Objectives Higher body mass index (BMI) is an important risk factor for atrial fibrillation (AF). The adipokines leptin, adiponectin and resistin are correlates of BMI, but their association with incident AF is not well known. We explored this relationship in a large cohort of postmenopausal women.
Methods We studied an ethnically diverse cohort of community-dwelling postmenopausal women aged 50–79 who were nationally recruited at 40 clinical centres as part of the Women's Health Initiative investigation. Participants underwent measurements of baseline serum leptin, adiponectin and resistin levels and were followed for incident AF. Adipokine levels were log transformed and normalised using inverse probability weighting. Cox proportional hazard regression models were used to estimate associations with adjustment for known AF risk factors.
Results Of the 4937 participants included, 892 developed AF over a follow-up of 11.1 years. Those with AF had higher mean leptin (14.9 pg/mL vs 13.9 pg/mL), adiponectin (26.3 ug/mL vs 24.5 ug/mL) and resistin (12.9 ng/mL vs 12.1 ng/mL) levels. After multivariable adjustment, neither log leptin nor log adiponectin levels were significantly associated with incident AF. However, log resistin levels remained significantly associated with incident AF (HR=1.57 per 1 log (ng/mL) increase, p=0.006). Additional adjustment for inflammatory cytokines only partially attenuated the association between resistin and incident AF (HR=1.43, p=0.06 adjusting for C-reactive protein (CRP); HR=1.39, p=0.08 adjusting for IL-6). Adjusting for resistin partially attenuated the association between BMI and incident AF (HR=1.14 per 5 kg/m2, p=0.006 without resistin; HR=1.12, p=0.02 with resistin).
Conclusions In women, elevated levels of serum resistin are significantly associated with higher rates of incident AF and partially mediate the association between BMI and AF. In the same population, leptin and adiponectin levels are not significantly associated with AF.
Statistics from Altmetric.com
Atrial fibrillation (AF) is the most prevalent cardiac arrhythmia. There are approximately 3.0 million people in the USA currently afflicted, and this number is expected to rise to 7.6 million by 2050.1 AF accounts for an increased risk of mortality even after adjustment for numerous cardiovascular risk factors and comorbidities.2 Women with AF have higher rates of stroke and overall mortality than men with AF2 ,3 making the early identification of AF even more imperative in this population. Obesity has recently been established as an important independent risk factor for incident AF.4 ,5 The mechanisms of this association, however, have not been well established. It has been postulated that adiposity may impact AF risk through the secretion of adipocyte-derived cytokines (adipokines) such as leptin, adiponectin and resistin. Each of these adipokines has been demonstrated to correlate with body mass index (BMI) and to have various effects on cardiovascular physiology.6–8 The association between these adipokines and AF, however, has been poorly characterised, particularly with respect to resistin.
Given these observations, we aimed to further explore the links between leptin, adiponectin, resistin and AF by studying their associations in a large cohort of ethnically diverse postmenopausal women. Additionally, we were interested in determining whether or not these adipokines play a role in mediating the relationship between BMI and incident AF.
The methods of women's health initiative (WHI) recruitment and follow-up are described elsewhere in detail.9 ,10 Briefly, 93 676 women were enrolled in the WHI observational study (OS) and 68 132 women were enrolled in at least one of the randomised control trials (CTs) of hormone therapy, dietary modification and calcium/vitamin D, between October 1993 and December 1998. Eligibility criteria for the OS and CT included being postmenopausal, 50–79 years of age, willing to complete study activities, and having ability to provide written informed consent with an intention to reside in the area for at least 3 years after enrolment. Participants were followed for cardiovascular disease (CVD) outcomes, including new-onset AF. Women from the WHI with a history of ischaemic stroke, colorectal cancer, breast cancer, endometrial cancer, coronary heart disease and hypertension (HTN) were randomly selected for measurement of baseline leptin, adiponectin and resistin levels as previously described.10–14 Participants with AF at baseline, as reported on the initial questionnaire or as evidenced on baseline ECG, those who had missing clinical data, and those lacking adipokine measurements were excluded.
Assessment of baseline characteristics
Participants completed self-administered or interview-administered questionnaires at study enrolment. Questionnaires assessed patient race/ethnicity, income, education level, history of HTN, congestive heart failure (CHF), diabetes mellitus (DM), hyperlipidaemia (HLD), coronary artery disease (CAD), stroke, peripheral arterial disease (PAD), smoking and alcohol use. Participants underwent measurement of baseline blood pressure, height and weight. BMI was calculated as weight (kg) divided by the square of measured height (m2). Alcohol use was defined by consumption of at least 12 alcoholic beverages in the participants' lifetime and information about ongoing alcohol use was obtained via questionnaires. Participants had fasting blood samples collected and sent to local laboratories where they were centrifuged and frozen to −70°C for storage. The WHI was overseen by ethics committees at all 40 clinical centres, by the coordinating centre and by a data and safety monitoring board. Each institution obtained human subjects committee approval. Each participant provided written informed consent. Further specifics about study design, study questionnaires, physical measurements, blood collection and quality assurance have been described previously.9 ,10 ,12 ,15
Ascertainment of incident AF
Women were followed annually with clinic visits and with either clinic visits or telephone calls in between annual visits. At each contact, women underwent a standardised interview and were asked about the development of AF and interim hospitalisations. Additionally, follow-up 12-lead ECG testing was conducted every 3 years after baseline. In the event of hospitalisation, medical records were reviewed and the implantable cardioverter defibrillator (ICD)-9 code for AF (427.31) was extracted. WHI data were linked with the Centers for Medicare and Medicaid Services data using social security numbers, birth dates and death dates, with 97% of Medicare-eligible WHI participants successfully linked. Among participants with Medicare coverage, incident AF was identified by first occurrence of ICD-9 code 427.31 in any diagnosis position within the inpatient (medicare provider analysis and review (MEDPAR)), outpatient and carrier files during years 1994–2011. Women with self-reported new-onset AF, AF on follow-up ECG or AF discovered upon review of hospital records were classified as having incident AF. Additional details about AF ascertainment within the WHI population have been previously described.4
Measurement of adipokine levels
Fasting blood samples from subjects obtained at baseline were stored at −70°C at the central repository and used to assess for concentrations of adiponectin, leptin, resistin, IL-6 and CRP. Plasma levels of total adiponectin, leptin, resistin and CRP were measured at different laboratories, with the majority of the measurements obtained via a multiplex assay (human adipokine panels A and B, respectively, Millipore, Billerica, Massachusetts, USA) based on Luminex technology (http://www.luminexcorp.com); the interassay coefficients of variation were 11.3% for adiponectin, 5.3% for leptin and 11.4% for resistin.13–15 The intraclass correlation coefficients of these adipokines were published previously.16 High-sensitivity CRP was measured by immunoturbidity initially, and then by immunonephelometry.15 ,17 IL-6 was measured by enzyme-linked immunosorbent assay (R&D Systems, Minneapolis, Minnesota, USA).15 Further details of laboratory adipokine measurement methods have been reported previously.13–15 ,17
Baseline characteristics of participants were compared across AF status, using t test for continuous variables or the χ2 test for categorical variables. The associations of incident AF with baseline levels of leptin, adiponectin, resistin and were measured using multivariable Cox proportional hazard regression models. The models were adjusted for covariates that have known or suspected associations with AF, including age, race/ethnicity, education, HTN, HLD, BMI, DM, CAD, CHF, PAD, cancer history, ischaemic stroke, alcohol use and smoking status. Additionally, all models included a time-dependent covariable for Medicare coverage.
To adjust for differences in measurement methods and in the distribution of adipokine levels among different study sites and laboratories, the adipokine values were normalised using inverse probability weighting. Additionally, given that adipokine levels were not normally distributed, they were log transformed and divided into quartiles for the purposes of analyses. Primary analyses were conducted to assess the association between continuous log-transformed adipokine levels and incident AF. Secondary categorical analyses were performed comparing quartiles of adipokine levels, using the first quartile as a reference group. Linear trends across incremental quartiles were also estimated. Associations were reported as HRs with 95% CIs. A p value of <0.05 was considered statistically significant.
To determine if BMI was a potential confounder of the associations between the adipokines and incident AF, multivariable analyses were performed with and without BMI as a covariate. Cubic spline linear regression analyses, using three knots, were performed to explore the presence of inflection points in the continuous relationships between adipokine levels and incident AF. If the spline regression analysis suggested the presence of a non-linear relationship, sensitivity analyses were performed via piecewise modelling. A separate multivariable Cox hazard regression analysis was also conducted using CRP levels to confirm the known association between this inflammatory marker and AF in our cohort. Additional analyses were conducted to evaluate the association between resistin and incident AF by adjusting for CRP and separately IL-6 as potential confounders.
To determine if resistin mediated the association between BMI and AF, the Pearson correlation between BMI and log resistin levels was first measured within our cohort. Following, resistin was added as a covariate to our primary multivariable Cox hazard regression model assessing the association between BMI and incident AF. A significant reduction in the HR after this addition was interpreted as evidence of mediation.
Subgroup analyses were also performed to determine the effect of serum adipokine levels on AF stratified by the following factors: age, ethnicity, BMI, HTN, CAD, DM, smoking status and alcohol use. Interactions between adipokine levels and subgroup variables were assessed by adding the interaction terms to the multivariable Cox hazard regression model. The analyses were performed using SAS statistical software (V.9.1; SAS Institute, Cary, North Carolina, USA).
Of the 6023 women enrolled in the Women's Health Initiative who had leptin, adiponectin and resistin levels measured, 340 women with baseline AF and 746 women with missing data were excluded, leaving 4937 participants available for analysis. Over an average follow-up time of 11.1 years, 892 women developed new-onset AF. Compared with women without incident AF, those who developed AF were older (68.9 years vs 65.0 years, p<0.001), more overweight (BMI 28.1 vs 27.6, p=0.03) and had higher rates of HTN (61.9% vs 47.1%, p<0.001), CAD (3.7% vs 1.6%, p<0.001), PAD (2.8% vs 1.6%, p=0.01) and CHF (0.6% vs 1.5%, p=0.005). Detailed patient demographics can be found in table 1.
Leptin, adiponectin, resistin levels and risk of AF
Continuous log leptin values were not significantly associated with incident AF in neither age/race-adjusted nor multivariable-adjusted models when analysed both as a continuous variable and divided into quartiles (table 2). The multivariable-adjusted cubic spline regression curve of log leptin and the risk of incident AF did reach statistical significance (p=0.03) (figure 1A). Given suggestion of a non-linear relationship between leptin and incident AF risk around 3.1 log leptin, however, a piecewise model looking at different parts of the spline curve was performed but was not statistically significant (p=0.32<3.1 and p=0.67>3.1).
Log adiponectin levels were likewise not associated with incident AF when analysed continuously or in quartiles in both age/race and fully adjusted multivariate models (table 2). Multivariate spline regression analysis of log adiponectin levels and the risk of incident AF confirmed the absence of a significant relationship (p=0.38) (figure 1B). A piecewise model was not conducted given the absence of obvious inflection points in the spline curve.
In contrast, continuous log resistin levels were associated with a higher incidence of AF after multivariate adjustment, including BMI (HR=1.57 per 1 log (ng/mL) increase, p=0.006). When analysed categorically, compared with women in the lowest quartile of resistin, those in the highest quartile had a significantly elevated multivariate risk of developing incident AF (HR=1.48, 95% CI 1.10 to 1.97; trend, p=0.05). There was additionally a statistically significant positive linear trend across increasing resistin quartiles with an HR of 1.14 for each increase in quartile (p=0.005). Multivariable-adjusted cubic spline regression analysis of log resistin levels and the risk of incident AF was statistically significant (p=0.007) (figure 1C). A piecewise model likewise demonstrated a statistically significant relationship between resistin levels and AF beyond 2.1 log resistin (p<0.001).
In our sensitivity analyses assessing the role of obesity in the resistin–AF relationship, removing BMI from the multivariable-adjusted model modestly attenuated the association between continuous log resistin and AF (HR=1.57 with BMI; HR=1.66 without BMI). Similar findings were obtained in evaluating the role of inflammatory markers in the resistin–AF relationship. Although the relationship between continuous log CRP levels and AF risk was robust in our cohort (multivariable-adjusted HR=1.17 per 1 log (mg/L) increase, p=0.03), introducing CRP into the multivariable-adjusted resistin–AF model resulted in only a modest attenuation of the AF risk associated with resistin (HR=1.43 (95% CI 0.99 to 2.06), p=0.06). The impact of adding IL-6 to the model likewise resulted in only modest reduction of the AF risk (HR=1.39 (95% CI 0.96 to 2.03), p=0.08). Pearson correlations for resistin, leptin and adiponectin with CRP were r=0.19, 0.27 and −0.15, respectively; and with IL-6 were r=0.20, 0.25 and −0.15, respectively.
BMI mediation analysis
Consistent with prior results in the entire WHI cohort, BMI was associated with an increased risk of incident AF in the present subset of patients after multivariate adjustment (HR=1.14 (1.04 to 1.26) per 5 kg/m2, p=0.006) (figure 2). Pearson correlations with BMI were 0.17, 0.64, −0.26, 0.30 and 0.28 for resistin, leptin, adiponectin, CRP and IL-6, respectively. After introducing log resistin levels into the multivariable model assessing the BMI–AF relationship, there was a small decrease in the association between BMI and incident AF (HR=1.12 (1.02 to 1.24) per 5 kg/m2, p=0.02). This decrease in the AF risk suggests that resistin partially mediates the association between BMI and AF, although an apparently small effect.
Associations between each adipokine and AF risk stratified by selected covariates are shown in figure 3. There were no statistically significant interactions between any of the examined covariates and the association between leptin and incident AF (figure 3A). The highest quartile of adiponectin levels conferred a greater risk of incident AF in participants with DM (HR=6.56, 95% CI 2.3 to 18.9, interaction p=0.004), although the number of AF cases is relatively small in the upper quartile of adiponectin and the CI is wide (figure 3B). Additionally, the highest quartile of resistin levels was associated with a significantly higher risk of AF development in patients who are current alcohol users (HR=1.86, 95% CI 1.3 to 2.8, interaction p=0.005) (figure 3C).
In this prospective cohort study of postmenopausal women from the WHI, we observed a significant association between resistin levels and incident AF after adjustment for multiple covariates including BMI and inflammatory biomarkers. Resistin levels were also found to partially mediate the relationship between obesity and AF, although the mediation effect was relatively small and the biological relevance of this mediation as well as its potential for targeted clinical intervention is unclear. We did not find an association between leptin and adiponectin with incident AF.
Obesity is an important independent risk factor for AF that is steadily becoming more prevalent in our population. The link between obesity and AF is complex and incompletely understood but various mechanisms have been proposed to explain the association. These mechanisms include atrial remodelling from obesity-related HTN and obstructive sleep apnoea, concomitant insulin resistance and hyperglycaemia affecting atrial tissue conduction properties, obesity associated systemic and localised inflammation and changes in myocardial energetics induced by obesity.18 An additional mechanistic explanation has been proposed through the direct effect of various adipocyte-derived peptides on cardiovascular physiology. The adipokines leptin, adiponectin and resistin are attractive candidates for this role as their levels are known to be significantly impacted by obesity and their effects on various cardiovascular properties have been well documented.6 ,7 ,8 Adipokines may, therefore, serve as additional tools for the risk stratification of patients in regards to the development of AF. Whether adipokine levels add meaningful prognostic information or improved reclassification of an individual's absolute risk for AF beyond conventional risk assessment methods remains to be determined.
A major finding of our study is the significant positive relationship between elevated resistin levels and the risk of developing incident AF. Although this association has been observed in prior studies, it is still controversial. One of the first reports of the link between resistin and AF comes from a study of 40 patients post coronary artery bypass graft surgery in which elevated resistin levels preoperatively were associated with increased rates of postoperative AF.19 Following, Rienstra et al reinforced these findings in 2847 subjects from the Framingham cohort followed >10 years.20 In their analysis, patients with elevated baseline levels of resistin were found to be at a significantly higher risk of developing new-onset AF (HR=1.17 per SD increase in log-transformed resistin). CRP was shown to attenuate this relationship. Since then, an additional case–control study of 100 patients with AF demonstrated that patients with paroxysmal and permanent AF had higher serum resistin levels than matched controls.21 Most recently, however, the association between resistin and AF was demonstrated to be less robust in a study conducted by Muse et al.22 This study focused on a sex-balanced, ethnically diverse, cohort of 1913 participants followed >5.5 years for CVD outcomes including incident AF. While resistin levels were shown to be closely associated with other forms of CVD, there was no association identified between resistin and AF. Factors that may have accounted for the disparate findings between this study and our own include a relatively low AF event rate (2.7%), shorter follow-up period, and a more ethnically diverse population in the Muse et al study.
The mechanism by which resistin increases AF risk is incompletely understood but several hypotheses exist. First, elevated resistin levels have been demonstrated in a number of conditions known to be associated with the development of AF such as CHF, CAD and HTN,7 however, adjusting for each condition in our multivariable analysis did not dramatically alter the increased risk of AF associated with higher resistin levels. Second, resistin is implicated in systemic inflammation as it is secreted by leucocytes in human beings23 and strongly correlated with inflammatory markers such as IL-6 and CRP.24 Given the known elevated risk of AF in inflammatory states,25 inflammation may serve as a possible mechanistic connection between resistin and AF risk. Supporting these prior observations, CRP was found to be significantly associated with AF risk within our cohort. However, adjusting for CRP and IL-6 only partially attenuated the effect of resistin on AF risk, implying that, while these inflammatory markers are confounders of this relationship, resistin influences AF risk through pathways additional to inflammation. Finally, it may be that resistin has direct effects on the heart and its electrical properties. In a recent study, resistin was shown to directly modulate electrophysiological properties of cardiac myocytes making them more prone to isoproterenol induced arrhythmias.26 Additional studies have demonstrated that resistin depresses cardiac contractility, promotes cardiac hypertrophy, and has effects on vasculature through the promotion of smooth muscle proliferation.7
Given the positive association of resistin with AF, we were interested in learning whether this association could explain the increased AF risk conferred by obesity. The relationship between obesity and AF was confirmed in our cohort; and we verified that BMI and resistin levels were positively correlated. Introducing resistin into the multivariable model assessing the association between BMI and AF resulted in only a modest attenuation of AF risk (≈2% difference in relative risks). This finding suggests that resistin may partially mediate the increased AF risk conferred by obesity, but is clearly not the sole contributor.
An interesting finding from our subgroup analysis was the significant association between elevated resistin levels and AF in current alcohol users. Alcohol use is a well-known risk factor for AF and has also been associated with elevated resistin levels in rats and human beings.27–29 In our study, there was no difference in the proportion of alcohol users between participants who developed AF and those who did not, suggesting that measurement of resistin levels may be a good way of predicting AF risk within this population. Additionally, resistin may prove to be a link in the mechanism by which alcohol predisposes patients to AF.
Prior studies investigating the relationship between adiponectin and AF have yielded conflicting results. One of the first studies to look at this association did so in a cross-sectional analysis of 304 hospitalised patients.30 In this study, Shimano et al found that patients with permanent AF had relatively higher levels of adiponectin and those with paroxysmal AF had lower levels compared with controls. An additional hospital-based case–control trial of 60 patients by Choi et al demonstrated similar findings of lower adiponectin concentrations in patients with paroxysmal AF and concentrations similar to controls in those with permanent AF.31 In contrast, the analysis by Rienstra et al described earlier, found no clear association between adiponectin and AF risk after multivariable adjustment.20 The findings of our study are most closely in line with those of Rienstra et al as is the size of our cohort and the methodology employed. Reasons for the differences between our findings and those of the earlier studies include variations in mean adiponectin levels, the study of a larger, community-based cohort and possible differences in the particular form of adiponectin measured. Given the lack of evidence connecting adiponectin to AF within large cohorts, it is unlikely that this adipokine plays a significant direct role in mediating the relationship between obesity and AF within the general population.
There have been no prior studies of the association between leptin and AF in human beings. One study looking at the role of leptin in AF pathogenesis in mice, demonstrated that leptin-deficient knockout mice did not develop atrial interstitial fibrosis and subsequent AF inducibility in response to infusions of angiotensin-II compared with controls.32 These findings suggested that leptin could play a role in the pathogenesis of atrial fibrosis seen in certain patients with AF. Our findings, however, failed to demonstrate a direct association between leptin levels and AF.
Our study has the strengths of being derived from a large, community-based cohort of women not selected on the basis of factors related to clinical CVD, significant duration of follow-up, a relatively large number of incident AF events for analysis, and a well-standardised method of clinical data collection. However, there are several limitations that warrant mention. One of the main limitations is the fact that our cohort was restricted to postmenopausal women with specific comorbidities and demographic makeup. While this can limit the generalisability of our results, there is no known reason to suspect it threatens the internal validity of study findings which have been mirrored in populations of women and men with diverse age characteristics such as the Framingham cohort.20 Additionally, it is possible that we may have been underpowered to detect a small association between leptin or adiponectin and incident AF. However, ours is one of the largest studies of adipokine levels with a large number of incident AF cases. It is also reassuring that established relationships between AF and biomarkers such as CRP were confirmed within our cohort.
In our community-based cohort of postmenopausal women, elevated resistin levels were associated with an increased risk of AF even after adjustment for BMI and other known AF risk factors as well as the proinflammatory cytokines CRP and IL-6. Resistin appears to partially mediate the association between BMI and AF, the clinical relevance of which requires clarification. Neither leptin nor adiponectin were significantly associated with increased rates of incident AF in our general cohort. Our findings suggest that resistin may play a role in the pathogenesis of AF and could potentially serve to enhance AF risk prediction in older women. Further studies are necessary to better characterise this association and its clinical utility.
What is already known on this subject?
Obesity is a well-known risk factor for the development of atrial fibrillation in women. It has been hypothesised that the risk of atrial fibrillation may be influenced by various cytokines that are known to be correlates of body mass index. Leptin, resistin and adiponectin are particularly attractive candidates for this role as their levels correlate with the degree of obesity and each has been demonstrated to have a number of effects on cardiovascular physiology. Their role in atrial fibrillation, however, has not been well documented.
What might this study add?
This prospective population-based study followed 4937 community-dwelling postmenopausal women from the Women's Health Initiative database and demonstrated that elevated baseline resistin levels are significantly associated with the risk of incident atrial fibrillation even after controlling for known risk predictors including body mass index. Women in the highest quartile of resistin levels were 25% more likely to develop incident atrial fibrillation compared with women in the lowest quartile. Leptin and adiponectin levels did not impact the rate of incident atrial fibrillation in this population.
How might this impact on clinical practice?
Better understanding the role of these adipokines in the mechanism of atrial fibrillation may lead to improved risk prediction methods which may be used to provide personalised therapeutic recommendations for patients. Further exploration of the underlying mechanism of obesity associated atrial fibrillation may additionally lead to the development of novel and more targeted therapeutic agents.
The authors thank the Women's Health Initiative investigators, staff and study participants for their outstanding dedication and commitment. A full listing of Women's Health Initiative investigators can be found at https://http://www.whi.org/researchers/Documents%20%20Write%20a%20Paper/WHI%20Investigator%20Long%20List.pdf.
Contributors MVP and SE were responsible for conception and design of the study as well as the interpretation of the data and drafting of the manuscript. JCL was responsible for the statistical analyses and compilation of the tables and figures. The remaining authors (FA, MLS, MJL, WL, KMT, LWM, RN, ES-B, CMA, JEM, TLA and MAH) contributed through scientific discussion at various stages of the study development, careful review of the manuscript and significant editorial comments. SE is the corresponding author. All authors take responsibility for all aspects of the reliability and freedom from bias of the data presented and their discussed interpretation.
Funding The WHI programme is funded by the National Heart, Lung, and Blood Institute, National Institutes of Health, US Department of Health and Human Services through contracts HHSN268201100046C, HHSN268201100001C, HHSN268201100002C, HHSN268201100003C, HHSN268201100004C and HHSN271201100004C.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
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.