Associations of Urinary Biomarkers of Phthalates, Phenols, Parabens, and Organophosphate Esters with Glycemic Traits in Pregnancy: The Healthy Start Study

Abstract

Background

Certain endocrine-disrupting chemicals (EDCs) are widespread in consumer products and may alter glucose metabolism. However, the impact of EDC exposures on glucose and insulin regulation during pregnancy is incompletely understood, despite potential adverse consequences for maternal and infant health. We estimated associations between 37 urinary biomarkers of EDCs and glucose-insulin traits among pregnant women.

Methods

Seventeen phthalate or phthalate substitute metabolites, six environmental phenols, four parabens, and ten organophosphate ester metabolites were quantified in mid-pregnancy urine from 298 participants in the Healthy Start Study. Fasting blood glucose, insulin, and hemoglobin A1c were assessed concurrently, and Homeostasis Model Assessment 2-Insulin Resistance (HOMA2-IR) was calculated. Gestational diabetes diagnoses and screening results were obtained from medical records for a subset of participants. We estimated associations between each EDC and outcome separately using linear and robust Poisson regression models and analyzed EDC mixture effects.

Results

The EDC mixture was positively associated with glucose, insulin, and HOMA2-IR, although overall associations were attenuated after adjustment for maternal BMI. Two mixture approaches identified di(2-ethylhexyl) phthalate (DEHP) metabolites as top contributors to the mixture’s positive associations. In single-pollutant models, DEHP metabolites were positively associated with fasting glucose, fasting insulin, and HOMA2-IR even after adjustment for maternal BMI. For example, each interquartile range increase in log₂-transformed mono(2-ethyl-5-oxohexyl) phthalate was associated with 2.4 mg/dL (95% confidence interval (CI): 1.1, 3.6) higher fasting glucose, 11.8% (95%CI: 3.6, 20.5) higher fasting insulin, and 12.3% (95%CI: 4.2, 21.1) higher HOMA2-IR. Few EDCs were associated with hemoglobin A1c or with a combined outcome of impaired glucose tolerance or gestational diabetes.

Discussion

Exposures to phthalates and particularly DEHP during pregnancy are associated with altered glucose-insulin regulation. Disruptions in maternal glucose metabolism during pregnancy may contribute to adverse pregnancy outcomes including gestational diabetes and fetal macrosomia, and associated long-term consequences for maternal and child health.

Introduction

Gestational diabetes mellitus (GDM) affects 6–15% of pregnancies (McIntyre et al., 2019) and is a major risk factor of adverse perinatal and long-term health outcomes for both the mother and the offspring (Lowe et al., 2019; McIntyre et al., 2019; Saravanan et al., 2020). However, the risk of adverse health outcomes extends to glycemic levels below the diagnostic thresholds for GDM: higher fasting and post-load maternal glucose levels during pregnancy are associated with increased risk of high birthweight, fetal hyperinsulinemia, childhood insulin resistance, and preeclampsia (Scholtens et al., 2019; The HAPO Study Cooperative Research Group, 2008). Higher levels of other glycemic measures during pregnancy, such as hemoglobin A1c (HbA1c) and glucose from the GCT, were also associated with adverse pregnancy outcomes (Lowe et al., 2012) and higher risk of cardiometabolic diseases in the mother (Retnakaran and Shah, 2019) and the offspring (Franks et al., 2006). Given these serious health consequences, understanding modifiable risk factors for GDM and less severe forms of hyperglycemia is important for informing interventions to protect women and children’s health.

A recent systematic review and meta-analysis reported that exposures to certain endocrine-disrupting chemicals (EDC) in pregnancy are associated with increased risk of GDM, but also highlighted a need for studies on less commonly measured chemicals, as well as attention to the effects of complex EDC mixtures (Yao et al., 2023). The mixture of EDCs to which pregnant people are regularly exposed continues to evolve. Many of these chemicals, such as low-molecular-weight phthalates, benzophenone-3, and parabens, are added to shampoo, lotion, and other personal care products as solvents, active ingredients, or antimicrobial agents (Guo and Kannan, 2013; Han et al., 2016; Zamoiski et al., 2015). Others, such as high-molecular-weight phthalates, bisphenol A (BPA), and certain organophosphate esters (OPEs), are used in the production of plastics, which are then used in a wide variety of consumer products: toys, food and beverage containers, construction materials, textiles, and others (Blum et al., 2019; Mustieles et al., 2020; Zota et al., 2014). OPEs are also added to electronics and furniture as flame retardants (Blum et al., 2019). EDCs are ubiquitous in the environment, and widespread human exposure occurs through ingestion, inhalation, and dermal absorption (Blum et al., 2019; Calafat et al., 2010, 2008a, 2008b; Hartle et al., 2016; Meeker et al., 2009). Biomonitoring studies in both the general population (Calafat et al., 2010; Ospina et al., 2018; van Woerden et al., 2021; Zota et al., 2014) and pregnant women (Bommarito et al., 2021; Braun et al., 2011; Buckley et al., 2022; Fruh et al., 2022) have reported frequent detection (>80% detection frequency) in urine of BPA, methyl paraben, and several phthalate and OPE biomarkers.

Toxicological evidence suggests that exposure to these EDCs may increase the risk of hyperglycemia and GDM by impairing pancreatic beta cell function, insulin action, and glucose regulation (Heindel et al., 2017; Sargis and Simmons, 2019). However, the epidemiologic evidence concerning whether exposure to EDCs commonly found in personal care and other consumer products during pregnancy adversely affects maternal glycemia is limited. Most of the previously published studies examined the potential glycemic effects of phthalates and BPA (Yao et al., 2023), and much less is known on the effects of other environmental phenols, parabens, and OPEs. Epidemiologic studies of the health effects of OPE exposure in pregnant people are particularly scarce (Crawford et al., 2020; Doherty et al., 2019; Yang et al., 2022; Yao et al., 2021). Furthermore, existing studies have typically focused on medical-record derived outcomes including GDM or continuous glucose from the GCT (Bellavia et al., 2019; James-Todd et al., 2022, 2018; Robledo et al., 2015; Vuong et al., 2021), and few have considered the spectrum of glucose metabolism with markers such as fasting glucose and insulin resistance. Examining glucose metabolism markers is important because they predict adverse perinatal outcomes even in women without GDM (Benhalima et al., 2019; Lowe et al., 2012; The HAPO Study Cooperative Research Group, 2008) and may also provide insights into toxicological mechanisms.

In this study, we examined the associations between 37 urinary biomarkers of EDCs frequently found in personal care and other consumer products, including phthalates/phthalate substitutes, environmental phenols, parabens, and OPEs, and concurrently measured glucose-insulin traits in a racially/ethnically diverse cohort of pregnant women. We also examined the EDC biomarkers’ associations with two medical-record derived outcomes, including continuous glucose from GCT and a combined outcome of impaired glucose tolerance (IGT) or GDM, defined based on data obtained from medical records. We hypothesized that individual chemical biomarkers and their mixture would be associated with higher levels of concurrently measured glucose and insulin traits, higher GCT glucose, and higher prevalence of impaired glucose tolerance (IGT) or GDM.

Methods

Study Population

The Healthy Start Study is a prospective cohort study that enrolled 1,410 mother-child dyads based in Denver, Colorado. Pregnant women were recruited from obstetrics clinics affiliated with the University of Colorado Hospital between 2009 and 2014. Eligibility criteria were aged ≥ 16 years, < 24 weeks gestation, singleton pregnancy, no history of stillbirth or preterm birth before 25 weeks gestation, and no history of diabetes, cancer, asthma managed with steroids, or psychiatric illness. Women completed questionnaires, attended a research visit at mid-pregnancy to provide biospecimens (median gestational age = 27 weeks; range: 18–36 weeks), and consented to the abstraction of medical records. The Healthy Start Study protocol was approved by the Colorado Multiple Institutional Review Board (COMIRB). All participants provided written informed consent.

Of 1,410 pregnancies, 726 had urine samples collected at the mid-pregnancy visit. Of these, 446 urine samples were selected for the quantification of metabolites of phthalates/phthalate substitutes, environmental phenols, and parabens, as reported in a previous study (Polinski et al., 2018). Another subset of 577 urine samples was selected for the quantification of OPE metabolites through the Environmental Influences on Child Health Outcomes (ECHO) Program (Buckley et al., 2020). The two sets of samples partially overlapped, resulting in 304 pregnancies with biomarker data available for the four classes of EDCs. We considered a pregnancy eligible for this analysis if all four classes of EDCs had been previously measured in urine, at least one glucose-insulin trait at the mid-pregnancy visit was available, and complete data was available for key covariates (urinary creatinine, maternal age, calendar year of pregnancy, gestational age at urine sample collection, race/ethnicity, education level, smoking, and pre-pregnancy body mass index). Of the 304 potentially eligible pregnancies, we excluded 1 missing all glucose-insulin traits, 4 pregnancies in which participants reported using hypoglycemic medications at the mid-pregnancy visit, and 1 repeat pregnancy to a woman who had an earlier pregnancy in the dataset. Therefore, the analytic sample included 298 women, each contributing data from one pregnancy. A subset of the 298 women, who had GCT or OGTT results available on the same day as or after the mid-pregnancy urine sample collection in the medical records (n = 153) were included in the analyses of GDM screening/diagnosis results (GCT glucose and IGT/GDM). We restricted the analyses of GDM screening/diagnosis results to this subset to preserve the temporal order of exposure and outcome assessments. A participant flow diagram is available in Supplemental Figure 1.

EDC Biomarkers

At mid-pregnancy, women provided spot urine samples after an overnight fast. Most visits occurred in the morning. Urine samples were collected in sterile cups and stored at −80 °C before shipping on dry ice to partner laboratories for analysis. Urinary concentrations of select phthalate/phthalate substitute metabolites, environmental phenols, and parabens were quantified at the Centers for Disease Control and Prevention (CDC) as previously described (Silva et al., 2007; Ye et al., 2005). The analysis of de-identified specimens at the CDC laboratory was determined not to constitute human subjects research. Urinary concentrations of OPE metabolites were measured at Wadsworth Center’s HHEAR Laboratory using a method similar to that described previously (Wang et al., 2019). Seventeen metabolites of phthalate/phthalate substitutes, six phenols, four parabens, and ten OPE metabolites were measured. The full list of EDC biomarkers, including their names, acronyms, limits of detection (LOD), and detection frequencies, is available in Supplemental Table 1.

Glucose-Insulin Traits, Impaired Glucose Tolerance, and GDM

Participants provided fasting blood samples at the same mid-pregnancy visit when the urine was collected. Fasting glucose was measured in plasma with an AU400e Chemistry Analyzer (Olympus America, Center Valley, PA, USA), fasting insulin was measured using radioimmunoassay (RIA) (Millipore Corporation, Burlington, MA, USA), and HbA1c was measured in whole blood with a DCA Vantage analyzer (Siemens, PA, USA) at the University of Colorado Hospital Clinical and Translational Research Center. Homeostasis Model Assessment 2-Insulin Resistance (HOMA2-IR) was calculated from fasting glucose and insulin using the HOMA2 calculator (Diabetes Trials Unit, The Oxford Centre for Diabetes, Endocrinology and Metabolism, University of Oxford, 2023; Wallace et al., 2004).

Results from 1-hour 50-g glucose challenge tests (GCT) and 3-hour 100-g oral glucose tolerance tests (OGTT) were abstracted from medical records. Administered as part of routine gestational diabetes screening, the GCT was typically offered to participants at 24–28 weeks of gestation, followed by an OGTT if GCT glucose ≥ 140 mg/dL. To preserve the temporal order of exposure and outcome assessments, we considered only GCT and OGTT results obtained on the same day as or after the mid-pregnancy urine sample collection. Glucose tolerance status (normal glucose tolerance, IGT, and GDM) was defined using both medical diagnoses and continuous GCT and OGTT results. IGT was defined as GCT glucose ≥ 140 mg/dL or having only one of the glucose values on the OGTT meeting or exceeding the Carpenter-Coustan Criteria (The American College of Obstetricians and Gynecologists, 2018); and GDM was defined as having at least two of the glucose values on the OGTT meeting or exceeding the Carpenter-Coustan Criteria (The American College of Obstetricians and Gynecologists, 2018) or having a physician diagnosis of GDM in the medical records. For analysis, we combined IGT and GDM into one category due to the small number of GDM cases (n = 9). Additionally, we analyzed continuous GCT results as a separate outcome, where available. Although GCT results are not used alone for clinical purposes, continuous GCT glucose predicts future cardiovascular disease risk in women (Retnakaran and Shah, 2019). Presenting analyses for this outcome also allows comparison with several previous studies (Bellavia et al., 2019, 2018; Chiu et al., 2017; James-Todd et al., 2018, 2016; Robledo et al., 2015; Shaffer et al., 2019; Vuong et al., 2021; Wang et al., 2020; Yang et al., 2022; Zukin et al., 2021) on this topic that analyzed associations between EDCs and continuous GCT results during pregnancy. We did not analyze continuous glucose from OGTT as an outcome because only 32 participants had OGTT results.

Covariates

Urinary creatinine, used to account for urine dilution, was measured at the CDC with a Roche/Hitachi Cobas 6000 Analyzer (Roche Diagnostics, Indianapolis, IN). Confounders selected a priori included maternal age, calendar year, gestational age at the time of urine sample collection, educational attainment, race/ethnicity, smoking, and pre-pregnancy body mass index (BMI). Time elapsed between urine sample collection and GCT (in weeks) was an additional covariate in the models for GCT glucose only. Age was calculated based on the woman’s date of birth and the date of urine sample collection. Gestational age at the time of urine sample collection was calculated based on the date of urine sample collection and gestational age recorded in dated medical records. Educational attainment (high school or less, some college/associate degree, college degree, and graduate degree), race/ethnicity (classified as people who identified as Non-Hispanic White, Non-Hispanic Black, Hispanic, and a combined category that included people who identified as non-Hispanic as well as Asian or Pacific Islander, American Indian or Alaska Native, or who did not identify with any of the racial groups listed), and smoking at mid-pregnancy (never, past, current) were self-reported via surveys and questionnaires. We acknowledge that race and ethnicity are social constructs (Flanagin et al., 2021). Race/ethnicity was included in analyses as a proxy for structural racism, which can lead to differences in EDC exposures (Chan et al., 2021; Zota and Shamasunder, 2017) and health disparities (Williams et al., 2019) among different racial/ethnic groups. Pre-pregnancy body weight was abstracted from the medical records or self-reported. Height was measured with a stadiometer. We used these values to calculate pre-pregnancy BMI as pre-pregnancy body weight (kg)/(height (m))². The Healthy Eating Index-2010 (HEI-2010), used in sensitivity analyses to evaluate confounding by diet quality, was calculated based on one or more 24-hour dietary recalls throughout pregnancy (Francis et al., 2021; Shapiro et al., 2016). Although the index did not capture diet quality before exposure and outcome assessments specifically, it represented the typical diet quality over pregnancy.

Statistical Methods

Urinary concentrations of chemical biomarkers detected in at least 70% of samples, hereafter referred to as “frequently detected chemical biomarkers” (n = 26), were analyzed as continuous variables. All instrument-output concentrations were used, including those below the LOD. To facilitate log₂-transformation, concentrations reported as zero or negative values (ranging from 0.3% (n = 1) of observations for mono(3-carboxypropyl) phthalate (MCPP), BPA, benzophenone-3, and dibutyl phosphate and di-isobutyl phosphate mixture (DBUP+DIBP) to 17% (n = 51) of observations for bis(2-chloroethyl) phosphate (BCETP)) were replaced with LOD/2 or the minimum positive reported concentration/2, whichever was smaller (Arbuckle et al., 2015). Urinary concentrations of biomarkers detected in less than 70% of samples, hereafter referred to as “less frequently detected chemical biomarkers” (n = 11), were dichotomized as detected vs not detected for analyses.

Each frequently detected chemical biomarker concentration was adjusted for urine dilution using the covariate-adjusted creatinine standardization method (O’Brien et al., 2016). In brief, each concentration was divided by the ratio of measured to predicted urinary creatinine. Variables included as predictors in this standardization model were maternal age, gestational age at urine sample collection, race/ethnicity, pre-pregnancy BMI, and height (Kuiper et al., 2022). Descriptive statistics (median (1^st and 3^rd quartiles)) for continuous variables and N (%) for categorical variables) were obtained for participant characteristics, outcomes, and each chemical biomarker. Spearman correlation coefficients among the frequently detected chemical biomarkers were calculated.

In single-pollutant models, we used multiple linear regression to examine the association between the urinary concentration of each frequently detected chemical biomarker and each continuous outcome, including fasting glucose, fasting insulin, HOMA2-IR, HbA1c, and GCT glucose. Fasting insulin and HOMA2-IR were natural log-transformed to meet the normality assumption, while the other continuous outcomes were analyzed in their original scales. We used Poisson regression with a robust error variance (Mansournia et al., 2021; Zou, 2004) to examine the association between each frequently detected chemical biomarker and the dichotomous outcome of IGT/GDM versus normal glucose tolerance. All chemical biomarker concentrations were log₂-transformed to reduce the impact of outliers. We assumed linear dose-response relationships between log₂(chemical biomarker concentration) and each (transformed) outcome because penalized splines from fully adjusted generalized additive models (GAM) showed few meaningful non-linear dose-response relations; non-linear smoothed terms were statistically non-significant or appeared driven by outliers (Supplemental Figures 2–7). We fit a series of single-pollutant models. Model 1 was adjusted for maternal age, gestational age at urine collection, and calendar year. Model 2 was additionally adjusted for education, race/ethnicity, and prenatal smoking. Model 3 was additionally adjusted for pre-pregnancy BMI. For GCT glucose, all models were additionally adjusted for time elapsed (weeks) between urine sample collection and GCT. This adjustment was made because post-load/postprandial glucose levels change throughout mid-to-late pregnancy (Bochkur Dratver et al., 2022). Accounting for the timing of GCT allowed us to capture this important source of variation in GCT glucose to increase the precision of effect estimates. All effect estimates were scaled by the interquartile range (IQR) of the log₂-transformed chemical biomarker concentration.

We used two methods for evaluating the associations between EDC mixtures and each outcome: quantile g-computation (Keil et al., 2020) and Bayesian kernel machine regression (BKMR) (Bobb et al., 2018). For both methods, we fit multi-pollutant models adjusted for the same sets of covariates as in Models 2 and 3 of the single-pollutant models to examine the potential impact of adjusting for pre-pregnancy BMI.

In quantile g-computation, we used the “qgcomp.noboot” function in the “qgcomp” package to examine (percent) differences in each continuous outcome and the conditional odds ratio of IGT/GDM associated with one quartile increase in all 26 frequently detected chemical biomarkers. The biomarker quartiles were defined based on the biomarkers’ distributions in the full analytic sample and were assumed to be associated with the (transformed) outcome linearly and additively. Each chemical biomarker contributed to either the positive or the negative partial effect of the mixture, which summed to the overall mixture effect (Keil et al., 2020). The importance of each biomarker was quantified by positive or negative weights, which represent the proportion of positive or negative partial effects due to the biomarker (Keil et al., 2020). We were unable to obtain the prevalence ratio of IGT/GDM associated with the mixture because the algorithm did not converge during bootstrapping.

In BKMR, we examined the association between the mixture of the 26 frequently detected chemical biomarkers and each continuous outcome. Hierarchical variable selection was incorporated into each model to identify important biomarkers. Specifically, phthalate metabolites from the same parent compound (di-n-butyl phthalate (DnBP): MnBP and MHBP; di-isobutyl phthalate (DiBP): mono-isobutyl phthalate (MiBP) and mono-hydroxy-isobutyl phthalate (MHiBP); di(2-ethylhexyl) phthalate (DEHP): mono(2-ethylhexyl) phthalate (MEHP), mono(2-ethyl-5-hydroxyhexyl) phthalate (MEHHP), mono(2-ethyl-5-oxohexyl) phthalate (MEOHP), and mono(2-ethyl-5-carboxypentyl) phthalate (MECPP)) were grouped, while the other biomarkers were selected individually. Prior to model fitting, all log₂-transformed biomarker concentrations and continuous covariates were standardized. Using the “bkmrhat” package, we fitted each BKMR model with four parallel Markov chains, each with 100,000 iterations, including 50,000 burn-in iterations. Model convergence was confirmed with trace plots, autocorrelation plots, and Gelman-Ruben convergence statistics (Zhang et al., 2023). We used a similar approach to fit probit BKMR for the binary outcome of IGT/GDM. However, the model failed to converge even after collapsing some covariate categories to reduce model complexity and extending each Markov chain to 250,000 iterations. Therefore, we report results for continuous outcomes only.

From each BKMR model, we obtained 1) the joint association of the mixture with the outcome, defined as the (percent) difference in the outcome when all biomarkers were at the 75^th percentile compared to when they were all the 25^th percentile; 2) the group and conditional posterior inclusion probabilities (PIP) of each biomarker; 3) the dose-response relationship between each biomarker and the outcome when the other biomarkers were at the 50^th percentile, and 4) the dose-response relationship between a biomarker and the outcome when a second biomarker was fixed at its 10^th, 50^th, and 90^th percentiles, holding all other biomarkers at their 50^th percentile. The last sets of outputs allowed us to examine potential interaction between two biomarkers. Because there were 26 × 25 = 650 two-way interactions, and not all biomarkers were equally important, we present two-way interactions among biomarkers with the highest two group PIPs only.

In sensitivity analyses, we examined the potential for confounding by diet quality by additionally adjusting for HEI-2010 in both single- and multi-pollutant models, in the subset of participants with complete diet data. To evaluate potential selection bias in the subset of participants included in analyses of GDM screening/diagnosis results, we compared participant characteristics and chemical biomarker concentrations between participants with and without GDM screening/diagnosis results in the time window of interest.

We applied the same analytic framework to the less frequently detected chemical biomarkers, except that they were dichotomized as “detected vs not detected” in models, and no multi-pollutant analyses were conducted. Statistical significance was defined as two-tailed p-value < 0.05. Data management was conducted with SAS 9.4 (SAS Institute Inc., Cary, NC), and statistical analyses were conducted with R 4.2.2 (R Core Team, 2022), using the packages of mgcv 1.8–41, stats 4.2.2, sandwich 3.1–0, qgcomp 2.10.1 (Keil, 2023), bkmrhat 1.1.3 (Keil, 2022), and bkmr 0.2.2 (Bobb, 2022), in addition to base R.

Results

In this analysis of 298 pregnant women, median age was 29.2 years (Quartile (Q) 1, Q3: 24.2, 32.7) (Table 1). Approximately 62% of participants were non-Hispanic White, 13% Non-Hispanic Black, 18% Hispanic, and 6% from other racial/ethnic groups. Approximately half of the women had college degrees or higher. The median pre-pregnancy BMI was 24.4 kg/m² (Q1, Q3: 20.7, 27.6). The prevalence of overweight and obesity was 29.9% and 17.4%, respectively.

Table 1.

Sociodemographic and glycemic characteristics of 298 women in the Healthy Start Study

	Median (Q1, Q3) or N (%)1
Age (years)	29.2 (24.2, 32.7)
Pre-pregnancy BMI (kg/m2)	24.4 (20.7, 27.6)
Gestational age at mid-pregnancy visit (weeks)	27.4 (25.6, 29.3)
Fasting glucose (mg/dL)	78.0 (74.0, 82.0)
Fasting insulin (uIU/mL)	13.0 (10.0, 17.0)
HOMA2-IR	1.39 (1.03, 1.82)
HbA1c (%)	5.0 (4.8, 5.1)
GCT glucose (mg/dL)2	111.0 (93.0, 126.0)
Race/ethnicity
Non-Hispanic White	186 (62.4%)
Non-Hispanic Black	40 (13.4%)
Hispanic	55 (18.5%)
Others	17 (5.7%)
Educational attainment
High school or less	81 (27.2%)
Some college/associate degree	64 (21.5%)
College degree	75 (25.2%)
Graduate degree	78 (26.2%)
Smoking
Never	223 (74.8%)
Past	55 (18.5%)
Current	20 (6.7%)
Pre-pregnancy BMI category
Normal/underweight (<25 kg/m2)	157 (52.7%)
Overweight (25–29.99 kg/m2)	89 (29.9%)
Obesity (≥ 30 kg/m2)	52 (17.4%)
Gestational glucose tolerance status 3
Normal glucose tolerance	131 (85.6%)
Impaired glucose tolerance	13 (8.5%)
Gestational diabetes mellitus	9 (5.9%)

¹Abbreviations: Q1 = 1^st quartile; Q3 = 3^rd quartile; HOMA2-IR: Homeostasis Model Assessment 2-Insulin Resistance; HbA1c: hemoglobin A1c; GCT: glucose challenge test; BMI: body mass index.

²Only GCT glucose obtained on the same day as or after urine sample collection (n = 143) was included in analyses.

³Gestational glucose tolerance status was defined using results from GCT and oral glucose tolerance tests conducted on the same day as or after urine sample collection and was available for 153 participants.

Urine and fasting blood samples were collected at a median gestational age of 27.4 weeks (Q1, Q3: 25.6, 29.3) (Table 1). Median (Q1, Q3) fasting glucose, insulin, HOMA2-IR, and HbA1c were 78.0 (74.0, 82.0) mg/dL, 13.0 (10.0, 17.0) uIU/mL, 1.39 (1.03, 1.83), and 5.0 (4.8, 5.1) %, respectively. The median time elapsed between urine sample collection and GCT was 1.7 weeks (Q1, Q3: 0.4, 3.5). The median (Q1, Q3) GCT glucose was 111 (93, 126) mg/dL (Table 1). Among the subset of participants with GDM screening/diagnosis results in the time window of interest (n=153), the prevalence of IGT and GDM was 8.5% and 5.9%, respectively (Table 1).

Twenty-six chemical biomarkers were detected in at least 70% of urine samples (Supplemental Table 1). Their detection frequencies ranged from 72.1% for MHBP to 100% for MECPP, mono-carboxyisooctyl phthalate (MCOP), methyl paraben, and propyl paraben. Median concentrations were the highest for methyl paraben (110.2 ng/mL, Q1, Q3: 39.9, 287.2), benzophenone-3 (99.8 ng/mL, Q1, Q3: 28.2, 373.1), and MEP (25.5 ng/mL, Q1, Q3: 12.9, 81.7) and the lowest for DBUP + DIBP (0.1 ng/mL, Q1, Q3: 0.1, 0.2), bisphenol S (BPS) (0.3 ng/mL, Q1, Q3: 0.2, 0.5) and 2,4-dichlorophenol (0.5 ng/mL, Q1, Q3: 0.3, 1.0) (Supplemental Table 2). Metabolites of the same parent phthalates (DnBP: MnBP and MHBP; DiBP: MiBP and MHiBP; DEHP: MEHP, MEHHP, MEOHP, and MECPP) were highly correlated with each other (Supplemental Figure 8). Other phthalate metabolites were moderately correlated. Correlation between biomarkers from different chemical classes – for instance, between parabens and OPE metabolites – was generally low.

Minimally adjusted single-pollutant models (Model 1) showed that among the frequently detected chemical biomarkers, most phthalate metabolites were positively associated with fasting glucose, fasting insulin, and HOMA2-IR (Supplemental Tables 3–5). Adjustment for education, race/ethnicity, and smoking attenuated some of these associations. An IQR increase in the concentrations of most log₂-transformed phthalate metabolites was associated with 1.1–2.8 mg/dL higher fasting glucose (MnBP: 1.1 mg/dL (95% confidence interval (CI): 0.05, 2.1); MEHHP: 2.8 mg/dL (95% CI: 1.5, 4.1)), 7.4–17.0% higher fasting insulin (MCPP: 7.4% (95% CI: 0.3, 14.9); MEOHP: 17.0% (95% CI: 8.3, 26.5)), and 6.4–17.6% higher HOMA2-IR (DBUP+DIBP: 6.4% (95% CI: 0.9, 12.2); MEOHP: 17.6% (95% CI: 8.8, 27.0)) (Supplemental Tables 3–5). Additional adjustment for pre-pregnancy BMI further reduced the magnitude of these associations, though several positive associations persisted with 95% CIs that did not contain the null. In the fully adjusted models, an IQR increase in log₂-transformed DEHP metabolites (MEHP, MEHHP, MEOHP, and MECPP), MCOP, mono-carboxy-isononyl phthalate (MCNP), and MCPP was associated with 1.2–2.5 mg/dL higher fasting glucose (MEHP: 1.2 mg/dL (95% CI: 0.1, 2.3); MEHHP: 2.5 mg/dL (95% CI: 1.3, 3.8)); an IQR increase in log₂-transformed MEHHP, MEOHP, and MECPP was associated with 8.7–11.3% higher fasting insulin (MECPP: 8.7% (95% CI: 1.0, 17.0); MEOHP: 11.8% (95% CI: 3.6, 20.5)) and 9.4–11.9% higher HOMA2-IR (MECPP: 9.4% (95% CI: 1.7, 17.7); MEOHP: 12.3% (95% CI: 4.2, 21.1)) (Supplemental Tables 3–5; Figure 1). 2,4-dichlorophenol was the only analyte that was statistically significantly, inversely associated with fasting glucose, insulin, and HOMA2-IR (Supplemental Tables 3–5; Figure 1). Few other chemical biomarkers were associated with fasting glucose, insulin, and HOMA2-IR (Supplemental Tables 3–5; Figure 1). None of the frequently detected EDC biomarkers were statistically significantly associated with HbA1c, GCT glucose, or IGT/GDM, although we noted non-significant positive associations between BPA and GCT glucose, and between BPS and GCT glucose and IGT/GDM (Supplemental Tables 6–8; Figures 1 and 2). Additionally, among the less frequently detected biomarkers, detectable ethyl paraben was associated with a lower prevalence of IGT/GDM (prevalence ratio 0.29, 95% CI 0.10, 0.80) compared to non-detected ethyl paraben (Supplemental Table 9).

Figure 1.

Differences/percent differences in fasting glucose, fasting insulin, HOMA2-IR, and HbA1c per interquartile range increase in log₂-transformed urinary concentrations of the frequently detected chemical biomarkers. “Frequently detected” was defined as detection frequency ≥ 70% in the analytic sample. Effect estimates were obtained from single-pollutant models using multiple linear regression and were adjusted for maternal age, gestational age at urine sample collection, calendar year, education, race/ethnicity, smoking, and pre-pregnancy BMI. Abbreviations: HOMA2-IR: Homeostasis Model Assessment 2-Insulin Resistance; HbA1c: hemoglobin A1c. Sample sizes: 297 for fasting glucose; 292 for fasting insulin and HOMA2-IR, and 285 for HbA1c.

Figure 2.

Differences in GCT glucose and prevalence ratios for IGT/GDM per interquartile range increase in log₂-transformed urinary concentrations of the frequently detected chemical biomarkers. “Frequently detected” was defined as detection frequency ≥ 70% in the analytic sample. Effect estimates were obtained from single-pollutant models using multiple linear (GCT glucose) or robust Poisson (IGT/GDM) regression and were adjusted for maternal age, gestational age at urine sample collection, calendar year, education, race/ethnicity, smoking, and pre-pregnancy BMI. For GCT glucose, time elapsed between urine collection and GCT was additionally adjusted. Abbreviations: GCT: glucose challenge test.; IGT: Impaired glucose tolerance; GDM: Gestational diabetes mellitus. Sample sizes: 143 for GCT glucose and 153 for IGT/GDM.

Results from quantile g-computation are shown in Table 2, Figure 3, and Supplemental Figure 9. Table 2 reports the overall mixture effects, while Figure 3 and Supplemental Figure 9 report the partial effects of the mixture and the weight of each chemical biomarker. For each model, the partial effects summed to the overall mixture effect, and the weights were the proportion of partial effects due to a biomarker. In models adjusted for demographic factors and smoking (Model 2), each quartile increase in all 26 frequently detected chemical biomarkers was associated with 2.2 mg/dL (95% CI: −0.3, 4.8) higher fasting glucose, 22.5% (95% CI: 4.5, 43.6) higher fasting insulin, and 22.7% (95% CI: 4.8, 43.6) higher HOMA2-IR (Table 2). Further adjustment for pre-pregnancy BMI somewhat attenuated the magnitude of these associations by 9–42%, but the direction of the associations remained positive. In the fully adjusted models, each quartile increase in the mixture was associated with 2.0 mg/dL (95% CI: −0.6, 4.6) higher fasting glucose, 13.1% (95% CI: −3.3, 32.4) higher fasting insulin, and 13.3% (95% CI: −3.1, 32.4) higher HOMA2-IR (Table 2).

Table 2.

The joint associations between the mixture of frequently detected¹ chemical biomarkers and glycemic outcomes as estimated with quantile g-computation

		Per quartile increase in all chemical biomarker concentrations
	N	Model 22	Model3
Difference in fasting glucose (mg/dL) (95% CI)	297	2.2 (−0.3, 4.8)	2.0 (−0.6, 4.6)
% Difference in fasting insulin (95% CI)	292	22.5 (4.5, 43.6) 3	13.1 (−3.3, 32.4)
% Difference in HOMA2-IR (95% CI)4	292	22.7 (4.8, 43.6)	13.3 (−3.1, 32.4)
Difference in HbA1c (percentage point) (95% CI)	285	−0.02 (−0.12, 0.07)	−0.04 (−0.13, 0.06)
Difference in GCT glucose (mg/dL) (95% CI)	143	−1.0 (−15.7, 13.8)	−0.8 (−16.0, 14.4)
Odds ratio for IGT/GDM (95% CI)	153	0.22 (0.01, 3.27)	0.13 (0.01, 2.95)

¹“Frequently detected” was defined as detection frequency ≥ 70% in the analytic sample.

²Model 2: Adjusted for maternal age, gestational age at urine sample collection, calendar year of pregnancy, education, race/ethnicity, and smoking. For GCT glucose, time elapsed between urine sample collection and GCT was additionally adjusted.

Model 3: Model 2 + pre-pregnancy BMI

³Bold: p-value < 0.05

⁴Abbreviations: HOMA2-IR: Homeostasis Model Assessment 2-Insulin Resistance; HbA1c: hemoglobin A1c; GCT: glucose challenge test.; IGT: Impaired glucose tolerance; GDM: Gestational diabetes mellitus; CI: confidence interval.

Figure 3.

Weights of the chemical biomarkers in quantile g-computation models for fasting glucose, insulin, and HOMA2-IR. Model 2 was adjusted for maternal age, gestational age at urine sample collection, calendar year of pregnancy, education, race/ethnicity, and smoking. Model 3 was additionally adjusted for pre-pregnancy BMI. Units for partial effects were mg/dL, log uIU/mL, and log unit, for fasting glucose, fasting insulin, and HOMA2-IR, respectively. Abbreviation: HOMA2-IR: Homeostasis Model Assessment 2-Insulin Resistance.

Consistent with our findings in single-pollutant models, two DEHP metabolites made the strongest contributions to the positive effects of the mixture. MECPP had the highest positive weights in models for fasting glucose across both adjustment sets (Figure 3), and MEOHP had the highest positive weights in models for fasting insulin and HOMA2-IR across both adjustment sets (Figure 3). Also consistent with the single-pollutant model results, the mixture was not significantly associated with HbA1c, GCT glucose, or IGT/GDM (Table 2). The partial positive and negative effects and weights from quantile g-computation models for these outcomes are available in Supplemental Figure 9.

BKMR models revealed positive joint associations between the chemical biomarker mixture and fasting glucose, insulin, and HOMA2-IR, although effects were attenuated somewhat by adjustment for pre-pregnancy BMI (Figures 4 and 5 and Table 3). In models adjusted for demographic factors and smoking (Model 2), compared to the 25^th percentile, the 75^th percentile of the chemical biomarker mixture was associated with 2.6 mg/dL (95% credible interval: 0.13, 5.1) higher fasting glucose, 20.9% (95% credible interval: 1.14, 44.4) higher fasting insulin, and 20.7% (95% credible interval: −3.0, 50.2) higher HOMA2-IR (Figure 4). Adjustment for pre-pregnancy BMI attenuated these associations, particularly for fasting insulin and HOMA2-IR. In the fully adjusted models (Model 3), compared to the 25^th percentile, the 75^th percentile of the chemical biomarker mixture was associated with 1.7 mg/dL (95% credible interval: −0.73, 4.1) higher fasting glucose, 4.7% (95% credible interval: −10.8, 23.0) higher fasting insulin, and 3.7% (95% credible interval: −14.8, 26.2) higher HOMA2-IR (Figure 4).

Figure 4.

The joint associations between the mixture of frequently detected chemical biomarkers and fasting glucose, insulin, and HOMA2-IR as estimated with Bayesian kernel machine regression (BKMR). “Frequently detected” was defined as detection frequency ≥ 70% in the analytic sample. Model 2 was adjusted for maternal age, gestational age at urine sample collection, calendar year of pregnancy, education, race/ethnicity, and smoking. Model 3 was additionally adjusted for pre-pregnancy BMI. Abbreviation: HOMA2-IR: Homeostasis Model Assessment 2-Insulin Resistance.

Figure 5.

Group posterior inclusion probabilities (PIP) from Bayesian kernel machine regression (BKMR) models for fasting glucose, insulin, and HOMA2-IR. Model 2 was adjusted for maternal age, gestational age at urine sample collection, calendar year of pregnancy, education, race/ethnicity, and smoking. Model 3 was additionally adjusted for pre-pregnancy BMI. Abbreviation: HOMA2-IR: Homeostasis Model Assessment 2-Insulin Resistance.

Table 3.

Conditional posterior inclusion probabilities (PIP) for DEHP metabolites as estimated with Bayesian kernel machine regression (BKMR) models

	Fasting glucose		Fasting insulin		HOMA2-IR
	Model 21	Model 3	Model 2	Model 3	Model 2	Model 3
MEHP	0.001	0.006	0.002	0.05	0.003	0.02
MEHHP	0.32	0.35	0.22	0.25	0.22	0.25
MEOHP	0.21	0.23	0.70	0.61	0.61	0.58
MECPP	0.46	0.40	0.07	0.09	0.17	0.15

¹Model 2 was adjusted for maternal age, gestational age at urine sample collection, calendar year of pregnancy, education, race/ethnicity, and smoking. Model 3 was additionally adjusted for pre-pregnancy BMI. Abbreviation: HOMA2-IR: Homeostasis Model Assessment 2-Insulin Resistance.

The most important biomarkers for fasting glucose were DEHP metabolites (similar to what we observed in single-pollutant models and quantile g-computation weights) and MCPP, but their relative rankings differed between models with and without pre-pregnancy BMI adjustment, and the group PIP exceeded 0.5 only for DEHP metabolites in Model 2 (Figure 5). Among DEHP metabolites, MECPP was the most important as it had the highest conditional PIP across both adjustment sets (Table 3). For fasting insulin and HOMA2-IR, the most important biomarkers were DEHP metabolites and triclosan, with the group PIPs exceeding 0.5 in the fully adjusted model for HOMA2-IR (Figure 5). Among DEHP metabolites, MEOHP was the most important as it had the highest conditional PIP for fasting insulin and HOMA2-IR across both adjustment sets (Table 3).

There were no notable non-linear dose response relations observed between EDC biomarkers and fasting glucose (Supplemental Figure 10), and no pairwise interactions among DEHP metabolites and MCPP were identified (Supplemental Figure 11). For fasting insulin and HOMA2-IR, MEOHP was positively associated with both outcomes in the fully adjusted models, while triclosan appeared to have an inverted U-shape dose-response relationship with both outcomes (Supplemental Figures 12 and 14). However, non-linear associations identified in BKMR may be driven by outliers and should be interpreted with caution. As for interaction, the associations between MEOHP and fasting insulin and HOMA2-IR appeared somewhat stronger at the 10^th and 50^th percentiles of triclosan compared to when triclosan was at the 90^th percentile (Supplemental Figures 13 and 15), suggesting a weak antagonism of effects. Consistent with both single-pollutant models and quantile g-computation, the chemical biomarker mixture was not associated with HbA1c or GCT glucose in BKMR models (Supplemental Figures 16–19).

Sensitivity analyses did not meaningfully change the interpretation of our findings. Results from both single- and multi-pollutant models for the frequently detected chemical biomarkers remained similar after additionally adjusting for HEI-2010 (n ranging from 150 to 294; Supplemental Tables 10 and 11 and Supplemental Figure 20). Compared to those excluded from the analyses of GDM screening/diagnosis results, participants included in the analyses were older, more likely to be non-Hispanic White persons, received GDM screening approximately 1.5 weeks later, and attended the mid-pregnancy visit approximately 3 weeks earlier (Supplemental Table 12). However, the proportion of participants receiving their first GDM screening before 24 weeks gestation, a clinical practice indicating the clinician’s judgment of elevated GDM risk (The American College of Obstetricians and Gynecologists, 2018), was similar between the two groups (Supplemental Table 12). The urinary concentrations of most of the chemical biomarkers were also similar between those included versus excluded from the analyses of GDM screening/diagnosis results (Supplemental Table 13).

Discussion

In a Denver, Colorado-based cohort of women, we examined the associations between 37 urinary biomarkers of EDCs frequently found in personal care and other consumer products, including select phthalates, environmental phenols, parabens and organophosphate esters, and concurrently measured glucose-insulin traits as well as medical-record derived GCT glucose and the presence of IGT/GDM. Across multiple analytic approaches, we found that certain metabolites of DEHP were consistently associated with higher fasting glucose, fasting insulin, and HOMA2-IR at mid-pregnancy, though results were statistically significant in single-pollutant models only. We did not observe significant positive associations with glucose-insulin traits among the other chemical classes studied, and the EDC mixture effects were generally not statistically significant. None of the frequently detected chemical biomarkers was associated with HbA1c, GCT glucose, and IGT/GDM. Overall, our results suggest that DEHP exposure at mid-pregnancy may influence glucose metabolism, leading to increased estimated insulin resistance and fasting glucose. However, our results also highlight the complexities in understanding the relationships between EDC exposures and hyperglycemia/GDM in pregnancy, as the findings varied considerably across chemical biomarkers and with respect to specific glucose/insulin outcomes.

DEHP is one of the most common EDCs found in consumer products. Used mainly as a plasticizer for polyvinyl chloride (PVC) plastics, DEHP can be added to numerous PVC applications, including furniture upholstery, garden hoses, shower curtains, the plastic sheathing of wires and cables, clothing, and plastic food packaging and containers (Agency for Toxic Substances and Disease Registry (ATSDR), 2022). Although the use of DEHP and the general population’s exposure appear to have declined in the United States since the late 2000s (Zota et al., 2014), exposure to DEHP – believed to result largely from dietary intake of contaminated foods (Serrano et al., 2014; Zota et al., 2016) – is still widespread (Centers for Disease Control and Prevention, 2022).

Epidemiologic studies in non-pregnant adults generally support positive associations between DEHP exposure and fasting glucose and markers of insulin resistance (Kim et al., 2013; James-Todd et al., 2012; Huang et al., 2014; Dales et al., 2018; Duan et al., 2019; Peng et al., 2023; Radke et al., 2019). DEHP exposure has also been associated with incident diabetes in some (Sun et al., 2014), though not all cohort studies in non-pregnant women (Peng et al., 2023). In pregnant women, the few studies examining DEHP exposure and fasting-state glucose-insulin traits reported inconsistent associations with fasting glucose (Fisher et al., 2018; Gao et al., 2021; Wang et al., 2023) and no associations with HOMA2-IR (Fisher et al., 2018). Furthermore, studies on DEHP exposure and GCT glucose, IGT, or GDM in pregnancy generally reported null associations (Yao et al., 2023; Yan et al., 2022; Fisher et al., 2018; Shaffer et al., 2019; James-Todd et al., 2022; Shapiro et al., 2015; Zukin et al., 2021; Robledo et al., 2015; James-Todd et al., 2018; Vuong et al., 2021; James-Todd et al., 2016). Unlike previous studies on phthalate exposures and fasting glucose and HOMA2-IR in pregnancy (Fisher et al., 2018; Gao et al., 2021; Wang et al., 2023), our study measured and examined DEHP metabolites and these outcomes concurrently, an approach similar to studies in non-pregnant adults. The consistency between our findings and those in non-pregnant adults regarding fasting-state glucose-insulin traits suggests that DEHP may have acute effects on insulin resistance and fasting glucose in both non-pregnant and pregnant people. Further investigations in other pregnancy cohorts, preferably with a longitudinal design, can help confirm our findings and fully evaluate the potential glycemic impact of DEHP exposure during pregnancy.

Our finding that higher urinary concentrations of DEHP metabolites were associated with higher fasting glucose and estimated insulin resistance during pregnancy has important implications. Elevated fasting glucose and insulin resistance in pregnancy have been associated with numerous adverse pregnancy and neonatal outcomes (Benhalima et al., 2019; The HAPO Study Cooperative Research Group, 2008). For instance, in the HAPO study, each 6.9 mg/dL increase in fasting glucose during pregnancy was associated with 38% higher odds of baby being born large for gestational age (LGA) and 21% higher odds of preeclampsia. This means that each 2.4 mg/dL increase in fasting glucose – the expected difference in fasting glucose comparing pregnancies with urinary MECPP concentrations at the 75^th to the 25^th percentiles based on the single-pollutant model – would be associated with 12% higher odds of LGA and 7% higher odds of preeclampsia. This implies a pathway by which increased DEHP exposure during pregnancy may lead to elevated risks of LGA and preeclampsia, conditions with long-term health implications for the child or the mother (Scifres, 2021; Turbeville and Sasser, 2020). Importantly, because DEHP exposure is ubiquitous, if DEHP is causally related to higher fasting glucose, insulin resistance, and their adverse perinatal and long-term consequences, the impact of reducing DEHP exposure during pregnancy on the health of women and children at the population level is considerable.

One intriguing finding of this study is the null associations between DEHP metabolite concentrations and GCT glucose and IGT/GDM, which has also been observed in many previous studies in pregnant women (Yao et al., 2023; Yan et al., 2022; Fisher et al., 2018; Shaffer et al., 2019; James-Todd et al., 2022; Shapiro et al., 2015; Zukin et al., 2021; Robledo et al., 2015; James-Todd et al., 2018; Vuong et al., 2021; James-Todd et al., 2016). These null associations may reflect common limitations of this and many previous studies – for instance, abstracting GCT and OGTT results from medical records instead of conducting OGTT at the same time as or soon after exposure assessment (James-Todd et al., 2022, 2018, 2016; Robledo et al., 2015; Shaffer et al., 2019; Shapiro et al., 2015; Vuong et al., 2021; Zukin et al., 2021). If DEHP has an acute, or even temporary, effect on glucose tolerance, this effect may not have been detected due to the time lag between exposure assessment and GCT/OGTT. However, null results could also be due to our limited sample size, or the combination of IGT with GDM as a category for analysis.

The difference between our observed results for fasting glucose versus GDM screening-related measures could also suggest that DEHP has more limited effects on post-load glucose compared to fasting glucose. Fasting glucose levels are determined primarily by the rate of hepatic glucose production (Alatrach et al., 2017), while post-load glucose levels depend largely on insulin-dependent glucose uptake into peripheral tissues, particularly skeletal muscle (Abdul-Ghani et al., 2006; Merz and Thurmond, 2020). Toxicological studies have shown that DEHP-fed rodents developed injuries and altered glucose metabolism in the liver (Martinelli et al., 2006; Zhang et al., 2017). In addition, human hepatocytes treated with DEHP showed reduced expression of insulin receptors, which can potentially lead to hepatic insulin resistance (Zhang et al., 2017). It is interesting to note that DEHP or its metabolites may also induce insulin resistance in skeletal muscle (Wei et al., 2020) and reduce pancreatic beta cell viability (Karabulut and Barlas, 2021). The relative importance of each pathway for DEHP’s toxicity is unknown. Regardless, our findings suggest that examining multiple glucose metabolism markers, in addition to clinical disease during pregnancy, is needed to provide a more complete picture of glycemic effects of EDCs and help generate additional insights into the potential mechanisms of EDC-related GDM pathogenesis.

Apart from DEHP metabolites, few other chemical biomarkers were associated with glucose and insulin traits. Some of our findings are consistent with previous studies. For example, BPA was not associated with IGT or GDM in two recent systematic reviews (Taheri et al., 2021; Yao et al., 2023). Other findings are new additions to the small existing evidence base (Chen et al., 2022; Jin et al., 2023; Ouyang et al., 2018; Wang et al., 2020; Yang et al., 2022). In toxicological studies, many phthalates or their metabolites, environmental phenols, parabens, OPEs or their metabolites affect similar physiological processes crucial to glucose homeostasis, including pancreatic β-cell function, adipogenesis, insulin sensitivity in adipose tissue and skeletal muscle, and the metabolism of sex and thyroid hormones (Yao et al., 2023; Heindel et al., 2017; Taxvig et al., 2012; Cano-Sancho et al., 2017). It is unclear why we observed null associations for most chemical biomarkers. The lower concentrations of many of the chemical biomarkers in the Healthy Start Study compared to the general population and other pregnancy cohorts may play a role (Polinski et al., 2018). Alternatively, EDCs sharing structural and toxicological similarities may not have the same health effects in human populations. Thus, to fully understand whether exposure to EDCs contributes to hyperglycemia and GDM in pregnancy, additional epidemiologic investigations are needed in diverse populations with different exposure profiles.

The study findings must be considered in light of several limitations. Our sample size was modest, which has limited the study’s power and our ability to examine categorical endpoints; for example, we were unable to estimate prevalence ratios for the association between the EDC mixture and IGT/GDM, or implement BKMR for the binary outcome due to the relatively small number of cases. Additionally, we are unable to examine how associations may differ among racial and ethnic groups within the US, and particularly limited in representation of people who identify as groups other than white, Black, or Hispanic. Urinary concentrations of the chemical biomarkers evaluated reflect recent exposure, and biomarkers were quantified in one spot urine sample, which likely led to non-differential exposure measurement error and hence attenuation of effect estimates. The study was restricted to women with ongoing pregnancies at the time of biospecimen collection (median 27 weeks); although no fetal demise was recorded among participants with available chemical exposure data, women with early pregnancy losses would have been excluded from this population. If higher EDC concentrations were associated with early pregnancy loss, then the study population could underrepresent participants with the highest exposures and/or greatest risk of pregnancy complications, potentially producing bias toward the null. Finally, all outcomes were assessed at one point in time. The limited temporality for some outcomes and our inability to examine changes/incidence of outcomes suggest caution in using our findings for causal inference.

This study also has several strengths. We examined multiple glucose-insulin traits, which provided information on the potential glycemic impact of EDCs beyond just a GDM diagnosis. We also examined an extensive list of chemical biomarkers to provide a more comprehensive evaluation of EDCs frequently found in personal care and other consumer products. Of note, our analyses on the OPE metabolites are among the few available (Feng et al., 2016; Jin et al., 2023; Yang et al., 2022) for these emerging environmental contaminants. Our use of two complementary multi-pollutant models also helped us understand the potential glycemic effects of the mixture and identify the most important components, which may assist in prioritizing public health interventions. Finally, the analytic sample was diverse in terms of race/ethnicity and socioeconomic status, relative to many of the previous studies cited here. This diversity enhanced the generalizability of our findings to reproductive-aged women located in the Denver, Colorado metropolitan area.

Conclusions

Using both single- and multi-pollutant approaches, we found that higher urinary concentrations of certain DEHP metabolites were associated with higher fasting glucose, fasting insulin, and HOMA2-IR in a Denver, Colorado-based cohort of pregnant women, though results were statistically significant in single-pollutant models only. Other EDC biomarkers were generally not associated with glucose or insulin traits during pregnancy. These findings suggest that DEHP may uniquely affect fasting glucose, marking increased glucose hepatic output, and insulin resistance during pregnancy. The short- and long-term health implications of such potential impact in women and their offspring need to be evaluated.

• DEHP metabolites were positively associated with fasting glucose at mid-pregnancy.

• DEHP metabolites were positively associated with HOMA2-IR at mid-pregnancy.

• Results were generally consistent between single- and multi-pollutant models.

• Few other endocrine disruptors were associated with glucose-insulin traits.

Biography

Mia Peng is currently an employee at the US Food and Drug Administration, Center for Tobacco Products (FDA/CTP). This work was conducted when Dr. Peng was employed at the University of Colorado. Although Dr. Peng is an FDA/CTP employee, this work was not done as part of her official duties. This publication reflects the views of the authors and should not be construed to reflect the FDA/CTP’s views or policies.