The National Health and Nutrition Examination Survey (NHANES) is a nationwide US survey assessing the health and nutrition of children and adults. Using multistage probability sampling, it collects data through interviews, physical exams, and lab tests every 2 years. The NHANES procedures are approved by the ethics committee of the National Center for Health Statistics (NCHS), with written consent from all adult participants. Further data analysis follows NCHS guidelines (23). The official website¹ provides free access to NHANES.

The initial sample for this study consisted of 30,468 individuals enrolled in three consecutive NHANES cycles from 2009 to 2014. Participants younger than 18 years old (n = 11,964) were excluded first. Then, those missing serum BFRs data (n = 13,046), periodontitis information (n = 2,136), covariates data (n = 867), and weight information (n = 10) were excluded. Ultimately, 2,445 individuals were included in the study (Figure 1).

The NHANES Laboratory Procedures Manual provides detailed instructions for the collection, storage, and processing of blood specimens. In the NHANES database, 12 different BFRs are quantified using automated liquid–liquid extraction and sample purification methods. These BFRs include PBDE17, PBDE28, PBDE47, PBDE66, PBDE85, PBDE99, PBDE100, PBDE153, PBDE154, PBDE183, PBDE209, and PBB153. Quantitative analysis is conducted using Isotope Dilution High-Resolution Gas Chromatography/Mass Spectrometry (GC/IDHRMS), as described by Johnson et al. (23). When levels fell below the detection limit, they were imputed as the detection limit divided by the square root of 2, following established protocols. To ensure the robustness of our study, we focused on nine BFRs with detection rates exceeding 70%, namely PBDE28, PBDE47, PBDE85, PBDE99, PBDE100, PBDE153, PBDE154, PBDE209, and PBB153 (22).

Periodontitis diagnosis relied on measurements of periodontal pocket probing depth (PD) and attachment loss (AL). Following CDC criteria (24), mild periodontitis was defined as having two or more interproximal sites with AL ≥ 3 mm and two or more interproximal sites with PD ≥ 4 mm (not on the same tooth), or one interproximal site with PD ≥ 5 mm. Moderate periodontitis was diagnosed when two or more interproximal sites had AL ≥ 4 mm or two or more interproximal sites had PD ≥ 5 mm, but not on the same tooth. Severe periodontitis was diagnosed when two or more interproximal sites had AL ≥ 6 mm (not on the same tooth), and at least one interproximal site had PD ≥ 5 mm. Patients not meeting any of these conditions were diagnosed with “no periodontitis.” The primary outcome of interest was the presence of moderate or severe periodontitis, while all other cases were categorized into the reference group.

Socio-demographic factors, physical examination results, and health-related factors that might confound the correlation between BFRs and periodontitis were considered covariates. The covariates in the study included age, gender (male or female), race/ethnicity (non-Hispanic Black, non-Hispanic White, other Hispanic, Mexican American, or other race), education level (less than high school, high school diploma, or more than high school), body mass index (BMI), total energy intake, smoking status, alcohol use (yes or no), family poverty-to-income ratio (PIR), sleep trouble (yes or no), diabetes mellitus (yes or no), hyperlipidemia (yes or no), and hypertension (yes or no).

BMI was calculated as weight (kg) divided by height (m²). Individuals were classified as having normal BMI (<25), overweight (25.0–29.9), or obese (≥30). Total energy intake was obtained from a 24-h food recall. For categorization, low intake was defined as males consuming less than 2,000 kcal/day and females consuming less than 1,600 kcal/day. Adequate intake was considered for males consuming between 2,000 and 3,000 kcal/day and females consuming between 1,600 and 2,400 kcal/day. High intake was determined for males consuming over 3,000 kcal/day and females consuming over 2,400 kcal/day. Smoking status was categorized into three groups: never use, former use, and current use, based on two questions: (a) “Have you smoked at least 100 cigarettes in your entire life?” (SMQ020), and (b) “Do you now smoke cigarettes?” (SMQ935). Never use was identified as having smoked fewer than 100 cigarettes in their entire life. Former use was identified as having smoked at least 100 cigarettes in their entire life, but not currently smoking. Current use was identified as having smoked at least 100 cigarettes in their entire life, and currently smoking. Alcohol use was defined as consuming 12 or more alcoholic drinks in 1 year. PIR was categorized as follows: ≤1.30 for low-income, 1.31–3.50 for middle-income, and >3.50 for high-income. Sleep trouble was assessed through the question, “Have you reported to a healthcare provider that you experience difficulty sleeping?” (SLQ050). Responses included “Yes,” “No,” “Refused,” and “Do not know.” Participants who answered “Yes” were classified as having sleep trouble, while “Refused” and “Do not know” responses were considered missing data. Diabetes mellitus was defined by any of the following: physician-diagnosed diabetes, hemoglobin A1c (HbA1c) ≥ 6.5%, fasting blood glucose ≥7.0 mmol/L, 2-h oral glucose tolerance test (OGTT) blood glucose ≥11.1 mmol/L, or the use of prescribed diabetes medication or insulin. When any of these conditions were met, hyperlipidemia was diagnosed: triglyceride levels ≥150 mg/dL, total cholesterol levels ≥200 mg/dL, LDL cholesterol levels ≥130 mg/dL, HDL cholesterol levels <40 mg/dL for men or <50 mg/dL for women, or if the person was on lipid-lowering medications. Hypertension was determined based on in-home interviews and blood pressure measurements taken at the mobile examination center (MEC). Participants were considered to have hypertension if they met any of the following criteria: (1) Average systolic blood pressure ≥ 140 mmHg; (2) Average diastolic blood pressure ≥ 90 mmHg; (3) Self-reported hypertension; and (4) Ever prescribed anti-hypertensive medications (25).

The NHANES data analysis followed rigorous statistical procedures, adhering to NHANES weighting recommendations. Weighted analyses were conducted utilizing the 2-year examination weight (WTMEC2YR), primary sampling units (SDMVPSU), and strata (SDMVSTRA). Continuous variables with normal distribution were presented using means ± standard errors (SE), while variables with non-normal distribution were presented using medians and interquartile ranges (IQR). Categorical variables were expressed as numbers and proportions. T tests and Kruskal–Wallis H tests were used to test differences for continuous parameters. Pearson’s chi-squared test for categorical variables. Considering the right-skewed distributions of exposures, serum BFRs concentrations were ln-transformed. Spearman’s correlation coefficients were used to examine correlations between each pair of ln-transformed BFRs. Survey-weighted generalized linear regressions were conducted to investigate the associations between BFRs exposure and periodontitis. Ln-transformed BFRs served as quantitative variables and were then divided into four quartile groups to be used as qualitative variables. Three models were constructed: Model 1, the crude model, was not adjusted. Model 2 was adjusted for age, gender, race, PIR, education levels, BMI, energy intake, smoking status and alcohol use. Model 3, built upon Model 2, included adjustments for sleep trouble, hypertension, diabetes and hyperlipidemia. To investigate the potential nonlinear relationship between BFRs and periodontitis, we conducted a restricted cubic spline (RCS) analysis in the fully adjusted model. Finally, we employed quantile weighted quantile sum (WQS) regression and quantile g-computation (QGC) analysis to examine the overall effect of BFRs on periodontitis after adjusting for confounders. WQS regression constructs a weighted index to test associations in both directions (26). In this study, the data were randomly divided into a training set (40%) and a validation set (60%). An individual weight exceeding 1/9 (since there were 9 chemical components in the study) was considered above the threshold. QGC analysis uniformly integrates effects without directional constraints and is a straightforward method for estimating the coefficients of both individual and combined exposure variables on outcome variables (27).

Statistical analyses were completed by R software (version 4.3.2). p value <0.05 on two sides was seen as statistically significant.

A total of 2,445 participants were ultimately included in this study. Table 1 displays the baseline characteristics of these participants. The average age of the participants was 51 years old, with 48.85% being male. Significant differences were observed between the periodontitis group and the control group in terms of age, race, education levels, energy intake, PIR, smoking status, DM, and hypertension. However, the data did not reveal any significant differences between these two groups in terms of gender, BMI, alcohol consumption, sleep trouble, or hyperlipidemia. Additionally, periodontitis patients were found to have a higher likelihood of exposure to elevated levels of various BFRs, including PBDE28, PBDE47, PBDE85, PBDE99, PBDE100, PBDE154, PBDE209, and PBB153, based on observations.

In the fully adjusted model, several ln-transformed BFR components were positively correlated with periodontitis, including serum PBDE28 (OR: 1.79; 95% CI: 1.43–2.23; p < 0.001), PBDE47 (OR: 1.50; 95% CI: 1.22–1.84; p < 0.001), PBDE85 (OR: 1.28; 95% CI: 1.09–1.50; p = 0.004), PBDE99 (OR: 1.34; 95% CI: 1.13–1.58; p < 0.001), PBDE100 (OR: 1.30; 95% CI: 1.07–1.59; p = 0.010), PBDE154 (OR: 1.33; 95% CI: 1.21–1.58; p = 0.003), PBB153 (OR: 1.45; 95% CI: 1.22–1.72; p < 0.001). When expressed in quartiles, PBDE28 (P for trend = 0.025), PBDE85 (P for trend = 0.011), PBDE100 (P for trend = 0.009), PBDE154 (P for trend = 0.019), and PBB153 (P for trend = 0.011) displayed an increased OR with periodontitis. The relationship between the BFR components and periodontitis is shown in Table 2.

An analysis of the RCS regression is shown in Figure 2. After adjusting for all covariates, we found a significant nonlinear correlation between ln-transformed PBDE28 (p = 0.021), PBED47 (p = 0.010), PBDE85 (p = 0.005), PBDE100 (p = 0.020), PBDE154 (p = 0.040) and PBB153 (p = 0.001) with periodontitis in the RCS regression. Among them, PBDE85 exhibited a distinctive inverted U-shaped association, while PBDE47, PBDE154, and PBDE100 showed a more plateau-like trend.

The correlations between each pair of ln-transformed BFRs, as indicated by the Spearman’s rank correlation coefficient, are presented in Figure 3. Among these components, PBDE47, PBDE85, PBDE99, PBDE100, and PBDE154 exhibited strong correlations with each other. The analysis using WQS regression and QGC analysis found a positive association between the exposure to mixture BFRs and the prevalence of periodontitis. The WQS index of mixture BFRs exposure was positively associated with periodontitis prevalence (OR: 1.52; 95% CI: 1.30–1.79, p < 0.001), with PBB153, PBDE28, PBDE209 and PBDE99 having a relatively stronger impact (Figure 4). However, no significant association was observed when analyzing in the negative direction (OR: 0.91; 95% CI: 0.82, 1.02; p = 0.10). In the QGC analysis, exposure to mixture BFRs also demonstrated similar outcomes, showing a positive association with the prevalence of periodontitis (QGC β: 1.29; 95% CI: 1.24–1.36, p < 0.001). Within the serum BFRs, PBB153, PBDE28, PBDE154, PBDE209, PBDE99, and PBDE47 exhibited positive weights, while PBDE153, PBDE85, and PBDE100 displayed negative weights (Figure 5).

Firstly, the data were derived from a nationally representative sample of adults in the United States, enhancing the generalizability and applicability of the findings. Secondly, the study not only analyzed the relationship between individual BFR components and periodontitis but also examined the impact of BFR mixtures, providing a more comprehensive perspective. Thirdly, complex statistical models were developed to control for confounding factors as much as possible, thereby increasing the reliability and scientific validity of the conclusions.

Firstly, as a cross-sectional study, it cannot establish a causal relationship between BFR exposure and periodontitis. Secondly, due to the lack of relevant data in the NHANES dataset, this study did not include newer BFRs developed as replacements for PBDEs. Thirdly, although we made efforts to control for confounding variables, there may still be unaccounted factors influencing the results. Therefore, further research is needed to improve control of potential confounders, and large-scale longitudinal studies are necessary to better elucidate the causal relationship between BFR exposure and periodontitis.