NHANES survey measures the health and nutrition of American adults, adolescents and children in the most comprehensive way. This survey combines information from interviews and physical examinations, which information from interviews and physical examinations is combined. NHANES is conducted under the National Center for Health Statistics (NCHS), an agency that falls under the jurisdiction of the Centers for Disease Control and Prevention (17).Data collection and laboratory analysis methods for the NHANES are described elsewhere. Questions on demographics, socioeconomics, dietary habits, and health are included in the interviews. Participants gave written consent, and the study’s methods and materials were ethics board-approved. This study used data from the 2013–2016 NHANES to determine blood HbEO, serum TT, E2, and SHBG levels for the United States population. Referring to previous NHANES literature, we deleted participants with missing covariates (18, 19). A total of 20146 participants were enrolled at first, after the exclusion of individuals who were pregnant (n=135), missed data on blood EO (n=15189) and unavailable data on sex hormones (n=601), 4221 participants were included in our final analysis. Figure 1 illustrates the selection process for participants.

HbEO, with a longer in vivo half-life than EO, is utilized as an indicator of EO exposure. The measurement of HbEO levels was performed according to the guidelines outlined in the NHANES Laboratory/Medical Technologist Procedures Manual (https://wwwn.cdc.gov/Nchs/Nhanes/2013-2014/ETHOX_H.htm). Initially, samples of erythrocytes were gathered, preserved at −30°C, and then sent to the National Center for Environmental Health for assessment. Subsequently, The HbEO quantity was measured utilizing high-performance liquid chromatograph-tandem mass spectrometry (HPLC-MS/MS) and the modified Edman reaction. Lastly, as a result, the levels of HbEO were expressed as picomoles adducts per gram of hemoglobin. Additionally, a comprehensive quality control program, incorporating both external and internal surveillance, was established to oversee and assess the accuracy and reliability of the analytical testing process.

A serum sample was prepared, preserved at -20°C, and shipped to the National Center for Environmental Health to be tested. To measure TT and E2, isotope dilution-liquid chromatograph-tandem mass spectrometry (ID-LC-MS/MS) was applied. The levels of SHBG were measured using a chemiluminescent assay. Detailed technical information about these methods are available in other sources (20, 21). The lower limit of detection (LLOD) was 0.75 ng/mL for TT, 2.994 pg/mL for E2, and 0.800 nmol/L for SHBG. For estimating the concentration of circulating free testosterone, the free androgen index (FAI) was used to estimate the level of circulating free testosterone by dividing TT by SHBG (22). The TT to E2 ratio (TT/E2) was utilized as an indirect measure of aromatase activity (23).

Potential confounding factors were also collected as covariates, including age, race/ethnicity, education level, BMI, poverty income ratio (PIR), cotinine and the time of sample collection. Race/ethnicity was classified into Hispanics, non-Hispanic Black, non-Hispanic White, and other races. Education level was classified as less than high school, more than high school, high school, or general educational development (24). BMI was computed by dividing weight (in kilograms) by the square of height (in meters). According to the International Obesity Task Force (IOTF), children were classified as overweight (BMI ≥ IOTF-25), obesity (BMI ≥ IOTF-30) and normal (BMI< IOTF-25) (25). In the adult group, BMI category was classified as < 25 kg/m2 and ≥ 25 kg/m2 (9). The PIR, which signifies the family’s collective socioeconomic standing, was computed by dividing the family income according to the poverty guidelines pertinent to the participant’s household size, coupled with the year and state data (26). Serum cotinine levels were measured and group into 2 categories: <LOD (below the LOD, 0.015 ng/mL) and ≥LOD to take the environmental tobacco exposure into account (27). Moreover, to account for the variations in hormone levels, a six-month time span (November 1 to April 30, May 1 to October 31) incorporated as covariates in the statistical analysis. Physical activity was calculated based on a detailed physical activity survey described previously (28). Total energy intake was calculated from three-day dietary-recall food composition tables (29).Whether the participant has taken prescription medicine is assessed by NHANES interviewers through the following question: “In the past 30 days, have you used or taken any medication that requires a prescription? Please do not include any prescription vitamins or minerals that you may have already mentioned.” according to the NHANES analysis guideline (https://wwwn.cdc.gov/Nchs/Data/Nhanes/Public/2013/DataFiles/RXQ_RX_H.htm#RXDCOUNT).

Potential confounding factors were also collected as covariates, including age, race/ethnicity, education level, BMI, poverty income ratio (PIR), cotinine and the time of sample collection. Race/ethnicity was classified into Hispanics, non-Hispanic Black, non-Hispanic White, and other races. Education level was classified as less than high school, more than high school, high school, or general educational development (24). BMI was computed by dividing weight (in kilograms) by the square of height (in meters). According to the International Obesity Task Force (IOTF), children were classified as overweight (BMI ≥ IOTF-25), obesity (BMI ≥ IOTF-30) and normal (BMI< IOTF-25) (25). In the adult group, BMI category was classified as < 25 kg/m2 and ≥ 25 kg/m2 (9). The PIR, which signifies the family’s collective socioeconomic standing, was computed by dividing the family income according to the poverty guidelines pertinent to the participant’s household size, coupled with the year and state data (26). Serum cotinine levels were measured and group into 2 categories: <LOD (below the LOD, 0.015 ng/mL) and ≥LOD to take the environmental tobacco exposure into account (27). Moreover, to account for the variations in hormone levels, a six-month time span (November 1 to April 30, May 1 to October 31) incorporated as covariates in the statistical analysis. Physical activity was calculated based on a detailed physical activity survey described previously (28). Total energy intake was calculated from three-day dietary-recall food composition tables (29). Whether the participant has taken prescription medicine is assessed by NHANES interviewers through the following question: “In the past 30 days, have you used or taken any medication that requires a prescription? Please do not include any prescription vitamins or minerals that you may have already mentioned.” According to the NHANES analysis guideline (https://wwwn.cdc.gov/Nchs/Data/Nhanes/Public/2013/DataFiles/RXQ_RX_H.htm#RXDCOUNT). Urine specimens were obtained during field examination visits and cryopreserved at −20°C prior to analytical processing. Chemical selection criteria required ≥85% detection frequency across samples, resulting in the inclusion of: two phenolic compounds (bisphenol A [BPA] and bisphenol S [BPS]), three phthalate derivatives (monobenzyl phthalate [MBzP], monoisobutyl phthalate [MiBP], and monocarboxyoctyl phthalate [MCOP]). BPA and BPS were quantified via automated online solid-phase extraction coupled with isotope-dilution high-performance liquid chromatography-tandem mass spectrometry (SPE-HPLC-MS/MS). MBzP, MCOP, and MiBP analysis employed HPLC separation with electrospray ionization tandem mass spectrometric detection (HPLC-ESI-MS/MS). Methodological specifications and quality assurance protocols adhere to NHANES laboratory standards (https://wwwn.cdc.gov/Nchs/Data/Nhanes/Public/2013/DataFiles/EPHPP_H.htm#URD14DLC), (https://wwwn.cdc.gov/Nchs/Data/Nhanes/Public/2013/DataFiles/PHTHTE_H.htm), polychlorinated biphenyls (PCBs) were measured in serum by high-resolution gas chromatography/isotope-dilution high-resolution mass spectrometry (https://wwwn.cdc.gov/Nchs/Data/Nhanes/Public/2013/DataFiles/PCBPOL_H.htm). Based on prior literature, we identified three polychlorinated biphenyls (PCBs) with potential endocrine-disrupting effects on sex hormones (30).

Appropriate weighting, in accordance with the recommendations of the NCHS, were employed for each analytical process. Weighted mean value (± standard deviation [SD]) were calculated employing NHANES primary sampling units and strata, with the results being subjected to a weighted t-test for statistical analysis. Frequencies (proportions) for categorical variables were analyzed using the weighted chi-square test. The serum HbEO and sex hormone indicator distributions exhibited a right-skewed pattern, prompting log2 transformation to normalize them for descriptive and regression analyses. The log2-transformed HbEO values were then analyzed as continuous and categorical (divided into quartiles) variables (31). Weighted quartiles for log2-transformed HbEO were calculated within specific sex-age and sex-puberty subgroups, as shown in Supplementary Table S1 .

A weighted multiple linear regression was used to calculate β (Standardized coefficients) values and corresponding 95% confidence intervals (CIs) to examine the connections between individual HbEO levels and sex steroid hormone indicators. Restricted cubic spline (RCS) analysis was applied to further examine linear and non-linear relationships between HbEO and sex hormones after adjusting for various potential covariates (32, 33). The Akaike information criterion (AIC) was used to choose the most suitable knots that had the smallest AIC (34).

Considering the marked variations in sex hormone levels between genders and at different stages of development, this analysis was conducted based on age (children: ≤11 years, adolescents: 12-19 years, adults: >19 years) for males and females (35). Grouping 6-19 years old participants as “children” or “adolescents” based solely on their ages may include both pre-pubertal and pubertal individuals in the same groups. Such grouping could result in subgroups with exceptionally high or low sex hormone levels, which might bias the regression analysis relationship between HbEO and sex steroid hormones. Moreover, the impact of log2-HbEO on sex hormones may be influenced by the pubertal stage. To tackle this concern, we further subcategorized individuals into pubertal and prepubertal subgroups based on their serum sex hormone levels and menarcheal status. Individuals were considered to have entered puberty (i.e., classified as pubertal) if their TT levels were equal to or greater than 50 ng/dL (in males) or E2 levels were 20 pg/mL or higher (in females), or if they had experienced the onset of menstruation (in females) (36–38). Those not meeting these criteria were classified as pre-pubertal (i.e., classified as prepubertal). Data regarding girls’ menarche status was collected through inquiries such as: “Have menstrual periods commenced?” (found in the medical questionnaire) and “What age was the first menstrual period?” (included in the reproductive health questionnaire). Female whose answer was “Yes” or provided their age at first menstrual period was recorded as having started menstrual cycles. Furthermore, this analysis performed the multiple linear regression analyses again, stratified by pubertal status in both males and females.

Subgroup analyses with multiplicative interaction terms were performed to show whether the relationship between individual HbEO levels and sex steroid hormones varied by race, age (<45 years or ≥45 years), BMI (<24 or ≥24 kg/m2), energy intake (<2,400 or ≥ 2,400 kcal), physical activity (<200 or ≥ 200 METs-hour/week) and taken prescription medicine (Yes or No) (39).

To evaluate the robustness of our results, 4 sensitivity analyses were conducted to examine the relationship between individual HbEO levels and sex steroid hormones: 1) Multiple imputation for missing data was conducted using the ‘mice’ package (Multivariate Imputation by Chained Equations) in R (40); 2) Energy intake, physical activity, and prescription medication was adjusted additionally; 3) exposure of bisphenol A, phthalates and polychlorinated biphenyls was adjusted additionally; 4) excluded participants aged ≥65 years in the adults. All statistical analyses were performed using R software version 4.1.2, and a two-tailed P value with a significance level of 0.05 was used for all statistical tests to determine significance.

Table 1 summarizes the demographic features of the enrolled participants categorized by sex and age. The study sample comprised a total of 4221 individuals, including 2089 males and 2101 females. The individuals had a mean age of 41.70 years. The detection rates for TT, E2, SHBG, and HbEO in the entire sample were 99.81%, 85.93%, 100%, and 97.49%, respectively. The values below the LLOD were replaced by the lower limit of detection divided by the square root of 2 (LLOD/sqrt (2)) according to the NHANES analysis guideline (https://wwwn.cdc.gov/Nchs/Data/Nhanes/Public/2013/DataFiles/ETHOX_H.htm#LBXEOA) ( Supplementary Table S2 ). However, the detection rate for E2 in male children was less than 50%. As a result, the analyses for E2 and TT/E2 were omitted for this particular subgroup. Comparatively, adult participants exhibited greater ethylene oxide exposure levels than adolescents, while adolescents showed higher exposure levels than children, across both genders ( Supplementary Table S3 ).

Given the crucial involvement of sex hormones in the initiation of puberty, the statuses of pre-puberty and puberty were also examined ( Supplementary Table S4 ). The proportion of E2 detections in prepubertal males and females was below 50%. Consequently, assessments of E2 and TT/E2 were not conducted for these demographic subsets. EO exposure was significantly higher in the puberty than that of pre-puberty in males. While there were no significances between pre-puberty and puberty in females.

Figure 2 and Supplementary Table S4 present the relationships between HbEO levels, categorized as continuous and quartiles, and sex hormones across various stages of development. Among males, no substantial correlations were found between HbEO and sex steroid hormones among children and adolescents ( Supplementary Table S4 ). However, in male adults, HbEO doubling corresponds to a 2.85% decrease in E2, a 2.10% increase in SHBG, and a 4.3% rise in the TT/E2 (for E2: p= 0.04, OR = -0.03, 95% CI: -0.06 to 0.00; for SHBG: p= 0.03, OR = 0.03, 95% CI: 0.00 to 0.06; for TT/E2: p< 0.001, OR = 0.06, 95% CI: 0.03 to 0.09). Moreover, consistent associations of serum E2 and TT/E2 levels with HbEO were also observed across quartiles of EO exposure. Serum E2 levels were decreased along with the increased exposure of EO (p for trend=0.014). Similarly, TT/E2 levels were increased accompanied by the increased exposure of EO (p for trend=0.003) ( Figure 2 ). No correlations were identified between EO exposure and sex hormones in female subjects across various categories.

The relationship between the continuous and quartile distributions of HbEO and sex steroid hormones in prepubertal and pubertal status were presented in Supplementary Table S5 . There were no significant correlations found between HbEO and sex steroid hormones in prepuberty and puberty, irrespective of gender. When evaluated by quartiles of EO exposure, HbEO (quartile 4 versus 1) showed a positive correlation with SHBG in pre-pubertal males (P = 0.03, OR = 0.21, 95% CI: 0.02, 0.4). In contrast, HbEO (quartile 3 versus 1) was negatively correlation with SHBG in pubertal males (P = 0.03, OR = -0.22, 95% CI: -0.42, -0.03).

A restricted cubic spline (RCS) analysis was utilized for further investigating the linear and nonlinear relationships between HbEO with sex hormones after adjusting for various confounding factors, as depicted in Figure 3 and Supplementary Figure S1 . In male children, A nonlinear association was found between HbEO and SHBG (NL-P value = 0.049). A non-linear relationship was found between HbEO and TT in male adolescents (NL-P value = 0.042). In male adult, non-linear relationships were discovered between HbEO with E2, SHBG and FAI (E2: NL-P value = 0.0026; SHBG: NL-P value =0; FAI: NL-P value =0.017). Furthermore, nonlinear associations were found between FAI, SHBG, and HbEO in female adult participants. (NL-P value = 0.041; FAI: NL-P value = 0.024) ( Figure 3 ).

What’s more, our RCS analysis uncovered a nonlinear relationship between HbEO and SHBG in the prepubertal group among males (NL-P value = 0.0264). While, HbEO has a non-linear association with FAI in female pubertal individuals (NL-P value = 0.0476) ( Supplementary Figure S1 ).

Subgroup analyses were conducted among adult males categorized by age, race, BMI, energy intake, physical activity, and prescription medication to further investigate the consistency of the relationship between the log2-transformed HbEO levels and sex hormones across different groups. These results confirmed a robust association between the log2-transformed HbEO levels and TT, E2, FAI, TT/E2 across different subgroups (P for interaction > 0.05) ( Figures 4A, B, D, E ). We found that the effect of HbEO on SHBG was significantly dependent on energy intake (interaction P < 0.001). In adult males with an energy intake ≥ 2400 kcal, HbEO was positively associated with SHBG ( Figure 4C ).

To validate the robustness of our findings, we conducted 4 sensitivity analyses. In all sensitivity analyses, the negative association of HbEO with E2 and TT/E2 remained significant, demonstrating the robustness of our results ( Supplementary Tables S6–S12 ).