Our findings show that people with lower incomes, who may be more likely to suffer from other disparities in health and exposures, have a greater burden of exposure to BPA. The results for children are especially troubling. Children overall had higher urinary BPA concentrations than teenagers or adults, but children whose food security was very low or who received emergency food assistance - in other words, the most vulnerable children - had the highest levels of any demographic group. Their urinary BPA levels were twice as high as adults who did not receive emergency food assistance. Concerns about health effects from BPA exposure are strongest for young children and neonates because they are still undergoing development [3
]. Results for BPA by race/ethnicity, adjusting for income, revealed that Non-Hispanic Whites and Blacks had similar urinary levels, and being Mexican American appeared to be highly protective.
Findings for PFCs revealed differences by socioeconomic position in the opposite direction. Participants with the highest incomes had the highest serum concentrations. We did not see the same vulnerability in younger age groups as with BPA; associations with income were strongest in adults. However, NHANES did not measure PFCs in 6-11 year-olds. While there was some variation by race/ethnicity, Non-Hispanic Whites had the highest levels for two of the four PFCs and being Mexican American again appeared to be protective.
The possible pathways by which SEP is associated with differential exposures to BPA and PFCs may be elucidated by comparing results for the individual SEP variables. Family income was by far the most consistent and important predictor of concentrations; it had a clear dose-response pattern for all chemicals, and remained the strongest when included in models with other SEP variables simultaneously. Conceptually, income reflects access to material goods; a family's current household income is the most specific measure of their immediate financial resources [28
]. Thus, income may affect exposure to BPA and PFCs via types of foods consumed or via other consumer products used (or not used) in the home. Past studies that have examined the effect of income, education, and occupation on diet quality have consistently found that income is the most important and strongest predictor of diet [43
]. Given that diet is assumed to be a major pathway of exposure to these chemicals, differences in food purchasing patterns by income seems one likely explanation for the observed differences.
The literature on specific differences in diet by measures of SEP is large; a review offers this summary of socioeconomic status (SES) and dietary intake: "available evidence suggests that consumption of whole grains, lean meats, fish, low-fat dairy products, and fresh vegetables and fruit was consistently associated with higher SES groups, whereas the consumption of fatty meats, refined grains, and added fats was associated with lower SES groups" [45
]. Cost of food is a compelling hypothesis for why this differential exists, as foods with higher energy density are cheaper per amount of energy, but also tend to be nutrient-poor [45
]. This is thought to be an important reason why consumption of fresh fruits and vegetables in particular is lower in people with lower incomes. Income is also a strong determinant of where a person lives, and there is a growing body of literature on the lack of access to large supermarkets and ample fresh fruits and vegetables in lower income neighborhoods [46
This is the first study to examine the relationship between body burdens of BPA and PFCs and two measures of food security as possible proxies for SEP. We conceptualized these variables as representing the intersection of income and dietary behavior, and assumed that those with very low food security, or those who received emergency food, were an especially vulnerable population in terms of accessible dietary options. Particularly in regards to BPA exposure, we hypothesized that they would be more likely to eat canned foods. Recent research indicates that eating canned and packaged foods can contribute to BPA body burdens [47
]. We did see associations in the hypothesized direction between BPA and the food security measures; there was a particularly strong signal with very low food security compared to low and marginal food security, and an association of slightly lower magnitude in those who received emergency food. These associations were attenuated when controlling for income, though coefficients remained positive. More striking were the associations between BPA and food security in 6-11 year-olds, which were of the greatest magnitude of any age group. This could be due to greater consumption of foods containing BPA, or the fact that children consume more food per body weight than adults. Though use of emergency food was also associated with PFCs, food security status was not as important a predictor for these compounds.
We saw fewer and less consistent associations between education and concentrations of BPA and PFCs, particularly PFCs. Education is a long-term indicator of SEP, and embodies the transition in SEP from childhood to adulthood [28
]. In terms of the specific ways in which education may impact exposure to chemicals, it is thought to represent the ability to access and interpret health-related information [28
]. With exposure to BPA and PFCs, however, this type of knowledge may not be useful in reducing exposures, as consumers most often do not know that they are being exposed to these chemicals, nor how exposure is occurring. The recent flurry of media and political action around BPA in baby bottles and PC water bottles may be changing this dynamic for BPA, and may explain the slightly stronger associations we observed with education, but increased attention began only in the last few years and is not relevant for the bulk of the study period [48
]. Similar to previous studies, we found some discordance between education and current income [44
]. For BPA, which has a very short half-life, we would be more interested in current and not long-term income, since the foods and products a person purchased in the very recent past directly contribute to urinary levels.
There appeared to be little association between BPA and PFCs and occupation; however, our ability to draw conclusions about these relationships is limited by the fact that we had smaller numbers as data were only available for adults from 2003-2004. We did not see many associations between the compounds and occupation classified into five skill- and work relations-based categories. It seems unlikely that work-related psychosocial stress would affect exposure, though the physical conditions of a workplace could contribute to exposures. Examples include working in an office with new carpeting or furniture that contains PFC precursors [21
] or a job in retail that involves handling credit card receipts that contain BPA [50
The weaker associations observed for education and occupation may also be partly related to the fact that both are measured on the level of the individual, whereas family income and food security are family-level measures [44
]. The latter two variables may be more accurate measures of SEP insofar as family purchasing patterns are concerned. This distinction could be important for food purchasing behavior, as it is not clear who in the family (i.e. the participant or some other family member) makes the food shopping decisions.
Our results clearly show differences in BPA and PFC body burdens by measures of SEP that were not explained by race/ethnicity, and vice versa. It is likely that cultural behaviors and patterns are associated with race/ethnicity independent of SEP. The strikingly lower concentrations of both chemicals in Mexican Americans, even after controlling for income, was the most notable result regarding race/ethnicity. This is particularly unexpected for BPA, where Mexican Americans do not fit with the observed pattern of lower income groups having higher urinary concentrations. Mexican Americans and Hispanics have been shown to have higher intake of fruits and vegetables compared to Non-Hispanic Blacks and Whites in different population-based surveys, including NHANES 2003-2004 [51
], the 2005 California Health Interview Survey [51
], and the 2000 National Health Interview Survey [52
]. Eating more fresh fruits and vegetables is likely to be associated with eating less canned foods, which may explain the lower urinary BPA levels seen in Mexican Americans compared to other groups.
In addition, we observed that foreign-born Mexican Americans had markedly lower serum concentrations of PFCs than U.S.-born Mexican Americans, except for PFNA. This is consistent with the fact that PFCs have long half-lives, and exposure from many years past (i.e. when living in Mexico, where exposures may be lower) could impact current serum levels. Similar patterns have been seen for some other persistent lipophilic chemicals [40
]. For BPA, foreign- and U.S.-born Mexican Americans had similar levels, which makes sense given that BPA has a short half-life, and lower exposures from years past would not matter.
A final aspect of SEP and its relationship with race/ethnicity that must be mentioned is wealth. Wealth can be thought of as the "accumulated assets" of an individual or family, usually in the form of savings, real estate, and inherited items, and represents economic security [54
]. While there are no direct measures of wealth in NHANES, differences in wealth by race/ethnicity are reported to be much larger than differences by income; for the same income, the amount of wealth for African Americans and Hispanics has been shown to be much lower than for Whites [28
]. Thus, adjusting for income alone may underestimate the real effect of SEP [28
], and differences by race/ethnicity may suffer from residual confounding due to inability to adjust for wealth.
Our findings have various practical implications for environmental epidemiology. It is standard in environmental epidemiology studies to include some measure of SEP as a covariate in models. But, it is rare to see a discussion of the rationale for the choice of SEP variable. In many instances, there seems to be an assumption that different measures, particularly income and education, serve as surrogates for the same underlying phenomenon, and that they can be used interchangeably. Our study finds that, for urine and serum concentrations of BPA and PFCs, this is not the case; the SEP measures we studied do not overlap entirely with one another, and had different estimates of effect in our regression models. We conclude that, in the context of this study, income, education, occupation, and food security do not represent the same socioeconomic constructs, but rather seem to capture different aspects of how SEP may be related to exposure to BPA and PFCs. While constraints regarding data availability and the need to maximize sample size will always be an issue, the question of which SEP measure to use is an important methodologic concern, and merits more consideration by researchers in the field.
As discussed, family income was the most important SEP predictor in our investigation. We found that adequate gradations must be used, however, to see the full extent of the effect. When modeled as a dichotomous variable with a cut point of $20,000, which is tempting to do in NHANES as there are fewer missing participants, the full effect of SEP was underestimated. There was also some indication that income measures that adjusted for household size, such as adjusted family income and PIR, were stronger predictors. Regarding education, we found that using a dichotomous variable with a cut point of high school graduation did not fully capture the SEP difference. Participants with some college or an associate's degree were more similar to high school graduates than college graduates. A disadvantage of using education as a measure of SEP is that it is not a useful measure for children, a concern that also applies to occupation. In addition, our findings related to occupation are limited due to the smaller sample size, but it may be the case that a different approach to categorizing occupation, such as one based on type of industry, would be more closely related to the outcomes of interest. Food security, though not as commonly used to assess SEP, revealed important information about a vulnerable population - children whose families have very low food security or receive emergency food aid - information that other SEP measures failed to provide.
Regarding the measurement of race and ethnicity in studies such as these, our findings show that useful information can be gleaned from considering country of origin, particularly for Mexican Americans. This follows previous examples [40
Inherent in our study are a number of limitations. One concern is possible confounding by geography, which we cannot assess with publicly-available NHANES data. Regional and local populations vary in measures of SEP and race/ethnicity; if BPA and PFC exposures also differ with geography, there could be confounding. This geographic variation in chemical exposures could occur through differences in environmental contamination-localized contamination with certain PFCs has been reported in the USA (e.g., Hoffman et al. [55
])- or in consumption patterns of foods and other products that lead to exposure. In particular, the striking findings for Mexican Americans must be taken with caution. Zota et al. [40
] received permission to access state-level geographic information for NHANES 2003-2004 PBDE data and showed that, because Californians overall had higher serum concentrations of PBDEs and a large proportion of Mexican Americans sampled by NHANES lived in California, residence in California confounded results for Mexican Americans. Further investigation of geographical differences in body burdens would be greatly aided by the public release of NHANES data indicating region of the USA, something that would appear unlikely to threaten confidentiality.
We relied on a single biomonitoring measurement of the chemicals of interest. This is less of an issue for PFCs, which have long half-lives and we would not expect concentrations to vary significantly within an individual. For BPA, however, this is more of a concern. Mahalingaiah et al. [56
] showed a single spot urine sample to be predictive of exposure over weeks to months, despite within person variability. However, they assessed a single sample's ability to classify participants into tertiles, which is not how we modeled our data. And, they concluded that a second sample offered improvements in classifying individuals.
For both compounds, there are complications involved in interpreting results from biomonitoring data. While biomonitoring measurements provide a useful estimate of internal dose, there is likely inter- and intra-individual variation in measurements as a result of various factors that influence the chemical's pharmacokinetics, i.e. its distribution among compartments of the body, metabolism, and excretion [57
]. These factors include genetics, biological characteristics such as gender, body fatness, and liver function, and environmental factors such as diet, all of which may affect a chemical's pharmacokinetics. Though these concerns may be particularly relevant for BPA, which is measured as a urinary metabolite, many questions remain about the pharmacokinetic behavior of both BPA and PFCs in the body.
The measures of SEP we studied are all based on self-reported data. Getting participants to report personal income in particular is notoriously difficult [28
]. However, the NHANES approach of asking people to report in income categories seemed to work reasonably well, as less than 4% of participants were missing family income data. A limitation in our assessment of SEP was the availability of only two years of data for occupation.
Our study has several strengths, including a large sample size, unrivalled in studies of this nature that involve costly biomonitoring measurements. The NHANES sampling methodology of oversampling certain racial/ethnic, income, and age groups was critical in providing an excellent distribution of participants across different categories. Thus, we had ample power to detect associations between different SEP and racial/ethnic groups, and were able to consider modification by age and gender. Another key advantage was the availability of robust data on a variety of SEP measures. This enabled us to compare different SEP-related variables.
This paper has primarily explored associations between body burdens and measures of SEP and race/ethnicity. More research is needed on the specific aspects of diet, consumer products, and other activities or circumstances that provide the links between SEP/race/ethnicity and body burdens (Figure ). This question might be approached using dietary intake data, measures of indoor exposure, and other techniques.