Serum androgens and risk of atrial fibrillation in older men: The Cardiovascular Health Study

Abstract

Background

Decline in serum androgens is common among older men and has been associated with cardiovascular disease, although its role in risk of atrial fibrillation (AF) has not been well defined.

Hypothesis

Low serum androgens are associated with an increased risk of AF.

Methods

We examined the prospective associations between testosterone, its more active metabolite dihydrotestosterone (DHT), and sex hormone–binding globulin (SHBG) with risk of AF among 1019 otherwise healthy men with average age 76.3 ±4.9 years in the Cardiovascular Health Study.

Results

After median follow‐up of 9.5 years, 304 (30%) men developed AF. We detected a nonlinear association with risk of incident AF in both free and total DHT, in which subjects with the lowest levels had a higher risk of incident AF. After adjustment for demographics, clinical risk factors, left atrial diameter, and serum NT‐proBNP levels, men with free DHT <0.16 ng/dL were at increased risk compared with men with higher levels (hazard ratio: 1.48, 95% confidence interval: 1.01–2.17, P <0.05). Sensitivity analyses confirmed that the increased risk was not cutpoint‐specific, with a significant association noted up to cutpoints <~0.2 ng/dL. We also detected a complex nonlinear association between SHBG and incident AF, in which subjects in the middle quintile (52.9–65.3 nmol/L) had increased risk.

Conclusions

Among older men, low free DHT is associated with an increased risk of incident AF. Further studies are needed to explore mechanisms for this association.

1. INTRODUCTION

Increased age is among the strongest risk factors for atrial fibrillation (AF), with estimates that of the expected 5.6 million patients with AF by the year 2050, >50% will be age > 80 years.1 A number of studies have linked low testosterone with increased mortality in men,2, 3, 4 at least in part due to an association with increased cardiovascular disease (CVD). One observational cohort study very recently noted an association between AF and lower total testosterone levels in certain age groups.5 However, this study did not examine other androgens, such as the more potent endogenous androgen dihydrotestosterone (DHT), and did not include adjustments for sex hormone–binding globulin levels (SHBG), which also increase with age and therefore affect total androgen levels without necessarily reflecting changes in free androgen. Another recent study identified an increased risk of AF associated with free testosterone6 but did not examine DHT. Clearly, more studies are needed to examine the association of risk of AF with levels of circulating androgens.

In this study, we examined the association of total and free testosterone and DHT, and SHBG concentrations, with incident AF in a large population of older men, after adjustment for potential confounders.

2. METHODS

2.1. Study population

The design and objectives of the Cardiovascular Health Study (CHS) have been previously described.7 In brief, CHS is a longitudinal study of men and women age ≥65 years recruited from a random sample of Medicare‐eligible residents of Pittsburgh, PA; Forsyth County, NC; Sacramento, CA; and Hagerstown, MD. The original cohort of 5201 participants was enrolled in 1989–1990; a second cohort of 687 African Americans was recruited in 1992–1993. Our study sample consisted of men from both cohorts who participated in the CHS examination in 1994 (when serum androgen samples were obtained), who had no history of CVD (defined as myocardial infarction, coronary artery bypass surgery, percutaneous coronary intervention, heart failure, or stroke), and who had sera available for measurement of circulating androgens (n = 1128). The institutional review board at each center approved the study, and each participant gave informed consent. Clinic examinations were performed annually from 1989 to 1999, and again in 2005. Determination of prevalent CVD was performed using methods previously validated by Psaty et al.8 Participants were contacted every 6 months for follow‐up, alternating between a telephone interview and a clinic visit through 1999 and by telephone interview only after that until 2005. Fasting laboratory measurements included glucose, serum cystatin C9, and N‐terminal pro‐B‐type natriuretic peptide (NT‐proBNP).10 Both NT‐proBNP and cystatin C levels were obtained from 2 years prior to the 1994 exam in which the androgens were measured (see below).

For the current analysis, we excluded men with prevalent AF (n = 63), as well as those taking medications that would suggest a diagnosis of AF or cardiac arrhythmia, including 34 taking digoxin, 10 taking warfarin, 1 taking class I antiarrhythmic medication, and 1 taking class III antiarrhythmic medication.

2.2. AF diagnosis

An annual resting electrocardiogram (ECG) was obtained yearly through the ninth year of follow‐up, and discharge diagnoses for all hospitalizations were collected. We identified cases of AF in 2 ways: (1) annual study ECGs interpreted by the Epidemiological Cardiology Research Center (EPICARE) ECG reading center, where the diagnoses of AF or atrial flutter were verified11; and (2) hospital discharge diagnoses that included codes for AF and flutter, although AF/flutter diagnoses that were made during the same hospitalization as coronary artery bypass surgery or heart valve surgery were not counted. Prior evaluation in CHS determined the positive predictive value of hospital discharge diagnosis to be 98.6% for diagnosis of AF,11 and a Holter substudy identified that only 1 in 819 subjects (0.1%) had persistent or intermittent AF not identified by the above measures.12

2.3. Androgen measurements

Description of methods and validation for measurement of circulating androgen levels are described in detail elsewhere.13 Briefly, samples were obtained from men participating in the 1994–1995 CHS examination. Total testosterone and DHT were measured simultaneously using a liquid chromatography–tandem mass spectrometry assay.14 SHBG was assayed using a time‐resolved fluoro‐immunoassay (Delfia; Perkin Elmer, Norton, OH). Free testosterone and free DHT were calculated using the Mazer method.15

2.4. Statistical analysis

We used Cox proportional hazards regression modeling with stratification by year of age and adjustment for demographic variables, clinical risk factors, and medications. Stepwise Cox regression was used to select medications and clinical risk factors for inclusion in the full models, with P value for inclusion of 0.10. In a subanalysis, 810 individuals were analyzed for whom the biomarker NT‐proBNP and left atrial (LA) diameter on echocardiography were available. Free and total testosterone and DHT levels were analyzed using linear models and nonlinear approaches. To assess for nonlinear associations, free and total testosterone, free and total DHT, and SHBG were examined using quintiles, restricted cubic splines (4 knots), and multivariable restricted spline models (“mvrs” function in Stata), with close attention to the nature of any nonlinear association detected. The mvrs function is an automated stepwise regression program that selects the best‐fitting nonlinear restricted cubic splines model using a closed‐test procedure.16 The proportional hazards assumption was checked with Schoenfeld residuals.

As a follow‐up to nonlinear associations identified for free and total DHT, we selected cutpoints based on the lowest quintile of values. Sensitivity analysis for these points was performed using a loop through various possible cutpoints in increments of 0.005 ng/dL from 0.02 ng/dL up to 0.50 ng/dL for free DHT and from 0.02 ng/mL to 1.00 ng/mL for total DHT using multivariate Cox regression. For further validation, we then analyzed the association of low free DHT (based on the cutpoint of 0.16 ng/dL) with incident AF using Mahalanobis distance matching17 (“mahascores” package in Stata) for individuals with low free DHT compared with matched “normal” values across clinical risk factors, with restricted matching by age (perfect matching required). Matching was assessed by examining distance measures of matched pairs compared with overall average distance measures, as well as matching across covariates. Statistical analyses were performed using Stata IC, version 14.2 for Mac (StataCorp LP, College Station, TX).

3. RESULTS

3.1. Androgens and AF–linear association

Characteristics of the population are shown in Table 1. Among 1019 individuals followed for a median of 9.5 years (interquartile range [IQR], 0.02–20.1 years), there were 304 cases of incident AF. In separate models stratified by age and adjusted for demographics (race, education, income, clinic location, and smoking status) and for demographics plus clinical risk factors (diabetes mellitus [DM], body mass index [BMI], loop diuretics, height, hypertension [HTN], depressed left ventricular ejection fraction, kidney function, and systolic blood pressure), we did not identify any linear association between any of the circulating androgen levels and incident AF at P < 0.05 (see Supporting Information, Table 1, in the online version of this article).

Table 1.

Characteristics of study population (N = 1019 males)

Demographics	Value
Mean age, y	76.3 ± 4.9
Caucasian race	857 (84.1)
Income ≥$25 000/ya	610 (59.9)
Greater than high school education	517 (50.7)
Medical history
DM	140 (14.4)
CHF	1 (0.1)
Current smokinga	114 (11.3)
Treated HTN	457 (44.9)
Medications
β‐Blockers	102 (10.0)
CCBs	166 (16.3)
ACEIs	96 (9.4)
Loop diuretics	31 (3.1)
Anthropometrics
BMI, kg/m2	26.7 ±3.7
Mean standing height, cm	172.7 ±6.4
Mean weight, lbs.	175.9 ±27.1
Biomedical measurements
SBP, mm Hg	132.2 ± 19.5
DBP, mm Hg	71.0 ± 10.7
Cystatin C, mg/Lb	1.1 ± 0.3
NT‐proBNP, pg/dLb	173.0 ± 426.5
Depressed LV functionc	97 (10.1)
LA diameter, cm	4.1 ± 0.6
Testosterone values
Free testosterone, ng/dL (Mazer)d	5.2 ± 2.2
Free DHT, ng/dL (Mazer)d	0.26 ± 0.12
Total testosterone, ng/dL	382.0 ± 178.4
Total DHT, ng/mL	0.44 ± 0.23
SHBG, nmol/L	64.3 ± 29.3

Abbreviations: ACEI, angiotensin‐converting enzyme inhibitor; BMI, body mass index; CCB, calcium channel blocker; CHF, congestive heart failure; DBP, diastolic blood pressure; DHT, dihydrotestosterone; DM, diabetes mellitus; HTN, hypertension; LA, left atrial; LV, left ventricular; LVEF, left ventricular ejection fraction; MI, myocardial infarction; NT‐proBNP, N‐terminal pro‐B‐type natriuretic peptide; SBP, systolic blood pressure; SHBG, sex hormone–binding globulin. Data are presented as n (%) or mean ± SD.

^aIncome and smoking status are from baseline.

^bCystatin C and NT‐proBNP levels were available from 2 years prior to baseline.

^cDepressed LV function is defined as LVEF <55% by echocardiography.

^dMazer formula for calculation of free sex hormone level.

No patients had a history of MI or stroke.

3.2. Androgens and AF–nonlinear association

Using age‐stratified univariable Cox regression with quintiles, restricted cubic splines, as well as stepwise multivariable spline Cox regression (see Supporting Information, Figures 1–3, respectively, in the online version of this article), we identified a consistent nonlinear trend in which individuals with the lowest levels of free and total DHT appeared to be at increased risk of incident AF. Both free and total DHT had the same nonlinear pattern; for simplicity, we focused on free DHT. We selected the lowest quintile as a cutpoint (<0.16 ng/dL) for “low” free DHT and noted a significant association with incident AF after stratification by age and adjustment for SHBG, demographics, and clinical risk factors, both overall and in the subgroup with NT‐proBNP and LA diameter measures (Table 2). We performed sensitivity analyses of the selected cutpoints by looping through other possible values of free DHT above and below these values (Figure 1), which verified that values of free DHT of up to 0.2 ng/dL could also have been used and had a significantly increased risk of incident AF at P < 0.05. We then performed Mahalanobis distance matching based on demographic and clinical risk factors, with a requirement of perfect age matching, to identify a cohort of subjects in whom all variables except free DHT levels were reasonably matched (see Supporting Information, Table 2, in the online version of this article). In this pruned cohort of 330 subjects, the median distance was 7.5 (IQR, 4.7–11.8), which compared favorably with the entire cohort, where the median distance was 19.3 (IQR, 13.7–26.3). After matching, we identified a significantly increased risk of incident AF in the group with low free DHT (Figure 2).

Table 2.

Association of low free DHT (<0.16 ng/dL) and incident AF

Model	HR	95% CI	P Value
Model 1: Age‐stratified + SHBG + demographicsa (N = 1002)	1.44	1.07–1.94	0.016
Model 2: Model 1 + clinical risk factorsb (N = 938)	1.43	1.04–1.96	0.026
Model 3: Model 2 + biomarkersc (N = 696)	1.48	1.01–2.17	0.044

Abbreviations: AF, atrial fibrillation; BMI, body mass index; CI, confidence interval; DHT, dihydrotestosterone; DM, diabetes mellitus; HR, hazard ratio; LA, left atrial; LV, left ventricular; NT‐proBNP, N‐terminal pro‐B‐type natriuretic peptide; SBP, systolic blood pressure; SHBG, sex hormone–binding globulin.

^aDemographics included clinic site, level of education, income, smoking status, and race.

^bClinical risk factors included BMI, standing height, DM, use of antihypertensive medications, SBP (mm Hg), depressed LV function, serum cystatin C level, and use of loop diuretics.

^cBiomarkers included serum NT‐proBNP level and LA diameter.

Figure 1.

Sensitivity analysis of cutpoint for free DHT. Predicted HR (left‐sided y axis, navy line) for various cutpoints of “low free DHT” from Cox regression model with adjustment for demographics (clinic site, level of education, income, smoking status, and race) and clinical risk factors (BMI, standing height, DM, use of antihypertensive medications, SBP [mm Hg], depressed LV function, serum cystatin C level, and use of loop diuretics). 95% CI provided by dashed navy lines. Right‐sided axis (green line) with proportion of subjects assigned as “low free DHT” based on each given cutpoint. Vertical dashed red line indicates cutpoint of 0.16 ng/mL, used for this analysis. Abbreviations: AF, atrial fibrillation; BMI, body mass index; CI, confidence interval; DHT, dihydrotestosterone; DM, diabetes mellitus; HR, hazard ratio; LV, left ventricular; SBP, systolic blood pressure

Figure 2.

Kaplan–Meier survival estimates for pruned cohort based on Mahalanobis distance matching (see article text for details). Matching performed for exact age match, and distance match by demographic and clinical risk factors. One observation from the low–free DHT group ended before entering follow‐up period (see Supporting Information, Table 2, in the online version of this article). HR for low free DHT vs normal free DHT was 1.59 (95% CI: 1.06–2.4, P = 0.02). Abbreviations: AF, atrial fibrillation; CI, confidence interval; DHT, dihydrotestosterone; HR, hazard ratio

Neither free nor total testosterone displayed evidence of any significant pattern of nonlinear association with incident AF (see Supporting Information, Figures 1–3, in the online version of this article). SHBG demonstrated a complex nonlinear association with risk of incident AF in which increased levels were generally associated with decreased risk, although there was significant nonlinearity to the association, with an increased risk noted among subjects with average to high levels (corresponding to quintiles 3 and 4; see Supporting Information, Figure 1E, in the online version of this article).

3.3. Other associations

In exploratory analyses, we identified no significant interaction between low free DHT and any other covariables (including BMI, DM, smoking, LA size, kidney function, HTN, or NT‐proBNP) on AF risk. Compared with individuals with normal levels of free DHT, those with low free DHT were on average older and non‐Caucasian, with greater average weight, BMI, LA size, and NT‐proBNP levels, and a larger proportion had increased prevalent DM and treated HTN (Table 3). These individuals also had lower levels of free and total testosterone, but increased levels of SHBG on average.

Table 3.

Associations with low free DHT

	Low Free DHTa	Normal Free DHT	P Valueb
No.	199	820
Mean age, y	78.1 ±5.9	75.9 ±4.5	<0.0001
Caucasian	158 (79)	699 (85)	0.04
Grade > 12	97 (49)	420 (51)	0.53
Income <$25 000/y	119 (60)	491 (60)	0.98
Current smoker	26 (14)	88 (11)	0.11
Standing height, cm	173.3 ± 6.6	172.6 ±6.4	0.22
Weight, lbs.	182.9 ± 33.5	174.2 ± 25.1	<00001
BMI, kg/m2	27.6 ± 4.4	26.5 ±3.5	0.0002
DM	41 (23)	98 (12)	0.002c
Impaired fasting glucose	22 (12)	98 (12)	—
Antihypertensive medication	103 (52)	354 (43)	0.03
SBP, mm Hg	132.7 ± 21.0	132.0 ± 19.2	0.69
DBP, mm Hg	69.9 ±10.4	71.3 ±10.8	0.09
LA size	4.19 ± 0.66)	4.06 ± 0.60	0.01
Abnormal LV function	19 (10)	78 (10)	0.92
Cystatin C	1.14 ± 0.28	1.07 ±0.24	0.002
NT‐proBNP	268 ± 717	150 ± 317	0.002
Free testosterone	3.13 ± 2.18	5.71 ± 1.82	<0.0001
Total testosterone	257 ± 186	412 ±163	<0.0001
SBHG	72.0 ± 37.1	62.4 ±26.8	<0.0001

Abbreviations: ANOVA, analysis of variance; DBP, diastolic blood pressure; DHT, dihydrotestosterone; DM, diabetes mellitus; LA, left atrial; NT‐proBNP, N‐terminal pro‐B‐type natriuretic peptide; SBP, systolic blood pressure; SD, standard deviation; SHBG, sex hormone–binding globulin. Data are presented as n (%) or mean ± SD.

^aLow free DHT defined as <0.16 ng/dL.

^b P value calculated from 1‐way ANOVA or χ² test between groups.

^cDM and impaired fasting glucose calculated from 1 test; P value refers to overall test.

4. DISCUSSION

In this prospective cohort of older men, we identified a significant nonlinear association between free DHT and risk of incident AF, in which only subjects with the lowest levels of free DHT had an increased risk after adjustment for age, demographics, and other clinical risk factors. We failed to detect any association, linear or nonlinear, of serum testosterone with incident AF, and we also identified a complex nonlinear association of SHBG with incident AF. Our studies indicated that a level of circulating free DHT <0.16 ng/dL was a reasonable cutpoint for designation as a “low” level, although sensitivity analysis suggested that a cutpoint below or above this level up to about 0.2 ng/dL would have been satisfactory. Although we were unable to detect any significant interaction of low free DHT with other risk factors on AF risk, we did note that individuals with low levels tended to have increased age, prevalence of DM and obesity (increased weight and BMI), as well as increased LA diameter and serum NT‐proBNP levels, both of which have been independently associated with risk of AF.10, 18

The association between low androgen hormone levels and risk of CVD has been a somewhat controversial topic recently, particularly with regard to testosterone‐replacement therapy, where there have been conflicting results.19, 20, 21, 22, 23, 24, 25 An association between low levels of circulating testosterone and incident AF has been observed in certain large cohorts,5 as well as in men with lone AF (476 ng/dL vs 514 ng/dL),26 although other studies have suggested no association.27 In animal models, decreased androgen level has been noted to cause changes in calcium handling that are proarrhythmic28 and could increase risk of AF. More specifically, recent studies in mouse models have found that castration reduced DHT levels, increased late sodium current in atrial myocytes, and increased electrically induced episodes of AF; importantly, these changes were reversed with DHT treatment.29 In humans, a recent study found that testosterone‐replacement therapy to therapeutic circulating levels (that would be expected to increase DHT levels as well) was associated with a decreased risk of AF,30 although more studies would be needed to frame this finding within the range of other cardiovascular effects that have been attributed to testosterone replacement therapy.

DHT, also known as 5α‐dihydrotestosterone, is synthesized through the action of the enzyme 5α‐reductase, which catalyzes its formation from testosterone. 5α‐reductase is located in the prostate gland, seminal vesicles, epididymis, skin, hair follicles, liver, and brain. DHT has a higher affinity for the androgen receptor than does testosterone, and importantly it cannot be converted by aromatase into estrogen, which is in contrast to testosterone. As such, measurement of free DHT is more likely to reflect primarily activation of the androgen receptor pathways than testosterone, which is also capable of being converted to estradiol and acting via estrogen receptors. A number of prior studies have identified low DHT as a better marker for risk of CVD, and cardiovascular risk factors, than testosterone alone. Joyce and colleagues recently demonstrated that low DHT, but not testosterone or SHBG, was associated with the risk of incident DM.31 In this same population as this study, Shores et al. also noted that low levels of free DHT, but not testosterone, were associated with risk of ischemic stroke,13 raising the possibility of AF as a mechanism for stroke in some of these individuals. These studies support our finding that DHT may play a larger role in CVD than testosterone, which may be more of a marker of DHT levels in other studies, such as those that have included testosterone and other androgens but excluded DHT.

In a large prospective study of circulating androgens and AF, Magnani and colleagues examined testosterone, estradiol, and dehydroepiandrostendione (DHEA) in the Framingham cohort and noted an association between testosterone and incident AF, in which the greatest risk was in men age 55 to 69 years and > 80 years.5 That analysis did not examine free or total DHT, nor free testosterone, which is important given that levels of SHBG also increase with age, an effect that we noted in our study as well and may have partially accounted for the nonlinear effects described in that study. A recent analysis of the Multi‐Ethnic Study of Atherosclerosis (MESA) cohort actually identified a positive association between free testosterone levels6 in men and risk of incident AF, although they noted the opposite association in women. As 64% of the MESA cohort was nonwhite (13% Chinese American, 27% Black, 24% Hispanic), this finding raises the interesting question of whether the association is primarily limited to Caucasian individuals; although there were other key differences between this cohort and ours, particularly age (average age of males, 62 ±10 years in MESA vs 76.5 ±5.0 years in CHS), which is a strong risk factor for AF (although the MESA study did not detect a significant interaction with age and hormone level [P = 0.30]).

The connection between circulating androgen levels and DM with CVD is increasingly being acknowledged,32 with several studies, including one previously conducted in the CHS,31 identifying an association between low testosterone and/or DHT and risk of DM or metabolic syndrome.31 Our findings that individuals with low free DHT were more likely to have increased BMI, HTN, and DM support the possibility that metabolic syndrome may have had a proarrhythmic effect33, 34, 35, 36 associated with low androgen levels; although, interestingly, only HTN was independently associated with an increased risk of incident AF in this study, and there was no evidence of an interaction with low levels of DHT and any of these covariables. Clearly, more work is needed regarding the role and association between circulating androgens, metabolic syndrome, and risk of AF.

4.1. Study limitations

Among the key limitations of this study were that we were only able to assess circulating levels of androgens at a single point in time, rather than longitudinally. As such, it is unknown whether a given individual's testosterone or DHT level decreased over time and whether that decrease itself played a role in incident AF, or whether the low level was merely a marker of other risk factors (eg, metabolic syndrome, as above) that independently resulted in the risk of AF. This limitation is important because it also fails to provide insight into the controversial question of whether hormone replacement to achieve “normal” levels of testosterone and/or DHT might help mitigate risk of AF. Clearly, further study is needed to assess both of these questions.

Another key limitation of this study was the post hoc decision to subdivide the subjects according to a cutpoint for “low” free DHT. We used sensitivity analysis to demonstrate that the cutpoint selected (<0.16 ng/dL) was not cherry‐picked specifically to allow a significant result, and we employed several nonlinear methods to determine that there was indeed evidence of an increased risk isolated to those with low levels of serum DHT. Androgen deficiency is a clinical condition that has been recognized for many years,37 and it is not entirely unfeasible that there would be a nonlinear effect once levels of DHT or testosterone drop below a certain threshold. Clearly, more work is needed to uncover the possible mechanisms for such an effect. However, it is likely that, as more and more biomarker data become available in observational datasets, that such nonlinear effects will be increasingly recognized.38

5. CONCLUSION

We found that in a large cohort of older men, there was evidence for a nonlinear association in which only men with low serum free DHT were at an increased risk of incident AF, after adjustment for age, demographics, and clinical risk factors. We were unable to detect a linear or nonlinear association with serum testosterone and risk of incident AF. More work is needed to examine the mechanisms and implications, particularly for hormone replacement therapy, of this finding.

Conflicts of interest

The authors declare no potential conflicts of interest.

Supporting information

Supplemental Figure 1. Age‐stratified univariable Cox regression by quintile of hormone level. All graphs are margin plots with error bars indicating 95% confidence intervals.

Supplemental Figure 2. Age‐stratified univariable Cox regression with restricted Cubic splines (knots = 4).

Supplemental Figure 3. Age‐stratified multivariable restricted spline Cox regression, based on mvrs function in Stata. All models adjusted for clinical risk factors (body mass index, standing height, diabetes, use of antihypertensive medications, systolic blood pressure (mmHg), depressed left ventricular function, serum Cystatin C level, and use of loop diuretics) and demographics (clinic site, level of education, income, smoking status and race). Models in A‐D (free and total testosterone and DHT) include SHBG. Model E includes total T and DHT.

Supplemental Table 1. Linear Associations with Incident AF

Supplemental Table 2. Comparison of Groups After Mahalanobis Distance Matching

Rosenberg MA, Shores MM, Matsumoto AM, et al. Serum androgens and risk of atrial fibrillation in older men: The Cardiovascular Health Study. Clin Cardiol. 2018;41:830–836. 10.1002/clc.22965