Introduction

Cognitive dysfunction in the elderly population is becoming a significant global health concern due to its negative impact on quality of life and social burden1. The symptoms of cognitive dysfunction typically include sensory impairment, learning and memory decline, and reduced attention and judgment capabilities2. These mild cognitive symptoms can degenerate over time and increase the risk of developing Alzheimer’s disease and mortality. As there is currently no effective treatment for cognitive dysfunction, it is imperative to identify its risk factors to delay its development and progression. Several demographic, disease, and environmental factors have been linked with the development of cognitive dysfunction3. However, recent evidence suggests that exposure to chemicals particularly heavy metals can increase the risk of developing cognitive dysfunction4. The levels of heavy metals within the environment have increased dramatically during the last few years due to mining, increased production of electronics, improper waste management, and excessive use of fertilizers5,6. As heavy metal pollution has become one of the major environmental problems worldwide there is a need to identify the impact of heavy metal exposure on human health, particularly cognitive function.

The United States Environmental Protection Agency (EPA) identified cadmium (Cd), chromium (Cr), arsenic (As), lead (Pb), copper (Cu), zinc (Zn), and nickel (Ni) as major pollutants7. These heavy metals can contaminate the air, water, and food chain8. Prolonged or excessive exposure to heavy metals has been linked to the development of several systemic diseases including Minamata disease, stroke, and Parkinson’s disease9. Long-term exposure to heavy metals has also been linked with worse performance in standardized cognitive function tests10,11,12. For example, As, Cd, Mn, Pb mixed exposure, and Tl for prenatal exposure were negatively associated with the child cognitive composite score13. Early exposure to airborne Pb in childhood is associated with reduced cognitive function14. Sun et al. discovered that exposure to the mixture of multiple metals were associated with detrimental effects on cognitive flexibility in children15. High blood levels of As, Cd, manganese (Mn), and Pb levels were also linked to poor cognitive function in Bangladeshi adolescents16. Exposure to Cd was associated with impaired cognitive function in adults and increased risk of developing behavioral, physiological, and molecular abnormalities in patients with Alzheimer’s disease17. Higher Cu intake was negatively correlated with poor performance on the Digit Symbol Substitution Test (DSST) and Consortium to Establish a Registry for Alzheimer’s Disease (CERAD)4. Pb, Cd, and Hg were found to exhibit synergistic toxic effects, exceeding the effects of a single metal or a combination of two metals9.

The neurotoxic effects of metals in pediatric populations have been widely investigated, yet the literature concerning the impact of heavy metals multiple exposure on cognitive functions in older adults remains somewhat sparse18,19. As human grew older, brains might be more prone to the detrimental impacts of environmental toxins, lead to alterations in physiological and cognitive capacities associated with aging. Recent study had hinted at a possible association between heavy metal exposure and cognitive decline in the elderly20. These insights align with the growing understanding that environmental factors could contribute to the risk of cognitive aging and neurodegenerative conditions21. In addition, these studies did not carefully take into account the impact of confounding factors, such as sex, age, ethnicity, education, income, and body mass index (BMI) on cognitive function. Moreover, the complexity of heavy metals exposure, coupled with variability in research methodologies and challenges in exposure assessment, limited our capacity to draw definitive conclusions regarding the precise nature and extent of this association. Recent epidemiological studies have evaluated the impact of exposure to 1 or 3 heavy metals and did not evaluate the synergistic toxic or protective effects of several metals such as Pb, Cd, Hg, Se, Cu, and Zn10,22. Therefore, this study aimed to explore the association between cognitive function and mixed exposure to 6 metals using machine learning (ML) methods.

ML is increasingly being used in epidemiological studies to analyze vast datasets and predict patterns of disease transmission, risk factors, and outcomes to inform public health interventions and policies6,9. Several machine learning methods could be used to explore the relationship between exposure to heavy metals and cognitive function, including Quantile regression (QR), the least square regression (OLS), generalized linear models (GLM), restricted cubic splines (RCS), Bayesian kernel machine regression (BKMR), random forest (RF), and convolutional neural networks (CNN)12. However, QR, RCS, and BKMR provide the best tools to explore the relationship between exposure to heavy metals and cognitive function. QR models are often used to analyze the relationship between arbitrary quantiles23. RCS provides important information on nonlinear effects and dose-response relationships of individual factors while BKMR could be used to examine the synergistic effects of different exposure factors5,12.

Therefore in this study, we aimed to make use of machine learning (QR, RCS, and BKMR) to evaluate the association between exposure to single or multiple heavy metals on cognitive function using epidemiological data extracted from the National Health and Nutrition Examination Survey (NHANES) conducted between 2011 and 2014. The findings of this study will provide new epidemiological evidence to support the association between heavy metal exposure and cognitive function, and thus facilitate the development of environmental heavy metal monitoring and prevention strategies.

Materials and methods

Study population

The National Health and Nutrition Examination Survey (NHANES) is a cross-sectional population health survey conducted by the National Center for Health Statistics (NCHS). The NHANES employs a complex, multistage probability sampling design to ensure that the data was representative of the non-institutionalized civilian population in the United States. Data from two cycles of the NHANES conducted between 2011 and 2014 were acquired. This survey contained information related to the blood levels of heavy metals and cognitive function. The data was collected by the Centers for Disease Control and Prevention (CDC), with participants in the NHANES receiving a full physical examination at the Mobile Examination Centers (MEC). Beyond the physical assessments, individuals also engaged in extensive interviews that ascertain demographic data and cognitive function test results. The NHANES data did not include information on individuals with at higher risk of developing dementia for genetic susceptibility. Furthermore, blood samples were collected from the participants and subsequently transported to laboratories for the quantification of various metallic concentrations. Only the records of individuals above 60 with complete blood Pb, Cd, Hg, Se, Cu, and Zn levels, cognitive performance tests, and sociodemographic data were included in the study. Written informed consent was obtained from all the participants or proxies. The survey protocol was approved by the Research Ethics Review Board of the NCHS. The selection process of the eligible participants is illustrated in Fig. 1.

Fig. 1
figure 1

Participants selection flowchart for the study.

Full size image

Measurement of cognitive function

The cognitive function test included the CERAD, the Animal Fluency Test (AFT), and DSST. The CERAD consists of an instant recall test (IRT) and a delayed recall test (DRT). AFT requires the participants to list as many animal names as possible to evaluate categorical language fluency. DSST assessed processing speed, continuous attention, and working memory by having participants match symbols to numbers. None of the tests have established benchmarks for poor cognitive performance. However, for all cognitive tests, higher scores indicate better cognitive performance12.

Measurements of blood heavy metals levels

Whole blood samples were processed, stored, and transported to the NHANES, Centers for Disease Control and Prevention (CDC) for analysis. The blood dilution step prior to analysis involved a simple dilution of 1 part sample + 1 part water + 48 parts diluent. The concentrations of Pb, Cd, Hg, and Se in whole blood were determined by inductively coupled plasma mass spectrometry (ICP-MS) technology based on quadrupole ICP-MS. Serum samples were diluted at a ratio of 1 + 1 + 28 with water and gallium (Ga) containing diluent for multiple internal standardization. The levels of serum Cu and Zn were measured by inductively coupled plasma dynamic reaction cell mass spectrometry (ICP-DRC-MS) on processed serum samples. The quality assurance and quality control (QA/QC) protocols for the measured data of blood samples complied with the requirements of the Clinical Laboratory Improvement Amendments of 1988. The heavy metal concentrations below the limit of detection ( LOD) were recorded as LOD / sqrt 2.

All test results met the established quality control standards of laboratory science for accuracy and precision. The methods used to obtain the laboratory results are described detail on the NHANES website at https://www.cdc.gov/nchs/nhanes/index.htm.

Covariates

Household interviews were used to acquire demographic information, including age (60–69 years; 70–79 years; 80 + years), sex (male and female), ethnicity (Mexican American; Other Hispanics; Non-Hispanic White; Non-Hispanic Black; Non-Hispanic-Asian; Other Races-Including Multi-Racial), educational level (less than high school graduate, high school graduate / GED or equivalent, more than high school graduate), marital status (married, widowed, divorced, separated, never married, live with partner), annual household income ($0-$24999, $25000-$54999, $55000-$74999, and more than $75000) and BMI (normal (≤ 24.9 kg/m2), overweight (25–29.9 kg/m2), obese (≥ 30 kg/m2)). Participants were asked about their alcohol consumption during the past year. Participants who consumed 12 or more alcoholic drinks in the past year were classified as drinkers. The participants were also asked about their smoking status. Those participants who smoked at least 100 cigarettes throughout their lifetime and still smoked daily or occasionally were classified as current smokers, while those who stopped smoking were classified as former smokers. Conversely, the participants who smoked less than 100 cigarettes in a lifetime were classified, and non longer smoked were classified as non-smokers.

Statistical analysis

The demographic data were summarized using descriptive statistics according to gender. The continuous variables were expressed as mean (standard deviation, SD), and the categorical variables were expressed as number of instances ( n ) and frequency (%). The Shapiro-Wilk test was used to analyze whether the data was normally distributed. Directed acyclic maps (DAG) were used to screen for associated covariates24 (Fig. 2). The QR, RCS, and BKMR models were used to analyze the impact of heavy metal exposure on the development of cognitive impairment.

Fig. 2
figure 2

Selection of confounders factors in Directed Acyclic Graphs (DAGs). Note: The left-most column in the figure represented the independent variables, the middle column represented the confounders, and the right-most column represented the outcome variables. The solid line represented the association of confounders with the independent and dependent variables. The dashed line represented that we speculated about an association based on previous literature.

Full size image

QR model analysis

Since the data did not conform to a normal distribution the QR model applicable to analyzing non-normally distributed data was selected to predict the conditional mean, median, and any other quantiles25. QR was used to analyze the association between heavy metal exposure and cognitive function across different points of the cognitive function distribution according to 5 different quantiles (0.10, 0.25, 0.50, 0.75, 0.90). This approach is particularly useful for handling skewed response distributions and can reveal varying impacts that traditional mean regression might miss. The coefficient and 95% confidence intervals (95% CI) were used to describe the relationship between exposure to heavy levels and the development of cognitive function.

RCS model analysis

RCS was used to explore the nonlinear effects and dose-response relationships between the 6 heavy metals and cognitive function. Given the large sample size, four percentiles (0.05, 0.35, 0.65, 0.95) were chosen as nodes for the spline function to create a smooth curve. The p-values and 95% CI were used to describe the nonlinear of the associations between heavy metals and cognitive function.

BKMR model analysis

The BKMR mode was used to evaluate26:

1) the contribution of each heavy metal element to the overall effect of combined heavy metals

2) the overall effect of combined heavy metals exposure on cognitive function

3) the dose-response relationship between the median single heavy metal levels and cognitive function

4) the impact of the single heavy metal exposure at the 75th and 25th percentile, with all the other metals fixed at the 25th, 50th, and 75th percentiles. In addition, the bivariate exposure-response profiles were used to assess the potential interactions between the slope of the curve for one chemical at the 20th, 40th, 60th, and 80th modifications of another chemical with the remaining variables fixed at the median.

The interactions between heavy metals and cognitive function were verified using the generalized cross-validation kernel integration (CVEK). p < 0.05 indicated an interaction between the heavy metals.

All models were adjusted for sex, age, ethnicity, education level, marital status, annual household income level, BMI, smoking status, and alcohol use. The data were analysed using the R software (version 4.3.2) and statistical package for social sciences software (SPSS) (version 24.0). The model analysis was performed using the “quantreg”, “rms”, and “bkmr” packages available on the R software.

Results of the study

Basic characteristics of the study subjects

A total of 19,931 completed the NHANES survey. Out of the total respondents, 3632 were aged 60 or above. After excluding the individuals with missing blood Pb, Cd, Hg, Se, Cu, Zn (n = 2567), cognitive performance test (n = 157), or demographic characteristics (n = 107) a total of 811 individuals remained of whom 395 were males and 412 females (Table 1). Most of the individuals were aged between 60 and 69 years (54.1%), non-Hispanic white (50.3%), more than high school graduate (52.8%), and low-income earners (37.9%). The normality test (Shapiro-Wilk) showed that the heavy metals and cognitive function data followed a skewed distribution (Table s1). Hence, the concentration levels of heavy metals were log-transformed for the subsequent data processing.

Table 1 Basic characteristics of the study subjects.
Full size table

Association between single heavy metal exposure and cognitive function according to the QR model

Since the data did not follow a normal distribution the QR model was used to analyze the data. The results showed that the association of heavy metals with cognitive function at different sites (Fig. 3). Cd was found to be inversely associated with IRT at the 0.5th quartile level (p = 0.048, 95%CI: -2.7, -0.1). Zn was positively associated with IRT at the 0.5th quartile level (p = 0.022, 95%CI: 0.55, 11.73). Se was positively associated with DSST at the 0.25th quantile level (p = 0.011, 95%CI: 6.62, 47.74), while Cu was negatively associated with DSST at the 0.25th tertile level (p = 0.049, 95%CI: -37.75, -0.09).

Fig. 3
figure 3

Quantile regression analysis of association between heavy metals and Cognitive function. Using different color discrimination, the positive (red) and negative (blue) phase effects of heavy metals on cognitive function tests. Note: IRT, the instant recall test; DRT, the delayed recall test; AFT, the Animal Fluency Test; DSST, the Digital Symbol Replacement Test. Model was adjusted for sex, age, ethnicity, education level, marital status, annual household income level, BMI, smoking status, and alcohol use.

Full size image

Association of single heavy metal exposure with cognitive function according to the RCS model

The dose-response relationship between heavy metals exposure and cognitive function was further investigated using the restricted square spline (Fig. S1-S4). The RCS model showed that Pb levels had an “inverted U-shaped” nonlinear association with the DSST (p-overall = 0.008, p-nonlinear = 0.005). Se showed a positive phase nonlinear relationship with DSST (p-overall = 0.001, p-nonlinear = 0.011). This association remained stable even for Se concentration above 2.29 (log10Se). A negative association was noted between Se and DSST at Se concentrations of 2.17 µg/L (95%CI: -4.16, -1.01) to 2.195 µg/L (95%CI: -0.18, -0.04) (log10Se). However, for Se concentration between 2.197 µg/L (95%CI: 0.004, 0.15) to 2.29 µg/L (95%CI: 2.56, 7.64) (log10Se), a positive association was noted between Se and DSST. Cu showed a negative phase linear dose-response relationship with DSST (p-overall = 0.020, p-nonlinear = 0.411) (Fig. 4). In addition, a negative association was noted between DSST and Cu levels for concentrations ranging between 2.16 (95%CI: -6.07, -0.02) to 2.25 (95%CI: -10.71, -2.48) (log10Cu).

Fig. 4
figure 4

Restricted cubic spline plot between heavy metal elements and cognitive functions. (a) : Pb correlated with DSST. (b) : Se correlated with DSST. (c) : Cu correlated with DSST. Note: Model was adjusted for sex, age, ethnicity, education level, marital status, annual household income level, BMI, smoking status, and alcohol use.

Full size image

Association of multiple heavy metal exposure with cognitive function according to the BKMR model

Table 2 summarizes the results of the association between the 6 heavy metals and cognitive function according to the BKMR model. Exposure to Zn had the highest impact on the IRT score, exposure to Se had the highest impact on the DRT score, and exposure to Cu had the highest impact on the DSST score (Table 2). The combined exposure to the 6 heavy metals showed a positive linear association with IRT and DRT scores, and a negative linear association with DSST score (Fig. 5).

Table 2 Summary results from BKMR analysis in the whole population.
Full size table
Fig. 5
figure 5

Combined effects of the metal as a mixture on cognitive function in elderly people. a, represented IRT; b, represented DRT; c, represented AFT; d, represented DSST. Note: Model was adjusted for sex, age, ethnicity, education level, marital status, annual household income level, BMI, smoking status, and alcohol use. IRT, the instant recall test; DRT, the delayed recall test; AFT, the Animal Fluency Test; DSST, the Digital Symbol Replacement Test.

Full size image

A negative association with IRT was then noted when the Hg concentration levels ranged between − 0.959 µg/L (95%CI: -1.813, -0.127) to -0.546 µg/L (95%CI: -0.983, -0.003) (log10Hg). Se and DSST showed a negative phase linear association relationship. Se was negatively associated with DSST at Se concentration levels between 2.02 µg/L (95%CI: -9.272, -0.648) to 2.24 µg/L (95%CI: -2.511, -0.029) (log10Se). Cu showed a negative association with DSST at Cu concentration levels between 2.19 (95%CI: -3.118, -0.002) µg/L to 2.38 µg/L (95%CI: -7.083, -0.497) (log10Cu) (Fig. 6). When the concentrations of other metals were positioned at the 25th, 50th, and 75th percentiles, the alterations in the concentrations of individual metals exhibited no association with the outcome variable (Fig. S5).

Fig. 6
figure 6

Univariate exposure respon cognitive functions and 95% confidence interval for each heavy metal with the other metals fixed at the median in elderly people. a, represented IRT; b, represented DRT; c, represented AFT; d, represented DSST. Note: Model was adjusted for sex, age, ethnicity, education level, marital status, annual household income level, BMI, smoking status, and alcohol use. IRT, the instant recall test; DRT, the delayed recall test; AFT, the Animal Fluency Test; DSST, the Digital Symbol Replacement Test.

Full size image

The parallel exposure-response relationship indicated that there was interaction between heavy metals which impacted the cognitive function (Fig. S6). After validated by the generalized cross-validation kernel ensemble (CVEK) model, the interaction between Cd with Hg was found to have an negative impact on the IRT(p = 0.002) and DRT(p = 0.008) scores. In addition, an interaction was also noted between Cd and Se which negatively impacted the IRT (p = 0.031) and DSST (p<0.001) scores. Similarly, the interaction was also noted between Cd and Zn which negatively impacted the IRT (p = 0.023) and DSST (p = 0.001) scores. The interaction between Cd and Cu negatively impacted the IRT score (p = 0.014).

Discussion

As the levels of heavy metals within the environment continue to increase there is an urgent need to understand the impact of these heavy metals on human health. Although numerous studies have been conducted to evaluate the impact of heavy metals on human health the impact of multiple metal exposure on cognitive function remains under-explored. In this study, we made use of several statistical methods to evaluate the association between Pb, Cd, Hg, Se, Cu, and Zn 6 heavy metals and cognitive function in older adults. Our findings indicated that exposure to Pb, Cd, Hg, Se, Cu, and Zn has an impact on cognitive function. Furthermore, this study also observed an interaction between several heavy metals.

An “inverted U-shaped” trend relationship was noted between Pb and DSST. Pb is known to be a neurotoxin. Pb exposure in early life or prolonged life exposure was linked with accelerated cognitive deterioration in later life22. Higher blood Pb levels in elderly aged 65 and older were linked with cognitive impairment in elderly people living in one of the nine longevity regions of China27. While our findings hinted at a potential positive association between Pb exposure and cognitive function, this association lacks empirical support from existing literature. We estimated the observed anomaly could be a consequence of unaccounted-for confounding variables which were not factored into our analysis. At the same time, we speculated that the mechanisms driving this association may involve complex biological processes. Further research were imperative to validate the our findings and elucidated the specific conditions or biomarkers which may underlie this association. The established toxicity of Pb far exceeded any potential protective effects, and thus, in the lack of compelling biological evidence, we maintain that Pb exposure was deleterious to cognitive function. Future studies should delve into the complex interplay between Pb exposure and cognitive function, and there should be an ongoing emphasis on minimizing Pb exposure to safeguard public health.

Similar to a study conducted in China, this study showed a negative association between Cd and cognitive function in older adults28. Cognitive decline and Alzheimer-like neuropathology are common manifestations of Cd poisoning29. Cd exposure could structurally and functionally disrupted the mitochondrial integrity, and ultimately lead to neurotoxicity30. A study conducted in Bangladesh revealed an inverse association between Cd and spatial recognition memory16.

Hg was inversely associated with IRT in the BKMR model. The existing scientific literature clearly implicated Hg, specifically its organic form of methylmercury had association with cognitive impairment, neurodegenerative disorders31,32. The neurotoxic mechanisms of Hg included interfering with the release of neurotransmitter, inhibiting the activity of nerve growth factors, causing oxidative stress, and causing apoptosis33. These biochemical change affected the structure and function of the brain, which affected cognitive functions such as memory, attention, language skills, and executive functions34. Individuals living in Raima, brazil with methylmercury MeHg levels above 6.0 µg/g had worse cognitive decline than those with lower levels35. A cross-sectional study conducted in the United States found that 18% of patients with idiopathic axonal neuropathy and 9% with small fiber neuropathy had blood mercury levels above 10 µg/L36. Given the potential negative effects of Hg on cognitive function, it was important to take preventive measures to reduce Hg exposure in the population, especially in children and women of reproductive age. For already exposed individuals, timely intervention and support measures such as cognitive training and nutritional support may help to mitigate the potential effects of Hg on cognitive function. Future studies are needed to further explore effective intervention strategies to protect the most vulnerable population the neurotoxic effects of Hg.

High concentrations of Cu was negatively associated with DSST in this study. Only a very small amount of Cu is required for smooth brain functioning37. Cu imbalance has been linked with the development of various neurological disorders including, Alzheimer’s disease, Parkinson’s disease, and Huntington’s disease38. A non-linear inverse association between dietary Cu intake and cognitive decline was noted in older adults39. Gu et al. also found an association between high whole blood Cu concentration and mild cognitive impairment in elderly patients11. Breaking the balance of Cu concentrations in vivo is not clear and requires further confirmation.

In this study, we also observed a positive association between Se and Zn on cognitive function. Se is an antioxidant that has an important role in protecting brain cells from oxidative stress40. As a result, studies have shown that Se can improve memory and overall brain health41. Not surprisingly an Amercian study showed that patients with Alzheimer’s disease had low concentrations of blood Se and Zn42. Moreover, Cheng et al. also showed that Se reduced the negative effects of As, Cd, and Pb in Chinese community-dwelling older adults43. However, it was noteworthy that the concentration and morphology of Se exert distinct influences on cognitive function. Se consumption should be carefully regulated, as levels that exceed the required amount may lead to neurotoxicity44. In studies of amyotrophic lateral sclerosis (ALS) patients, it has been observed that Se concentration was linked to the presence and severity of the disease45,46. Consequently, we speculated that low concentrations of Se may have a certain protective effect on cognitive function, but excessive selenium intake still poses a risk of cognitive impairment. Additionally, the chemical speciation of Se was crucial in determining its neurotoxic potential. Studies have suggested a correlation between Se and certain selenium-containing compounds, such as selenoprotein P, which enhanced the risk of mild cognitive impairment advancing to dementia47. It is necessary to delve deeper into the role of Se in preventing and treating neurological diseases in future studies.

The Chinese Health and Nutrition Survey (CHNS) found that maintaining appropriate dietary Zn levels could prevent cognitive decline48. Similarly, a strong association between cognitive decline and blood Zn deficiency was observed in community-dwelling older Brazil adults49. This evidence suggests that the maintenance of normal Se and Zn levels in humans may reduce the risk of cognitive impairment and these elements should therefore be included in the diet.

The combined exposure of the 6 heavy metals showed a positive linear association with IRT, DRT, and a negative linear association with DSST. Among the PIP values, Zn had the largest proportion in IRT. Se had the largest proportion in DRT, and Cu had the largest proportion in DSST. Zn were identified as protective factors and were positively associated with cognitive function4,40. Conversely, high blood concentrations of either Cd and Hg, Cu resulted in increased cognitive dysfunction.

The strength of our study was that it was the first to analyze the association between combined exposure to Pb, Cd, Hg, Se, Cu, Zn and cognitive function. Based on the study of single heavy metal exposure, the effects of combined exposure on cognitive function were further explored to provide a new basis for the effects of heavy metal on cognitive function. A widely used and representative database provides more scientific data support for this study. This study combined three methods: QR, RCS, and BKMR, to better predict the association of heavy metals and cognitive function.

Of course, this study still had limitation. Although we tried to exclude various confounding factors such as age and gender during the analysis, there may be other confounding factors that were not taken into account.

Conclusions

Overall, this study showed that Pb, Cd, Hg, Se, Cu, and Zn were associated with cognitive function. High levels of Pb, Cd, Hg, and Cu were negatively correlated with cognitive function. Conversely, Zn were positively associated with cognitive function. Joint exposure to the 6 heavy metals showed a positive linear association with IRT, DRT, contrarily, a negative linear association with DSST. The interaction between Cd and Hg, Cu resulted in worse cognitive function. However further research is required to confirm these findings.