This article requires a subscription to view the full text. If you have a subscription you may use the login form below to view the article. Access to this article can also be purchased.
Abstract
Background and objectives The incidence of atrial fibrillation is high in ESRD, but limited data are available on the incidence of atrial fibrillation across a broad range of kidney function. Thus, we examined the association of eGFR and urine albumin-to-creatinine ratio with risk of incident atrial fibrillation.
Design, setting, participants, & measurements We meta-analyzed three prospective cohorts: the Jackson Heart Study, the Multi-Ethnic Study of Atherosclerosis, and the Cardiovascular Health Study. Cox regression models were performed examining the association of eGFR and urine albumin-to-creatinine ratio with incident atrial fibrillation adjusting for demographics and comorbidity. In additional analyses, we adjusted for measures of subclinical cardiovascular disease (by electrocardiogram and cardiac imaging) and interim heart failure and myocardial infarction events.
Results In the meta-analyzed study population of 16,769 participants without prevalent atrial fibrillation, across categories of decreasing eGFR (eGFR>90 [reference], 60–89, 45–59, 30–44, and <30 ml/min per 1.73 m2), there was a stepwise increase in the adjusted risk of incident atrial fibrillation: hazard ratios (95% confidence intervals) were 1.00, 1.09 (0.97 to 1.24), 1.17 (1.00 to 1.38), 1.59 (1.28 to 1.98), and 2.03 (1.40 to 2.96), respectively. There was a stepwise increase in the adjusted risk of incident atrial fibrillation across categories of increasing urine albumin-to-creatinine ratio (urine albumin-to-creatinine ratio <15 [reference], 15–29, 30–299, and ≥300 mg/g): hazard ratios (95% confidence intervals) were 1.00, 1.04 (0.83 to 1.30), 1.47 (1.20 to 1.79), and 1.76 (1.18 to 2.62), respectively. The associations were consistent after adjustment for subclinical cardiovascular disease measures and interim heart failure and myocardial infarction events.
Conclusions In this meta-analysis of three cohorts, reduced eGFR and elevated urine albumin-to-creatinine ratio were significantly associated with greater risk of incident atrial fibrillation, highlighting the need for further studies to understand mechanisms linking kidney disease with atrial fibrillation.
Introduction
Atrial fibrillation (AF) is the most common sustained arrhythmia in the general population. The prevalence of AF is particularly high among patients with kidney disease, occurring in 7%–20% of those with ESRD on dialysis, which is two- to threefold higher than in the general population (1–3). The US Renal Data System reports that the prevalence of AF has increased over time among patients with ESRD (4).
Several studies (5–17) have reported a high incidence and prevalence of AF among the larger population of patients with CKD not yet requiring dialysis. One recent study estimated the prevalence of AF to be 18% in a multicenter cohort of participants with CKD (10). An analysis of over 10,000 middle-aged and older white and black participants from the Atherosclerosis Risk in Communities (ARIC) Study found a strong graded association between lower eGFR and higher urine albumin-to-creatinine ratio (UACR) with risk of incident AF (18). AF affects selection of cardiovascular (CV) therapies (19) and is associated with poor outcomes (20,21); thus, better characterization of risk factors for incident AF is important.
Prior studies of kidney function in relation to AF risk have notable limitations. Most general population studies have not been adequately powered to examine persons with CKD. Other studies have estimated GFR using only serum creatinine rather than the more accurate combined serum creatinine and cystatin C equation (22). Furthermore, few studies have comprehensively examined whether the association of eGFR with AF is modified by the extent of UACR and conversely, whether the association of UACR with AF is modified by eGFR. Most longitudinal studies have not accounted for the contribution of subclinical cardiovascular disease (CVD) or interim CV events. It is well established that patients with CKD are at higher risk of conduction abnormalities and abnormal cardiac structure/function as well as events, such as heart failure and myocardial infarction, which are important risk factors for the development of AF. Finally, important differences in risk of incident AF have been noted in particular patient subgroups by demographics and comorbidity, which have not been adequately addressed by prior studies (23).
To address these questions, we performed a meta-analysis of three diverse community-based cohorts with a broad range of kidney function (as determined by eGFR calculated from serum creatinine and cystatin C) to examine the associations of eGFR and UACR with risk of incident AF. We hypothesized that there are strong, graded associations of lower eGFR and higher UACR with risk of incident AF, even after accounting for a broad range of confounders and across subgroups by demographics and comorbidity.
Materials and Methods
Study Populations
Three community-based cohorts were included in this analysis: the Jackson Heart Study (JHS), the Multi-Ethnic Study of Atherosclerosis (MESA), and the Cardiovascular Health Study (CHS). Each site’s institutional review board approved the study, and all participants provided informed consent. For the meta-analysis, we included a total of 16,769 participants from the three study cohorts (Supplemental Figure 1).
The JHS is a community-based cohort study of blacks designed to evaluate risk factors for CVD (24,25). Participants (n=5306) ages 21–94 years old were recruited during 2000–2004 from the tricounty region (Hinds, Madison, and Rankin) of metropolitan Jackson, Mississippi (26). For our study, participants were excluded if they had prevalent AF on study entry (n=67) or were missing baseline measures of serum creatinine or cystatin C needed to calculate eGFR (n=173). This left a final analytic sample of 5066 participants (Supplemental Figure 1).
The MESA is a community-based cohort study of clinical and subclinical CVD (27). Between 2000 and 2002, the MESA enrolled 6814 adults 45–84 years of age from six field centers (New York and Bronx Counties, New York; Baltimore City and County, Maryland; Forsyth County, North Carolina; Chicago, Illinois; St. Paul, Minnesota; and Los Angeles, California). Only individuals without known prevalent CVD, defined as myocardial infarction, angina, stroke, transient ischemic attack, heart failure, use of nitroglycerin, prior angioplasty, coronary artery bypass grafting, valve replacement, pacemaker or defibrillator implantation, or any surgery on the heart or arteries, were eligible to participate. For our study, participants were excluded if they had prevalent AF on study entry (n=69), were missing baseline measures of serum creatinine or cystatin C needed to calculate eGFR (n=63), or had unavailable follow-up after the baseline study visit (n=5). This left a final analytic sample of 6677 participants.
The CHS is a cohort study of older community-dwelling adults ages 65 years old or older (28). Participants were recruited from Health Care Financing Administration Medicare eligibility lists at four locations: Forsyth County, North Carolina; Sacramento County, California; Washington County, Maryland; and Pittsburgh, Pennsylvania. An initial 5201 participants were recruited between 1989 and 1990. An additional 687 black participants were added to the study in 1992 and 1993. For our study, participants were excluded if they had prevalent AF on study entry (n=162) or were missing baseline measures of serum creatinine or cystatin C needed to calculate eGFR (n=700). This left a final analytic sample of 5026 participants.
Exposures
For all three cohorts, eGFR was calculated from serum concentrations of creatinine and cystatin C measured at baseline using the 2012 Chronic Kidney Disease Epidemiology Collaboration equation (22). CKD was defined as eGFR<60 ml/min per 1.73 m2. In the JHS, serum creatinine was measured using the Jaffe method (29). In the MESA, serum creatinine was measured by rate reflectance spectrophotometry using thin film adaptation of the creatine amidinohydrolase method on the Vitros analyzer (Johnson & Johnson Clinical Diagnostics Inc.). In the CHS, creatinine was measured using a colorimetric method (Ektachem700; Eastman Kodak, Rochester, NY) (30). In the JHS and the MESA, serum cystatin C was measured by a particle-enhanced immunonephelometric assay (N Latex Cystatin C; Siemens AG, Munich, Germany). In the CHS, serum cystatin C was measured using a BN II nephelometer (N Latex cystatin C; Dade Behring, Munich, Germany) (31).
UACR was available at baseline in the JHS and the MESA. In the JHS, UACR was obtained from either 24-hour urine collections or spot urine samples at examination 1 (32,33). A subset of participants (n=223) collected urine samples using both methods. UACR was calculated for both sample types and found to be highly correlated (r=0.97). Urine albumin was measured by nephelometric immunoassay (Dade Behring) and urine creatinine by the enzymatic Jaffe method. In the MESA, urinary albumin concentration was determined from spot urine samples by nephelometry using the Array 360 CE Protein Analyzer (Beckman Instruments Inc.). Urinary creatinine was measured from spot urine samples by the rate Jaffe method using the Vitros 950IRC instrument (Johnson & Johnson Clinical Diagnostics Inc.).
Outcomes
The primary outcome was incident AF (including atrial flutter) in all three cohorts. Incident AF was identified similarly in all three cohorts from any of three sources: electrocardiograms (ECGs) performed at study visits, International Classification of Diseases 9 (ICD-9) codes from hospitalization surveillance, and for participants enrolled in fee-for-service Medicare, ICD-9 codes from Medicare inpatient or outpatient claims (18,34–38). Participants were censored at death, loss to follow-up, or last available study follow-up (the JHS: December 31, 2012; the MESA: December 31, 2012; and the CHS: June 30, 2014).
Covariates
Demographic characteristics (age, sex, and race/ethnicity) and education level were determined by self-report. Diabetes was defined as fasting glucose ≥126 mg/dl or use of oral hypoglycemic medications or insulin. Information on tobacco use was collected from self-report (never, former, or current). Physical examination measures (height, weight, and BP) were obtained at the baseline study visit. Medication use was determined by medication inventory and recorded by study personnel.
History of heart failure, myocardial infarction, and stroke on study entry were determined by self-report (the CHS and the JHS) and review of prior medical records for confirmation (the CHS only). The MESA did not include participants with prior CVD. Clinically recognized heart failure and myocardial infarction that occurred during follow-up were captured by study visits and/or telephone interviews with participants and their kin. During these regular contacts, interim health events, including diagnostic tests, new diagnoses, hospitalizations, and death, were ascertained. Information on events in each of the three cohorts triggered review of ICD-9 Clinical Modification diagnosis codes, procedure codes, discharge summaries, and medical records.
Subclinical CV measures (left ventricular mass, left ventricular ejection fraction, and PR interval duration) obtained by echocardiogram, cardiac magnetic resonance imaging (MRI), or ECG were included. In the JHS, echocardiograms were performed at examination 1 by certified ultrasonography technicians. The calculation of left ventricular mass and left ventricular ejection fraction has been previously described (32). Both variables were reported as continuous variables. In the MESA, cardiac MRI was performed at baseline and quantified centrally at a single reading center (39). Cardiac MRI was performed with 1.5-T magnets with determination of left ventricular mass and left ventricular ejection fraction as previously described (39,40). Both were reported as continuous variables. In the CHS, echocardiograms were performed at baseline. Methods for quantification of left ventricular mass (continuous) and left ventricular ejection fraction (reported as normal, borderline, or abnormal) have been previously described (41,42). For all three cohorts, baseline ECGs were performed using standardized methods, and PR duration was quantified.
Statistical Methods
We compared participant characteristics across each of the study cohorts by the presence or absence at baseline of CKD defined as eGFR<60 ml/min per 1.73 m2 or microalbuminuria (UACR≥30 mg/g). Measures of baseline UACR were available only in the JHS and the MESA.
We then calculated the unadjusted incidence rates of AF in each cohort (per 1000 person-years) overall and by eGFR level (<60 and ≥60 ml/min per 1.73 m2) and UACR level (<30 versus ≥30 mg/g). Kaplan–Meier curves were generated to examine the cumulative AF-free survival in each cohort in participants stratified by eGFR and UACR categories. Log rank tests were performed to statistically compare the curves within each cohort. Penalized regression splines were generated to examine the functional form of the associations of eGFR and UACR with risk of incident AF in each cohort and confirmed linear associations for each (Supplemental Figure 2).
Participants were censored at death, loss to follow-up, or last available study follow-up. For analyses examining the adjusted association of eGFR with incident AF, we first modeled associations within each cohort using a Cox proportional hazards model. For these analyses, eGFR was modeled continuously (per 20 ml/min per 1.73 m2 lower eGFR) as well as in categories (≥90, 60–89, 45–59, 30–44, and <30 ml/min per 1.73 m2) per the Kidney Disease Improving Global Outcomes guidelines (43). We performed a series of nested models adjusting for possible confounders. Model 1 adjusted for age, age2 (to account for the large range of ages in the study population), sex, race/ethnicity, education (high school, some college, or college), height, weight, current smoking, and history of diabetes. Model 2 adjusted for variables in model 1 plus history of baseline of heart failure (the JHS and the CHS only), myocardial infarction (the JHS and the CHS only), and stroke (the JHS and the CHS only); systolic BP; diastolic BP; and use of antihypertensive medication use. The resulting cohort-specific estimates of association were combined using a fixed effects meta-analysis. For each model 2, we formally assessed the heterogeneity present in our meta-analysis using Cochran heterogeneity statistic and I2, which quantifies the degree of inconsistency in the individual studies’ results (44).
We used a similar approach to test the association of UACR with risk of incident AF in a meta-analysis of data from the JHS and the MESA. UACR was modeled continuously (per doubling calculated by taking the log2) as well as in clinically based categories: <15, 15–29, 30–299, and ≥300 mg/g.
We tested the proportional hazards assumption in unadjusted and adjusted (model 2) models. We found a violation of the assumption in unadjusted models testing the association of eGFR with incident AF in the JHS (P=0.05) and the CHS (P=0.007) but not in the MESA (P=0.25). There was no violation of the assumption in the models that tested the association of UACR with incident AF in the JHS and the MESA. In response, we examined Schoenfeld residual plots and accordingly, chose a cutoff of 5 years on the basis of when the risk of AF across time seemed to differ. We then performed a sensitivity analysis truncating follow-up time to 11 years and allowing for differing risks during years 0–5 and years 5–11 in the meta-analyzed study population.
In secondary analyses, we performed stratified analyses in prespecified subgroups: age (<65 and ≥65 years old), sex, race/ethnicity (white or black; Hispanic and Chinese MESA participants were excluded from this analysis), baseline prevalent CVD (yes/no; defined as prevalent heart failure, myocardial infarction, or stroke), baseline prevalent diabetes, the alternative measure of kidney function (eGFR or UACR), and study cohort. We tested for the multiplicative interactions and included the main effect of the covariate of interest in these models.
A primary goal of this analysis was to characterize the interaction between eGFR and UACR on the risk of incident AF. In addition to the stratified analyses examining the relative risk of incident AF across combinations of eGFR and UACR categories, we also examined differences in adjusted incidence rates in participants with (1) eGFR≥60 ml/min per 1.73 m2and UACR<30 mg/g, (2) eGFR≥60 ml/min per 1.73 m2 and UACR≥30 mg/g, (3) eGFR<60 ml/min per 1.73 m2 and UACR<30 mg/g, and (4) eGFR<60 ml/min per 1.73 m2 and UACR≥30 mg/g. We tested for the additive interaction in these models where we estimated the hazard ratio for categories 2–4 compared with category 1. If the excess relative risk for participants in group 4 is greater than those for participants in groups 2 and 3 added together, it suggests departure from additivity, indicating possible biologic interaction between UACR and eGFR.
We performed two sensitivity analyses. The first adjusted for cardiac imaging and electrocardiographic measures (left ventricular mass, left ventricular ejection fraction, and PR duration). The second adjusted for new diagnoses of heart failure and myocardial infarction that were made after baseline but before or at the same time as the AF diagnosis. This analysis only included the MESA and the CHS, because the JHS did not collect data on heart failure events until 5 years after the baseline visit.
Missing covariates were handled through multiple imputation. There was a large degree of missingness in subclinical CV measures included in model 3 (in the JHS, n=1818; in the MESA, n=1798; and in the CHS, n=2180). Most of the missingness in all three cohorts was seen with the left ventricular mass variables (particularly in blacks in the CHS due to the timing of the echocardiogram). A much smaller degree of missingness (<3%) was present for other covariates in each cohort. All subjects’ values were multiply imputed using chained equations (45), which were then combined using Rubin rules to account for the variability in the imputation procedure (46).
All analyses were conducted using R 3.3.0; a two-sided P value of <0.05 was considered statistically significant for all analyses.
Results
Study Population
Overall, there were 251 (5%) participants with eGFR<60 ml/min per 1.73 m2 in the JHS, 554 (8%) participants with eGFR<60 ml/min per 1.73 m2 in the MESA, and 1467 (29%) participants with eGFR<60 ml/min per 1.73 m2 in the CHS. Participants with eGFR<60 ml/min per 1.73 m2 were older, had higher BP, were more likely to be taking antihypertensive medications, and had a higher burden of comorbidity compared with participants without CKD across all three cohorts (Table 1). Similarly, in the JHS and the MESA, participants with higher UACR were older, had higher BP, were more likely to be taking renin-angiotensin-aldosterone system inhibitors, were more likely to have diabetes, and had lower eGFR compared with participants with lower UACR (Supplemental Table 1).
Characteristics of participants in the Jackson Heart Study, the Multi-Ethnic Study of Atherosclerosis, and the Cardiovascular Health Study by CKD status (defined as eGFR<60 ml/min per 1.73 m2)
Incidence Rates of AF in Each Cohort
In the JHS, there were 361 deaths and 238 incident AF events over a mean follow-up time of 8.5 (SD=2.7) years (overall incidence rate of AF was 5.5 per 1000 person-years). In the MESA, there were 503 deaths and 720 incident AF events over a mean follow-up time of 10 (SD=3.0) years (overall incidence rate of AF was 10.8 per 1000 person-years). In the CHS, there were 2829 deaths and 1569 incident AF events over a mean follow-up time of 12.5 (SD=7.1) years (overall incidence rate of AF was 25 per 1000 person-years). In all three cohorts, the unadjusted incidence rates for AF (per 1000 person-years) were significantly higher in participants with eGFR<60 ml/min per 1.72 m2 versus ≥60 ml/min per 1.73 m2: 15.9 versus 5.1 in the JHS, 28.4 versus 9.5 in the MESA, and 33.6 versus 22.5 in the CHS. Similarly, unadjusted incidence rates for AF were higher in those with UACR≥30 mg/g compared with <30 mg/g in the JHS (12.6 versus 4.3 per 1000 person-years) and the MESA (22.1 versus 9.8 per 1000 person-years). Kaplan–Meier curves for each cohort showed a significant difference in the unadjusted cumulative AF-free survival in participants with eGFR<60 versus ≥60 ml/min per 1.73 m2 as well as participants with UACR≥30 versus <30 mg/g (log rank test P<0.001 for all cohorts) (Figure 1).
The cumulative AF-free survival is lowest among participants low eGFR and high urine albumin-to-creatinine ratio. (A) Kaplan–Meier curves of incident atrial fibrillation in participants by eGFR categories (<60 and ≥60 ml/min per 1.73 m2). (B) Kaplan–Meier curves of incident atrial fibrillation in participants by urine albumin-to-creatinine ratio (UACR) categories (<30 and ≥30 mg/g). CHS, Cardiovascular Health Study; JHS, Jackson Heart Study; MESA, Multi-Ethnic Study of Atherosclerosis.
Association of eGFR with Incident AF
In the JHS, participants with eGFR<45 ml/min per 1.73 m2 had greater risk of incident AF, although there were few participants in the lowest eGFR category (Supplemental Table 2). In the MESA and the CHS, there was a graded association of lower eGFR categories with risk of incident AF (Supplemental Tables 3 and 4). We observed low heterogeneity across the three study cohorts (I2=0% for model 2). In the meta-analyzed study population, there was a stepwise increase in risk of incident AF across a broad range of eGFR categories (Table 2). In adjusted models, compared with eGFR≥90 ml/min per 1.73 m2, the risk of incident of AF was significantly greater among participants with eGFR<60 ml/min per 1.73 m2, with the greatest risk among participants with eGFR<30 ml/min per 1.73 m2 (hazard ratio, 2.03; 95% confidence interval, 1.40 to 2.96) (Table 2).
Association of eGFR and urine albumin-to-creatinine ratio with risk of incident atrial fibrillation in 16,769 participants from community-based cohorts
As a sensitivity analysis to address the violation in the test for proportional hazards, we stratified follow-up time as 0–5 and 5–11 years in each cohort. In these analyses, the adjusted associations of categorical and continuous eGFR with incident AF were stronger when the follow-up time was 0–5 versus 5–11 years (Supplemental Table 5).
Association of UACR with Incident AF
In the JHS, the associations of UACR with incident AF were similar across UACR categories (Supplemental Table 2). In the MESA, there was a stepwise association of higher UACR with risk of incident AF (Supplemental Table 3). We observed low heterogeneity across the two cohorts (I2=9.9% for model 2). In the meta-analyzed study population of the JHS and the MESA, there was a stepwise association between higher UACR and risk of incident AF, with the greatest risk among participants with UACR≥300 mg/g (hazard ratio, 1.76; 95% confidence interval, 1.18 to 2.62) compared with those with UACR<15 mg/g (Table 2).
As conducted above for eGFR, a sensitivity analysis to address the violation in the test for proportional hazards was conducted for UACR. In these analyses, the adjusted associations of categorical and continuous UACR with incident AF were stronger when the follow-up time was 0–5 versus 5–11 years (Supplemental Table 5).
Stratified Analyses across Participant Subgroups
We tested the adjusted association of eGFR and UACR with risk of incident AF in analyses stratified by age, sex, race, prevalent CVD, prevalent diabetes, alternative kidney measure (eGFR or UACR), and study cohort. The associations of eGFR<60 ml/min per 1.73 m2 (versus ≥60 ml/min per 1.73 m2) with incident AF were relatively consistent across all subgroups (although some subgroup analyses had wide confidence intervals) (Figure 2A). Tests for multiplicative interaction revealed a statistically significant interaction of eGFR with sex (P=0.01), with greater risk among men. The multiplicative interaction of eGFR with UACR was not statistically significant.
Multivariable association of eGFR or urine albumin-to-creatinine ratio with risk of incident atrial fibrillation is consistent across most participant subgroups (yrs and participants with prevalent CVD, respectively). (A) Multivariable association of eGFR<60 ml/min per 1.73 m2 (versus ≥60 ml/min per 1.73 m2) with risk of incident atrial fibrillation in three community-based cohorts across participant subgroups. (B) Multivariable association of urine albumin-to-creatinine ratio (ACR) ≥30 mg/g (versus <30 mg/g) with risk of incident atrial fibrillation in two community-based cohorts across participant subgroups. 95% CI, 95% confidence interval; CHS, Cardiovascular Health Study; CVD, cardiovascular disease; HR, hazard ratio; JHS, Jackson Heart Study; MESA, Multi-Ethnic Study of Atherosclerosis.
In stratified analyses, the associations between UACR≥30 mg/g (versus <30 mg/g) and incident AF were also consistent across all subgroups (Figure 2B). However, the association did not reach statistical significance in participants with prevalent CVD, perhaps due to small sample sizes (Figure 2B). The P values for multiplicative interaction were not statistically significant across these subgroups.
Incidence Rates of AF across eGFR and UACR Categories
In addition to examining the relative risk of incident AF across eGFR and UACR categories, we also studied the adjusted incidence rates of AF across four categories of eGFR and UACR (Figure 3). The adjusted incidence rate of AF in participants with eGFR≥60 ml/min per 1.73 m2 and UACR<30 mg/g was 4.13 (95% confidence interval, 3.64 to 4.63) per 1000 person-years. Participants with either eGFR<60 ml/min per 1.73 m2 with UACR<30 mg/g (5.01; 95% confidence interval, 3.78 to 6.25 per 1000 person-years) or UACR≥30 mg/g with eGFR≥60 ml/min per 1.73 m2 (5.54; 95% confidence interval, 4.27 to 6.81 per 1000 person-years) had comparable adjusted incidence rates for AF. Finally, the adjusted incidence rate for those with eGFR<60 ml/min per 1.73 m2 and UACR≥30 mg/g was 7.21 (95% confidence interval, 4.84 to 9.58) per 1000 person-years (Figure 3). The P value for the additive interaction was not statistically significant (P=0.58).
Adjusted incidence rates of atrial fibrillation are high in participants with either low eGFR or high UACR.
Sensitivity Analyses: Adjustment for Cardiac Imaging/ECG Measures and Interim/Concurrent Heart Failure and Myocardial Infarction Events
In the meta-analyzed population of all three cohorts, when models were further adjusted for subclinical CV measures (left ventricular mass, left ventricular ejection fraction, and PR duration), associations were mildly attenuated, but overall findings were similar to the main analysis (Supplemental Table 6).
To understand the contribution of interim CV events to risk of subsequent AF among those with low eGFR or high UACR, we adjusted for interim/concurrent heart failure and myocardial infarction events in the meta-analyzed study population, which consisted of the CHS and the MESA. In total, there were 84 interim/concurrent heart failure and myocardial events in the MESA and 570 interim/concurrent heart failure and myocardial events in the CHS. With this adjustment, the association of low eGFR and high UACR remained materially unchanged (Supplemental Table 7).
Discussion
In this meta-analysis of data from nearly 17,000 diverse participants from three community-based cohorts, we found a strong stepwise association with incident AF for both reduced eGFR and elevated UACR. The associations were robust, even with adjustment for subclinical CV measures and interim CV events. Our study extends findings from prior studies by examining a large, multicenter, diverse patient population; using finer categories of eGFR (43); examining the association with incident AF across a more comprehensive range of eGFR (including participants without clinical CKD with eGFR>60 ml/min per 1.73 m2); and using the more recent combined cystatin C-creatinine eGFR equation to more accurately characterize levels of kidney function (22). Our findings provide further conclusive evidence that abnormal markers of kidney function are significant risk factors for the development of AF.
In our study, eGFR<60 ml/min per 1.73 m2 was significantly associated with greater risk of incident AF, and there was a stepwise increase in risk of AF with lower eGFR and progressive worsening of CKD categories. The risk of incident AF was particularly large in the CHS, a cohort of older participants with higher burden of comorbidity. These findings are consistent with prior literature (13,16,17,47). In the ARIC Study, Alonso et al. (18) reported a strong graded association between cystatin C–based eGFR and incident AF, with risk estimates similar to our risk estimates. The associations in the ARIC Study analysis were less robust when using creatinine-based eGFR. Notably, some JHS participants are also ARIC Study participants; thus, n=1626 individuals were included in both the ARIC Study analysis and this analysis. Our analyses extended this prior work by using the more accurate combined creatinine-cystatin C equation to define eGFR, examining a more racially/ethnically diverse population, and adjusting for a wide range of covariates, including cardiac imaging measures. In the CHS, investigators reported significant associations of elevations in cystatin C with risk of incident AF in unadjusted models, which were attenuated with multivariable adjustment (35). This analysis differs from the previous CHS analysis, in that we used eGFR from both creatinine and cystatin C, started the follow-up period from baseline in the CHS, and had both a longer follow-up time and a greater event rate.
UACR is an important potentially modifiable measure of kidney function and has been linked with poor CV outcomes and death in populations with and without CKD (48–50). We found that UACR>30 mg/g was significantly associated with greater risk of incident AF. A few prior studies have examined the association of UACR and risk of incident AF. In the ARIC Study, participants with UACR>30 mg/g had a twofold higher risk of incident AF (18). In studies of older adults (6) and patients with type 2 diabetes (7), albuminuria has been linked with incident AF (12,51). Our study extends and confirms these prior analyses by studying a more diverse and larger study population.
Notably, in our study, we did not find a multiplicative interaction of eGFR with UACR. This is consistent with the ARIC Study, which did not find a significant interaction between eGFR and UACR (18). We also did not find a significant additive interaction between UACR and eGFR with risk of incident AF, which suggests that the association of high UACR and low eGFR with incident AF in combination is not greater than that of either marker alone. In this study, the observed associations were robust, even with adjustment for subclinical CVD, suggesting that the association of abnormal measures of kidney function with incident AF is, in part, independent of cardiac structural and conduction abnormalities. Studies have reported that abnormal cardiac structure and function as well as conduction delays are associated with increased risk of AF (52–57). The analysis in the ARIC Study also adjusted for ECG measures and found an independent association of eGFR and UACR with incident AF. However, we were able to adjust for left ventricular mass, an important risk factor and confounder, as ascertained from high-quality research echocardiograms and cardiac MRIs. It is interesting that, although somewhat attenuated, the findings did not materially differ from the main analyses, suggesting that presence of subclinical CVD in patients with CKD does not entirely account for the observed associations.
With adjustment for interim heart failure and myocardial infarction, the associations between lower eGFR and higher UACR remained strong. Clinical CVD is a known risk factor for development of AF (58), and patients with CKD have greater risk of heart failure and myocardial infarction compared with those without CKD (59).
Several possible mechanisms may explain the link between kidney dysfunction and risk of incident AF beyond traditional CV risk factors. Myocardial fibrosis is a well established feature of AF and also linked to impaired kidney function (60–62). Cardiac inflammation is a key driver of AF. Animal models and human studies have shown a strong link between inflammation and left atrial fibrosis in CKD (63,64). Patients with kidney dysfunction are more likely to have abnormalities of electrolytes, which may contribute to the risk of AF (65). Disorders of mineral metabolism, manifest by altered levels of calcium, phosphorus, vitamin D, parathyroid hormone, and fibroblast growth factor-23, are common in CKD and linked to AF. Animal models have shown that the sodium-calcium exchanger and calcium handling in the cardiomyocyte is altered in CKD and may lead to electrophysiologic dysfunction and AF (66). Clinical studies have identified elevated phosphorus (67) and low 25(OH)D (68) as risk factors for AF (67). In the MESA, baseline measures of and fibroblast growth factor-23 were associated with incident AF (69). Subclinical volume overload may also lead to cardiac stretch and thus, be a risk factor for AF in patients with CKD. Further study of novel therapies to treat CKD-specific AF risk factors may help prevent AF and its associated complications.
The findings of this study have important clinical implications. Our data suggest that patients with kidney disease, as defined by either reduced eGFR or elevated UACR, are a high risk for AF and may identify a population that would benefit from targeted AF prevention therapies. Specifically, renin-angiotensin-aldosterone system inhibitors have been shown in some settings to reduce risk of incident AF (70–72). Novel therapies, such as antioxidants, are also promising (73). Studies are needed to test the efficacy of AF prevention therapies in patients with kidney disease.
Our study had several strengths. We studied three diverse community-based patient populations. GFR was estimated using both serum creatinine and cystatin C. AF ascertainment was standardized in each of the three cohorts. We were able to account for important confounders, including antihypertensive medication use and subclinical measures of CVD. We recognize a few limitations as well. Measures of serum creatinine and cystatin C were not performed using the same assay across all three study cohorts, which may have introduced error in our quantification of eGFR. Ascertainment of AF relied, in part, on administrative codes, which are imperfect. However, prior work has shown reasonable validity with this approach, with a reported positive predictive value of 77% (74). Increased left atrial diameter and inferior vena cava diameter, risk factors for AF, were not systematically quantified in all three cohorts and therefore, could not be included as mediating factors. Measures of left ventricular mass and left ventricular ejection fraction were ascertained by echocardiograms in the CHS and the JHS but cardiac MRI in the MESA, which may have led to some differences in the quantification of these parameters. The study design was observational, and therefore, we cannot establish a causal relationship. Finally, although we adjusted for a large number of possible confounders, there is still the potential for residual confounding.
In conclusion, in this large meta-analysis of data from three diverse community-based cohorts with a broad range of kidney function, we found that eGFR<60 ml/min per 1.73 m2 and UACR ≥30 mg/g were significantly associated with greater risk of incident AF. These associations were consistent across patient subgroups and after accounting for numerous possible confounders, including presence of subclinical CVD and clinical CV events. These data highlight the need for further studies to examine the mechanistic link between kidney disease and AF.
Disclosures
None.
Acknowledgments
Because I.H.d.B. is a Deputy Editor of the Clinical Journal of the American Society of Nephrology, he was not involved in the peer review process for this manuscript. Another editor oversaw the peer review and decision-making process for this manuscript. Rajnish Mehrotra, the Editor-in-Chief is at the same institution as some of the authors, including the Deputy Editor, and was therefore also not involved in the peer-review process for this manuscript.
The authors thank the other investigators, the staff, and the participants of the Jackson Heart Study (JHS), the Multi-Ethnic Study of Atherosclerosis (MESA), and the Cardiovascular Health Study (CHS) for their valuable contributions.
This study was supported by the following funding sources: National Institute of Diabetes and Digestive and Kidney Diseases grants K23DK088865 (to N.B.), R01DK103612 (to N.B.), and R01DK102134 (to B.Y.) and National Heart, Lung, and Blood Institute (NHLBI) grants R21HL121348 (to S.R.H.) and R01HL127659 (to S.R.H.). Additional support was provided by American Heart Association grant 16EIA26410001 (to A.A.). B.Y. is also supported, in part, by funding from the Veterans Affairs Puget Sound Health Care System. This research was also supported by an unrestricted gift from the Northwest Kidney Centers to the Kidney Research Institute. The Jackson Heart Study is supported by contracts HHSN268201300046C, HHSN268201300047C, HHSN268201300048C, HHSN268201300049C, and HHSN268201300050C from the NHLBI and the National Institute on Minority Health and Health Disparities. The Cardiovascular Health Study (CHS) is supported by contracts HHSN268201200036C, HHSN268200800007C, N01HC55222, N01HC85079, N01HC85080, N01HC85081, N01HC85082, N01HC85083, and N01HC85086 and grants U01HL080295 and U01HL130114 from the NHLBI, with additional contribution from the National Institute of Neurological Disorders and Stroke. Additional support was provided by grant R01AG023629 from the National Institute on Aging. The MESA is supported by contracts N01-HC-95159, N01-HC-95160, N01-HC-95161, N01-HC-95162, N01-HC-95163, N01-HC-95164, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168, and N01-HC-95169 from the NHLBI and grants UL1-TR-000040 and UL1-RR-025005 from the National Center for Research Resources.
A full list of principal CHS investigators and institutions can be found at CHS-NHLBI.org. A full list of participating MESA investigators and institutions can be found at http://www.mesa-nhlbi.org.
Footnotes
Published online ahead of print. Publication date available at www.cjasn.org.
This article contains supplemental material online at http://cjasn.asnjournals.org/lookup/suppl/doi:10.2215/CJN.01860217/-/DCSupplemental.
- Received February 16, 2017.
- Accepted May 15, 2017.
- Copyright © 2017 by the American Society of Nephrology
References
In this issue
