Table 1 List of sensitivity analyses

1. In addition to the covariates identified by DAG, we included other covariates based on prior evidence of likely confounders such as chronic obstructive pulmonary disease, other respiratory diseases, cancer, immunosuppression, chronic kidney disease, general practice attendance rate in the year prior to cohort entry, and A&E attendance rate in the year prior to cohort entry in the fully adjusted models (stratified by general practice)

To test the robustness of the covariate selection

2. Additionally adjusted for ethnicity in DAG and fully adjusted models. In the fully adjusted models, additional covariates included chronic obstructive pulmonary disease, other respiratory diseases (not including asthma), cancer, immunosuppression, chronic kidney disease, General Practice attendance rate in the year prior to cohort entry, and Accident and Emergency attendance rate in the year prior to cohort entry

In the main analysis, we did not adjust for ethnicity as a sizable proportion of individuals with missing ethnicity (~23%). We undertook complete case analysis to address missing data

3. Repeated main analysis excluding people prescribed antiplatelets 4 months before study start date

To explore the impact of use of antiplatelet which can reduce the risk of blood clots

4. Repeated main analysis excluding people who were prescribed both warfarin and DOACs on the day of the latest OAC prescription

To assess the sensitivity of exposure definition

5. Repeated main analysis excluding people who ever had warfarin prescription 4 months before study start date in the DOAC group

As warfarin is hypothesised to have harmful effect on severe COVID-19 compared with DOAC, this analysis was to assess the sensitivity of exposure definition

6. Time-updated the OAC exposure variable

To evaluate the impact of national recommendation on drug switching from warfarin to DOACs due to COVID-19 pandemic [24]