Do exposure to heavy metals mediate the associations between socioeconomic indicators and self-rated health among the US adult Populations: A Weighted Quantile Sum Mediation Approach


 This study aimed to examine the associations between socioeconomic status (SES) and self-rated health (SRH) among US general adult populations and the extent to which blood and urinary metal mixtures explain these associations. We used 14 years of repeated cross-sectional data that consists of seven consecutive NHANES cycles from 2003-04 to 2015-16 (N = 9497). SRH was measured using a 5-point Likert scale, and SES was measured by Family Income to Poverty Ratio (FMPIR), levels of education, and employment status. Blood concentration of lead, mercury, and cadmium, and urinary concentrations of ten heavy metals (arsenic, barium, cadmium, cesium, cobalt, lead, mercury, molybdenum, thallium, tungsten) were used as metal mixtures. The total effects of SES on SRH were examined by linear regression. The direct effects of SES on metal mixtures were examined by weighted quantile sum (WQS) regression with Repeated Holdout Validation method, and causal mediation effect of mixtures was examined by model-based causal mediation technique. Results showed that SES (education β: 0.17; 95% CI: 0.15, 0.18; employment β: 0.16; 95% CI: 0.12, 0.21; and FMPIR β: 0.09; 95% CI: 0.08, 0.11) were positively, and the WQS indices of blood and urine metal mixtures (blood β: -0.04; 95% CI: -0.05, -0.03, urine β: -0.07; 95% CI: -0.13, -0.004) were significantly inversely associated with SRH in the US adult population. The novel finding was the mechanism between SES and SRH that exposure to heavy metals may explain socioeconomic inequalities in SRH in the US. Longitudinal studies are needed to corroborate this study results.


Introduction
People with a higher socioeconomic status (SES: higher educational attainment, having a greater wealth, and being employed in a white-collar job) are more likely to have better self-rated health (SRH) and higher life expectancy compared to those from lower SES groups (Hill and Needham 2006;Phelan et al. 2010;Suresh et al. 2011;Vonneilich et al. 2019). Firstly, people with higher SES have access to proper healthcare, nutritious food, better housing, healthier lifestyles, and less stressful events, therefore, reporting better health status (Power et al. 1998;Frankenberg and Jones 2004;Adler and Rehkopf 2008;Phelan et al. 2010;Suresh et al. 2011;Zajacova and Dowd 2011;Platts and Gerry 2017;Vonneilich et al. 2019;Budhathoki et al. 2020). Secondly, they are more likely to avoid environmental exposures and have multiple channels to prevent themselves from being continuously exposed to environmental toxins (Evans and Kantrowitz 2002;Brender et al. 2011;Chakraborty et al. 2011;Morello-Frosch et al. 2011;Bell and Ebisu 2012;Tyrrell et al. 2013). This unequal distribution of exposure to environmental toxins is partially due to differences in SES in the population, especially occupation, income, and education. Thus, people with low SES confront environmental injustice and health inequalities (Evans and Kantrowitz 2002;Evans and Kim 2010;Morello-Frosch et al. 2011;Bell and Ebisu 2012;Tyrrell et al. 2013;Cullati et al. 2014).
Epidemiological studies have shown that people who are highly exposed to a range of environmental toxins tend to have poor health status, as indicated by higher prevalence of respiratory disease, obesity, hypertension, cardiovascular disease, diabetes, cancer, liver damage, kidney ailment, vision impairment, and lower cognitive function in adults (Evans and Kantrowitz 2002;Evans and Kim 2010;Mendy et al. 2012;Cullati et al. 2014;Chowdhury et al. 2018). Two previous studies assessed the effects of toxic heavy metals on SRH and found that people with higher exposures to urinary heavy metals were more likely to rate their health as poor, fair, and good (Shiue 2015;Brailsford et al. 2018). Hence, some scholars theoretically argued that non-social environmental factors, such as exposure to pollutants, may constitute a signi cant pathway connecting SES and SRH, which has not yet been measured and studied empirically (Evans and Kantrowitz 2002;Evans and Kim 2010;Chakraborty et al. 2011). Moreover, most of the existing studies were limited to examine the effect of a single metal biomarker on health outcomes (Shiue 2013(Shiue , 2015Nigra et al. 2016;Brailsford et al. 2018;Moody et al. 2018).
The observed association between SES and the combined effects of metals is often overlooked in previous studies (Mendy et al. 2012;Tyrrell et al. 2013;Awata et al. 2017;Brailsford et al. 2018). To our best knowledge, no study has examined both blood and urinary metal mixture effects on SRH and its mediating role in the associations between SES and SRH using the mixture approach. Thus, this study aimed to examine the association between SES and SRH among US general adult populations and to what extent this association is explained by blood and urinary metal mixtures applying the weighted quantile sum (WQS) mediation approach.

Study Population and Data Sources
We used 14 years of repeated cross-sectional data (2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016) from the National Health and Nutrition Examination Survey (NHANES), which includes the consistent format of demographic, socioeconomic, dietary, and health-related questions over the years. NHANES is conducted by the Centers for Disease Control and Prevention (CDC) and the National Center for Health Statistics (NCHS) based on a multistage probability cluster sampling design to ensure a representative sample in noninstitutionalized US residents (NHANES 2018).
This study sample consisted of seven consecutive NHANES cycles: 2003-04, 2005-06, 2007-08, 2009-10, 2011-12, 2013-14, and 2015-16. Two earlier cycles (1999-2000and 2001-2002 and the latest cycle 2017-2018 of NHANES were not included in this study since some of the important heavy metals were not measured in the earlier cycles and were not yet released latest cycle. A total of 17934 participants had blood, and urinary measurements of heavy metals in the included seven consecutive cycles of NHANES. We excluded participants aged less than 20 years old (n=5986) because NHANES consider 20+ as the adult population. We also dropped the samples with missing data in the self-rated health (n=1042), SES, and confounders (n=1112) via listwise deletion. The nal eligible analytic sample was 9794 (Fig. 1). Detailed information on NHANES questionnaires, datasets, and related documentation is available at https://wwwn.cdc.gov/nchs/nhanes/Default.aspx.

Measurements of Metal Mixtures Exposures
Blood concentration of lead, mercury, and cadmium was measured in each included cycle of NHANES by an Inductively Coupled Plasma Mass Spectrometer with Dynamic Reaction Cell Technology (ELAN® DRC II) the National Center for Environmental Health. Venous whole blood samples were collected by phlebotomists, and details are published elsewhere (NHANES 2015(NHANES , 2018CDC 2018). Urinary concentrations of ten heavy metals (arsenic, barium, cadmium, cesium, cobalt, lead, mercury, molybdenum, thallium, tungsten) were measured in each included cycle of NHANES. Inductively coupledplasma dynamic reaction cell mass spectrometry (ICP-DRC-MS) was employed to analyze all the heavy metals concentration in urine by the Division of Laboratory Sciences, National Center for Environmental Health, Atlanta, Georgia (NHANES 2015(NHANES , 2018CDC 2018). The urinary concentrations of heavy metals were measured in micrograms per liter (ug/L) (NHANES 2018) and standardized by creatinine concentration to account for urine dilution (Mendy et al. 2012). We used the imputed values provided by NHANES for metal concentrations below limits of detection to be consistent with CDC's National Report on Human Exposure to Environmental Chemicals (CDC 2018). All blood and urinary heavy metal concentrations were log-transformed due to highly skewed distributions (Shiue 2015).

Socioeconomic Status (SES) Measurements
Socioeconomic status (SES) was measured employing commonly used indicators: educational attainment, family income to poverty ratio (FMPIR), and employment status of the participants (Hill and Needham 2006;Phelan et al. 2010;Suresh et al. 2011;Arcaya et al. 2015;Vonneilich et al. 2019). In NHANES, the levels of education were categorized as ve ordinal categories (less than 9th grade, 9-11th grade, high school graduate, some college or associate in arts (AA) degree, and college graduate or higher) (NHANES 2018). FMPIR, a continuous value, is calculated by dividing family income by poverty guidelines of the Department of Health and Human Services (HHS) speci c to the survey year (NHANES 2018). The employment status was a dichotomous variable: being currently employed=1 and unemployed=0 (NHANES 2018). We did not use a composite index of SES measures because these indicators tend to have different pathways of causation in relation to exposures to heavy metals (Brailsford et al. 2018). SES indicators were used as continuous exposures except for employment to develop statistical models in this study.

Self-rated Health (SRH) Measure
The overall SRH, a globally recognized subjective indicator of general health status, was considered as the outcome variable in this study (Suresh et al. 2011;Brailsford et al. 2018;Vonneilich et al. 2019). SRH was measured using a 5-point Likert scale. The highest score 5 denotes excellent SRH and the lowest score 1 denotes poor SRH. DeMaris notes that with a large degree of freedom, ordinal variables "with at least ve levels" can be treated as "approximately interval" in ordinary least squares (OLS) regression (DeMaris 2002).

Statistical Analysis
Descriptive statistics are presented as frequencies and percentages. Trends of SRH and exposure to heavy metals across all seven cycles and the pairwise correlations of blood and urine heavy metals are presented in a heatmap. SRH and metal concentration trends were tested using the Spearman rank correlation coe cient (Gauthier 2001). Statistical models were developed in three steps. We adjusted for age, gender, race/ethnicity, and year of the survey related to SES, heavy metals exposure, and SRH as confounders in all multivariable analyses.
In the rst step, the total effect of SES on SRH was measured through a linear regression adjusted for the aforementioned confounders. This allowed us to estimate the effect due to both paths represented in g.
2, described by the straight arrow pointing from SES to SRH, representing all potential mechanisms, and the path that goes through the metal mixture represented by the weighted quantile sum (WQS) index.
In the second step, a WQS regression model was used to examine the direct associations of SES with SRH (the arrow pointing to SRH from SES in Fig. 2), adjusting for mixtures of heavy metals effects (WQS index) and aforementioned confounders for each biological matrix. A WQS approach in conjunction with linear function considering SRH as the continuous outcome was employed, which considered all the measured heavy metals and included heavy metals were constrained to have the same direction of effects for SRH (Carrico et al. 2015;Czarnota et al. 2015a, b). The WQS regression model calculated a weighted index, a linear combination of the different metals standardized into quintiles, which denoted the whole-body burden of all ten metals. The calculated weight for each chemical represented how much a single chemical contributed to the WQS index of metals.
WQS regression involves the split of the dataset in a training and validation part on which the weight and the regression parameters occur, respectively (usually 40% of the data are used for the test and 60% for validation). In order to use all the observations to estimate both the weights and the regression coe cients, a repeated holdout validation was implemented (Tanner et al. 2019). A hundred repeated holdout validations were performed besides the 100 bootstraps to estimates the weights. The repeated holdout procedure also allowed us to characterize better distribution of the weights associated with each mixture element.  Fig. 2. The impact of WQS on SRH was then estimated through a linear regression model (arrow pointing to SRH from WQS in Fig. 2). The product of these two effects provides the estimate of the average indirect association between SES and SRH. Con dence intervals for all estimated regression parameters were assessed using bootstrap resampling. We applied two-sided statistical tests assuming a signi cance level of 5%. All analyses accounted for survey weights and were performed in R version 4.1.0 (R Core Team 2021).

Study Population Characteristics
The overall characteristics of the study population are presented in Table 1. A total of 9794 US adults aged 20-80 years were included in the analysis. The highest percentage (20.1%) of respondents were between 40-49 years old, and the lowest percentage of respondents were between 70-80 years old. There was almost similar proportion of men and women. The sampling from other racial/ethnic groups increased across the seven cycles. The percentage of participants with college or university level education increased from the 2009-10 cycle to 2016 cycle. The highest prevalence of unemployment rate (37%-43%) was recorded between 2009 and 2014. Overall, about 21% respondents belong to FMPIR <=1.30 and these percentages increased consistently until 2014 (20.5% to 23.6%).

Trends of Self-Rated Health in the US Adult Population
The trends of average SRH score reported by 9794 US adults across seven consecutive NHANES 2003-2016 cycles are shown in Fig. 3. The highest score 5 denotes excellent self-rated physical health, while the lowest score 1 refers to poor physical health. A general decreasing trend of SRH was found through the Spearman rank correlation coe cient (rho = -0.671; p = 0.099. SRH scores showed a downward trend from 2003 to 2014.

Trends of Blood and Urinary Heavy Metals Concentration and Correlations
The trends of creatinine-adjusted ten urinary heavy metals concentration in seven consecutive NHANES cycles (2003 to 2016) are depicted in Fig.4. Blood and urinary lead (BPb, UPb), cadmium (BCd, UCd), and urinary mercury (BHg, UHg) concentrations remarkably decreased across all seven NHANES cycles (p < 0.05). A signi cant decreasing trend was also found for blood Hg and urinary Cesium (Cs). An overall decreasing trend in urinary concentrations of Arsenic (UAs), Barium (UBa), and Molybdenum (UMo) was also observed, particularly in the last three cycles (2011-12, 2013-14, and 2015-16). However, the trends were not statistically signi cant. Urinary Tungsten (UW) also showed a non-signi cant overall decreasing trend. Conversely, Cobalt (UCo) and Thallium (UTl) had an increasing trend over the seven cycles, but only UCo was signi cant (p=0.012). A correlation heatmap shows the pairwise Spearman correlations among blood and urinary heavy metals in the US adult population (Fig. 5). Overall, we have observed weak to strong positive correlations among blood and urinary metals with a range of 0 to 0.59, with the strongest correlation between UCs and UPb (r=0.59) within the urine metals, while blood metal correlations ranged from 0 to 0.35. However, we have observed weak negative correlations among few bloods and urinary metals.
Total Effects of SES on SRH Table 2 shows the total effects of SES on SRH, and SES characteristics were positively signi cantly associated with SRH after adjusting for age, race, gender, and NHANES cycles (Table 2). Participants with higher levels of education (β: 0.17; 95% CI: 0.15, 0.18), being employed (β: 0.16; 95% CI: 0.12, 0.21) and with a higher FMPIR (β: 0.09; 95% CI: 0.08, 0.11) were more likely to have better SRH compared to their counterparts, which suggests socioeconomic inequalities in SRH.  Tables 3 and 4 show the direct and indirect effects of SES and WQS index on SRH, respectively, for blood and urine metals. The direct effects of SES on SRH were consistently signi cant after further adjustments for both blood and urine metal mixtures (mediator) along with other potential confounders (Table 3). The WQS indices of blood and urine metal mixtures (blood β: -0.04; 95% CI: -0.05, -0.03, urine β: -0.07; 95% CI: -0.13, -0.004) were signi cantly inversely associated with SRH. It indicates that the participants who have a higher index of metal mixtures concentrations in their blood or urine were more likely to have poorer SRH than participants who have a lower index of metal mixtures concentration. The estimated metals weight for the Repeated Holdout WQS index is shown in Fig.6. This suggests that both blood and urine Cd (Weighted 0.99 and 0.75 respectively) and urine Tl (Weighted 0.22) were the two highest weighted heavy metals in SRH. between signi cant and non-signi cant weights equal to the inverse of the number of elements in the mixture. Models were adjusted age, race, gender, education, employment, FMPIR, and NHANES cycles.
SES was inversely associated with the WQS index of mixtures that also indicates socioeconomic inequalities in exposure to metal mixtures (Table 4). Results suggest that participants with higher FMPIR (β: -0.06, 95% CI: -0.09, -0.04) and education (β: -0.03, 95% CI: -0.04, -0.02) are exposed to lower levels of metal mixtures in their urine compared to participants who report lower levels of FMPIR and education. We observed a similar result with employment status.

Mediation Effect of WQS Index
To untangle the causal relationship between SES, metal mixtures, and SRH, we examined the average causal mediation effect of the WQS index on SRH and presented it in Fig.7A-F. We observed a statistically signi cant but weak average indirect effect for SES through blood and urine heavy metal mixtures (Table  5 shows the mediated proportion of the effects ranged from 0.6% to 2.1%).

Fig. 7: Mediation effect of WQS index of heavy metals mixtures on SRH
The only non-signi cant mediated effect was for urine metal mixtures when considering the association between employment and SRH. Alternatively, the association between SES and SRH was signi cantly mediated by the WQS index. Discussion This is the rst study, to our best knowledge, that used one of the novel multipollutant approaches (WQS regression) to examine the association between SES and SRH among US general adult populations and the mediating effect of multiple blood and urinary metal mixtures in these associations. Our ndings show that the overall trend of SRH scores in US adults was downward from 2003 to 2014. Our results show that creatinine-adjusted urinary concentrations of cobalt and thallium increased in the period between 2003 and 2016. However, blood and urinary concentration of cadmium, mercury, and lead markedly decreased in this period. The SES indicators were positively associated with SRH after adjusting for age, race, gender, and NHANES cycles. SES were inversely associated with exposure to the WQS index of blood and urinary metal mixtures, and this WQS index was also inversely associated with SRH. In other words, the associations between SES and SRH were signi cantly mediated by the WQS index of blood and urinary metal mixtures, although the effect sizes were small. Consistent with previous studies, our ndings show that SRH had substantially deteriorated between 2003 and 2016, which was more pronounced between 2011 and 2016 (Salomon et al. 2009;Shiue 2015). This deterioration in SRH over the years may be explained by the growing socioeconomic and racial inequalities in health and SRH in the US (Suresh et al. 2011;Zajacova and Dowd 2011;Tellez-Plaza et al. 2012;Beck et al. 2014;Vonneilich et al. 2019), which is con rmed by our empirical assessment of signi cant positive association between SES and SRH.
Studies suggest that over the years, the successful public health interventions in the US have led to a substantial decrease in the environmental exposure to heavy metals (Muntner et al. 2005;Tellez-Plaza et al. 2012;Ruiz-Hernandez et al. 2017), which is consistent with our results. (Tellez-Plaza et al. 2012;Shiue 2013Shiue , 2015Ruiz-Hernandez et al. 2017). However, there are socioeconomic inequalities in the environmental exposure to heavy metal mixtures in the US adult populations. In other words, people from the lower SES group are signi cantly more exposed to environmental heavy metal mixtures compared with those from the higher SES group, leading to higher burden of environmental and health inequalities, which is in agreement with previous studies (Evans and Kim 2010;Phelan et al. 2010;Brender et al. 2011;Chakraborty et al. 2011;Morello-Frosch et al. 2011;Bell and Ebisu 2012;Tyrrell et al. 2013;Brailsford et al. 2018). People throughout their life are exposed to multiple heavy metals, which are widely dispersed in the environment, such as in food, substandard housing, industrial fumes, water, and air (Billionnet et al. 2012;Alloway 2013;Braun et al. 2016;Awata et al. 2017). These multiple metals can introduce interactions between correlated metals, model misspeci cation, confounding by co-pollutants, and overlooking synergistic or antagonistic effects in single pollutant analytic method (Billionnet et al. 2012;Braun et al. 2016). The unequal distribution of exposure to metal mixtures is because higher SES people may have multiple channels to avoid continuous exposure and reduce metal mixtures (Evans and Kantrowitz 2002;Brender et al. 2011;Chakraborty et al. 2011;Bell and Ebisu 2012;Tyrrell et al. 2013).
To untangle the complex relationship among SES, environmental exposure to metal mixtures, and SRH, we used a novel and innovative multipollutant WQS approach to overcome these complex pattern of metal exposures and understand the mechanism of socioeconomic inequalities in SRH (Billionnet et al. 2012;Braun et al. 2016;Bellavia et al. 2019;Renzetti et al. 2019). Our mediation analysis shows that the environmental metal mixtures mediated the association between SES and SRH. Each of them mediated this relationship in its unique way. For example, SES indicators, including education level, employment status, and FMPIR, were associated with the blood and urinary concentration of mixtures. On the other hand, the blood and urinary metal mixtures were also associated with SRH. Therefore, the association between SES and SRH may be explained by the exposure to multiple metal mixtures. People from the lower SES group were exposed to higher levels of metal mixtures and thus had lower SRH. A previous study explored heavy metals exposure as a mediator in the relationship between SES and SRH (Brailsford et al. 2018). This study was, however, conducted on a much smaller sample (singly 2007-2008 NHANES cycle) and used a conventional statistical method that was unable to examine simultaneous effects of common metals due to complex exposure patterns, high collinearities, and interactions among metals (Billionnet et al. 2012;Taylor et al. 2016;Wang et al. 2018;Rana 2019;Zhang et al. 2019). Our study has a much larger sample size (seven cycles of NHANES from 2003 to 2016) and rst time empirically measured metal mixtures as mediator applying a novel mixture analysis method. The results from our study strongly support the theory that the embodiment of heavy metals plays a substantial mediating role in the association between SES and SHR (Evans and Kantrowitz 2002;Evans and Kim 2010;Brailsford et al. 2018).
The main strengths of our study are the application of mixture analysis method-WQS with Repeated Holdout Validation approach, and mediating role of blood and urinary metal mixtures using large sample size and the high quality of the NHANES data with availability of detailed information about the demographic and socioeconomic characteristics of the participants. The blood and urinary heavy metals were also measured under rigorous laboratory quality control conditions, which ensures reliability and comparability of the concentrations from different years (CDC 2018;NHANES 2018). The main limitation is that although our sample size was quite big, the data came from multiple cross-sectional surveys conducted between 2003 and 2016. Therefore, it may not be possible to make strong causal inferences about the associations between SES and SRH and causal mediation effects of metal mixtures. However, our results were highly signi cant but small effect sizes, which suggests that it is highly likely that such associations, modifying, and mediating effects truly exist. Another possible limitation would be the use of creatinine-adjusted urinary concentrations of heavy metals as an indicator for environmental exposure. Depending on the elimination half-time of each studied heavy metal, urinary concentration may not always be the most valid indicator for assessing environmental exposure. However, we used blood metals, a better biomarker for heavy metal exposure assessment that has similar effects. Nevertheless, the NHANES dataset is the only one that contains this information for thousands of people, and there is evidence suggesting that the urinary concentrations of heavy metals from NHANES re ect the population level environmental exposure to a great extent (Mendy et al. 2012;CDC 2018;Wang et al. 2018;Zhang et al. 2019).

Conclusion
Our study found that SES were positively associated with SRH in the US adult population. We also found that the SES were inversely associated with blood and urinary concentration of heavy metal mixtures in the US population. The novel nding was the mechanism between SES and SRH that the heavy metal mixtures may play a mediating role in the association between SES and SRH. We can, therefore, infer that socioeconomic inequalities in SRH in the US may be explained by the exposure to multiple heavy metal mixtures. It could be a viable mechanism of SES inequalities in SRH.
Further research and longitudinal studies are needed to corroborate this study results and to make robust causal inferences. We suggest with reference to previous studies that neighborhoods with high concentration of poverty and people of low income and education are extremely vulnerable to the higher exposure to toxic heavy metals and their deleterious effects on SRH (Evans and Kantrowitz 2002;Evans and Kim 2010;Brender et al. 2011;Morello-Frosch et al. 2011;Tyrrell et al. 2013;Ard et al. 2016;Brailsford et al. 2018). Thus, crafting public health interventions speci cally tailored to these disadvantaged communities (Evans and Kantrowitz 2002;Adler and Rehkopf 2008;Evans and Kim 2010;Brender et al. 2011;Chakraborty et al. 2011;Tyrrell et al. 2013;Brailsford et al. 2018) would be a successful way of preventing the combined damaging effects of metal mixtures on SRH that low SES and heavy metal mixtures have together.

Funding
This research did not receive any speci c grant from funding agencies in the public, commercial, or not-for-pro t sectors.

Con icts of interest/Competing interests
The authors have no known con ict/competing interests that could have in uenced the results of this study.

Availability of data and material
All data les are available from the NHANES program of the National Center for Health Statistics: https://wwwn.cdc.gov/nchs/nhanes/continuousnhanes/default.aspx. The datasets generated during and analyzed during the current study are available from the corresponding author on reasonable request.