Sage Journals: Discover world-class research

Abstract

Background

Chronic kidney disease is a global health issue, with environmental metal mixtures potentially contributing to its risk, although interactions between metals remain unclear. The present study aims to systematically explore the association between exposure to environmental metal mixtures and the risk of chronic kidney disease as well as their potential interactive effects.

Methods

Data from 3514 adults (aged ≥20 years) who participated in the National Health and Nutrition Examination Survey 2011–2016 were analyzed. Blood concentrations of lead, cadmium, mercury, selenium, manganese, serum copper, and zinc were measured. Chronic kidney disease was defined by an estimated glomerular filtration rate of <60 mL/min/1.73 m² or urinary albumin–creatinine ratio of ≥30 mg/g. Logistic regression, restricted cubic splines, weighted quantile sum, Bayesian kernel machine regression, and machine learning were used for data analysis.

Results

Higher concentrations of lead, cadmium, and copper were linked to an increased risk of chronic kidney disease; zinc showed protective effects. Weighted quantile sum indicated that exposure to metal mixtures was positively associated with chronic kidney disease risk (odds ratio: 1.58, 95% confidence interval: 1.30–1.94), with lead and cadmium showing the highest contribution. Bayesian kernel machine regression confirmed the cumulative/interactive effects of metals. XGBoost (area under the curve: 0.801) showed good predictive performance.

Conclusions

Environmental metal mixtures, especially lead and cadmium, increase the risk of chronic kidney disease. Metal interactions modulate renal effects, highlighting the need for exposure reduction strategies.

Keywords

Chronic kidney disease heavy metals metal mixture exposure Bayesian kernel machine regression weighted quantile sum regression machine learning National Health and Nutrition Examination Survey

Introduction

Chronic kidney disease (CKD) has become a significant global public health issue. According to the Global Burden of Disease Study, the global prevalence of CKD reached 9.1% in 2017, and the all-age mortality rate of CKD increased by 41.5% from 1990 to 2017.¹ Data from the United States show that the disability-adjusted life years related to CKD increased by 52.6% from 2002 to 2016, and the mortality rate increased by 58.3%.² Although the burden of CKD in developed countries is clear, the actual burden in developing countries may be more severe.³

The kidney, as an important metabolic organ, has various functions, such as excretion of metabolic waste, maintenance of body fluid balance, and endocrine functions. However, it is also a target organ for various metals and is susceptible to their toxic effects. Animal experiments have shown that exposure to multiple heavy metals can lead to oxidative stress, inflammatory response, DNA damage, and cell apoptosis, resulting in kidney dysfunction.⁴ Lead (Pb) exposure has adverse effects on kidney function,⁵ and even low-dose Pb exposure may be associated with a decrease in estimated glomerular filtration rate (eGFR) and the occurrence of CKD.⁶ Similarly, cadmium (Cd) in the environment can accumulate in the kidney and cause glomerular dysfunction, and it is an important risk factor for CKD.⁷ In addition, arsenic, copper (Cu), and zinc (Zn) are believed to affect kidney function. Arsenic exposure is associated with an increased risk of CKD, especially in areas with high arsenic concentrations in drinking water, as confirmed by a study in Taiwan.⁸ Cu and Zn may further affect kidney function by interacting with other metals.⁹ Selenium (Se), an essential trace element, has antioxidant and kidney-protective effects. It can bind to Cd to reduce its nephrotoxicity, but its protective effect may be affected by the synergistic effects of other metals.¹⁰

In real life, people are often exposed to multiple metals simultaneously rather than a single metal.¹¹ The impact of metal mixtures on health cannot be measured based on a single metal because of the complex interactions between metals. A study involving adults aged 40 years or older in the United States found that the concentration of blood metal mixture was positively correlated with renal function decline, among which cobalt, Cd, and Pb contributed greatly, while manganese (Mn) was negatively correlated with renal dysfunction.¹² Another study used data from the National Health and Nutrition Examination Survey (NHANES) to explore the effect of mixed heavy metal exposure on renal function in older people, and the results revealed a relationship between mixed metal exposure and proteinuria and CKD, in which uric acid played a mediating role.¹³ However, existing studies are mostly limited to specific populations with small sample sizes, and evidence regarding opposite interaction between metals is still limited. The opposite role of metals is also worth exploring; for instance, Se and Mn can offset the effect of other metals on eGFR and thus have an opposite effect on CKD.¹³

Several methods are available for metal mixture analysis. Weighted quantile sum (WQS) regression can analyze the effects of multiple metal mixtures, Bayesian kernel machine regression (BKMR) models can tackle complex nonlinear relationships, and machine learning methods excel in processing large datasets and intricate relationships. In our study, we included seven metals: Pb, Cd, mercury (Hg), Se, and Mn in the blood and Zn and Cu in the serum. They may have opposite effects on CKD. Building on classical single metal research, we also employed WQS regression, BKMR models, and interpretable machine learning to fully explore the association between metal mixtures and CKD, aiming to provide a scientific basis for risk assessment and preventive strategies related to environmental metal exposure and CKD.

Materials and methods

Study population

The NHANES is a nationally representative cross-sectional study designed to assess the health and nutrition status of United States (US) residents. It is conducted by the National Center for Health Statistics (NCHS) under the Centers for Disease Control and Prevention (CDC). The survey was approved by the NCHS Research Ethics Review Board, and all participants provided written informed consent. This study analyzed data from three cycles (2011–2012, 2013–2014, and 2015–2016). After excluding the target variable and core covariate omissions, 3514 participants over the age of 20 years were included in the final analysis (Figure S1).

Exposure assessment

The staff obtained biological samples from the participants in a mobile examination center, where the environment was controlled to ensure that the samples were collected under the same conditions. The samples were stored at −30°C and transported to the Division of Laboratory Sciences, National Center for Environmental Health, and Centers for Disease Control for analysis. Strict quality control was implemented during all detection processes. Inductively coupled plasma mass spectrometry was used for the determination of Pb, Cd, Hg, Mn, Se, Zn, and Cu. Metal concentrations below the limit of detection (LOD) were replaced with values equal to the LOD divided by the square root of two.

Outcome

CKD was evaluated according to the criteria of the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI),¹⁴ and the eGFR calculated using calibrated serum creatinine (Scr) combined with the albumin–creatinine ratio (ACR) was used to evaluate renal function. In the case of males, if the value of Scr is less than or equal to 0.9 mg/dL, the formula for calculating eGFR is as follows: eGFR = 144 × (Scr/0.9)^−0.411 × (0.993)^Age. Conversely, when Scr exceeds 0.9 mg/dL, the calculation formula is as follows: eGFR = 144 × (Scr/0.9)^−1.209 ×(0.993)^Age. In case of females, when Scr is at most 0.7 mg/dL, eGFR is calculated as follows: eGFR = 144 × (Scr/0.7)^−0.329 ×(0.993)^Age. However, if Scr is greater than 0.7 mg/dL, the formula is as follows: eGFR = 144 × (Scr/0.7)^−1.209 × (0.993)^Age. A dichotomous outcome variable was used to define CKD. Specifically, CKD was diagnosed when eGFR was less than 60 mL/min/1.73 m² or ACR was equal to or greater than 30 mg/g.

Covariates

Demographics included age, sex, ethnicity, educational level, poverty income ratio (PIR), and marital status. Ethnicity was classified as Mexican American, Non-Hispanic White, Non-Hispanic Black, and other. Educational level was classified as below high school, high school, and university and above. PIR was calculated by dividing family income by the poverty threshold specific to family size and the corresponding year;¹⁵ families living below the poverty line had a PIR of less than 1.0. Body mass index (BMI) was classified as normal or lean (≤25 kg/m²), overweight (25.1–29.9 kg/m²), and obese (≥30 kg/m²). Marital status was categorized into two groups: individuals who were married or were living with their partner were classified as married and others as unmarried. Participants who responded no to the question “Smoked at least 100 cigarettes in life” were considered nonsmokers, and all other cases were defined as smokers. Participants drinking more than 12 times per year were defined as alcohol drinkers, and others were considered nondrinkers. Participants were considered to have diabetes if they met any of the following criteria: (a) participants answered yes to any of the following questions: “Doctor told you have diabetes” or “Taking insulin now” or “Take diabetic pills to lower blood sugar”; (b) fasting glucose level ≥126 mg/dL; (c) glycated hemoglobin A1c (HbA1c) level ≥6.5%; and (d) oral glucose tolerance test (OGTT) ≥200 mg/dL. Hypertension was defined as the presence of at least one of the following conditions: (a) systolic blood pressure (SBP) ≥140 mmHg or diastolic blood pressure (DBP) ≥90 mmHg; (b) present medications for hypertension treatment; and (c) self-reported hypertension.¹⁶

Statistical analysis

All statistical analyses were conducted using R software, version 4.4.2, and p-values less than 0.05 were considered to indicate statistical significance. Continuous variables that were not normally distributed were expressed as median (interquartile range), and categorical variables were expressed as percentages. Component comparisons were performed using the Mann–Whitney U test and chi square test. Metal concentrations were natural log-transformed after unifying the units to μg/L to improve distribution. Our analysis did not incorporate survey weights from NHANES to render the results of logistic regression (LR) analyses and BKMR comparable.

Monometallic analysis

We first formed quartiles of the transformed metal concentrations, which were used to explore the association and trend between metal concentrations and CKD. Subsequently, the concentrations of the transformed metals were included as continuous variables in the model to further explore their associations with CKD. Model 1 was the crude model, and model 2 was obtained after adjusting for all variables. Then, we performed a restricted cubic spline (RCS) analysis to explore the potential nonlinear dose–response relationships.

Mixed metal exposure analysis

We then performed WQS regression to explore the associations between metal mixtures and CKD. WQS is a common method for environmental mixture analysis, wherein a single integrated score (weighted quantile sum) is obtained through a supervised framework that is included into a multivariable regression model to assess the overall effect of the mixture on the outcome.¹⁷ The model also calculates the weight of each metal to reflect the extent to which the individual metal contributes to the WQS index. We split the data into a 40% test set and 60% validation set and conducted a bootstrap 5000 times to guarantee the stability of the results.

BKMR is a statistical method for analyzing the effects of mixtures, which is particularly useful for studying the complex interactions among multiple exposure factors, and provides a flexible nonparametric approach to detect and estimate the effects of mixtures and potential interactions between them, with its core objective being to model exposure through kernel methods.^18,19 In this study, the Markov Chain Monte Carlo method was used to conduct 20,000 iterations to fit the BKMR model. The posterior inclusion probabilities ranged from 0 to 1, with a threshold of 0.5 used to determine significance, to understand the relative importance of individual metals to the outcome variable. When analyzing the mixture exposure fixed at the median (50th percentile) as a reference, the changes in cumulative risk or probability of binary outcome events were compared through gradient variations of 25th–75th percentiles. When analyzing interactions, the exposures were fixed separately at the 10th, 50th, and 90th percentiles, and differences in the effect of one exposure on the outcome when the exposure was increased by one IQR were examined through the slope changes of the bivariate dose–response function.

Machine learning

A total of nine machine learning models—LR, random forest (RF), XGBoost (XGB), support vector machine (SVM), K-nearest neighbors (KNN), glmnet (GLMNET), multilayer perceptron (MLP), Light Gradient Boosting Machine (LGBM), and naïve Bayes (NB)—were adopted to explore the association between metal mixtures and CKD. The data were divided into training set and test set in a ratio of 8:2, using 10-fold cross-validation and grid search to optimize model parameters and conducting interpretability analysis based on the optimal model.

Results

Basic characteristics

As shown in Table 1, a total of 3514 participants were included in this study, with 51.18% of them being female and 48.82% male. The proportion of females was higher in the CKD group than in the non-CKD group (55.33%, p = 0.007). The CKD population was mainly composed of older people (aged ≥60 years) (58.34%, p < 0.001), and the proportion of non-Hispanic Blacks in the CKD group (14.38%) was significantly higher than that in the non-CKD group (8.76%; p < 0.001). The proportions of individuals with obesity, smoking, diabetes, and hypertension in the CKD group were all significantly higher than those in the non-CKD group (all p < 0.001). In terms of metal exposure, the concentrations of Pb, Cd, and Cu in the CKD group were significantly higher than those in the non-CKD group (all p < 0.001), while the concentration of Zn was lower (p = 0.037). The differences in the concentrations of the remaining metals were not statistically significant. As shown in the correlation heatmap (Figure S2), the correlation coefficients among the seven metals were low, with no obvious collinearity risk.

Table 1.

The basic characteristics of the study population.

Variable	Total (n = 3514)	Non-CKD (n = 2876)	CKD (n = 638)	p
Sex, n (%)				0.007
Male	1888 (51.18)	1549 (52.22)	339 (44.67)
Female	1626 (48.82)	1327 (47.78)	299 (55.33)
Age, n (%), years				<0.001
20–39	644 (19.07)	611 (20.99)	33 (7.09)
40–59	1792 (55.13)	1598 (58.43)	194 (34.57)
≥60	1078 (25.80)	667 (20.58)	411 (58.34)
Race, n (%)				<0.001
Mexican American	451 (7.71)	378 (7.87)	73 (6.68)
Non-Hispanic White	1527 (70.59)	1247 (70.66)	280 (70.12)
Non-Hispanic Black	717 (9.54)	539 (8.76)	178 (14.38)
Other	819 (12.17)	712 (12.70)	107 (8.82)
Marital status, n (%)				0.836
Unmarried	1446 (37.35)	1165 (37.27)	281 (37.86)
Married	2068 (62.65)	1711 (62.73)	357 (62.14)
Educational level, n (%)				<0.001
Below high school	671 (12.70)	489 (11.35)	182 (21.10)
High school	770 (20.61)	606 (19.83)	164 (25.42)
University and above	2073 (66.69)	1781 (68.81)	292 (53.48)
BMI, n (%)				0.001
Normal or lean	1042 (29.5)	877 (29.8)	165 (27.6)
Obesity	1351 (37.8)	1066 (36.5)	285 (45.5)
Overweight	1121 (32.7)	933 (33.6)	188 (26.8)
Smoking status, n (%)				<0.001
Yes	1727 (47.98)	1349 (45.95)	378 (60.60)
No	1787 (52.02)	1527 (54.05)	260 (39.40)
PIR, n (%)				0.021
Not poor	2808 (86.76)	2316 (87.26)	492 (83.69)
Poor	706 (13.24)	560 (12.74)	146 (16.31)
Alcohol consumption, n (%)				0.779
Yes	72 (2.21)	55 (2.17)	17 (2.40)
No	3442 (97.79)	2821 (97.83)	621 (97.60)
Diabetes, n (%)				<0.001
Yes	644 (14.85)	407 (12.16)	237 (31.66)
No	2870 (85.15)	2469 (87.84)	401 (68.34)
Hypertension, n (%)				<0.001
Yes	1272 (32.31)	863 (28.11)	409 (58.53)
No	2242 (67.69)	2013 (71.89)	229 (41.47)
Metal concentration, M (Q₁, Q₃), μg/L
Blood lead, M (Q₁, Q₃)	9.70 (6.30, 15.20)	9.20 (6.00, 14.40)	13.30 (9.20, 20.20)	<0.001
Blood cadmium, M (Q₁, Q₃)	0.27 (0.17, 0.51)	0.26 (0.16, 0.46)	0.40 (0.23, 0.66)	<0.001
Blood mercury, M (Q₁, Q₃)	0.77 (0.42, 1.50)	0.77 (0.41, 1.48)	0.74 (0.43, 1.53)	0.948
Blood selenium, M (Q₁, Q₃)	195.23 (180.55, 209.45)	195.48 (180.71, 209.68)	193.62 (179.07, 206.92)	0.076
Blood manganese, M (Q₁, Q₃)	9.04 (7.29, 11.14)	9.08 (7.34, 11.14)	8.81 (7.13, 11.05)	0.208
Serum copper, M (Q₁, Q₃)	1120.00 (980.00, 1308.00)	1114.00 (971.00, 1294.00)	1205.00 (1036.00, 1385.00)	<0.001
Serum zinc, M (Q₁, Q₃₃)	809.00 (718.00, 906.00)	812.00 (722.00, 907.00)	794.00 (702.00, 886.00)	0.037

BMI: body mass index; CKD: chronic kidney disease; PIR: poverty income ratio.

Monometallic analysis

The relationship between metals and CKD risk was assessed using quartile analysis (Table 2) and continuous variable analysis (Table 3). In the unadjusted model, elevated concentrations of Pb, Cd, Mn (only significant in the crude model), and Cu were all associated with an increased risk of CKD, while Zn was negatively correlated. No significant risk associations were observed for Hg and Se. After adjusting for major confounders (Model 2), the effect estimates changed slightly for some metals, but the positive associations of Pb (odds ratio (OR) of the highest quartile: 1.93, 95% confidence interval (CI): 1.38–2.73, p for trend <0.001), Cd (OR: 1.67, 95% CI: 1.19–2.34, p for trend < 0.001), and Cu (only highest quartile OR: 1.41, 95% CI: 1.02–1.94, p for trend = 0.038) with CKD risk remained significant, and dose–response trends were observed. These trends persisted in the continuous variable analysis (Table 3). Zn continued to show a protective effect (highest quartile OR: 0.63, 95% CI: 0.47–0.83, p for trend < 0.001). Further description of the dose–response relationship through RCS regression (Figure S3) revealed linear positive correlation trends between Pb, Cd, and Cu and CKD risk; Zn showed a linear negative correlation trend. Notably, the dose–response relationship of Cd with CKD risk had certain nonlinear characteristics (p for nonlinear = 0.010), suggesting that the risk increase was more pronounced in the low-to-middle concentration interval.

Table 2.

The association between the quartiles of blood metal concentrations and the risk of CKD (results of logistic regression analysis).

Variable	Model	Q₁	Q₂		Q₃		Q₄
			OR (95% CI)	p value	OR (95% CI)	p value	OR (95% CI)	p value	p for trend
ln(blood lead)	Model 1	Ref	1.75 (1.30–2.37)	<0.001	2.73 (2.06–3.64)	<0.001	4.48 (3.43–5.92)	<0.001	<0.001
	Model 2	Ref	1.18 (0.85–1.65)	0.337	1.51 (1.09–2.12)	0.015	1.93 (1.38–2.73)	<0.001	<0.001
ln(blood cadmium)	Model 1	Ref	1.39 (1.05–1.85)	0.022	2.48 (1.92–3.22)	<0.001	2.67 (2.07–3.47)	<0.001	<0.001
	Model 2	Ref	1.09 (0.79–1.49)	0.603	1.50 (1.11–2.04)	0.009	1.67 (1.19–2.34)	0.003	<0.001
ln(blood mercury)	Model 1	Ref	0.99 (0.78–1.26)	0.914	1.02 (0.80–1.30)	0.873	0.97 (0.76–1.24)	0.825	0.901
	Model 2	Ref	0.93 (0.71–1.22)	0.588	0.95 (0.72–1.25)	0.717	0.93 (0.70–1.24)	0.615	0.664
ln(blood selenium)	Model 1	Ref	0.90 (0.71–1.14)	0.39	0.95 (0.74–1.20)	0.642	0.86 (0.67–1.09)	0.217	0.294
	Model 2	Ref	1.05 (0.80–1.38)	0.741	1.23 (0.94–1.63)	0.134	1.12 (0.85–1.49)	0.422	0.259
ln(blood manganese)	Model 1	Ref	0.73 (0.58–0.92)	0.009	0.70 (0.55–0.88)	0.003	0.66 (0.52–0.84)	<0.001	<0.001
	Model 2	Ref	0.81 (0.62–1.05)	0.113	0.76 (0.58–1.00)	0.052	0.78 (0.59–1.04)	0.09	0.073
ln(serum copper)	Model 1	Ref	1.29 (0.99–1.69)	0.063	1.80 (1.39–2.33)	<0.001	2.03 (1.58–2.62)	<0.001	<0.001
	Model 2	Ref	0.94 (0.70–1.27)	0.675	1.20 (0.89–1.62)	0.231	1.41 (1.02–1.94)	0.038	0.011
ln(serum zinc)	Model 1	Ref	0.92 (0.73–1.16)	0.489	0.89 (0.70–1.12)	0.315	0.72 (0.56–0.92)	0.009	0.011
	Model 2	Ref	0.86 (0.66–1.12)	0.264	0.75 (0.57–0.98)	0.033	0.63 (0.47–0.83)	0.001	<0.001

Q₁–Q₄ represent the quartile groups of the concentrations of various metals, with Q₁ serving as the reference group (Ref). Model 1 is the unadjusted model, and Model 2 is the model obtained after adjusting for age, sex, race, educational level, poverty income ratio, marital status, BMI, smoking, alcohol consumption, diabetes, and hypertension. All metal concentrations were included in the analysis after being transformed by natural logarithm.

BMI: body mass index; CKD: chronic kidney disease; CI: confidence interval; OR: odds ratio.

Table 3.

Association between blood metal concentrations and the risk of chronic kidney disease.

Variable	Model	OR (95% CI)	p value
ln(blood lead)	Model 1	2.19 (1.93–2.49)	<0.001
	Model 2	1.58 (1.35–1.86)	<0.001
ln(blood cadmium)	Model 1	1.49 (1.35–1.65)	<0.001
	Model 2	1.33 (1.16–1.52)	<0.001
ln(blood mercury)	Model 1	0.98 (0.90–1.07)	0.7
	Model 2	1.00 (0.90–1.11)	>0.9
ln(blood selenium)	Model 1	0.58 (0.29–1.15)	0.12
	Model 2	0.85 (0.40–1.79)	0.7
ln(blood manganese)	Model 1	0.57 (0.44–0.73)	<0.001
	Model 2	0.77 (0.57–1.03)	0.075
ln(serum copper)	Model 1	2.63 (1.84–3.77)	<0.001
	Model 2	2.12 (1.30–3.44)	0.002
ln(serum zinc)	Model 1	0.54 (0.34–0.86)	0.01
	Model 2	0.44 (0.26–0.74)	0.002

Model 1 is the unadjusted model, and Model 2 is the model obtained after adjusting for age, sex, race, educational level, poverty income ratio, marital status, BMI, smoking, alcohol consumption, diabetes, and hypertension. All metal concentrations were included in the analysis after being transformed by natural logarithm.

CI: confidence interval; OR: odds ratio; BMI: body mass index.

Mixed metal exposure analysis

After adjusting for covariates, WQS regression showed that exposure to the metal mixture was positively associated with CKD risk (OR: 1.58, 95% CI: 13.0–1.94, p < 0.001). After visualizing the weight of each metal (Figure 1), the weights of Pb and Cd in the WQS model were considerably higher than those of other metals and played a key role in the association between metal mixture exposure and CKD, while Mn, Se, and Zn had extremely low weights, indicating a weak effect on the risk of CKD mixed exposure.

Figure 1.

The average weight distribution of the contributions of various metals to the risk of chronic kidney disease (CKD) in the weighted quantile sum (WQS) regression model. Blood lead (Pb) and blood cadmium (Cd) have the highest weights for the risk of CKD during exposure to metal mixtures and are the main contributing factors; other metals (mercury (Hg), copper (Cu), manganese (Mn), selenium (Se), and zinc (Zn)) have lower weights and relatively small overall contributions to the risk of CKD.

The BKMR model analyzes the joint effects of metal mixtures on CKD. Table 4 shows the group posterior inclusion probabilities (groupPIP) and conditional posterior inclusion probabilities (condPIP) via BKMR, with each group main effect being similar. Mn was most prominent in condPIP (0.99958), followed by Pb and Cu, reflecting its independent role after controlling for other metals. In terms of overall exposure–response relationship, as shown in Figure 2(a), there was a positive correlation between the joint quantile increase of metal mixed exposure and CKD risk, and the mean difference and CIs of the high-quantile group also supported the cumulative adverse effects of mixed exposure on CKD. The exposure–response relationships of individual metals (Figure S4) further showed that the increase in Pb and Cd concentrations was positively associated with an increased CKD risk. Figure 2(b) quantitatively summarizes the effects of each metal on CKD risk under different quantile fixed conditions. As shown in Figure 3, Pb and Cd, Pb and Mn, Cu and Mn, and Zn and Mn may have interactive effects. Zn showed a trend of protective effects on CKD, but its point estimation was negative and the CI was wide. Mn showed a relatively stable negative impact, while Pb and Cd showed positive effects, indicating that their exposure increases CKD risk.

Table 4.

The group posterior inclusion probability (groupPIP) and conditional posterior inclusion probability (condPIP) of each metal for the risk of CKD in the BKMR model.

Variable	Group	groupPIP	condPIP
Pb	1	0.9266	0.61666
Cd	1	0.9266	0.38334
Hg	2	0.9500	0.00000
Se	2	0.9500	0.00042
Mn	2	0.9500	0.99958
Cu	3	0.9986	0.52313
Zn	3	0.9986	0.47687

The groupPIP indicates the posterior probability of the combined effect of each metal (by group) on the risk of CKD. The condPIP represents the independent posterior probability of a single metal on the risk of CKD after controlling for the influence of other metals. A larger value indicates a higher contribution of the metal to the risk of CKD.

BKMR: Bayesian kernel machine regression; CKD: chronic kidney disease; Pb: lead; Cd: cadmium; Hg: mercury; Se: selenium; Mn: manganese; Cu: copper; Zn: zinc.

Figure 2.

(a) The metal mixture is fixed at the 50th percentile as the reference, and it is gradually increased from the 25th percentile to the 75th percentile to compare the changes in probability. The results show that with the increase in the quantiles of combined heavy metal exposure, the risk of CKD shows an upward trend. (b) The estimated effects of each single metal element (Zn, Cu, Mn, Se, Hg, Cd, and Pb) on the outcome variable at different quantiles (25th, 50th, and 75th percentiles). The dots represent the effect estimates, and the horizontal lines indicate the 95% confidence intervals. The results indicate that the effects of some metal elements are affected by the changes in their quantiles. CKD: chronic kidney disease; Zn: zinc; Cu: copper; Mn: manganese; Se: selenium; Hg: mercury; Cd: cadmium; Pb: lead.

Figure 3.

Effect diagram of the BKMR model for the bivariate interaction among metal mixtures. Each small graph represents the interaction effect of two metals at different quantiles (0.1, 0.5, 0.9), with the remaining metals fixed at specific quantile points. The horizontal axis represents the exposure level of one metal, and the vertical axis represents the estimated combined effect. Curves of different colors represent changes in the quantiles of the interacting metals. The results reveal the possible nonlinear and interactive relationships among the exposures of multiple metals. BKMR: Bayesian kernel machine regression.

Machine learning

The evaluation results of all models in 10-fold cross-validation, as shown by the receiver operating characteristic (ROC) curve (Figure S5) and various evaluation indicators, indicated that all models had good classification performance. The XGBoost model showed the best performance, with an area under the curve (AUC) of 0.801 (95% CI: 0.791–0.810), accuracy of 0.838, specificity of 0.974, and sensitivity of 0.225. The AUCs for the RF, LGBM, GLMNET, and MLP models were close to 0.79–0.80, and they showed good accuracy (all above 0.81). The KNN model had the lowest AUC (0.743), while the remaining models had AUCs between 0.786 and 0.798. Therefore, we selected XGBoost for both global and local interpretations (Figure 4). Shapley Additive Explanations (SHAP) analysis (Figure 4(a)) showed that age, hypertension, diabetes, and BMI contributed the most to the prediction of CKD. A break-down plot in Figure 4(b) shows the effects of different features on individual CKD. Various metal exposure variables (Pb, Cu, and Cd) had high SHAP values in the model, indicating that they also play an important role in predicting CKD risk. Among them, the association between Cu and CKD risk may be U-shaped, while Mn was negatively correlated with CKD risk.

Figure 4.

(a) Visualization of feature importance and the direction of influence. Each row represents a variable. The horizontal axis represents the SHAP value of the variable. Different colors are used to indicate the high and low values of the variable (yellow for high values and purple for low values), showing the contribution of each variable to the model prediction and the direction of influence and (b) the waterfall chart of SHAP analysis for a single sample gradually shows the specific influence of different features (such as hypertension, age, metal exposure, and educational level) on the predicted value of the sample. Red represents a negative influence, and green represents a positive influence, jointly forming the final model prediction result. SHAP: Shapley Additive Explanations.

Discussion

Based on the NHANES data, this study systematically investigated the individual and combined effects of exposure to seven metals on CKD using various statistical models and machine learning methods. WQS regression analysis revealed a positive association between mixed metal exposure and CKD risk (OR: 1.58, 95% CI: 13.0–1.94, p < 0.001). Multiple analysis methods indicated that Pb and Cd were positively correlated with CKD risk. Furthermore, Cu, Mn, and Zn had complex relationships with CKD, while the effects of Hg and Se were weak.

This study found that environmental metal mixtures, especially Pb and Cd, increase the risk of CKD, which is consistent with the results of several previous studies. A study involving US adults aged 40 years and above showed that blood metal mixture concentrations were positively correlated with decreased renal function, with Cd and Pb contributing significantly, consistent with their primary role in metal mixtures in this study.¹² Another study that used NHANES data to explore the impact of mixed heavy metal exposure on renal function in the older population found an association of mixed heavy metal exposure with proteinuria and CKD. The results of the present study further support this association and supplement the specific roles and interactive effects of different metals.¹³

Due to the widespread presence of metals in the environment and their bioaccumulative characteristics, they pose a potential threat to human health. Their exposure increases the risk of CKD, and long-term exposure is more likely to develop into end-stage renal disease.²⁰ Our research found that the exposure concentrations of Pb and Cd were significantly positively correlated with the risk of CKD onset. Regardless of using single metal regression analysis, RCS curves, or WQS and BKMR mixed exposure models, Pb and Cd are always the main risk contributors to CKD, and they may also have a positive interactive effect. The machine learning SHAP explanation further confirmed their global and individual risk weights. Pb is known to cause nephrotoxicity with cumulative effects, and even low doses of Pb can cause renal toxicity.²¹ Its accumulation in the kidneys, especially in proximal tubular cells, can trigger oxidative stress, inflammatory responses, and vascular dysfunction, ultimately leading to tubulointerstitial fibrosis and decreased renal function.²² Consistent with the study by Yao and Xu,²³ our research found a positive association between Pb and CKD risk, with a dose–response relationship.

Cd is a widely available environmental pollutant. Similar to Pb, our research found that Cd is consistently the most significant risk contributor to CKD across various analytical methods. Cd absorption in the human body plays no useful role, and even low doses of Cd can demonstrate nephrotoxicity and carcinogenicity.²⁴ Almost all of the Cd in the body is excreted through the kidneys, and it easily accumulates in the renal cortex, with its biological half-life reaching 10–30 years.²⁵ It also targets the proximal tubules, inducing pathological changes such as tubular degeneration, edema, and fibrosis.²⁶ The concentration of Cd in the blood is positively correlated with the incidence of CKD, especially in patients with hypertension or nondiabetic patients, where this correlation is more pronounced.²⁷ Two meta-analyses^28,29 have suggested that Cd exposure is closely related to an increased risk of CKD. Our study’s dose–response analysis further examined that especially within the low to medium Cd concentration range, the risk of CKD rapidly increases with the rise in blood Cd concentrations, suggesting that even at environmentally acceptable doses, continuous low-dose exposure should not be overlooked. Notably, combined exposure to Pb and Cd may exacerbate kidney function damage.³⁰

The disruption of Cu balance, which may lead to Cu poisoning or deficiency, causes multiorgan damage.³¹ Cu plays a complex role in kidney health. It supports metabolism and exerts antioxidant effects,³² but it also causes damage when present in excess.³¹ An appropriate amount of Cu helps maintain mitochondrial function and reduces fibrosis, whereas excess Cu may lead to increased cell damage, inflammation, and fibrosis through various mechanisms.³³ The study found that individuals with higher blood concentrations of Cu had a significantly increased risk of CKD, and this association was dose-dependent. Specifically, compared with the lowest quartile of blood Cu concentrations, the odds of CKD in the highest quartile increased by 65%.³⁴ Moreover, it has been indicated that the presence of Mn may have an antagonistic effect on the renal toxicity of Cu. In our study, the association between Cu and CKD risk was U-shaped, with a high concentration of Cu exposure showing an elevated CKD risk trend, but a protective effect might be observed at medium and low concentrations. Similar to our results, Zhang et al. found a U-shaped relationship between dietary Cu intake and incident CKD, indicating that too low and too high Cu intake may increase the risk of CKD.³⁵

Our results showed that Mn displayed weak negative associations under multi-models, indicating that high concentrations may reduce the CKD risk. A study suggested that individuals with lower blood Mn concentrations were more likely to have kidney dysfunction.³⁶ On the contrary, some studies have found that Mn exposure may adversely affect children’s kidney function,³⁷ indicating that the effects of Mn may differ according to population and exposure dose. Thus, further studies on the association between Mn and CKD risk are needed. Se was believed to have antioxidant and heavy metal toxicity-alleviating effects.³⁸ A study from China showed that appropriate Se intake may have positive effects on CKD, and individuals with higher Se intake had a lower prevalence of CKD.³⁹ The Se–CKD association may be affected by other factors. A study found that individuals with higher blood Pb and Cd concentrations had higher odds of CKD, while high plasma Se concentrations could reduce this risk. This suggested that Se plays a protective role in environmental toxicant-induced CKD.³⁸ However, its independent or significant protective effect was not detected in this cohort. This may be because the Se intake is relatively adequate in this study population. Zn has antioxidant and anti-inflammatory properties and may play a role in reducing inflammation. Zn concentrations are associated with levels of matrix metalloproteinase (MMP-9) and oxidative stress markers, suggesting that Zn plays an important role in the inflammatory and fibrotic processes of CKD patients.⁴⁰ Zn presented a protective trend against CKD in this study, and high Zn concentrations may reduce CKD risk, which is related to the antioxidant and anti-inflammatory mechanism of Zn.^8,9 However, its independent effect was weakened under mixed exposure scenarios, suggesting the presence of metal–metal interactions.

We confirmed that the dominant factors under exposure of multiple metals are still Pb and Cd, using methods such as WQS and BKMR. Moreover, BKMR analysis suggested a certain interaction between Pb and Cd and Mn, Cu, and Zn. Additionally, the harmful effect of Pb and Cd can only be partially “buffered” by Zn, but it cannot be completely offset. The overall risk remains significant. This indicates that the synergistic effects of metal co-exposure should be given attention in actual environmental health management.

This study also has certain limitations. Despite being adjusted for multiple confounders, the cross-sectional design of this study limited the causal inference ability, and the link between exposure to metal mixtures and CKD requires a cohort study with long-term observation to explore causality. Metal exposure and CKD assessment were both based on single detection, which is difficult to reflect the cumulative effect of long-term exposure to metals and the dynamics of disease progression. Although our analysis excluded significant collinearity, it is undeniable that there may still be residual confounding. In addition, the data only contained information about adults in the United States, and the extrapolation of conclusions in other regions and children’s populations should be cautioned. Finally, unweighted data were used for analysis, which might have led to some sample bias and estimation error. In the future, multi-omics technologies such as metabolomics and epigenetics can be combined to conduct in-depth research on the molecular mechanisms underlying CKD induced by metal mixtures as well as to identify key pathways and biomarkers. Meanwhile, intervention studies can be designed for high-risk populations (such as occupationally exposed groups and residents in areas with excessive metal concentrations in drinking water) to evaluate the actual effect of reducing metal exposure on CKD prevention.

Supplemental Material

sj-pdf-1-imr-10.1177_03000605251378695 - Supplemental material for Evaluating the impact of environmental metal mixtures on chronic kidney disease risk: Insights from the National Health and Nutrition Examination Survey 2011–2016

Supplemental material, sj-pdf-1-imr-10.1177_03000605251378695 for Evaluating the impact of environmental metal mixtures on chronic kidney disease risk: Insights from the National Health and Nutrition Examination Survey 2011–2016 by Sen Zhang and XiaoJuan Fu in Journal of International Medical Research

Supplemental Material

sj-pdf-2-imr-10.1177_03000605251378695 - Supplemental material for Evaluating the impact of environmental metal mixtures on chronic kidney disease risk: Insights from the National Health and Nutrition Examination Survey 2011–2016

Supplemental material, sj-pdf-2-imr-10.1177_03000605251378695 for Evaluating the impact of environmental metal mixtures on chronic kidney disease risk: Insights from the National Health and Nutrition Examination Survey 2011–2016 by Sen Zhang and XiaoJuan Fu in Journal of International Medical Research

Supplemental Material

sj-pdf-3-imr-10.1177_03000605251378695 - Supplemental material for Evaluating the impact of environmental metal mixtures on chronic kidney disease risk: Insights from the National Health and Nutrition Examination Survey 2011–2016

Supplemental material, sj-pdf-3-imr-10.1177_03000605251378695 for Evaluating the impact of environmental metal mixtures on chronic kidney disease risk: Insights from the National Health and Nutrition Examination Survey 2011–2016 by Sen Zhang and XiaoJuan Fu in Journal of International Medical Research

Footnotes

Acknowledgments

Not applicable.

Author contributions

Sen Zhang: Conceptualization, Data Curation, Formal Analysis, Methodology, Software, Writing—Original Draft. Xiaojuan Fu: Conceptualization, Data Curation, Methodology, Supervision, Writing—Review and Editing.

Data availability statement

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: .

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Sen Zhang

Supplemental material

Supplemental material for this article is available online.

References

GBD Chronic Kidney Disease Collaboration. Global, regional, and national burden of chronic kidney disease, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet 2020; 395: 709–733.

Bowe

Xie

, et al. Changes in the US burden of chronic kidney disease from 2002 to 2016: an analysis of the Global Burden of Disease Study. JAMA Netw Open 2018; 1: e184412.

Luyckx

Tuttle

Garcia-Garcia

, et al. Reducing major risk factors for chronic kidney disease. Kidney Int Suppl (2011) 2017; 7: 71–87.

Singh

Bist

Choudhary

Co-exposure to multiple heavy metals and metalloid induces dose dependent modulation in antioxidative, inflammatory, DNA damage and apoptic pathways progressing to renal dysfunction in mice. Environ Toxicol Pharmacol 2024; 111: 104537.

Weaver

Lee

Ahn

, et al. Associations of lead biomarkers with renal function in Korean lead workers. Occup Environ Med 2003; 60: 551–562.

Harari

Sallsten

Christensson

, et al. Blood lead levels and decreased kidney function in a population-based cohort. Am J Kidney Dis 2018; 72: 381–389.

Smereczański

Brzóska

MM.

Current levels of environmental exposure to cadmium in industrialized countries as a risk factor for kidney damage in the general population: a comprehensive review of available data. Int J Mol Sci 2023; 24: 8413.

Hsu

Hsieh

Wang

, et al. Arsenic exposure from drinking water and the incidence of CKD in low to moderate exposed areas of Taiwan: a 14-year prospective study. Am J Kidney Dis 2017; 70: 787–797.

Meng

Kuang

, et al. Association of urinary exposure to multiple metal(loid)s with kidney function from a national cross-sectional study. Sci Total Environ 2023; 882: 163100.

10.

Chen

Zhu

, et al. Epidemiological study of kidney health in an area with high levels of soil cadmium and selenium: does selenium protect against cadmium-induced kidney injury? Sci Total Environ 2020; 698: 134106.

11.

Sexton

Adgate

Fredrickson

, et al. Using biologic markers in blood to assess exposure to multiple environmental chemicals for inner-city children 3–6 years of age. Environ Health Perspect 2006; 114: 453–459.

12.

Nan

Yang

, et al. Associations between individual and combined metal exposures in whole blood and kidney function in U.S. adults aged 40 years and older. Biol Trace Elem Res 2024; 202: 850–865.

13.

Pan

Niu

Duan

, et al. Uric acid mediates the relationship between mixed heavy metal exposure and renal function in older adult people. Front Public Health 2024; 12: 1403878.

14.

Levey

Stevens

Schmid

; CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration)et al. A new equation to estimate glomerular filtration rate. Ann Intern Med 2009; 150: 604–612.

15.

Johnson

Paulose-Ram

Ogden

, et al. National health and nutrition examination survey: analytic guidelines, 1999–2010. Vital Health Stat 2 2013; 1–24.

16.

Whelton

Carey

Aronow

, et al. 2017. ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH/ASPC/NMA/PCNA guideline for the prevention, detection, evaluation, and management of high blood pressure in adults: executive summary: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Hypertension 2018; 71: 1269–1324.

17.

Czarnota

Gennings

Wheeler

DC.

Assessment of weighted quantile sum regression for modeling chemical mixtures and cancer risk. Cancer Inform 2015; 14: 159–171.

18.

Bobb

Claus Henn

Valeri

, et al. Statistical software for analyzing the health effects of multiple concurrent exposures via Bayesian kernel machine regression. Environ Health 2018; 17: 67.

19.

Bobb

Valeri

Claus Henn

, et al. Bayesian kernel machine regression for estimating the health effects of multi-pollutant mixtures. Biostatistics 2015; 16: 493–508.

20.

Tsai

Kor

, et al. Prospective associations between environmental heavy metal exposure and renal outcomes in adults with chronic kidney disease. Nephrology (Carlton) 2018; 23: 830–836.

21.

Ekong

Jaar

Weaver

VM.

Lead-related nephrotoxicity: a review of the epidemiologic evidence. Kidney Int 2006; 70: 2074–2084.

22.

Kwon

Bae

Noh

, et al. Erythrophagocytosis of lead-exposed erythrocytes by renal tubular cells: possible role in lead-induced nephrotoxicity. Environ Health Perspect 2015; 123: 120–127.

23.

Yao

Relationships between blood concentrations of cadmium, lead, mercury, selenium, and manganese and the risk of chronic kidney disease: a cross-sectional study based on NHANES 2011–2018. Arch Med Sci 2024; 20: 1822–1830.

24.

Rani

Kumar

Lal

, et al. Cellular mechanisms of cadmium-induced toxicity: a review. Int J Environ Health Res 2014; 24: 378–399.

25.

Satarug

Moore

MR.

Adverse health effects of chronic exposure to low-level cadmium in foodstuffs and cigarette smoke. Environ Health Perspect 2004; 112: 1099–1103.

26.

Johri

Jacquillet

Unwin

Heavy metal poisoning: the effects of cadmium on the kidney. Biometals 2010; 23: 783–792.

27.

Yeon

Kang

Park

, et al. Association between blood cadmium levels and the risk of chronic kidney disease in Korea, based on the Korea National Health and Nutrition Examination Survey. Asian Biomed (Res Rev News) 2025; 19: 60–66.

28.

Doccioli

Sera

Francavilla

, et al. Association of cadmium environmental exposure with chronic kidney disease: a systematic review and meta-analysis. Sci Total Environ 2024; 906: 167165.

29.

Jalili

Kazemi

Cheng

, et al. Associations between exposure to heavy metals and the risk of chronic kidney disease: a systematic review and meta-analysis. Crit Rev Toxicol 2021; 51: 165–182.

30.

Liu

Yuan

Xiao

, et al. Associations of plasma metal concentrations with the decline in kidney function: a longitudinal study of Chinese adults. Ecotoxicol Environ Saf 2020; 189: 110006.

31.

Zhu

Zhou

Niu

, et al. COX17 restricts renal fibrosis development by maintaining mitochondrial copper homeostasis and restoring complex IV activity. Acta Pharmacol Sin 2023; 44: 2091–2102.

32.

Kang

YJ.

Copper and homocysteine in cardiovascular diseases. Pharmacol Ther 2011; 129: 321–331.

33.

Weiskirchen

Targeting copper to combat macrophage-driven inflammation: a potential advanced therapeutic strategy. Signal Transduct Target Ther 2023; 8: 339.

34.

Guo

Lin

Meng

, et al. Association of copper exposure with prevalence of chronic kidney disease in older adults. Clin Nutr 2022; 41: 2720–2728.

35.

Zhang

Gan

Xiang

, et al. U-shaped association between dietary copper intake and new-onset chronic kidney disease: a 30-year follow-up study from young adulthood to midlife. Mol Nutr Food Res 2025; 69: e202400761.

36.

Koh

Kim

Yoon

, et al. Association of blood manganese level with diabetes and renal dysfunction: a cross-sectional study of the Korean general population. BMC Endocr Disord 2014; 14: 24.

37.

Nascimento

Baierle

Göethel

, et al. Associations among environmental exposure to manganese, neuropsychological performance, oxidative damage and kidney biomarkers in children. Environ Res 2016; 147: 32–43.

38.

Wong

Chung

, et al. The association between plasma selenium and chronic kidney disease related to lead, cadmium and arsenic exposure in a Taiwanese population. J Hazard Mater 2019; 375: 224–232.

39.

Xie

Zeng

Shi

, et al. Association between selenium status and chronic kidney disease in middle-aged and older Chinese based on CHNS data. Nutrients 2022; 14: 2695.

40.

Mohtashamian

Soleimani

Gilasi

, et al. Association of zinc status with matrix metalloproteinases, advanced glycation end-products, and blood pressure in patients with chronic kidney disease. Biol Trace Elem Res 2023; 201: 4275–4285.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.50 MB

0.13 MB

0.46 MB