprived) 3,474 (19.0) 9,608 (18.2)

Smoking status

Never 6,992 (38.3) 19,167 (36.3)

Former 9,422 (51.6) 29,662 (56.1)

Current
1,803 (9.9) 3,909 (7.4)

Missing 54 (0.3) 94 (0.2)

Hazardous alcohol use 2,371 (13.0) 6,537 (12.4)

Care home residence 219 (1.2) 323 (0.6)

Comorbidities

Hypertension 7,622 (41.7) 22,061 (41.8)

Heart Failure 1,424 (7.8) 5,700 (10.8)

Myocardial infarction 769 (4.2) 1,640 (3.1)

Peripheral arterial disease 208 (1.1) 426 (0.8)

10
 Stroke/transient ischaemic attack 336 (1.8) 1,041 (2.0)

Venous thromboembolism 320 (1.8) 547 (1.0)

Diabetes

Controlled (HbA1c < 58 mmols/mol)

1,850 (10.1) 4,324 (8.2)

Uncontrolled (HbA1c ≥ 58 mmols/mol) 772 (4.2) 2,112 (4.0)

HbA1c not measured 60 (0.3) 72 (0.1)

COPD 1,548 (8.5) 5,267 (10.0)

Other respiratory diseases

754 (4.1) 2,105 (4.0)

Cancer 2,837 (15.5) 8,001 (15.1)

Immunosuppression
258 (1.4) 441 (0.8)

Chronic kidney disease

2,047 (11.2) 8,477 (16.0)

Dementia
518 (2.8) 1,217 (2.3)

Primary care consultations

Median, IQR 7 (3-12) 10 (5-16)

Min, Max 0, 202 0, 198

A&E attendance

Median, IQR 0 (0-0) 0 (0-1)

Min, Max 0, 39 0, 54

Flu vaccination 11,200 (61.3) 39,910 (75.5)

Medications

Oestrogen/ oestrogen-like drugs 221 (1.2) 308 (0.6)

Antiplatelets 3,538 (19.4) 1,691 (3.2)

Abbreviations: COPD, Chronic obstructive pulmonary disease

11
Table S3. Results in the DAG and fully adjusted models comparing current use of oral anticoagulants with non-use.
Total Rate
Number person- per
of events weeks 1,000 Unadjusted Age/Sex Adjusted DAG Adjusted* Fully adjusteda
HR 95% CI HR 95% CI HR 95% CI HR 95% CI
Tested for
SARS-CoV-2
1.00 1.00 1.00 1.00
non-use 2941 507282 5.8 (ref) (ref) (ref) (ref)
current use 8249 1486790 5.55 0.95 0.91 - 0.99 0.98 0.94 - 1.03 0.99 0.95 - 1.04 0.96 0.91 - 1.00
Testing positive for
SARS-CoV-2
1.00 1.00 1.00 1.00
non-use 151 538126 0.28 (ref) (ref) (ref) (ref)
current use 329 1569391 0.21 0.75 0.62 - 0.91 0.78 0.64 - 0.95 0.77 0.63 - 0.95 0.7 0.56 - 0.87
COVID-19 related
hospital admissionb
1.00 1.00 1.00 1.00
non-use 63 539242 0.12 (ref) (ref) (ref) (ref)
current use 163 1571612 0.1 0.89 0.67 - 1.19 0.88 0.65 - 1.18 0.85 0.62 - 1.15 0.76 0.55 - 1.06
COVID-19
deathb,c
1.00 1.00 1.00 1.00
non-use 55 540181 0.1 (ref) (ref) (ref) (ref)
current use 130 1573615 0.08 0.81 0.59 - 1.11 0.78 0.57 - 1.08 0.74 0.53 - 1.04 0.66 0.45 - 0.97
*Adjusted for age, sex, obesity, smoking, dementia, hypertension, heart failure, myocardial infarction, peripheral arterial disease, stroke/transient ischemic attack, venous thromboembolism, diabetes, flu vaccination, antiplatelet use,
oestrogen and oestrogen-like therapy use, and Index of Multiple Deprivation.
a
Adjusted for age, sex, obesity, smoking, dementia, hypertension, heart failure, myocardial infarction, peripheral arterial disease, stroke/transient ischemic attack, venous thromboembolism, diabetes, flu vaccination, antiplatelet use,
oestrogen and oestrogen-like therapy use, Index of Multiple Deprivation, chronic obstructive pulmonary disease, other respiratory diseases, cancer, immunosuppression, chronic kidney disease, general practice attendance and A&E
attendance and stratified on general practice.
b
For outcomes of COVID-19 related hospital admission and COVID-19 death, we classified people with a diabetes diagnosis but not having HbA1c measures in the past year as uncontrolled diabetes in DAG adjusted and fully adjusted
models.as the parameter for not having HbA1c measures did not converge and people with a diabetes diagnosis but not having HbA1c measures in the past year, are likely to have uncontrolled diabetes due to their potential lack of
monitoring and management of diabetes.
c
Due to low event count for parameters of stroke/transient ischaemic attack and oestrogen use, they did not converge in the model and were dropped from the main analysis.

12
Table S4. Additionally adjusted for ethnicity in DAG models.
Number Total
of person- Rate per DAG Adjusted with DAG Adjusted Fully Adjusted with Fully Adjusted
events weeks 1,000 Unadjusted Age/Sex Adjusted ethnicity* without ethnicitya ethnicityb without ethnicityc
HR 95% CI HR 95% CI HR 95% CI HR 95% CI HR 95% CI HR 95% CI
Tested for SARS-CoV-2
1.00 1.00 1.00 1.00 1.00 1.00
non-use 2248 385301 5.83 (ref) (ref) (ref) (ref) (ref) (ref)
current use 6263 1124422 5.57 0.95 0.91 - 1.00 0.98 0.94 - 1.03 0.99 0.94 - 1.05 0.99 0.94 - 1.05 0.95 0.90 - 1.00 0.95 0.90 - 1.00
Testing positive for
SARS-CoV-2e
1.00 1.00 1.00 1.00 1.00 1.00
non-use 109 409048 0.27 (ref) (ref) (ref) (ref) (ref) (ref)
current use 248 1188131 0.21 0.78 0.63 - 0.98 0.81 0.64 - 1.01 0.80 0.63 - 1.02 0.8 0.63 - 1.02 0.74 0.57 - 0.96 0.75 0.57 - 0.97
For outcomes of COVID-19 related hospital admission and COVID-19 death, parameters of ethnicity did not converge.
*Adjusted for age, sex, ethnicity, obesity, smoking, dementia, hypertension, heart failure, myocardial infarction, peripheral arterial disease, stroke/transient ischemic attack, venous thromboembolism, diabetes, flu vaccination, antiplatelet
use, oestrogen and oestrogen-like therapy use, and Index of Multiple Deprivation.
a
Adjusted for age, sex, obesity, smoking, dementia, hypertension, heart failure, myocardial infarction, peripheral arterial disease, stroke/transient ischemic attack, venous thromboembolism, diabetes, flu vaccination, antiplatelet use,
oestrogen and oestrogen-like therapy use, and Index of Multiple Deprivation.
b
Adjusted for age, sex, ethnicity, obesity, smoking, dementia, hypertension, heart failure, myocardial infarction, peripheral arterial disease, stroke/transient ischemic attack, venous thromboembolism, diabetes, flu vaccination, antiplatelet
use, oestrogen and oestrogen-like therapy use, Index of Multiple Deprivation, chronic obstructive pulmonary disease, other respiratory diseases, cancer, immunosuppression, chronic kidney disease, general practice attendance and A&E
attendance and stratified on general practice.
c
Adjusted for age, sex, obesity, smoking, dementia, hypertension, heart failure, myocardial infarction, peripheral arterial disease, stroke/transient ischemic attack, venous thromboembolism, diabetes, flu vaccination, antiplatelet use,
oestrogen and oestrogen-like therapy use, Index of Multiple Deprivation, chronic obstructive pulmonary disease, other respiratory diseases, cancer, immunosuppression, chronic kidney disease, general practice attendance and A&E
attendance and stratified on general practice.
d
We classified people with a diabetes diagnosis but not having HbA1c measures in the past year as uncontrolled diabetes in DAG adjusted and fully adjusted models.as the parameter for not having HbA1c measures did not converge and
people with a diabetes diagnosis but not having HbA1c measures in the past year, are likely to have uncontrolled diabetes due to their potential lack of monitoring and management of diabetes.

13
Table S5. Excluded people prescribed antiplatelets 4 months before study start date.
Unadjusted Age/Sex Adjusted DAG Adjusted* Fully adjusteda
HR 95% CI HR 95% CI HR 95% CI HR 95% CI
Tested for SARS-CoV-2
1.00 1.00 1.00 1.00
non-use (ref) (ref) (ref) (ref)
current use 0.93 0.89 - 0.98 0.97 0.92 - 1.01 0.96 0.92 - 1.01 0.93 0.88 - 0.97

Testing positive for

SARS-CoV-2
1.00 1.00 1.00 1.00
non-use (ref) (ref) (ref) (ref)
current use 0.75 0.61 - 0.92 0.78 0.63 - 0.96 0.8 0.64 - 0.99 0.74 0.58 - 0.93
COVID-19 related
hospital admissionb
1.00 1.00 1.00 1.00
non-use (ref) (ref) (ref) (ref)
current use 0.91 0.66 - 1.25 0.89 0.64 - 1.22 0.87 0.63 - 1.21 0.81 0.57 - 1.16

COVID-19 deathb,c
1.00 1.00 1.00 1.00
non-use (ref) (ref) (ref) (ref)
current use 0.78 0.56 - 1.09 0.75 0.53 - 1.05 0.75 0.53 - 1.06 0.69 0.47 - 1.03
*Adjusted for age, sex, obesity, smoking, dementia, hypertension, heart failure, myocardial infarction, peripheral arterial disease, stroke/transient ischemic attack, venous thromboembolism, diabetes, flu vaccination, oestrogen and oestrogen-
like therapy use, and Index of Multiple Deprivation.
a
Adjusted for age, sex, obesity, smoking, dementia, hypertension, heart failure, myocardial infarction, peripheral arterial disease, stroke/transient ischemic attack, venous thromboembolism, diabetes, flu vaccination, oestrogen and oestrogen-
like therapy use, Index of Multiple Deprivation, chronic obstructive pulmonary disease, other respiratory diseases, cancer, immunosuppression, chronic kidney disease, general practice attendance and A&E attendance and stratified on
general practice.
b
For outcomes of COVID-19 related hospital admission and COVID-19 death, we classified people with a diabetes diagnosis but not having HbA1c measures in the past year as uncontrolled diabetes in DAG adjusted and fully adjusted
models.as the parameter for not having HbA1c measures did not converge and people with a diabetes diagnosis but not having HbA1c measures in the past year, are likely to have uncontrolled diabetes due to their potential lack of
monitoring and management of diabetes.
c
Due to low event count for parameters of stroke/transient ischaemic attack, oestrogen use and peripheral artery disease, they did not converge in the model and were dropped from the main analysis.

14
Table S6. Limited the study cohort to people who aged 55 or above.
Unadjusted Age/Sex Adjusted DAG Adjusted* Fully adjusteda
HR 95% CI HR 95% CI HR 95% CI HR 95% CI
Tested for SARS-CoV-2
1.00 1.00 1.00 1.00
non-use (ref) (ref) (ref) (ref)
current use 0.96 0.92 - 1.01 0.98 0.94 - 1.03 1.00 0.95 - 1.04 0.96 0.92 - 1.01

Testing positive for

SARS-CoV-2
1.00 1.00 1.00 1.00
non-use (ref) (ref) (ref) (ref)
current use 0.73 0.60 - 0.89 0.76 0.62 - 0.93 0.77 0.62 - 0.95 0.71 0.56 - 0.89
COVID-19 related
hospital admission b
1.00 1.00 1.00 1.00
non-use (ref) (ref) (ref) (ref)
current use 0.86 0.64 - 1.17 0.86 0.64 - 1.17 0.84 0.61 - 1.16 0.78 0.55 - 1.09

COVID-19 deathb,c
1.00 1.00 1.00 1.00
non-use (ref) (ref) (ref) (ref)
current use 0.78 0.57 - 1.07 0.79 0.57 - 1.08 0.75 0.53 - 1.05 0.66 0.45 - 0.97
*Adjusted for age, sex, obesity, smoking, dementia, hypertension, heart failure, myocardial infarction, peripheral arterial disease, stroke/transient ischemic attack, venous thromboembolism, diabetes, flu vaccination, antiplatelet use,
oestrogen and oestrogen-like therapy use, and Index of Multiple Deprivation.
a
Adjusted for age, sex, obesity, smoking, dementia, hypertension, heart failure, myocardial infarction, peripheral arterial disease, stroke/transient ischemic attack, venous thromboembolism, diabetes, flu vaccination, antiplatelet use,
oestrogen and oestrogen-like therapy use, Index of Multiple Deprivation, chronic obstructive pulmonary disease, other respiratory diseases, cancer, immunosuppression, chronic kidney disease, general practice attendance and A&E
attendance and stratified on general practice.
b
For outcomes of COVID-19 related hospital admission and COVID-19 death, we classified people with a diabetes diagnosis but not having HbA1c measures in the past year as uncontrolled diabetes in DAG adjusted and fully adjusted
models.as the parameter for not having HbA1c measures did not converge and people with a diabetes diagnosis but not having HbA1c measures in the past year, are likely to have uncontrolled diabetes due to their potential lack of
monitoring and management of diabetes.
c
Due to low event count for parameters of stroke/transient ischaemic attack, and oestrogen use, they did not converge in the model and were dropped from the main analysis.

15
Table S7. Subgroup analysis on positive COVID-test according to care home residence in the DAG and fully adjusted models comparing current
use with non-user.
Total Rate
Number person- per
of events weeks 1,000 Unadjusted Age/Sex Adjusted DAG Adjusted* Fully adjusteda
HR 95% CI HR 95% CI HR 95% CI HR 95% CI
P(interaction)= P(interaction)= P(interaction)= P(interaction)=
0.491 0.352 0.416 0.716
Not living in care-home
non-use 132 533212 0.25 1.00 (ref) 1.00 (ref) 1.00 (ref) 1.00 (ref)
current use 300 1561506 0.19 0.78 0.63 - 0.95 0.80 0.65 - 0.98 0.78 0.62 - 0.96 0.72 0.57 - 0.91
Living in care-home
non-use 19 4883 3.89 1.00 (ref) 1.00 (ref) 1.00 (ref) 1.00 (ref)
current use 29 7884 3.68 0.96 0.54 - 1.72 1.06 0.60 - 1.90 1.00 0.56 - 1.79 0.62 0.29 - 1.33
*Adjusted for age, sex, obesity, smoking, dementia, hypertension, heart failure, myocardial infarction, peripheral arterial disease, stroke/transient ischemic attack, venous thromboembolism, diabetes, flu vaccination, antiplatelet use,
oestrogen and oestrogen-like therapy use, and Index of Multiple Deprivation.
a
Adjusted for age, sex, obesity, smoking, dementia, hypertension, heart failure, myocardial infarction, peripheral arterial disease, stroke/transient ischemic attack, venous thromboembolism, diabetes, flu vaccination, antiplatelet use,
oestrogen and oestrogen-like therapy use, Index of Multiple Deprivation, chronic obstructive pulmonary disease, other respiratory diseases, cancer, immunosuppression, chronic kidney disease, general practice attendance and A&E
attendance and stratified on general practice.

16
Table S8. Bias-analyses for DAG-adjusted hazard ratios.
Bias- Observed hazard ratio E-value Cornfield condition
adjusted Lower Upper
Point Lower CI Upper CI Point Lower CI Upper CI Point
hazard CI CI
estimate bound bound estimate bound bound estimate
ratio bound bound
Testing positive for
1 0.77 0.63 0.95 1.92 2.55 1.29 1.30 1.59 1.05
SARS-CoV-2
COVID-19 related
1 0.85 0.62 1.15 1.63 2.61 - 1.18 1.61 -
hospital admission
COVID-19 death 1 0.74 0.53 1.04 2.04 3.18 - 1.35 1.89 -
As a simplification, we consider higher-risk health behaviour to be a binary variable and assume no interaction between the unmeasured confounder and measured covariates on the outcome. To apply the bias analysis formulas to hazard
ratios, we assume the outcome is rare.

17