Per- and polyfluoroalkyl substances (PFAS) exposure and cardiovascular risk: lipid profile as a mediator Authors: Xianghai Wang, Yongping Cao, Xuli Chen, Guojin Jian, Fei Ma, Hengyang Zhang, Qiuyu Wang, Wenwen Xiao Published: 2025 ABSTRACT AbstractBackgroundIn recent years, amidst the phase-out of long-chain per- and polyfluoroalkyl substances (PFAS), the true relationship between PFAS exposure and lipid metabolism or cardiovascular disease (CVD), as well as the role of lipid profiles in this association, remains unclear.MethodsData from 25 531 NHANES participants (age ≥ 20 years old) enrolled between 2015 and 2020 were examined. To assess links between individual PFAS and cardiovascular disease (CVD) as well as lipid measures, both logistic and multivariable linear regression analyses were performed. Nonlinear exposure–response patterns with CVD were fitted using restricted cubic splines. In addition, the combined impact of PFAS mixtures on CVD risk was investigated via Bayesian kernel machine regression (BKMR), weighted quantile sum (WQS) regression, and quantile g-computation (Q-gcomp), and mediation analyses were evaluated using causal mediation models.ResultsAfter comprehensive adjustment, each log-unit increase in PFNA, n-PFOA, n-PFOS, and Sm-PFOS was inversely associated with CVD risk. There was a significant inverse trend in the associations of n-PFOA and Sm-PFOS with CVD. PFDeA, PFHxS, PFNA, PFUA, n-PFOA, n-PFOS, and Sm-PFOS were significantly positively associated with total cholesterol (TC). PFUA was significantly negatively associated with triglycerides (TG), and specific PFAS also showed significant positive associations with low-density lipoprotein (LDL) and high-density lipoprotein (HDL). Mixture exposure analysis indicated a significant inverse trend between PFAS mixtures and CVD, with Sm-PFOS contributing the most weight in the mixture index. Through mediation analysis, we found that total cholesterol and LDL cholesterol serve as significant intermediaries in the relationships between PFNA, n-PFOA, n-PFOS, Sm-PFOS, and cardiovascular disease.ConclusionThese findings imply that contemporary PFAS exposure profiles may confer differential cardiovascular effects, in part through lipid-mediated pathways, and highlight the need for continued monitoring and mechanistic studies to inform risk assessment and regulatory decisions.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12944-025-02811-3.Keywords: Per- and polyfluoroalkyl substances (PFAS), Cardiovascular disease (CVD), Lipid profile, National health and nutrition examination survey (NHANES), Mediation effect. Introduction Per- and polyfluoroalkyl substances (PFAS) are a large class of synthetic fluorinated organic compounds characterized by strong carbon-fluorine bonds, which confer exceptional thermal stability and resistance to degradation [1]. First synthesized in the 1940 s, PFAS quickly found widespread application in industry and everyday products, ranging from nonstick cookware and water-resistant fabrics to greaseproof food wrappers, firefighting foams, and various chemical processes [2, 3]. Humans encounter these compounds through several routes, including drinking contaminated water or eating tainted food, inhaling indoor air or household dust, and skin contact with PFAS-treated materials [4]. Over the past decades, the production and use of long-chain PFAS such as perfluorooctanoic acid (PFOA) and perfluorooctane sulfonate (PFOS) have raised serious environmental and health concerns due to their persistence, bioaccumulation, and potential toxicity [5]. In response, global regulatory agencies have progressively phased out legacy long-chain PFAS like PFOA and PFOS, guided by initiatives such as the U.S. Environmental Protection Agency’s 2010 Stewardship Program and the Stockholm Convention on Persistent Organic Pollutants. However, alternative PFAS, including short-chain compounds and those with novel chemical structures, have entered the market, many of which remain poorly characterized in terms of toxicity and long-term health effects [6, 7]. In recent years, public and scientific concern about PFAS has grown significantly, with major health authorities such as the U.S. National Academies and the European Food Safety Authority (EFSA) calling for enhanced regulation, biomonitoring, and toxicological evaluation of PFAS. In the past decade, accumulating epidemiological evidence has linked PFAS exposure to increased risk of cardiovascular disease (CVD). Cross-sectional analyses in the United States have demonstrated significant associations between serum concentrations of legacy PFAS, such as PFOA, PFOS, and perfluorononanoic acid (PFNA), and elevated risk of CVD outcomes, including coronary heart disease, stroke, and heart failure [8, 9]. Similarly, a large-scale Italian study reported increased mortality from ischemic heart disease and other cardiovascular causes in populations residing in areas with PFAS-contaminated drinking water [10]. Recent large cohort studies and meta-analyses have further reported positive associations between PFAS homologues, including PFHxS, PFNA, and PFOS, and cardiovascular events or mortality [11, 12]. However, as regulatory interventions have shifted production patterns and altered human exposure profiles from long-chain PFAS to mixtures of short- and long-chain compounds, there is an urgent need to re-evaluate these associations using contemporary exposure data. Lipid metabolism plays a pivotal role in the development of CVD. Alterations in lipid profiles, such as elevated low-density lipoprotein cholesterol (LDL-C) and triglyceride levels, are well-recognized contributors to CVD risk [13–15]. Emerging evidence linked PFAS exposure to dyslipidemia, characterized by higher total cholesterol (TC), LDL-C, and triglycerides (TG) [16, 17]. For example, a meta-analysis pooling 29 investigations found that elevated serum PFOA and PFOS corresponded with increased blood lipids [18]. In addition, PFAS-induced oxidative stress and inflammation may disrupt lipid homeostasis and promote atherogenesis [19–21]. Altogether, these observations highlight the need to explore lipidomic pathways as mediators in the PFAS-CVD connection. Given the evolving PFAS exposure landscape resulting from regulatory actions and industrial shifts, it is crucial to elucidate current associations between both individual and mixed PFAS exposures and CVD risk in the general population. Moreover, with increasing evidence that lipid metabolism may mediate the effects of environmental exposures on cardiovascular outcomes, exploring the potential mediating role of lipidomic disturbances is essential. Therefore, this study aimed to investigate the associations between serum PFAS concentrations and CVD prevalence using data from the two latest NHANES cycles (2015–2016 and 2017–2020), focusing on both single-exposure and mixture modeling approaches. We further assessed whether specific lipid parameters mediate the observed PFAS–CVD associations. By leveraging biomonitoring data and applying advanced mixture modeling and mediation analysis techniques, this study provides up-to-date and robust epidemiological evidence on PFAS-related cardiovascular risks. The results enhance our understanding of how modern PFAS exposures affect cardiometabolic health and provide valuable insights for risk assessors and public health policy. Method Study population NHANES is a nationwide health survey program conducted by the National Center for Health Statistics (NCHS), a division of the U.S. Centers for Disease Control and Prevention (CDC). Its primary objective is to assess the health and nutritional status of the U.S. population and to provide scientific evidence for public health policy, disease prevention, and health promotion. In this study, to investigate the latest associations between PFAS exposure and CVD, as well as the potential role of lipid profiles, data from the two recent survey cycles (2015–2016 and 2017–2020) were analyzed. Figure 1 illustrates the inclusion and exclusion criteria for the study population. Flowchart of the population included in the study The final analytic sample size of 1,383 participants was derived from an initial pool of 25,531 individuals. Specifically, 20,466 participants were excluded due to missing data on chemical exposures, and an additional 2,704 were excluded for missing lipid profiles. Further exclusions included 378 participants aged under 20 years, 271 with missing family income ratio data, 1 with missing serum cotinine, and 21 with missing BMI data. Additionally, 256 participants were excluded due to missing alcohol drinking status, 1 due to missing hypertension status, 36 due to missing diabetes status, and 14 due to missing CVD outcomes. Detection of serum PFAS PFAS in serum samples were measured using high-performance liquid chromatography coupled with tandem mass spectrometry (HPLC-MS/MS). Serum samples underwent protein precipitation and solid-phase extraction to isolate PFAS compounds. Quantification of each analyte was performed using isotope dilution with ^13 C-labeled internal standards. Strict quality control procedures were implemented, including the use of blank samples, spiked controls, and repeat analyses to monitor precision and consistency. This method enabled the simultaneous detection of multiple PFAS, with limits of detection (LOD) typically as low as 0.1 ng/mL. For analyte concentrations below the LOD, values were imputed by dividing the LOD by the square root of 2. In this study, the PFAS selected as the exposure variables of interest included perfluorodecanoic acid (PFDeA), perfluorohexane sulfonic acid (PFHxS), 2-(N-methylperfluorooctanesulfonamido) acetic acid (Me-PFOSA-AcOH), perfluorononanoic acid (PFNA), perfluoroundecanoic acid (PFUA), n-perfluorooctanoic acid (n-PFOA), n-perfluorooctane sulfonic acid (n-PFOS), and perfluoromethylheptane sulfonic acid isomers (Sm-PFOS). Assessment of CVD CVD risk was defined as self-reported physician-diagnosed cardiovascular events. Cardiovascular health status was assessed using a standardized medical questionnaire. Participants were asked by trained interviewers, “Has a doctor or other health professional ever told you that you had any of the following conditions: heart attack, coronary heart disease, angina pectoris, stroke, or congestive heart failure?” In this study, participants who reported having any of these cardiovascular events were classified as CVD-positive, while those who answered “no” to all related questions were classified as CVD-negative. Lipid profile After an overnight fast, blood was drawn and processed according to standardized laboratory protocols to quantify lipid levels. Total cholesterol and triglycerides were assayed enzymatically using automated analyzers, whereas HDL-C was measured directly. LDL-C concentrations were then estimated chiefly via the Friedewald formula [22]: Covariates The model was adjusted based on the following covariates. Demographic and lifestyle variables were collected by trained personnel using standardized questionnaires during the NHANES household interview. Age (years), sex (male/female), race/ethnicity (non-Hispanic white, non-Hispanic black, Mexican American, other Hispanic, other), education (less than 9th grade, 9–11th grade, high school graduate/GED or equivalent, some college or AA degree, college graduate or above), and household income and poverty ratio (≤ 1.30, 1.31–3.50, > 3.50) were self-reported by the subjects at the time of the interview. Smoking status was defined by the question “Do you smoke now?” (yes/no), reflecting current tobacco use. Serum cotinine was measured serum cotinine (ng/mL) in fasting blood to objectively reflect recent active and passive smoking. Alcohol consumption was defined as a binary variable (yes/no) based on the survey item ALQ151: ‘In the past 12 months, have you had at least 12 drinks of any type of alcoholic beverage?‘. Participants answering ' yes ' were classified as alcohol consumers, others were classified as non-consumers. Hypertension and diabetes were defined by self-report of physician diagnosis (‘Has your doctor ever told you that you have hypertension/diabetes?‘), reflecting any diagnosis before the medical examination. Body mass index (BMI) was measured at the mobile medical examination center (MEC) based on standardized height and weight readings (≤ 25, 25–30, > 30). Statistical analysis Given that we were interested in biological associations rather than population prevalence estimates, we did not use the NHANES sampling weights but instead adjusted for key demographic variables (age, sex, race/ethnicity) and other confounders that constitute the survey weights. The characteristics of study participants were summarized using means ± SD for continuous measures and counts (percentages) for categorical variables. Serum PFAS levels were described by percentiles (min, 5th, 10th, 25th, 50th, 75th, 90th, 95th). Pearson correlation coefficients were used to assess correlations among PFAS. PFAS concentrations were log-transformed to approximate normal distribution. Logistic regression models were used to investigate associations between both continuous and quartile-based PFAS concentrations and CVD. Model 1 included demographic covariates (age, sex, education, race/ethnicity, income-to-poverty ratio), while Model 2 additionally adjusted for lifestyle (smoking, alcohol use), clinical history (hypertension, diabetes), BMI, and serum cotinine. Multivariable linear regression was likewise used to link PFAS exposures with lipid profiles. Nonlinear relationships between PFAS concentrations and CVD were evaluated using restricted cubic spline functions. Multiple statistical approaches were applied to assess the effects of PFAS mixture exposure on CVD, including Bayesian Kernel Machine Regression (BKMR), Weighted Quantile Sum (WQS) regression, and Quantile g-Computation (Q-gcomp) models. In the BKMR model, the association between PFAS mixture exposure and CVD (treated as a binary outcome) was assessed with 20,000 iterations of a Markov Chain Monte Carlo (MCMC) algorithm. The overall mixture effect, changes in CVD risk across the 25th to 75th percentiles of individual PFAS while holding other PFAS at the 25th, 50th, or 75th percentile, and single-variable exposure–response functions with other PFAS fixed at the 50th percentile were estimated. We did not apply BKMR hierarchical (group-level) selection because chain-length grouping would yield an uninformative seven long-chain vs. one short-chain PFAS that could obscure PFHxS’s contribution, instead, we used component-wise variable selection treating all eight PFAS equally. For WQS regression, bidirectional associations between PFAS mixtures and CVD were evaluated. Each WQS index was derived through 1,000 bootstrap resamplings, and the weight contribution of each PFAS in the mixture was estimated. In the Q-gcomp model, both the weights and the overall linear association between PFAS mixture exposure and CVD were calculated. Finally, for PFAS showing significant CVD links, mediation analyses (using TC, TG, LDL-C, HDL-C as potential mediators) were conducted. The procedure comprised three steps: Firstly, multivariable regression of each PFAS exposure on the lipid mediator (TC, TG, LDL-C, or HDL-C), adjusting for all covariates. Secondly, multivariable regression of PFAS exposure on CVD status, additionally adjusting for the lipid mediator to estimate the direct effect. Thirdly, computation of the average causal mediation effect (ACME), the average direct effect (ADE), and the proportion of the total effect mediated by each lipid parameter. To assess the stability of the results, sensitivity analyses were performed by comparing the associations between PFAS exposure and both CVD and lipid levels across two distinct regression models. All analyses were carried out in R 4.4.1, employing the “bkmr”, “gWQS”, “qgcomp”, and “mediation” packages. P < 0.05 was considered significant. Results Population characteristics Characteristics of the study population were summarized in Table 1. A total of 1,383 participants were included, of whom 150 (10.8%) had a history of cardiovascular disease (CVD) and 1,233 (89.2%) did not. Participants with CVD were older than those without CVD and more likely to be male. The racial distribution differed by CVD status, with higher proportions of Mexican American and Other Hispanic participants among CVD cases. Smoking, hypertension, and diabetes were more prevalent in the CVD group (P < 0.05). Mean BMI was higher in participants with CVD than in those without, and obesity (BMI ≥ 30 kg/m²) was more frequent. Mean total cholesterol and LDL-C were lower in participants with CVD, whereas differences in triglycerides and HDL-C did not reach significance. Table 1Characteristics of the study in NHANES 2015–2020CharacteristicsOverallNon-CVDCVDP valueN13831233150Age (years, mean (SD))49.98 (17.13)48.26 (16.86)64.07 (12.17)< 0.001Gender (%)0.008Male703 (50.8)611 (49.6)92 (61.3)Female680 (49.2)622 (50.4)58 (38.7)Race (%)0.018Non-Hispanic white185 (13.4)174 (14.1)11 (7.3)Non-Hispanic black143 (10.3)129 (10.5)14 (9.3)Mexican American536 (38.8)468 (38.0)68 (45.3)Other Hispanic329 (23.8)285 (23.1)44 (29.3)Other190 (13.7)177 (14.4)13 (8.7)Education (%)0.203Less than 9th grade90 (6.5)79 (6.4)11 (7.3)9-11th grade148 (10.7)125 (10.1)23 (15.3)High school graduate/GED or equivalent315 (22.8)282 (22.9)33 (22.0)Some college or AA degree492 (35.6)437 (35.4)55 (36.7)College graduate or above338 (24.4)310 (25.1)28 (18.7)Family poverty-income ratio (mean (SD), %)2.70 (1.63)2.71 (1.64)2.60 (1.61)0.457< 1.3366 (26.5)328 (26.6)38 (25.3)1.3–3.5533 (38.5)471 (38.2)62 (41.3)>=3.5484 (35.0)434 (35.2)50 (33.3)Smoking (%)< 0.001Yes665 (48.1)562 (45.6)103 (68.7)No718 (51.9)671 (54.4)47 (31.3)Alcohol (%)0.292Yes213 (15.4)185 (15.0)28 (18.7)No1170 (84.6)1048 (85.0)122 (81.3)Hypertension (%)< 0.001Yes483 (34.9)379 (30.7)104 (69.3)No900 (65.1)854 (69.3)46 (30.7)Diabetes (%)< 0.001Yes208 (15.0)158 (12.8)50 (33.3)No1175 (85.0)1075 (87.2)100 (66.7)BMI (kg/m2, mean (SD), %)29.94 (7.64)29.75 (7.54)31.47 (8.24)0.009< 25363 (26.2)339 (27.5)24 (16.0)25–29.9.9432 (31.2)378 (30.7)54 (36.0)>=30588 (42.5)516 (41.8)72 (48.0)Serum cotinine (ng/mL, mean (SD))59.19 (129.38)59.15 (128.38)59.59 (137.72)0.968Total cholesterol (mg/dL, mean (SD))184.84 (40.61)186.52 (39.98)171.02 (43.18)< 0.001Low-density lipoprotein (mg/dL, mean (SD))109.50 (35.78)111.16 (35.38)95.83 (36.22)< 0.001Triglyceride (mg/dL, mean (SD))104.21 (62.27)103.12 (61.95)113.13 (64.36)0.063High-density lipoprotein (mg/dL, mean (SD))54.50 (16.11)54.75 (16.06)52.51 (16.43)0.108Abbreviations Q quartile, SD standard deviation; BMI body mass index. Categorical variables were analyzed using chi-square tests continuous variables were analyzed using analysis of variance Characteristics of the study in NHANES 2015–2020 Abbreviations Q quartile, SD standard deviation; BMI body mass index. Categorical variables were analyzed using chi-square tests continuous variables were analyzed using analysis of variance Distribution of PFAS Table S1 summarizes the 2015–2020 serum PFAS measurements by reporting the minimum, selected percentiles, and maximum values. The median (50th percentile) and interquartile range (25th–75th percentiles) for each compound were: PFDeA, 0.2 ng/mL (0.1–0.3); PFHxS, 1.2 ng/mL (0.7–2.1); Me-PFOSA-AcOH, 0.07 ng/mL (0.07–0.2); PFNA, 0.5 ng/mL (0.3–0.9); PFUA, 0.1 ng/mL (0.07–0.2); n-PFOA, 1.4 ng/mL (0.9–2.25); n-PFOS, 3.2 ng/mL (1.8–5.75); and Sm-PFOS, 1.4 ng/mL (0.7–2.5). As illustrated in Fig. S1, boxplot distributions for these analytes were tightly clustered. Spearman’s rank correlations (Fig. S2) showed that pairwise correlations among PFAS ranged from weak to strong (r = 0.09–0.75), indicating variable interdependence across compounds. Associations between serum PFAS concentrations and CVD The results of the relationship between serum PFAS concentration and CVD were presented in Table 2. In the minimally adjusted models (Model 1), each log-unit increase in PFNA, n-PFOA, n-PFOS, and Sm-PFOS was inversely associated with CVD risk (PFNA: OR, 0.72, 95% CI, [0.57, 0.91], P = 0.007; n-PFOA: OR, 0.60, 95% CI, [0.45, 0.80], P = 0.001; n-PFOS: OR, 0.71, 95% CI, [0.57, 0.89], P = 0.003; Sm-PFOS: OR, 0.64, 95% CI, [0.50, 0.83], P = 0.001). Quartile analyses showed that participants in the highest exposure quartile (Q4) of n-PFOA and Sm-PFOS had markedly lower odds of CVD compared with Q1 (n-PFOA Q4 vs. Q1: OR, 0.13, 95% CI, [0.03, 0.46], P-trend = 0.001; Sm-PFOS Q4 vs. Q1: OR, 0.45, 95% CI, [0.16, 1.37], P-trend = 0.020). PFNA also exhibited a significant dose-response trend (P-trend = 0.038), with Q4 vs. Q1 OR 0.49 (95% CI, [0.19, 1.21]). Table 2Association between PFAS and CVD in NHANES 2015–2020VariablesOR 95% CI (Continuous)P valueQ1Q2Q3Q4P-trendPFDeAModel 10.78 (0.61, 1.00)0.051Ref0.92 (0.61, 1.39)0.64 (0.35, 1.12)0.57 (0.18, 1.50)0.101Model 20.82 (0.64, 1.06)0.126Ref0.97 (0.63, 1.47)0.69 (0.37, 1.24)0.59 (0.18, 1.59)0.180PFHxSModel 10.76 (0.59, 0.98)0.038Ref0.52 (0.24, 1.23)0.41 (0.19, 1.00)0.39 (0.12, 1.19)0.068Model 20.81 (0.63, 1.05)0.112Ref0.54 (0.24, 1.31)0.49 (0.21, 1.22)0.40 (0.12, 1.27)0.183Me-PFOSA-AcOHModel 10.99 (0.77, 1.27)0.937Ref1.03 (0.61, 1.69)0.95 (0.54, 1.62)0.99 (0.44, 2.08)0.914Model 20.96 (0.74, 1.24)0.732Ref1.10 (0.64, 1.83)0.93 (0.52, 1.62)0.76 (0.33, 1.64)0.587PFNAModel 10.72 (0.57, 0.91)0.007Ref0.72 (0.41, 1.30)0.55 (0.30, 1.01)0.49 (0.19, 1.21)0.038Model 20.73 (0.57, 0.94)0.013*Ref0.77 (0.43, 1.43)0.59 (0.32, 1.13)0.48 (0.18, 1.21)0.052PFUAModel 10.88 (0.66, 1.16)0.368Ref1.20 (0.76, 1.87)0.50 (0.25, 0.93)1.03 (0.37, 2.51)0.249Model 20.92 (0.69, 1.24)0.583Ref1.26 (0.78, 2.01)0.53 (0.26, 1.00)1.13 (0.41, 2.82)0.400n-PFOAModel 10.60 (0.45, 0.80)0.001Ref0.48 (0.25, 0.99)0.35 (0.17, 0.73)0.13 (0.03, 0.46)0.001Model 20.64 (0.48, 0.86)0.003*Ref0.59 (0.29, 1.23)0.43 (0.21, 0.91)0.14 (0.03, 0.52)0.002*n-PFOSModel 10.71 (0.57, 0.89)0.003Ref0.67 (0.29, 1.68)0.57 (0.25, 1.44)0.29 (0.09, 0.88)0.033Model 20.76 (0.60, 0.96)0.023*Ref0.88 (0.38, 2.27)0.80 (0.34, 2.06)0.41 (0.13, 1.29)0.127Sm-PFOSModel 10.64 (0.50, 0.83)0.001Ref0.96 (0.39, 2.74)0.62 (0.25, 1.75)0.45 (0.16, 1.37)0.020Model 20.65 (0.50, 0.85)0.001*Ref1.08 (0.42, 3.17)0.66 (0.26, 1.90)0.48 (0.17, 1.53)0.025*Abbreviations Q quartile, Ref reference, OR odd ratio, CI confidence interval. Model 1 adjusted for age, gender, race, education level and family poverty-income ratio. Model 2 further adjusted for smoking status, alcohol consumption, hypertension, diabetes, BMI, and serum cotinine. Odds ratios derived from unweighted logistic regression Association between PFAS and CVD in NHANES 2015–2020 Abbreviations Q quartile, Ref reference, OR odd ratio, CI confidence interval. Model 1 adjusted for age, gender, race, education level and family poverty-income ratio. Model 2 further adjusted for smoking status, alcohol consumption, hypertension, diabetes, BMI, and serum cotinine. Odds ratios derived from unweighted logistic regression After full adjustment, inverse associations remained robust for PFNA (OR, 0.73, 95% CI, [0.57, 0.94], P = 0.013), n-PFOA (OR, 0.64, 95% CI, [0.48, 0.86], P = 0.003), n-PFOS (OR, 0.76, 95% CI, [0.60, 0.96], P = 0.023) and Sm-PFOS (OR, 0.65, 95% CI, [0.50, 0.85], P = 0.001). Participants in Q4 of n-PFOA and Sm-PFOS continued to have significantly reduced CVD odds (n-PFOA: Q4 vs. Q1: OR 0.14, 95% CI, [0.03, 0.52], P-trend = 0.002; Sm-PFOS: Q4 vs. Q1: OR, 0.48, 95% CI, [0.17, 1.53], P-trend = 0.025). No consistent associations were observed for Me-PFOSA-AcOH or PFUA, and initial inverse relations for PFDeA and PFHxS in Model 1 were attenuated after full adjustment. Associations between serum PFAS concentrations and lipid profile The results of the relationship between PFAS and lipid profile were shown in Table 3. In multivariable linear regression models, each log-unit increase in PFDeA, PFHxS PFNA, PFUA, n-PFOA, n-PFOS, and Sm-PFOS was significantly associated with higher total cholesterol (TC). Quartile analyses revealed clear dose-response trends for PFUA (P-trend = 0.012), n-PFOA (P-trend < 0.001), n-PFOS (P-trend = 0.001), and Sm-PFOS (P-trend < 0.001), whereas PFDeA, PFHxS, and PFNA exhibited borderline trends (P-trend = 0.098, 0.077, and 0.050, respectively). Table 3Association between PFAS and lipid profile in NHANES 2015–2020Variablesβ 95% CI (Continuous)P valueQ1Q2Q3Q4P-trendTCPFDeA3.08 (0.22, 5.94)0.035Ref1.23 (−3.47, 5.94)4.93 (−1.76, 11.61)7.33 (−4.73, 19.39)0.098PFHxS3.54 (0.65, 6.44)0.017*Ref6.45 (−1.73, 14.63)11.09 (2.29, 19.89)3.21 (−9.70, 16.12)0.077Me-PFOSA-AcOH0.35 (−2.85, 3.55)0.828Ref1.38 (−4.85, 7.61)−1.41 (−8.32, 5.50)1.52 (−9.48, 12.52)0.996PFNA3.38 (0.57, 6.18)0.018*Ref2.78 (−3.61, 9.16)8.52 (1.66, 15.38)1.13 (−9.93, 12.18)0.050PFUA4.44 (1.07, 7.81)0.010*Ref2.36 (−3.30, 8.02)8.77 (2.00, 15.55)7.40 (−4.60, 19.40)0.012*n-PFOA6.09 (2.68, 9.50)< 0.001*Ref8.09 (0.08, 16.09)15.73 (7.20, 24.26)10.79 (−3.06, 24.64)< 0.001*n-PFOS4.44 (1.65, 7.23)0.002*Ref1.55 (−7.81, 10.91)9.36 (−0.44, 19.16)12.87 (−0.20, 25.94)0.001*Sm-PFOS6.93 (3.93, 9.92)< 0.001*Ref3.02 (−5.64, 11.68)13.50 (4.46, 22.54)15.18 (4.00, 26.35)< 0.001*TGPFDeA−3.68 (−7.88, 0.53)0.087Ref−5.46 (−12.37, 1.44)−9.41 (−19.22, 0.41)−0.32 (−18.02, 17.38)0.114PFHxS0.76 (−3.54, 5.06)0.729Ref−5.37 (−17.51, 6.77)2.83 (−10.22, 15.89)−10.59 (−29.75, 8.57)0.533Me-PFOSA-AcOH−1.76 (−6.49, 2.97)0.466Ref4.87 (−4.35, 14.08)−5.47 (−15.68, 4.75)−1.47 (−17.74, 14.79)0.585PFNA−0.57 (−4.71, 3.57)0.789Ref3.98 (−5.46, 13.41)1.34 (−8.80, 11.47)−8.06 (−24.40, 8.28)0.475PFUA−6.13 (−11.10, −1.16)0.016*Ref−3.00 (−11.33, 5.32)−11.95 (−21.93, −1.98)−17.27 (−34.94, 0.40)0.006*n-PFOA3.13 (−1.93, 8.19)0.226Ref3.57 (−8.33, 15.48)6.97 (−5.71, 19.64)4.55 (−16.04, 25.14)0.290n-PFOS−0.01 (−4.15, 4.12)0.995Ref−0.09 (−13.95, 13.77)3.31 (−11.21, 17.82)−11.81 (−31.17, 7.54)0.765Sm-PFOS2.68 (−1.79, 7.14)0.240Ref0.67 (−12.21, 13.55)9.39 (−4.06, 22.83)2.28 (−14.33, 18.90)0.262LDLPFDeA2.41 (−0.14, 4.96)0.064Ref1.59 (−2.60, 5.77)3.21 (−2.74, 9.17)7.07 (−3.67, 17.81)0.130PFHxS2.02 (−0.56, 4.61)0.124Ref7.32 (0.03, 14.61)8.26 (0.42, 16.10)1.14 (−10.36, 12.64)0.465Me-PFOSA-AcOH−0.08 (−2.92, 2.77)0.958Ref0.70 (−4.85, 6.25)−1.72 (−7.87, 4.44)0.13 (−9.66, 9.93)0.781PFNA3.15 (0.66, 5.65)0.013*Ref3.37 (−2.31, 9.05)8.07 (1.97, 14.17)−0.23 (−10.06, 9.61)0.070PFUA3.07 (0.06, 6.07)0.046*Ref1.82 (−3.22, 6.86)6.71 (0.67, 12.74)3.03 (−7.66, 13.72)0.057n-PFOA3.54 (0.50, 6.58)0.023*Ref5.87 (−1.28, 13.01)10.19 (2.58, 17.80)5.06 (−7.30, 17.42)0.023*n-PFOS3.69 (1.20, 6.17)0.004*Ref4.40 (−3.93, 12.73)10.19 (1.47, 18.92)13.59 (1.96, 25.23)0.001*Sm-PFOS5.70 (3.01, 8.38)< 0.001*Ref4.28 (−3.48, 12.03)12.49 (4.40, 20.58)14.40 (4.40, 24.40)< 0.001*HDLPFDeA1.40 (0.40, 2.41)0.006*Ref0.74 (−0.91, 2.40)3.60 (1.25, 5.95)0.29 (−3.94, 4.52)0.029*PFHxS1.38 (0.36, 2.40)0.008*Ref0.25 (−2.63, 3.13)2.32 (−0.78, 5.42)4.29 (−0.26, 8.84)0.007*Me-PFOSA-AcOH0.79 (−0.32, 1.91)0.164Ref−0.29 (−2.47, 1.88)1.42 (−0.99, 3.83)1.71 (−2.13, 5.54)0.227PFNA0.34 (−0.65, 1.32)0.504Ref−1.41 (−3.66, 0.84)0.16 (−2.26, 2.57)2.99 (−0.90, 6.88)0.115PFUA2.60 (1.42, 3.78)< 0.001*Ref1.18 (−0.80, 3.16)4.44 (2.07, 6.81)7.83 (3.64, 12.03)< 0.001*n-PFOA1.92 (0.72, 3.11)0.002*Ref1.51 (−1.31, 4.32)4.13 (1.13, 7.12)4.83 (−0.03, 9.70)< 0.001*n-PFOS0.76 (−0.22, 1.75)0.129Ref−2.90 (−6.20, 0.40)−1.51 (−4.96, 1.94)1.60 (−3.00, 6.20)0.162Sm-PFOS0.71 (−0.34, 1.77)0.186Ref−1.41 (−4.45, 1.63)−0.84 (−4.02, 2.34)0.34 (−3.58, 4.27)0.619Abbreviations Q quartile, Ref reference, OR odd ratio, CI confidence interval, TC total cholesterol, TG triglyceride, LDL low-density lipoprotein, HDL high-density lipoprotein. Model 1 adjusted for age, gender, race, education level and family poverty-income ratio. Model 2 further adjusted for smoking status, alcohol consumption, hypertension, diabetes, BMI, and serum cotinine. Odds ratios derived from unweighted multiple linear regression Association between PFAS and lipid profile in NHANES 2015–2020 Abbreviations Q quartile, Ref reference, OR odd ratio, CI confidence interval, TC total cholesterol, TG triglyceride, LDL low-density lipoprotein, HDL high-density lipoprotein. Model 1 adjusted for age, gender, race, education level and family poverty-income ratio. Model 2 further adjusted for smoking status, alcohol consumption, hypertension, diabetes, BMI, and serum cotinine. Odds ratios derived from unweighted multiple linear regression For triglyceride (TG), only PFUA was significantly associated with serum TG. Participants in the third quartile of PFUA had a TG that was 11.95 mg/dL lower than those in the first quartile (95% CI, [−21.93, −1.98]), and there was a significant dose-response trend across quartiles (P trend = 0.006). Each log-unit increase in PFNA, PFUA, n-PFOA, n-PFOS, and Sm-PFOS was independently associated with higher LDL-C. In quartile analyses, participants in the highest Sm-PFOS quartile had 14.40 mg/dL (95% CI, [4.40, 24.40]) greater LDL‐C than those in the lowest quartile (P-trend < 0.001), while n-PFOS and n-PFOA also showed significant dose–response trends (P-trend = 0.001, P-trend = 0.023). Each log-unit increase in PFDeA, PFHxS, PFUA, and n-PFOA was significantly associated with higher HDL-C. Quartile analyses confirmed dose–response relationships for PFDeA (Q3 vs. Q1: β, 3.60, 95% CI, [1.25, 5.95], P-trend = 0.029), PFHxS (Q4 vs. Q1: β, 4.29; 95% CI, [−0.26, 8.84], P-trend = 0.007), PFUA (Q4 vs. Q1: β, 7.83, 95% CI, [3.64, 12.03], P-trend < 0.001) and n-PFOA (Q3 vs. Q1: β, 4.13, 95% CI, [1.13, 7.12], P-trend < 0.001). Nonlinear relationship curve between PFAS and CVD In the above analysis, it was observed that there might be a nonlinear trend between PFAS and CVD, so the RCS model was subsequently used to fit the nonlinear curve between them (Fig. 2). Statistically significant overall associations were observed for PFHxS, PFNA, n-PFOA, and Sm-PFOS. Tests for non-linearity were non-significant for all compounds (P-nonlinear > 0.05), indicating that the exposure–response functions did not depart from linearity. Nonlinear relationship between PFAS and CVD in NHANES 2015–2020 Relationship between PFAS mixed exposure and CVD Next, we further explored the effects of combined exposure to PFAS on CVD. Figure 3A illustrates the overall association between the PFAS mixture and the continuous CVD outcome. A significant inverse trend was observed: mixture exposures below the 50th percentile were positively associated with CVD, whereas exposures above the 50th percentile were negatively associated. When all other PFAS were fixed at their 75th, 50th, and 25th percentiles, n-PFOA remained inversely associated with CVD (Fig. 3B). Moreover, single-PFAS exposure-response curves for each analyte showed approximately linear dose-response relationships with CVD (Fig. 4). Posterior inclusion probabilities (PIPs) were obtained from the component-wise variable-selection procedure implemented in BKMR and are shown in Table S10. All eight PFAS exhibited high PIPs (≥ 0.95), with PFDeA, PFHxS, MPAH, PFNA, n-PFOA, and n-PFOS at 1.00, and PFUA (0.9690) and Sm-PFOS (0.9568) also demonstrating strong evidence of contributing to the mixture effect on CMI. BKMR analysis of mixed PFAS exposure and CVD. A The joint effect of the mixture on CVD when all chemicals at a particular percentile are compared to all chemicals at the 50th percentile according to the BKMR model. The estimate h(Z) is considered the relationship between the chemical and the potential continuous outcome. B Effects of individual PFAS, changes in potential continuous CVD outcomes when individual PFAS were at the 75th and 25th percentiles, and when all other PFAS were at the 25th, 50th, or 75th percentiles Dose-response relationships between single PFAS and CVD analyzed by BKMR. Estimation of univariate exposure-response functions between single PFAS exposure and CVD when other PFAS were fixed at the 50th percentile. The x-axis shows log transformed serum PFAS concentrations (original units are ng/mL) WQS regression confirmed a significant inverse association between the PFAS mixture and CVD risk (OR, 0.68, 95% CI, [0.48, 0.97]) (Table S11), with Sm-PFOS contributing the greatest weight to the WQS index (Fig. 5). Q-gcomp likewise indicated that Sm-PFOS had the highest effect weight in the negative mixture-CVD relationship (Fig. S3A), and demonstrated a linear inverse overall association between PFAS mixture exposure and CVD (Fig. S3B). WQS analysis of the relationship between PFAS and CVD. The model was adjusted for age, gender, race, education, family poverty-income ratio, serum cotinine, alcohol, smoking, hypertension, diabetes Lipid profile mediates the association between PFAS and CVD We next conducted causal mediation analyses to quantify the extent to which serum lipids mediate the inverse associations between individual PFAS and CVD. As shown in Fig. 6A, TC significantly mediated the relationships of PFNA, n-PFOA, n-PFOS and Sm-PFOS with CVD. For PFNA, the average causal mediation effect (ACME) was − 0.0020, corresponding to 7.56% mediation, while the average direct effect (ADE) remained − 0.019 (P = 0.008). Similarly, TC mediated 8.57% of the n-PFOA-CVD association (ACME = −0.0030, P = 0.018; ADE = −0.033, P < 0.001), 12.50% of the n-PFOS-CVD association (ACME = −0.0032, P = 0.012; ADE = −0.020, P = 0.036), and 12.35% of the Sm-PFOS-CVD association (ACME = −0.0030, P = 0.008; ADE = −0.022, P = 0.014). Lipid profile mediates the association between selected PFAS and CVD. A TG; B LDL. TG, triglyceride; LDL, low-density lipoprotein; ACME, average causal mediation effects; ADE, average direct effects Figure 6B shows analogous findings for LDL-C. LDL-C mediated 7.77% of the PFNA-CVD effect (ACME = −0.0021, P = 0.024; ADE = −0.019, P = 0.018), 5.87% of the n-PFOA–CVD effect (ACME = −0.0020, P = 0.028; ADE = −0.034, P < 0.001), 12.32% of the n-PFOS-CVD effect (ACME = −0.0031, P = 0.006; ADE = −0.019, P = 0.048), and 12.52% of the Sm-PFOS-CVD effect (ACME = −0.0033, P = 0.006; ADE = −0.022, P = 0.010). These results indicate that perturbations in both TC and LDL-C partially explain the protective associations observed between several PFAS exposures and CVD risk. Sensitivity analysis The results of sensitivity analysis are shown in Table S2-S5 (PFAS continuous variable analysis) and Table S6-S9 (PFAS quartile analysis). The estimated effect directions and significance levels of the two models on various lipid indicators (TC, TG, LDL, HDL) were basically consistent, indicating that the research conclusions have good robustness. Discussion This study integrated traditional single-exposure analyses with advanced mixture approaches, including BKMR, WQS regression, and Q-gcomp, to characterize both individual and joint effects of eight PFAS congeners on CVD risk. Across all statistical frameworks, we observed consistent inverse associations between PFNA, n-PFOA, n-PFOS, Sm-PFOS and CVD. In single-exposure models, each log-unit increase in these four PFAS was linked to 24–40% lower odds of CVD. Mixture analyses reaffirmed these findings: in BKMR, PFAS mixed exposure showed a negative trend with CVD; in WQS and Q-gcomp, the overall PFAS index was significantly inversely associated with CVD, with Sm-PFOS contributing the greatest weight. Causal mediation further revealed that perturbations in total and LDL cholesterol mediated 6–13% of these PFAS-CVD associations. Collectively, our results underscore both individual and joint PFAS exposures as determinants of CVD risk via lipid-mediated pathways. PFAS are chiefly absorbed via ingestion of tainted food and water, inhalation of contaminated dust, and skin contact with everyday products. Their strong carbon-fluorine bonds confer environmental persistence and bioaccumulation, leading to widespread internal exposure in the population [23]. For example, in a cross‑sectional analysis of 2,094 participants, elevated serum PFOA, PFOS, and PFNA levels corresponded with higher odds of self‑reported CVD [24]. Additionally, multivariable models identified positive associations of PFDeA, PFUA, and PFHxS with CVD prevalence [25]. In an Italian study of 59 147 residents of PFAS‑contaminated municipalities, there were 51 621 deaths versus 47 731 expected between 1985 and 2018, with ischemic heart disease mortality significantly elevated compared to the reference population [10]. Case-control research from Hebei, China (355 ACS cases and 355 matched controls) showed that each standard-deviation increase in plasma PFOA and PFOS was associated with 1.51-fold (95% CI, [1.07–2.15]) and 1.77-fold (95% CI, [1.15–2.72]) higher odds of acute coronary syndrome, respectively [12]. In addition, a cohort of 664 Taiwanese adolescents and young adults (aged 12–30 years) with a median serum PFOS concentration of 8.65 ng/mL (range 0.11–85.90 ng/mL) demonstrated a significant increase in carotid intima-media thickness across PFOS quartiles (0.434 mm, 0.446 mm, 0.458 mm, 0.451 mm; P for trend < 0.001) [26–28]. Additionally, analyses of U.S. health professional cohorts (101 CHD cases and 101 matched controls) found that each log-unit increase in plasma PFOS, Br-PFOS, and n-PFOS was associated with significantly higher odds of coronary heart disease: odds ratios were 3.66 (95% CI, [1.36–9.89]), 3.68 (95% CI, [1.55–8.76]), and 3.01 (95% CI, [1.16–7.86]), respectively [29]. Moreover, a global meta-analysis of 29 studies reported that PFOA/PFOS exposure was associated with a 10–20 mg/dL rise in total cholesterol and a roughly 1.1-fold increase in CVD risk [11]. These findings support the hypothesis that PFAS have adverse cardiovascular effects. However, with the phase-out of PFOA and PFOS and the emergence of replacement PFAS (short-chain and novel structures), these associations may be shifting. In contrast, our study observed that certain PFAS may act as protective factors against CVD. A meta-analysis of five Swedish cohort studies found that individuals with higher serum PFOA concentrations had a 20% lower incidence of CVD compared to those with low exposure [30]. In the U.S. C8 Health Project, elevated PFHxS and PFOS levels were linked to reduced stroke risk among diabetic adults, and no PFAS compound was associated with increased stroke incidence regardless of diabetic status [31]. Collectively, these studies converge with our findings, suggesting that under current, lower-level exposure profiles, certain PFAS congeners may be inversely associated with cardiovascular disease endpoints in diverse populations. The mechanisms underlying these paradoxical protective associations remain to be elucidated but may involve PFAS-induced modulation of lipid metabolism or competitive receptor interactions that differ by congener and exposure level. One explanation for our findings lies in changes to PFAS exposure over time. Early NHANES data (2003–2004) showed median serum n-PFOA and n-PFOS concentrations of roughly 3.9 ng/mL and 20.7 ng/mL, respectively. In contrast, during 2015–2020 these values fell to about 1.4 ng/mL for n-PFOA and 3.2 ng/mL for n-PFOS, reflecting the effective removal of long-chain PFAS from production and use [24]. Environmental pollutants may exhibit protective effects at low concentrations; as long-chain PFAS exposure has decreased, their interference with lipid metabolism and vascular inflammation may also be attenuated, potentially placing the net exposure–response relationship within a protective range [32, 33]. Meanwhile, though the medians of some short-chain alternatives (PFNA, PFHxS, Sm-PFOS) remain low, they have risen relative to early 21st-century values, consistent with increased usage. Short-chain PFAS such as PFHxS and PFHpA have significantly lower intrinsic potency toward key molecular targets compared to their long-chain counterparts [34]. Their shorter carbon backbones may also facilitate faster renal clearance, reducing bioaccumulation and limiting peak tissue concentrations that drive atherogenic signaling [35, 36]. As environmental PFAS shift toward these less potent, shorter-chain congeners, epidemiological findings may reflect modest protective effects. Our findings that several PFAS congeners (PFNA, n-PFOA, n-PFOS, Sm-PFOS) are inversely associated with CVD risk, and that TC and LDL-C mediate 6–13% of these associations—raise questions regarding underlying biological mechanisms. One plausible pathway involves activation of PPARα and lipid-regulatory networks. PPARα is a nuclear receptor that regulates fatty acid β-oxidation and lipoprotein synthesis. In vitro studies have shown that PFOA and PFOS strongly activate PPARα target genes, increasing the expression of fatty acid oxidation enzymes and apolipoproteins involved in HDL formation [37–39]. In animal models, PFNA and Sm-PFOS significantly elevate HDL-C levels via PPARα activation and alter very low-density lipoprotein (VLDL) synthesis, consistent with our observed positive β coefficients for HDL-C [40]. Moreover, PPARα activation also upregulates LDL receptor expression, enhancing the clearance of atherogenic LDL particles, which may underlie the inverse PFAS-CVD associations observed [41]. Another potential mechanism is disruption of bile acid synthesis and cholesterol elimination. PFAS can impair hepatic bile acid pathway enzymes, particularly cholesterol 7α-hydroxylase (CYP7A1), reducing conversion of cholesterol to bile acids and fecal excretion [42]. Transcriptomic analyses suggest that PFAS downregulate CYP7A1 and the bile salt export pump, leading to intracellular cholesterol accumulation and compensatory upregulation of cholesterol synthesis [43]. These changes may elevate serum TC and LDL-C, which in turn contribute to CVD risk through lipid-driven atherogenesis. Additionally, due to their amphiphilic nature, PFAS readily bind to apolipoprotein B-containing lipoproteins. Biophysical data show that PFAS partition into the surfaces of LDL and VLDL particles, stabilizing them against receptor-mediated clearance and increasing circulating cholesterol levels, thereby elevating TC and LDL-C [44, 45]. At the level of oxidative stress and endothelial inflammation, PFAS induce reactive oxygen species (ROS) and inflammatory cytokines in vascular cells, further disrupting lipid metabolism. In vitro studies indicate that PFOS exposure enhances NADPH oxidase driven ROS production, impairs nitric oxide signaling, and promotes endothelial dysfunction; chronic low-grade inflammation exacerbates lipid dysregulation [46]. Analyses have linked PFAS exposure to increased IL-6 and TNF-α, which are known to promote hepatic VLDL secretion and alter HDL composition [47]. Recent population studies report that a one-unit increase on the natural log scale in serum PFOA corresponds to a 1.94 mg/dL (95% CI, [1.48–2.41]) increase in total cholesterol, while equivalent ln-increments for PFOS and PFHxS are linked to 4.99 mg/dL (95% CI, [4.12–5.86]) and 2.02 mg/dL (95% CI, [1.45–2.58]) rises, respectively [48]. These results are close to our continuous β estimates. Moreover, data from the Diabetes Prevention Program and its Outcomes Study indicate that doubling serum levels of PFOA, PFHxS, and PFNA is associated with higher baseline total cholesterol concentrations [49]. These findings align with our results and help elucidate the mechanisms underlying the observed associations. Our study presents several novel contributions. First, we leveraged the two most recent NHANES cycles (2015–2016 and 2017–2020) to capture the evolving landscape of the phasedown of long-chain PFAS. This enabled us to examine PFAS-related associations in a population whose internal PFAS burden has notably shifted toward lower long-chain and emerging short-chain congeners. Second, beyond traditional single-pollutant regression, we concurrently applied BKMR, WQS regression, and Q-gcomp. All three mixture modeling approaches converged on a consistent inverse PFAS-CVD relationship. BKMR identified uniformly high PIPs for all eight PFAS, highlighting PFDeA, PFHxS, Me-PFOSA-AcOH, PFNA, n-PFOA, and n-PFOS (PIP = 1.00) as primary contributors, with PFUA (0.969) and Sm-PFOS (0.957) also important. WQS regression yielded an OR of 0.68 (95% CI, [0.48–0.97]) for the PFAS index, with Sm-PFOS carrying the greatest weight. Quantile g-computation similarly assigned Sm-PFOS the largest negative weight and demonstrated a linear inverse mixture-CVD association. This triangulated, cross-method mixture analysis approach strengthens causal inference regarding the relationship between real-world PFAS mixtures and CVD risk. Third, we identify PFNA, n-PFOA, n-PFOS, and Sm-PFOS as the primary drivers of both lipid perturbations and reduced odds of CVD. This congener-level insight informs which PFAS substitutions may pose relatively lower cardiometabolic risk. Finally, for the first time in a U.S. sample, we quantify the extent to which changes in lipid profiles explain the inverse association between PFAS and CVD. We demonstrate that disruptions in total and LDL cholesterol mediate 6–13% of the PFAS effects on CVD, establishing a direct link between exposure, a known risk factor, and disease. This study also has several limitations. First, in terms of study design and causal inference, NHANES is a cross-sectional survey in which exposure and outcome are measured simultaneously, making it impossible to determine temporal sequence and thus limiting causal interpretation. Prospective cohort studies or clinical trials are needed to more accurately estimate causal relationships. Causal mediation analysis relies on assumptions such as no interference and no unmeasured confounding, which are difficult to fully satisfy in cross-sectional data and may lead to biased estimates of mediation effects. By emphasizing robust covariate adjustment and complementary mixture and mediation models, we prioritized uncovering plausible mechanistic relationships that can guide future experimental and longitudinal research. Furthermore, our mediation findings should be interpreted with caution. In cross-sectional data, mediation models cannot confirm temporal order or exclude unmeasured confounding of the exposure-mediator-outcome triad. Therefore, these results generate hypotheses about biological pathways that need to be tested in prospective or experimental designs. Second, CVD status was determined by self-report of previous physician diagnoses, which may be subject to recall bias or misclassification, especially in populations with limited healthcare resources or low health literacy, which may attenuate or inflate observed associations. Misclassification may not be differential, biasing results toward the null hypothesis, or it may be differential if certain subgroups systematically under- or over-report disease. Future studies that incorporate adjudicated medical records or biomarker-based endpoints may help validate self-report results and reduce measurement error. Third, regarding exposure assessment, PFAS concentrations were measured at a single time point, which may not accurately reflect long-term exposure dynamics. Serum levels are also susceptible to recent intake and may introduce measurement error. Although we used multiple model, including BKMR, WQS, and Q-gcomp, to assess mixture effects, differences in how these models handle high-dimensional interactions and weight assignments could affect the consistency of the results. Fourth, in terms of covariate control and confounding, while we adjusted for demographic, lifestyle, and clinical factors, there may still be unmeasured or poorly measured confounders (such as occupational exposure, genetic susceptibility) that could bias the effect estimates. Fifth, although NHANES primarily reflects the U.S. adult population. The generalizability of our findings to other countries or specific high-risk groups remains to be validated. Moreover, our study includes data from the 2015–2020 cycles and does not cover more recent or longer-term exposure trends, which warrant continued monitoring in future research. Sixth, the selection of covariates resulted in some sample losses, which may have introduced selection bias. Specifically, the missing rates of the household income ratio and drinking variables were high, thus significantly increasing the proportion of participants excluded in the complete case analysis. This may have systematically removed individuals with low income or certain behavioral characteristics, which is an important limitation of this study. Finally, our analysis did not incorporate NHANES survey weights, which may limit the direct generalizability of our effect estimates to the entire U.S. population. Survey weighting is essential when the goal is to produce prevalence or mean estimates, but it is less critical for etiologic investigations focused on internal validity [50]. Here, our primary objective was to elucidate the biological pathways, specifically lipid-mediated mechanisms, through which PFAS may influence cardiometabolic risk, rather than to quantify population-level effect sizes. Conclusions In this analysis of 2015–2020 NHANES data, higher serum levels of PFNA, n-PFOA, n-PFOS, and Sm-PFOS were each robustly associated with lower odds of prevalent CVD. In addition, specific PFAS also had significant associations with lipid profiles. Advanced mixture models confirmed an overall inverse PFAS-CVD trend driven primarily by Sm-PFOS, and causal mediation indicated that perturbations in total and LDL cholesterol explain a meaningful fraction of these associations. These findings suggest that contemporary PFAS exposure profiles may confer differential cardiovascular effects, in part via lipid-mediated pathways, and underscore the need for ongoing monitoring and mechanistic research to inform risk assessment and regulatory decision-making. Supplementary Information Supplementary material 1. Acknowledgements Not applicable. Authors’ contributions Xianghai Wang analyzed the data, wrote the main manuscript, and was involved in funding acquisition. Yongping Cao, Xuli Chen, Fei Ma, Guojin Jian, and Hengyang Zhang reviewed the manuscript and provided critical comments. Qiuyu Wang and Wenwen Xiao conceptualized the project, provided direct oversight, wrote and reviewed the manuscript. All the authors read and approved the final manuscript. Funding This study was supported by Anhui Provincial University Scientific Research Project (Natural Science) Key Project [No. 2024AH051971]. Data availability The datasets generated and/or analysed during the current study are available in the NHANES repository, [https://www.cdc.gov/nchs/nhanes/index.html](https:/www.cdc.gov/nchs/nhanes/index.html). Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Qiuyu Wang, Email: Qiuyuwang2025@163.com. Wenwen Xiao, Email: wenwenxiao1996@163.com. References Associated Data Supplementary Materials Supplementary material 1. Data Availability Statement The datasets generated and/or analysed during the current study are available in the NHANES repository, [https://www.cdc.gov/nchs/nhanes/index.html](https:/www.cdc.gov/nchs/nhanes/index.html).