Atrial fibrillation as risk factor for cardiovascular disease and death in women compared with men: systematic review and meta-analysis of cohort studiesBMJ 2016; 352 doi: https://doi.org/10.1136/bmj.h7013 (Published 19 January 2016) Cite this as: BMJ 2016;352:h7013
- Connor A Emdin, DPhil student1,
- Christopher X Wong, research fellow2,
- Allan J Hsiao, PhD student3,
- Douglas G Altman, professor4,
- Sanne AE Peters, research fellow1,
- Mark Woodward, professor1 5,
- Ayodele A Odutayo, DPhil student4 6
- 1George Institute for Global Health, University of Oxford, Oxford, UK
- 2Nuffield Department of Population Health, University of Oxford, Oxford, UK
- 3Department of Economics, Massachusetts Institute of Technology, Cambridge, MA, USA
- 4Centre for Statistics in Medicine, Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences, University of Oxford, Oxford, UK
- 5George Institute for Global Health, University of Sydney, Sydney, Australia
- 6Faculty of Medicine, University of Toronto, Toronto, ON, Canada
- Correspondence to: C Emdin
Objective To determine whether atrial fibrillation is a stronger risk factor for cardiovascular disease and death in women compared with men.
Design Meta-analysis of cohort studies.
Data sources Studies published between January 1966 and March 2015, identified through a systematic search of Medline and Embase and review of references.
Eligibility for selecting studies Cohort studies with a minimum of 50 participants with and 50 without atrial fibrillation that reported sex specific associations between atrial fibrillation and all cause mortality, cardiovascular mortality, stroke, cardiac events (cardiac death and non-fatal myocardial infarction), and heart failure.
Data extraction Two independent reviewers extracted study characteristics and maximally adjusted sex specific relative risks. Inverse variance weighted random effects meta-analysis was used to pool sex specific relative risks and their ratio.
Results 30 studies with 4 371 714 participants were identified. Atrial fibrillation was associated with a higher risk of all cause mortality in women (ratio of relative risks for women compared with men 1.12, 95% confidence interval 1.07 to 1.17) and a significantly stronger risk of stroke (1.99, 1.46 to 2.71), cardiovascular mortality (1.93, 1.44 to 2.60), cardiac events (1.55, 1.15 to 2.08), and heart failure (1.16, 1.07 to 1.27). Results were broadly consistent in sensitivity analyses.
Conclusion Atrial fibrillation is a stronger risk factor for cardiovascular disease and death in women compared with men, though further research would be needed to determine any causality.
Atrial fibrillation (AF) is a leading cause of cardiovascular disease worldwide,1 with an estimated 33.5 million people affected in 2010.2 Recent estimates from the Global Burden of Disease study indicated that AF was associated with an age adjusted mortality rate of 1.7 per 100 000 people (95% uncertainty interval 1.4 to 2.1) in 20103 and that the prevalence is increasing in both developed and developing countries.2
Although AF is associated with an increased risk of stroke4 and death5 in men and women, a growing body of literature suggests that women and men experience risk factors for cardiovascular disease differently. Previous analyses have shown that smoking and diabetes are associated with greater proportional risks of coronary heart disease in women than in men.6 7 Diabetes is also associated with a greater relative risk of stroke in women.8 It is currently unclear, however, whether such sex differences exist for AF. While being female is a risk factor for stroke among individuals with AF,9 this could reflect differences in the multivariable adjusted risk of stroke by sex in the general population rather than differential effects of AF by sex. To explore sex differences in the effect of AF, it is necessary to compare sex specific estimates of the effect of AF on risk of death and cardiovascular disease. While some studies have suggested that AF is more strongly associated with the risk of stroke10 and death11 in women than men, others have suggested not.12 13 14 Sex differences in the association between AF and death and cardiovascular disease would have substantial implications for the estimation of the global and regional burden, for the targeting of treatment to manage it, and for future research into causes of sex differences.
Accordingly, we conducted a meta-analysis of cohort studies to estimate the association between AF and cardiovascular disease and death in women and men and to compare the sexes.
Search strategy and selection criteria
We conducted a systematic review of cohort studies that reported associations between AF and death or cardiovascular events in men and women. This was done in accordance with the MOOSE (meta-analysis of observational studies in epidemiology) guidelines.15 An experienced research librarian designed and conducted the search strategy. Medline and Embase were searched from 1966 to March 2015 with a combined text and MeSH subheading search with the following terms: “atrial fibrillation”, “mortality”, “death”, “cardiovascular disease”, “heart failure”, “myocardial infarction”, “death, sudden, cardiac”, “stroke”, “kidney”, “renal”, “peripheral”, and “risk factors”. We conducted a review of the references of identified studies.
We included any study that reported sex specific associations between AF and any of the following outcomes: all cause mortality, cardiovascular mortality, stroke (fatal and non-fatal), cardiac events (a composite of cardiac death and non-fatal myocardial infarction, excluding heart failure), heart failure, peripheral arterial disease, and chronic kidney disease. For inclusion in the analysis, studies were required to have a minimum of 50 participants with AF, a minimum of 50 participants without AF, and a median follow-up of at least six months. Importantly, we included only studies that reported associations for both men and women to restrict our analysis to comparisons of men and women within studies and to reduce the risk of heterogeneity between studies influencing our results. Additionally, we required all studies to adjust for, at a minimum, age and the presence of cardiovascular disease at baseline. We excluded studies that examined postoperative atrial fibrillation because the differing epidemiology and duration of postoperative atrial fibrillation relative to chronic atrial fibrillation.16 There were no language restrictions, and an investigator with extensive experience in epidemiological study translation (AJH) translated non-English studies. We contacted authors of studies that did not report separate associations for women and men to provide any unpublished data on adjusted sex specific associations.
Data were extracted, in duplicate, from studies deemed to meet the eligibility criteria. These included details on general study characteristics (study name, duration of follow-up, year of publication), information about the studied population (number of participants with and without AF, mean age, number of men and women, number of participants with a history of coronary heart disease, stroke, and heart failure), and information on the outcomes in the study (all cause mortality, cardiovascular mortality, stroke, cardiac events, heart failure, peripheral arterial disease, and chronic kidney disease). We extracted sex specific adjusted measures of relative risk (hazard ratios, relative risks, and odds ratio) and 95% confidence intervals. We used the maximally adjusted relative risk that was available and risk estimates corresponding to the longest period of follow-up. For cohorts that had multiple reports of the same outcome, we used the report with the largest number of events. Study quality was assessed with the Newcastle-Ottawa scale for cohort studies.17
For the primary analysis, we derived a ratio of relative risks with 95% confidence intervals of AF for each outcome in women compared with men, as previously described.8 This relative risk ratio for each study was then pooled with inverse variance weighted random effects meta-analysis. We also pooled relative risks for men and women separately. For one study, which reported separate hazard ratios for men and women in different age groups, we first used inverse variance weighted fixed effects meta-analysis to generate a summary hazard ratio for men and for women.18 We used funnel plots to examine if publication bias seemed to be present for outcomes that had at least 10 studies present (all cause mortality and stroke).19 If publication bias was present, we used the trim and fill method to adjust for publication bias.20 Heterogeneity was quantified with the I2 statistic and the Q test. P<0.05 was considered significant.
To estimate the difference in absolute risks associated with AF between women and men, we multiplied estimated sex specific excess incidence rates for all cause mortality,21 cardiovascular mortality,21 coronary heart disease21 (restricted to coronary heart disease mortality as an incidence rate for the composite of coronary heart disease death and non-fatal myocardial infarction could not be obtained), stroke,22 and heart failure23 in the United Kingdom general population by sex specific associations of each outcome with AF. The relative risk associated with each outcome in women was calculated by multiplying the pooled ratio of relative risks by the relative risk in men. We then subtracted the excess risk in men from the excess risk in women to estimate the difference in absolute risks associated with AF between men and women. Confidence intervals were derived through simulation with 10 000 draws from the distribution of the men’s relative risk and ratio of relative risks performed.
We undertook seven sensitivity analyses to determine if the ratio of relative risks in women versus men for mortality and stroke differed by methodological and study characteristics. Firstly, we stratified studies by whether AF was ascertained through electrocardiography at baseline or through medical records. Secondly, we stratified by region (Europe versus non-Europe). For these two stratified analyses (which stratified on categorical variables), we performed tests for interactions between subgroups. Thirdly, we stratified by size of the study (>100 000 versus <100 000 participants) to examine whether the results were consistent we excluded two large studies24 25 (each with more than 100 000 participants). Fourthly, we stratified by baseline year of enrolment of each cohort (1990 and before versus 1991 or later). Fifthly, we stratified by length of follow-up (≤10 versus >10 years). Sixthly, we stratified by median age (≤65 versus >65 at baseline). Finally, we stratified studies by the ratio of the event rate in women to men to examine whether our results were influenced by potentially lower absolute risk of outcomes in women than men. For the five analyses that stratified on continuous variables we performed tests for trend using meta-regression. The seven sensitivity analyses were restricted to stroke and all cause mortality because there were too few studies to conduct sensitivity analyses for other outcomes. All statistical analyses were conducted with R version 3.0.
No patients were involved in setting the research question or the outcome measures, nor were they involved in the design and implementation of the study. There are no plans to involve patients in dissemination.
The systematic search identified 3635 studies, of which 268 were examined in the full text review (fig 1⇓) and 238 were excluded. Accordingly, we included 30 cohort studies. None of these studies reported sex specific associations for either peripheral arterial disease or chronic kidney disease. Twenty studies provided published sex specific associations, while 10 studies provided unpublished associations.
Table 1⇓ shows the characteristics of the included studies. Ascertainment of AF status was through electrocardiography or continuous monitoring in 15 studies, medical records in eight studies, and a combination in six studies. One study did not report the method of ascertainment. Quality of included studies, as assessed by the Newcastle-Ottawa Scale, was high (appendix table A).
Studies included a total of 4 371 714 participants, 66 511 with AF (not counting one study34 with an overlapping cohort as a second study24). Twenty six studies reported the number of events. Of these, 14 studies reported on all cause mortality (18 563 events), four studies on cardiovascular mortality (7702 events), 11 studies reported on stroke (83 030 events), six studies on cardiac events (3583 events), and three studies on heart failure (27 468 events). Four additional studies reported relative risks for all cause mortality, cardiovascular mortality, and stroke associated with AF but did not report the number of events. Appendix table B shows pooled incidence rates for studies that reported number of events for men and women separately are provided.
Risk of cardiovascular disease and all cause mortality in women versus men
The pooled relative risk of all cause mortality for individuals with AF compared with those without AF was higher in women than in men (relative risk 1.69 (95% confidence interval 1.50 to 1.90) v 1.47 (1.32 to 1.65); fig 2⇓). The pooled ratio of relative risks for women versus men showed a 12% greater risk of all cause mortality associated with AF in women than in men (relative risk ratio 1.12, 95% confidence interval 1.07 to 1.17l; fig 3⇓). Although heterogeneity was observed in the relative risk of all cause mortality associated with AF in both women and men (I2=91% and I2=89%, respectively, P<0.001; fig 2⇓), no significant heterogeneity was observed in the pooled ratio of relative risks (I2=2%, P=0.43; fig 3⇓).
The relative risk of stroke was also greater in women than in men (relative risk 4.05 (95% confidence interval 2.52 to 6.50) v 1.77 (1.40 to 2.24); fig 4⇓). When we pooled the ratios of relative risks, AF was observed to be associated with twice the relative risk of stroke in women than in men (relative risk ratio 1.99, 95% confidence interval 1.46 to 2.71; fig 5⇓). There was significant heterogeneity between studies (I2=73%, P<0.001).
AF was associated with a higher relative risk of cardiovascular mortality in women than in men (relative risk ratio 1.93, 95% confidence interval 1.44 to 2.60; fig 6⇓; appendix fig A), with little heterogeneity observed between studies (I2=8%), and was associated with a 55% higher relative risk of cardiac events (cardiac death or non-fatal myocardial infarction) in women versus men (1.55, 1.15 to 2.08; fig 6⇓; appendix fig B). This was consistent when we restricted our analysis to three studies that reported only myocardial infarction, excluding the one study that reported sudden cardiac death30 (1.64, 1.15 to 2.34). AF was also associated with an increased risk of heart failure in women compared with men (1.16, 1.07 to 1.27; appendix fig C).
When we looked at events per 1000 patient years, corresponding absolute risk increases in outcomes associated with AF in women compared with men were 1.8 (95% confidence interval 1.1 to 2.6) for all cause mortality, 4.3 (1.9 to 7.5) for cardiovascular mortality, 3.1 (1.1 to 6.1) for stroke, 0.6 (0.03 to 1.3) for cardiac events, and 6.1 (2.1 to 12.7) for heart failure.
Ratios of relative risk for all cause mortality were broadly consistent in seven sensitivity analyses (appendix fig D). Results of tests of interaction were non-significant when we stratified studies by ascertainment of AF through electrocardiography versus use of medical records/self reported history and region (Europe, non-Europe; all P>0.05 for interaction). Results of tests for trend by number of participants, baseline year, and ratio of mortality rate in women to men were also non-significant. Ratios of relative risk, however, declined with increasing length of follow-up and increasing age (appendix fig D).
For ratios of relative risks for stroke, results of tests of interaction by ascertainment method and region were non-significant (appendix fig E). Results of tests for trend by number of participants, baseline year, length of follow-up, and median age were also non-significant. Relative risks were significantly higher in studies with a lower ratio of absolute rate of stroke in women compared with men (P=0.027 for trend for ratio of stroke rate in women to men). When we restricted the meta-analysis to three studies with a ratio of stroke rate in women to men greater than 1 (that is, a greater absolute rate of stroke in women than men), however, atrial fibrillation continued to be associated with a higher relative risk of stroke in women than men (1.47, 95% confidence interval 1.18 to 1.83).
We found no evidence of publication bias for all cause mortality (appendix fig F), though we observed significant publication bias for stroke (appendix fig G; P=0.002). When we used the trim and fill method to control for publication bias, AF was associated with a non-significantly higher risk of stroke in women than in men (relative risk ratio 1.28, 95% confidence interval 0.94 to 1.73).
In this systematic review and meta-analysis of 30 studies with 4 371 714 participants, we observed atrial fibrillation (AF) to be a significantly greater risk factor for death and cardiovascular disease in women than in men. AF was associated with a higher relative risk of all cause mortality, stroke, cardiovascular mortality, cardiac events, and heart failure in women compared with men.
Comparison with previous individual studies
Previous studies have presented conflicting evidence on the effect of AF on the risk of death and cardiovascular disease in women. Of the 19 studies on the association of AF with the risk of all cause mortality (fig 2⇑) we included in our analysis, the ratio of relative risks was greater than one but not significant for 11 studies. When we pooled these studies using random effects meta-analysis, the estimate for all cause mortality was significant, indicating that individual studies were underpowered to detect a differential effect of AF on risk of all cause mortality in women compared with men. Similarly, six of the 13 studies that reported a sex specific association of AF with stroke did not detect a significant interaction between women and men (fig 3⇑). When we pooled these studies, however, we observed a significant ratio of relative risks, again indicating a lack of power of some previous individual studies to detect an interaction by sex.
Recent meta-analyses have shown that type 2 diabetes is a greater risk factor for coronary heart disease7 47 (ratio of relative risk 1.44, 95% confidence interval 1.27 to 1.63) and for stroke8 (1.27, 1.10 to 1.46) in women than in men and that type 1 diabetes is also a greater risk factor for death and cardiovascular disease in women.48 Evidence of an increased risk of coronary heart disease associated with type 2 diabetes in women has been cited in European Society of Cardiology guidelines for the treatment of cardiovascular disease in patients with type 2 diabetes.49 Similarly, evidence of an increased risk of stroke associated with type 2 diabetes in women has been included in American Heart Association/American Stroke Association guidelines for stroke prevention in women.50 Our results show that AF is also a greater risk factor for death and cardiovascular disease in women than in men and extend these previous works showing that women experience the effects of some key risk factors for cardiovascular disease differently to men.
It is unclear what could cause the observed differences in risk of mortality and cardiovascular disease associated with AF between women and men. One possibility is that women with AF are undertreated relative to men. The results of a cohort study of Canadian patients with AF enrolled in 1990-94 support this hypothesis. Canadian women were half as likely as Canadian men to receive warfarin.51 Analyses of more contemporary cohorts, including a global registry of 17 814 patients with AF52 and an analysis of 83 513 patients with AF in Quebec,53 however, showed no differences in use of anticoagulants between men and women with AF.53 It is therefore unlikely that broad differences in treatment between the sexes are responsible for the increased relative risks we observed in women. Physiological or psychosocial differences between women and men could result in differential effects of AF on cardiovascular risk. For example, women are at a higher risk of torsade de pointes, an often lethal adverse event of antiarrhythmic drugs prescribed for AF.54 Response to oral anticoagulants could also differ between the sexes, with a higher risk of bleeding observed among women.55 Future research is needed to distinguish if one or many of these potential mechanisms underlie the differential effects of AF observed in our analysis.
Strengths and limitations
This analysis has several strengths. Firstly, as a systematic review and meta-analysis of all available studies of AF and risk of death and cardiovascular disease in women compared with men, it has greater power than any of the included individual studies to detect differences. This is evident in the meta-analysis of all cause mortality, in which 16 of the included 19 studies did not detect a difference. Secondly, we included only studies that reported the effect of AF on risk of cardiovascular disease and death separately in men and women. This ensured that our primary analysis (the ratio of relative risks) was a within study comparison, minimising the effect of heterogeneity between studies. Thirdly, we included only cohort studies that were adequately adjusted (prespecified requirement to adjust by at least age and history of cardiovascular disease) and that had a minimum of 100 participants with six months’ follow-up to reduce the risk of confounding and small study effects. Fourthly, pooled ratios of relative risk for all cause mortality or stroke were broadly consistent in several different sensitivity analyses.
This analysis also has several limitations. Firstly, while we attempted to contact and acquire unpublished data from eligible cohorts, our results might be influenced by publication bias because studies that detect an interaction between AF and risk of death and cardiovascular disease by sex might be more likely to be published. While there was evidence of publication bias for stroke, however, we found no evidence of this for all cause mortality. Furthermore, use of trim and fill procedures resulted in a non-significantly increased effect of AF on risk of stroke in women compared with in men similar in magnitude to other outcomes. Secondly, as a meta-analysis of observational studies, sex differences in the association of AF with risk of death and cardiovascular disease might be caused by unobserved confounding between sexes. For example, women might have had a greater number of comorbidities at the time of diagnosis of AF, which could not be fully adjusted for. However, we required included studies to adjust for, at minimum, age and presence of cardiovascular comorbidities, and we used the maximally adjusted model available. Thirdly, many of the studies had differences in design, duration of follow-up, outcome ascertainment, and populations. Indeed, we observed a greater absolute increase in cardiovascular death than all cause death because of differing studies contributing to either outcome. Because our primary analysis was focused on comparisons within studies (the ratio of relative risks between women and men), however, heterogeneity was low for outcomes other than stroke. Fourthly, our primary analysis focused on the ratio of relative risks, which might not reflect absolute differences in risk of death and cardiovascular disease associated with AF between sexes. We found no evidence of a difference in ratios of relative risk in studies in which the absolute risk of all cause mortality was higher in men (appendix fig D), and a higher ratio of relative risks continued to be observed in studies in which women had a higher absolute rate of stroke than men (appendix fig E). Finally, we were unable to ascertain the underlying cause of the sex differences in the association of AF with risk of mortality and cardiovascular disease.
Implications for clinicians, policy makers, and future research
This analysis has several implications. With respect to clinical care, our results indicate that AF is associated with worse outcomes in women than in men. Although female sex is incorporated as a risk factor for stroke in the widely used CHA2DS2-VASc score,56 AF seems to affect women and men differently. The American Heart Association recently recommended the development of a specific risk score for stroke in women as some risk factors for stroke are unique to women, others are more prevalent in women, and others differentially increase the risk of stroke in women.57 These results support the development of such a score. They also, however, show that AF is associated with an increased risk of all cause mortality, cardiovascular mortality, and cardiac events in women relative to men. Therefore, it might be appropriate for clinicians to consider more aggressive treatment of risk factors in women with AF as they seem to be at higher proportional risk of death and cardiovascular disease.
With regard to public health policy, these results indicate that sex differences in the effect of AF on risk of death and cardiovascular disease exist. Consequently, estimation of the global and regional burden of AF should be independent of sex. Allocation of public health resources for prevention and treatment of AF should also consider the differential effects of AF by sex. Future research should be encouraged to determine the underlying causes of the observed sex differences.
What is already known on this topic
Atrial fibrillation (AF) is associated with an increased risk of stroke and death in men and women
A growing body of literature suggests that women and men experience risk factors for cardiovascular disease differently, including diabetes, smoking, and body mass index
What this paper adds
AF is associated with a stronger relative risk of all cause mortality, cardiovascular mortality, stroke, ischaemic heart disease, and heart failure in women than in men
Whether the differential association of AF with death and cardiovascular disease in women relative to men is causal is unclear
We thank the following for contributing unpublished data: Julia Hippisley-Cox (QRisk); Janine Gronewold and Dick Hermann (Heinz Nixdorf Recall Study); Ale Algra (Life Long After Cerebral Ischaemia study); Charles Wolfe and Siobhan Crichton (South London Stroke Register and Barbados Register of Strokes, funded by the National Institute for Health Research (NIHR) Biomedical Research Centre based at Guy’s and St Thomas’ NHS Foundation Trust and King’s College London); Konstantinos Siontis, Bernard Gersh, and Jeffrey Geske (Mayo Clinic); Alberto Bouzas-Mosquera (Madrid, Spain); Gustavo Saposnik (Registry of the Canadian Stroke Network); Yannick Bejot (stroke registry of Dijon, France); Veronique Roger and Alanna Chamberlain (Olmsted County); and Emanuela Rossi and Simonetta Genovesi (Lombardi, Italy). Your contributions are greatly appreciated.
Contributors: CAE, CXW, and AAO were involved in the design, implementation, and analysis of the study and in writing the final manuscript. AJH was involved in the implementation of the study and in commenting on drafts of the final manuscript. MW, SAEP, and DGA were involved in the analysis of the study and in commenting on drafts of the final manuscript. CAE is guarantor.
Funding: This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. CAE, CXW, AJH, and AAO are funded by Rhodes Scholarships. CXW is also supported by a Neil Hamilton Fairly Fellowship from the National Health and Medical Research Council of Australia. No funders were involved in the design, implementation, analysis, or reporting of this study.
Ethical approval: Not required.
Competing interests: All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf. MW declares consultancy fees from Amgen and Novartis. All other authors declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work.
Data sharing: Data and code are available from the lead author on request.
Transparency: The lead author (the manuscript’s guarantor) affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.
This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 3.0) 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 and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/3.0/.