Associations Between Urinary Metal Levels and Incident Heart Failure: A Multi-Cohort Analysis

Abstract

Background:

Environmental metals are recognized cardiovascular disease risk factors, yet the role of metal exposure in heart failure (HF) risk remains understudied.

Objectives:

Evaluate the prospective association of urinary metals with incident HF across three geographically and ethnically/racially diverse cohorts: the Multi-Ethnic Study of Atherosclerosis (MESA) and the Strong Heart Study (SHS) in the United States, and the Hortega Study in Spain.

Methods:

Adults 18–85 years old in MESA (n=6,644), SHS (n=2,917), and Hortega (n=1,300) were followed up to 20 years. Urinary levels of a multi-metal panel were measured at baseline and corrected for urine dilution. Cox proportional hazards and Cox-Elastic Net models were used to estimate the multi-adjusted (sociodemographic/clinical/lifestyle covariates) hazard ratio (HR) of incident HF by individual metals and the mixture of five metals available in all cohorts, respectively. The pooled HR(95%CI) of HF by one unit increase in log2 transformed levels of individual metals (i.e., doubling of the dose) across cohorts was estimated using a fixed effects meta-analysis. Analyses by left ventricular ejection fraction (LVEF) were conducted in a subset.

Results:

A total of 1,001 participants developed HF. In adjusted models, significant associations [pooled HRs (95%CI) per doubling of urinary metal] were identified for cadmium [1.15(1.07,1.24)], molybdenum [1.13(1.05,1.22)], and zinc [1.22(1.14,1.32)]. The HRs(95%CI) for the association of one interquartile range increase in the multi-metal mixture levels of five metals (arsenic, cadmium, molybdenum, selenium and zinc) and incident HF were 1.38(1.00,1.86) in MESA, 1.55(1.28. 1.97) in SHS, and 1.08(0.85, 1.63) in Hortega. Stratified models by LVEF were consistent with the pooled results.

Conclusions:

Urinary metals are risk factors of HF across three diverse populations, supporting the role of reducing metal exposures to lower HF risk.

Introduction:

Contaminant metals such as arsenic, cadmium and lead are established risk factors of cardiovascular disease (CVD), with evidence primarily available for ischemic heart disease.^1–4 Higher levels of essential metals copper, cobalt and zinc in the urine have also been associated with increased risk of CVD and subclinical atherosclerosis.^3,5 Previous studies have documented associations between contaminant metals with echocardiographic changes in cardiac function and geometry.^6–8 Despite suggestive evidence on the potential role of metals on HF pathophysiology, there is a limited number of prospective studies of metals and heart failure (HF), including studies evaluating the role of metal mixtures – which are more likely to mimic real life exposures – .

Metals are widespread in the environment from natural geologic processes and anthropogenic activities.⁹ Metal exposure is influenced by sociodemographic and structural factors resulting in inequities across population groups, with higher levels of exposure documented across non-Hispanic Black, Hispanic/Latino, American Indian, and communities with lower socioeconomic status.^10,11 These population groups also suffer a higher burden of HF.^12,13

Metal exposure and metal dysregulation in the body can impact cardiovascular health through different molecular mechanisms including systemic inflammation and oxidative stress, endothelial damage, atherosclerosis, and epigenetic modifications.^4,14 These molecular pathways can contribute to the pathogenesis and progression of HF endpoints, including subtypes(i.e., heart failure with preserved, mildly-reduced or reduced ejection fraction (HFpEF, HFmrEF, and HFrEF, respectively).¹⁵

In this study, we leveraged prospective data with more than 20 years of follow-up across three racial and ethnically diverse cohorts from different geographic areas, i) the Multi-Ethnic Study of Atherosclerosis (MESA), a study of US adults from 6 urban-suburban areas, ii) the Strong Heart Study (SHS), a large epidemiological cohort of American Indian adults in the US, and iii) the Hortega study, a cohort of adults from a general population in Spain. The main goal was to assess the prospective association of a multi-element panel of non-essential and essential urinary metal levels, as well as the effect of metal mixtures, on incident HF and HF subtypes risk.

Materials and methods:

Study population:

Participants from three different population-based cohorts were included in this study: The MESA study recruited 6,814 men and women 45–84 years free of clinical CVD from 6 US urban and sub-urban areas (Baltimore County, Maryland; Chicago, Illinois; Forsyth County, North Carolina; Los Angeles County, California; Northern Manhattan and the Bronx, New York; and St. Paul, Minnesota) in 2000–2002. Participants self-identified as White (38%), Black (28%), Hispanic/Latino (23%) or Chinese-descent (11%). The study design and procedures have been described.¹⁶ For the present study, after excluding participants with incomplete information on HF endpoints (n=19), urinary metals (n=85), extreme urinary metal values defined as one order of magnitude higher than the highest following value (n=4), and missing data on relevant sociodemographic and clinical covariates (n=105), a total of 6,601 participants were included and followed through December 2019.

The SHS recruited 4,549 men and women 45–74 years from 13 different American Indian tribes in North Dakota and South Dakota (Northern Plains), Oklahoma (Southern Plains), and Arizona (Southwest) between 1989–1992. One community withdrew participation, leaving 3,517 participants eligible. Cohort details are published.¹⁷ After excluding participants without available information on urinary metals (n=465), HF endpoints (n=18) and information on relevant covariates (n=117), a total of 2,917 participants, free of HF at baseline, were included and followed through December 2019

The Hortega study included 1,502 adult participants 18–85 years in the city of Valladolid, an urban area in northern Spain between 2001–2003. Procedures details are published.¹⁸ For the present analysis, after excluding participants without urinary metals available (n=20), those with prevalent HF or lost to follow-up (n=94) ,and missing relevant covariates (n=88), a total of 1,300 participants, free of HF at baseline, were included and followed through December 2021. Detailed participant flowcharts are provided in Figure S1.

Institutional (MESA and Hortega), and Tribal (SHS) Research Review Boards at all participating centers across cohorts approved the study and all participants gave informed consent.^16–18

Chronic Heart Failure events:

Details on the study follow up procedures in MESA, SHS, and Hortega have been published.^16–18

A subset participants had available information on left ventricular ejection fraction (LVEF) concurrent, and up to a few months after the HF diagnosis in MESA (n=358) and SHS (N=218), respectively. As an exploratory analysis, participants with EF information available were classified according to their LV EF as LVEF ≥ 50% (HFpEF) or LVEF < 50% (HFmrEF and HFrEF). HFmrEF and HFrEF participants were analyzed together due to their pathophysiologic similarities. A detailed description of HF events ascertainment and definition, including HF subtypes, is provided in Table S1.

Covariates:

In MESA, SHS and Hortega, self-reported sex, age, race and ethnicity, smoking status, alcohol use, education, use of lipid lowering medication, and hypertension medication use, were obtained by standardized questionnaires at baseline. Total cholesterol, HDL-cholesterol, and plasma glucose were measured in fasting blood samples (SHS and MESA), and non-fasting samples (Hortega) at baseline. Measurements of systolic blood pressure (SBP), and height and weight (to calculate BMI) were obtained during a physical examination. Measurement procedures have been previously described for all cohorts,^16–18 a detailed description of variable definitions and measurements of covariates is provided in Table S2.

Urinary metals:

In MESA, spot urine samples were analyzed at the Trace Metals Core Laboratory at Columbia University using Inductively-Coupled Plasma Mass Spectrometry with dynamic reaction cell (ICP-MS-DRC) with a PerkinElmer NexION 350S (Waltham, MA, US) blinded to participant characteristics as previously described during the baseline visit (2000–2002).¹⁹ A total of 15 trace elements were measured. In the SHS, urinary levels of 6 metals were measured in spot morning urine samples at the Trace Metals Core Laboratory at the University of Graz, Austria via ICP-MS (Agilent 7700x ICPMS, Agilent Technologies, Germany) during the baseline visit (1989–1992) as previously described.²⁰ In the Hortega study, urinary levels of 7 metals were measured in spot urine samples collected at the baseline visit (2001–2003), at the Laboratory of Environmental Chemistry and Bioanalysis of Huelva University (Spain) by an ICP-MS Thermo XSeries2 (Thermo Scientific, Germany), as previously described.^18,21All laboratory metal analyses were conducted blinded to participant characteristics.

A consistent approach across cohorts was conducted to i) correct for urine dilution: individual urinary metal concentrations were divided by urine creatinine concentrations and included as μg metal/g creatinine in the statistical analyses; ii) deal with samples below the methods’ detection limits (MDLs): these samples were replaced by the MDL divided by the square root of two, and iii) account for organic arsenic species: total urinary arsenic levels were corrected for arsenobetaine levels in MESA and the Hortega study.^19–21 In the Hortega study, arsenobetaine levels for the full sample were calculated by multiple imputation methods as previously described.²¹ The sum of sum of inorganic arsenic (arsenate and arsenite) and methylated species (monomethylarsonate, MMA, and dimethylarsinate, DMA) was used as the biomarker of arsenic exposure in the SHS.²²

A detailed description of elements measured in each cohort and MDLs is presented in Table S3.

Statistical analysis:

Descriptive statistics were used to summarize sociodemographic and clinical variables at baseline by incident HF status. To evaluate the prospective association of urinary metal levels with incident HF, we used Cox proportional hazards models in each individual cohort. Time to event was calculated as the difference between the date of the baseline examination and the date of the event, the date of death or the administrative censoring dates (December 31^st , 2019 in MESA and SHS, and December 31^st, 2021, in Hortega), whichever occurred first. Models were progressively adjusted, and a harmonized use of consistent covariates was implemented across cohorts when possible (Table S1). Model 1 was adjusted for age, sex, smoking, BMI, race (MESA only), and eGFR, as glomerular filtration can influence urine metal levels. Model 2 was further adjusted for SBP, hypertension treatment, total cholesterol, HDL-cholesterol, lipid lowering medication, and diabetes status. Some of these clinical factors may act as mediators in the association. We present both models in the results. In MESA and SHS, models were stratified by study centers, allowing the baseline hazard function to vary across them. We identified no violations of the proportional hazards model assumption after visual examination of Schoenfeld residuals. P-values were not corrected for multiple testing.

Multiple statistical approaches were used to provide a detailed assessment of the dose-response. First, individual metals were categorized into quartiles comparing the second, third and fourth quantile to the first (reference). Second, urinary metals were log transformed and analyzed as a continuous variable comparing the 75th to the 25th percentile (equivalent to one interquartile range (IQR)). This approach minimizes the impact of outliers and also captures the log-linear dose responses.^1,23 Third, we used restricted quadratic splines on log-transformed metal levels as a flexible approach to assess potential non-linearities.

We tested alternative dilution-adjustment approaches by modeling urinary metals in μg/L and adding urinary creatinine concentrations as an additional covariate to the model. We conducted sensitivity further adjusting for other variables including education, alcohol use, urinary cotinine (MESA and Hortega), and PM_2.5 levels (MESA only), which yielded similar results.

We conducted a fixed effects meta-analysis to assess the pooled hazard ratios (HR) for the metals available across the three cohorts, by combining the study-specific HR of incident HF per doubling of the baseline metal levels (metals were log2 transformed and the HR per 1 unit change in urinary metal levels was extracted), and variance-covariance matrices using restricted maximum likelihood. A fixed effects model was chosen because under a limited number of studies (up to three), it provides a more conservative estimate of the overall effect and minimizes the influence of between-study variability.²⁴

To assess the joint effect of urinary metals as a mixture, we used a machine learning agnostic approach using modified Cox proportional hazards models with an elastic-net penalty.²⁵ This approach allows the inclusion of the baseline levels of multiple urinary metals simultaneously in the model (defined as the mixture) to obtain an estimate of the overall association of the mixture on time to HF. A set of 5 metals (arsenic, cadmium, molybdenum, selenium, and zinc) measured across cohorts were included as the mixture. We estimated the 10-year survival probability difference per one IQR change in the baseline metal mixture, to provide an additional estimator with a causal interpretation²⁶ and a shorter-term clinical interpretability. Separate models were run for each cohort. Additional details are provided in Supplementary Methods.

Results:

Baseline characteristics by cohort:

At baseline, the median (IQR) age (years) was 62 (53, 70), 55 (50, 63), and 52 (37, 73) and 47%, 42% and 50% of participants were male, respectively, in MESA, SHS and Hortega. Over the study follow up (median [IQR] years 17.6 [12.5, 18.5] in MESA, 18.6 [9.28, 26.4] in SHS, 18.8 [15.0, 19.6] in Hortega), a total of 1,000 participants developed HF. Some baseline differences existed across cohorts, ever smoking rates and diabetes prevalence were highest in the SHS (71.0% and 38.8%, respectively) (Table 1).

Table 1.

Baseline characteristics of study participants by cohort and incident HF status in MESA (N=6,601), SHS (N=2,914), and Hortega (N=1,300).

	MESA		SHS		Hortega
Incident Heart Failure
	No	Yes	No	Yes	No	Yes
N	6,177	424	2,440	477	1,200	100
Age (years)a	62.0 (53.0, 70.0)	69.0 (62.0, 75.0)	54.9 (49.1, 62.1)	56.9 (51.3, 63.9)	48.6 (35.98, 71.9)	77.1 (73.7, 79.5)
Male sexb	2892 (46.7)	242 (56.9)	1047 (43)	182 (38.2)	597 (49.8)	49 (49.0)
Smoking status
Never	3,150 (50.9)	181 (42.6)	740 (30.4)	104 (21.8)	545 (45.4)	61 (61)
Former	2,234 (36.1)	190 (44.7)	789 (32.4)	184 (38.6)	355 (29.6)	36 (36)
Current	808 (13)	54 (12.7)	908 (37.3)	189 (39.6)	300 (25)	3 (3)
BMI (kg/m2)	27.5 (24.4, 31.0)	28.6 (25.5, 33.8)	29.5 (26.1, 33.5)	30.8 (27.8, 34.9)	25.9 (23.4, 28.6)	28.2 (24.6, 31.5)
SBP (mmHg)	123 (111, 139)	135 (120, 152)	124 (112, 136)	126 (116, 140)	127 (115, 142)	143 (131, 158)
Hypertension medication	2,206 (35.6)	251 (59.1)	509 (20.9)	147 (30.8)	203 (16.9)	49 (49.0)
Total Cholesterol (mg/dL)	192 (171, 215)	189 (168, 212)	193 (168, 217)	193 (172, 221)	200 (173, 223)	200 (173, 232)
HDL-Cholesterol (mg/dL)	48 (40, 59)	47 (40, 57)	44 (37, 53)	42 (35, 49)	51.2 (42.0, 61.2)	46 (38.6, 54.6)
Diabetes status
No	4,633 (74.8)	246 (57.9)	1,203 (49.4)	147 (30.8)	1121 (93.4)	73 (73.0)
Impaired Fasting Glucose (mg/dL)	849 (13.7)	63 (14.8)	376 (15.4)	57 (11.9)	-	-
Diabetes	710 (11.5)	116 (27.3)	858 (35.2)	273 (57.2)	79 (6.6)	27 (27.0)
eGFR <60	166 (2.7)	44 (10.4)	85 (3.5)	17 (3.6)	100 (8.3)	32 (32.0)
Non-essential metals
Total arsenic (μg/g)	3.10 (1.95, 5.08)	2.81 (1.78, 4.77)	8.39 (5.07, 14.19)	8.57 (5.32, 14.61)	6.54 (4.05, 11.36)	6.52 (3.89, 10.77)
Cadmium (μg/g)	0.53 (0.36, 0.80)	0.54 (0.36, 0.85)	0.95 (0.61, 1.49)	1.03 (0.67, 1.60)	0.39 (0.23, 0.65)	0.43 (0.25, 0.72)
Tungsten (μg/g)	0.06 (0.04, 0.11)	0.06 (0.04, 0.11)	0.11 (0.06, 0.23)	0.12 (0.07, 0.23)
Uranium (μg/g)	0.005 (0.000, 0.010)	0.005 (0.000, 0.010)	-	-	-	-
Essential metals
Cobalt (μg/g)	0.39 (0.28, 0.57)	0.41 (0.29, 0.58)	-	-	0.23 (0.13, 0.49)	0.21 (0.13, 0.43)
Copper (μg/g)	12.4 (10.0, 15.7)	13.8 (10.6, 18.0)	-	-	6.26 (3.85, 10.12)	7.85 (5.32, 13.1)
Molybdenum (μg/g)	40.4 (28.7, 57.8)	40.3 (29.1, 57.0)	29.1 (20.0, 41.0)	32.4 (23.2, 45.9)	25.5 (13.6, 50.9)	36.2 (19.7, 59.4)
Selenium (μg/g)	43.4 (34.2, 55.5)	44.42 (33.4, 55.8)	47.7 (36.1, 66.0)	56.99 (42.1, 77.1)	49.2 (31.7, 80.7)	46.7 (22.8, 66.1)
Zinc (μg/g)	532.8 (359.7, 804.5)	679.0 (479.0, 987.6)	544.5 (389.2, 778.0)	701.9 (485.8, 961.7)	192.98 (98.5, 358.7)	293.7 (129.9, 533.0)

^aContinuous variables are presented as median (IQR).

^cCategorical variables are presented as N (%)

Median (IQR) urinary cadmium levels (μg/g) were higher in SHS (0.95 (0.61, 1.49)) compared to MESA (0.53 (0.36, 0.80)) and Hortega (0.39 (0.23, 0.65)) (Table 1). Copper and molybdenum levels were higher in MESA than in SHS and Hortega. Zinc levels were lowest in Hortega. The correlation between metals in urine ranged from low to moderate (Figure S2).

Pooled associations between urinary metals and incident HF

Urinary arsenic, cadmium, molybdenum, selenium, and zinc were measured in the three cohorts. The pooled HR (95%CI) for incident HF per doubling of urinary metal levels in the model adjusted for sociodemographic and clinical covariates (Model 2) were significant for cadmium (1.15 (1.07, 1.24)),molybdenum (1.13 (1.05, 1.22)), and zinc (1.22 (1.14, 1.32)). Positive non-statistically significant associations were identified for arsenic (pooled HR: 1.04 (0.97, 1.12)), selenium (pooled HR: 1.10 (0.97, 1.25)). Individual cohort and pooled HRs (95%CI) are displayed in Figure 1.

Figure 1.

Meta-analysis of results per doubling of the dose (1 unit change in log2 transformed levels of urinary metals) across cohorts.

Model was adjusted by: age, sex, race/ethnicity (MESA), eGFR, smoking, BMI, SBP, hypertension treatment, total cholesterol, HDL-cholesterol, diabetes diagnosis, and use of lipid lowering, and stratified by study center in SHS and MESA.

Urinary metal mixtures and incident HF:

The elastic-net model identified a positive association for the urinary 5-metal mixture (arsenic, cadmium, molybdenum, selenium, and zinc) and incident HF, with the largest magnitude of the association found in the SHS. The fully adjusted HR (95%CI) for incident HF when comparing the 75^th versus 25^th percentiles of the 5-metal mixture levels were 1.38 (1.00, 1.86) in MESA, 1.55 (1.28, 1.97) in SHS, and 1.08 (0.85, 1.63) in Hortega (Table 2 [Model 2]). The corresponding 10-year survival probability differences (95%CI) were −0.40% (−0.78, 0.00) in MESA, −1.24% (−2.20, −0.59) in SHS, and −0.03% (−0.21, 0.06) in Hortega.

Table 2.

Hazard ratio (95%CI) and 10-year survival probability difference per one interquartile range (IQR) increase in the mixture of 5 urinary metals at baseline in MESA , SHS and Hortega: arsenic, cadmium, molybdenum, selenium, and zinc.

Urinary metal mixture
	Hazard Ratio (95%CI)		10-year survival Probability Difference, %
	Model 1a	Model 2b	Model 1a	Model 2b
Study population
MESA	1.61 (1.17, 2.20)	1.38 (1.00, 1.86)	−0.70 (−1.30, −0.23)	−0.40 (−0.78, 0.00)
SHS	1.96 (1.62, 2.56)	1.55 (1.28. 1.97)	−2.86 (−4.30, −1.88)	−1.24 (−2.20, −0.59)
Hortega	1.21 (0.92, 1.69)	1.08 (0.85, 1.63)	−0.08 (−0.27, 0.03)	−0.03 (−0.21, 0.06)

^aModel 1 was adjusted by: age, sex, race/ethnicity (MESA), eGFR, smoking, and BMI, and stratified by study center in the SHS and MESA.

^bModel 2 was further adjusted by: SBP, hypertension treatment, total cholesterol, HDL-cholesterol, diabetes diagnosis, and use of lipid lowering medication.

Individual metal levels and incident HF and HF subtypes by cohort:

The HRs (95%CI) of incident HF per IQR increase in urinary metal levels in MESA, SHS and Hortega are reported in Table 3. In fully adjusted models (Model 2), statistically significant associations with incident HF were identified for cadmium, tungsten, zinc, copper, and uranium in MESA; for cadmium, molybdenum and zinc in SHS; and for cadmium, molybdenum and copper in Hortega. The associations remained consistent across different assessment of the dose-response relationship (i.e., quartiles, as shown in Table S4). In the spline models, the dose response was largely linear for cadmium and copper, and positive at higher levels for zinc and molybdenum. A positive association was identified for arsenic at higher levels in the SHS (Figure 2).

Table 3.

Hazard ratios (95%CI) of incident HF per one IQR in baseline levels of urine metals within the MESA (n=6,601), SHS (n=2,917) and Hortega (n=1,219) cohorts.

	MESA			SHS			Hortega
	IQR (μg/g)	Model 1a	Model 2b	IQR (μg/g)	Model 1a	Model 2b	IQR (μg/g)	Model 1a	Model 2b
N cases		424	424		477	477		100	100
Arsenicc
	3.11	1.13 (0.98, 1.30)	1.08 (0.93, 1.25)	9.16	1.10 (0.94, 1.28)	1.06 (0.90, 1.25)	7.31	1.01 (0.79, 1.29)	0.97 (0.76, 1.24)
Cadmium
	0.45	1.27 (1.07, 1.50)	1.26 (1.07, 1.50)	0.88	1.09 (0.96, 1.24)	1.14 (1.01, 1.29)	0.42	1.27 (0.99, 1.63)	1.31 (1.02, 1.68)
Tungstenc
	0.07	1.22 (1.08, 1.38)	1.15 (1.01, 1.31)	0.17	1.12 (1.00, 1.26)	1.11 (0.99, 1.24)
Molybdenum
	29.2	1.09 (0.97, 1.24)	1.06 (0.94, 1.19)	21.1	1.3 (1.16, 1.46)	1.23 (1.09, 1.39)	37.9	1.25 (0.95, 1.64)	1.21 (0.92, 1.60)
Seleniumc
	21.3	1.07 (0.94, 1.23)	1.02 (0.89, 1.17)	31.1	1.34 (1.16, 1.54)	1.14 (0.98, 1.32)	125.4	0.97 (0.75, 1.26)	0.98 (0.75, 1.27)
Zincc
	454.5	1.30 (1.16, 1.46)	1.17 (1.03, 1.33)	413.1	1.87 (1.66, 2.11)	1.51 (1.31, 1.73)	267.2	1.27 (0.98, 1.66)	1.18 (0.90, 1.55)
Cobalt
	0.28	1.17 (1.04, 1.31)	1.14 (1.01, 1.27)	-	-	-	0.34	1.06 (0.85, 1.32)	1.05 (0.84, 1.30)
Copperc				-	-	-
	5.80	1.32 (1.22, 1.42)	1.25 (1.15, 1.37)	-	-	-	6.39	1.33 (1.06, 1.68)	1.27 (1.01, 1.59)
Uraniumc				-	-	-	-	-	-
	0.01	1.28 (1.09, 1.50)	1.24 (1.06, 1.45)	-	-	-	-	-	-

Hazard ratios and their confidence intervals were obtained using Cox proportional hazards models.

^aModel 1 was adjusted for: age, sex (male, female), race/ethnicity (only in MESA), eGFR, smoking (none, former, current) and BMI. The model was stratified by study center in the SHS and MESA study.

^bModel 2 was further adjusted for: systolic blood pressure, hypertension treatment (yes/no), total cholesterol, HDL-cholesterol, diabetes diagnosis (no, impaired fasting glucose, confirmed diabetes) and use of lipid lowering medication (yes/no).

^cUrinary metals were log transformed because of their right skewed distribution

Figure 2.

Hazard Ratios for incident HF by baseline urinary metal concentrations

The solid lines represent adjusted HRs based on restricted quadratic splines for the log-transformed concentration of urinary metals, knots were set at 10^th, 50^th and 90^th percentiles. The shadowed areas represent the upper and lower 95% confidence intervals. The reference was set at the 10^th percentile

^aModel 1 was adjusted by: age, sex, race/ethnicity (MESA), eGFR, smoking, and BMI, and stratified by study center in the SHS and MESA.

^bModel 2 was further adjusted by: SBP, hypertension treatment, total cholesterol, HDL-cholesterol, diabetes diagnosis, and use of lipid lowering medication.

Ejection fraction (EF) information was available in 358 MESA and 218 SHS participants with a diagnosis of HF. In the fully adjusted models (Model 2) by ejection fraction, the HR (95%CI) for incident HF EF≥50% was 1.28 (1.00, 1.64) for arsenic, 1.67 (1.28, 2.19) for cadmium, 1.29 (1.05, 1.57) for tungsten, 1.26 (1.03, 1.55) for molybdenum, 1.30 (1.06, 1.61) for selenium, 1.28 (1.08, 1.52) for cobalt, and 1.36 (1.19, 1.56) for copper in MESA, while no significant association was found for any metal in SHS. For incident HF EF<50%, the corresponding HRs were 1.20 (1.00, 1.44) for zinc in MESA, and 1.63 (1.26, 2.11) for zinc, 1.33 (1.08, 1.63) for tungsten and 1.66 (1.33, 2.07) for molybdenum in the SHS (Table S5).

Discussion

In this multi-cohort study and meta-analysis including >10,000 adults from different geographic locations and diverse races and ethnicities, we identified consistent associations between higher levels of urinary non-essential and essential metals and incident HF risk over 20 years of follow up. Higher levels of urinary cadmium, molybdenum, and zinc were consistently associated with incident HF risk across cohorts. The hypothesis driven analyses for individual metals were consistent with the results from machine learning approaches, which identified associations for the urinary 5-metal mixture with incident HF risk, as well as with a decreased 10-year survival probability when comparing higher to lower baseline urinary metal mixture levels (Central Illustration). The largest magnitude for the metal-mixture effects on incident HF risk was identified among the SHS participants, a population with a long history of high burden of exposure to contaminant metals suffering high rates of CVD.¹³

Central Illustration:

Associations Between Urinary Metal Levels and Incident Heart Failure: A Multi-Cohort Analysis

Altogether, our findings support the role of urinary metals as robust risk factors of incident HF risk among disease-free adults.

Contaminant metals such as arsenic, cadmium and lead are established CVD risk factors.^1–4 Our prospective findings in MESA, SHS and Hortega for cadmium and incident HF risk are consistent with those reported in other prospective studies assessing urinary cadmium at similar exposure levels in non-smoking adults in Denmark (HR 1.10 (1.00, 1.20) per 0.19 μg/g cadmium).²⁷, elderly adults from Australia (HR: 1.17 (1.00, 1.35) per ~2.7 increase in cadmium,²⁸ and shorter term studies in SHS²⁹ and Hortega³⁰; as well as and cross sectional assessments in the National Health and Nutrition Examination Survey (NHANES)^31,32 These findings for urinary cadmium also align with studies assessing blood cadmium and HF risk in NHANES 1999–2018 (OR: of self-reported HF 1.77 (1.34, 2.34) comparing highest to lowest blood cadmium tertiles).³³

Arsenic is an established cardiotoxicant,^2,23 yet the role of chronic exposure to low levels of arsenic on HF risk is unclear. We identified a positive non-statistically significant association for urinary arsenic and incident HF risk in MESA and Hortega, and a statistically significant association at higher levels in the SHS (Figure 2). A previous study in young adults from the SHS identified associations between urinary arsenic and left ventricular hypertrophy and impaired function, suggesting the role of arsenic in pre-clinical HF stages.⁶ The differential associations observed across cohorts may be attributable to differences in arsenic sources and levels of exposure. The main source of arsenic exposure in MESA and Hortega participants is through fish and seafood - sources of organic non-toxic arsenic species- .³⁴ However, in the SHS, arsenic exposure occurs mostly in its inorganic toxic form from drinking water or air pollution.³⁵ A comprehensive assessment of arsenic toxicity at low levels requires additional strategies for analysis,³⁶ warranting further studies assessing the relationship between arsenic and HF pathophysiology.

Essential metals such as cobalt, copper, molybdenum and zinc play a key role in maintaining body homeostasis and cellular functions.^37,38 However, their balance and retention in the cells is affected during cardiometabolic dysregulation, resulting in their release to extracellular compartments including urine, which can serve as an indicator of loss of body reserves.^9,39

We consistently identified associations between higher levels of urinary copper and zinc with incident HF risk across cohorts. Higher urinary zinc levels have been associated with diabetes risk⁴⁰ – a well characterized HF risk factor⁴¹ – in epidemiological studies. In our models, the association between urinary zinc and incident HF remained statistically significant after adjustment for diabetes, supporting both a role of zinc on HF pathophysiology through diabetes risk, but also a potential independent effect.

Though studies of urinary copper and HF are scarce, our findings are consistent with previous studies based on serum biomarkers which identified associations of serum copper and HF prevalence⁴² and left ventricular dysfunction.⁵ These epidemiological findings are supported by mechanistic evidence identifying associations between copper homeostasis and mechanisms of HF such as oxidative stress and alterations of myocardial contractility.⁴³ Evidence for other essential metals in urine, such as molybdenum, selenium and cobalt is limited but suggestive of associations with increased CVD risk.^44,45

Taken together, our prospective findings in MESA, SHS and Hortega further suggest the potential of elevated levels of zinc and copper in urine (a non-invasive biomarker) to serve as predictors of HF risk among participants free of clinical symptoms.

Metal exposure and metal dysregulation in the body can impact the cardiovascular system through different molecular mechanisms including systemic inflammation and oxidative stress, direct endothelial damage, atherosclerosis, lipid and glucose dysregulation, and epigenetic modifications.^4,14,37 These molecular pathways can contribute to the pathogenesis and progression of HF by inducing direct damage to the myocardium, but also by increasing the risk of hypertension, diabetes, and other established risk factors for HF. Yet, further studies, including formal mediation analyses are needed to disentangle these complex relationships and identify specific mechanisms. Given the heterogeneous pathophysiology of HF subtypes,¹⁵ some metals may be more relevant for different HF subtypes. While studies are very limited, previous research identified differential associations of individual metals with different HF subtypes. For example, 24-h urinary cadmium levels has been associated with reduced regional longitudinal stain rate and radial left ventricular strain, indicators of systolic dysfunction,⁷ and arsenic and uranium with LV hypertrophy.^6,8 Yet, more research in this area is critically needed.

Strengths of this study include the large sample size and the inclusion of racially and ethnically diverse participants from different geographic locations. In addition, we assessed the joint association of urine metals as a mixture, finding consistent results across individual metal and mixtures approaches. Other strengths include high-quality data collection methods and surveillance of clinical HF events over a long follow-up, the use of harmonized sociodemographic and clinical indicators across cohorts, and highly sensitive laboratory methods for the measurement of urinary metals at low concentrations. The assessment of metals in the urine is an established and commonly used biomarker in epidemiological studies. Urine is the biomarker of choice for the assessment of internal dose for several metals such as cadmium, arsenic, tungsten or uranium, as urine integrates multiple sources and routes of exposure and multiple metals are eliminated through the urine.⁹Urine is not an established biomarker for other metals such as lead or mercury, which are not evaluated in this study for that reason.⁹ The interpretation, however, can be different for the different metals, as levels in the urine are determined by sources of exposure, metabolism and accumulation in the body, and routes of excretion, influencing their half-lives, as it has been recently reviewed.⁹ The study has some potential limitations. We relied on a single spot measurement of urinary metals, which may not reflect long term exposures. While urinary metals can be constant over time for metals with long half-lives (e.g., cadmium), or for metals for which exposure in the environment is constant (e.g., arsenic in drinking water in the absence of interventions), additional studies with serial urine metal measurements are needed to evaluate the role of long-term metal exposure.⁹ Differences in baseline levels of exposure across cohorts may be attributable to different collection times, or differences in environmental regulations which are not evaluated in the manuscript. The main outcome of this study is an aggregated HF definition. EF is only available in a subset of participants and the EF collection methods were heterogeneous across cohorts. Thus, this analysis may not accurately capture the nuances of the pathophysiologic mechanisms underlying HF subtypes. Future studies evaluating HF subtypes, as well as studies that evaluate the role of metals on HF risk factors and their mediating role if HF development are needed. Our meta-analysis includes a limited number of studies, future studies should include a larger number of cohorts. However, up to date, urinary metal biomarkers are rarely available in epidemiological studies. Additional sources of confounding due to built and natural factors in the environment including water and soil contamination, consumer products, housing conditions, as well as occupation and life styles (e.g. dietary habits, alcohol consumption, education), may be present.

In summary, we identified consistent associations between individual and mixture levels of both non-essential and essential urinary metals, individually and as a mixture, with increased risk of incident HF and HF subtypes across diverse populations. These findings support the role of metals as potential risk factors for incident HF and open new perspectives for improving cardiovascular health.

Clinical Perspectives

Competency in Medical Knowledge

• We present the results from the largest study of urinary metals and heart failure up to date.

• Consistent associations were identified across cohorts for individual metals and increased risk of heart failure across multiple assessments of the dose-response relationship, as well as for all 5 priority metals when analyzed as a mixture.

• Our findings support the role of urinary metals as risk factors for heart failure and can inform novel risk prediction and preventive strategies to improve cardiovascular health by reducing metal exposures across diverse populations.

Translational Outlook

This study identifies consistent associations between higher levels of urinary non-essential and essential metals and incident HF risk over 20 years of follow across three diverse population-based cohorts in the US and Europe. Further research is needed to identify the biological mechanisms underlying theses associations. In addition, future studies should explore the potential of interventions to reduce metals exposure and dysregulation as novel preventive strategies for HF, particularly across populations with high burden of environmental exposures. Interdisciplinary collaboration among epidemiologists, toxicologists, and cardiologists will be essential to translate these findings into public health and clinical applications.

Non-standard Abbreviations and Acronyms:

EF

Ejection Fraction

HF

Heart Failure

HFmrEF

Heart Failure with mildly-reduced ejection fraction

HFpEF

Heart Failure with preserved ejection fraction

HFmrEF

Heart Failure with reduced ejection fraction

HR

Hazard Ratio

ICP-MS-DRC

Inductively Coupled Plasma Mass Spectrometry with dynamic reaction cell

IQR

Interquartile range

MESA

Multi-Ethnic Study of Atherosclerosis

SHS

Strong Heart Study

Data Availability Statement

Data utilized in this work can be made available upon request to the MESA, SHS, and Hortega cohort data coordinating centers.