National Health and Nutrition Examination Survey (NHANES) is an ongoing, nationally representative series of surveys conducted every two years to monitor the health and nutritional status of non-institutionalized citizens in the United States (26). The study cohort was confined to the survey period from 2013 to March 2020, as EO measurements were unavailable before this timeframe (27, 28). In NHANES 2013-2020, a total of 33657 participants completed both interviews and medical examinations. We selected adults aged 18 to 65 years for the study. According to the NCEP ATP III-2005 criteria, the components defining metabolic syndrome primarily consist of plasma glucose, triglycerides, high-density lipoprotein cholesterol (HDL-C), and waist circumference. Therefore, participants with missing data on the components defining MetS will be excluded. Participants with missing demographic information such as marital status, severe drinking habits, family income-poverty ratio (PIR), educational level, and smoking status were also excluded. Ultimately, our study included 1842 participants ( Figure 1 ).

The modified Edman reaction measures HbEO in human whole blood or red blood cells. This method is applied to N-terminal hemoglobin adducts and has been optimized to enhance product yield, sensitivity, and automation (29). Thus, NHANES staff utilize this method to detect HbEO.

MetS was characterized based on the NCEP ATP III-2005 criteria (30, 31), which entail the presence of three or more of the following conditions: 1) increased waist circumference (EWC), defined as waist circumference ≥102 cm in men and ≥88 cm in women; 2) high blood pressure, indicated by blood pressure levels ≥130/85 mm Hg or the use of medication for previously diagnosed hypertension; 3) reduced levels of HDL-C, with values below <40 mg/dL in men and <50 mg/dL in women, or the use of specific treatment for low HDL-C; 4) elevated triglycerides (TGs), defined as TG levels ≥150 mg/dL or the use of medication for high TG levels; and 5) increased fasting glucose, represented by fasting glucose levels ≥100 mg/L or the use of medication for high glucose levels and a previous diagnosis of type 2 diabetes.

Based on previous research, smoking, and drinking status was categorized into three groups: “never,” “former,” and “current” (32, 33). Hypertension was defined as systolic blood pressure ≥140 mmHg, diastolic blood pressure ≥ 90 mmHg, self-reported physician diagnosis, or current use of antihypertensive medication (34). Information regarding marital status, educational attainment, and ethnicity was extracted from the fundamental demographic data in the NHANES database. Diabetes was delineated by criteria such as glycated hemoglobin (HbA1c) levels equal to or exceeding 6.5%, fasting blood glucose levels greater than or equal to 7 mmol/L, self-reported diabetes, or the present use of antidiabetic medication (35). These subgroups comprise individuals categorized as either normal weight/overweight (BMI < 30 kg/m²) or obese (BMI ≥ 30 kg/m²) (36). PIR was used to assess household income levels and classified into three groups (<1.3, 1.3 - 3.5, > 3.5) (37). The following data were collected to diagnose MetS: triglycerides, glucose, and HDL-C.

To better ascertain the relationship between HbEO levels and MetS, we categorized HbEO levels into four groups using quartiles (Q1<22.62, Q2: 22.62-33.515, Q3: 33.515-148, Q4 >148 pmol/g Hb) (28, 38, 39). Statistical analyses were conducted following the guidelines of NHANES, considering the complex sampling design of the survey to address the bias associated with sample selection, oversampling, and nonresponse (22). Therefore, weights were calculated using the WTSAF2YR weight calculation method for the biochemical markers.

When describing the baseline characteristics of the study population, the data are presented as weighted means ± standard deviation (SD) for continuous measurements, and as unweighted counts along with weighted percentages for categorical measurements. Due to the severe skewness in the distributions of triglycerides and plasma glucose, these variables are presented with the median [interquartile range (IQR)]. The Wilcoxon rank-sum test (ranksum test) was employed to compare independent samples of these variables. Statistical significance was assessed using Student’s t test for continuous variables and chi-square tests for categorical variables. Weighted multivariable logistic regression models was used to calculate odds ratios (ORs) with 95% confidence intervals (CIs). The multivariable weighted model was adjusted for age, sex, BMI (non-obese or obese), race, marital status, smoking status, alcohol drinking status, PIR, diabetes, and hypertension. The relationship between HbEO and MetS was also modeled using restricted cubic splines (RCS) with three knots positioned at the 1st, 50th, and 90th percentiles. To assess the robustness of the study, we conducted three sensitivity analyses. Firstly, the International Diabetes Federation-2009 criteria was adopted to redefine the metabolic syndrome (40, 41). Then, we performed the primary analysis on the participants redefined according to these criteria to assess the reliability and robustness of our results. Secondly, due to the skewed distribution of HbEO data, we logarithmically transformed serum ethylene oxide levels (27), and included the transformed log (HbEO) as a continuous variable in two models (38, 39). The study outcomes were defined as MetS according to ATP criteria. Lastly, dietary intake is a significant factor influencing metabolism, and in the United States, dietary quality is determined by the Healthy Eating Index-2015 (HEI-2015) (42). HEI-2015 is used to assess dietary quality based on a population-based scoring algorithm. Evaluation of HEI-2015 comprises 13 dietary components, including Dairy, total protein foods, and seafood and plant proteins, which encompass alternative dairy and protein products (43, 44). We included participants’ total HEI-2015 scores as covariates in the final model to evaluate whether diet quality affects model contributions. It is important to emphasize the use of weighting variable WTDRD1 based on the NHANES analysis guidelines using dietary recall data (excluding missing dietary data n=731, resulting in a final sample size of n=1111) for the third analysis. The visualization of the 13 components was conducted through radar plots ( Supplementary Figure 1 ).

To explore potential sources of variability in the relationship between HbEO and the studied outcomes, we extended our investigation through subgroup analyses based on sex, age, non-obese or obese status, hypertension, and diabetes. The selection of specific variables for subgroup analyses was based on their clinical relevance to liver diseases and their potential influence on the relationship between HbEO and the studied outcomes. Including these variables in our analysis allowed us to assess the presence of effect modifications (interactions). This was further examined by incorporating a product term of each stratifying variable and HbEO into the primary model, followed by a Wald test. The entire statistical analysis was conducted using the statistical computing and graphics software R (version 4.1.3) and STATA (version 17.0), with statistical significance set at P < 0.05.

The study involved 1842 participants with an average age of 44 ( Table 1 ), comprising 900 male participants and 942 female participants. MetS components include reduced high-density lipoprotein cholesterol, elevated triglycerides, elevated plasma glucose, elevated waist circumference, elevated systolic blood pressure and diastolic blood pressure. The study population was subsequently categorized into two groups: one without metabolic syndrome and another with metabolic syndrome. There were significant differences in MetS components between males and females (P < 0.001). Significant differences were also found in educational level, marital status, and diabetes status among different groups (P = 0.007, P = 0.003, P < 0.001, respectively). Furthermore, individuals with MetS were significantly more prevalent in the hypertension group compared to those without (P < 0.001).

In the multivariate weighted logistic analysis, the investigation focused on assessing the correlation between HbEO levels and the risk of MetS and the correlation between HbEO and the risk associated with individual MetS components. ( Table 2 ) The serum HbEO level exhibited a positive correlation with metabolic syndrome risk in Q2 level (OR=1.64, 95% CI: 1.04~2.48), Q3 level (OR=1.99, 95% CI: 1.29~3.08), and Q4 level (OR=2.89, 95% CI: 1.92~4.34) in model 1, with a significant p-value for trend (P=0.01), indicating that with the level of HbEO increased, the MetS risk increased. Regarding elevated waist circumference, a significantly positive association was found in the Q4 level (OR=2.28, 95% CI: 1.04~5.01) with a significant p-value for trend (P=0.05) in model 1. As for elevated blood pressure, a significantly positive association was found in Q4 level in model 1 (OR=1.64, 95% CI: 1.07~2.52). Regarding reduced high-density lipoprotein cholesterol, a significantly positive association was found in Q2 and Q4 in model 1 with a significant p-value for trend (P=0.002) and in Q4 in model 2 (OR=3.03, 95% CI: 1.60~5.75) with a significant p-value for trend (P=0.04). As for elevated total triglycerides, a significantly positive association was found in Q3 and Q4 in model 1 (OR=1.81, 95% CI: 1.15~2.84; OR=2.78, 95% CI: 1.66~4.65, respectively) with a significant p-value for trend (P=0.04) and in Q3 and Q4 in model 2 (OR=1.72, 95% CI: 1.05~2.81; OR=2.28, 95% CI: 1.19~4.37, respectively). Regarding diabetes, a significant negatively association was found in Q4 in model 1 (OR=0.62, 95%CI: 0.39~0.99) and in Q3 and Q4 in model 2 (OR=0.57, 95%CI: 0.33~0.98; OR=0.40, 95%CI: 0.19~0.84).

The dose-response association between HbEO level and MetS risk indicated that increasing levels of serum HbEO were associated with a higher risk of MetS ( Figure 2 ). However, the p-value for non-linearity was more significant than 0.05, suggesting a possible linear association between serum HbEO and MetS risk (P-overall=0.0359, P-non-linear=0.179). For the female population, there was a non-linear and L-shape association between HbEO level and MetS risk (P-overall<0.001, P-non-linear=0.0024) (refer to Supplementary Figure 2 ). Regarding the population without obesity, a significantly non-linear and inverted U-shape association was found between HbEO and MetS risk (P-overall=0.0107, P-non-linear=0.0055) (refer to Supplementary Figure 3 ). For the population aged 50 and older, a significantly non-linear and inverted U-shape association was found between HbEO and MetS risk (P-overall<0.001, P-non-linear=0.0157) (refer to Supplementary Figure 4 ).

In sensitivity analysis ( Supplementary Table 1 ) that used the IDF definition for MetS, the serum HbEO levels were positively associated with MetS risk in Q3 and Q4 levels (OR=1.71, 95% CI: 1.13~2.58; OR=2.60, 95% CI: 1.69~3.99, respectively) in model 1 with a significant p-value for trend (P=0.02). In model 2, serum HbEO levels were positively associated with MetS risk in Q4 level (OR=2.65, 95% CI: 1.35~5.19), with a non-significant p-value for trend (P=0.14). The analysis elucidated a positive correlation between logarithmic-transformed HbEO [Log (HbEO)] levels and the susceptibility to MetS (refer to Supplementary Table 2 ). In Model 1, the odds ratio (OR) denoting the association between Log (HbEO) levels and MetS risk was determined to be 1.31 (95% CI: 1.12~1.53), while in Model 2, it was 1.26 (95% CI: 1.03~1.55). The positive association persisted after adjusting for the Healthy Eating Index-2015 score within the model (refer to Supplementary Table 3 ). In Model 1, a discernible elevation in the risk of MetS was observed concomitant with EO levels. For instance, relative to the reference group, the OR for MetS was 2.29 (95% CI: 1.25~4.20) in the Q2 level of EO levels, 1.96 (95% CI: 1.23~3.41) in the Q3 level, and 2.77 (95% CI: 1.65~4.66) in Q4 level. Similarly, in Model 2, the risk of MetS exhibited a progressive augmentation with elevated EO levels (OR=2.17, 95% CI: 1.05~4.49) in Q2 level; OR=2.09, 95% CI: 1.13~3.87) in Q3 level; OR=2.23, 95% CI: 1.10~4.53) in Q4 level. These findings underscore a consistent and statistically significant positive association between EO exposure and the predisposition to MetS, even after adjustment for plausible confounding variables such as the Healthy Eating Index-2015 score.

Subgroup analysis was conducted to investigate potential sex, age, and BMI interactions with the relationship between serum HbEO and the risk of MetS ( Figure 3 ). Notably, the subgroup analysis of age showed a significant difference between age groups (P for interaction=0.028), suggesting an interaction between serum HbEO and MetS about age. The serum HbEO level in males in Q3 and Q4 was significantly associated with an increased risk of MetS (OR=3.02, 95%CI: 1.38~6.58; OR=2.78, 95%CI: 1.07~7.23, respectively). This relationship was also observed in females in Q4 (OR=4.02, 95%CI: 1.15~14). There was a significant association between an increased risk of MetS and serum HbEO level in individuals aged 50 and over in Q3 and Q4 (OR=2.58, 95%CI: 1.25~5.33; OR=5.83, 95%CI: 2.19~15.48, respectively). Furthermore, the level of serum HbEO in individuals without obesity in Q3 and Q4 was also significantly associated with an increased risk of MetS (OR=2.88, 95%CI: 1.34~6.17; OR=4.94, 95%CI: 1.85~13.2, respectively). Subgroup analysis for hypertension and diabetes suggested hypertension and diabetes had no impact on the prevalence of MetS (P=0.411, P=0.993, respectively) ( Supplementary Figure 5 ).