Evaluation of antithrombotic use and COVID-19 outcomes in a nationwide atrial fibrillation cohort

Objective To evaluate antithrombotic (AT) use in individuals with atrial fibrillation (AF) and at high risk of stroke (CHA2DS2-VASc score ≥2) and investigate whether pre-existing AT use may improve COVID-19 outcomes. Methods Individuals with AF and CHA2DS2-VASc score ≥2 on 1 January 2020 were identified using electronic health records for 56 million people in England and were followed up until 1 May 2021. Factors associated with pre-existing AT use were analysed using logistic regression. Differences in COVID-19-related hospitalisation and death were analysed using logistic and Cox regression in individuals with pre-existing AT use versus no AT use, anticoagulants (AC) versus antiplatelets (AP), and direct oral anticoagulants (DOACs) versus warfarin. Results From 972 971 individuals with AF (age 79 (±9.3), female 46.2%) and CHA2DS2-VASc score ≥2, 88.0% (n=856 336) had pre-existing AT use, 3.8% (n=37 418) had a COVID-19 hospitalisation and 2.2% (n=21 116) died, followed up to 1 May 2021. Factors associated with no AT use included comorbidities that may contraindicate AT use (liver disease and history of falls) and demographics (socioeconomic status and ethnicity). Pre-existing AT use was associated with lower odds of death (OR=0.92, 95% CI 0.87 to 0.96), but higher odds of hospitalisation (OR=1.20, 95% CI 1.15 to 1.26). AC versus AP was associated with lower odds of death (OR=0.93, 95% CI 0.87 to 0.98) and higher hospitalisation (OR=1.17, 95% CI 1.11 to 1.24). For DOACs versus warfarin, lower odds were observed for hospitalisation (OR=0.86, 95% CI 0.82 to 0.89) but not for death (OR=1.00, 95% CI 0.95 to 1.05). Conclusions Pre-existing AT use may be associated with lower odds of COVID-19 death and, while not evidence of causality, provides further incentive to improve AT coverage for eligible individuals with AF.


INTRODUCTION
Atrial fibrillation (AF) is a disturbance of heart rhythm affecting 37.5 million people globally 1 and significantly increases the risk of stroke. 2 Anticoagulants (AC), a subtype of antithrombotics (AT), reduce the risk of stroke 3 and are recommended for individuals with AF and at high risk of stroke (CHA 2 DS 2 -VASc score ≥2, the National Institute for Health and Care Excellence (NICE) threshold). 4 5 Despite improvements in AC uptake, previous evaluations suggest that up to one-third of individuals with AF and CHA 2 DS 2 -VASc score ≥2 in the UK may not be on AC, 6 with around 15% on no type of AT. 6 Hypotheses for this suboptimal medication centre around clinical overestimation of bleeding and fall risk in elderly patients, 6 7 but the potential drivers of AT use remain underexplored at the population scale.
COVID-19 has presented another risk factor for individuals with AF, who are at increased risk of poor outcomes if they become infected. 8 Observational evidence from Germany (n=6637) suggests that pre-existing AC use, but not antiplatelets (APanother subtype of AT), may reduce mortality in individuals hospitalised with COVID-19. 9 However, evidence is discordant, with a US study (n=3772) observing no difference in mortality in groups on AC or AP. 10 In the UK, a larger study (n=70 464 of 372 746) explored AC and AC subtypes (warfarin vs direct oral anticoagulants (DOACs)) in individuals with AF and observed that AC was associated with lower COVID-19-specific mortality. 11 This observational evidence is promising, but it does not compare all subtypes of AT and only covered the period up to 28 September 2020.
This study, therefore, set out to conduct the largest scale evaluation of AT use in individuals with AF to date in routinely updated, linked, population-scale electronic health record (EHR) data for 56 million people in England. 12 Using this statistical power, this study investigated what factors are associated with pre-existing AT use and whether pre-existing AT use (across subtypes) is associated with COVID-19-related hospitalisation and death.

Study design and data sources
We conducted a cohort analysis using the newly establishedNationalHealthService(NHS)Digital Trusted Research Environment for England, which provides secure, remote access to linked, personlevel EHR data for over 56 million people. 12 Available data sources cover primary care, secondary Cardiac risk factors and prevention care, pharmacy dispensing, death registrations and COVID-19 tests and vaccines. We used the General Practice Extraction Service Extract for Pandemic Planning and Research (GDPPR) for demographic and diagnostic data (eg, a diagnosis of AF) and the NHS Business Service Authority Dispensed Medicines (BSADM)formedicationexposuredata(eg,pre-existingATuse) as this is the most accurate available representation of the medication an individual takes. Hospital Episode Statistics (HES), COVID-19 Hospitalisations in England Surveillance System, Secondary Uses Service, and the Office for National Statistics (ONS) Civil Registration of Deaths were used for COVID-19 hospitalisation and death. Public Health England's Second Generation Surveillance System was used to identify COVID-19 test results, and the COVID-19 vaccination events data set was used for COVID-19 vaccine status.

Study populations
Individuals were included in the study if registered with a general practice (GP) in England (at least one record in the GDPPR data set with a valid person pseudo-identifier), ≥18 years old and alive on 1 January 2020, had available data on sex, ethnicity and GP location (based on the most recent, available data across primary care (GDPPR), secondary care (HES) and death registrations(ONS)),andhadadiagnosisofAF(codedinGDPPR)with a CHA 2 DS 2 -VAScscore≥2(calculatedfromthesumofcomponents 13 coded in GDPPR).
Individuals with contraindications to subtypes of AT (eg, DOACs in mitral stenosis, prosthetic mechanical valves, antiphospholipid antibody syndrome) were included as they are still eligible for other AT subtypes (eg, AP, warfarin).
To investigate exposure to pre-existing AT use on COVID-19-related hospitalisation and death, the inclusion criteria of a recorded COVID-19 event were applied. A COVID-19 event was defined as any positive test (PCR or lateral flow), a coded diagnosis in primary or secondary care, or a COVID-19 diagnosis on a death certificate (see Thygesen et al 14 andCALIBER 15 for further details and phenotyping algorithms).
All phenotyping algorithms used are available on GitHub (https://github.com/BHFDSC/CCU020/tree/main/england/ phenotypes) and online supplemental figure 1 provides a flow chart of individuals excluded at each stage.

Study variables Medication exposure
An individual was defined as taking a particular medication if theyhadoneormoredispensedprescription(codedintheNHS BSADM)intheprevious6months.Wepurposefullydefineda liberalthresholdtosupportevaluationofATusageuptoMay 2021 that may have included unusual buying patterns (eg, bulk buying) caused by the pandemic.
Mutually exclusive medication categories were constructed for AC only, AP only, AP and AC, and no AT. Apixaban, rivaroxaban, dabigatran and edoxaban were collectively categorised as DOACs for comparison with warfarin. For analysis, three mutually exclusive medication categories were tested (any AT vs no AT, AC only vs AP only, DOACs vs warfarin).

Outcomes
We defined two COVID-19 outcomes: COVID-19-related hospitalisation and COVID-19 death. COVID-19 hospitalisation included any hospital admission with a recorded COVID-19 diagnosis in any position (eg, not the primary diagnosis). COVID-19 death included individuals with a COVID-19 diagnosis on their death certificate in any position, a registered death within 28 days of their first recorded COVID-19 event or a discharge destination denoting death after a COVID-19 hospitalisation. Follow-up for COVID-19 outcomes ended on 1May2021,withthefinalfollow-updateaseitherthedateof the outcome of interest (eg, COVID-19 death) or the study end date(1May2021).

Covariates
Covariates were preselected based on potential associations with pre-existing AT use 6 or COVID-19 outcomes and included demographics (age, sex, ethnicity, geographical location, socioeconomic status, as measured by the Index of Multiple Deprivation decile), comorbidities that increase risk of stroke and bleeding (congestive heart failure, hypertension, stroke, vascular disease, diabetes, uncontrolled hypertension, renal disease, liver disease, prior major bleeding, hazardous alcohol use, history of fall, body mass index (BMI), smoking status) and other medications (antihypertensives, lipid-regulating drugs, proton pump inhibitors, non-steroidal anti-inflammatory drugs (NSAIDs), corticosteroids, other immunosuppressants and COVID-19 vaccination status, defined as at least one vaccine recorded in the COVID-19 vaccination events data set prior to the individual's COVID-19 event).
The same covariates (excluding COVID-19 vaccination status) were used as independent variables to test associations with pre-existing AT use (for any AT vs no AT, AP only vs AC only, DOACs vs warfarin) and to calculate a propensity score for use as an additional covariate in the COVID-19 outcome analysis (as demonstrated in Elze et al 16 ).

Statistical analysis
Descriptive statistics were used to summarise the study population characteristics and were stratified by medication category. Pairwise Pearson's correlation coefficients were used to check for potential collinearities between covariates. Multivariable logistic regression was used to test associations with pre-existing AT use and calculate the propensity score.
Multivariable logistic regression and Cox regression were used to test differences between exposure groups (any AT vs no AT, AC only vs AP only, DOACs vs warfarin) for COVID-19related hospitalisation and death. An additional post-hoc analysis compared dabigatran (a thrombin inhibitor) against factor Xa inhibitors (apixaban, edoxaban, rivaroxaban). Logistic and Cox regression methods were selected to evaluate potential differences between event-based (logistic regression) and timeto-event-based (Cox regression) analysis. All covariates including the propensity score were included in both methods (as demonstrated in Elze et al 16 ).Forvariableswithincompletedata(BMI: 9.3% missing), individual values were imputed with the cohort mean.
Two sensitivity analyses were conducted. First, to evaluate the potential impact of different time periods, analysis was repeated for 1 January 2020-1 December 2020, prior to the introduction of vaccines and the 29 December 2020 cases peak of the second wave. 17 Second, to validate the potential effect on COVID-19specific outcomes, analysis was repeated with COVID-19 hospitalisation and death defined exclusively as the primary recorded diagnosis (coded first on hospital record or death certificate).
Primary results are reported from the multivariable logistic regression models covering the full time period (1 January 2020-1 May 2021), with the other analyses reviewed for concordance.

Patient and public involvement
The UK National Institute for Health Research-British Heart Foundation (BHF) Cardiovascular Partnership lay panel comprising individuals affected by cardiovascular disease reviewed and approved this project.

Evaluation of AT use
From a total of 55 903 113 individuals registered with a GP practice in England, 972 971 (1.7%) had a diagnosis of AF and a CHA 2 DS 2 -VAScscore≥2on1January2020and88.0%(n=856336)had pre-existing AT use, with 74.3% (n=722 737) on AC only (see figure 1 for key study findings). The demographic and clinical characteristics of this cohort are summarised in tables 1-3. By May 2021, the proportion of individuals on any AT had fallen to 87.7%, but only AC had increased to 75.7% (see figure 2). For individuals on any AT, warfarin prescriptions fell from 24.8% in January 2020 to 17.1% in May 2021, while DOACs rose from 60.3% to 69.5% (see online supplemental figure 2).
Differences were also observed across demographics, ethnicity, socioeconomic status and geographical location, with women (OR=0.91, 95% CI 0.90 to 0.92) and individuals from ethnic minorities and lower socioeconomic positions associated with loweroddsofATuse(eg,ethnicityofblackorblackBritishvs white; OR=0.68, 95% CI 0.64 to 0.72).  In other AT subtypes (AC vs AP and DOACs vs warfarin), the results were broadly consistent (see online supplemental figures 3 and 4), with the primary exception of vascular disease which was associated with reduced odds of AC versus AP (OR=0.37, 95% CI 0.36 to 0.38).

AT use and COVID-19 outcomes
From 972 971 individuals who had a diagnosis of AF and a CHA 2 DS 2 -VASc score ≥2 on 1 January 2020, 8% (n=77364) had a recorded COVID-19 event, 3.8% (n=37 418) had a COVID-19-related hospitalisation and 2.2% (n=21 116) died whenfollowedupto1May2021.Thecharacteristicsofindividuals with a recorded COVID-19 event are summarised in online supplemental tables 1-3. Mean age (81) and comorbidities (mean CHA 2 DS 2 -VASc score 4.2) were both marginally higher compared with the full cohort. The proportion of individuals with pre-existing AT use was also marginally lower at 86.7%, but otherwise demographic and clinical characteristics were consistent.

Cardiac risk factors and prevention
These results were all directionally consistent across Cox regression analysis and the sensitivity analyses (see online supplemental figures [6][7][8].

Main findings
In 972 971 individuals with AF and a CHA 2 DS 2 -VASc score ≥2, we observed 88.0% (n=856336) with pre-existing AT use, which was associated with lower odds of COVID-19 death (OR=0.92, 95% CI 0.87 to 0.96). Although this association may not be causal, it provides further incentive to improve AT coverage for eligible individuals with AF.
Of the AF cohort analysed, 8% (n=77 364) had a recorded COVID-19 event, of which 3.8% (n=37 418) had a COVID-19-related hospitalisation and 2.2% (n=21 116) died. A marginally lower risk of COVID-19 death was observed for those with pre-existing AT use, which directionally aligns with the most comparable previous studies. 9 11 AT use was, however, associated with higher odds of COVID-19 hospitalisation. This observation remained consistent when including only hospitalisations and deaths where COVID-19 was the first coded diagnosis. Higher observed risk of hospitalisation could reflect increased health-seeking behaviour (both patient-driven or by a clinician) of those with pre-existing AT use or may indicate that any risk reduction associated with AT use only materialises in the most serious cases. The same pattern was observed in AC versus AP and supports the findings of Fröhlich et al 9 that AC may be associated with lower risk of death than AP. For DOACs versus warfarin, no difference was observed between groups for COVID-19 death, but DOACs were associated with marginally reduced odds of COVID-19 hospitalisation. Our analysis did not directly investigate the previously reported observation that vitamin K depletion through warfarin is harmful, 18 but more generally our findings suggest that it is unlikely that warfarin is associated with more severe COVID-19 outcomes compared with DOACs. 11 Although these associations across AT subtypes do not prove causality, they provide further incentive to improve AT coverage for individuals with AF that are already at high risk of stroke. Previous evaluations in the UK have estimated that around 15% of these individuals do not take any AT and around 17% take AP only rather than the recommended AC. 3 6 Our evaluation found around 12% on no AT and around 7% on AP only, which suggests national-level guidance 19 and primary care incentives such as the Quality and Outcomes Framework 20 continue to have a positive impact. Nonetheless, one in five individuals remain on a suboptimal medication regimen. Shifts from warfarin to DOACs observed in this study and others 21 were recommended by COVID-19 guidance 22 and demonstrate the potential impact of rapidly disseminated medications policy using populationscale EHR data.
Identifying which factors are associated with AT use is key to further lowering the proportion of individuals on suboptimal medication. NSAIDs displayed the strongest association with noATuseandlikelyreflectstheassociationbetweenNSAIDs and increased risk of major bleeding in individuals with AF. 23 For comorbidities, liver disease had the strongest association with no AT use, which is also supported by clinical evidence. 24 However, recent evidence suggests 25 26 more personalised risk calculations for bleeding and stroke may enable more individuals with liver disease to benefit from AT. History of falls was the comorbidity with the second strongest association with no AT use, suggesting it remains a key factor in AT medicating decisions and may be overweighted as a proxy for bleeding risk. 7 27 IntheUK,NICEguidancewasrecentlyupdated 4 to explicitly address this issue and it will be important to track the impact of this in future evaluations. On demographics, lower odds of AT use were observed in women, but this is likely influenced by using NICE's primary threshold for the CHA 2 DS 2 -VASc score of 2 for both sexes. The CHA 2 DS 2 -VASc score allocates 1 point to women and 0 for men, resulting in a larger proportion of comparatively healthy women (eg, 12% and 25% of women in the cohort have vascular disease and diabetes vs 21% and 33%, respectively, in men). However, demographic differences in AT use across ethnicity and socioeconomic status mirror systematic healthcare inequalities that have been reported previously. 28 29 Targeted outreach to these groups will be key to improving AT use further.

Strength and limitations
Routinely updated, linked, population-scale EHR data sets provide the statistical power to robustly analyse targeted subgroups and control for a wide range of potential confounders. The prevalence of individuals with AF and CHA 2 DS 2 -VASc score ≥2inourcohortissimilartothatobservedintheQualityand Outcomes Framework, 20 which provides an external validation for our data set. All code is open-source and an updated nationwide evaluation can be rapidly created for future time points.
The study does have limitations. First, the reported associations do not demonstrate causality and residual confounding is unlikely to have been fully eliminated. For example, in-hospital treatment regimens were not analysed so differences in COVID-19 outcomes due to additional targeted anticoagulation regimens 30 or other medications cannot be accounted for in our analysis. While we attempted to mitigate confounding through careful cohort selection, covariates and propensity score adjustment, our study design does not control for all potential factors associated with the initiation of AT use which may influence COVID-19 outcomes. Second, our decision (supported by Elze et al 16 ) to include all covariates and the propensity score for the COVID-19 analysis could theoretically lead to overfitting; however, Elze et al's 16 own analysis demonstrates limited differences between methods. Lastly, exposure to AT medication was defined as one or more dispensed prescriptions (recorded inNHSBSADM)intheprevious6months.Otherstudieshave used different time periods and prescription frequency counts 9 11 and adherence was not measured. We purposefully defined a liberalthresholdtosupportevaluationofATusageuptoMay 2021 that may have included unusual buying patterns (eg, bulk buying) caused by the pandemic. The trade-off is that for the COVID-19 outcome analyses it increases the probability of including a minority of 'exposed' individuals who had ceased regular, pre-existing AT medication.

CONCLUSIONS
Pre-existing AT use may be associated with lower odds of COVID-19 death and, while not evidence of causality, provides further incentive to improve AT coverage for eligible individuals with AF.
Data availability statement Data are available upon reasonable request. The deidentified data used in this study are available via the CVD-COVID-UK Consortium, coordinated by BHF Data Science Centre, for accredited researchers working on approved projects in NHS Digital's TRE for England, but as restrictions apply they are not publicly available. The authors and colleagues across the CVD-COVID-UK Consortium have invested considerable time and energy in developing the data resource used here and are keen to ensure that it is used widely to maximise its value. For enquiries about data access, please see www. healthdatagateway. org/ dataset/ 7e5f0247-f033-4f98-aed3-3d7422b9dc6d or email bhfdsc@ hdruk. ac. uk.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.
Open access This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/ licenses/by/4.0/.

IMAGE CHALLENGE
Pulmonary vein Doppler flow in a patient with fatigue and dyspnoea

CLINICAL INTRODUCTION
A woman in her 80s with a medical history of uncontrolled hypertension, hyperlipidaemia and diet-controlled pre-diabetes presented to a primary care physician's office with fatigue and dyspnoea on exertion of 2-3 months' duration. The patient reported no chest pain, paroxysmal dyspnoea or orthopnoea. Medications included atenolol 50mg once a day. Blood pressure in the clinic was 158/75 mm Hg. An echocardiogram was performed.