The Environmental Protection Agency (EPA) has defined cumulative risk as “the combined risks from aggregate exposures to multiple agents or stressors” [1
]. While much of the scientific community, including the EPA, has advocated for the study of cumulative risk models [1
], this is rarely done in epidemiological studies. When it is studied, researchers typically do not combine social and chemical exposures. Several studies have used various methodologies to study chemical mixtures [5
], and some scientists have combined psychosocial stressors [6
], but most researchers tend to construct separate models for chemical and nonchemical stressors. Recently, researchers have advocated for combining these exposures in order to have a more complete understanding of how exposures interact to affect outcomes [2
]. Although researchers have attempted to model the combined effects of chemical and nonchemical exposures to study lead and stress on cognition [9
]; benzene and stress on birthweight [10
]; and lead, cadmium, and polychlorinated biphenyls on blood pressure [11
], methods to study combined chemical and nonchemical are still lacking.
Studying multiple chemical and nonchemical exposures is of special interest to researchers studying perinatal health. One area of concern are indicators for fetal growth restriction, such as small for gestational age (SGA), which have shown to have both short-term and long-term effects on infants as they grow into adulthood [12
]. Despite extensive research into the exposures contributing to SGA, understanding the exact etiology has proven difficult. Several studies reveal possible associations with SGA, including environmental toxicants, race, ethnicity, income, maternal educational attainment, and maternal mental health [13
], but findings are often inconsistent and the biological rationale is not conclusively known. Studies are consistent in indicating that women who are non-Hispanic-black, low income, and have not attained degree past high school [15
] are more at risk for most adverse perinatal outcomes, including SGA.
Chemical exposures are also risk factors for reduced fetal growth. Heavy metals, which are common neurotoxins [20
], have been linked to impeded fetal development at high levels of occupational exposure. In utero exposure to certain heavy metals, like lead (Pb), cadmium (Cd), and mercury (Hg), have been associated with low birthweight (LBW) and preterm birth (PTB) [21
], although there are many studies that do not find an increased risk between chemical exposure and adverse birth outcomes [13
]. This is similar to studies researching other chemicals. For example, researchers studying the effects of phthalates and pesticides on fetal development have reached inconclusive results [13
]. However, there is evidence that, while at low doses there are no detectable risks when studied individually, there could be a cumulative risk when these low doses are studied in combination [31
]. Animal models have indicated that if modeled cumulatively, these chemicals do in fact impact fetal development [33
Beyond the possible link between environmental toxicants and birth outcomes, other researchers have focused on social stressors for their risk of adverse birth outcomes. Many point to the stress experienced during preconception and pregnancy, with evidence of an association between many indicators of stress and fetal growth [15
]. Women who report having depressive symptoms tend to have worse birth outcomes [36
], as well as a higher risk for SGA [14
]. Although women who experience these psychosocial issues trend towards having adverse birth outcomes, like chemical exposures, literature reviews have reported some ambiguity; the evidence is far from conclusive that stress disorders directly cause adverse birth outcomes [37
]. Inconclusive findings could be the result of inadequate methods for studying stress [43
], but it could also be a result of the type of populations that are studied. When studying low risk, healthy pregnant women, Voegtline et al. found that stress indicators, such as depressive symptoms, trait anxiety, and emotional well-being, tended to be non-significantly correlated with cortisol measured throughout pregnancy [44
Further complicating perinatal research is the possible interaction between chemical and nonchemical exposures. Common socioeconomic exposures, like race/ethnicity, income, and education, are correlated with each other [45
]. African Americans, for example, have higher incidences of PTB, impaired fetal growth, and maternal mortality [47
], with studies identifying stress induced by racism as an underlying factor [18
]. Levels of cortisol are higher among women of lower socioeconomic status (SES) [51
], with black women tending to report higher exposure to stress compared to other races [41
]. Race and ethnicity are commonly multi-correlated with other social factors [45
] and those that are disadvantaged are at additional risk to be ‘vulnerable’ to environmental exposures and psychosocial stress. DeFur et al. theorized that vulnerability, which is typically a result of low SES, interacts with physical and social environments and results in disparities in health outcomes [52
]. This theory is supported by numerous studies finding that lower-SES populations most commonly live in areas with higher environmental exposures, including proximity to pollution, industrial areas, and other areas containing large amounts of environmental toxicants [7
], and are thus more vulnerable to be exposed to psychosocial stress as a result of these circumstances [57
]. SES may also be an indicator of different dietary intakes or occupational exposures that may put people of lower SES at greater risk for environmental toxicants [59
Due to the multi-correlated nature of risk factors contributing to fetal growth, perinatal research could benefit from researching exposures cumulatively. This is especially true when studying relatively healthy women, as healthy women tend to be exposed to toxicants at levels lower than the regulatory concerned and associations of individual chemical and nonchemical exposures with birth outcomes are typically only evident at high levels of exposure [13
]. However, just as in chemical exposures that may only show risk at low doses with multiple exposures, there could be associations between the nonchemical and chemical risk factors and perinatal outcomes, if a combination of lower levels of exposure are measured cumulatively. These risks also may interact with each other, causing a high degree of multicollinearity. We have included a conceptual model to explain the hypothesized interrelationships of possible chemical and nonchemical stressors (Figure 1
). The objective of this study is to understand whether cumulative risk, indicated by the combination of exposure to heavy metals, psychosocial stress, and sociodemographic risk, contributes to incidence of SGA infants in a healthy cohort of pregnant women. We compare two possible ways of assessing this cumulative risk, one a relatively simple risk score that weights exposures equally, and one which allows for varying weights, weighted quantile sums [60
There were 2038 participants that met the inclusion criteria and provided a blood specimen. Women were divided evenly between non-Hispanic white, non-Hispanic black, and Hispanic (27.3%, 26.1%, and 27.4% respectively) (Table 1
). The majority of women either earned <$
= 489, 27.8%) or ≥$
= 521, 29.6%). Only 3% of women (n
= 58) scored high enough on the EPDS to be categorized as depressed. Seven percent of infants (n
= 150) were SGA.
A description of continuous variables is presented in Table 1
. About 7% (n =
148) of women had Pb levels below the limit of detection. Only 1% (n =
23) of women had Hg levels below the limit of detection and 8% (n =
171) of women had Cd levels below the limit of detection. The average Pb blood level was 0.51 micrograms per deciliter (μg/dL) and only 1% (n =
21) of the study population had a Pb blood level of greater than 5 μg/dL. Only one participant had an excessive level of Cd, measured at 5 μg/L, and one participant had blood Hg content (>5 μg/L).
Most variables were individually associated with SGA (Table 2
), with the exception of depression (OR: 0.32, 95% CI: 0.08, 1.33) and Hg (OR: 0.91, 95% CI: 0.78, 1.45). Non-Hispanic black women were at a higher risk of having SGA infants compared to non-black women (OR: 2.46, 95% CI: 1.59, 3.81). Higher levels of Cd and Pb also were associated with having an SGA infant (OR: 1.18, 95% CI: 1.04, 1.35 and OR: 1.22, 95% CI: 1.03, 1.35). Higher income was protective against SGA when compared to lower income (less than $
30,000) and attaining higher levels of education was also protective against SGA. Women with previous pregnancies, married women, and women who were older than 25 were less likely to have SGA infants compared to nulliparous, nonmarried, and younger women respectively.
Variables within domains tended to be correlated with each other (Table 3
); however, Hg and Pb were not significantly correlated. Though many of the variables were statistically significantly correlated, education and income appeared to be the only variables highly correlated to each other (
= 0.68, p
< 0.05). Depression and perceived stress were moderately correlated (
= 0.31, p
< 0.05). Only the two sociodemographic variables (education and income) were correlated with SGA (not included in table).
In unadjusted cumulative models (Table 4
), the demographic index and the total cumulative index were the only variables associated with SGA. The total cumulative index (Table 4
) was associated with SGA even with race/ethnicity included in the model (OR: 1.17, 95% CI: 1.02, 1.35). While not significantly associated with SGA, the unadjusted metals domain and the psychosocial domain appeared to have an association, given that the lower confidence level was close to the null (OR 1.16, 95% CI 0.97, 1.38 and OR: 1.20, 95% CI 0.96, 1.52 respectively). In the adjusted models, inputting the domains separately did not yield any associations between the indices or race/ethnicity and SGA.
The weights for the WQS regression models and the betas are described in Table 5
. The sociodemographic domain was the only domain associated with SGA (β = 0.41, SE = 0.20, p
= 0.04). According to the weights assigned in the WQS regression models, income appears to have the most influence on the index, as the weight for income was 0.69, while it was only 0.31 for education (Table 5
). Weights for adjusted and unadjusted WQS models, betas (β) and standard errors (SE) for final adjusted WQS regression models n WQS models, Fetal Growth Study.
This analysis indicates that a cumulative measure of multiple exposures could be associated with SGA. The index that included all three domains in the cumulative risk score analysis (heavy metals, psychosocial stress, and sociodemographics) was consistently associated with SGA, suggesting that modeling exposures cumulatively could help explain how lower levels of exposures jointly increase women’s risk of these adverse birth outcomes. However, in WQS regression models, the total cumulative model was not associated with SGA, and the only domain to be associated with the outcome was the sociodemographic index. Sociodemographics and race/ethnicity were so strongly associated with the outcome they outweighed effects of other exposures. Previous studies have indicated that controlling for indicators of socio-economic levels could control for other effects. This results in only the strongest SES terms being associated with the outcome, while producing null results among other indicators. In Lefmann et al’s study, for example, they believed that the reason why Medicaid status was the only predictor associated with SGA, LBW, and PTB was because Medicaid controlled for other correlated indicators, such as race and experiencing stressful events [75
]. They postulated that Medicaid status could be an indicator for poverty, thereby negating associations between other sociodemographic and stress variables in the models. A similar process could be operating with respect to exposures to toxicants; certain lower-SES populations are more vulnerable to environmental exposures [52
]. Several studies have indicated disadvantaged populations are at higher risk of exposure to both chemical and nonchemical exposures [3
]. DeFur et al., described the vulnerability in lower SES groups, meaning not just a susceptibility of lower SES groups to harmful exposures, but also an inability of these groups to recover. That results in disparities in outcomes between SES groups [52
]. DeFur et al.’s conclusion is plausible given that lower SES people tend to live in closer proximity to industrial areas and other areas with higher chemical exposures [7
]. However, depending on the population, other potential sources of pollutant exposure (diet, consumer products, and occupation) may not have the same correlation with SES: for instance, jobs with relatively high levels of exposure may also be high-paying, or women with higher incomes may be able to afford seafood that contains mercury. While the study does not allow us to determine why lower SES women had a higher risk of having SGA infants, the analyses do provide more evidence on the lasting and overwhelming impacts of lower income and educational attainment on perinatal outcomes.
WQS regression’s ability to weight predictors may be of special interest to researchers who study multiple correlated predictors, as it avoids a simplistic assumption of equal weighting provided by a simple cumulative risk score, such as our initial analysis provided. It should be noted, however, that WQS still assumes an equal increase/decrease in risk with every one-quantile change, which may be incorrect if an association is nonlinear. We found no other studies that used the WQS method to estimate the joint effect of both chemical and nonchemical exposures. Since its development, the approach has been primarily used to combine multiple chemical exposures [61
], although more recently, the approach has been used to create SES indices [82
], nutrient indices from self-reported food frequency questionnaires [11
], and an index for psychosocial stress [6
]. Though the authors of these studies chose to build models that did not combine chemical and nonchemical exposures, some authors did attempt to build WQS coefficients that combined several domains. Yorita, Christensen et al. created three WQS indices (heavy metals, co-planar polychlorinated biphenyls [PCBs], and non-dioxin-like PCBs) to predict alanine amino transferase (ALT) levels (which indicate liver damage). Czarnota et al. used one WQS index to estimate the combined effect of PCBs, polycyclic aromatic hydrocarbons (PAH), and pesticides [77
]. Risk factors across different domains are commonly correlated with each other, so using an approach that accounts for this correlation could be helpful for researchers in many fields of epidemiology. However, if exposure variables are strongly correlated with each other and weakly correlated with the outcome WQS will perform poorly [60
We conducted this analysis on a cohort of pregnant women with no previous history of very PTB or LBW infants, as well as no other pregnancy complications. Results cannot be generalized to other populations of women and are not representative of the entire population of the United States. A review of literature reveals that scientists are also currently unaware of the amounts of heavy metals that may cause adverse birth outcomes. While most women did not have blood samples that were higher than the CDC recommended limits, there is still evidence that lower levels of exposure could have adverse effects on perinatal outcomes (which was one of the objectives of this study) [25
]. As no clear cut-off point was available from the literature, we used the highest tertile to classify exposure.
The primary intention of this study was to understand whether cumulative risk, of chemical and nonchemical exposures, was associated with SGA in a healthy population of women, which included women who were obese but otherwise healthy. Investigating women who do not present prior health issues results in the exclusion of higher-risk women. If the examined risk factors are associated with intermediates, selection bias could exist; regardless, this fact limits generalizability. Studying the effects in more-exposed and/or less-healthy populations is an important next step in this research. While we chose variables that have effects on fetal growth, there may have been other variables that could be more appropriate for this analysis that were not measured for this study. Other chemical exposures, such as PCBs, air pollution, or other environmental pollutants, may interact with heavy metals, but cannot be studied in this secondary data analysis [11
]. We were similarly limited to the psychosocial stress measures studied in this analysis; we were unable to study other indicators of stress, such as childhood traumatic events or post-traumatic stress disorder (PTSD), which have also been associated with adverse birth outcomes [44
]. Each factor studied is affected by a degree of measurement error; some of them, such as stress and metal levels, may change over time as well. (Age is likely to be a susceptibility factor, but thoroughly examining this interaction would require multiple measures over time.) Combining multiple imperfectly measured variables can attenuate effects under some circumstances [88
]. As methods for combined exposure analysis develop, the degree to which errors are amplified will need to be assessed. The study also suffers from missing data, as illustrated in Figure 1
. One limitation of WQS, or any assessment studying cumulative risk, is that the many exposure variables included in the study will inevitably lead to more missing data compared to traditional studies, which only assess one exposure variable with one outcome. To the extent that missing data is informative or associated with exposure and outcome, this may bias the sample; most likely, women with complete data will be at lower risk than the overall cohort.