Assessing urinary phenol and paraben mixtures in pregnant women with and without gestational diabetes mellitus: a case-control study

Abstract

Prior studies have identified the associations between environmental phenol and paraben exposures and increased risk of gestational diabetes mellitus (GDM), but no study addressed these exposures as mixtures. As methods have emerged to better assess exposures to multiple chemicals, our study aimed to apply Bayesian kernel machine regression (BKMR) to evaluate the association between phenol and paraben mixtures and GDM.

This study included 64 GDM cases and 237 obstetric patient controls from the University of Oklahoma Medical Center. Mid-pregnancy spot urine samples were collected to quantify concentrations of bisphenol A (BPA), benzophenone-3, triclosan, 2,4-dichlorophenol, 2,5-dichlorophenol, butylparaben, methylparaben, and propylparaben. Multivariable logistic regression was used to evaluate the associations between individual chemical biomarkers and GDM while controlling for confounding. We used probit implementation of BKMR with hierarchical variable selection to estimate the mean difference in GDM probability for each component of the phenol and paraben mixtures while controlling for the correlation among the chemical biomarkers.

When analyzing individual chemicals using logistic regression, benzophenone-3 was positively associated with GDM [adjusted odds ratio (aOR) per interquartile range (IQR) = 1.54, 95% confidence interval (CI) 1.15, 2.08], while BPA was negatively associated with GDM (aOR 0.61, 95% CI 0.37, 0.99). In probit-BKMR analysis, an increase in z-score transformed log urinary concentrations of benzophenone-3 from the 10th to 90th percentile was associated with an increase in the estimated difference in the probability of GDM (0.67, 95% Credible Interval 0.04, 1.30), holding other chemicals fixed at their medians. No associations were identified between other chemical biomarkers and GDM in the BKMR analyses.

We observed that the association of BPA and GDM was attenuated when accounting for correlated phenols and parabens, suggesting the importance of addressing chemical mixtures in perinatal environmental exposure studies. Additional prospective investigations will increase the understanding of the relationship between benzophenone-3 exposure and GDM development.

1. Introduction

Gestational diabetes mellitus (GDM), defined as glucose intolerance with onset or first recognition during pregnancy (Carpenter and Coustan, 1982), is a common pregnancy complication that can lead to various adverse perinatal and maternal outcomes, including spontaneous abortion, fetal anomalies, intrauterine fetal demise, and preeclampsia (American Diabetes Association (ADA), 2016). Additionally, the long-term impact of GDM after the obstetrical and neonatal periods is recognized as mothers and their offspring are at increased risk of obesity and subsequent type 2 diabetes (T2DM) (Damm et al., 2016). According to the 2016 National Vital Statistics Birth Data, the crude national prevalence of GDM in the United States (US) was 6.0% (Deputy et al., 2018). In addition, diagnoses of GDM increased from age-standardized estimates of 47.6 to 63.5 per 1000 live births from 2011 to 2019 among nulliparous women aged 15 to 44 years with a singleton live birth in the US (Shah et al., 2021). Although maternal age, race/ethnicity, diet, and obesity are established risk factors for GDM (Zhang et al., 2016), the etiology of GDM remains unclear. As diabetes pathogenesis is potentially accelerated under the diabetogenic conditions of pregnancy (Torgersen and Curran, 2006), further clarity is needed regarding the possible involvement of multiple environmental chemicals in GDM development during this critical window of susceptibility.

Usage of phenols and parabens in food production, personal and home care, and medical-care products is ubiquitous (Sargis and Simmons, 2019). By acting as endocrine-disrupting chemicals (EDCs), these chemicals may play an important role in interfering with glucose homeostasis through various potential mechanisms during the susceptible period of pregnancy; therefore, they are suspected risk factors for GDM (Gaston et al., 2020). For example, animal studies have revealed that phenols and parabens alter the activity or expression of the peroxisome proliferator-activated receptors (PPARs), affecting adipogenesis and energy balance (Taxvig et al., 2012). In addition, phenols and parabens participate in controlling microRNA expression in trophoblast cells and sequentially induce exosome signaling from the placenta, which mediates GDM development (Ehrlich et al., 2016). Numerous epidemiological studies have examined the associations of certain phenols and parabens with elevated pregnancy glucose levels (Bellavia et al., 2018; Bellavia et al., 2019; Chiu et al., 2017; Wang et al., 2020) and GDM (Fisher et al., 2018; Li et al., 2019; Liu et al., 2019; Ouyang et al., 2018; Robledo et al., 2013; Shapiro et al., 2018; Shapiro et al., 2015; Wang et al., 2017; Zhang et al., 2019), with inconclusive results. However, most studies have only examined the associations of one chemical with GDM at a time. In reality, pregnant women are exposed to multiple environmental chemicals (Wang et al., 2021). Therefore, exploring the health impact of individual chemicals does not adequately represent the complexity of realistic exposures.

Although assessments of exposure to multiple chemicals have emerged, no study to date has explored the associations between phenol and paraben mixtures and the perinatal diagnosis of GDM. Pregnant women may be more susceptible to EDC-induced changes in pancreatic islets under the influence of pregnancy hormones (Sargis and Simmons, 2019). Thus, understanding the effect of phenols and parabens on GDM is of interest. In the present study, we aimed to implement a novel probit extension of Bayesian kernel machine regression (BKMR) to investigate the association of phenol and paraben mixtures with GDM. To compare results with prior studies evaluating individual chemical exposure biomarkers, our study additionally evaluated the associations between individual phenols and parabens and GDM using conventional logistic regression models.

2. Materials and Methods

2.1. Study design and population

In this clinic-based case-control study, we enrolled 314 pregnant women who received prenatal care at the University of Oklahoma Medical Center Women’s Clinic and High-Risk Pregnancy Clinics between August 2009 and May 2010. As described previously (Chen et al., 2021), the women included in the study were aged 18 years or older, attended their first clinic visit following glucose screening, did not report a pre-existing diagnosis of type 1 or type 2 diabetes, spoke English or Spanish, and resided within the nine counties surrounding the clinic. Restricting cases and controls to clinic patients who lived in the catchment area aimed to select controls from the same source population that generated the cases. Our analyses were restricted to 301 pregnant women who had stored urine available for quantification of phenol and paraben concentrations. The characteristics of the study participants did not differ between women who did and did not have urine specimens available.

The GDM diagnosis was assessed based on a two-step approach (Carpenter and Coustan, 1982), 1-hour 50-g glucose challenge test (GCT) followed by 3-hour 100-g oral glucose tolerance test (OGTT), which was in accordance with the clinical practice guidelines followed by the recruitment clinics when the study was conducted. Pregnant women were identified as having GDM (n = 64) if their GCT screening value was ≥ 200 mg/dL; or had an initial GCT screening ≥ 135 mg/dL, followed by the OGTT screening with two or more values exceeding 95 mg/dL, 180 mg/dL, 165 mg/dL, and 145 mg/dL, at fasting, 1-hour, 2-hour, 3-hour post-OGTT, respectively (Carpenter and Coustan, 1982). Pregnant women who tested negative for GDM were concurrently selected as controls (n = 237). All participants provided informed consent for their urine specimens to be collected and stored for analysis in this study. This study was approved by the Institutional Review Board of the University of Oklahoma Health Sciences Center [parent study participation and storage of specimens for future research (#14738, 7/16/2009) and current project involving secondary data analysis (#2477, 7/27/2010)]. The analysis of de-identified specimens at the Centers for Disease Control and Prevention (CDC) laboratory was determined not to constitute engagement in human subjects’ research.

2.2. Quantification of urinary phenols and parabens

Mid-pregnancy urine samples (median: 26.0 gestational weeks) were collected at enrollment in sterile polypropylene containers and stored at −20 °C. In November 2011, the frozen urine samples were shipped overnight to the CDC’s National Center for Environmental Health, Division of Laboratory Sciences for analysis. As described previously (Ye et al., 2006; Ye et al., 2005), urinary concentrations of the following five phenols [bisphenol A (BPA), benzophenone-3, triclosan, 2,4-dichlorophenol, 2,5-dichlorophenol] and three parabens (butylparaben, methylparaben, and propylparaben) were quantified by online solid-phase extraction coupled with isotope dilution-high-performance liquid chromatography-tandem mass spectrometry. The limits of detection (LODs) were 0.2 (2,4-dichlorophenol, 2,5-dichlorophenol, butylparaben, and propylparaben), 0.4 (BPA and benzophenone-3), 1.0 (methylparaben), and 2.3 (triclosan) ng/mL. A value equal to the LOD/√2 was substituted for biomarker concentrations below the LOD (Hornung and Reed, 1990).

Specific gravity (SG) was measured by a calibrated hand-held refractometer at the time of urine sample collection to adjust for variations in urine diluteness. We adjusted the urinary concentrations based on the approach that incorporates both covariate-adjusted standardization and the inclusion of SG as a covariate in the regression model (O’Brien et al., 2016). The formula for covariate-adjusted standardization is given as SG-adjusted chemical concentration (ng/mL) = observed chemical concentration × [(SG_p−1)/(SG_o−1)], where SG_p and SG_o are the predicted SG and observed SG for each observation, respectively (Kuiper et al., 2021). SG_p was estimated by employing a prediction model with SG_o, with the following factors known to affect SG: age, race/ethnicity, and pre-pregnancy body mass index (BMI) (Kuiper et al., 2021).

2.3. Covariates Measurement

All participants completed a short interview at enrollment for collection of data on demographic, behavioral, and medical characteristics. The factors evaluated included maternal age, race/ethnicity, educational level, parity, and history of GDM. Pre-pregnancy BMI (kg/m²) was calculated using self-reported pre-pregnancy height and weight. Alcohol consumption (yes/no) was defined as having ever consumed an alcoholic beverage during the first trimester. Pregnant women were defined as active smokers (yes/no) if they self-reported as currently smoking or had a urinary cotinine concentration greater than 15 ng/mL (Benowitz et al., 2009). The detailed measures of each covariate have been described previously (Chen et al., 2021).

2.4. Statistical Analyses

Demographic characteristics, lifestyle factors, and GDM history were summarized using percentages. Comparisons of characteristics between GDM cases and controls were conducted using Chi-square tests. The SG-adjusted urinary concentrations of phenols and parabens of cases and controls were summarized using geometric mean (GM) of concentrations with 95% confidence interval (CI) and 25th, 50th, and 75th percentiles. Differences in the geometric mean between cases and controls were evaluated using Student’s t-tests of the arithmetic means of the log-transformed values. Pairwise correlations among the five phenols and three parabens were assessed using Pearson’s correlation for the log-transformed values.

Multivariable logistic regression models were applied to estimate odds ratios (ORs) and 95% CIs for the associations of individual SG-adjusted urinary concentrations of phenols and parabens with GDM while controlling for potential confounders. Both continuous and categorical measures of individual biomarker concentrations were assessed in the models. For continuous measures, we evaluated the ORs per interquartile range (IQR) increase in the log-transformed concentrations of each chemical. For categorical measures, the urinary concentrations were classified into tertiles based on the control distribution, and the lowest tertile was used as the reference group. Due to the relatively low detection frequency of butylparaben, we classified butylparaben concentrations as non-detectable, being below the median of detectable concentrations among controls, or greater than or equal to the median of detectable concentrations. The linear trend of ORs was determined by treating each tertile as an ordinal variable in the models and reported its p-value. Since prior studies have suggested that pre-pregnancy BMI may modify the associations of certain phenols and parabens with GDM (Bellavia et al., 2018; Li et al., 2019; Liu et al., 2019; Wang et al., 2020; Zhang et al., 2019), we explored this potentially differential effect by conducting a stratified analysis by the obesity status.

This study used the manual forward selection approach in multivariable modeling to evaluate the potential confounding variables (Lash et al., 2021), since the number of events in this study population was too limited to evaluate a model that included all covariates simultaneously. The forward selection procedures start with a minimal model that includes only the exposure variable, followed by subsequent models that systematically add covariates to evaluate the evidence for confounding. Covariates were defined as confounders when the exposure ORs changed by approximately 10% or beyond when comparing the models with and without adjustment for the selected factor. To maximize the number of covariates that could be supported by the number of GDM events in our logistic regression models, we selected the covariates with the fewest categories if the change in the exposure OR was similar for different specifications of the same covariate. We identified the following variables as potential confounders of the associations between phenols, parabens, and GDM: maternal age (continuous measure), race/ethnicity (Non-Hispanic White/Others, Non-Hispanic Black, or Hispanic), education level (less than high school versus high school graduate or beyond), pre-pregnancy BMI (obese or not obese), and history of GDM (nulliparous or parous without GDM history versus parous with GDM history).

The effect of exposure to multiple phenols and parabens on GDM was performed using BKMR, a nonparametric approach for estimating multiexposure mixtures that flexibly models the joint effect of the mixture using the kernel function (Bobb et al., 2018; Bobb et al., 2015; Valeri et al., 2017). BKMR can be extended to analyze the association between mixtures and a binary outcome by implementing the probit link function. Using probit regression in BKMR (probit-BKMR), we can estimate the exposure-response function between biomarker concentrations and a latent continuous outcome, a continuous marker of the binary outcome, indicating the estimated difference in the probability of the outcome (Bobb et al., 2018). To accommodate correlation structures within phenols and parabens, we fitted probit-BKMR with a hierarchical variable selection approach to account for the correlation of the mixture components. According to the results of the exploratory factor analysis (EFA), we classified the phenols and parabens into five groups: dichlorophenols (2,4-dichlorophenol and 2,5-dichlorophenol), parabens (butylparaben, methylparaben, and propylparaben), BPA, benzophenone-3, and triclosan, respectively. The kernel function was specified as the Gaussian kernel when fitting the probit-BKMR models in the present study (Bobb et al., 2015; Valeri et al., 2017). We fitted the probit-BKMR model using the Markov chain Monte Carlo (MCMC) algorithm with 100,000 iterations in the R package ‘bkmr’. To ensure the comparability of the exposure variables, all log-transformed concentrations of SG-adjusted phenols and parabens were standardized to create z-scores when fitting the models. The confounders identified in the models evaluating individual chemical biomarkers were controlled for in the probit-BKMR models. Estimates of the exposure-response function and its uncertainty quantified as credible intervals (CrI) were obtained after fitting the probit-BKMR models. We characterized the estimated exposure-response function as follows: The dose-response relationships of each chemical biomarker with GDM were visualized by plotting the estimated difference in the probability of GDM against each z-score-transformed phenol and paraben concentrations while fixing other chemicals at their median levels. The single-exposure effects of each chemical on GDM were summarized as the estimated difference in the probability of GDM when a particular phenol and paraben concentration was increased from the 10th to the 90th percentile with other phenols and parabens fixed to their 25th, 50th, or 75th percentile. The overall effects of the phenol and paraben mixtures were represented as the estimated difference in the probability of GDM when all the mixtures were at a particular percentile ranging from the 10th to the 90th percentile and compared to the mixtures at the median levels. The interactive effects were presented as bivariate exposure-response functions between a certain phenol and paraben and the estimated difference in the probability of GDM when a second phenol and paraben concentration was at the 10th, 50th, or 90th percentile, while fixing other chemicals at their median levels. A formal interaction test by adding product terms to the logistic regression models was conducted for the presence of potential effect modification.

All statistical analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC, USA) and R version 4.0.2 (R Core Team, 2020). Two-sided p-values < 0.05 were considered statistically significant.

3. Results

3.1. Basic characteristics of study participants

Table 1 displays the demographics, lifestyle factors, and GDM history of the study participants. GDM cases and controls differed in age, race/ethnicity, education level, pre-pregnancy BMI, smoking status, and history of GDM. The age range of the GDM cases was 19-41 years, with the majority (37.5%) aged 30-34 years; the age range for the controls was 18-44 years, with most (56.5%) below 25 years of age. More than half of the cases had pre-pregnancy BMI classified as obese (52.4%). Most cases were Hispanic and less likely to be current smokers compared to the controls. Among the parous women, approximately one-third of cases had a prior history of GDM (31.0%), which was a higher percentage than the controls (4.9%) (p < 0.01).

Table 1.

Distribution of demographics, lifestyle factors, and GDM history for cases and controls.

Variables	Cases (n = 64) n (%)	Controls (n = 237) n (%)	P-value c
Maternal age (years)			< 0.01
< 25	18 (28.1)	134 (56.5)
25 – 29	12 (18.8)	63 (26.6)
30 – 34	24 (37.5)	24 (10.1)
≥ 35	10 (15.6)	16 (6.8)
Race/Ethnicity			< 0.01
Non-Hispanic White	16 (25.0)	70 (29.7)
Non-Hispanic Black	6 (9.4)	79 (33.4)
Hispanic	40 (62.5)	70 (29.7)
Native American	2 (3.1)	17 (6.8)
Asian	0	1 (0.4)
Educational level			0.02
Less than high school	32 (50.0)	73 (30.8)
High school	18 (28.1)	89 (37.6)
More than high school	14 (21.9)	75 (31.7)
Pre-pregnancy BMI (kg/m2) a			< 0.01
Normal (≤ 24.99)	13 (20.6)	107 (45.5)
Overweight (25.0-29.99)	17 (27.0)	52 (22.1)
Obese (≥ 30.0)	33 (52.4)	76 (32.3)
Alcohol consumption			0.65
No	58 (90.6)	210 (88.6)
Yes	6 (9.4)	27 (11.4)
Active smoker c			0.04
No	55 (85.9)	175 (73.8)
Yes	9 (14.1)	62 (26.2)
Self-reported history of GDM			< 0.01
Nulliparous	13 (20.3)	55 (23.2)
Parous without GDM history	35 (54.7)	173 (73.0)
Parous with GDM history	16 (25.0)	9 (3.8)

^aPre-pregnancy BMI data were missing for one case and two controls.

^bSelf-reported current smoking status or urinary cotinine levels ≥ 15 ng/mL for active smokers.

^cP-value based on the Chi-square test.

3.2. Urinary concentrations of phenols and parabens

The percentage of detectable concentrations, GM (95% CI), and percentiles distribution of SG-adjusted urinary concentrations of phenols and parabens for cases and controls are summarized in Table 2. The detection frequency ranged from 93.3% for BPA to 100% for 2,5-dichlorophenol, except for butylparaben (51.7%) and triclosan (73.2%). The urinary concentration of BPA was lower among the GDM cases compared to controls [GM (95% CI): 1.85 (1.51, 2.27) vs. 2.58 (2.29, 2.91) ng/mL, p = 0.01], whereas GDM cases had higher benzophenone-3 than controls [GM (95% CI): 48.17 (28.25, 82.16) vs. 18.60 (14.89, 23.24) ng/mL, p < 0.01]. Pearson’s correlations between the log-transformed SG-adjusted urinary concentrations of phenols and parabens are shown in Supplementary Figure 1. There was a high correlation between 2,4-dichlorophenol and 2,5-dichlorophenol (r = 0.90, p < 0.01). Among the parabens, butylparaben was moderately correlated with methylparaben (r = 0.32, p < 0.01) and propylparaben (r = 0.34, p < 0.01), while methylparaben was strongly correlated with propylparaben (r = 0.78, p < 0.01).

Table 2.

The percentage of detectable concentrations, geometric mean (ng/mL) and distribution percentiles of SG-adjusted urinary concentrations (ng/mL) of phenols and parabens for cases and controls (2009-2010).

Chemicals	Detection frequency (%) a	Cases				Controls				P-value b
		GM (95% CI)	P25	P50	P75	GM (95% CI)	P25	P50	P75
Phenols
Bisphenol A	93.3	1.85 (1.51, 2.27)	1.04	2.05	2.91	2.58 (2.29, 2.91)	1.42	2.33	4.53	0.01
Benzophenone-3	99.7	48.17 (28.25, 82.16)	9.71	31.45	225.65	18.60 (14.89, 23.24)	5.88	12.62	36.64	< 0.01
Triclosan	73.2	18.04 (10.55, 30.83)	2.78	13.72	136.24	11.89 (9.59, 14.75)	3.13	9.16	35.20	0.15
2,4-Dichlorophenol	93.6	1.68 (1.15, 2.44)	0.72	1.20	2.17	1.54 (1.29, 1.84)	0.65	1.15	2.37	0.67
2,5-Dichlorophenol	100	19.04 (11.36, 31.89)	3.62	10.62	42.21	20.47 (16.02, 26.15)	5.10	13.44	47.24	0.79
Parabens
Butylparaben	51.7	0.61 (0.40, 0.95)	0.13	0.33	2.61	0.65 (0.52, 0.83)	0.16	0.36	1.87	0.80
Methylparaben	99.3	132.43 (92.91, 188.80)	76.37	167.0	325.12	169.93 (139.60, 206.80)	69.58	193.56	524.90	0.24
Propylparaben	99.0	26.78 (16.36, 43.84)	7.08	53.53	102.04	27.81 (21.58, 35.82)	7.21	35.75	119.88	0.89

Abbreviations: SG, specific gravity; GM, geometric mean; CI, confidence interval.

Three subjects (1 case and 2 controls) were removed from analysis as missing for pre-pregnancy BMI.

^aThe limits of detection were 0.2 (2,4-dichlorophenol, 2,5-dichlorophenol, butylparaben, and propylparaben), 0.4 (BPA and benzophenone-3), 1.0 (methylparaben), and 2.3 (triclosan) ng/mL.

^bP-value from Student’s t-test of the arithmetic means of the log-transformed values between cases and controls.

3.3. Logistic regression analysis of individual phenols, parabens, and GDM

The associations between SG-adjusted urinary concentrations of individual phenols, parabens, and GDM were evaluated in Table 3. A reduced odds of elevated BPA was observed in women with GDM (adjusted OR per IQR = 0.61, 95% CI 0.37, 0.99). When urinary concentrations were categorized into tertiles, the odds of being in the highest BPA tertile decreased by 51% among cases compared to controls (adjusted OR = 0.49, 95% 0.19, 1.25), though the 95% CI crossed the null. In contrast, benzophenone-3 was positively associated with GDM (adjusted OR per IQR = 1.54, 95% CI 1.15, 2.08). The odds of having benzophenone-3 concentrations in the highest tertile were more than three times greater among the cases than the controls (adjusted OR for the highest benzophenone-3 tertile = 3.82, 95% CI 1.60, 9.08 relative to the lowest tertile). The direction of observed associations of BPA and benzophenone-3 with GDM remained consistent when stratified by the obesity status (Supplementary Table 1); however, the stratified estimates lacked precision due to the smaller sample size in subgroup analyses. Evidence of associations with GDM were not observed for triclosan, dichlorophenols, or parabens.

Table 3.

Associations between SG-adjusted urinary concentrations of phenols, parabens, and GDM.

Phenols and parabens (ng/mL)	Cases n (%)	Controls n (%)	Crude OR (95% CI) a	Adjusted OR (95% CI) b
Bisphenol A
Per IQR			0.60 (0.40, 0.89)	0.61 (0.37, 0.99)
≤ 1.66	25 (39.7)	78 (33.2)	1.00	1.00
1.67 – 3.34	26 (41.3)	78 (33.2)	1.05 (0.56, 1.98)	1.03 (0.49, 2.17)
≥ 3.35	12 (19.0)	79 (33.6)	0.48 (0.22, 1.03)	0.49 (0.19, 1.25)
			P trend = 0.08	P trend = 0.18
Benzophenone-3
Per IQR			1.57 (1.22, 2.03)	1.54 (1.15, 2.08)
≤ 7.40	11 (17.5)	78 (33.2)	1.00	1.00
7.41– 22.0	14 (21.2)	78 (33.2)	1.28 (0.55, 3.03)	1.20 (0.44, 3.23)
≥ 22.01	38 (60.3)	79 (33.6)	3.41 (1.63, 7.15)	3.82 (1.60, 9.08)
			P trend < 0.01	P trend < 0.01
Triclosan
Per IQR			1.36 (0.94, 1.97)	1.41 (0.93, 2.13)
≤ 4.71	24 (38.1)	78 (33.2)	1.00	1.00
4.72 – 22.68	15 (23.8)	78 (33.2)	0.62 (0.30, 1.27)	0.66 (0.28, 1.53)
≥ 22.69	24 (38.1)	79 (33.6)	1.00 (0.52, 1.90)	1.14 (0.54, 2.42)
			P trend = 0.99	P trend = 0.75
2,4-Dichlorophenol
Per IQR			1.06 (0.83, 1.35)	1.06 (0.80, 1.41)
≤ 0.80	18 (28.6)	78 (33.2)	1.00	1.00
0.81 – 1.83	29 (46.0)	78 (33.2)	1.63 (0.84, 3.18)	1.98 (0.90, 4.36)
≥ 1.84	16 (25.4)	79 (33.6)	0.88 (0.42, 1.85)	1.03 (0.44, 2.43)
			P trend = 0.75	P trend = 0.91
2,5-Dichlorophenol
Per IQR			0.96 (0.68, 1.35)	0.92 (0.61, 1.37)
≤ 6.65	25 (39.7)	78 (33.2)	1.00	1.00
6.66 – 31.69	18 (28.6)	78 (33.2)	0.71 (0.36, 1.41)	0.80 (0.36, 1.81)
≥ 31.70	20 (31.7)	79 (33.6)	0.79 (0.41, 1.54)	0.78 (0.35, 1.76)
			P trend = 0.48	P trend = 0.55
Butylparaben
Per IQR			0.94 (0.63, 1.40)	1.16 (0.73, 1.85)
Non-detectable	33 (52.4)	111 (47.2)	1.00	1.00
≤ 1.81	13 (20.6)	62 (26.4)	0.72 (0.35, 1.47)	0.96 (0.41, 2.24)
≥ 1.82	17 (27.0)	62 (26.4)	0.93 (0.48, 1.81)	1.53 (0.68, 3.43)
			P trend = 0.73	P trend = 0.34
Methylparaben
Per IQR			0.81 (0.56, 1.17)	0.83 (0.53, 1.29)
≤ 102.55	21 (33.3)	78 (33.2)	1.00	1.00
102.56 – 383.74	30 (47.6)	78 (33.2)	1.43 (0.75, 2.71)	1.35 (0.63, 2.85)
≥ 383.75	12 (19.1)	79 (33.6)	0.57 (0.26, 1.23)	0.58 (0.24, 1.39)
			P trend = 0.20	P trend = 0.28
Propylparaben
Per IQR			0.97 (0.66, 1.46)	1.13 (0.70, 1.83)
≤ 14.12	20 (31.7)	78 (33.2)	1.00	1.00
14.13 – 79.22	20 (31.7)	78 (33.2)	1.00 (0.50, 2.00)	1.13 (0.49, 2.60)
≥ 79.23	23 (36.6)	79 (33.6)	1.15 (0.58, 2.26)	1.54 (0.69, 3.46)
			P trend = 0.69	P trend = 0.29

Abbreviations: SG, specific gravity; OR, odds ratio; CI, confidence interval.

^aAdjusted for specific gravity.

^bAdjusted for specific gravity, age, race/ethnicity, education level, pre-pregnancy BMI, and history of GDM.

3.4. Probit-BKMR analysis of phenol and paraben mixtures and GDM

Figure 1 presents the estimated exposure-response functions of each phenol and paraben and its 95% CrI. We observed a positive dose-response relationship between benzophenone-3 and GDM, with the probability of GDM increasing with increasing concentrations of benzophenone-3. Although the relationship between triclosan and GDM was positive as well, the null was included in the 95% CrI at the highest concentration. BPA and methylparaben showed inverse relationships with GDM. There was little evidence of the associations between 2,4-dichlorophenol, 2,5-dichlorophenol, butylparaben, propylparaben, and GDM. Overall, the CrI was wide in each estimated single exposure-response function, which represented the high variability of these relationships. Figure 2 displays the associations of individual phenols and parabens with the estimated GDM probability difference when fixing all other phenols and parabens concentrations to their 25th, 50th, and 75th percentiles. We found that an increase in z-score transformed log urinary concentrations of benzophenone-3 from the 10th to 90th percentile corresponded to a 0.68 (95% CrI 0.0, 1.36), 0.67 (95% CrI 0.04, 1.30), and 0.63 (95% CrI −0.01, 1.28) increase in the estimated GDM probability when the other phenols and parabens were fixed to their 25th, 50th, and 75th percentiles. However, there was no evidence of GDM associations for the rest of the phenols and parabens while accounting for the confounding effect of other phenols and parabens. The estimated overall effect of the phenol and paraben mixtures on GDM was lacking as the credible intervals for these effect estimates at each particular percentile of the mixtures crossed the null (Supplementary Figure 2). Additionally, a potential interaction was observed between benzophenone-3 and methylparaben when other phenols and parabens were fixed at their median concentrations, with the positive association between benzophenone-3 and the estimated GDM probability attenuated with higher concentrations of methylparaben (Supplementary Figure 3). However, the precision of the product term of benzophenone-3 and methylparaben was limited (p for interaction = 0.26). The slopes of the other bivariate exposure-response function for a given phenol or paraben were parallel at different percentiles of the second phenol or paraben, which indicated no evidence of an interaction.

Figure 1. The estimated single exposure-response functions and 95% CrI for each phenol and paraben when all other phenols and parabens were fixed at their median levels.

The probit-BKMR model was adjusted for specific gravity, age, race/ethnicity, education level, pre-pregnancy BMI, and history of GDM.

Figure 2. The estimated effect and 95% CrI of each phenol and paraben on GDM when each phenol and paraben was increased from the 10th to the 90th percentile of its distribution while other phenols and parabens were fixed to their 25th, 50th, or 75th percentile.

The probit-BKMR model was adjusted for specific gravity, age, race/ethnicity, education level, pre-pregnancy BMI, and history of GDM.

4. Discussion

In the present study, we examined the individual and mixture effects of urinary concentrations of select phenols and parabens on GDM. Using logistic regression models for assessing each individual chemical biomarker, we determined that benzophenone-3 was positively associated with GDM, whereas BPA was inversely linked to GDM. However, using probit-BKMR to consider the correlation among the phenols and parabens examined, we only consistently observed that higher urinary concentrations of benzophenone-3 were associated with an increased estimated probability of GDM. In addition, we observed no evidence of the overall effects of the phenol and paraben mixtures on GDM. To the best of our knowledge, this study is the first to evaluate the associations of phenol and paraben mixtures with GDM.

BPA is a widespread contaminant commonly used in the manufacture of numerous plastic consumer products made of epoxy resins and polycarbonate (Sargis and Simmons, 2019). As BPA is an endocrine disruptor that humans are widely exposed to in daily life, the associations between urinary concentrations of BPA and pregnancy glucose levels and GDM have been examined in numerous studies; however, findings remain inconsistent. Among subfertile women in the Environment and Reproductive Health (EARTH) study from Massachusetts, Chiu et al. (2017) observed a positive association between concentrations of BPA and glucose levels. In a term birth pregnancy cohort of the LIFECODES study in Boston, Bellavia et al. (2018) found that BPA concentrations (1.32-2.10 ng/mL) were associated with higher glucose levels but only among women with obesity. Conversely, in a Chinese cohort based in Wuhan, Zhang et al. (2019) reported that BPA concentrations were associated with decreased glucose levels among women with obesity although the confidence intervals included the null value. By addressing the clinical diagnosis of GDM as a primary outcome, several studies reported no association between BPA and GDM (Fisher et al., 2018; Shapiro et al., 2015; Zhang et al., 2019). In a prior assessment of BPA in a small subset of the same population evaluated for the current study, the observed odds of GDM were lower with increasing tertiles, but the confidence intervals for these estimates were wide and included the null value (Robledo et al., 2013). A Chinese cohort study based in Shanghai reported that the odds of GDM decreased by 27% for per unit increase in the log-transformed concentrations of BPA (Wang et al., 2017). Additionally, the odds of GDM for the second (0.82-1.83 ng/mL) and third (1.83-154.60 ng/mL) tertiles of BPA were lower than that of the lowest tertile (< 0.1-0.82 ng/mL) by 52% and 60%, respectively. Results from Wang et al. (2017) were consistent with our findings of the individual exposure models that observed a decreased odds for GDM by 40% per IQR increase in the log-transformed concentrations of BPA. Although we also observed that the odds of being in the highest tertile of BPA (3.35-68.19 ng/mL) decreased by 51% in the cases compared to the controls, this observed association was attenuated and included the null value. Our study population had slightly higher SG-adjusted urinary concentrations of BPA (GM: 2.58 ng/mL) than that of the female population in the National Health and Nutrition Examination Survey (NHANES) 2009-2010 and pregnancy cohorts from the USA, Canada, and China (GM range: 0.87-1.73 ng/mL) (Bellavia et al., 2018; Centers for Disease Control and Prevention (CDC), 2022; Chiu et al., 2017; Shapiro et al., 2015; Zhang et al., 2019). Although different adjustment methods were applied for adjusting the urinary hydration status, notably the distribution of urinary BPA concentrations in the study by Wang et al. (2017) [GM (P25, P75) (μg/g creatinine): 2.40 (1.47, 4.36)] was comparable to that in our study [GM (P25, P75) (SG-adjusted ng/mL): 2.33 (1.42, 4.53)], which may reflect similar exposure patterns in these populations. In the mixture analysis, we observed a similar negative trend of the estimated exposure-response function of BPA and GDM as observed in individual analysis; however, the 95% CrI did not exclude the null value when increasing BPA concentrations. Findings from an animal study showed that exposure to BPA may induce increased insulin, leading to the development of chronic hyperinsulinemia and resulting in the alteration of glucose homeostasis (Alonso-Magdalena et al., 2006). In gestational animal models, pregnant mice treated with BPA at doses of either 10 or 100 μg/kg/day were associated with decreased glucose tolerance and increased concentrations of plasma insulin compared to control mice (Alonso-Magdalena et al., 2010). Although experimental cell research has demonstrated environmental relevant doses of BPA (1 nM) can enhance the glucose-induced insulin secretion in human pancreatic islet cells (Soriano et al., 2012), no other study has evaluated the involvement of environmentally relevant doses of BPA in the association between a rapid increase in plasma insulin and adverse effects on glucose metabolism; the possibility of this association needs to be investigated in future studies.

Benzophenone-3 is an ultraviolet blocker widely used in sunscreens, cosmetics, and personal care products, and plastic surface coatings for food packaging (Dodson et al., 2020). Few epidemiological studies have examined the association between exposure to benzophenone-3 and GDM. Among subfertile women in the EARTH study, Wang et al. (2020) reported an inverse association between first-trimester urinary concentrations of benzophenone-3 and glucose levels. This study additionally found that the odds of impaired glucose tolerance (GCT ≥ 140 mg/dL) decreased by 88% among women with second-trimester benzophenone-3 in the third quartile (170.7-811.8 ng/mL) compared to women in the lowest quartile (< LOD-36.3 ng/mL). However, one study from the United Kingdom (UK) reported a null association between the serum concentrations of benzophenone-3 and GDM (Fisher et al., 2018). Inconsistent with previous findings, our results showed a positive association between benzophenone-3 and GDM in the individual exposure biomarkers analysis. This positive association was consistently observed when accounting for the effect of other phenols and parabens in our mixture analysis. The inconsistency between our results and those of previous studies may relate to the differences in distribution of urinary benzophenone-3 concentrations in the study populations. The women in our study population exhibited slightly lower urinary concentrations of benzophenone-3 compared to the female population in NHANES 2009-2010 (GM: 18.60 vs. 32.0 ng/mL) (Centers for Disease Control and Prevention (CDC), 2022). However, in the EARTH cohort, the urinary concentrations of benzophenone-3 found in subfertile women during preconception and first- and second-trimesters were approximately 8-fold higher than those found in our study population (Wang et al., 2020). With substantially higher concentrations of benzophenone-3 in the EARTH study, their referent group (lowest quartile: < LOD-36.3 ng/mL) was characterized by relatively higher concentrations than the referent group representing low exposure in our study population (< LOD-7.40 ng/mL). Prior studies have found that benzophenone-3 was associated with altered thyroid hormone levels (Aker et al., 2018) and increased oxidative stress (Ferguson et al., 2019) during pregnancy, contributing to adverse pregnancy outcomes. Evidence from an in vivo study showed that female rats administered with benzophenone-3 dermally during pregnancy had increased extracellular glutamate concentration and enhanced lipid peroxidation (Skórkowska et al., 2020). Both extracellular glutamate and lipid peroxidation play an important role in the pathogenesis of diabetes (Takahashi et al., 2019; Tangvarasittichai, 2015). Another gestational animal study found that exposure to benzophenone-3 may lead to dysregulation of metabolites in the tricarboxylic acid cycle, followed by increased xanthurenic acid levels and an impact on glucose homeostasis (Han et al., 2022). Additional animal studies are warranted to understand the mechanism underlying the effect of benzophenone-3 on the abnormal changes in pregnancy glucose levels.

Triclosan is a chlorinated antimicrobial chemical that has been widely used in personal care products such as hand washes, toothpaste, and deodorants (Dodson et al., 2020). Experimental studies using gestational animal models have shown that the activity of placental glucose transporters and serum thyroxine levels were reduced, leading to insulin resistance, in pregnant mice exposed to a high dose of triclosan (8 mg/kg/day) compared to that in the control mice (Cao et al., 2017; Hua et al., 2017). A recent study demonstrated that exposure to triclosan during pregnancy was associated with increased oxidative stress among US pregnant women (Ferguson et al., 2019). Epidemiological studies evaluating the effect of triclosan on GDM are limited. We found limited evidence of association between the urinary concentrations of triclosan and GDM in individual exposure analysis, which is consistent with the results from Canadian and Chinese studies (Ouyang et al., 2018; Shapiro et al., 2018). Although we observed a positive relationship between the estimated exposure-response function of triclosan and GDM in our mixture analysis, the width of the 95% credible interval increased with increasing concentration and did not exclude the null value. However, this result contradicted the findings of a study from the UK that observed an inverse association between the serum concentrations of triclosan and GDM (Fisher et al., 2018). The discrepancy between these results may be because triclosan was assessed using different matrices. As non-persistent chemicals metabolize quickly in the human body, the levels of their hydrophilic metabolites in the blood are substantially lower than in urine (Calafat et al., 2013). The low levels measured in the blood are more susceptible to extraneous sources and may not reflect actual daily habits of exposure (Calafat et al., 2013).

2,4-Dichlorophenol and 2,5-dichlorophenol, which are used in herbicides and pesticides, can be found in contaminated drinking water and may be linked to immunotoxicity or carcinogenicity (Dodson et al., 2020). Pregnant women with high concentrations of 2,4-dichlorophenol and 2,5-dichlorophenol have been found to have increased oxidative stress and changes in thyroid hormone levels (Aker et al., 2018; Berger et al., 2018; Ferguson et al., 2019). In our study, we observed no association between 2,4-dichlorophenol, 2,5-dichlorophenol, and GDM. Only one study has evaluated the associations of urinary concentrations of 2,4-dichlorophenol and 2,5-dichlorophenol with diabetes-related outcomes. Results of a cross-sectional study using data from the 2007-2010 NHANES showed a positive dose-dependent association between increasing quartiles of 2,5-dichlorophenol and T2DM, and no association was found between 2,4-dichlorophenol and T2DM (Wei and Zhu, 2016). However, this study was conducted in the general population instead of pregnant women. Additional studies can help better understand the potential impact of 2,4-dichlorophenol and 2,5-dichlorophenol in dysglycemia during pregnancy and its mechanisms.

Parabens are a group of esters of p-hydroxybenzoic acid with alkyl substituents that are extensively used as preservatives in personal care products, cosmetics, pharmaceuticals, and food (Błędzka et al., 2014; Dodson et al., 2020). Parabens can potentially impact glucose homeostasis through specific mechanisms. Through operating as estrogenic EDCs, parabens may alter hormone signaling (van Meeuwen et al., 2008). In addition, parabens may stimulate adipocytes to release leptin and adiponectin and affect the adipogenic effect by activating PPARs (Taxvig et al., 2012). Certain parabens were also associated with decreased thyroid hormone levels and increased oxidative stress among pregnant women (Aker et al., 2018; Berger et al., 2018; Ferguson et al., 2019). Among subfertile women in the EARTH study, glucose levels increased with higher urinary concentrations of propylparaben and butylparaben when all parabens were mutually adjusted in the same model (Bellavia et al., 2019). However, according to our individual exposure biomarker analysis, butylparaben, methylparaben, and propylparaben showed no association with GDM, which is in accordance with the results of two Chinese studies conducted in Wuhan (Li et al., 2019; Liu et al., 2019). Our mixture analysis observed an inverse relationship between the estimated exposure-response function of methylparaben and GDM, but the credible intervals crossed the null. The differences in the exposure biomarkers concentrations may be a potential reason for these inconsistent results among studies. The urinary concentrations of methylparaben and propylparaben in our study (GM for methylparaben: 169.93; for propylparaben: 27.81 ng/mL) were higher than those determined in the NHANES and the cohorts of the studies in Wuhan and the EARTH study (GM range for methylparaben: 5.54-124.1; for propylparaben: 0.49-25.3 ng/mL) (Bellavia et al., 2019; Centers for Disease Control and Prevention (CDC), 2022; Li et al., 2019; Liu et al., 2019). Thus, the referent group of lowest tertile in our study would represent relatively higher concentrations than in the Wuhan and the EARTH studies and does not similarly assess the dose-response relationship between methylparaben, propylparaben, and GDM when compared to these lower exposure biomarker concentrations.

Several methodological considerations need to be considered when comparing our results with those of other studies. Chemical exposure assessment in different trimesters may be a possible reason for inconsistent results in studies evaluating the associations of phenols and parabens with GDM. Using the data from the EARTH pregnancy cohort, Chiu et al. (2017) assessed BPA exposure in different trimesters of urine specimens to determine the timing of exposure in relation to blood glucose levels. They observed that increased second-trimester BPA concentrations were associated with evaluated glucose levels. However, no association was found between BPA and glucose levels when measuring biomarkers concentrations in the first-trimester (weeks 1-12). These findings suggest that the second-trimester (weeks 13-26) may be the etiologically relevant time window during which susceptibility to BPA exposures may be related to the pregnancy glucose levels (Chiu et al., 2017). Our study quantified the urinary concentrations of phenols and parabens at the clinical visit immediately following glucose screening during mid-pregnancy. Therefore, the timing of our exposure assessment may have occurred outside the etiologically relevant window for GDM.

The evaluation of the different outcomes for dysglycemia during pregnancy may lead to discrepant results across studies. Several previous studies assessed the association of phenol and paraben exposures with non-fasting GCT glucose levels because they were underpowered with a limited number of GDM events (Bellavia et al., 2018; Bellavia et al., 2019; Chiu et al., 2017; Wang et al., 2020). These studies also evaluated GCT glucose as a binary outcome, which categorized GCT ≥ 140 mg/dL as an indicator of impaired glucose tolerance. Non-fasting GCT can be influenced by the timing of meals and testing (Gupta et al., 2015). Thus, using non-fasting GCT to evaluate impaired glucose tolerance may be more susceptible to misclassification errors. Although our study used the clinical diagnosis of GDM as a primary outcome, it does not exclude some women who may have glucose levels that are higher than normal glucose but not high enough to be diagnosed as GDM (pre-diabetes). Additionally, GDM case identification in our study did not apply the more recent diagnostic criteria adopted by the International Association of Diabetes and Pregnancy Study Groups (IADPSG) (Metzger et al., 2010) because these criteria had not been adopted when our participants were being tested for GDM. When applying the diagnosis criteria by IADPSG, only a single-step OGTT is performed and single abnormal OGTT values are defined as GDM. Thus, cases of GDM may have been missed when using the former two-step approach requiring two or more OGTT values exceeding thresholds. In our study, there were 41 women in the control group who exhibited GCT > 135 mg/dL during the initial screening. After prompting referral of these women for a diagnostic OGTT, only 7 of these women presented a single abnormal value on the 3-hour OGTT. The inclusion of these women with pre-diabetes in the control group may bias observed associations toward the null; however, the numbers of women with pre-diabetes are limited.

Few studies to date have accounted for the confounding effect due to exposure to other related chemicals in the analysis of associations between phenols, parabens, and GDM. In the present study, we observed that certain phenols and parabens were highly correlated with their related chemicals; thus, multicollinearity may be present when including all chemical biomarkers in the linear or logistic regression models at the same time. Multicollinearity would influence regression estimates and bias standard errors (Vatcheva et al., 2016). Using data from the EARTH pregnancy cohort, Bellavia et al. (2019) evaluated the association of exposure to three parabens (methylparaben, propylparaben, and butylparaben) and glucose levels measured by GCT using multiple linear regression and BKMR models. This study reported that the first-trimester urinary concentrations of propylparaben were not associated with glucose levels in the individual paraben model. While mutually adjusting all parabens in the same model, increased propylparaben concentrations were significantly associated with a decreased glucose level with extremely wide confidence intervals. No statistically significant association was observed when further analyzing parabens as a mixture in the BKMR model. These inconsistent results of individual, mutually adjusted, and BKMR models demonstrated the adverse impact of multicollinearity on the estimate and precision of the linear regression model. Our study employed BKMR with a hierarchical variable selection approach to model multiple chemicals and GDM. Thereby, we accounted for the confounding effect across other highly correlated chemicals and avoided multicollinearity and model misspecification, which cannot be handled through the traditional parametric regression approach (Bobb et al., 2018; Bobb et al., 2015; Valeri et al., 2017). So far, limited studies have assessed the nonlinear association between phenols, parabens, and GDM. Using restricted cubic splines, Li et al. (2019) and Liu et al. (2019) identified the possible nonlinear association between the urinary concentrations of parabens and GDM. Zhang et al. (2019) reported a U-shaped association between the urinary concentrations of BPA and glucose levels, as measured using OGTT. In line with previous studies, we observed non-monotonic associations of BPA and triclosan with the estimated difference in the probability of GDM when evaluated using probit-BKMR models. These results suggest that the traditional regression method with the assumption of linearity may not correctly describe the exposure-response functions when evaluating the association of phenol and paraben exposures with GDM. Using BKMR in our study, we captured the presence of a complex nonlinear exposure-response function while fixing other chemical biomarkers at certain concentrations and explored the jointed effect of the chemicals. Our study highlights the importance of evaluating the effect of phenols as a mixture using BKMR in studies of multiple perinatal environmental exposures.

This study expanded the scope of evidence evaluating the association between phenols, parabens, and GDM using a mixture analysis of probit-BKMR. However, it is necessary to consider several limitations when interpreting these results. The relatively small sample size of our study is a major limitation, which limited statistical power, precision of point estimates, and our ability to simultaneously evaluate multiple potential confounders in the multivariable models. Additionally, it is possible that our observed associations may have been impacted by potential selection bias due to unknown selection factors. Our study restricted the population to patients residing in the counties surrounding the clinic to ensure that controls were selected from the source population that provided the cases. We cannot, however, rule out the possibility that potential selection bias may exist if the exposure distribution of the controls failed to represent the source population. We assessed the urinary concentrations of phenols and parabens using a single spot urine sample. As phenols and parabens are rapidly metabolized and excreted from the body and urinary concentrations are known to vary throughout pregnancy (Vernet et al., 2018), the measurements of urinary concentrations would primarily reflect exposures in the days immediately before specimen collection. Several prior studies assessed the variability of urinary phenols and parabens during pregnancy and reported that certain chemicals had moderate reliability, with intraclass correlation coefficients (ICC) ranging from 0.62 to 0.70 for benzophenone-3, 0.58 to 0.61 for triclosan, 0.56 to 0.64 for butylparaben, 0.44 to 0.61 for methylparaben, and 0.54 to 0.56 for propylparaben (Braun et al., 2012; Philippat et al., 2013; Yazdy et al., 2018), except for BPA with rather low ICC ranging from 0.11 to 0.31 (Casas et al., 2018). This suggests that a single urine sample may be reasonable to assess individual exposure to most phenols and parabens. Nevertheless, concentrations of urinary phenols and parabens may also be potentially influenced by the time of urine collection and non-fasting status. Since sample collection occurred across morning and afternoon clinic visits, variation in time of day can be considered another limitation of the use of single spot urine samples to assess exposure. Thus, we cannot rule out the possibility of measurement error due to the variability of urinary concentrations that may result in attenuation or inflation of the true association. Considering that changes in placental and maternal hormones over the course of pregnancy are involved in the development of maternal insulin resistance (Newbern and Freemark, 2011), future research with multiple urine biomarker measurements during pregnancy will improve exposure assessment and help elucidate the sensitive windows of environmental chemical exposures and GDM. In addition, the detection frequency of certain parabens (i.e., butylparaben) was relatively low in this study. We substituted a value equal to the LOD/√2 for concentrations less than the LOD, which may be prone to biased estimation. However, this substitution makes little difference in the geometric mean estimate if the detection frequency is above 40% (Centers for Disease Control and Prevention (CDC), 2022). Furthermore, our study was unable to account for unmeasured factors such as diet and physical activity (Mijatovic-Vukas et al., 2018). Diet is a recognized health-related factor during pregnancy and is a primary source of exposure to certain phenols (Pacyga et al., 2019). Detailed dietary components and patterns should be obtained in future studies to consider the role of diet in the observed association. Lastly, since the specimen collection in this case-control study design occurred at the clinical visit following glucose screening, the temporality of the relationship between the environmental phenols and parabens and GDM cannot be established. Moreover, due to this limitation, it is possible that the urinary concentrations of chemical biomarkers may have been affected by GDM development.

5. Conclusions

In this case-control study among pregnant women in Oklahoma, a positive association between the urinary concentrations of benzophenone-3 and GDM was observed either assessed individually or evaluated as a mixture, while accounting for other phenols and parabens. This study applied a novel statistical framework for evaluating the effect of exposure to multiple phenols and parabens on GDM. An inverse association was observed between urinary concentrations of BPA and GDM in the individual exposure biomarker model, but this association was attenuated and included the null when addressing phenol and paraben mixtures in the analysis. The differences in the effects of individual chemicals and chemical mixtures revealed the importance of accounting for correlation among the chemical mixture components in studies on the relationship between perinatal environmental exposure and pregnancy complications, given women are exposed to multiple chemicals. Future prospective studies involving repeated biospecimen collection and fasting glucose measures during different windows of exposure throughout pregnancy may provide insights on the strength of evidence for a causal relationship between phenol and paraben exposure and pregnancy dysglycemia. Additional prospective investigations are warranted for verifying the link between benzophenone-3 exposure and GDM.