NHANES is a large-scale national survey led by the Centers for Disease Control and Prevention (CDC). It combines interviews, physical examinations, laboratory tests, and health behavior questionnaires to cover various health indicators and environmental exposure factors, assessing the health and nutritional status of the U.S. population. For this study, we selected 19,399 individuals from the 2013–2014 and 2017–2018 cycle. After excluding those younger than 20 years (n = 8,031), and those lacking OA questionnaire data (n = 0) or EO assay results (n = 7,922), a total of 3,386 participants were included in the final analysis (Figure 1). Since the research data was obtained from a public database (https://www.cdc.gov/nchs/nhanes/index.htm), ethical review was conducted by the CDC (https://www.cdc.gov/nchs/nhanes/irba98.htm).

OA was determined by reviewing the medical conditions section (variable name prefix MCQ) in the self-reported personal interview data. Participants were asked two questions regarding OA: “Has a doctor or other health professional ever told you that you had arthritis?” and “Which type of arthritis was it?” A response of “Yes” to either question was recorded as a positive case of OA.

According to the NHANES Laboratory Procedures Manual (LPM) (https://wwwn.cdc.gov/Nchs/Nhanes/2013-2014/ETHOX_H.htm), blood samples were collected and processed for testing at the National Center for Environmental Health. The testing procedure included the following steps: first, preparation of the specimen for measurement of hemoglobin adducts of ethylene oxide; second, measurement of total hemoglobin in the sample solution used for hemoglobin adduct measurements; third, modified Edman reaction in the sample solution and isolation of Edman products; lastly, analysis of Edman products by high-performance liquid chromatography coupled with tandem mass spectrometry (HPLC-MS/MS) and processing of results. The results were reported in pmol adduct per gram of hemoglobin. The entire process employed various quality controls to ensure the accuracy and precision of the test results.

MET-minute is an indicator used to measure the intensity and duration of physical activity. MET (Metabolic Equivalent of Task) represents the metabolic equivalent of an activity. MET-minute is calculated by multiplying the MET value of the activity by its duration (in minutes), and is used to quantify the total energy expenditure of an individual over a specific period. In this study, weekly MET-minute composite scores were used to assess physical activity levels, based on the NHANES suggested MET Scores, with the following calculation formula (21): (8.0 MET scores × minutes of vigorous work-related activity) + (4.0 MET scores × minutes of moderate work-related activity) + (4.0 MET scores × minutes of walking or bicycling for transportation) + (8.0 MET scores × minutes of vigorous leisure-time physical activity) + (4.0 MET scores × minutes of moderate leisure time [recreational] physical activity). According to the Physical Activity Guidelines for Americans, physical activity was divided into four categories (22): sedentary (no regular physical activity), insufficient (performing some regular activity, but < 500 MET-minutes per week), moderate (500–1,000 MET-minutes per week), and vigorous (more than 1,000 MET-minutes per week).

Demographic characteristics, including age, gender, ethnicity, education level, and family poverty income ratio (PIR), were extracted from the demographics file. PIR is defined as the ratio of family income to the poverty threshold, with participants categorized into low income (< 1.3), middle income (1.3–3.5), and high income (≥3.5) groups. Body mass index (BMI) was calculated as weight (kg) divided by height squared (m²), and classified into < 25 kg/m² and ≥25 kg/m² categories. Current smoker status was determined based on the response to the question, “Do you now smoke cigarettes?” Past-year alcohol drinking was categorized according to the response to “In the past 12 months, on those days that you drank alcoholic beverages, on average, how many drinks did you have?” into Non-drinker, 1–3 drinks, and ≥4 drinks groups. Diabetes mellitus and hypertension were determined based on medical diagnoses. Total calcium levels were classified according to clinical standards as insufficient (< 8.5 mg/dL) or sufficient (≥8.5 mg/dL). Similarly, 25-hydroxyvitamin D2+D3 levels were categorized as deficient (25–50 nmol/L), insufficient (50–75 nmol/L), and sufficient (≥75 nmol/L) (23).

NHANES employs a complex sampling design. This study used the subsample weight WTSA2YR for weighting calculations to correct for representativeness bias. To maintain data integrity, missing values in covariates were handled as follows: continuous variables with missing values below 10% were imputed with the mean if normally distributed and with the median if not, while continuous variables with missing values above 10% and all categorical variables were addressed using multiple imputation. All continuous variables were categorized, and results were presented as absolute values (n) or percentages (%), with chi-square tests used for evaluation. The skewed EO values were log10-transformed, analyzed using the Kruskal-Wallis test, presented as M (Q₁, Q₃), and then grouped into quartiles.

First, weighted univariate and multivariate logistic regression analyses were conducted to calculate odds ratios (ORs) and 95% confidence intervals (CIs) to explore the relationship between log10-transformed EO levels and OA risk. Potential confounders were adjusted to ensure the reliability of the results. The crude model did not adjust for any covariates. Model I adjusted for age, sex, race, PIR, and education level. Model II adjusted for all covariates. Next, restricted cubic splines (RCS) were used to explore potential complex nonlinear relationships between log10-transformed EO levels and OA risk, using the 10th, 50th, and 90th percentiles as knots. Finally, subgroup analysis and interaction test were then performed on potential confounding variables to examine the consistency of the relationship between EO and OA across different subgroups and to identify sources of variation. All statistical analyses were conducted using R software (version 4.4.1). P-value < 0.05 was considered statistically significant.

As detailed in Table 1, the study included 3,386 participants from the NHANES 2013–2014 and 2017–2018 cycles. Of these, 952 were diagnosed with OA, while 2,434 were not, yielding a weighted OA prevalence of 27.21%. Participants were categorized into four groups based on quartiles of log10-transformed EO levels: Q1 (0.87 ≤ log10 EO < 1.32 pmol/g Hb, n = 849), Q2 (1.32 ≤ log10 EO < 1.48 pmol/g Hb, n = 845), Q3 (1.48 ≤ log10 EO < 1.76 pmol/g Hb, n = 845), and Q4 (1.76 ≤ log10 EO ≤ 3.24 pmol/g Hb, n = 847). Significant differences were observed among groups regarding age, gender, ethnicity, education level, PIR, BMI, current smoker status, past-year alcohol drinking, diabetes mellitus, and 25-hydroxyvitamin D2+D3 levels (all P < 0.05). No meaningful differences were found for hypertension and total calcium levels. Additionally, Figure 2 shows no significant differences in EO concentrations between OA and non-OA participants (P > 0.05).

Weighted univariate logistic regression analysis indicated that the association between log10-transformed EO and OA was not statistically significant in Supplementary Table 1, (OR = 1.18, 95%CI: 0.91, 1.52, P = 0.2). However, weighted multivariate logistic regression analysis revealed a significant association between EO and OA after adjusting for covariates. As shown in Table 2, In both Model 1 and Model 2, log-10 transformed EO was positively associated with the risk of OA (Model 1: OR = 1.54, 95%CI: 1.16–2.05, P = 0.005; Model 2: OR = 1.76, 95%CI: 1.12–2.77, P = 0.025). Furthermore, the risk was significantly higher in the Q4 group compared to the Q1 group (Q4 in Model 1: OR = 1.74, P = 0.003; Q4 in Model 2: OR = 2.02, P = 0.052), suggesting that higher EO concentration exposure may increase the risk of OA. In addition, trend analyses indicated a significant association between log-10 transformed EO concentration and OA risk in both Model 1 (P for trend = 0.014) and Model 2 (P for trend = 0.041).

The RCS model further investigated the non-linear relationship between EO and the risk of OA. Figure 3A reveals a U-shaped nonlinear association between log-10 transformed EO and OA risk in the unadjusted model (P for overall < 0.001, P for non-linearity = 0.001). Figure 3B demonstrates a J-shaped non-linear relationship between log-10 transformed EO and OA risk after adjusting for all confounding variables (P for overall < 0.001, P for non-linearity = 0.001).

Figure 4 presents the results of the stratified analysis and interaction test for all covariates. Overall, the association between EO exposure and OA risk was not significant. However, in specific subgroups, EO exposure was strongly associated with either a positive or negative risk of OA. For instance, in the 20–40 and 40–65 age groups, as well as in groups with low calcium and low 25-hydroxyvitamin D2 + D3 levels, EO exposure was associated with a increased OA risk. Conversely, in never smokers (Not at all), EO exposure was linked to a decreased OA risk. Interaction tests indicated that age, smoking status, total calcium level, and 25-hydroxyvitamin D2 + D3 level are potential modifiers of the relationship between EO exposure and OA risk (all P < 0.05), suggesting these factors may influence the impact of EO exposure on OA risk.