Article Text

Original research
Discontinuation of oral anticoagulation in atrial fibrillation and risk of ischaemic stroke
  1. Luis Alberto García Rodríguez1,
  2. Lucía Cea Soriano2,
  3. Stine Munk Hald3,
  4. Jesper Hallas3,
  5. Yanina Balabanova4,
  6. Gunnar Brobert5,
  7. Pareen Vora4,
  8. Mike Sharma6,
  9. David Gaist3
  1. 1Spanish Centre for Pharmacoepidemiologic Research (CEIFE), Madrid, Spain
  2. 2Complutense University of Madrid, Madrid, Spain
  3. 3University of Southern Denmark, Odense, Denmark
  4. 4Bayer AG, Berlin, Germany
  5. 5Bayer AB, Stockholm, Sweden
  6. 6McMaster University, Hamilton, Ontario, Canada
  1. Correspondence to Dr Luis Alberto García Rodríguez, Spanish Centre for Pharmacoepidemiologic Research (CEIFE), Madrid, Spain; lagarcia{at}


Objective To evaluate associations between oral anticoagulant (OAC) discontinuation and risk of ischaemic stroke (IS) among patients with atrial fibrillation (AF).

Methods We undertook a population-based cohort study with nested case–control analysis using UK primary care electronic health records (IQVIA Medical Research Data-UK) and linked registries from the Region of Southern Denmark (RSD). Patients with AF (76 882 UK, 41 526 RSD) were followed to identify incident IS cases during 2016–2018. Incident IS cases were matched by age and sex to controls. Adjusted ORs for OAC discontinuation (vs current OAC use) were calculated using logistic regression.

Results We identified 616 incident IS cases in the UK and 643 in the RSD. ORs for IS with any OAC discontinuation were 2.99 (95% CI 2.31 to 3.86, UK) and 2.30 (95% CI 1.79 to 2.95, RSD), for vitamin K antagonist discontinuation they were 2.38 (95% CI 1.72 to 3.30, UK) and 1.83 (95% CI 1.34 to 2.49, RSD), and for non-vitamin K antagonist oral anticoagulant discontinuation they were 4.59 (95% CI 2.97 to 7.08, UK) and 3.37 (95% CI 2.35 to 4.85, RSD). ORs were unaffected by time since discontinuation and duration of use. Annually, up to 987 IS cases in the UK and 132 in Denmark could be preventable if OAC therapy is not discontinued.

Conclusions Our results suggest that patients with AF who discontinue OAC therapy have a significant twofold to threefold higher risk of IS compared with those who continue therapy. Addressing OAC discontinuation could potentially result in a significant reduction in AF-attributed IS.

  • atrial fibrillation
  • stroke
  • epidemiology
  • electronic medical records

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See:

View Full Text

Statistics from


Atrial fibrillation (AF) is the most common chronic cardiac arrythmia affecting approximately 2%–3% of the population of Europe and the USA.1 It is associated with a fivefold increased risk of ischaemic stroke (IS) across all ages.2 Approximately one-third of IS events are attributed to AF, and these are frequently severe leading to death or significant disability.3 However, there is clear evidence that they are preventable with oral anticoagulant (OAC) therapy. Vitamin K antagonists (VKAs) reduce the risk of stroke by around two-thirds compared with controls,4 and non-vitamin K antagonist oral anticoagulants (NOACs) afford at least a comparable benefit but with the advantage of less intracranial bleeding.5

Clinical guidelines thus recommended life-long OAC therapy for patients with AF at increased stroke risk.6 7 Continuous OAC therapy without interruption is necessary to maintain this benefit. Observational studies indicate that patients with AF who discontinue OAC therapy have an approximately twofold increased risk of stroke/transient ischaemic attack (TIA),8 9 yet, 1-year OAC discontinuation rates are reportedly high—between 30% and 70%.10 11 To help address this significant care gap, robust data from recent clinical practice are needed to better determine the magnitude of the potential benefit for the AF population if OAC discontinuation is avoided. Of the limited studies on this topic, few, if any, have explored whether the risk of ischaemic stroke varies by time since OAC discontinuation, duration of OAC use before discontinuation or OAC class. We therefore conducted a large population-based cohort study with nested case–control analysis, using two separate datasets from the UK and Denmark to evaluate associations between OAC discontinuation and risk of IS among patients with AF.

Materials and methods

Data sources

We used the IQVIA Medical Research Data-UK (IMRD-UK), formerly known as The Health Improvement Network (THIN). The IMRD-UK is a database of de-identified longitudinal primary care electronic health records (EHRs) reflecting routine patient care. The database covers approximately 6% of the UK population, to which it is generalisable in terms demography, prevalence of major diseases and death rates.12 It includes coded entries (Read codes, the standard clinical coding system used by the UK’s National Health Service) and captures all prescriptions issued in primary care. Information received from secondary care visits is entered retrospectively. We also accessed data from five linked registries covering the Region of Southern Denmark (RSD; population ~1.2 million), which is representative of the Danish population regarding demographics, healthcare and medication use.13 Details on four of these registers—the National Danish Patient Registry, the Danish National Prescription Registry, the Danish Stroke Registry and the Danish Civil Registry—have been published previously.14–17 We also used the Register of Laboratory Results for Research, which contains results of blood tests conducted within Danish hospitals for inpatients/outpatients and those requested by general practitioners.

Study design and source population

The study design is shown in online supplemental figure 1. The source populations included individuals aged 20–89 years between 1 January 2016 and 31 December 2018 (Denmark) or between 1 July 2016 and 30 June 2018 (UK). These study periods were based on obtaining a 2-year period defined by the latest available data updates. The upper age cut-off was applied due to the high proportion of people above this age living in long-term care facilities, and thus with potentially more incomplete medical records. Individuals in IMRD-UK were required to be permanently registered with their practice, and to have at least 3 years’ registration before their first recorded prescription. In Denmark, individuals were required to have been residents of Southern Denmark for at least 10 years. We assigned each individual a ‘start date’ (start of follow-up) defined as the start of the study period or, for Denmark, the date of immigration to RSD (from elsewhere in Denmark or abroad) if this was later. We assigned a ‘stop date’ (end of follow-up), defined as the earliest of the following: the first coded entry of IS after the start date (using the Stroke Registry for Denmark, which records all IS hospitalisations), the emigration date from RSD (for Denmark), date of death or the end of the study period.

Public and patient involvement

No patients were involved in any aspect of the study design, implementation of the study or the interpretation and writing of the results. There are no plans to involve patients in the dissemination of the study results.

AF cohort

We restricted the source cohorts to patients with AF diagnosed before the start date (prevalent AF) and those with AF diagnosed between the start and stop date (incident AF). For the latter, the start date was reassigned to the date of AF diagnosis. For Denmark, AF was identified from the Danish National Patient Registry using International Classification of Diseases (ICD)−10 codes that have a 93% positive predictive value (PPV).18 We are unaware of AF validation studies in IMRD-UK, yet a validation study of a similar primary care database reported a 92.6% PPV for AF Read code entries.19 We excluded patients with a code for mitral valve stenosis or mechanical valve surgery any time before, and up to 14 days after, the start date.

Outcome identification

For the UK, patients were required to have a hospitalisation recorded between 15 days before and 30 days after the IS code entry date. This was determined through manual review of patients’ EHR by one investigator (LCS) with review by a second investigator (LAGR) where necessary to ensure that the hospitalisation related to the first incident IS during follow-up. The date of hospitalisation was the index date. For Denmark, the PPV (90%) and sensitivity (91%) of the ICD-10 IS codes in the stroke registry are high.20 We did not exclude patients with a history of stroke (ischaemic or haemorrhagic) before the start date, thus the stroke could have been either a first or recurrent event.

Control selection

Controls were randomly selected from the country-specific AF cohort by risk-set sampling and individually matched to cases (UK 5:1, Denmark 10:1) by age and sex. To do this, for each case, we identified all cohort members who were still at risk of a first IS and of the same age and sex (the case set). Within each case set, 5 controls (UK) or 10 controls (Denmark) were selected at random using the sttocc command in STATA. Matched controls were assigned the index date of their corresponding case. Cases were eligible to be controls for another case until the date of their IS. By this design, the OR is an unbiased estimate of the incidence rate ratio that would have emerged from a cohort study based on the source population.

Exposure to oral anticoagulants

Exposure to OACs for IS cases and controls up to (and including) the index date was determined from prescription records in IMRD-UK and the Danish Prescription Registry (see online supplemental methods). Based on the most recent episode of OAC use before the index date, OAC exposure was categorised as follows: current use, when the prescription supply lasted until/over the index date or ended within the seven previous days; intermediate use, when the prescription supply ended 8–30 days before the index date; discontinued, when the prescription supply ended ≥31 days before the index date; and non-use, where there was no prescription before the index date. An OAC treatment episode comprised the length of consecutive prescriptions disregarding gaps between prescriptions of ≤30 days (grace period). Current users of both a VKA and a NOAC (ie, switchers) were assigned to a separate category. Discontinuers of NOAC therapy were required to have no VKA use between the date of NOAC discontinuation and the index date, and vice versa.

Comorbidity and other potential confounders

We obtained data on comorbidities and other potential confounders as described in online supplemental methods.

Statistical analysis

Descriptive analyses were undertaken with categorical data presented using frequency counts and percentages, and continuous data using means with SD. Incidence rates were calculated as the number of incident IS cases during follow-up divided by the total person-years, with 95% CIs based on assuming a Poisson distribution. Incidence rates were stratified by age and sex. We estimated the annual number of potentially preventable IS cases among patients with AF that would occur nationwide in the absence of any episode of OAC discontinuation, based on data from the last complete year of the study period and population census data (see online supplemental methods). Nested case–control analyses were conducted using conditional logistic regression to calculate ORs as measures of the relative risk of IS with OAC discontinuation (vs current OAC use) adjusted for confounders. We used a stepwise approach retaining variables in the model that changed the OR by 10% or more. We included the number of referral/outpatient visits in the final model as an informative proxy variable for overall comorbidity. Stratified analyses were performed by time since OAC discontinuation (30–365 or ≥365 days before the index date) and by duration of the last episode of OAC use before discontinuation (≤120 or >120 days, ≤365 or >365 days). Sensitivity analysis of the UK data was performed to evaluate the effect of excluding patients who discontinued OAC therapy due to major bleeding (see online supplemental methods). We performed a post hoc analysis changing the reference group to ‘never use of an OAC’.

Analyses were undertaken using Stata, V.12 (UK), V.16 (Denmark).


Incidence of IS

The AF study cohorts comprised 76 882 (UK) and 41 526 (Denmark) patients. Mean age (SD) at the start of follow-up was 72.9 years (±11.4, UK) and 71.8 years (±11.5, Denmark). Men accounted for 58.7% (UK) and 58.8% (Denmark) of the study cohort. We identified 616 incident IS cases during 114 461 person-years (UK) and 643 incident IS cases during 95 236 person-years (Denmark). Corresponding incidence rates of IS were 53.8 per 10 000 person-years (95% CI 49.7 to 58.2, UK) and 67.5 (95% CI 62.4 to 72.9, Denmark). The mean age of IS cases was 77.5 years (SD 9.2, UK) and 78.4 years (SD 9.0, Denmark). Men accounted for 56.0% (UK) and 55.7 (Denmark) of IS cases. Incidence rates of IS increased with age (online supplemental figure 2 and table 1).

Factors associated with incident IS

Characteristics of IS cases and controls in the UK and Danish cohorts are shown in table 1 and online supplemental table 2. In both UK and Danish cohorts, having at least one hospitalisation (due to any cause) in the year before the index date was associated with a higher risk of IS. Other factors shown to be associated with a higher risk of IS were having a history of IS, current use of antiplatelets and, for Denmark, a diagnosis of peripheral artery disease. Current use of antiarrhythmics was associated with a lower risk of IS.

Table 1

Characteristics of incident IS cases and controls

Discontinuation (previous use) of OAC therapy among incident cases of hospitalised IS

The number of IS cases in each OAC exposure category is shown in table 2. A total of 28.6% (176/615) of UK IS cases and 22.1% (142/643) of Danish IS cases had no OAC prescription before their IS. Among IS cases who had ever used an OAC, 31.2% (UK, 137/439) and 26.1% (Denmark, 131/501) of cases discontinued OAC therapy before their hospitalisation. Among IS cases ever prescribed a VKA, 34.0% (UK, 84/247) and 29.6% (Denmark, 75/253) discontinued, while among IS cases ever prescribed a NOAC, 26.8% (UK, 53/198) and 21.8% (Denmark, 56/257) discontinued. Sixty-three IS cases 2017 (UK) and 42 IS cases in 2018 (Denmark) were OAC discontinuers. Using population estimates of the data sources and nationwide, we estimated that annually, 1481 IS cases in the UK and 198 IS cases in Denmark occurred among patients with AF who discontinued OAC therapy (table 3). As randomised controlled trials have documented that OAC therapy reduces IS risk by approximately two-thirds,4 5 we estimated that 987 (ie, two-thirds of 1481) IS cases in the UK and 132 (two-thirds of 198) IS cases in Denmark are potentially preventable if OAC therapy is not discontinued.

Table 2

OAC exposure among incident cases of IS in the UK and Denmark

Table 3

Nationwide estimates for the UK and Denmark of the number of IS cases potentially occurring among patients with AF who had discontinued OAC therapy at the time of stroke

OAC discontinuation and risk of IS

OAC discontinuation was associated with an increased risk of IS in both datasets (figure 1, table 4). For any OAC discontinuation, adjusted ORs were 2.99 (95% CI 2.31 to 3.86, UK) and 2.30 (95% CI 1.79 to 2.94, Denmark), for VKA discontinuation they were 2.38 (95% CI 1.72 to 3.30, UK) and 1.83 (95% CI 1.34 to 2.48, Denmark), and for NOAC discontinuation they were 4.59 (95% CI 2.97 to 7.08, UK) and 3.37 (95% CI 2.35 to 4.84, Denmark). ORs from analyses stratified by time since OAC discontinuation (table 4) and duration of OAC use before discontinuation (figure 1, online supplemental file 1) were not significantly different from the main estimates. In the sensitivity analysis, after removing 25/137 (18.2%) IS cases and 34/313 (10.91%) controls in the UK who had discontinued OAC therapy due to bleeding, the OR for IS (2.88, 95% CI 2.19 to 3.78) was only minimally different from the main analysis estimate. In the post hoc analysis, using never use as the reference group, current OAC use was associated with a significantly reduced risk of IS, and OAC discontinuation was associated with an increased risk of IS (online supplemental table 4).

Figure 1

ORs (95% CI) for associations between oral anticoagulant (OAC) discontinuation and risk of ischaemic stroke (IS).

Table 4

ORs (95% CI) for the association between time since OAC discontinuation and the risk of IS among patients with AF


In our large population study using data from the UK and Denmark, we found that patients with AF who discontinue OAC therapy have a twofold to threefold higher risk of IS compared with those who maintain therapy. There was no evidence that this risk changed appreciably by time since OAC discontinuation or duration of OAC use or differed by OAC class. We estimated that close to 987 IS cases in the UK and 132 IS cases in Denmark could be preventable every year if OAC therapy was not discontinued.

A key strength of our study is the population-based settings in two European countries with different healthcare systems and the similar results observed from these two settings. Our findings have good generalisability because the study cohorts included elderly patients and those with multiple comorbidities, reflecting the spectrum of patients with AF in clinical practice, and the individuals in the IMRD-UK and the RSD are representative of their respective nationwide populations.12 13 We evaluated both VKAs and NOACs—the latter being increasingly prescribed over the last decade,21 22 and the OAC class recommended for stroke prevention in this patient population in American guidelines.6 The nested case–control analysis enabled an accurate assessment of OAC exposure—a variable that changes over time. To further avoid misclassification of OAC therapy, we did not include patients who discontinued OAC therapy in the month before the index date in our main discontinuer category, that is, patients more prone to exposure misclassification. Discontinuation of OAC may represent an appropriate response to, for example, a major bleeding event, yet our sensitivity analysis indicated that the associations seen are not explained by bleeding occurrence. Although some misclassification of OAC exposure may still have occurred as we assumed that all patients took their medicine, any misclassification would have been non-differential between IS cases and controls and biased the risk estimates towards the null. Another limitation is that some IS cases in the UK may have been missed if information from secondary care was not recorded. This could be one reason why IS incidence rates in the UK were lower than those in RSD where stroke cases were identified from a hospital-based database. In addition, the percentage of IS cases with a previous IS (a risk factor for subsequent stroke) was lower in the IMRD-UK (20.7%) than in the RSD (32.5%), and incidence rates of IS in Denmark were notably higher than UK rates in the Global Burden of Disease study.23 While we adjusted for confounding factors in our analyses, residual confounding cannot be excluded.

The twofold to threefold higher risk of IS associated with OAC discontinuation observed in our study is roughly the inverse of the benefit obtained with OAC use for stroke prevention in AF seen in randomised trials.4 5 This level of increased IS risk is also consistent with previous smaller observational studies on this topic,8 9 24 25 including those from Martinez et al24 in the UK who similarly found that time since VKA discontinuation was not an influencing factor on IS risk. Other smaller observational studies have reported even high relative risks for stroke with VKA cessation.26 27 Although the point estimates in our study suggest IS risk could be higher when the OAC discontinued is a NOAC, this could be explained by greater exposure misclassification for VKAs than NOACs. One could expect misclassification of VKA exposure to be more common because dosing is guided by the international normalised ratio. This could lead to greater inaccuracy in determining VKA prescription length using prescription data alone, thus diluting the true increased relative risk under a non-differential assumption.

Our findings have high clinical relevance as they point towards a significant potential to reduce AF-related IS if OAC discontinuation levels are reduced. OAC discontinuation is common—in our study, nearly one-third of IS cases in the UK and about a quarter of those in Denmark were classed as discontinuers at the time of their IS, and it should be noted that these patients may have had previous episodes of discontinuation. In addition, we found that a substantial percentage of IS cases (28.6% UK, 22.1% Denmark) were not prescribed OAC therapy before their stroke, in line with a previous study from Denmark.28 Our estimates could help support physician–patient communication about the extent of IS risk if they were to discontinue OAC therapy. Encouraging therapy persistence is increasingly important as ageing populations will see greater numbers of people living with AF and requiring OAC therapy. So far, evidence regarding a potential benefit of educational/behavioural interventions on OAC persistence among patients with AF is lacking,29 30 thus further research into ways of increasing therapy OAC persistence among patients with AF is needed. The quantitative evidence provided from our study, of the importance of continuing OAC therapy, may help towards greater prioritisation of this topic, as well as being helpful for physicians in their efforts to educate patients about the importance of OAC persistence.


What is already known on this subject?

  • The majority of patients with atrial fibrillation (AF) require long-term oral anticoagulant (OAC) therapy to reduce their risk of ischaemic stroke (IS), yet levels of OAC discontinuation among this patient population are high.

What might this study add?

  • Our study suggests that patients who discontinued OAC therapy have a two- to three-fold higher IS risk than those who maintain therapy, irrespective of OAC class, time since discontinuation, or OAC duration.

How might this impact on clinical practice?

  • Encouraging persistence with OAC therapy in patients with AF could potentially result in a significant reduction in AF-attributed IS.


We thank Susan Bromley of EpiMed Communications Ltd, Abingdon, UK for providing medical writing assistance funded by Bayer AG. IQVIA provided the IMRD-UK incorporating data from THIN, a Cegedim Database. Reference made to THIN is intended to be descriptive of the data asset licensed by IQVIA.


View Abstract

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • Presented at Two abstracts containing data from this study have been accepted for presentation at the European Stroke Organisation Conference to be held in November 2020.

  • Contributors LAGR, DG, YB, GB and PV contributed to the design of the study. LCS, LAGR and DG extracted the data and performed the statistical analysis. All authors contributed to the interpretation of the data and the review of manuscript drafts, and all approved the final manuscript. LAGR is the guarantor. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.

  • Funding The UK part of this study was funded by Bayer AG. The Danish part of this study did not receive funding; the activities of David Gaist are supported by a grant from Odense University Hospital. Bayer had no role in the study apart from salaries paid to YB, GB and PV whose contributions to the study are mentioned above. All authors had full access to all of the data (including statistical reports and tables) in the study and can take responsibility for the integrity of the data and the accuracy of the data analysis is also required.

  • Competing interests LAGR works for CEIFE, which has received other research funding from Bayer AG. LAGR has also received honoraria for serving on advisory boards for Bayer AG. DG has received honoraria from AstraZeneca (Sweden) for participation as a co-investigator on a research project outside the submitted work, and receiving speaker honorarium from Bristol-Myers Squibb outside the submitted work. YB and PV are employees of Bayer AG. GB is an employee of Bayer AB. MS has served on the steering committees and led sub-studies from trials sponsored by Bayer and has served as a consultant and received speaker’s honoraria from Bayer. MS has also served as a consultant to Portola, Bristol Myers Squibb and Janssen. LCS, SMH and JH report no potential conflicts of interest.

  • Patient consent for publication Not required.

  • Ethics approval The UK and Denmark studies were approved by the Independent Scientific Research Committee for IMRD-UK (reference no. 19THIN057) and the Danish Data Protection Agency, respectively. Data collection for IMRD-UK was approved by the South East Multicentre Research Ethics Committee in 2003 and individual studies using IMRD-UK data do not require separate ethical approval if only anonymised data are used. Danish law does not require approval from an ethics board or informed consent for register studies.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data availability statement Data are available from the corresponding author upon reasonable request.

  • 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.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles