Urine Tungsten and Chronic Kidney Disease in Rural Colorado

Abstract

Background:

Chronic kidney disease (CKD) is a cause of global morbidity and mortality in agricultural communities. The San Luis Valley (SLV) is a rural agricultural community in southern Colorado with geographic and sociodemographic risk factors for CKD, including a water supply contaminated by heavy metals.

Methods:

We obtained pre-existing sociodemographic, clinical, and urine trace metal data for 1659 subjects from the San Luis Valley Diabetes Study, a prospective cohort study. We assessed prospective associations between urine tungsten (W) and time-to-CKD using accelerated failure time models (n=1,659). Additionally, logistic models were used to assess relationships between urine W and renal injury markers (NGAL, KIM1) using Tobit regression (n=816), as well as epidemiologically-defined CKD of unknown origin (CKDu) using multiple logistic regression (n=620).

Results:

Elevated urine W was strongly associated with decreased time-to-CKD, even after controlling for hypertension and diabetes. Depending on how CKD was defined, a doubling of urine W was associated with a 27% (95% CI 11%, 46%) to 31% (14%, 51%) higher odds of developing CKD within 5 years. The relationship between urine W and select renal injury markers was not significant, although urine NGAL was modified by diabetes status. Elevated (> 95%ile) urinary W was significantly associated with CKDu (OR 5.93, 1.83, 19.21) while adjusting for known CKD risk factors.

Conclusions:

Our data suggest that increased exposure to W is associated with decreased time-to-CKD and may be associated with CKDu. Given persistence of associations after controlling for diabetes and hypertension, W may exert a primary effect on the kidney, although this needs to be evaluated further in future studies.

Graphical Abstract

1. Introduction

The San Luis Valley (SLV) is a vast intermountain region in southern Colorado that houses the headwaters of the Rio Grande River (Figure 1). It is home to a largely bi-ethnic population of approximately 46,000 people, including approximately 5,000 agricultural workers. High rates of poverty relative to the rest of the state, in conjunction with high rates of diabetes, CKD, and depression, make the SLV a particularly vulnerable population.^1–4 This sense of vulnerability is heightened by regional water insecurity. Global climate change has increased temperatures and dramatically reduced snowpack in the western U.S., including the Rio Grande River Basin, to the point that water demand threatens to outstrip supply.^5–7

Figure 1.

Approximate location of the San Luis Valley (shaded in gray) in Colorado, USA. Graphic produced using Google Drawings (http://docs.google.com/drawings).

As water sources deplete in the SLV, there has been growing concern that heavy metals—prominent components of the rock comprising the Sangre de Cristo and San Juan mountain ranges that bracket the SLV⁸—could be concentrating in the water supply and thereby increasing risk for toxicity. Indeed, elevated levels of arsenic, uranium, cadmium, and tungsten (W) have been measured in water drawn from wells within the SLV.^9,10 Furthermore, drinking water is a predominate source of exposure to these metals; this is reflected in correlations with urinary metal concentrations.¹⁰ Specifically, unpublished data from a population of residents in the SLV (n=135) demonstrates that concentrations of metals (including tungsten) in household drinking water were highly correlated with urinary tungsten levels corrected for urine creatinine (rho=0.78). Previous studies have reported an increased risk of cardiovascular disease and diabetes related to arsenic in the water in the SLV,^9,11 and cadmium is being investigated as a driver of CKD in the region.¹² While several heavy metals have well-described toxicity in humans,^13–15 there is a paucity of literature describing the health effects of W exposure in humans. Numerous studies using data from the National Health and Nutrition Examination Survey (NHANES) have reported epidemiological associations between elevated urinary W and a host of chronic diseases, including vascular disease,^16–18 asthma,¹⁹ hypertension,²⁰ and diabetes.²¹ However, these studies are limited by cross-sectional designs. The findings presented here are part of a larger study investigating the independent associations between heavy metals with established presence in drinking water in the SLV (cadmium, arsenic, tungsten). Our hypothesis is that elevated urinary levels of tungsten are associated with the development of CKD.

Any identified relationship can also be evaluated in the context of the escalating epidemic of CKD manifest in agricultural communities elsewhere in the world.²² This disease, termed chronic kidney disease of unknown origin (CKDu) due to its uncertain etiology, disproportionately affects individuals who lack the traditional risk factors for CKD, such as diabetes and hypertension. CKDu appers to have a propensity for outdoor laborers such as agricultural workers, has reportedly killed tens of thousands of people, and may have begun as early as the 1970s.^23,24 Several competing mechanisms—including serial heat stress injury,^25,26 medication use,^27,28 agrochemicals, heavy metals,²⁹ and infectious etiologies³⁰—have been proposed as drivers of the CKDu epidemic, however none of these in isolation can explain the emergence of CKDu. Although individuals with CKDu may have increased exposure to heavy metals, to our knowledge tungsten has not been investigated as a risk factor for CKDu. We present ancillary findings investigating the association between urinary tungsten and development of CKDu.

Given the unique risk factors inherent to the SLV and the concern regarding CKD and CKDu worldwide, we sought to investigate a potential link between W and chronic kidney disease in the region. A better understanding of this relationship could inform future policies that could improve the health and resilience of the inhabitants of the region.

2. Methods

2.1. Study Population and Outcome Measures

The study population was selected from the San Luis Valley Diabetes Study (SLVDS, n= 1795). The SLVDS was a population-based prospective cohort study in a rural, bi-ethnic community in southcentral Colorado designed to determine risk factors for diabetes. In brief, subjects with diabetes were identified by review of regional hospital and medical records, or otherwise volunteered for the study after learning about it from advertisements in local newspapers, presentations, and radio shows. Residents of Conejos and Alamosa counties who were aged 20–74, carried a documented diagnosis or self-report of diabetes mellitus, spoke English or Spanish, and had capacity to consent were eligible for participation. Non-cases were selected via a two-stage sampling procedure: in the first stage, the occupants of approximately 21% of all structures in the geographic study area were interviewed by bilingual field staff, who collected basic demographic and medical history data; in the second stage, controls were randomly selected within different age, sex, ethnic group, and county strata reflecting those of Hispanic subjects with diabetes. When stratified by county, the study population was similar to that described by the 1980 census. All subjects were invited to an initial clinic visit between 1984–1988 at which demographic, behavioral, biometric, and clinical data was collected. Subjects were then followed prospectively and invited to two follow-up visits (at which similar data was collected) between 1988 and 1992, and 1997 and 1998, respectively. The study retention rate was approximately 86%. A comprehensive description of the methodology of the SLVDS methods is available.² Figure 2 illustrates the selection process from the SLVDS for our analyses (n=1690).

Figure 2.

Study subject selection scheme for NGAL/KIM1, time-to-CKD, and CKDu/AKIu analyses. Age is in units of years, eGFR mL/min/1.73 m², ACR mg/g. ^a An additional subject was missing hypertension and education status and was excluded in models that included those covariates. ^b 5-year odds of developing CKD model was limited to those who did not have CKD at time of W measurement.

2.1.1. Chronic Kidney Disease (CKD)

Subjects with an estimated glomerular filtration rate (eGFR) < 45 mL/min/1.73 m², or those with a urine albumin-to-creatinine ratio (ACR) > 300 mg/g and an eGFR between 45–60 mL/min/1.73 m², as calculated by the Modified Diet in Renal Disease (MDRD) formula,³¹ on two or more consecutive visits, were categorized as having CKD in accordance with Kidney Disease Improving Global Outcomes (KDIGO) criteria.³² Those who had been told by a physician previously that they had kidney issues and who had an eGFR < 45 mL/min/1.73 m² at enrollment were diagnosed with CKD on the initial visit. Subjects who had a single visit with an eGFR < 45 ml/min/1.73 m² with recovery of normal renal function (eGFR > 60 mL/min/1.73 m², ACR < 30 mg/g and serum creatinine < 1.3 mg/dL) in subsequent visits were labeled as having AKI. Of note, there were 33 subjects who had an eGFR < 45 mL/min/1.73 m² on their final visit and were categorized as “CKD indeterminate” due to missing follow up assessment of eGFR to determine subsequent decreased eGFR or recovery.

2.1.2. Chronic Kidney Disease of Unknown Origin (CKDu)

No CKDu criteria have been uniformly accepted for use in epidemiological studies, but most criteria reflect that CKDu arises in individuals without diabetes or hypertension, and manifests with minimal or no proteinuria. We defined CKDu as CKD (above), in a subject age ≤ 60 years at time of urine sample collection, with no concurrent or history of diabetes or hypertension (self-report), and no proteinuria (ACR < 30 mg/g). An acute form of the disease, labeled AKIu (i.e. “acute kidney disease of unknown origin”), shared the same age, comorbidity, and ACR criteria as CKDu, but was distinguished by recovery of eGFR at the subsequent visit. After restriction, there were 21 cases of CKDu, 43 cases of AKIu, and 556 non-cases available for analysis.

2.1.3. Exposure assessment and imputation of W values below limit of detection

To assess exposure to potentially nephrotoxic heavy metals, morning spot urine samples were collected from SLVDS study participants and deep-frozen at −80 ˚C in analyte vials within 24 hours of sample collection. Urine samples were analyzed for 17 different trace metals, including W, arsenic, and cadmium, at either the Colorado Department of Public Health and Environment Chemistry Laboratory or Columbia University Trace Metals Core Lab using inductively-coupled plasma mass spectrometry methods that have been described previously.¹⁰ These two labs had two different limits of detection: 0.2 µg/L and 0.0313 µg/L, respectively. However, in a small subset (n = 25) of samples analyzed at both laboratories, the correlation between values was high (rho = 0.93). In order to account for the left censoring of W values that were below the limit of detection, we imputed W values using methods described by Richardson and Ciampi.³³ To impute W values, we fit a Tobit regression model,³⁴ which accounts for left-censoring, using ln(W) as the outcome and 11 predictors hypothesized to be associated with W levels as covariates, including: age, sex, smoking status (non-smoker, former smoker, or current smoker), pack-years, natural log of urine creatinine, BMI, ethnicity (Hispanic or non-Hispanic white [NHW]), diabetes status (no diabetes, impaired glucose tolerance [IGT], or diabetes), hypertension status (hypertension yes/no), waist-to-hip ratio and education level (less than high school, high school graduate, or college graduate). After backwards selection, the final model was used to impute ln(urine W concentration [µg/L]) for individuals with W level below the limit of detection included smoking status, urine creatinine (interquartile range: 63.0, 153.0 mg/dl; 4.6% below 30 mg/dl), ethnicity, diabetes, education level, and BMI as covariates. Urine W concentrations were not corrected for the urine creatinine concentration (which was instead included as a covariate in statistical models described below) in accordance with previous studies.²¹

2.2. Statistical methods

2.2.1. Time-to-CKD analysis

Accelerated failure time models were used to test the association between W levels and time-to-CKD, accounting for interval censoring of CKD diagnosis times and controlling for important covariates (age at W measurement, sex, ethnicity [white vs. NHW], smoking status [never, former, current], education level, hip-to-waist ratio, and the natural log of urine creatinine levels), among 1,659 subjects. Time was measured as days since a subject’s first W measurement. Since study visits were spaced approximately 5 years apart, the exact timing of CKD onset is unknown. For subjects that developed CKD during the study, time-to-CKD was treated as interval-censored, with right-censoring at the time of the last healthy (non-CKD) visit and left censoring at the time of the first visit when CKD was noted. In addition, many subjects were diagnosed with CKD at study entry. These subjects were assigned a time-to-CKD of one day to reduce left truncation. Subjects that did not develop CKD during the study were treated as right-censored at their last study visit. Diabetes (normal, IGT, diabetes) and hypertension (yes/no) were evaluated as potential confounders, mediators or effect modifiers.

As a sensitivity analysis, we fit models replacing the ln(W) covariate with an indicator of whether W was less than or equal to 0.2 µg/L. As several subjects’ CKD status could not be definitively ascertained, the analysis was performed with subjects with indeterminate CKD status treated as (1) right-censored or not developing CKD during the study period; (2) as having developed CKD; and (3), excluded from the analysis altogether.

Logistic regression was used to assess the odds of developing CKD within 5 years of the baseline W measurement. This analysis was limited to subjects that did not have CKD at the time of their baseline W measurement (“incident CKD”). Backwards selection was performed to develop a final model, considering the same covariates in the time-to-CKD analyses (including diabetes and hypertension). Logistic regression analyses of incident CKD were repeated three times, due to the presence of CKD indeterminate cases, as described previously.

2.2.2. Renal injury marker analysis

The urine samples of a random sub-cohort (n = 816) of SLVDS participants were analyzed for levels of neutrophil gelatinase-associated lipocalin (NGAL) and kidney injury molecule-1 (KIM1) (n = 816). The distribution of those selected for the NGAL and KIM1 analysis was statistically similar to the SLVDS population by age, sex, and ethnicity.

We tested for associations between NGAL and KIM1 and W levels (ln W) with W treated as both a continuous variable and dichotomous variable (above or less than or equal to 0.2 µg/L, the highest limit of detection in our dataset) in the sub-cohort of 816 individuals, using Tobit regression to account for left censoring of renal injury marker levels. We assumed log normal distributions for NGAL and KIM1. Models controlled for age, sex, smoking status, natural log of urine creatinine, ethnicity, waist-to-hip ratio and education level.

In both the continuous and dichotomous W models, we tested for interaction or effect modification by diabetes and hypertension. If there was no evidence of effect modification, we evaluated diabetes and hypertension as potential confounders or mediators. We compared these models to the base model using likelihood ratio tests to determine if diabetes and/or hypertension were significantly associated with renal injury marker levels. If the likelihood ratio test was significant, we then calculated the percent changes in the coefficient for ln(W) (or dichotomous W) to determine if confounding or mediation may be present.

2.2.3. CKDu/AKIu analysis

An analysis (n = 620) using multiple logistic regression models assessed the odds of CKDu and AKIu in relation to urine W levels. CKDu and AKIu were coded as dichotomous outcomes. The base 2 log of the urine W level was the primary explanatory variable. Due to sample size restriction, backwards selection used a p-to-stay of 0.25 for the final model. Age, sex, ethnicity, smoking status, education level, hip-to-waist ratio, and the natural log of the urine creatinine were included as covariates. As a sensitivity analysis, urine W was coded as a dichotomous exposure variable (Y/N), and defined as exceeding the 95^th percentile of the restricted study population values.

Tobit regression and time-to-CKD analyses were performed in R (v. 1.1.456) using the ‘survival’, ‘lmtest’, and ‘truncnorm’ packages. CKDu/AKIu analyses were performed using SAS software, version 9.4 (SAS Institute, Cary, NC).^35,36

These analyses were approved by the Colorado Multiple Institutional Review Board (protocol #18–0696) as was the previous SLVDS (COMIRB #00–031).

3. Results

3.1. Study cohort

Descriptive statistics of the prospective time-to-CKD and CKDu sub-cohorts are presented in Table 1a. The mean age of participants in these sub-cohorts was 54.3 (SD = 12.2) and 46.1 (SD = 9.3) years, respectively. In both groups, a majority of participants were female, Hispanic, and either current or former smokers. In the time-to-CKD group, most participants did not have diabetes (63.5%) or hypertension (63.4%).

Table 1a.

Descriptive statistics of study population

	Time-to-CKD cohort					CKDu cohort
	Kidney Disease Status					Kidney Disease Status
	Total (n = 1659)	Non-case (n = 1245)	AKI (n = 111)	CKD indeterminate (n = 122)	CKD (n = 181)	Total (n = 620)	CKDu (n = 21)	AKIu (n = 43)	Non-case (n = 556)
ln(W) Mean (SD)	−1.45 (1.32)	−1.49 (1.30)	−1.44 (1.35)	−1.40 (1.39)	−1.20 (1.42)	−1.17 (1.11)	−0.69 (1.59)	−0.98 (1.27)	−1.21 (1.07)
ln(standardized W), Mean (SD)a
W Measurement Status, N (%)	−1.42 (1.33)	−1.47 (1.31)	−1.36 (1.28)	−1.35 (1.38)	−1.16 (1.47)	−1.15 (1.21)	−0.82 (1.66)	−1.03 (1.38)	−1.17 (1.17)
Above Limit of Detection	1121 (67.6%)	832 (66.8%)	76 (68.5%)	84 (68.9%)	129 (71.3%)	429 (69.2%)	15 (71.4%)	30 (69.8%)	384 (69.1%)
Below Limit of Detection	538 (32.4%)	413 (33.2%)	35 (31.5%)	38 (31.1%)	52 (28.7%)	191 (30.8%)	6 (28.6%)	13 (30.2%)	172 (30.9%)
Ln(W) for measurements > LOD	−0.830 (1.17)	−0.871 (1.15)	−0.836 (1.22)	−0.776 (1.23)	−0.594 (1.23)	−0.807 (1.14)	−0.182 (1.63)	−0.555 (1.31)	−0.852 (1.01)
Age, Mean (SD)	54.3 (12.2)	52.1 (12.1)	59.3 (9.3)	61.3 (10.4)	61.7 (9.8)	46.1 (9.28)	53.3 (5.9)	49.2 (6.4)	45.6 (9.4)
Sex, N (%)
Female	891 (53.7%)	684 (54.9%)	55 (49.5%)	62 (50.8%)	90 (49.7%)	333 (53.7%)	9 (42.9%)	19 (44.2%)	305 (54.9%)
Smoking Status, N (%)
Never-smoker	763 (46.0%)	563 (45.2%)	55 (49.5%)	58 (47.5%)	87 (48.1%)	255 (41.1%)	9 (42.9%)	17 (39.5%)	229 (41.2%)
Former smoker	506 (30.5%)	367 (29.5%)	34 (30.6%)	44 (36.1%)	61 (33.7%)	182 (29.3%)	5 (23.8%)	17 (39.5%)	160 (28.8%)
Smoker	390 (23.5%)	315 (25.3%)	22 (19.8%)	20 (16.4%)	33 (18.2%)	183 (29.5%)	7 (33.3%)	9 (20.9%)	167 (30.0%)
Ethnicity, N (%)
Hispanic	885 (53.3%)	672 (54.0%)	43 (38.7%)	61 (50.0%)	109 (60.2%)	357 (57.6%)	16 (76.2%)	30 (69.8%)	311 (55.9%)
NHWb	774 (46.7%)	573 (46.0%)	68 (61.3%)	61 (50.0%)	72 (39.8%)	263 (42.4%)	5 (23.8%)	13 (30.2%)	245 (44.1%)
Education level, N (%)
Less than high school	543 (32.7%)	376 (30.2%)	36 (32.4%)	49 (40.2%)	82 (45.3%)	113 (18.2%)	2 (9.5%)	4 (9.3%)	107 (19.2%)
High school graduate	812 (48.9%)	621 (49.9%)	56 (50.5%)	63 (51.6%)	72 (39.8%)	358 (57.7%)	12 (57.1%)	27 (62.8%)	319 (57.4%)
College graduate	304 (18.3%)	248 (19.9%)	19 (17.1%)	10 (8.2%)	27 (14.9%)	149 (24.0%)	7 (33.3%)	12 (27.9%)	130 (23.4%)
Agricultural Work
No	1437 (86.6%)	1086 (87.2%)	89 (80.2%)	105 (86.1%)	157 (86.7%)	537 (86.6%)	18 (85.7%)	37 (86.1%)	482 (86.7%)
Yes	222 (13.4%)	159 (12.8%)	22 (19.8%)	17 (13.9%)	24 (13.3%)	83 (13.4%)	3 (14.3%)	6 (14.0%)	74 (13.3%)
Outdoor Work
No	1351 (81.4%)	1019 (81.8%)	87 (78.4%)	98 (80.3%)	147 (81.2%)	500 (80.7%)	18 (85.7%)	35 (81.4%)	447 (80.4%)
Yes	308 (18.6%)	226 (18.2%)	24 (21.6%)	24 (19.7%)	34 (18.8%)	120 (19.4%)	3 (14.3%)	8 (18.6%)	109 (19.6%)
Ln(urine creatinine), Mean (SD)	4.57 (.63)	4.58 (0.63)	4.53 (0.67)	4.55 (0.56)	4.56 (0.61)	4.51 (0.98)	4.74 (0.65)	4.66 (0.74)	4.49 (1.01)
Waist-to-hip ratio, Mean (SD)	0.937 (0.077)	0.935 (0.078)	0.937 (0.076)	0.945 (0.073)	0.943 (0.071)	0.94 (0.08)	0.94 (0.08)	0.95 (0.07)	0.93 (0.08)
Diabetes status, N (%)
No diabetes	1053 (63.5%)	847 (68.0%)	69 (62.2%)	59 (48.4%)	78 (43.1%)	620 (100%)	21 (100%)	43 (100%)	556 (100%)
IGT	189 (11.4%)	135 (10.8%)	20 (18.0%)	12 (9.8%)	22 (12.2%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)
Diabetes	417 (25.1%)	263 (21.1%)	22 (19.8%)	51 (41.8%)	81 (44.8%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)
Hypertension status, N (%)
Hypertension	606 (36.5%)	402 (32.3%)	40 (36.0%)	63 (51.6%)	101 (55.8%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)
No hypertension	1052 (63.4%)	843 (67.7%)	71 (64.0%)	58 (47.5%)	80 (44.2%)	620 (100%)	21 (100%)	43 (100%)	556 (100%)
Missing	1 (0.1%)	0 (0%)	0 (0%)	1 (0.8%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)

^astandardized for urine creatinine

^bNon-Hispanic white

For the time-to-CKD analysis, 31 subjects were excluded due to missing covariates, leaving 1,659 subjects for analysis. For the odds of incident CKD analyses, when CKD was defined as only confirmed CKD cases, there were 957 controls and 61 incident confirmed CKD cases available for analysis; excluding CKD indeterminate subjects, the number of controls was reduced to 912. When CKD was defined as either confirmed or indeterminate CKD, there were 956 controls and 94 incident cases available for analysis. In the CKDu analysis, the final sample size was 620 after excluding individuals with diabetes, hypertension, age > 60, and an ACR ≥ 30 mg/g. In the renal injury marker analysis, the final sample size was 816 after excluding subjects with missing W, NGAL, and KIM1 measurements, or missing education and hypertension status. The composition of this sub-cohort is presented in Table 1b, and was similar to that of the time-to-CKD and CKDu sub-cohorts.

Table 1b.

Descriptive statistics of renal biomarker cohort.

	NGAL/KIM1 cohort
	Diabetes Status				Hypertension
	Total (n = 816)	No diabetes (n = 512)	IGT (n = 106)	Diabetes (n = 198)	Yes (n = 303)	No (n = 513)
ln(W) Mean (SD)	−1.35 (1.31)	−1.29 (1.33)	−1.35 (1.18)	−1.50 (1.31)	−1.37 (1.31)	−1.34 (1.31)
ln(standardized W), Mean (SD)a	−1.28 (1.30)	−1.30 (1.32)	−1.19 (1.29)	−1.25 (1.26)	−1.28 (1.27)	−1.27 (1.32)
W Measurement Status, N (%)
Tungsten > 0.2	587 (71.9%)	359 (70.1%)	81 (76.4%)	147 (74.2%)	219 (72.3%)	368 (71.7%)
Tungsten < 0.2	229 (28.1%)	153 (29.9%)	25 (23.6%)	51 (25.8%)	84 (27.7%)	145 (28.3%)
ln(W) for measurements > LOD	−0.81 (1.15)	−0.69 (1.13)	−0.96 (1.07)	−1.04 (1.21)	−0.80 (1.16)	−0.84 (1.14)
Ln(NGAL) Mean (SD)	1.23 (1.50)	0.86 (1.33)	1.05 (1.45)	2.29 (1.47)	1.41 (1.57)	1.13 (1.46)
Ln(KIM1) Mean (SD)	−1.02 (0.83)	−1.02 (0.83)	−1.15 (0.74)	−0.94 (0.89)	−0.92 (0.86)	−1.08 (0.81)
Age, Mean (SD)	55.5 (11.5)	53.5 (11.5)	58.2 (10.4)	59.2 (10.7)	58.6 (10.5)	53.7 (11.6)
Sex, N (%)
Female	421 (51.6%)	250 (48.8%)	54 (50.9%)	117 (59.1%)	157 (51.8%)	264 (51.5%)
Male	395 (48.4%)	262 (51.2%)	52 (49.1%)	81 (40.9%)	146 (48.2%)	249 (48.5%)
Smoking Status, N (%)
Never-smoker	286 (35.0%)	170 (33.2%)	33 (31.1%)	83 (41.9%)	123 (40.6%)	163 (31.8%)
Former smoker	292 (35.8%)	180 (35.2%)	45 (42.5%)	67 (33.8%)	120 (39.6%)	172 (33.5%)
Smoker	238 (29.2%)	162 (31.6%)	28 (26.4%)	48 (24.2%)	60 (19.8%)	178 (34.7%)
Ethnicity, N (%)
Hispanic	441 (54.0%)	237 (46.3%)	63 (59.4%)	141 (71.2%)	172 (56.8%)	269 (52.4%)
NHWb	375 (46.0%)	275 (53.7%)	43 (40.6%)	57 (28.8%)	131 (43.2%)	244 (47.6%)
Education level, N (%)
Less than high school	296 (36.3%)	140 (27.3%)	47 (44.3%)	109 (55.1%)	125 (41.3%)	171 (33.3%)
High school graduate	377 (46.2%)	269 (52.5%)	42 (39.6%)	66 (33.3%)	129 (42.6%)	248 (48.3%)
College graduate	143 (17.5%)	103 (20.1%)	17 (16.0%)	23 (11.6%)	49 (16.2%)	94 (18.3%)
Outdoor Work
No	669 (82.0%)	411 (80.3%)	82 (77.4%)	176 (88.9%)	412 (80.3%)	257 (84.8%)
Yes	147 (18.0%)	101 (19.7%)	24 (22.6%)	22 (11.1%)	101 (19.7%)	46 (15.2%)
Agricultural Work
No	709 (86.9%)	446 (87.1%)	87 (82.1%)	176 (88.9%)	442 (86.2%)		267 (88.1%)
Yes	107 (13.1%)	66 (12.9%)	19 (17.9%)	22 (11.1%)	71 (13.8%)		36 (11.9%)
Ln(urine creatinine), Mean (SD)	4.53 (0.59)	4.62 (0.60)	4.44 (0.65)	4.36 (0.50)	4.52 (0.58)		4.54 (0.60)
Waist-to-hip ratio, Mean (SD)	0.939 (0.074)	0.939 (0.076)	0.943 (0.070)	0.939 (0.071)	0.942 (0.070)		0.938 (0.076)
Kidney disease status, N (%)
Non-case	578 (70.8%)	393 (76.8%)	71 (67.0%)	114 (57.6%)	191 (63.0%)		387 (75.4%)
AKI	69 (8.5%)	42 (8.2%)	14 (13.2%)	13 (6.6%)	26 (8.6%)		43 (8.4%)
CKD Indeterminate	80 (9.8%)	36 (7.0%)	10 (9.4%)	34 (17.2%)	40 (13.2%)		40 (7.8%)
CKD	89 (10.9%)	41 (8.0%)	11 (10.4%)	37 (18.7%)	46 (15.2%)		43 (8.4%)
Diabetes status, N (%)
No diabetes	512 (62.7%)	512 (100%)	0 (0%)	0 (0%)	138 (45.5%)		374 (72.9%)
IGT	106 (13.0%)	0 (0%)	106 (100%)	0 (0%)	51 (16.8%)		55 (10.7%)
Diabetes	198 (24.3%)	0 (0%)	0 (0%)	198 (100%)	114 (37.6%)		84 (16.4%)
Hypertension status, N (%)
Yes	303 (37.1%)	138 (27.0%)	51 (48.1%)	114 (57.6%)	303 (100%)		0 (0%)
No	513 (62.9%)	374 (73.0%)	55 (51.9%)	84 (42.4%)	0 (0%)		513 (100%)

^astandardized for urine creatinine

^bNon-Hispanic white

3.2. Time-to-CKD

The results of the time-to-CKD analysis are summarized in Table 2. In the unadjusted models, the association between W and time-to-CKD was statistically significant, regardless of how CKD indeterminate subjects were grouped in the analysis. When indeterminate subjects were treated as right-censored (not developing CKD during the study period), time-to-CKD decreased by 62.5% (95% CI 42.0%, 75.7% decrease) for every doubling of the urine W level. Similar results were found when CKD indeterminate subjects were excluded from the analysis. When indeterminate subjects were considered as CKD cases, time to indeterminate or confirmed CKD decreased by 48.9% (31.1%, 62.0% decrease) for every doubling of the W level. We did not find evidence that the effect of W on time-to-CKD differed by diabetes or hypertension status. In separate models controlling for diabetes and hypertension status, W levels remained significantly associated with time-to-CKD. The magnitudes of the associations were similar to those found in the base models.

Table 2.

Estimated Percent Decrease in Time-to-CKD for a Doubling of W Levels

	Event = Confirmed CKD			Event = Confirmed or CKD Indeterminate			Excluding CKD Indeterminate
Model	Est. (%)	95% CI	p-value	Est. (%)	95% CI	p-value	Est. (%)	95% CI	p-value
Basea	62.5	42.0, 75.7	<0.001	48.9	31.1, 62.0	<0.001	64.9	45.3, 77.5	<0.001
Base + Diabetes	60.5	39.7, 74.1	<0.001	46.5	28.7, 59.8	<0.001	63.6	44.0, 76.3	<0.001
Base + Hypertension	62.1	41.9, 75.4	<0.001	48.7	31.2, 61.7	<0.001	64.6	45.1, 76.2	<0.001
Base + Diabetes + Hypertension	60.6	40.0, 74.1	<0.001	46.7	29.1, 59.9	<0.001	63.5	44.1, 76.2	<0.001

^abase model controlled for age, sex, smoking status, waist-to-hip ratio, education, ethnicity, and natural log of the urine creatinine

These associations persisted in sensitivity analyses using dichotomized W levels. Subjects with W>0.2 had 21.7 times decreased time-to-CKD compared to subjects with W≤0.2 (p < .0001, 95% CI 4.0, 118.6). This association was significant when controlling for hypertension and diabetes, although the magnitude of the association was somewhat reduced (17.8 times, p < .0001, 3.5, 91.7). Similar results were found when CKD indeterminate subjects were excluded from the analyses. When CKD indeterminate subjects were classified as CKD cases, subjects with W>0.2 had an estimated 9.1 decreased time-to-CKD compared to subjects with W≤0.2 (p < 0.0001, 2.9, 29.0). This association also persisted after controlling for diabetes and hypertension (7.5 times, p < 0.0001, 2.5, 22.7).

In sensitivity analyses excluding participants with urine creatinine levels outside of the normal range (<20 mg/dl or >320 mg/dl for men or > 275 mg/dl for women, ~2.2% excluded), there was no significant change in the association. We also fit additional models controlling for agricultural and outdoor work, and results remained unchanged (Supplementary Table 1). Furthermore, to evaluate the possibility that W could be co-eluting with the known nephrotoxic heavy metals cadmium and arsenic, additional sensitivity analyses were performed in which urine arsenic and cadmium were forced into the time-to-CKD model. Results were unchanged, and are available in Supplementary Table 4.

3.3. Odds of CKD Diagnosis within 5 years of W measurement

The final logistic regression model for the analysis treating indeterminate CKD status as right-censored/not developing CKD as well as excluding CKD indeterminate status subjects controlled for age, ethnicity, diabetes, hypertension and ln(urine creatinine). When CKD indeterminate subjects were treated as right-censored/not having developed CKD, for each doubling of W levels, the odds of developing CKD in 5 years increased by 1.27 times (95% CI, 1.11, 1.46, p =0.001). The analysis treating indeterminate as having developed CKD controlled for age, waist-to-hip ratio, diabetes, hypertension and ln(urine creatinine). When classifying CKD indeterminate subjects as having developed CKD, there was a significant association between a two-fold increase in urine W and the 5-year odds of developing CKD (OR 1.18, 95% CI 1.05, 1.32, p = .006). Similar results were found when indeterminate subjects were excluded from the analysis, with odds of developing CKD increasing by 1.31 times for a doubling of W levels (95% CI 1.14, 1.51, p = 0.0003).

These associations persisted in sensitivity analyses dichotomizing W (≤ than and > than 0.2 µg/L); the final model controlled for age, diabetes status, hypertension, and urine creatinine. In the analysis that treated CKD indeterminate subjects as right-censored, subjects with W levels over 0.2 had 1.84 times higher odds of developing CKD within 5 years of their W measurement compared to subjects with W levels less than 0.2 µg/L (95% CI 1.05, 3.25, p = 0.035). Similar results were found excluding indeterminate subjects from the analysis, with subjects with W levels over 0.2 µg/L having 1.99 times higher odds of developing CKD compared to those with W levels less than 0.2 (1.11, 3.55, p = 0.020). When classifying CKD indeterminate subjects as having developed CKD, odds of CKD were 1.73 times higher for subjects with W > 0.2 µg/L compared to subjects with W ≤ 0.2 µg/L (1.09, 2.76, p = .020). In a sensitivity analysis, participants with urine creatinine levels outside of the normal range were excluded (~2.2% excluded) and there was no significant change in the association. We also fit additional models controlling for agricultural and outdoor work, and results remained unchanged (Supplementary Table 2). Furthermore, to evaluate the possibility that W could be co-eluting with cadmium and arsenic, additional sensitivity analyses were performed in which urine arsenic and cadmium were forced into the 5-year odds of CKD model. Results were unchanged, and are available in Supplementary Table 5.

3.4. Urine NGAL, KIM1, and W

Demographic characteristics of the renal biomarker cohort are presented in Table 1b. In both the NGAL and KIM1 base models, ln(W) (denoted W) was not significantly associated with either ln(NGAL) (denoted NGAL) or ln(KIM1) (denoted KIM1). In a sensitivity analysis of the base model using dichotomized W levels (urine W greater than 0.2 µg/L [denoted W>0.2]), urine W was not significantly associated with NGAL or KIM1. Since urine NGAL and KIM1 can be elevated in individuals with early kidney injury secondary to diabetes and hypertension,^37–39 these comorbidities were investigated as effect modifiers and confounders. The results of these analyses are presented in Tables 3a and 3b. In a sensitivity analysis, models controlling for agricultural and outdoor work were run, and results remained unchanged (Supplementary Tables 3a and 3b). As in the time-to-CKD analyses, co-existence of arsenic and cadmium was considered as a potential confounder in the renal biomarker analysis. Urine arsenic and cadmium were forced into an additional sensitivity analysis; the finding of effect modification by diabetes status on the association between urine W and NGAL persisted in these models, which are shown in Supplementary Tables 6a and 6b.

Table 3a.

Effect of W on Urine NGAL by Comorbidity^a

Continuous model	NGAL increase per doubling urine W (-fold)	95% CI	p-value
Base Model	1.04	0.97, 1.11	0.28
Considering Effect Modification by Diabetes Status			p=0.031*
No diabetes	1.08	1.00, 1.16	0.06
IGT	0.87	0.73, 1.04	0.13
Diabetes	0.95	0.84, 1.06	0.34
Controlling for Hypertension Status	1.04	0.97, 1.11	0.30
Dichotomous model	NGAL increase for W>0.2 compared to W≤0.2 (-fold)	95% CI	p-value
Base Model	1.21	0.95, 1.54	0.13
Considering Effect Modification by Diabetes status			p=0.033 *
No diabetes	1.45	1.09, 1.93	0.01
IGT	0.70	0.38, 1.29	0.25
Diabetes	0.87	0.56, 1.35	0.54
Controlling for Hypertension Status	1.20	0.94, 1.75	0.14

^acontrolling for age, sex, smoking status, natural log of urine creatinine, ethnicity, waist-to-hip ratio, and education level

^*p-value from partial F test for interaction between DM status and ln(W)

Table 3b.

Effect of W on Urine KIM1 by Comorbidity^a,b

Continuous models	KIM1 increase per doubling urine W (-fold)	95% CI	p-value
Base Model	1.01	0.97, 1.05	0.60
Controlling for Diabetes Status	1.01	0.97, 1.04	0.75
Controlling for Hypertension Status	1.01	0.97, 1.05	0.65
Dichotomous models	NGAL increase for W>0.2 compared to W≤0.2 (-fold)	95% CI	p-value
Base model	1.02	0.89, 1.16	0.80
Controlling for Diabetes Status	1.006	0.81, 1.15	0.93
Controlling for Hypertension Status	1.01	0.89, 1.16	0.85

^acontrolling for age, sex, smoking status, natural log of urine creatinine, ethnicity, waist-to-hip ratio, and education level

^beffect modification in KIM1 analysis was not significant in any model

3.5. CKDu, AKIu, and W exposure

In the crude continuous W models, a doubling of the urine W was associated with increased odds of CKDu (OR 1.28, p = 0.042, 95% CI 1.01, 1.63). After backwards selection the final logistic regression model controlled for age, education level, and ln(urine creatinine), the effect of urine W on the odds of CKDu was no longer significant (OR 1.16, p = 0.27, 95% CI 0.90, 1.49). A one-year increase in age was significantly associated with an increased odds of CKDu (OR 1.14, p = 0.0004, 95% CI 1.06, 1.22), although this was expected given that renal function (captured by eGFR) tends to decrease with age.³¹

We did not find a statistically significant association between urine W and AKIu. In the univariate model, the odds of AKIu increased 1.12 times for every doubling of the urine W level (p = 0.20, 95% CI 0.94, 1.35). The final AKIu regression model included education level, waist-to-hip ratio, age, and ln(urine creatinine) as covariates. As in the CKDu model, age was significantly associated with increased odds of AKIu (OR 1.06 for every one-year increase in age, p = 0.006, 95% CI 1.02, 1.10).

After dichotomizing W (> or ≤ than the 95^th percentile W of the cohort), there were 31 subjects with urine W levels > the 95^th percentile, five of whom had CKDu and four of whom had AKIu. Exceeding the percentile urine W level was significantly associated with CKDu (OR 7.59, p = 0.0003, 95% CI 2.55, 22.58) but not AKIu (OR 2.48, p = 0.12, 95% CI 0.79, 7.81) in the crude model. The effect of exceeding the percentile urine W value on the odds of CKDu remained significant after adjustment for age, education level, and the ln(urine creatinine): the odds of CKDu were 5.93 times higher in subjects with urine W levels above the population percentile (p = 0.003, 95% CI 1.83, 19.21). Age was also significant in both the dichotomous CKDu (OR 1.14 per one-year increase in age, p = 0.0005, 95% CI 1.06, 1.22) and AKIu (OR 1.06 per one-year increase, p = 0.005, 95% CI 1.02, 1.10) models. In a sensitivity analysis including subjects over the age of 60 years (n=844), associations of W with CKDu did not change significantly.

4. Discussion

To our knowledge, this is first study of the association of renal injury biomarkers and urine W in Colorado, and among the early studies of CKDu in the U.S. In our study population, elevated urine W was strongly associated with decreased time-to-CKD, even after controlling for hypertension and diabetes, among other variables. This finding persisted after excluding individuals with CKD at baseline, which likely contributed to our large effect size: regardless of how CKD indeterminate subjects were grouped in our analysis, a doubling of urine W was associated with a 27 percent to 31 percent higher odds of developing CKD within 5 years. In addition, elevated urinary W was significantly associated with CKDu, but not AKIu, in dichotomous 95^th percentile models. While there was not a significant relationship between urine W with urine NGAL or KIM1 overall, we found that the effect of urine W on urine NGAL was modified by diabetes status, with non-diabetic subjects having a stronger and significant association between CKD and W. These findings persist even when accounting for agricultural work, outdoor work, and the known nephrotoxic heavy metals arsenic and cadmium in sensitivity analyses. We hypothesize that W acts in the interstitial areas and/or tubules, as we did not note any significant associations between urine W and proteinuria (a hallmark of glomerular disease). This appears plausible given that other heavy metals, such as cadmium and lead, have been demonstrated to cause tubular injury and interstitial nephritis, respectively.⁴⁰

Our findings of an association between urine W and accelerated time-to-CKD, as well as an association between high (95^th percentile) urine W and epidemiologically-defined CKDu, raise the question of whether the presence of W in the drinking water of the SLV could be leading to kidney injury. If true, this would have important implications for the broader CKD epidemic in agricultural communities around the world, for which a nephrotoxin has been postulated as a potential driver of the CKDu epidemic. To our knowledge, none of the CKDu studies to date have reported an association with W. This may in part reflect the extremely limited data on the human health effects of W exposure. Most of the studies on the health effects of W have been epidemiological, and have reported associations between elevated urinary W and prevalent diabetes, peripheral artery disease, asthma, hypertension, and survey-based history of cardiovascular and cerebrovascular disease.^16–21 There are even fewer studies of W and renal disease; an analysis of urine metals in the NHANES found that urinary W increased linearly with increasing eGFR,⁴¹ allaying concerns that the findings in our study could be due to accumulation (i.e. that decreased renal function could lead to increased retention of W, and thereby create a spurious association). In addition, we considered that the association between urine W and CKD could be occurring via diabetes or hypertension. Elevated urine W has been associated with type 2 diabetes in epidemiological studies, and it is conceivable, given that W has been linked to endothelial dysfunction, that it could cause glomerulosclerosis (an effect challenging to isolate from hypertension given that it can cause and be caused by high blood pressure).^16,17,41 However, our CKDu analysis (which excluded individuals with hypertension and diabetes) still found an association between W and CKDu, and we found no evidence of effect modification or confounding by diabetes or hypertension in our time-to-CKD analyses. Our data are consistent with W exerting a primary effect on the kidney, although this needs to be evaluated further. Lastly, the prevalence of kidney disease in our study population did not differ by sex. This sets our study apart from previous literature on CKDu, which has shown a predilection for males. However, since we suspect that tungsten exposure occurs from heavy metal contamination of drinking water (and therefore not a strictly occupational exposure per se), we would not expect major differences in disease prevalence based on sex alone.

The present study is a first step, but more research is needed to determine the association between W and CKD seen in the SLV. One avenue of interest is to establish the presence of W in kidney tissue in situ. Methods for detecting and quantifying W in tissue and bodily fluids have been described.⁴² Thus, assessing the presence of W in kidney tissue using mass spectrometry or other methods is an exciting and logical next step. Furthermore, the SLVDS database contains extensive data on cause of death and medical comorbidities—including cardiovascular disease, diabetes, and peripheral arterial disease; future analyses should assess the relationship between urinary W and these outcomes in this rural agricultural population. Our findings also underscore the need for further studies on the health effects of W in animal models, which are needed to develop a more comprehensive toxicological profile of W. As mentioned previously, we suspect that exposure to W in the SLV occurs primarily through consumption of contaminated water. This is supported by the high heavy metal content in the conglomerate and volcanic rock comprising the nearby mountain ranges,⁸ previous reports of elevated heavy metal concentrations in the water supply,⁴³ including in this cohort where water data from previous research demonstrated in an ancillary analysis for this paper, elevated levels of W in household water supply of CKDu subjects when compared to non-CKD subjects.¹⁰ Better understanding of the health effects of W are necessary to inform W drinking water standards, which currently do not exist in the U.S.⁴⁴

This study had limitations. Identification of CKD cases was limited by the design of the SLVDS, which only allotted subjects a maximum of three follow-up visits after the initial recruiting visit. Subjects who presented with an abnormal eGFR on the final visit could not be followed further to determine if they developed CKD, and were thus labeled “CKD indeterminate”. Fortunately, we were able to account for these subjects and determined that their inclusion and exclusion from the analysis did not change our conclusions. In the definition of AKI, a limitation is that only one follow up visit was used to assess recovery when ideally two follow up eGFR values would be most clinically relevant. One visit was used to define AKI due to availability of data, it is expected that misclassification bias potentially introduced by this would be non-differential and would bias the estimate towards the null. In the CKDu and AKIu analysis, we were only able to assess the association with W. Future studies should investigate these relationships prospectively given the small sample size in this pilot ancillary study. Some of the CKDu sub-analyses produced small sample sizes (e.g. n = 6 for W ≥95^th percentile in CKDu subjects). This is partly due to the rather restrictive definition of CKDu used in an effort to reduce misclassification, and partly due to our exclusion of CKD indeterminate subjects. Additionally, in the CKDu analysis, we considered other risk factors including being an agriculture worker, agrochemicals in water, and heat and humidity. The sample size for agriculture workers was too small to evaluate (n=5), heat and humidity ranges did not significantly vary during the follow-up period, and in other research, detectable levels of agrochemicals in drinking water were found in only 18% of wells and did not exceed EPA limits.⁴⁵ In concert, these limitations underscore that CKDu in the SLV warrants further research in a longitudinal study in an occupational cohort.

Such limitations notwithstanding, the findings of this study support the need for further investigation of environmental exposure to W as a risk factor of CKD, and as a potential contributor to the risk of CKDu. If kidney health in agricultural communities is indeed influenced by a nephrotoxin, it is imperative that we identify it in order to improve the livelihoods of workers and their communities at home and abroad.

Highlights:

• Elevated urine W strongly associated with decreased time-to CKD.

• Doubling of urine W associated with 27–31% higher odds of CKD within 5 years.

• No significant relationship between urine W and either urine NGAL or KIM1.

• Extreme (95%ile) urine W associated with epidemiologically-defined CKDu.

• Tungsten may exert a primary effect on the kidney.