Correlates of non-persistent endocrine disrupting chemical mixtures among reproductive-aged Black women in Detroit, Michigan.

SELF is a prospective cohort study of reproductive-aged Black women residing in the Detroit-metropolitan area that was designed to assess associations of environmental and lifestyle factors with uterine leiomyomata (UL). Full details of SELF recruitment have been described previously ( Baird et al., 2015 ). In brief, women (n= 1,693) were eligible for recruitment during 2010–2012 if they had an intact uterus, were 23–35 years of age, self-identified as Black or African American, and had no prior diagnosis of UL, autoimmune disease or cancer that required regular medication use. Enrollment for pregnant women was delayed until 3 months postpartum. The current analysis used baseline data from a case-cohort substudy originally designed to examine associations between EDCs and ULs. This substudy included a random sample of 592 women selected at baseline who were free of UL, as well as 162 incident cases that accrued during 60 months of follow-up that were not part of the random baseline sample. We excluded 3 women who did not have measurements for at least one non-persistent EDC biomarker for a final sample size of 751 women ( Figure S1 ). The SELF protocol was approved by Institutional Review Boards at the Henry Ford Health System (HFHS), National Institute of Environmental Health Sciences (NIEHS), and Boston University. The involvement of the Centers for Disease Control and Prevention (CDC) did not constitute engagement in human subjects’ research. Urine samples were collected during baseline study visits at the HFHS and shipped overnight on dry ice to the NIEHS for storage in the biorepository at −80°C. Urine samples were then shipped overnight on dry ice to the CDC for quantification of 16 phthalates, 7 phenols, 4 parabens, and triclocarban using online solid phase extraction coupled with high performance liquid chromatography isotope dilution tandem mass spectrometry ( Silva et al., 2007 ; Ye et al., 2006 , 2005 ). Phthalates included mono-n-butyl phthalate (MBP), mono-hydroxybutyl phthalate (MHBP), mono-isobutyl phthalate (MiBP), mono-hydroxyisobutyl phthalate (MHiBP), MEP, monobenzyl phthalate (MBzP), mono-carboxyisononyl phthalate (MCNP), mono-isononyl phthalate (MNP), mono-carboxyisooctyl phthalate (MCOP), mono-3-carboxypropyl phthalate (MCPP), mono(2-ethyl-5-carboxypentyl) phthalate (MECPP), mono(2-ethyl-5-hydroxyhexyl) phthalate (MEHHP), mono(2-ethylhexyl) phthalate (MEHP), and mono(2-ethyl-5-oxohexyl) phthalate (MEOHP). We also quantified concentrations of two metabolites of phthalate alternatives: 1,2-cyclohexane dicarboxylic acid-monohydroxy isononyl ester (MHiNCH), and 1,2-cyclohexane dicarboxylic acid-monocarboxy isooctyl ester (MCOCH). Phenols included BPA, bisphenol F (BPF), bisphenol S (BPS), benzophenone-3 (BP3), 2,4-dichlorophenol, 2,5-dichlorophenol, and triclosan. Parabens included methyl paraben, ethyl paraben, propyl paraben, and butyl paraben. We classified low molecular weight phthalate metabolites as those deriving from phthalate esters with side alkyl chain lengths of 1 to 4 carbons ( National Research Council (US) Committee on the Health Risks of Phthalates, 2008 ; Wang et al., 2019 ). The parent compounds included di-n-butyl phthalate (DnBP, metabolites: MBP and MHBP), di-isobutyl phthalate (DiBP, metabolites: MiBP and MHiBP), and diethyl phthalate (DEP, metabolite: MEP). High molecular weight parent compounds, with ester side alkyl chains of ≥5 carbons ( National Research Council (US) Committee on the Health Risks of Phthalates, 2008 ; Wang et al., 2019 ), include benzylbutyl phthalate (BzBP, metabolite: MBzP), diisodecyl phthalate (metabolite: MCNP), diisononyl phthalate (DiNP, metabolites: MCOP and MNP), dioctyl phthalate (metabolite: MCPP, also a metabolite of other high molecular weight phthalates), and di(2-ethylhexyl) phthalate (DEHP, metabolites: MECPP, MEHHP, MEHP, and MEOHP). Limits of detection ranged from 0.2 – 1.2 ng/mL for phthalates, 0.1 – 1.7 ng/mL for phenols, and 0.1 – 1.0 ng/mL for parabens ( Table 1 ). Creatinine concentrations in urine were measured at the NIEHS using Beckman Coulter clinical analyzer AU400e. Quality assurance/quality control methods have been previously described in detail ( Bethea et al., 2020 ; Wesselink et al., 2020 ). Data on correlates were collected at baseline through interviews and self-administered questionnaires. Participants also reported their intake frequency and serving size of 110 foods in the past 12-months on a food frequency questionnaire ( BLOCK et al., 1986 ). We selected potential correlates for this analysis a priori based on the literature and on prior findings in the SELF cohort ( Bethea et al., 2020 ; Wesselink et al., 2020 ). Demographic correlates included age (years, continuous), education (≤high school diploma or GED, some college/Associate’s degree/technical degree, or ≥Bachelor’s degree), annual household income ($50,000), employment status (unemployed, employed <30 hours/week, or employed ≥30 hours/week), and marital status (never married, previously married, or currently married). Categorical cutoffs for demographic variables were based on distributions of the data to ensure adequate sample sizes within each category. Behavioral and anthropometric correlates included smoking status (never, former, current <10 cigarettes/day, or current ≥10 cigarettes/day), alcohol consumption (low [<10 drinks in last year], moderate [more than low, but not heavy], or heavy [≥6 drinks on a day when drinking alcohol or ≥4 drinks at a single sitting twice per month or more]), and body mass index (BMI, kg/m 2 ), which was calculated from height and weight measured at baseline (continuous). Reproductive correlates included age at menarche (≤10, 11, 12, 13, ≥14 years), parity (continuous), ever use of oral contraceptives (yes or no), ever use of progestin-only injectable (yes or no), current hormonal contraceptive use (yes or no), and current use of vaginal ring, a form of hormonal contraception worn inside the vagina (yes or no). Dietary correlates included daily servings of beef, pork or lamb (continuous), daily servings of dairy (milk equivalent servings, continuous), drank water from a plastic bottle in the past 24-hours (yes or no), and eaten food from a can in the past 24-hours (yes or no). Product use data included the use of perfume, creams or lotion before bed, nail polish, make-up (foundation, blush, eye make-up, lipstick, lip gloss, chapstick, lip balm, Vaseline), vaginal douche, vaginal powder, vaginal deodorant sprays or wipes, or condoms in the past 24 hours (yes or no), as well as the frequency of sunscreen use at baseline (never/hardly ever, sometimes, often, always/nearly always). We first assessed the detection frequencies and distributions of all EDC biomarkers. Biomarkers with a detection frequency <65% were excluded from further analyses. Biomarker concentrations below the limit of detection (LOD) were imputed using the LOD/√2 ( Hornung and Reed, 1990 ). We adjusted for urine dilution by dividing biomarker EDC concentrations by creatinine concentrations. Summary statistics were calculated for all EDC biomarkers ( Table 1 ). We provided mean concentrations of each biomarker from the 2011–2012 National Health and Nutrition Examination Survey (NHANES) for Black Americans and white Americans for comparison with our study population. Some biomarkers were not measured in NHANES in the 2011–2012 cycle; for these biomarkers, we used the earliest possible cycle where the biomarkers were measured (2013–2014 or 2014–2015). We also calculated Spearman correlation coefficients between EDC biomarkers measured in SELF ( Figure S2 ). All EDC biomarkers were natural log-transformed and Z-standardized for the mixtures analysis. We used two dimension reduction techniques, principal component analysis (PCA) and k -means clustering, which allowed us to compare the robustness of our findings across two commonly-used mixtures approaches. We have previously described both of these methods in detail ( Schildroth et al., 2021 ). Briefly, PCA linearly transforms data into uncorrelated principal components (PCs) that each explain a percentage of variance in the data, where the first PC explains the most variance. PCA calculates loadings that describe correlations between PCs and biomarker concentrations, and the loadings are used to calculate PC scores for each subject ( Jolliffe and Cadima, 2016 ). We used the PC scores, which represent EDC biomarker mixture profiles, as the outcome variable in subsequent regression models. We a priori selected the number of PCs that explained ~50% of the variance in the data for analysis. K -means clustering uses Euclidean geometry to assign participants with similar EDC exposure biomarker profiles to clusters. We examined a range of clusters (2, 3, 4, and 5 clusters), while ensuring each cluster had a reasonable sample size. We used twenty-three selection indices to determine the optimal number of clusters for this dataset ( Charrad et al., 2014 ; “NbClust function | R Documentation,” n.d. ). Cluster membership was used as the outcome variable in subsequent regression models. We fit multivariable linear regression and multinomial logistic regression models to estimate associations between our a priori selected correlates and EDC biomarker mixture profiles identified using PCA and k -means clustering, respectively. Associations between correlates and PCs were estimated as the mean difference (β) in PC scores across categories of the correlate. The interpretation of the direction of the β coefficient is dependent on the direction of the loading; for example, a negative β coefficient for a given PC would indicate the correlate is associated with lower scores on that PC, which represents higher concentrations of chemical biomarkers that loaded negatively onto the PC. Similarly, a positive β coefficient would indicate that the correlate is associated with higher PC scores, and thus higher concentrations of chemical biomarkers that loaded positively onto the PC. Associations between correlates and cluster membership were estimated as the odds ratio (OR) of cluster membership. Ninety-five percent confidence intervals (CIs) were calculated for all measures of association. Betas and ORs were adjusted for all other correlates, an approach used previously in the SELF cohort ( Bethea et al., 2020 ; Schildroth et al., 2021 ; Wesselink et al., 2020 ). There were few missing data in SELF (<3% for correlates and <1% for EDCs). We imputed missing correlate and EDC biomarker data using the Markov chain Monte Carlo method ( Zhou et al., 2001 ). We generated 15 datasets using all available EDC biomarker and correlate data. We are not aware of any current method that allows for the statistical combination of imputed datasets for PCA or k -means clustering. Therefore, as we and others have done previously in similar applications of mixtures data ( Kalloo et al., 2018 ; Schildroth et al., 2021 ), we averaged the imputed values across all imputed datasets and used the averaged data as our analytic dataset. We conducted four sets of sensitivity analyses. First, we re-ran our linear regression analyses using the number of PCs that explained at least an additional 5% of variance in the data per PC following the initial 50% variance cutoff. Similarly, we re-ran our multinomial logistic regression models using the numbers of clusters that were not identified as the optimal number of clusters. Second, we ran minimally adjusted models that included only a select number of correlates (age, education, BMI, current use of vaginal ring, and nail polish use in the past 24-hours) to ensure we did not over-adjust models or introduce collider bias. Correlates for these models were selected based on our prior findings in SELF ( Bethea et al., 2020 ; Wesselink et al., 2020 ). Third, we re-ran our main analyses restricting to women who were included in the original randomly selected sub-cohort at baseline (n= 592) to ensure we did not induce selection bias by including incident UL cases. Lastly, we identified influential biomarker concentrations using Rosner’s generalized extreme Studentized deviate method ( Rosner, 1983 ) and excluded these observations, as both PCA and k -means clustering are sensitive to extreme values ( Jolliffe, 2006 ; Raykov et al., 2016 ).

The mean age of participants was 28.5 years (SD: 3.5 years). Twenty-nine percent were currently married, 78.4% had some college/technical education or higher, 60.6% were employed at least part-time, and 54.7% had a household income of at least $20,000 per year. Seventy-three percent of subjects were non-smokers, and the majority (51.1%) of women reported moderate alcohol consumption. The average BMI was 33.7 km/m 2 . The majority had used perfume (65.0%) and make-up (73.9%) in the past 24 hours, while few had used lotion or creams (34.5%), nail polish (9.6%), vaginal douche (2.3%), vaginal powder (12.9%), vaginal deodorant (10.0%), or condoms (4.8%) in the past 24 hours. Fifteen percent were currently using a vaginal ring, and most women never or hardly ever (74.8%) wore sunscreen at baseline ( Table S1 ). All EDC biomarkers were detected in at least 65% of samples, except for MHiNCH (24.0%), MCOCH (11.5%), and butyl paraben (57.5%), which were excluded from further analyses ( Table 1 ). The phenols with the highest median concentrations were BP3 (13.3 μg/g creatinine), triclosan (7.2 μg/g creatinine), and 2,5-dichlorophenol (2.4 μg/g creatinine). The phthalate metabolites with the highest median concentrations were MEP (59.5 μg/g creatinine), MCOP (19.9 μg/g creatinine), and MBP (16 μg/g creatinine), and the highest median concentration for parabens was observed for methyl paraben (116.1 μg/g creatinine). The median concentration of triclocarban was 0.2 μg/g creatinine (range: 0.1, 291.1 μg/g creatinine). SELF participants tended to have higher median concentrations of phthalates, including metabolites of DEHP, compared to mean concentrations among Black and white Americans in NHANES ( Table 1 ). Black women in SELF and Black Americans in NHANES both had higher concentrations of parabens compared with white Americans, while white Americans had the highest concentrations for several phenols, including BP3, BPA, BPF, and triclosan. Spearman correlations for EDC biomarkers among SELF participants tended to be higher for biomarkers within the same class ( Figure S2 ). The highest correlation among the phenols was between 2,4-dichlorophenol and 2,5-dichlorophenol (0.50). For phthalates, the highest correlations were observed between metabolites of the same parent compound: 0.85 for MBP and MHBP (metabolites of DnBP), 0.76 for MCOP and MNP (metabolites of DiNP), 0.85 for MHiBP and MiBP (metabolites of DiBP), and 0.87 for MECPP and MEHHP, 0.88 for MECPP and MEOHP, and 0.94 for MEHHP and MEOHP (metabolites of DEHP). We found moderate correlations (0.15 – 0.30) of BPA, BPS, and BPF with several phthalate metabolites, suggesting a common exposure source for these chemicals (e.g., plastics, diet). The correlations between the parabens ranged between 0.31 and 0.80, but parabens were not strongly correlated with other EDC biomarkers. Triclocarban was not strongly associated with any of the other EDC biomarkers. Ten of the 23 k -means selection indices identified 3 clusters as the optimal number of clusters. The first cluster (N= 232) was characterized by the lowest biomarker concentrations, except for BPS, and was set as the reference cluster. The second cluster was characterized by the highest concentrations of the parabens, two phenols (BPF and triclosan), metabolites of low molecular weight phthalates (MBP, MEP, MiBP and MHBP), and one metabolite of high molecular weight phthalates (MBzP). Cluster 3 was characterized by the highest concentrations for metabolites of high molecular weight phthalates (MCNP, MCOP, MNP, MCPP, MECP, MEHHP, MEHP, MEOHP), triclocarban, and most phenols (except BPF and triclosan, Figure 1 ). When we examined associations between socioeconomic correlates and cluster membership, income and employment were associated with higher odds of membership to both cluster 2 and cluster 3 compared with cluster 1, indicating income and employment were associated with higher concentrations of all EDC biomarkers ( Table 2 ). These associations were generally stronger for women in cluster 3, which indicates that income and employment are more strongly associated with biomarkers of high molecular weight phthalates, most phenols, and triclocarban. Women who smoked (current, <10 cigarettes/day: OR= 2.25, 95% CI= 1.11, 4.55, vs. non-smokers) or had a higher BMI (per 1-unit increase, OR= 1.03, 95% CI= 1.01, 1.05) were also more likely to belong to cluster 3. Consumption of dairy in the last year was associated with decreased odds of membership to cluster 2 (OR= 0.81, 95% CI= 0.64, 1.04), suggesting decreased exposure to low molecular weight phthalates, parabens, and some phenols. Current use of a vaginal ring was associated with increased membership to both clusters, particularly to cluster 3 (OR= 2.42, 95% CI= 1.28, 4.59), reflecting stronger associations with higher molecular weight phthalates, most phenols, and triclocarban. Though imprecise, there was also evidence that ever use of oral contraceptives was associated with increased odds of membership to cluster 2 (OR=1.27, 95% CI= 0.84, 1.94). Nail polish use in the past 24 hours was associated with increased odds of membership to cluster 2 (OR= 2.44, 95% CI= 1.22, 4.87) and cluster 3 (OR= 2.26, 95% CI= 1.06, 4.84). Similarly, condom use in the past 24 hours was associated with membership to both clusters, with a stronger association in cluster 2 (OR= 4.91, 95% CI= 1.61, 15.01) than in cluster 3 (OR= 2.21, 95% CI= 0.60, 8.22), suggesting stronger associations with low molecular weight phthalates, parabens, and some phenols. Use of vaginal deodorant in the past 24-hours was associated with both cluster 2 (OR= 1.60, 95% CI= 0.82, 3.13) and cluster 3 (OR= 2.38, 95% CI= 1.16, 4.89) membership. We observed a non-monotonic association between frequency of sunscreen use at baseline and cluster membership ( Table 2 ). Four PCs explained ~50% of the variance in the data (PC-1: 21%, PC-2: 12%, PC-3: 9%, PC-4: 7%). We characterized PC-1 as indicative of exposure to all phthalates and BPA, whose biomarkers loaded negatively onto PC-1 ( Figure 2 ). The pattern of biomarkers that loaded onto PC-2 was similar to what we observed in our k -means clustering analysis. Namely, PC-2 was indicative of higher concentrations of phthalates, where metabolites of low molecular weight phthalates (MEP, MBP, MHBP, MiBP and MHiBP) loaded positively onto the PC, while metabolites of high molecular weight phthalates (MCPP, MECPP, MEHP, MNP, MCOP, and MCNP) loaded negatively onto the PC, indicating that they contribute to the component in opposite directions. The exception was MBzP, a metabolite of the high molecular weight phthalate BzBP, that loaded with metabolites of the low molecular weight phthalates. PC-3 was indicative of exposure to parabens, three phenols or their precursors (BP3, triclosan and 2,4-dichlorophenol), and MEP, which loaded positively onto the PC, as well as three additional phthalate metabolites (MBP, MHBP and MBzP) that loaded negatively onto the PC. Lastly, PC-4 was indicative of exposure to parabens, which loaded positively onto the PC, and three phenols or their precursors (2,4-dichlorophenol, 2,5-dichlorophenol, triclosan), which loaded negatively onto the PC. Similar to findings from k -means clustering, we found that employment was associated with increased exposure to high molecular weight phthalates in PC-2 (compared to women not employed, <30 hours/week: β= −0.38, 95% CI= −0.79, 0.03; ≥30 hours/week: β= −0.47, 95% CI= −0.78, −0.16; Table 3 ). Employment was also associated with higher concentrations of parabens in PC-3 (<30 hours/week: β= 0.39, 95% CI= 0.06, 0.72) and PC-4 (<30 hours/week: β= 0.22, 0.08, 0.53; ≥30 hours/week: β= 0.14, 95% CI= −0.09, 0.38). As with k -means clustering, smoking was associated with increased biomarker concentrations of triclosan and the dichlorophenols (or their precursors) that loaded on PC-4 (compared to never smokers, former smokers: β= −0.42, 95% CI= −0.78, −0.05). Compared with nulliparous women, parous women consistently had higher biomarker concentrations of some phenols in PC-4 (1 birth: β= 0.30, 95% CI= 0.04, 0.56; 2 births: β= 0.26, 95% CI= −0.05, 0.57; ≥3 births: β= 0.33, 95% CI= 0.00, 0.66). Eating food from a can in the past 24 hours was modestly associated with BPA and a mixture of phthalates in PC-1 (β= −0.27, 95% CI= −0.64, 0.11). Nail polish use in the past 24 hours (β= −0.57, 95% CI= −1.13, −0.01), vaginal deodorant use in the past 24 hours (β= −0.56, 95% CI= −1.12, 0.01), and average frequency of sunscreen use at baseline (often: β= −0.74, 95% CI= −1.50, 0.02) were all associated with higher biomarker concentrations of a mixture of phthalates and BPA in PC-1. In PC-2, perfume use in the past 24-hours (β= −0.18, 95% CI= −0.46, 0.09) was associated with biomarkers of high molecular weight phthalates, while nail polish use in the past 24-hours was associated with higher concentrations of MBzP and to low molecular weight phthalates (β= 0.34, 95% CI= −0.09, 0.76). PC-3 scores, reflecting a mixture of parabens, some phenols, and MEP, were associated with perfume use (β= 0.19, 95% CI= −0.03, 0.41), lotion use before bed (β= 0.36, 95% CI= 0.14, 0.58), make-up use (β= 0.52, 95% CI= 0.27, 0.76), and vaginal powder use (β= 0.39, 95% CI= 0.08, 0.70) in the past 24 hours, as well as average frequency of sunscreen use at baseline (sometimes: β= 0.33, 95% CI= 0.03, 0.64; always/nearly always: β= 0.31, 95% CI= −0.13, 0.75). Vaginal deodorant use (β= −0.32, 95% CI= −0.60, −0.03) and always/nearly always use of sunscreen at baseline (β= −0.36, 95% CI= −0.76, 0.05) in the last year were also associated with some phenols in PC-4. Current use of a vaginal ring was associated with PC-1 (β= −0.99, 95% CI= −1.50, −0.47), PC-2 (β= −0.32, 95% CI= −0.71, 0.07), PC-3 (β= 0.32, 95% CI= 0.01, 0.64), and PC-4 (β= 0.45, 95% CI= 0.16, 0.75), reflecting exposure to most EDCs, but especially phthalates and parabens. Similarly, ever use of oral contraceptives was associated with PC-3 (β= 0.22, 95% CI= −0.02, 0.46), which was indicative of higher biomarker concentrations of MEP, parabens, and some phenols. These findings were generally consistent with our findings from k -means clustering. In sensitivity analyses with additional PCs, we found that PC-5 and PC-6 each explained an additional 5% of variance in the data. Findings for the first 4 PCs were similar to our main results in which we restricted to 4 PCs ( Figure S3 , Table S2 ). PC-5 was indicative of exposure to DEHP, while PC-6 was characterized by exposure to phenols or their precursors. Similar to our main findings, current use of the vaginal ring was associated with higher concentrations of DEHP metabolites in PC-5, as well as with phenols in PC-6. Results from sensitivity analyses using alternate numbers of clusters are reported in Figure S4 – S6 and Table S3 – S5 . Of note, current use of a vaginal ring (OR= 1.76, 95% CI= 1.09, 2.85), recent use of vaginal deodorant (OR= 1.64, 95% CI= 0.96, 2.77), and recent use of condoms (OR= 2.01, 95% CI= 0.96, 4.20) were associated with increased odds of membership to a cluster with higher concentrations of all EDC biomarkers ( Table S3 ). Our results were similar when we restricted the sample to the random sub-cohort selected at baseline (n= 592; Figures S7 – S8 , Tables S6 – S7 ), though we found a much stronger association for current vaginal ring use in k -means clustering analyses (cluster 2: OR= 2.10, 95% CI= 1.01, 4.37; cluster 3: OR= 3.28, 95% CI= 1.53, 7.05). Findings from minimally adjusted models were equivalent to our main findings for age, BMI, and product use for both PCA and k -means clustering, though we did find stronger associations for current use of vaginal ring using both methods ( Tables S8 – S9 ). Our results were comparable to main models after removing influential biomarker concentrations, though findings for some correlates were attenuated ( Figures S9 – S10 , Tables S10 – S11 ).

We identified several correlates that were associated with profiles of biomarkers of non-persistent EDC mixtures in Black women. Employment and income were associated with all EDC biomarkers. Current use of vaginal ring was similarly associated with increased biomarker concentrations of phthalates and parabens, while smoking was associated with increased concentrations of phenols. While eating food from a can in the past 24 hours was modestly associated with increased biomarker concentrations of phthalate metabolites and BPA, increased consumption of dairy (milk, yogurt, cheese) was associated with decreased urinary concentrations of biomarkers of all EDCs, especially metabolites of low molecular weight phthalates, parabens, and two phenols (BPF and triclosan). Use of several products in the past 24 hours was associated with several profiles of exposure biomarkers of non-persistent EDCs, especially phthalate metabolites. Perfume use was associated with biomarkers of a mixture of high molecular weight phthalate metabolites, as well as with a mixture that contained parabens and MEP, a metabolite of diethyl phthalate (DEP). DEP, high molecular weight phthalates, and parabens have been detected in perfumes ( Al-Saleh and Elkhatib, 2016 ; Guo and Kannan, 2013 ). Our findings are also consistent with prior studies in women that identified perfume as a correlate of exposure to parabens and the metabolites of DEP (MEP) and DiBP (MiBP), when assessed individually ( Ashrap et al., 2018 ; Braun et al., 2013 ; Cantonwine et al., 2014 ; Parlett et al., 2012 ; Wesselink et al., 2020 ). We also identified nail polish use in the past 24-hours as a correlate of exposure to parabens and phthalates, especially low molecular weight phthalates. Metabolites of DEP (MEP) and dibutyl phthalate (MHBP and MBP) have been detected in nail polishes ( Food and Drug Administration, n.d. ; Guo and Kannan, 2013 ), and nail polish use was previously associated with concentrations of low molecular weight phthalate metabolites, including MEP, MHBP, and MBP, as well as with parabens, MBP, and MEP ( Ashrap et al., 2018 ; Braun et al., 2013 ; Buckley et al., 2012 ; Cantonwine et al., 2014 ; Parlett et al., 2012 ; Wesselink et al., 2020 ). Using make-up and lotion before bed in the past 24 hours was associated with biomarkers of a mixture of parabens, phenols (BP3, 2,4-dichlorophenol, triclosan) and MEP in our study. Multiple studies, including SELF, have reported associations of make-up and lotion use with urinary concentrations of MEP ( Braun et al., 2013 ; Parlett et al., 2012 ; Wesselink et al., 2020 ), while some have reported associations with MiBP, MBP, parabens, triclosan, and BP3 ( Ashrap et al., 2018 ; Braun et al., 2013 ; Meeker et al., 2013 ; Parlett et al., 2012 ; Sandanger et al., 2011 ; Wesselink et al., 2020 ). These findings are consistent with known uses of certain phthalates, triclosan, and parabens in these products ( European Union, n.d. ; Food and Drug Administration, n.d. ; Guo and Kannan, 2013 ). Women who used sunscreen at baseline also had biomarker profiles suggestive of mixtures containing phthalates, BPA, parabens, and phenols (i.e., triclosan, BP3, 2,4-dichlorophenol, and 2,5-dichlorophenol). Prior studies have similarly identified positive associations between sunscreen use and biomarkers of exposure such as BP3, parabens, MBP (parent compound: DnBP), and 2,5-dichlorophenol ( Ashrap et al., 2018 ; Bethea et al., 2020 ; Ferguson et al., 2017 ; Meeker et al., 2013 ; Philippat et al., 2015 ; Zamoiski et al., 2015 ), reflecting known uses of BP3, BPA, and parabens in sunscreen products ( Dodson et al., 2012 ; Guo and Kannan, 2013 ; Yap et al., 2017 ). Diet, including consumption of dairy, was associated with decreased concentrations of a mixture of MEP, phenols, and parabens. This is similar to other studies that reported associations between milk, cheese, and yogurt consumption with decreased concentrations of high molecular weight phthalate biomarkers ( Cantonwine et al., 2014 ; He et al., 2019 ; Polinski et al., 2018 ). These findings may in part reflect healthier eating habits, including lower consumption of packaged and canned foods. We identified strong associations between mixtures of high molecular weight phthalate metabolites and current use of a vaginal ring, which is consistent with our prior findings for individual phthalate metabolites in SELF ( Wesselink et al., 2020 ). Use of a vaginal ring was also associated with paraben and phenol biomarker concentrations using both PCA and k -means clustering, suggesting vaginal rings may be a source of exposure to multiple EDC classes. There were suggestive associations between ever use of oral contraception and exposure profiles that included phthalates, parabens and phenols. Generally, there is less evidence in the literature examining hormonal contraceptives as sources of EDC exposure in women, relative to other correlates (e.g., diet, personal care products). Where hormonal contraceptives are commonly used by reproductive-aged women ( Daniels and Abma, 2018 ), additional research can help to better characterize this exposure route and examine the potential health implications for women. Menstrual and vaginal products (i.e., vaginal deodorant and powders) were associated with biomarkers of phthalate metabolites and phenols. Phthalates have been associated with menstrual and vaginal products in the literature ( Branch et al., 2015 ; Wesselink et al., 2020 ), and may in part contribute to exposure disparities between Black and white women as Black women are more likely to use menstrual and vaginal products ( Zota and Shamasunder, 2017 ). Parity was positively associated with biomarker concentrations of phenols and triclosan. However, previous findings in the literature are inconsistent ( Bethea et al., 2020 ; Meeker et al., 2013 ; Weiss et al., 2015 ). SELF is one of the first studies to examine correlates of mixtures of these EDC classes in Black women. Our findings suggest that Black women are exposed to multiple EDCs in consumer products, a notion supported by prior literature that found commonly used consumer products tend to contain mixtures of EDCs. Sunscreens, for example, can contain parabens, phthalates, phenols, and other classes of EDCs in the same products ( Dodson et al., 2012 ). Chemicals within and across these classes of EDCs have the potential for cumulative or interactive health impacts in humans. Phthalate mixtures, for example, have been jointly associated with decreased birthweight ( Chiu et al., 2018 ) and decreased probability of implantation, clinical pregnancy, and live birth ( Mínguez-Alarcón et al., 2019 ). Mixtures of phthalates and phenols were also jointly associated with increased risk of preterm birth ( Zhang et al., 2021 ) and obesity ( Zhang et al., 2019 ), with potential phenol-phenol interactions ( Zhang et al., 2019 ). Further, Black women have higher incidences of multiple reproductive diseases associated with non-persistent EDCs, such as uterine fibroids, infertility and pregnancy loss ( James-Todd et al., 2016 ), highlighting the importance of mitigating exposure to mixtures for public health equity efforts. These disparities may reflect in part racist policies and marketing, as some racial/ethnic groups are targeted by cosmetic advertisements ( Zota and Shamasunder, 2017 ). For example, Black women are far more likely to use certain beauty products, like hair relaxers and scented menstrual and vaginal products, than other racial/ethnic groups; some of these products may also contain higher concentrations of parabens, phthalates, and other EDCs ( Helm et al., 2018 ; Zota and Shamasunder, 2017 ). Disparities in product use reflect in part the idealization of white beauty norms, leading to internalized racism and body shame ( Zota and Shamasunder, 2017 ). Racial segregation of neighborhoods may also impact dietary patterns of Black women, where stores in predominantly Black neighborhoods tend to have more canned goods and less fresh produce than stores in predominantly white neighborhoods ( Morland and Filomena, 2007 ; Ranjit et al., 2010 ). Consumption of canned goods and less healthy foods has been associated with increased EDC exposure ( Ranjit et al., 2010 ), which we also observed in this study. The main strength of this study is its focus on Black women, who have historically been underrepresented in exposure science studies. We used data from a well-established cohort with extensive baseline data collection that allowed us to assess and control for potential confounding by a variety of demographic, reproductive, dietary and product use correlates. Our focus on identifying correlates of non-persistent EDC mixtures is also a strength of this study: our results suggest that some correlates may be sources of exposure to multiple non-persistent EDCs, and these findings can be used to prioritize targeted exposure reduction strategies. Further, our findings were consistent across two dimension reduction statistical methods that have been previously used in mixtures research ( Kalloo et al., 2020 , 2018 ), supporting the robustness of our results. However, each of these methods has drawbacks. The interpretation of associations in PCA is challenging, particularly when biomarkers load in opposite directions ( Braun et al., 2016 ). While the interpretation of k -means clustering is more straightforward, this method is limited by decreasing sample sizes with increasing numbers of clusters ( Raykov et al., 2016 ). Further, both methods are sensitive to outlier values ( Jolliffe, 2006 ; Raykov et al., 2016 ), though our main findings did not materially differ from sensitivity analyses that removed chemical outlier values. There is also little guidance on selecting the optimal number of PCs and clusters. Researchers can use a priori knowledge of the literature and selection indices to guide these analytic decisions, but the selection of PC and cluster numbers is nonetheless subjective. PCs and clusters identified in our data may not be generalizable to populations outside of the SELF cohort ( Braun et al., 2016 ). Our analysis had additional limitations. The case-cohort design used in SELF included 162 women who were not randomly selected at baseline, and may have introduced selection bias. However, the magnitude and direction of associations of key correlates were similar when we restricted our analysis to the randomly selected sub-cohort in sensitivity analyses, suggesting minimal selection bias. We took the average of imputed values from our imputed datasets, as has been done previously ( Kalloo et al., 2018 ; Schildroth et al., 2021 ), because we were not aware of methods to statistically combine results across PCA and k -means analyses. Therefore, our results may underestimate the true standard errors. We also adjusted for all correlates in models, an approach which may have unintentionally adjusted for mediators or colliders. However, our findings for key correlates were similar in reduced models, suggesting adjustment for all of our a priori factors of interest did not introduce significant bias. The correlates examined in this study were quantified in baseline structured self-administered questionnaires and interviews. Therefore, it is possible that the correlates were misclassified, though we would expect any misclassification to be non-differential with respect to EDC biomarker concentrations. Few women in our study also reported recent use of nail polish (9.6%), vaginal douche (2.3%), vaginal powder (12.9%), vaginal deodorant (10.0%), vaginal ring (15.2%), or condoms (4.8%), and results for these correlates should be interpreted with caution. Our study was also cross-sectional in nature, limiting any causal interpretations of our findings. Moreover, urine measurements of non-persistent EDCs reflect recent exposure. We previously reported that the ICCs between concentrations measured at baseline and 20-month follow-up were variable (phthalates: 0.07 – 0.47; phenols: −0.01 – 0.59; parabens: 0.02 – 0.36; triclosan: 0.15; triclocarban: 0.20) ( Bethea et al., 2020 ; Wesselink et al., 2020 ), and baseline EDC biomarkers likely do not fully reflect the relevant exposure windows for some correlates (i.e., ever use of oral contraceptives); this may partially explain null findings. In summary, we identified multiple correlates, particularly use of personal care products and of vaginal rings, that were associated with mixtures of non-persistent EDCs in Black women. Black women have higher body burdens of several EDCs, possibly reflecting racial disparities in product use and dietary habits. Our findings highlight avenues of potential individual-level public health interventions to reduce exposure disparities in Black women, including reducing personal care product use and reducing consumption of canned or packaged foods. Further interventions that address underlying structural racism, including eliminating racist marketing and policies, the removal of EDCs from consumer products targeting Black women, and ensuring equitable access to quality foods, can also prove instrumental in reducing population-level exposure to EDC mixtures.

Endocrine disrupting chemicals (EDCs) interfere with the normal function of the endocrine system ( Diamanti-Kandarakis et al., 2009 ). Phthalates, phenols, and parabens are classes of EDCs that are considered non-persistent given their rapid breakdown in the environment, their relatively short biologic half-lives (ranging from hours to days), and their fast metabolism and elimination from the body following exposure ( Centers for Disease Control and Prevention, n.d. ; Kim and Choi, 2014 ; Sandborgh-Englund et al., 2006 ; Somani and Khalique, 1982 ; Stahlhut et al., 2009 ; Völkel et al., 2002 ; Wei et al., 2021 ; Ye et al., 2014 ). Each of these EDC classes has been associated with adverse reproductive and gynecologic health outcomes. Higher urinary concentrations of phthalate metabolites, for example, have been associated with increased risk of endometriosis, breast cancer, adverse birth outcomes, and uterine fibroids ( Ahern et al., 2019 ; Cai et al., 2019 ; Kamai et al., 2019 ; Radke et al., 2019 ; Weuve et al., 2010 ), though null or inconsistent findings for some health outcomes have also been reported ( Fruh et al., 2021 ; Upson et al., 2013 ; Wesselink et al., 2021 )Increasing biomarker phenol concentrations have been associated with infertility, polycystic ovary syndrome, and breast cancer ( Rochester, 2013 ), while parabens have been associated with altered menstrual cycles and gestational diabetes ( Wei et al., 2021 ). Exposure to non-persistent EDCs can occur through inhalation, ingestion or dermal contact with consumer products ( Dodson et al., 2020 ; Liao et al., 2013 ; Wang and Qian, 2021 ). Phthalates are commonly used as plasticizers and are generally split into two group based on the number of carbons in their alkyl side chains: low molecular weight (1–4 carbons) and high molecular weight (5+ carbons) ( National Research Council (US) Committee on the Health Risks of Phthalates, 2008 ; Wang et al., 2019 ). Phthalates are commonly used in personal care products, including nail polish, fragrances, lotions, hair products, and cosmetics ( Centers for Disease Control and Prevention, n.d. ; Philippat et al., 2015 ; Phthalate Exposure Assessment in Humans, 2008 ), as well as in building materials, medical devices, medications, adhesives, detergents, and clothing ( Centers for Disease Control and Prevention, n.d. ; Phthalate Exposure Assessment in Humans, 2008 ). Because phthalates are used in food packaging materials ( Giuliani et al., 2020 ), food consumption is also an important route of exposure to certain phthalates ( Centers for Disease Control and Prevention, n.d. ; Edwards et al., 2021 ). Similarly, diet (e.g., canned goods, packaged foods, bottled water) is a route of exposure to bisphenols, such as bisphenol A (BPA), that are used in the manufacture of plastics and epoxy resins ( Faniband et al., 2014 ; Meeker et al., 2009 ). Some phenols are used as ultraviolet filters (e.g., benzophenone-3) ( Faniband et al., 2014 ; Philippat et al., 2015 ) in sunscreen products. Both phenols (e.g., triclosan) and triclocarban are commonly used as antimicrobials in personal care products ( Dhillon et al., 2015 ; Dodson et al., 2012 ; Philippat et al., 2015 ), and parabens are often used as preservatives, and are the most common additive in personal care products after water ( Wei et al., 2021 ). Given the ubiquitous use of consumer products, the majority of the U.S. population has detectable concentrations of biomarkers of several non-persistent EDCs in their urine ( Calafat et al., 2010 ; Lehmler et al., 2018 ; Silva et al., 2004 ). Further, biomarker concentrations of non-persistent EDCs tend to be higher among women than men, and among Black Americans compared with white Americans ( Calafat et al., 2010 ; Lehmler et al., 2018 ; Silva et al., 2004 ; Varshavsky et al., 2016 ; Wei et al., 2014 ). Disparities in exposure to EDCs among Black women may result from inconsistent access to healthy foods ( Morland and Filomena, 2007 ; Ranjit et al., 2010 ) or differential patterns of use of certain products, such as menstrual and vaginal products and hair products that specifically target Black consumers ( Charles, 2011 ; Ferranti, 2011 ; James-Todd et al., 2012 ; Zota and Shamasunder, 2017 ). Products that are specifically advertised for Black women have been found to have higher concentrations of some EDCs (e.g., phthalates) ( Helm et al., 2018 ; James-Todd et al., 2021 ; Zota and Shamasunder, 2017 ). Use of such products (e.g., hair relaxers, skin lightening creams) among women of color may result, in part, from racist ideas of beauty that idealize white beauty norms and devalue Black hair texture and skin color ( Gaston et al., 2019 ; Zota and Shamasunder, 2017 ). In the Study of Environment, Lifestyle and Fibroids (SELF), a cohort of reproductive-aged Black women, we previously identified correlates of exposure to individual phthalates, phenols, parabens, and triclocarban. Personal care product use, including vaginal products, nail polish, make-up, lotions, and perfume were positively associated with urinary concentrations of phthalate biomarkers, while sunscreen use was associated with higher concentrations of biomarkers of phenols and parabens ( Bethea et al., 2020 ; Wesselink et al., 2020 ). These findings are consistent with the literature ( Ashrap et al., 2018 ; Braun et al., 2013 ; Cantonwine et al., 2014 ; Meeker et al., 2013 ; Parlett et al., 2012 ; Sandanger et al., 2011 ) and reflect known uses of these EDCs in products ( Al-Saleh and Elkhatib, 2016 ; Dodson et al., 2012 ; European Union, n.d. ; Food and Drug Administration, n.d. , n.d. ; Guo and Kannan, 2013 ; Yap et al., 2017 ). Socioeconomic, demographic, and behavioral factors have been inconsistently associated with biomarkers of these EDC classes ( Bethea et al., 2020 ; Cantonwine et al., 2014 ; He et al., 2019 ; Jin et al., 2019 ; Meeker et al., 2013 ; Parlett et al., 2012 ; Polinski et al., 2018 ; Reeves et al., 2019 ; Stacy et al., 2017 ; Weiss et al., 2015 ; Wenzel et al., 2018 ; Wesselink et al., 2020 ). However, none of these studies considered mixtures of non-persistent EDCs, which may have additive or interactive health effects ( Chiu et al., 2018 ; Mínguez-Alarcón et al., 2019 ; Zhang et al., 2021 , 2019 ). We are aware of only one publication that assessed profiles of biomarkers of exposure to chemical mixtures in women that included non-persistent EDCs ( Kalloo et al., 2018 ). In that study, daily consumption of fruits or vegetables was inversely associated with a mixture profile that included BPA, and household income was positively associated with a profile that included parabens and monoethyl phthalate (MEP). This study also found that self-identified Black women had higher exposure biomarkers of a profile that included parabens and MEP, and lower exposure biomarkers of two mixture profiles: the first included phenols and triclosan, and the second included phthalates. However, this study did not identify mixture biomarker profiles specifically among Black women that may differ from women of other racial/ethnic groups given social and structural forces leading to altered patterns of EDC exposure ( Kalloo et al., 2018 ). In the current study, we analyzed baseline data from the SELF cohort to characterize profiles of biomarker mixtures of non-persistent EDCs and identify correlates of exposure among Black women.