Pre-diagnostic free androgen and estradiol levels influence heart failure risk in both women and men: a prospective cohort study in the UK Biobank

Abstract

Aims:

Serum sex hormones have been linked to CVD risk. However, their roles in the pathogenesis of HF in both men and women are unclear. We investigated the associations between free androgen, testosterone, and estradiol, and future risk of HF.

Methods and results:

This prospective cohort study evaluated UK Biobank participants free of prevalent CVD and HF at baseline. Unitless free androgen, testosterone, and estradiol indices were generated using serum concentrations of total testosterone (nmol/L), estradiol (pmol/L), sex hormone binding globulin (SHBG, nmol/L), and albumin (g/L) in blood collected at enrollment. Multivariable Cox regression was used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) of incident HF in relation to quartiles (Q) of free androgen, testosterone, estradiol, and potential confounders. There were 180,712 men (including 5585 HF cases with free androgen and 571 HF cases with free estradiol), and 177,324 women (including 2858 HF cases with free androgen and 314 HF cases with free estradiol) with complete data. Increased free androgen was associated with decreased HF risk in both men (HR_{Q4 vs. Q1}=0.86, 95%CI: 0.79–0.94, p-trend_continuous<0.0001) and post-menopausal women (HR_{Q4 vs. Q1}=0.83, 95%CI: 0.73–0.95). Similar inverse associations were observed for free testosterone only in men (HR_{Q4 vs. Q1}=0.91, 95%CI: 0.83–0.98). Higher free estradiol was significantly associated with decreased HF risk among men (HR_{Q4 vs. Q1}=0.76, 95%CI: 0.59–0.98), but was positively associated among pre-menopausal women (HR_{Q4 vs. Q1}=2.16, 95%CI: 1.11–4.18).

Conclusions:

Sex hormones potentially influence heart failure pathogenesis and may offer pathways for interventions.

Graphical Abstract

Introduction

Heart failure (HF) syndrome is a complex clinical syndrome that can result from cardiovascular diseases (CVDs) that impair cardiac function, including coronary heart disease, heart inflammation, high blood pressure, and cardiomyopathy.^1,2 HF is a major global health concern in Europe and worldwide, leading to high mortality, frequent hospitalizations, and heavy economic costs.³ Understanding biological mechanisms underlying HF pathogenesis could alleviate the public health burden of HF by identifying intermediate causal biomarkers that could be targeted using clinical or population-level interventions.

Testosterone is the primary ‘male sex hormone’, while estradiol is the predominant bioactive form of estrogen that is considered the primary ‘female sex hormone’. However, testosterone and estradiol are produced in both men and women,⁴ and influence health outcomes in both sexes.⁵ They have been shown to be potent mitogens and important regulators of gene expression through interactions with tissue-specific androgen receptors (AR) and estrogen receptors (ER), respectively.^6,7 Given that AR and ER motifs are localized in genes of different cardiac tissues and are differentially expressed among men and women,⁸ testosterone and estradiol can influence cardiovascular outcomes in both sexes.

Epidemiological studies have found associations between alterations to sex hormones and HF risk, as well as CVDs on the causal pathway to HF syndrome. However, the findings have been largely inconsistent and controversial.^9,10 Further, the cardiovascular effects of estradiol among men and testosterone among women are less appreciated. To shed light on the sex-specific roles of circulating androgens and estrogens in the pathogenesis of HF, we estimate the associations between free androgen, free testosterone, and free estradiol and future risk of HF in both men and women in the UK Biobank.

Methods

Study design

The UK Biobank has been described in detail elsewhere (http://www.ukbiobank.ac.uk/).^11,12 Briefly, the source population was adults aged 40–69 years who lived ≤40 km of 22 study assessment centers situated throughout the UK constituent countries of England, Wales, and Scotland. Approximately 9.2 million people registered in the National Health Service (NHS) were mailed invitations to participate in the study. Among these people, 503,317 (5.5%) visited the assessment centers in 2006–2010 and were enrolled.¹¹ The participants were given touchscreen questionnaires, physical examinations, and provided biological samples for molecular/genetic analyses. Our dataset from August 2022 included 502,409 participants (project number: 28072).

Heart failure diagnosis

We defined HF using in-patient hospital diagnoses coded according to the International Classification of Disease (ICD) version 9 (428.0 and 428.1) and ICD-10 (I50.0, I50.1, I50.9, I11.0, I13.0, and I13.2) classifications.^13,14 Compared to chart reviews, ICD-10 codes were previously found to accurately predict HF (sensitivity=68.6%, specificity=99.3%, positive predictive value=90.2%, and negative predictive value=97.2%).¹⁵

Inclusion / exclusion criteria

Among the 502,409 study participants at baseline, we sequentially excluded 372 participants with discrepancy between genetic and self-reported sex, 40 voluntary dropouts since obtaining the dataset, 2257 participants with prevalent HF, and 23,199 participants with prevalent CVDs that are on the causal pathway to HF or are strong risk factors (i.e., atrial fibrillation, angina pectoris, acute myocardial infarction, subsequent ST elevation and non-ST elevation myocardial infarction and its complications, chronic/acute ischemic heart diseases, cardiomyopathy, myocarditis, intermediate coronary syndrome, coronary atherosclerosis, acute pericarditis, occlusion/stenosis of precerebral arteries, and pulmonary edema (which is comorbid with HF)) (Figure 1).

Figure 1:

Flow chart of the UK Biobank study population

Prospective follow-up

The prospective follow-up started for each participant at the date of visit to the assessment center in 2006–2010 and ended at the date of first incident HF diagnosis, death, or administrative censoring (i.e., September 20^th, 2021, for England and Wales and October 31^st, 2021, for Scotland), whichever came first. Vital status and the primary underlying cause of death for participants were provided by the NHS Information Centre and the NHS Central Register. The UK Biobank study was approved by the National Information Governance Board for Health and Social Care and the NHS North West Multicenter Research Ethics Committee. All participants provided electronic informed consent.

Measurement of serum sex hormone concentrations

Blood biospecimens were collected at the time of enrollment in serum separator tubes, transported at 4 °C to the central UK Biobank processing laboratory, aliquoted, and stored at − 80 °C.¹⁶ Serum testosterone (nmol/L) was analyzed using a one-step competitive immunoassay, estradiol (pmol/L) was measured using a two-step competitive immunoassay, and sex hormone binding globulin (SHBG, nmol/L) was measured using a two-step sandwich immunoassay. These hormones were measured using a Beckman Coulter Unicel DxI 800 as described: (https://biobank.ndph.ox.ac.uk/showcase/refer.cgi?id=5636). Additionally, albumin (g/L) was measured using colorimetric assays on a Beckman Coulter AU5800. Full information on serum biochemistry and quality control procedures can be found at: (https://biobank.ndph.ox.ac.uk/showcase/refer.cgi?id=1227). The lower limits of detection for testosterone, estradiol, SHBG and albumin were <0.35 nmol/L, <175 pmol/L, <0.33 nmol/L, and <15 g/L, respectively. The UK Biobank did not distinguish between values above and below the limits of detection, therefore, missing values for these reasons could not be imputed.

A unitless free androgen index (FAI) was generated by dividing baseline total testosterone concentrations (nmol/L) by SHBG concentrations (nmol/L) and multiplying by a factor of 100.¹⁷ Similarly, a unitless free estradiol index (FEI) was generated by dividing baseline estradiol concentrations (nmol/L; converted from pmol/L) by SHBG concentrations (nmol/L) and multiplying by a factor of 100,000.^18–21 The free testosterone index (FTI) incorporated information on total testosterone, SHBG, and albumin and was calculated and validated as described. ²²

Statistical analyses

Spearman rank correlations were estimated between free androgen index, free testosterone index, free estradiol index, leukocyte telomere length (LTL), and other biological markers previously used to generate a Biological Health Score (BHS) reflective of biological aging.^23,24

Multivariable Cox regression models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) of incident HF (first hospitalization) in relation to quartiles (Q) of free androgen, testosterone, and estradiol indices, in separate (single-hormone) and mutually-adjusted (bi-hormone) analyses. All analyses were conducted independently for men and women. Among men, the models were adjusted for potential confounders including study assessment center, age at recruitment (continuous), race/ethnicity/ancestry (European-ancestry, Black/African ancestry, mixed, South Asian, and other groups), educational attainment (i.e., qualifications), smoking status (never, former, and current), body mass index (BMI; <18.5, 18.5≤ to <25, 25≤ to <30, 30≤ to <35, and ≥35 kg/m²), material deprivation (Townsend deprivation index, socioeconomic status, continuous), alcohol intake (never, former, current occasional, current <1 drink/day, current 1–3 drinks/day, current >3 drinks/day, and unknown), diabetes status (none, diabetic, and unknown), glycated hemoglobin (HbA1c, mmol/mol, continuous), and hypertension status based on average systolic and diastolic blood pressure (DBP) at baseline (normal, elevated, stage 1 and 2 hypertension, hypertensive crisis, and unknown). Among women, we adjusted for the same covariates as men in addition to ever oral contraceptive use, ever hormone replacement therapy use, and menopausal status (pre-menopause, post-menopause, artificial menopause, and missing/unknown). For the underlying timescale, we used follow-up time to improve consistency with previous analyses.²⁵ We tested for multiplicative interaction between continuous free androgen and estradiol using cross-product terms. Additionally, we evaluated combined associations (‘joint effects’) by dichotomizing free androgen and free estradiol at their 75^th percentile and creating a cross-classification variable (i.e., low free androgen index and low free estradiol index (reference); low free androgen and high free estradiol; high free androgen and low free estradiol; high free androgen and high free estradiol).

Non-linear relationships were evaluated using separate cubic spline models with five manual knots. Here, we adjusted for the same covariates noted above, but removed study assessment site when needed because of unstable estimates from sparse data.

We conducted additional a priori specified sensitivity analyses to test the robustness of the overall findings, as well as stratified analyses by ancestry (i.e., European and Black/African), age (by 4-year age groups), and BMI (<18.5 kg/m², 18.5≤ to <25 kg/m², 25≤ to <30 kg/m², 30≤ to <35 kg/m², and ≥35 kg/m²) when there were sufficient case numbers. Among women, we also stratified by menopausal status (i.e., pre-menopause, natural post-menopausal, and artificial menopause from hysterectomy or bilateral oophorectomy usually from cancer care). We conducted attenuation analyses including LTL and biomarkers that compose the BHS. All analyses were conducted using SAS v. 9.4 (Cary, North Carolina, USA). Two-sided p-values <0.05 were considered statistically significant.

Results

Descriptive characteristics at enrollment

The final analytic sample included 476,541 participants. Baseline characteristics of study population are shown in Table 1. Women comprised 55.7% of the study population. The vast majority of the participants were of European ancestry (94.18% for women, 93.93% for men), while those of Black/African ancestry made up a small fraction of the study population (1.7% for women, 1.56% for men). The average ages at recruitment were 56.19 (8.00 SD) years for women and 56.32 (8.21 SD) years for men. The average total estradiol levels were 537.90 (468.88 SD) pmol/L for women and 223.17 (82.83 SD) pmol/L for men. Furthermore, the average total testosterone concentrations were 1.12 (0.64 SD) nmol/L for women and 12.04 (3.71 SD) nmol/L for men.

Table 1:

Descriptive characteristics of women and men from the overall UK Biobank study population

Characteristic	WOMEN			MEN
	n=265,440			n=211,101
Age at recruitment, years, n, mean, SD	265,440	56.19	8.00	211,101	56.32	8.21
Body mass index, kg/m2, n, %
<18.5	1,843	0.69		479	0.23
≥18.5 to <25	101,222	38.13		52,076	24.67
≥25 to <30	96,745	36.45		103,395	48.98
≥30 to <35	40,785	15.37		40,100	19.00
≥35	20,187	7.61		11,226	5.32
Missing/unknown	4,658	1.75		3,825	1.81
Smoking status, n, %
Never	158,179	59.59		105,148	49.81
Former	82,306	31.01		78,253	37.07
Current	23,521	8.86		26,439	12.52
Unknown	1,434	0.54		1,261	0.60
Self-reported Race/Ethnicity/Ancestry, n, %
White/European	250,002	94.18		198,297	93.93
Mixed	1,815	0.68		1,043	0.49
South Asian	3,594	1.35		3,828	1.81
Black/African	4,503	1.70		3,289	1.56
Other groups	4,306	1.62		3,275	1.55
Missing/unknown	1,220	0.46		1,369	0.65
Townsend deprivation index (socioeconomic status), n, mean, SD	265,121	−1.36	3.03	210,829	−1.29	3.14
Education (qualifications), n, %
College or University degree	83,152	31.33		72,563	34.37
A levels/AS levels or equivalent	31,317	11.80		21,815	10.33
O levels/GCSEs or equivalent	61,529	23.18		39,044	18.50
CSEs or equivalent	14,283	5.38		11,639	5.51
NVQ or HND or HNC or equivalent	11,731	4.42		18,734	8.87
Other professional qualifications	15,075	5.68		9,149	4.33
Unknown/Missing/No Answer	48,353	18.22		38,157	18.08
Alcohol intake, n, %
Never	15,132	5.70		5,730	2.71
Former	9,338	3.52		6,985	3.31
Current occasional	73,971	27.87		33,797	16.01
Current <1 drink/day	76,988	29.00		42,534	20.15
Current 1–3 drinks,day	79,720	30.03		91,116	43.16
Current >3 drinks/day	9,435	3.55		30,118	14.27
Unknown	856	0.32		821	0.39
Diabetes status, n, %
None	255,079	96.10		197,128	93.38
Diabetic	9,229	3.48		12,710	6.02
Unknown	1,132	0.43		1,263	0.60
Glycated hemoglobin, HbA1c, mmol/mol, n, mean, SD	245,644	35.69	5.79	196,816	36.21	7.27
Hypertension status, n, %
Normal	49,692	18.72		16,032	7.59
Elevated	31,536	11.88		21,726	10.29
Stage 1 or 2	154,922	58.36		150,039	71.07
Hypertensive crisis	4,928	1.86		4,703	2.23
Unknown	24,362	9.18		18,601	8.81
Menopausal Status, n, %
Pre-menopause	63,626	23.97		N/A
Post-menopause	159,688	60.16
Hysterectomy	29,680	11.18
Bilateral Oophorectomy	183	0.07
Unknown/Missing	12,263	4.62
Ever Oral Contraceptive Use, n, %
Yes	214,869	80.95		N/A
No	49,235	18.55
Unknown/Missing	1,336	0.50
Ever Hormone Replacement Therapy, n, %
Yes	99,463	37.47		N/A
No	164,478	61.96
Unknown/Missing	1,499	0.56
Ever Methyltestosterone/Testosterone Replacement Medication Use, n %
Yes	N/A			338	0.16
No				210,763	99.84
Total Serum Estradiol (E2) concentration, pmol/L, n, mean, SD	57,403	537.90	468.88	16,838	223.17	82.83
Total Serum Testosterone (T) concentration, nmol/L, n, mean, SD	206,765	1.12	0.64	196,263	12.04	3.71
Total Serum Sex Hormone Binding Globulin (SHBG) concentration, nmol/L, n, mean, SD	222,182	62.21	31.01	181,728	39.49	16.73
Total Serum Albumin concentration, g/L, n mean, SD	224,781	44.96	2.60	183,166	45.57	2.60
Free Estradiol Index (FEI), n, mean, SD	51,219	877.2	811.2	15,543	638.66	646.82
Free Androgen Index (FAI), n, mean, SD	187,009	2.3	2.1	180,712	33.80	14.98
Free Testosterone Index (FTI), n mean, SD	186,874	0.015	0.011	180,581	0.214	0.063

Abbreviations: standard deviation (SD)

With respect to key anthropometric and lifestyle factors that influence sex hormone levels, 24.31% of men and 22.97% of women had a high BMI ≥30 kg/m²; 8.86% of women and 12.52% of men were current smokers; only 3.55% of women and 14.27% of men were heavy drinkers (>3 drinks/day). With respect to reproductive factors that influence sex hormone levels among women, 60.16% were in post-menopause, 11.25% had either hysterectomy or bilateral oophorectomy (artificial menopause), 80.95% had ever used oral contraceptives, and 37.47% had ever used hormone replacement therapy. Among men, only a miniscule proportion (0.16%) had ever used methyltestosterone/testosterone replacement medication.

Associations between free androgen, free testosterone, free estradiol, and future risk of heart failure among women and men

After excluding prevalent HF and CVDs, we analyzed 211,101 men and 265,440 women. Among these subjects, 180,712 men (including 5585 HF cases with free androgen and 571 HF cases with free estradiol), and 177,324 women (including 2858 HF cases with free androgen and 314 HF cases with free estradiol) had non-missing analyzable data.

The free androgen and free testosterone indices were highly correlated among men (Spearman rho=0.81, p<0.0001) and overall women (Spearman rho=0.98, p<0.0001). In the single-hormone analyses, we found that participants with the highest free androgen levels had significantly decreased risk of HF compared to those with the lowest free androgen levels, which was consistent in both sexes (Table 2). Among men, the protective effect of increased free androgen was observed for the second (HR=0.91, 95% CI: 0.85–0.97), third (HR=0.83, 95% CI: 0.77–0.89), and fourth (HR=0.86, 95% CI: 0.79–0.94) quartiles (p-trend<0.0001). Similar inverse associations with HF risk were observed for free testosterone among men (Table 3). Here, the inverse association with free testosterone was non-monotonic, with a nadir at the 2^nd quartile (HR=0.90, 95% CI: 0.84–0.96; Table 3). Among women, the protective effect of free androgen was observed only among those with the highest levels (HR=0.87, 95% CI: 0.77–0.97) and the trend was non-monotonic (Table 2). However, associations between free testosterone and HF risk were not detected in women (Table 3)

Table 2:

Pre-diagnostic free androgen and estradiol levels and HF risk among women and men from the UK Biobank

	Women							Men
Free Androgen Index	Categorical Range	No. of incident cases	HR	95% CI Lower	95% CI Upper	P-value		Categorical Range	No. of incident cases	HR	95% CI Lower	95% CI Upper	P-value
Quartile 1 (Reference)	0.15≤ to <1.13	634	1.00					0.40≤ to <25.33	2044	1.00
Quartile 2	1.13≤ to <1.79	677	0.96	0.86	1.07	0.47		25.33≤ to <31.46	1548	0.91	0.85	0.97	5E-03	*
Quartile 3	1.79≤ to <2.84	750	0.95	0.85	1.06	0.38		31.46≤ to <39.43	1134	0.83	0.77	0.89	7E-07	*
Quartile 4	≥2.84	797	0.87	0.77	0.97	0.01	*	≥39.43	859	0.86	0.79	0.94	7E-04	*
					P-trend	0.17						P-trend	<0.0001	*
Free Estradiol Index
Quartile 1 (Reference)	79.33≤ to <408.52	94	1.00					73.22≤ to <402.70	179	1.00
Quartile 2	408.52≤ to <646.95	65	0.84	0.61	1.17	0.31		402.70≤ to <538.88	157	0.90	0.72	1.12	0.35
Quartile 3	646.95≤ to <1068.22	78	1.14	0.83	1.56	0.42		538.88≤ to <745.13	132	0.87	0.69	1.10	0.24
Quartile 4	≥1068.22	77	1.40	1.02	1.93	0.04	*	≥745.13	103	0.76	0.59	0.98	0.03	*
					P-trend	5E-03						P-trend	0.01	*

Multivariable Cox regression models were used to estimate hazard ratios (HR) and 95% confidence intervals (CI) of incident HF in relation to free androgen index and free estradiol index, separately. Fee testosterone index and free estradiol index were primarily analyzed as quartiles. Among men, the models were further adjusted for potential confounders including study assessment center, age at recruitment (continuous), race/ethnicity (White European, Black/African ancestry, mixed, South Asian, and other groups), educational attainment (i.e., qualifications), smoking status (never, former, current), body mass index (BMI; <18.5, ≥18.5 to <25, ≥25 to <30, ≥30 to <35, and ≥35 kg/m²), material deprivation (Townsend deprivation index, continuous), alcohol intake (never, former, current occasional, current <1 drink/day, current 1–3 drinks/day, current >3 drinks/day, unknown), diabetes status (none, diabetic, unknown), glycated hemoglobin (HbA1c, mmol/mol, continuous), and hypertension status based on average systolic and diastolic blood pressure at baseline (normal, elevated, stage 1 and 2 hypertension, hypertensive crisis, and unknown). Among women, we adjusted for the same covariates as men in addition to ever oral contraceptive use, ever hormone replacement therapy use, and menopausal status (pre-, post-, artificial, or missing/unknown). Free androgen index was generated by dividing baseline testosterone concentrations (nmol/L) by sex hormone binding globulin (SHBG) concentrations (nmol/L) and multiplying by a factor of 100. Free estradiol index was generated by dividing baseline estradiol concentrations (nmol/L) by SHBG concentrations (nmol/L) and multiplying by a factor of 100,000. Tests of trend were calculated using continuous free androgen index and free estradiol index values separately, adjusted for same the covariates noted above. We analyzed participants with complete free androgen index, free estradiol index, and covariate data. The categories for missingness were not shown.

^*P-values <0.05.

Table 3:

Pre-diagnostic free testosterone and heart failure risk in the UK Biobank

	Men						Pre-menopausal Women a					Pre-menopausal Women
Free Testosterone Index	No. incident cases	HR	95% CI Lower	95% CI Upper	P-value		No. incident cases	HR	95% CI Lower	95% CI Upper	P-value	No. incident cases	HR	95% CI Lower	95% CI Upper	P-value
Quartile 1 (Reference)	2047	1.00					31	1.00				505	1.00
Quartile 2	1474	0.90	0.84	0.96	2.8E-03	*	34	0.92	0.56	1.50	0.73	558	1.01	0.89	1.14	0.93
Quartile 3	1155	0.87	0.81	0.94	2.7E-04	*	45	0.93	0.58	1.50	0.77	586	0.98	0.87	1.11	0.76
Quartile 4	904	0.91	0.83	0.98	0.02	*	76	1.11	0.71	1.74	0.64	591	0.89	0.79	1.01	0.08
				P-trend	5.5E-05	*				P-trend	0.36				P-trend	0.14

Multivariable Cox regression models were used to estimate hazard ratios (HR) and 95% confidence intervals (CI) of incident heart failure in relation to free testosterone index (FTI) analyzed as quartiles. Among men, the models were further adjusted for potential confounders including study assessment center, age at recruitment (continuous), race/ethnicity (White European, Black/African ancestry, mixed, South Asian, and other groups), educational attainment (i.e., qualifications), smoking status (never, former, current), body mass index (BMI; <18.5, ≥18.5 to <25, ≥25 to <30, ≥30 to <35, and ≥35 kg/m2), material deprivation (Townsend deprivation index, continuous), alcohol intake (never, former, current occasional, current <1 drink/day, current 1–3 drinks/day, current >3 drinks/day, unknown), diabetes status (none, diabetic, unknown), glycated hemoglobin (HbA1c, mmol/mol, continuous), and hypertension status based on average systolic and diastolic blood pressure at baseline (normal, elevated, stage 1 and 2 hypertension, hypertensive crisis, and unknown). Among women, we stratified by menopausal status and adjusted for the same covariates as men in addition to ever oral contraceptive use and ever hormone replacement therapy use. Tests of trend were calculated using continuous FTI adjusted for same the covariates noted above. FTI category for missingness is not shown.

^aWe note that some of the covariate estimates were unstable in the analyses of pre-menopausal women.P-values <0.05.

With respect to estradiol, the highest free estradiol level was associated with a significantly increased risk of HF compared to the lowest level among women (HR=1.40, 95% CI: 1.02–1.93; p-trend=0.005; Table 2). Conversely, the highest free estradiol level was associated with a significantly decreased risk of HF compared to the lowest free estradiol level among men (HR=0.76, 95% CI: 0.59–0.98), and the trend was non-monotonic (p-trend=0.01).

In the bi-hormone analyses, we found inverse associations between free androgen and HF risk in both sexes (Figure 2A and 2B) similar to the single-hormone analyses (Table 2). We also found ‘opposing effects’ of free estradiol between women and men (Figure 2C and 2D) similar to the single-hormone analyses (Table 2); however, the inverse association among men was non-significant. We found no evidence for multiplicative interaction between free androgen and free estradiol among women (P-interaction=0.73) and men (P-interaction=0.66). Additionally, we did not find evidence of synergistic combined associations between free androgen and free estradiol using a cross-classification variable (data not shown).

Figure 2: Mutually-adjusted pre-diagnostic free androgen and estradiol levels and heart failure risk in the UK Biobank.

Multivariable Cox regression models were used to estimate hazard ratios (HR) and 95% confidence intervals (CI) of incident heart failure in relation to mutually-adjusted free androgen index (FAI) and free estradiol index (FEI) (i.e., Two-hormone models). FAI and FEI were primarily analyzed as quartiles. Among men, the models were further adjusted for potential confounders including study assessment center, age at recruitment (continuous), race/ethnicity (White European, Black/African ancestry, mixed, South Asian, and other groups), educational attainment (i.e., qualifications), smoking status (never, former, current), body mass index (BMI; <18.5, ≥18.5 to <25, ≥25 to <30, ≥30 to <35, and ≥35 kg/m²), material deprivation (Townsend deprivation index, continuous), alcohol intake (never, former, current occasional, current <1 drink/day, current 1–3 drinks/day, current >3 drinks/day, unknown), diabetes status (none, diabetic, unknown), glycated hemoglobin (HbA1c, mmol/mol, continuous), and hypertension status based on American Heart Association/American College of Cardiology cutoffs using average systolic and diastolic blood pressure at baseline (normal, elevated, stage 1 and 2 hypertension, hypertensive crisis, and unknown). Among women, we adjusted for the same covariates as men in addition to ever oral contraceptive use, ever hormone replacement therapy use, and menopausal status (pre-, post-, artificial, or missing/unknown). FAI was generated by dividing baseline testosterone concentrations (nmol/L) by sex hormone binding globulin (SHBG) concentrations (nmol/L) and multiplying by a factor of 100. FEI index was generated by dividing baseline estradiol concentrations (nmol/L) by SHBG concentrations (nmol/L) and multiplying by a factor of 100,000. Tests of trend were calculated using continuous FAI and FEI values separately, adjusted for same the covariates noted above. P-values <0.05. Tests of trend were calculated using mutually-adjusted continuous FAI and FEI values adjusted for same the covariates noted above. P-values <0.05. Among women, there were a total of 4052 incident cases among 233,030 analyzable participants with complete data. Among men, there were a total of 6153 incident cases among 196,563 analyzable participants with complete data. The categories for missingness are not shown.

When stratifying women by menopausal status, we found that the inverse association between free androgen and HF risk was consistent in direction across subgroups; however, the estimates were only statistically significant among post-menopausal women (HR_{Q4 vs Q1}: 0.83, 95% CI: 0.73–0.95; Table 4). Furthermore, the association between elevated free estradiol and increased HF risk was statistically significant only among pre-menopausal women (HR_{Q4 vs Q1}: 2.16, 95% CI: 1.11–4.18; Table 4).

Table 4:

Pre-diagnostic free androgen and estradiol levels and HF risk among women by menopausal status in the UK Biobank

		Pre-menopausea					Post-menopause					Artificial menopauseb
Free Androgen Index	Categorical Range	HR	95% CI Lower	95% CI Upper	P-value		HR	95% CI Lower	95% CI Upper	P-value		HR	95% CI Lower	95% CI Upper	P-value
Quartile 1 (Reference)	0.15≤ to <1.13	1.00					1.00					1.00
Quartile 2	1.13≤ to <1.79	0.98	0.59	1.62	0.94		0.96	0.84	1.08	0.48		0.99	0.74	1.32	0.95
Quartile 3	1.79≤ to <2.84	1.08	0.66	1.76	0.75		0.95	0.84	1.07	0.40		0.93	0.69	1.24	0.61
Quartile 4	≥2.84	1.00	0.62	1.62	1.00		0.83	0.73	0.95	6E-03	*	0.94	0.70	1.25	0.66
				P-trend	0.77				P-trend	0.96				P-trend	0.22
Free Estradiol Index
Quartile 1 (Reference)	79.33≤ to <408.52	1.00					1.00					1.00
Quartile 2	408.52≤ to <646.95	1.25	0.60	2.59	0.55		0.82	0.51	1.32	0.41		0.89	0.47	1.65	0.70
Quartile 3	646.95≤ to <1068.22	1.94	1.00	3.79	0.05	*	1.17	0.72	1.90	0.52		0.78	0.38	1.61	0.50
Quartile 4	≥1068.22	2.16	1.11	4.18	0.02	*	1.13	0.59	2.18	0.71		1.43	0.70	2.91	0.32
				P-trend	0.20				P-trend	0.03	*			P-trend	0.54

Multivariable Cox regression models were used to estimate hazard ratios (HR) and 95% confidence intervals (CI) of incident HF in relation to mutually-adjusted free androgen index and free estradiol index. Free androgen index and free estradiol index were primarily analyzed as quartiles. The models were adjusted for potential confounders including study assessment center, age at recruitment (continuous), race/ethnicity (White European, Black/African ancestry, mixed, South Asian, and other groups), educational attainment (i.e., qualifications), smoking status (never, former, current), body mass index (BMI; <18.5, ≥18.5 to <25, ≥25 to <30, ≥30 to <35, and ≥35 kg/m²), material deprivation (Townsend deprivation index, continuous), alcohol intake (never, former, current occasional, current <1 drink/day, current 1–3 drinks/day, current >3 drinks/day, unknown), diabetes status (none, diabetic, unknown), glycated hemoglobin (HbA1c, mmol/mol, continuous), and hypertension status, ever oral contraceptive use, and ever hormone replacement therapy use. Free androgen index was generated by dividing baseline testosterone concentrations (nmol/L) by sex hormone binding globulin (SHBG) concentrations (nmol/L) and multiplying by a factor of 100. free estradiol index was generated by dividing baseline estradiol concentrations (nmol/L) by SHBG concentrations (nmol/L) and multiplying by a factor of 100,000.

^*P-values <0.05. Tests of trend were calculated using continuous free androgen index and free estradiol index values separately, adjusted for same the covariates noted above.

Among pre-menopausal women, 35,412 participants including 128 HF cases had free androgen index data, and 48,108 participants including 187 cases had free androgen index data. Among post-menopausal women, 7373 participants including 120 HF cases had free estradiol index data, and 110,174 participants including 2242 cases had free androgen index data. Among women in artificial menopause, 4370 participants including 66 HF cases had free estradiol index data, and 19,042 participants including 429 cases had free androgen index data. We analyzed participants with complete free androgen index, free estradiol index, and covariate data. The categories for missingness were not shown.

^aWe note that some of the covariate estimates were unstable in the analyses of pre-menopausal women.

^bArtificial menopause was defined as women who underwent hysterectomy or bilateral oophorectomy.

Sensitivity and attenuation analyses

The findings were robust even when further adjusting for parity (0, 1, 2, ≥3), or metabolism-altering medications such as oral contraceptives and hormone replacement therapy (data not shown). The trends were also similar when stratifying by age and BMI categories (data not shown). When restricting the bi-hormone analyses to European-ancestry participants, we found similar associations between free androgen and estradiol, and HF risk (online supplementary Table S1) as the overall analyses (Figure 2). When restricting to those of Black/African ancestry, we only had sufficient data to analyze free androgen alone (online supplementary Table S2). Here, we found a similar inverse association between free androgen and HF risk among Black/African men, but the results were non-significant among comparable women (online supplementary Table S2). Lastly, adjusting the models for LTL and each individual biomarker composing the BHS did not influence the estimates >10% (data not shown).

Cubic spline analyses to assess non-linear trends

Given that we found evidence of non-monotonic relationships between free androgen, free estradiol, and HF risk in the single-hormone and bi-hormone quartile analyses, we conducted ad hoc analyses to further characterize non-linear relationships (online supplementary Figures S1–S4). Among men, we found that the association between high free androgen and decreased risk of HF was non-linear and reached a nadir at approximately free androgen=30, which then leveled off at higher values (p_non-linear<0.0001). Conversely, we did not find compelling evidence that inverse association between free estradiol and HF risk was non-linear among men (p_non-linear=0.41). Among women, there was a bathtub-curved relationship between free androgen and HF risk; however, the test of non-linearity was non-significant (p_non-linear=0.09). Lastly, the cubic spline analyses showed no evidence of non-linear relationships between free estradiol and HF risk among women (p_non-linear=0.71).

Correlations between free androgen, free estradiol, and biomarkers related to biological aging

We considered whether the observed associations between free androgen, free estradiol, and HF risk among women and men could be potentially driven by correlations with some noteworthy biomarkers that were previously found to be related to risk of HF (online supplementary Table S3).^23,26 Among men and women, we found that free androgen and free estradiol had low correlations with LTL, overall BHS and its constituent biomarkers.

Discussion

In a large CVD-free study population, we found that increased free androgen in pre-diagnostic serum was associated with decreased risk of newly diagnosed HF in both men and post-menopausal women, with non-linear relationships. Similar inverse associations were found for free testosterone, but only among men. Additionally, free estradiol had a protective effect among men, but was positively associated among pre-menopausal women. When stratified by racial/ethnic subgroups, the results and trends were similar to the main analyses.

Our study expands upon previous reports with novel and important findings. Consistent with our results, the prospective ARIC study found inverse associations between total testosterone in plasma and HF risk among men.⁹ The peripheral effects of testosterone may explain its beneficial effects in the pathophysiology of HF syndrome in some subgroups. Animal studies suggest that testosterone affects vascular reactivity by acting on synthesis and release of nitric oxide (NO), a powerful vasodilator.²⁷ However, the sex-specific mechanistic understanding of biological responses to testosterone in the pathophysiology of HF syndrome is unclear, and cardiac effects of testosterone in men and women need to be explored.

With respect to estradiol, we observed a protective effect of increased free estradiol on HF risk among men. Supporting our findings, epidemiologic and experimental studies have shown a cardioprotective effect of estrogens.²⁸ However, we found a positive association between free estradiol and HF risk among pre-menopausal women. Taken together, these results are consistent with the “timing hypothesis”, in which the opposite effect of estrogen on CVD outcomes is observed among pre- and post-menopausal women.²⁸ In the Nurses’ Health Study, women initiating hormone replacement therapy at or near menopause were observed to experience significant coronary heart disease protection, whereas the those who started treatment over 10-years after menopause did not.²⁹ Additionally, analyses in the Women’s Health Initiative estrogen plus progestin trial suggested that reduced coronary heart disease risk may appear only after several years of treatment.³⁰

Several previous studies suggested the roles played by sex hormones in predicting mortality in patients who already have established decompensation. Among men with stable chronic HF and reduced left ventricular ejection fraction (LVEF), regardless of gonadal and adrenal androgen deficiencies as well as conventional clinical prognostic indicators, a poor prognosis was predicted by both high and low concentrations of circulating estradiol.³¹ Additionally, men with either decreased or increased concentrations of serum estradiol have different clinical characteristics, indicating potential differences in the underlying pathophysiological mechanisms. Men with decreased concentration of estradiol have increased serum total testosterone, decreased serum dehydroepiandrosterone sulfate (DHEA-S), advanced New York Heart Association (NYHA) class, impaired renal function, and decreased total fat tissue mass compared with the middle quintal of estradiol. Men with increased concentration of estradiol have increased serum bilirubin and liver enzymes, and decreased serum sodium compared with the middle quintal of estradiol.³¹ Furthermore, testosterone deficiency has been associated with clinical parameters and prognosis in HF patients.^32,33 Among men with chronic HF, those with normal levels of all anabolic hormones revealed the best 3-year survival rate compared with those with anabolic hormone deficiency,³⁴ and multiple hormonal and metabolic deficiency syndrome was independently associated with increased all-cause mortality and cardiovascular hospitalization.³⁵

Our study had numerous strengths. First, the prospective cohort study design allowed for the establishment of temporality between the serum sex hormones, potential confounders, and HF diagnosis, which strengthened our inferences. Second, we excluded participants with prevalent CVD at enrollment that are on the causal pathway to or are strong risk factors for HF, thus minimizing the influence of disease-effect bias on baseline sex hormone levels. Third, the UK Biobank was linked to national hospital registries, thus we were confident that the vast majority of in-patient HF cases were captured in the study population during the follow-up period.

Our study had some limitations. First, although we found evidence that our testosterone findings were consistent among a limited number of African-ancestry participants, the study population was predominantly composed of European-ancestry participants, which limits the generalizability of our findings to other ethnic groups. Since the available data suggest that the aetiology of HF may differ depending on the income level of the country, the role of sex hormones in the pathogenesis HF may vary across countries. In western developed countries, coronary artery disease by itself or in combination with hypertension seems to be the most common cause of HF.³⁶ However, HF is more commonly caused by infectious, nutritional deficiencies, and inflammatory in sub-Saharan Africa than in middle- and high-income countries.³⁷ Second, we did not have information on HF subtypes and ejection fraction during the follow-up. However, this kind of misclassification would be non-differential and the observed effects would likely be conservative underestimates. Third, serum sex hormone concentrations were only measured once at enrollment irrespective of menstrual cycle, which potentially introduces non-differential misclassification of estradiol concentrations among pre-menopausal women. Fourth, we did not have data on other sex hormones such as dehydroepiandrosterone, which may have cardiovascular influence. Lastly, the analyses were restricted to participants with sex hormone measurements within the detectable range of the assays. Estradiol was only in the detectable range for ~7% of men and ~19% of women, which potentially limits generalizability. However, it is unlikely that participant enrollment was strongly related downstream to both estradiol levels and cardiovascular disease (i.e., collider bias).

In summary, we detected protective associations between free androgen and testosterone, and new HF diagnosis among men. In women however, the association differed by menopausal status as increased free androgen was associated with elevated risk of HF before menopause but not after menopause. Further, increased free estradiol levels was associated to elevated HF risk among pre-menopausal women but not post-menopausal women and men. Further epidemiologic studies of sex-specific etiologic mechanisms, particularly in diverse populations, are warranted. Our findings suggest that sex hormones influence the pathogenesis HF and may offer potential pathways for interventions.