Statistical analysis

Mean and standard deviation were used to report continuous variables with a normal distribution, whereas median and interquartile range were used to report skewed data. Categorical variables were reported as frequency and proportion. For categorical variables, the Chi-square test was utilized, while the t-test or Mann–Whitney U test was utilized for continuous variables, based on whether the data were normally distributed or not, to compare group differences between participants with SSD or self-reported trouble sleeping and the control group.

The odds ratios (ORs) and 95% confidence intervals (CIs) for PAHs associated with SSD or self-reported trouble sleeping were evaluated by employing the weighted multivariate logistic regression model. The statistical model comprised creatinine-adjusted PAHs as both a continuous variable (with log-transformation for non-normal distribution) and a categorical variable having the lowermost quartile as the reference group. The crude model was not adjusted, while Model I was adjusted for age and gender, along with adjusting Model II for age, gender, ethnic background, educational status, family PIR, and general health.

Furthermore, the current study utilized restricted cubic spline (RCS) curves with four knots to depict the dose–response relationship between PAHs and the prevalence of SSD and self-reported trouble sleeping following the adjustment for model variables (52). Additionally, the combined associations of all six PAH metabolites with SSD or self-reported trouble sleeping were assessed, and the relative contribution of each component in the mixture was determined for the positive association using weighted quantile sum (WQS) regression, a novel statistical method in environmental epidemiology (53). The WQS regression model was created to assess how mixed exposure affected health outcomes. All PAH metabolite concentrations were initially ranked in quartiles, and all data were then randomly divided into training and validation sets. The weighted index of each PAH metabolite represented its contribution to the positive association and was constrained to a range of 0–1, summing up to 1. A total of 1,000 bootstrap replicates were performed to estimate the effect size and 95% CIs based on previously published studies, followed by the random distribution of data into a 60% validation set and a 40% test set.

Sensitivity analysis was further performed to evaluate the robustness of our results. Adults with chronic conditions including (hypertension, cardiovascular disease, diabetes mellitus, and chronic obstructive pulmonary disease) were excluded to explore the associations. Participants were diagnosed as hypertension by: average systolic pressure ≥ 140 mmHg or average diastolic pressure ≥ 90 mmHg; or self-reported hypertension; or taking anti-hypertension drugs. Participants were diagnosed as diabetes mellitus by: doctor told you have diabetes; or glycohemoglobin HbA1() > 6.5%; or fasting glucose ≥7.0 mmoL/L; or random blood glucose ≥11.1 mmoL/L; or two-hour OGTT blood glucose ≥11.1 mmoL/L; or use of diabetes drugs or insulins. Participants were diagnosed as cardiovascular disease by doctor r told congestive heart failure, coronary heart disease, or angina pectoris, or heart attack, or stroke. Participants were diagnosed as COPD by: FEV1/FVC < 0.7; or ever told have emphysema; or use drugs including selective phosphodiesterase-4 inhibitors, mast cell stabilizers, leukotriene modifiers, inhaled corticosteroids. Besides, we conducted a primary study using train datasets (2005–2010 NHANES waves) for discovery of associations between urinary PAH metabolites with the prevalence of SSD and self-reported trouble sleeping, and used test datasets (2011–2016 NHANES waves) to replicate analysis.

The R software (version 4.1.3) was utilized for conducting all statistical analyses in the current study. The significance level for all statistical tests was set at two-tailed p < 0.05.