Article Text

other Versions

Download PDFPDF
Original research
Female sex as a risk factor for ischaemic stroke varies with age in patients with atrial fibrillation
  1. Victor Chien-Chia Wu1,
  2. Michael Wu2,
  3. Victor Aboyans3,
  4. Shang-Hung Chang1,
  5. Shao-Wei Chen4,
  6. Mien-Cheng Chen5,
  7. Chun-Li Wang1,
  8. I-Chang Hsieh1,
  9. Pao-Hsien Chu1,
  10. Yu-Shen Lin6,7
  1. 1 Division of Cardiology, Chang Gung Memorial Hospital, Linkou Medical Center, Taoyuan City, Taiwan
  2. 2 Divison of Cardiovascular Medicine, Arrhythmia Services Section, Rhode Island Hospital, Warren Alpert School of Medicine, Brown University, Providence, Rhode Island, USA
  3. 3 Department of Cardiology, Dupuytren University Hospital, Limoges, France
  4. 4 Department of Cardiothoracic and Vascular Surgery, Chang Gung Memorial Hospital, Linkou Medical Center, Taoyuan City, Taiwan
  5. 5 Division of Cardiology, Department of Internal Medicine, Chang Gung Memorial Hospital, Kaohsiung Medical Center, Chang Gung University College of Medicine, Kaohsiung, Taiwan
  6. 6 Division of Cardiology, Chang Gung Memorial Hospital, Chiayi, Taiwan
  7. 7 Graduate Institute of Clinical Medical Sciences, College of Medicine, Chang Gung University, Taoyuan, Taiwan
  1. Correspondence to Dr Yu-Shen Lin, Department of Cardiology, Chang Gung Memorial Hospital, Chiayi 33305, Taiwan; dissertlin{at}


Objectives Female sex is an inconsistent ischaemic stroke risk factor in patients with atrial fibrillation (AF). We hypothesised that the ischaemic stroke risk varies with age among women compared with men.

Methods We retrieved the patients with newly diagnosed AF during 2001–2013 from Taiwan’s National Health Insurance Research Database. Patients with missing information, age <20 years, history of valvular heart disease and surgery, rheumatic heart disease, hyperthyroidism or anticoagulation and/or antiplatelet use were excluded. Propensity score matching (PSM) included patient comorbidities, medications and index date stratified by age and sex groups. Primary outcome was defined as ischaemic stroke at follow-up.

Results After exclusion criteria, 87 369 men and 71 853 women remained for analysis (aged 73.1±14.4 years). After 1:1 PSM, we included 59 583 men (aged 73.5±13.7 years) and 59 583 women (aged 73.4±13.8 years) for analysis. We also stratified patients by age. The ischaemic stroke risk varied with age in women compared with men: lower in the ≤55 years (subdistribution HR (SHR)=0.75, 95% CI 0.62 to 0.90) and 56–65 years (SHR=0.87, 95% CI 0.78 to 0.98) groups, neutral in the 66–75 years group (SHR=1.01, 95% CI 0.94 to 1.08) and adverse in the >75 years group (SHR=1.14, 95% CI 1.09 to 1.19).

Conclusions The female/male ischaemic stroke risk ratio varied with age. Only women aged >75 years had a higher risk, whereas women aged <65 years had a lower risk compared with men. These findings challenge the ‘sex category’ component of the CHA2DS2-VASc score, used to make decision regarding anticoagulation treatment in AF patients.

  • atrial fibrillation
  • female sex
  • anticoagulation
  • ischemic stroke

Statistics from

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.


Atrial fibrillation (AF) is prone to ischaemic stroke and may engender incapacitating disabilities in patients without proper management. Accurate risk prediction and effective stroke prevention thus are imperative in patients with AF. According to the European and US guidelines on AF management,1–3 female sex is equivalent to a CHA2DS2-VASc score of 1. However, controversies exist regarding the weight of female sex in the score calculation.

A Swedish study reported that anticoagulation should be considered for women with AF with a moderately increased risk of stroke, but not for women aged <65 years without risk factors.4 A Danish study also revealed that female sex was a risk modifier rather than a risk factor.5 Moreover, a Japanese study concluded that female sex could generally be excluded from the calculation of the CHA2DS2-VASc score because it does not confer the same weight for ischaemic stroke prediction as the rest of the components.6 Other studies have debated the excess of stroke risk in women aged <65 years.7 8 Such inconsistencies questioned the validity of anticoagulation in these patients with AF, where side effect may be associated with life-threatening bleeding. Therefore, the aim of the study was to investigate the implication of female sex as a risk factor for ischaemic stroke in patients with AF. We hypothesised that the excess risk in women is age dependent.


Data source

Taiwan’s National Health Insurance started in 1995 and covers 99.5% (23 million) of the residents in Taiwan. The National Health Insurance Research Database (NHIRD) provides the data of all inpatient and outpatient services, diagnoses, emergency room visits, prescriptions, examinations, operations and expenditures and are updated biannually. Diseases were diagnosed using International Classification of Diseases, ninth revision, Clinical Modification (ICD-9-CM) and version 2001 codes. Since >95% of the Taiwanese population consists of Han Chinese, our study can be considered to be uniform in ethnic background.

Study patients

Electronic medical records from the NHIRD between 1 January 2001 and 31 December 2013 were retrieved for patients with a discharge diagnosis or at least two consecutive outpatient clinic diagnoses of AF (ICD-9-CM V.2001: 427.31). The date on which AF was first diagnosed was assigned as the index date. The coverage period of our database was from 1997 to 2013; therefore, we first excluded patients with AF diagnosis between 1997 and 2000. Furthermore, we excluded patients who with missing information (<0.1%), aged <20 years, history of valvular surgery, rheumatic heart disease, hyperthyroidism, or received anticoagulation/antiplatelet during the entire follow-up. The remaining patients were classified according to sex. Propensity score matching (PSM) included patient comorbidities, medications and index date, with stratification being conducted according to age and sex groups (figure 1).

Figure 1

Flowchart of study design and screening criteria for inclusion of patients with atrial fibrillation.

Study outcome, covariates and follow-up

Our primary outcome was ischaemic stroke and was defined according to principal diagnoses at admission or emergency visits. These diagnoses have been previously validated in NHIRD studies.9 The first incidence of ischaemic stroke at any time during follow-up was assigned as the endpoint. Each patient was followed up until the date of ischaemic stroke occurrence, death or 31 December 2013, whichever occurred first.

Covariates included comorbidities, namely hypertension, diabetes mellitus, dyslipidaemia, ischaemic heart disease, peripheral arterial disease, chronic obstructive pulmonary disease, renal status, abnormal liver function, gout, systemic thromboembolism, myocardial infarction, stroke, heart failure and medications. These comorbidities have been defined according to discharge diagnosis and/or two outpatient visits. Patients with systemic thromboembolism, myocardial infarction, stroke and heart failure were defined as having any inpatient diagnosis before the index date. Most of these diagnoses were validated in a previous NHIRD study. Similarly, data on medication usage were retrieved on claim-based data of the previous year.

Ascertainment of AF and CHA2DS2-VASc score

All patients who had AF diagnosis (ICD-9-CM code 427.31) defined according to diagnoses during hospitalisation or ≥2 consecutive clinic visits. The accuracy of AF diagnosis using ICD-9-CM code in the NHIRD has been confirmed in previous studies.10 11 Comorbidities in CHA2DS2-VASc score were hypertension, diabetes mellitus, vascular disease, stroke and heart failure, where validation have been made12–14 and appropriate medical prescriptions were utilised to increase the diagnostic accuracy. Vascular disease, including coronary artery disease and peripheral arterial disease, have also been validated.12 15 16

Statistical analysis

To adjust for disparities in comorbidities between men and women, we performed PSM. The patients were first split into 10 strata according to age (≤50, 51–55, 56–60, 61–65, 66–70, 71–75, 76–80, 81–85, 86–90 and >90 years), and then PSM was performed separately for each age stratum. We aggregated the 10 data sets when conducting analysis. Covariates included in the propensity score calculation were the 13 comorbidities, 16 types of medication and index date. The matching was processed using a greedy nearest neighbour algorithm with a calliper of 0.2 times the SD of the logit of propensity score. The quality of matching was assessed using the standardised mean difference (SMD) between sex groups after matching, where a value <0.1 indicated negligible difference (table 1).

Table 1

Baseline characteristics of the study patients before and after propensity score matching

The stroke risk of both sexes was compared in the entire matched cohort and stratified by age group by using Fine and Gray’s subdistribution hazard model, which considered all-cause death as a competing risk (online supplementary table 1). The interaction of sex and continuous age was calculated, after which continuous age was divided into age groups. We used the subdistribution hazard function to determine the cumulative incidence of ischaemic stroke events stratified by age (ie, ≤55 years, 56–65 years, 66–75 years and >75 years). Finally, forest plots comparing the risk of ischaemic stroke between female and male patients aged ≤55, 56–65, 66–75 and >75 years were performed according to different CHA2DS2-VA scores without considering the sex category.

Supplemental material

A p value of <0.05 was considered statistically significant. No adjustment for multiple testing (multiplicity) was performed in this study. All statistical analyses were performed using commercial software (SAS V.9.4), including the ‘psmatch’ procedure for PSM, ‘phreg’ procedure for survival analysis, and ‘%cif’ macro for generating a cumulative incidence function through Fine and Gray’s method.

Sensitivity analysis

To test the robustness of the primary analysis, we conducted three sensitivity analyses. First, excluding patients who underwent oral anticoagulant (OAC) treatment during the follow-up may lead to underestimations of stroke risk, particularly in patients at low risk of stroke.17 Therefore, we censored patients who initiated anticoagulation treatment during the follow-up instead of excluding them (online supplementary table 2). Second, we compared the risk of stroke between men and women by using the whole cohort data before PSM with traditional multivariable adjustment (online supplementary table 3: age was treated as a continuous variable with mean centering; online supplementary table 4: age was treated as a categorical variable with 10 levels). The covariates included all the variables listed in table 1, except for CHA2DS2-VASc score, follow-up years and propensity score. Third, we used the Cox proportional hazard model in addition to the Fine and Gray competing model (online supplementary table 1).

Supplemental material

Supplemental material

Supplemental material

Patient and public involvement

Since this study is a retrospective cohort study based on national insurance database, in no part or stage of the research were the patients/public involved.


Study population

We identified a total of 334 680 patients newly diagnosed of AF during 2001–2013 in the NHIRD. After exclusion criteria, 159 222 patients with AF (aged 73.1±14.4 years) were eligible for analysis. Of these eligible patients, 87 369 were men (aged 71.3±14.8 years) and 71 853 were women (aged 75.3±13.6 years). After PSM, 59 583 male patients (aged 73.5±13.7 years) and 59 583 female patients (aged 73.4±13.8 years) were analysed for comparison of ischaemic stroke risk (figure 1). The SMD values of all age groups were 0, and those of the remaining variables were <0.1, except for the value of the CHA2DS2-VASc score. These results indicated balanced distributions of these variables between male and female patients (right panel in table 1).

Determining ischaemic stroke risk through age-stratified analysis after PSM

Before stratifying the patients according to age groups, we assessed whether the sex difference in stroke risk was constant, proportional or non-linear across ages. Therefore, prior to PSM, we used Fine and Gray’s model, which included continuous age (with mean centred), square of age, sex, interaction of age by sex and interaction of square of age by sex. The result revealed that the interaction of square of age with sex was significant, indicating that there was a quadratic relationship between impact of sex on stroke risk and ages (p=0.001, online supplementary table 3). To understand how age may modify the association between sex and ischaemic stroke, we split the patients into 10 age strata; we then performed PSM and compared the risk of ischaemic stroke between sex groups in each stratum. Compared with men, women had a significantly decreased ischaemic stroke risk in the ≤50 years (subdistribution HR (SHR) 0.71, 95% CI 0.54 to 0.94), 51–55 years (SHR 0.64, 95% CI 0.50 to 0.82) and 56–60 years (SHR 0.77, 95% CI 0.64 to 0.94) age groups. The sex difference in stroke risk was non-significant in the 61–65 years (SHR 0.89, 95% CI 0.77 to 1.03), 66–70 years (SHR 0.99, 95% CI 0.89 to 1.11) and 71–75 years (SHR 1.02, 95% CI 0.94 to 1.12) age groups. Then, women had a significantly increased ischaemic stroke risk in the 76–80 years (SHR 1.13, 95% CI 1.05 to 1.22), 81–85 years (SHR 1.19, 95% CI 1.10 to 1.28) and 86–90 years (SHR 1.17, 95% CI 1.06 to 1.30) age groups. The sex difference was again non-significant in the >90 years age group (SHR 0.99, 95% CI 0.84 to 1.16) (online supplementary table 1). Compared with men, the risk of ischaemic stroke in women increased with age in patients with AF; the risk profile changed from protective to neutral, then to adverse and finally back to neutral in non-agenarian patients (figure 2).

Figure 2

Subdistribution HR of female versus male sex for ischaemic stroke risk stratified by age group.

We combined age groups and used the following age categories to generate cumulative incidence graphs (figure 3) and forest plots according to different CHA2DS2-VA scores (figure 4): ≤55, 56–65, 66–75 and >75 years. The female-to-male ischaemic stroke risk profile changed from protective in the ≤55 years age group (SHR 0.75, 95% CI 0.62 to 90), slightly protective in the 56–65 years age group (SHR 0.87, 95% CI 0.78 to 0.98) and neutral in the 66–75 years age group (SHR 1.01, 95% CI 0.94 to 1.08) to adverse in the >75 years age group (SHR 1.14, 95% CI 1.09 to 1.19).

Figure 3

Cumulative incidence function of ischaemic stroke in male and female patients aged (A) ≤55, (B) 56–65, (C) 66–75 and (D) >75 years (D).

Figure 4

Forest plots representing female versus male ischaemic stroke risk in (A) ≤55, (B) 56–65, (C) 66–75 and (D) >75 years age groups.

In the first sensitivity analysis, we censored the patients who initiated anticoagulants during the follow-up period instead of excluding them, and the analysis results were similar to those of the primary analysis (online supplementary table 2). In the second sensitivity analysis, we used the whole cohort with traditional multivariable adjustment. Without stratification on the basis of age, we observed a significant interaction between sex by square of continuous age (online supplementary table 3). By categorising age into 10 groups, we also noted that female sex exhibited the same trend of stroke risk (ie, protective, neutral, adverse and again neutral) with increasing age when compared with the male sex (online supplementary table 4). In the third sensitivity analysis, the Cox model results in the predefined age groups were generally comparable to the results of the primary analysis by Fine and Gray’s model (online supplementary table 1).


The main findings and characteristics of current study were: (1) female sex conferred dynamic and varying ischaemic stroke risk profiles, ranging from protective in the ≤55 year age group, slightly protective in the 56–65 year age group and neutral in the 66–75 year age group to adverse in the >75 year age group. (2) Our study has the longest mean follow-up period (3 years) compared with previous studies of 1–2 years.4–6

Prior to PSM, incidence density was used to evaluate the ischaemic stroke risk in each component such that CHA2DS2-VASc score was equal to 1. During follow-up, patients diagnosed as having ischaemic stroke had a CHA2DS2-VASc score of 1. Among components of CHA2DS2-VASc for which the score is equal to 1, the incidence density was lowest in female sex (online supplementary figure 1).

Supplemental material

Overall, female sex was associated with significantly increased ischaemic stroke compared with the male sex (SHR 1.06, 95% CI 1.03 to 1.09) (online supplementary table 1). This may be attributed to the fact that in both male and female patients, diagnosis of AF was mainly distributed in the >75 years age group, accounting for >50% of the entire population (online supplementary figure 2).

Supplemental material

In patients with AF, the use of the CHA2DS2-VASc score to appropriately guide anticoagulation therapy is crucial. In the seminal study which introduced the CHA2DS2-VASc score, univariate analysis revealed that female sex, history of vascular disease, prior stroke or transient ischaemic attack and diabetes were associated with an increased incidence of thromboembolism (all p<0.05), whereas multivariate analysis demonstrated that only female sex (OR 2.53, p=0.29) was significantly associated with an increased incidence of thromboembolism.18 This was confirmed in a Danish study, where female sex was given a weight of 1 point in the CHA2DS2-VASc score, representing an intermediate risk level.19 However, a Swedish study challenged these results and argued against prescribing anticoagulants to women aged <65 years without other risk factors.4

This paradoxical recommendation has led to a Danish study suggesting that the decision on OAC treatment should be guided by a CHA2DS2-VA score excluding the sex category and that the ‘Sc’ risk component should enhance the decision for OAC treatment based on ≥2 additional stroke risk factors.5 However, in patients with AF with ≥2 risk factors, the decision to use OAC is already certain. In a Japanese study, the authors suggested that female sex component should be excluded and that only CHA2DS2-VA score be used for risk stratification and OAC treatment decision.6 They found no sex difference in thromboembolic events when they used CHADS2, modified CHA2DS2-VASc and modified CHA2DS2-VA risk scores, where V component was coronary artery disease.6 In a meta-analysis, female patients with AF were at an increased risk of stroke compared with male patients, particularly in the >65 years age group.19 Female sex may act as a stroke risk modifier, particularly in elderly and very elderly patients with AF, conferring a significant increase in stroke risk.20 In a Chinese study without using PSM, the cumulative incidence of stroke or systemic embolism in men versus women was found to be not different across age groups of <65 (p=0.95), 65–74 (p=0.58) and >75 (p=0.17) years.21

In the current study, in order to eliminate the effect of additional scores due to age in CHA2DS2-VASc score, we directly compared female and male patients in all age groups. We performed PSM in 10 age strata, followed by comparing the ischaemic stroke risk between male and female patients. Initially, the ischaemic stroke risk decreased in female patients in the ≤55, 51–55, 56–60 and 61–65 years age groups compared with male patients; the risk subsequently became neutral in the 66–70 and 71–75 years age groups (figure 2 and online supplementary table 1). Then, the risk increased in female patients in the 76–80, 81–85 and 86–90 years age groups compared with male patients. Finally, there was an equipoise between sexes in patients at age >90 years. By combining age groups, female sex as a risk factor for ischaemic stroke varied from protective at age ≤55 years (SHR 0.75, 95% CI 0.62 to 90) and age 56–65 years (SHR 0.84, 95% CI 0.75 to 0.95), to neutral at 66–75 years (SHR 1.01, 95% CI 0.94 to 1.08) and then to adverse at >75 years (SHR 1.13, 95% CI 1.08 to 1.18). Our study showed that female patients in older age groups had a higher risk of ischaemic stroke compared with male patients and is consistent with earlier studies that enrolled older patients.22 In our cohort, patients with AF aged >75 years accounted for >50% of the entire AF population, resulting in female patients having an overall increased ischaemic stroke risk compared with their male counterparts. Nevertheless, it would be oversimplified to consider women as a group having an excess ischaemic stroke risk compared with men, because at a younger age (≤65 years) women had a reduced risk compared with men in our study.

In summary, studies have examined and compared the stroke risk between sexes in patients with AF and have found that women had a comparable 23 ,24 or higher risk25 26 compared with men. The conventional CHA2DS2-VASc score thus has limitations to appropriately guide the stroke risk evaluation in female patients with AF. Given this knowledge gap, we conducted current study with a large population of age-matched and sex-matched patients with AF to clarify the relationship between female and male risks for ischaemic stroke. Our investigation revealed that female sex as a risk factor for ischaemic stroke in patients with AF varied with age, from protective in the ≤55 years age group, slightly protective in the 56–65 years age group, neutral in the 66–75 years age group and then adverse in the >75 years age group.


Several limitations are associated with epidemiologic data from the NHIRD. First, using ICD-9-CM codes for patient screening may result in some cases missed due to incorrect coding. Second, some patients who had ischaemic stroke with indistinct symptoms may be missed due to unavailability of clinical characteristics or image studies. Third, certain misclassification of disease leading to the miscalculation of the CHA2DS2-VASc score may have occurred due to retrospective nature of the study. Finally, we did not compare our study result directly with the ischaemic stroke risk calculated using the CHA2DS2-VASc score for different age groups (65–74 and ≥75 years). Therefore, further investigation may be required on the clinical applicability of our results.


Our findings do not support an increased ischaemic stroke risk in all female patients with AF as compared with male patients. The female/male ischaemic stroke risk ratio varied according to age. Only women aged >75 years had a higher risk, whereas women aged <65 years had a lower risk as compared with men. These findings challenge the ‘sex category’ component of the CHA2DS2-VASc score, which is used to make decision regarding anticoagulation treatment in patients with AF.

Key messages

What is already known on this subject?

  • Studies have revealed that female sex is an inconsistent ischaemic stroke risk factor in patients with atrial fibrillation (AF).

What might this study add?

  • The female/male ischaemic risk ratio varies according to age. Only women aged >75 years with AF have a higher risk, whereas women aged <65 years have a lower risk as compared with men.

How might this impact on clinical practice?

  • Female patients with AF and aged <65 years presented a decreased risk of stroke compared with their male counterparts; therefore, anticoagulation treatment in these patients may not be warranted.

Supplemental material


We thank Zoe Ya-Jhu Syu and Alfred Hsing-Fen Lin for their assistance in statistical analyses in this study.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.


  • VC-CW and MW contributed equally.

  • Contributors VCW, MW and YSL contributed to the conception or design of the work and final approval of the version to be published. VCW and YSL contributed to the data collection. VCW, MW, VA, SHC and SWC contributed to the data analysis and interpretation. VCW and MW drafted the article. MCC, CLW, ICH and PHC contributed to the critical revision of the article.

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Competing interests None declared.

  • Patient consent for publication Not required.

  • Ethics approval This study was approved by Institutional Review Board of Chang Gung Memorial Hospital (IRB No. 201 801 354B0).

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

  • Data availability statement Data are available upon reasonable request.