Preterm birth (PTB) affects 9%–10% of pregnancies in the United States, and is associated with increased risk of maternal and neonatal morbidity and mortality (1, 2). There are pronounced racial and ethnic disparities in PTB in the United States, with rates disproportionately higher in non-Hispanic Black women than non-Hispanic White women (14% vs. 9%) (3, 4). These disparities may further contribute to higher infant mortality (4) among non-Hispanic Black relative to non-Hispanic White infants. For other groups, studies showed no significant difference in PTB rate comparing Asian or Hispanic women to White women, although the risk appeared higher in certain Asian subgroups (5). Therefore, identifying the potentially modifiable risk factors of PTB, especially those that are unevenly distributed across racial and ethnic groups, is important, to help understand and reduce the disparities in PTB.

The existing literature suggests that disparities in PTB are largely attributable to environmental factors rather than genetic variation (3, 6, 7). These include social stressors, physical stressors (such as environmental chemicals and pollutants), neighborhood variation, healthcare access/quality, and individual cultural practices (8). One study suggested that certain sociodemographic and perinatal health factors contributed to the Black vs. White disparity in PTB, although they reported that more than 60% of the disparities in PTB remained unexplained (9). Other studies also showed that the Black vs. White disparities in PTB persisted after accounting for socioeconomic status, access to care, or medical interventions (10–12). For environmental pollutants, multiple studies revealed associations of air pollution, lead, phthalates, and other chemicals with increased risk of PTB, and found higher exposure levels among non-Hispanic Black women compared with the non-Hispanic White women (13–20). However, it remains unclear whether and what proportion of racial and ethnic disparities in PTB is attributable to different exposures to these environmental factors. A causal mediation analysis (21) is needed to further explore the role of multiple environmental factors on the racial and ethnic disparities in PTB.

There have been recent calls for and developments in the methodology of evaluating environmental factors as potential mediators of health disparities (22–24). Furthermore, given that people are often simultaneously exposed to multiple environmental factors (25, 26), a growing body of conceptual models and statistical methods integrating the joint effects of multiple pollutants into a mediation analysis framework has been proposed, especially in the field of environmental health disparities (27–30). These methods can help quantify the proportion of disparity due to environmental factors, as well as the proportion of disparity that would remain if interventions were made to reduce the levels of these environmental factors. Despite the discussions on this framework and the related methods, a real-world, population-based application of these methods in evaluating the contribution of environmental chemicals/pollutants as a mixture to a health disparity question remains lacking.

One class of environmental chemicals, known as polybrominated diphenyl ethers (PBDEs), has been used as a flame retardant since the 1970s and remains to be detected in the US population even a decade after the voluntary phase out that began in 2004 (31–33). PBDEs have the potential to shed or volatilize into the environment (34). Human beings are exposed to PBDEs via inhalation of contaminated air, ingestion of contaminated food, and contact with indoor dust. We hypothesize that PBDEs might be potential mediators for the racial and ethnic disparities in PTB given the following evidence: (1) multiple studies showed higher exposure levels to PBDEs among non-Hispanic Black women compared with non-Hispanic White women (35–37); (2) studies have found associations between certain PBDE congeners and elevated risk of PTB (38–42). In this study, we aimed to use real-world data from a large, multicenter, multiracial/ethnic cohort of singleton pregnancies in the United States to evaluate whether and the extent to which exposure to PBDEs may contribute to the racial and ethnic disparity in PTB and gestational age at delivery, through applying causal mediation analyses incorporating these chemicals (individually and as mixtures) as potential mediators. Race and ethnicity are socially constructed, and racial/ethnic health disparities are driven by the root cause of structural/institutional racism. With that in mind, we present a causal diagram (43) of our research questions in Figure 1.

Among the 2008 women included in the study, 552 (27.5%) self-identified as non-Hispanic White, 504 (25.1%) self-identified as non-Hispanic Black, 568 (28.3%) self-identified as Hispanic, and 384 (19.1%) self-identified as Asian/Pacific Islander (Table 1). There were several differences in characteristics across these groups (Table 1). On average, compared with non-Hispanic White women, non-Hispanic Black women were younger, had higher BMI, lower education level, and less family income, and were more likely to be unmarried. Non-Hispanic Black women also had the highest plasma cotinine level and total and sedentary activity levels compared with other groups. Hispanic women had the highest mean BMI and plasma total lipid level, the lowest percentage of being nulliparous, and the highest percentages of attaining less than a college degree or being long-term immigrants. Asian/Pacific Islander women had the highest mean age, the lowest mean BMI, plasma cotinine level, and total activity level, as well as the highest percentage of being recent immigrants. Non-Hispanic Black women had shorter mean gestational ages at delivery (39.0 vs. 39.3 weeks) and higher risks of PTB (9.1% vs. 5.1%) compared with non-Hispanic White women. The outcomes among Hispanic or Asian/Pacific Islander women were similar to those of the non-Hispanic White women.

In Table 2, non-Hispanic Black women have higher GMs of all six PBDE congeners and ∑PBDEs than non-Hispanic White women. Across all groups, Hispanic and Asian/Pacific Islander women had the highest GMs of PBDE 28 and PBDE 154, respectively. When comparing those with vs. without PTB in the study population, four PBDEs (i.e., PBDE 28, 47, 100, and 153) and ∑PBDEs had higher GMs.

We used a WQS index to estimate the mixture effect of six PBDEs on gestational age at delivery (weights for each PBDE shown in Figure 2). The association of a 1-unit increase in the WQS index with gestational age at delivery was β (95% CI) = −0.20 (−0.35, −0.05) weeks, adjusted for race and ethnicity and other covariates. Table 3 provides results from the mediation analysis, where the WQS index was considered a potential mediator for the racial and ethnic disparity of gestational age at delivery or PTB. Comparing non-Hispanic Black women with non-Hispanic White women, the covariate-adjusted difference in gestational age at delivery was β_TE (95% CI) = −0.30 (−0.54, −0.05) weeks. The CDEs (95% CIs) when fixing everyone's WQS index levels at the 25th, 50th, and 75th percentiles were −0.18 (−0.46, 0.10), −0.23 (−0.49, 0.02), and −0.32 (−0.57, −0.07) weeks, respectively. Overall, a suggestive NIE of β_NIE (95% CI) = −0.04 (−0.07, 0.00) weeks were mediated through the WQS index (proportion mediated = 11.8%). The odds ratio of PTB comparing non-Hispanic Black with non-Hispanic White women was OR_TE (95% CI) = 1.82 (1.00, 3.31), yet no statistically significant NIE was found. In addition, no statistically significant disparity was found when comparing Hispanic or Asian/Pacific Islander women with non-Hispanic White women.

The correlation coefficients between PBDEs are shown in Supplementary Figure S1. PBDE 28, 47, 99, and 100 were moderately to highly correlated, and PBDE 153 and 154 were weakly correlated. Using BKMR with hierarchical variable selection (based on the correlation structure, PBDE 28, 47, 99, and 100 were assigned as Group 1, and PBDE 153 and 154 were assigned as Group 2), we found that Group 2 was of relatively greater importance, and PBDE 153 was the most important chemical within Group 2 [reflected by the posterior inclusion probabilities (PIPs) shown in Supplementary Table S1] that was associated with shorter gestational age at delivery (Supplementary Figure S2). There were no qualitative interactions between the PBDEs (Supplementary Figure S3). Thus, for the single-mediator model, we included PBDE 153 as the mediator. Table 4 provides results from the mediation analysis, where only PBDE 153 was considered as a potential mediator. The CDEs (95% CIs) when fixing everyone's PBDE 153 levels at the 25th, 50th, and 75th percentiles were −0.18 (−0.45, 0.09), −0.18 (−0.45, 0.09), and −0.29 (−0.54, −0.05) weeks, respectively, yet with a non-significant NIE mediated via PBDE 153 (proportion mediated = 7.9%). No statistically significant NIE was found for the non-Hispanic Black vs. non-Hispanic White disparity in PTB.

Table 5 provides results when all six PBDEs were included simultaneously as multiple mediators. The CDEs (95% CIs) when fixing all PBDEs at the 25th, 50th, and 75th percentiles were −0.25 (−0.53, 0.02), −0.14 (−0.47, 0.17), and −0.27 (−0.61, −0.03) weeks, respectively, yet with a non-significant NIE jointly mediated via PBDE 28, 46, 99, 100, 153, and 154 (proportion mediated = 16.3%). No statistically significant NIE was found for the disparity in PTB. No exposure–mediator(s) or mediator–mediator interaction was found for any of the aforementioned analyses (p-values for interactions >0.05).

As secondary analysis, the outcomes stratified by further specified Hispanic origin or Asian background are provided in Supplementary Table S2. Several subgroups had shorter mean gestational ages at delivery than non-Hispanic White participants. Among them, those reporting Filipino background also had higher mean WQS index levels than non-Hispanic White women (Supplementary Figure S4). Mediation analysis results comparing Filipino to non-Hispanic White women are provided in Supplementary Table S3. The proportion mediated by WQS index for the shorter gestational age at delivery was 16% (95% CI: −11%, 61%) when comparing the n = 45 women with Filipino background with non-Hispanic White women.

The results from the sensitivity analyses to evaluate the robustness of our main results are shown in the Supplementary Materials. In summary, evaluating PTB in subcategories showed no significant findings (Supplementary Table S4), and evaluating the absolute risk of PTB on the risk difference scale showed similar findings (Supplementary Table S5). Using g-estimation yielded similar patterns of mediation (Supplementary Table S6). Modeling the WQS index as a binary mediator dichotomized at the median had very little impact on the indirect effect estimates (Supplementary Table S7). The associations between PBDE 153 or the WQS index and gestational age at delivery were linear (p-trend < 0.05, Supplementary Table S8), and modeling WQS or PBDE 153 as quantiles showed similar mediation effects (Supplementary Tables S9, S10). Using a weighted exposure index of 4 PBDEs (PBDE 28,99, 100, and 153) based on quantile g-computation analysis (selecting the PBDEs with weights toward an association with shortened gestational age at delivery, Supplementary Figure S5) yielded similar mediation patterns as the WQS index (Supplementary Table S11). Finally, Supplementary Figure S6 showed similar estimates across various magnitudes of potential mediator measurement error.

In this multiracial/ethnic cohort of pregnant women in the United States, we demonstrated shorter gestational age at delivery, higher risk of PTB, and higher exposure levels to PBDEs among non-Hispanic Black than in non-Hispanic White women, and we evaluated potential mediation by PBDEs for the racial and ethnic disparities in gestational age at delivery and PTB utilizing several recently developed causal mediation approaches. In particular, we observed that a weighted index summarizing PBDEs as a mixture had a suggestive mediating role in the Black vs. White disparity in gestational age at delivery that accounted for 11.8% of the total disparity. No significant mediation was found for the disparity of PTB, or from evaluating other forms of PBDE mediators. We also revealed disparities in gestational age at delivery comparing the Filipino subgroup with non-Hispanic White women, although no significant mediation via PBDEs was found. While further validation using larger datasets are needed, our results point to the possibility of PBDE mixtures acting as mediators for the existing racial and ethnic disparities in gestational age at delivery.

Our observation of a higher risk of PTB and shorter mean gestational age at delivery among non-Hispanic Black women compared with non-Hispanic White women is consistent with previous reports (4, 9, 67). However, the PTB risks were lower than the general population since this study consisted of relatively healthy, non-obese individuals. We also observed higher average concentrations of PBDEs comparing non-Hispanic Black with non-Hispanic White women, which aligned with previous studies that reported similar disparity patterns among certain PBDEs (e.g., 28, 47, 99, and 100) (36, 68). This disparity may be explained by differences in social and contextual factors contributing to sources of PBDE exposures, such as differences in residential neighborhoods (69), housing (e.g., indoor dust) (70), and furniture PBDE exposures (68). Furthermore, we observed higher mean levels of four PBDE congeners and ∑PBDEs in those who delivered preterm compared with non-preterm, suggesting that PBDEs, either individually or as mixtures, might be potential mediator(s) accounting for part of the disparities in PTB or gestational age at delivery. In our causal mediation analyses, we found that a weighted index of all PBDEs (i.e., the WQS index) accounted for 11.8% of the total Black vs. White disparity in gestational age at delivery. Particularly, we found that the CDEs were closer to the null when fixing everyone's WQS index at lower levels, suggesting the potential benefit in reducing the existing disparity in gestational age at delivery by intervening on PBDE levels in the entire population. Conversely, the proportion mediated by the WQS index for the Black vs. White disparity in PTB was only 4% (and non-significant), which could be explained either by lower statistical power owing to a limited number of events, or that the magnitude of mediation for gestational age at delivery might be relatively small to make a noticeable impact on the risk of PTB in this healthier population. Given this is the first study that evaluated the potential mediation role of PBDEs for this disparity, future studies of larger sample sizes or conducted among a higher-risk population might be needed to validate our findings. Past studies have revealed other mediators (such as socioeconomic and health factors, and access to healthcare) for the racial and ethnic disparity in PTB, but a large proportion of the disparity remained (9–11, 71–73). If PBDEs truly mediate part of the racial and ethnic disparity in length of gestation or PTB, then this class of chemicals might be an additional modifiable factor to help further alleviate this disparity.

Similarly to previous literature (5), we did not observe significant differences in gestational age at delivery or PTB comparing Hispanic or Asian/Pacific Islander women with non-Hispanic women. However, we did find a 38 per 1,000 births higher risk of PTB and 0.5-week shorter mean gestational age at delivery comparing a subgroup of Filipino women with non-Hispanic White women, which was consistent with previous studies showing that Filipino women had higher relative risk of PTB (compared with non-Hispanic White) than other Asian subgroups (74). Despite these Filipino women also having higher exposure levels to PBDEs, the results from mediation analysis were non-significant. This might be due to the small number of participants with various Asian backgrounds, although we could not rule out the possibility that there might be unmeasured confounding such as cultural, psychosocial, or early life factors that are driving this disparity, especially when more than half of the Filipino women in this study were immigrants. Future studies with more specific focus on these racial/ethnic minority subgroups that collect acculturation-related variables are needed to further explore this mediation.

In this study, we compared three different approaches of incorporating PBDEs as potential mediators of racial and ethnic disparities in gestational age at delivery and PTB: reducing to a WQS index, selecting a single PBDE 153 congener according to prior knowledge and its relative importance from the hierarchical BKMR selection, and including six PBDE congeners as multiple mediators. Overall, the estimated proportion mediated via the single PBDE 153 congener was smaller than that via the WQS index. This is possibly owing to the limitation of selecting mediator(s) a priori based on the mediator-outcome association alone, which might leave out important mediator(s) weakly associated with the outcome that may also contribute to the indirect effect. The estimated proportion jointly mediated by multiple PBDEs was higher than the proportion mediated by the WQS index, but with much wider CIs due to potential overfitting or multicollinearity. Our example showed that the WQS approach carries the advantage of reducing the PBDEs to a single score to avoid overfitting or multicollinearity, while preserving the information from each congener, serving as a suitable approach to explore the overall contribution of a chemical mixture to a health disparity question (28).

We acknowledge several limitations of this study. First, this study consisted of women with low-risk of adverse health outcomes at baseline and without obesity, so our findings might not be fully generalizable to the overall US population. Second, unmeasured confounding was inevitable, such as other geographic, psychosocial, or lifestyle factors. Third, statistical power was limited when evaluating potential mediation within certain subgroups. Lastly, we were not able to directly measure historical or contemporary environmental racism or adverse environments in these data that are contributing to (or on the causal pathway for) the observed disparities where race and ethnicity act as a proxy for these complex processes (75). Further studies are needed to inform interventions on the policies and systems level.

This study has many unique strengths. First, this study utilized a prospective cohort design in a large, racially/ethnically diverse population with clinically validated outcomes and a comprehensive set of covariates. Second, we applied different statistical approaches to evaluate mediation(s) through individual as well as mixtures of PBDEs. Third, efforts were made to evaluate mediation for disparities in subcategories of PTB, or among other under-studied racial and ethnic subgroups (e.g., based on Asian backgrounds). Lastly, we conducted various sensitivity analyses to validate the robustness of our findings.

In conclusion, in this multiracial/ethnic cohort of pregnant women in the United States, we found that non-Hispanic Black women had shorter gestational ages at delivery, higher risk of PTB, and higher exposures to PBDEs compared with non-Hispanic White women. Our mediation analysis provided suggestive evidence that the Black vs. White disparity in gestational age at delivery might be partially mediated by disparities in exposures to PBDEs. Lowering the PBDE exposures at the population level may help reduce this disparity.