Associations between per- and polyfluoroalkyl substances, liver function, and daily alcohol consumption in a sample of U.S. adults

Abstract

Background and Aim:

Per- and polyfluoroalkyl substances (PFAS) are ubiquitous in the environment and in the serum of the U.S. population. We sought to evaluate the association of PFAS independently and jointly with alcohol intake on liver function biomarkers in a sample of the U.S. general population.

Methods:

Using data from the National Health and Nutrition Examination Survey (2003–2016; N=11,794), we examined the five most historically prevalent PFAS with >75% detection rates. We estimated odds ratios (OR) and 95% confidence intervals (CI) for the association between PFAS (quartiles and log-transformed continuous, ng/mL) and high levels (>95^th percentile) of liver injury biomarkers using logistic regression models adjusted for key confounders. We evaluated interactions between PFAS and alcohol consumption and sex via stratified analyses and conducted sub-analyses adjusting for daily alcohol intake among those with available drinking history (N=10,316).

Result:

Serum perfluorooctanoic acid (PFOA) was positively associated with high levels of alanine transferase (ALT) without monotonic trend (OR_Q4vsQ1=1.45, CI: 0.99–2.12; p-trend=0.18), and with increased aspartate transaminase when modeled continuously (OR_con=1.15, CI: 1.02–1.30; p-trend=0.03). Perfluorooctane sulfonate (PFOS) and perfluorohexane sulfonate (PFHxS) were both inversely associated with alkaline phosphatase while the trend was statistically significant only for PFHxS (p=0.02). A non-monotonic inverse association was observed with PFOA (p-trend=0.10). The highest quartile of PFOS was associated with high total bilirubin (TB) (OR_Q4vsQ1=1.57, CI: 1.01–2.43, p-trend=0.02). No significant associations were found between any PFAS and γ-glutamyl transpeptidase. We found no associations for perfluorodecanoic acid and perfluorononanoic acid. We observed some suggestive interactions with alcohol intake, particularly among heavy drinkers.

Conclusion:

Consistent with other studies, serum levels of PFOA, PFHxS and PFNA were positively associated with high levels of ALT, and we also observed weak positive associations between selected PFAS and TB. Positive associations observed among heavy drinkers warrant additional evaluation.

Introduction

Per- and polyfluoroalkyl substances (PFAS) are a family of compounds widely used in industry and commercial products since their introduction in the 1940s (Hekster et al., 2003; Liew et al., 2018; Sunderland et al., 2019). PFAS have many favorable physical and chemical properties for manufacturing including waterproofness and lipophobicity, and the strong bond between the carbon and fluorine atom results in their physical and chemical stability and persistence in both biota and the environment (Brendel et al., 2018; Fromme et al., 2009). PFAS are detected worldwide in aquatic environments (Hu et al., 2016), soil (Dalahmeh et al., 2018), agricultural crops and food (Ao et al., 2019; Gebbink et al., 2015; Ghisi et al., 2019; Papadopoulou et al., 2019) and indoor house dust (Ao et al., 2019; Winkens et al., 2018), suggesting that humans can be exposed through multiple pathways. More than 10 specific PFAS have been identified in human biospecimens (Jian et al., 2018), and gender differences in serum concentrations have been observed (Ericson et al., 2007; Jain and Ducatman, 2019a; Jian et al., 2018; Liu et al., 2021).

Exposure to PFAS is related to many adverse health outcomes such as thyroid diseases, obesity, diabetes, hyperuricemia, and hypertension; and cancer (Costello et al., 2022; Li et al., 2022; Liu et al., 2021; Shearer et al., 2021; Steenland and Winquist, 2021). A recent systematic review of cross-sectional studies found that several PFAS were associated with increased levels of liver injury markers including alanine aminotransferase (ALT), aspartate aminotransferase (AST) and gamma-glutamyl transferase (GGT) (Costello et al., 2022). Alkaline phosphatase (ALP) and total bilirubin (TB) are also indicators of liver damage and are often used to test liver function and to diagnose liver diseases in clinical settings (Pratt and Kaplan, 2000). Studies have suggested that PFAS are hepatotoxic through various mechanisms. They are suspected to disrupt the endocrine system, induce inflammation, and increase oxidative stress (Fenton et al., 2021; Lilienthal et al., 2017; Sunderland et al., 2019). The most consistent evidence of endocrine disruption from PFAS exposure is in the area of lipid and glucose metabolism (Sunderland et al., 2019). The liver is one of the main sites of PFAS accumulation (Pérez et al., 2013). Studies in mouse models have shown that PFAS can increase liver weight and induce steatosis by activating genes involved in fatty acid metabolism, including peroxisome proliferator-activated receptor (Ppar)-α and cluster of differentiation 36 (Cd36) (Das et al., 2017; Wu et al., 2017; Zhang et al., 2018). An analysis from the C8 Health Study in the Mid-Ohio Valley demonstrated a positive relationship between serum perfluoroalkyl acids (PFAAs) and cytokeratin 18 M30 (CK18 M30), a marker of hepatocyte apoptosis (Bassler et al., 2019). A study found different distributions of several PFAS between tumor liver and non-tumor liver samples in humans and suggested that these differences might be due to the oxidative stress associated with the antioxidant capacity in the liver (Liu et al., 2021). An analysis of U.S. National Health and Nutrition Examination Survey (NHANES) data found that certain PFAS are positively associated with increased levels of liver function biomarkers among obese patients compared with non-obese patients (Jain and Ducatman, 2019b). Collectively, the current evidence, predominantly derived from rodent studies, suggest that PFAS disrupt normal liver function, alter liver histology and morphology, and result in liver damage. Thus, the translation of these findings to humans is uncertain.

Alcohol consumption is a prevalent lifestyle factor around the world. More than 85.6% adults in the U.S. reported that they drank alcohol at some point in their lifetime (SAMHSA, 2020). Epidemiologic and toxicologic studies have suggested that alcohol consumption causes liver damage and leads to liver diseases through multiple pathways (Chen et al., 2020; Chu et al., 2021; Long et al., 2020; Roerecke et al., 2019). Animal studies have shown that PFAS can interact with alcohol, jointly inducing and exacerbating fatty liver disease (Gripshover, 2021; Gripshover et al., 2022), but epidemiologic data on their potential joint effects are lacking.

In this study, we aimed to investigate the association between PFAS exposure and liver function in a representative sample of the U.S. population. We also evaluated whether self-reported alcohol consumption modified the effect of PFAS on liver function.

Methods

Study design and population

NHANES is an ongoing nationwide survey of the U.S. non-institutionalized population. Details of the study population and survey design have been described elsewhere (CDC, 2021). Briefly, the survey applied a stratified, multistage probability sampling design to collect health and nutrition information in the U.S general population. Participants underwent a physical examination and laboratory tests in a mobile examination center (MEC). Demographic, socioeconomic, dietary, and health-related questions were collected in the MEC or in the participants’ home by computer-assisted questionnaires or trained interviewers.

NHANES began monitoring PFAS in the U.S. population in 1999; serum PFAS were measured in a random, one-third subsample of participants ≥ 12 years of age. Data from 1999–2000 were measured in surplus serum samples and used different extraction and preconcentration methods than those in later cycles (Calafat et al., 2007), and the 2001–2002 cycle only had summed data for PFAS rather than individual PFAS. Therefore, we combined data from seven other NHANES cycles with available individual PFAS measurements (2003–2016) for the current study (2003–04 N= 2368; 2005–06 N= 2323; 2007–08 N= 2294; 2009–20 N= 2432; 2011–12 N =2177; 2013–14 N= 2339; 2015–16 N= 2170). After combining cycles, a total of 16,103 participants had available PFAS measurements. We excluded individuals for analysis who were less than 18 years old at the time of participation (N=2710) or were missing any data for PFAS (N=1494) or liver function biomarkers (N=1219). Participants with a positive serum hepatitis B surface antigen or positive serum hepatitis C antibody (N=211) were also excluded because viral pathogens may also affect liver function (Figure S1). The final sample eligible for analysis included 11,794 participants. The NHANES study protocol was approved by the ethical review committee of the National Center for Health Statistics and all adult participants gave written informed consent.

Exposure assessment: serum PFAS

PFAS were measured in serum using an online solid phase extraction coupled to high performance liquid chromatography-turbo ion spray ionization-tandem mass spectrometry (online SPE-HPLC-TIS-MS/MS) (Kuklenyik et al., 2005).

We examined the five most prevalent PFAS with a detection rate of approximately 75% or greater in the samples: perfluorodecanoic acid (PFDA), perfluorohexane sulfonic acid (PFHxS), perfluorononanoic acid (PFNA), perfluorooctanoic acid (PFOA) and perfluorooctane sulfonic acid (PFOS). Starting in 2013, NHANES reported the structural isomers of PFOA and PFOS. Concentrations of linear PFOA (n-PFOA), branched isomers of PFOA (Sb-PFOA, branched PFOA isomers), linear PFOS (n-PFOS), and perfluoromethylheptane sulfonate isomers (Sm-PFOS, monomethyl branched PFOS isomers) were summed to generate the sum of PFOA and PFOS, respectively in the 2013–2014 and 2015–2016 cycles. The results below the detection limit in the original data were imputed with a value equal to the limit of detection divided by the square root of 2.

Outcome assessment

Serum ALT, AST, GGT, ALP and TB were measured by NHANES Collaborative Laboratory Services according to a standardized NHANES protocol. Briefly, serum liver biomarkers were measured either on a Beckman Synchron LX20 (2003–2006) using an enzymatic rate method, or on a Beckman UniCel DxC800 Synchron (2007–2016) using a kinetic rate method.

Confounder evaluation

Covariates were selected for consideration based on known or suspected relationships with either PFAS or liver function, and included age (continuous and squared), sex (male or female), race/ethnicity (Non-Hispanic white, non-Hispanic black, Mexican American, or other race/ethnicities), poverty-income-ratio (PIR, based on family income, family size and annual change of living cost, <1 indicating below the federal established poverty threshold), education level (less than high school, high school or equivalent, some college or more), body mass index (BMI), and smoking history (non-smoker, former or current smokers). We also included survey cycle to account for temporal variation because of the known decline in serum PFOA and PFOS in the U.S. population over the past 20 years (Dong et al., 2019; Kato et al., 2011). As a sensitivity analysis, we separately evaluated the influence of adjustment for coffee intake, given it has been shown to influence liver biomarkers (ALP in particular) and is a potential source of PFAS exposure.

Approximately 87.5% (N=10,316) of participants with PFAS measurements also had an available self-reported history of alcohol intake, which consisted of drinking status (non-, former, or current drinkers) and daily alcohol consumption (drinks/day for former/current drinkers). According to the National Institute on Alcohol Abuse and Alcoholism (NIAAA), moderate drinking is defined as up to 1 drink per day for women and up to 2 drinks per day for men. NIAAA also defines heavy drinking as more than 3 drinks on any day for women and more than 4 drinks for men (NIAAA, 2021). We categorized daily alcohol consumption into 5 groups for analysis to capture a range of intake that covered none, light, moderate, and heavy alcohol intakes [missing, 0 (non-drinkers), 1–2 drinks/day, 2–4 drinks/day, >4 drinks/day].

Statistical analysis

We performed our analyses in SAS (v. 9.4) and used SAS survey procedures to account for the complex NHANES survey design and sampling weights. In descriptive analyses, we evaluated PFAS distributions by participant demographic and lifestyle characteristics. The SAS macro %SurveyCorrCov was applied to test the rank-based correlation between each individual PFAS.

We estimated the association between levels of PFAS and liver biomarkers in weighted linear models (i.e., using PROC SURVEYREG). We initially evaluated the relationship between each PFAS and liver biomarker with a basic model adjusted for age, age², sex and race/ethnicity (Model 1), and a full model further adjusted for BMI, PIR, education level, smoking history, and survey cycle (Model 2). Serum PFAS concentration (continuous) was log₂-tranformed to improve model fit. The beta coefficients were estimated with 95% confidence intervals (95% CI). In logistic models similarly accounting for sample weights, we estimated the odds ratio (OR) and 95% CI for the association between quartiles of PFAS and a dichotomized outcome variable for liver function biomarkers (high vs. low). These upper normal limits (i.e., high) were defined as the sex-specific 95^th percentiles of the concentration of biomarkers, as has been done in prior studies as an indicator of potential liver damage (Kolahdoozan et al., 2020; Prati et al., 2002; Valenti et al., 2021). We also included log₂-transformed PFAS as continuous terms in the logistic models. Linear trend tests were applied to a continuous variable derived from the within-category median of each PFAS quartile. For analyses stratified by time period, we also categorized the survey cycles into three groups based on their corresponding median PFAS concentrations, which declined over time: 2003–2008, 2009–2012, and 2013–2016. We stratified linear models by a binary BMI variable (BMI: <30 kg/m², ≥30, i.e., obese or not) to assess the potential modifying effect of obesity on overall associations. We employed a restricted cubic spline analysis to evaluate potential non-linear relationships. In sensitivity analyses, we also additionally mutually adjusted the final models of main effects for the most strongly correlated (Spearman’s ρ> 0.70) PFAS: PFNA and PFDA (ρ=0.77), and PFOA and PFOS (ρ=0.72). We also further excluded n=175 participants who reported a history of liver conditions in the 2003–2010 cycles and who also reported still having those conditions in the 2011–2016 cycles, and those who reported a liver cancer diagnosis at the time when the survey was conducted.

To evaluate potential interactions between PFAS and daily alcohol consumption on liver biomarkers, we conducted analyses among participants stratified based on their daily drinking habits. We also included daily alcohol consumption in the logistic models as interaction terms and evaluated statistical significance (p-interaction) with likelihood ratio tests comparing models with and without interaction terms. We also examine its role as a confounder by adjusting for both categorized and continuous daily alcohol consumption. We used two-tailed p-values at an α=0.05 statistical significance level for all analyses.

Results

All PFAS included in our analysis were detected in over 98% of participants except for PFDA (74.1%; Table S1). PFOS had the highest median concentration (10.09 ng/mL) while concentrations of PFDA were the lowest (0.20 ng/mL). The five PFAS were weakly to moderately correlated with the highest correlation found between PFNA and PFDA (ρ=0.77) and the lowest between PFHxS and PFDA (ρ=0.28; Table S2).

Serum levels of PFOA (median=4.13 [2003–2004] and 1.65 [2015–2016], ng/mL) and PFOS (median=21.17 [2003–2004] and 5.21 [2015–2016]) declined in the NHANES population over time, while PFHxS, PFNA and PFDA remained relatively consistent. Among all participants, higher PFAS levels were found in males (median=3.38 [PFOA], 12.42 [PFOS], 2.00 [PFHxS], 1.00 [PFNA], ng/mL) and older individuals (median=3.19 [PFOA], 14.24 [PFOS], 2.00 [PFHxS], 0.99 [PFNA], ng/mL), with the exception of PFDA. Non-Hispanic white participants had the highest median levels of PFOA, PFHxS, and PFNA, while non-Hispanic black participants had the highest median level of PFOS. Median levels of PFDA were comparable among different race/ethnicity groups.

In both unadjusted and adjusted linear models, TB had weak but positive associations with all five PFAS (Table 2). In Model 2, a doubling in serum levels of PFOA, PFHxS and PFNA were positively associated with a 0.87 (0.49–1.25), 0.62 (0.33–0.92) and 0.66 (0.25–1.08) increase in ALT, respectively. For every twofold rise in PFOA levels, there was a positive increase in AST (β=0.65, 95% CI=0.23–1.08). Similarly, for each twofold elevation in PFHxS levels, a 0.32 (0.05–0.59) increase in AST was observed. PFOS, PFHxS and PFDA were negatively associated with ALP (β=−0.77 [−1.44, −0.09], 0.47 [−0.86, −0.08], −0.92 [−1.53, −0.32], respectively). No clear association was found between any PFAS and GGT after adjustment for covariates. In the analysis stratified by BMI categories, we found that the aforementioned positive associations were generally slightly stronger in the obese group compared with the non-obese group (data not shown). In models by time period, linear associations between ALT and PFAS were somewhat stronger and all PFAS showed associations in the earliest cycles (Table S3). Likewise, associations in logistic models were similar or slightly stronger in earlier cycles, although more imprecise (data not shown).

Table 2.

Linear associations between log-transformed serum PFAS and liver function biomarkers, NHANES 2003–2016 ^a,b

		Unadjusted				Model 1 c				Model 2 d
		β	95% CI		P	β	95% CI		P	β	95% CI		P
ALT	PFOA	1.09	0.74	1.43	<0.0001	0.70	0.36	1.04	<0.0001	0.87	0.49	1.25	<0.0001
	PFOS	0.49	0.21	0.78	<0.0001	0.19	−0.09	0.46	0.18	0.23	−0.11	0.57	0.19
	PFHxS	0.92	0.64	1.19	<0.0001	0.50	0.21	0.79	0.0008	0.62	0.33	0.92	<0.0001
	PFNA	0.82	0.39	1.25	<0.0001	0.53	0.14	0.93	0.01	0.66	0.25	1.08	0.002
	PFDA	0.09	−0.26	0.45	<0.0001	0.00	0.34	0.34	0.99	0.19	−0.15	0.53	0.27
AST	PFOA	0.71	0.39	1.02	<0.0001	0.51	0.16	0.86	0.005	0.65	0.23	1.08	0.003
	PFOS	0.17	−0.05	0.40	0.13	−0.07	−0.31	0.16	0.54	−0.13	−0.44	0.17	0.38
	PFHxS	0.51	0.28	0.75	<0.0001	0.28	0.02	0.53	0.03	0.32	0.05	0.59	0.02
	PFNA	0.35	−0.002	0.71	0.05	0.18	−0.19	0.55	0.005	0.22	−0.19	0.63	0.29
	PFDA	−0.02	−0.33	0.28	0.89	−0.11	−0.42	0.21	0.50	−0.08	−0.42	0.26	0.63
ALP	PFOA	−0.02	−0.60	0.56	0.96	−0.04	−0.70	0.62	0.91	−0.52	−1.22	0.18	0.15
	PFOS	0.19	−0.29	0.67	0.43	−0.17	−0.73	0.38	0.53	−0.77	−1.44	−0.09	0.03
	PFHxS	−0.17	−0.55	0.20	0.37	−0.43	−0.84	−0.01	0.04	−0.47	−0.86	−0.08	0.02
	PFNA	−0.09	−0.77	0.59	0.79	−0.31	−1.03	0.41	0.39	−0.40	−1.12	0.33	0.28
	PFDA	−0.73	−1.34	−0.12	0.02	−0.92	−1.55	−0.30	0.004	−0.92	−1.53	−0.32	0.003
GGT	PFOA	1.81	0.70	2.92	0.002	1.36	0.14	2.58	0.03	1.63	−0.06	3.32	0.06
	PFOS	0.23	−0.61	1.07	0.59	−0.60	−1.44	0.23	0.16	−0.99	−2.23	0.24	0.11
	PFHxS	0.27	−0.41	0.95	0.43	−0.41	−1.25	0.44	0.35	−0.33	−1.29	0.63	0.49
	PFNA	1.23	0.05	2.41	0.04	0.54	−0.75	1.82	0.41	0.72	−0.80	2.24	0.35
	PFDA	0.11	−0.82	1.03	0.82	−0.34	−1.29	0.61	0.48	−0.08	−1.17	1.01	0.88
TB	PFOA	0.06	0.05	0.06	<0.0001	0.04	0.03	0.05	<0.0001	0.016	0.008	0.024	0.0001
	PFOS	0.04	0.04	0.05	<0.0001	0.03	0.03	0.04	<0.0001	0.006	0.000	0.013	0.05
	PFHxS	0.04	0.03	0.05	<0.0001	0.02	0.02	0.03	<0.0001	0.014	0.008	0.021	<0.0001
	PFNA	0.04	0.03	0.05	<0.0001	0.03	0.03	0.04	<0.0001	0.014	0.007	0.022	0.0003
	PFDA	0.04	0.03	0.04	<0.0001	0.03	0.02	0.04	<0.0001	0.011	0.003	0.018	0.007

^a:Perfluorooctanoic acid (PFOA), perfluorooctane sulfonic acid (PFOS), perfluorohexanesulfonic acid (PFHxS), perfluorononanoic acid (PFNA), perfluorodecanoic acid (PFDA)

^b:Alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), gamma-glutamyl transferase (GGT), total bilirubin (TB)

^c:Adjusted for age, age², gender and race/ethnicity

^d:Adjusted for age, age², gender, race/ethnicity, body mass index, poverty income ratio, education level, smoking status and survey cycle

In the fully adjusted logistic regression models, the highest quartile of serum PFOA showed an association with a high level of ALT; however, the trend across quartiles was not monotonic (OR_Q4vsQ1=1.45, 95% CI: 0.99–2.12; p-trend=0.18; Table 3). Continuous PFOA was significantly associated with a high level of AST (OR_con= 1.15, 95% CI: 1.02–1.30, p=0.03), but quartile associations were not significantly increased. Similarly, continuous PFHxS was associated with high ALT (OR_con=1.09, 95% CI: 1.00–1.19, p-trend=0.06) but not in quartiles. We observed no significant associations between any PFAS and high levels of GGT. The highest quartile of PFOS was associated with high levels of TB with a significant trend (OR_Q4vsQ1=1.57, 95% CI: 1.01–2.43, p-trend=0.02). A positive association was found between PFHxS and high TB levels across quartiles (OR_Q2vsQ1=1.23, 95% CI: 0.87–1.74, OR_Q3vsQ1=1.48, 95% CI: 1.08–2.02, OR_Q4vsQ1=1.37, 95% CI: 0.99–1.90; p-trend=0.09), and the association was similarly positive in the continuous model (OR_con=1.14, 95% CI: 1.05–1.25). There was also a linear trend in the associations between PFNA quartiles and high TB levels and the relationship was positive in the continuous exposure model (OR_con =1.12, 95% CI: 0.99–1.27; p-trend=0.05). We found inverse associations between PFOS quartiles and high levels of ALP (p-trend=0.08) and when PFOS was modeled continuously (OR_con=0.83, 95% CI: 0.74–0.92, p<0.001). PFDA was also inversely associated with ALP in each quartile and when modeled continuously (OR_Q2vsQ1=0.73, 95% CI: 0.55–0.97, OR_Q3vsQ1=0.53, 95% CI: 0.37–0.77, OR_Q4vsQ1=0.72, 95% CI: 0.53–0.98; p-trend=0.07; OR_con=0.89, 95% CI: 0.79–1.00, p=0.05). Similarly, PFHxS was associated with low ALP both in quartiles (p-trend=0.02) and continuously (p<0.0001). The findings obtained from the spline analysis demonstrated consistency with the main results obtained from both linear and logistic models (data not shown). After additionally adjusting both linear and logistic models for daily alcohol consumption, the effect estimates did not materially change (<10%; data not shown). When mutually adjusting linear and logistic main effects models for correlated PFAS, we generally found only modest changes in effect sizes and wider confidence intervals (data not shown).

Table 3.

Odds ratios and 95% confidence intervals for associations between PFAS and high (>95^th percentile) levels of liver function biomarkers, NHANES 2003–2016 ^a,b,c

	ALT				AST				ALP				GGT				TB
	N	OR	95%CI		N	OR	95%CI		N	OR	95%CI		N	OR	95%CI		N	OR	95%CI
PFOA
≤ 1.79	159	Ref.			172				224				168				120
1.80 – 2.89	152	1.50	1.05	2.15	145	1.08	0.76	1.52	155	0.66	0.50	0.88	158	1.17	0.84	1.64	104	0.77	0.54	1.09
2.80 – 4.66	133	1.09	0.76	1.57	130	1.02	0.74	1.41	171	0.75	0.56	1.02	153	1.18	0.81	1.71	150	1.06	0.75	1.48
>4.66	129	1.45	0.99	2.12	137	1.28	0.92	1.79	138	0.67	0.48	0.93	153	1.29	0.88	1.89	158	1.17	0.81	1.70
p-trend d		0.18				0.13				0.10				0.24				0.13
continuous e	573	1.17	1.03	1.32	584	1.15	1.02	1.30	688	0.83	0.75	0.93	632	1.08	0.96	1.21	532	1.09	0.96	1.25
PFOS
≤ 5.30	177	Ref.			170				190				160				107
5.31 – 10.09	150	0.97	0.68	1.38	146	0.90	0.64	1.27	149	0.60	0.44	0.82	159	1.07	0.74	1.56	113	0.99	0.67	1.46
10.10 – 18.40	141	0.77	0.52	1.14	140	0.81	0.59	1.10	172	0.65	0.46	0.93	177	1.03	0.77	1.37	148	1.29	0.89	1.86
>18.40	105	0.80	0.52	1.22	128	0.73	0.50	1.07	177	0.59	0.39	0.87	136	0.75	0.51	1.10	164	1.57	1.01	2.43
p-trend d		0.24				0.11				0.08				0.04				0.02
continuous e	573	0.98	0.88	1.09	584	0.95	0.86	1.05	688	0.83	0.74	0.92	632	0.94	0.86	1.02	532	1.14	1.01	1.28
PFHxS
≤ 0.90	176	Ref.			169				217				181				120
0.91 – 1.59	134	1.06	0.75	1.49	139	1.00	0.72	1.37	153	0.86	0.63	1.16	145	1.10	0.81	1.50	121	1.23	0.87	1.74
1.60 – 2.79	133	1.00	0.67	1.47	146	0.88	0.63	1.23	168	0.67	0.49	0.92	169	0.97	0.69	1.38	152	1.48	1.08	2.02
>2.80	130	1.16	0.81	1.64	130	0.96	0.68	1.37	150	0.68	0.49	0.93	137	0.95	0.67	1.33	139	1.37	0.99	1.90
p-trend d		0.41				0.83				0.02				0.56				0.09
continuous e	573	1.09	1.00	1.19	584	1.00	0.91	1.09	688	0.86	0.80	0.93	632	0.98	0.90	1.07	532	1.14	1.05	1.25
PFNA
≤ 0.60	157	Ref.			175				150				166				114
0.61 – 0.92	135	0.86	0.63	1.16	122	0.71	0.49	1.01	150	0.83	0.60	1.15	126	0.81	0.55	1.20	117	1.07	0.76	1.50
0.93 – 1.47	138	0.95	0.72	1.25	122	0.76	0.55	1.04	155	0.85	0.63	1.14	155	1.01	0.74	1.39	142	1.13	0.82	1.57
>1.48	143	1.14	0.84	1.56	165	1.05	0.77	1.45	173	0.83	0.61	1.13	185	1.07	0.77	1.49	159	1.40	0.96	2.02
p-trend d		0.22				0.32				0.34				0.36				0.05
continuous e	573	1.08	0.95	1.24	584	1.02	0.90	1.16	688	0.92	0.81	1.03	632	1.04	0.93	1.15	532	1.12	0.99	1.27
PFDA
≤0.14	171	Ref.			176				191				191				122
0.15 – 0.20	152	0.84	0.62	1.12	138	0.76	0.56	1.03	191	0.73	0.55	0.97	157	0.80	0.58	1.09	130	0.86	0.61	1.19
0.21 – 0.39	107	1.00	0.73	1.38	96	0.81	0.57	1.14	94	0.53	0.37	0.77	93	0.62	0.46	0.84	98	1.07	0.73	1.58
>0.39	143	0.91	0.67	1.23	174	0.91	0.66	1.24	184	0.72	0.53	0.98	191	0.85	0.62	1.16	182	1.20	0.86	1.67
p-trend d		0.80				0.85				0.07				0.40				0.10
continuous e	573	1.02	0.91	1.14	584	0.99	0.88	1.10	688	0.89	0.79	1.00	632	0.97	0.87	1.07	532	1.09	0.96	1.24

^a:Perfluorooctanoic acid (PFOA), perfluorooctane sulfonic acid (PFOS), perfluorohexanesulfonic acid (PFHxS), perfluorononanoic acid (PFNA), perfluorodecanoic acid (PFDA)

^b:Alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), gamma-glutamyl transferase (GGT), total bilirubin (TB)

^c:Adjusted for age, age2, gender, race/ethnicity, body mass index, poverty income ratio, education level, smoking status and survey cycle

^d:Linear trend tests were conducted on a continuous variable obtained from the median of each PFAS within each quartile.

^e:Continuous models reflect association per 1-unit increase in serum PFAS on the log base 2 scale, corresponding to an approximate doubling in analyte levels.

We observed positive associations between continuous alcohol intake (drinks per day) with ALT (β=0.47 [0.20–0.73]), AST (β=0.30 [0.15–0.45]), and GGT (β=1.08 [0.62–1.54]), as has been reported in the full NHANES population (Tsai et al., 2012a; Tsai et al., 2012b). In analyses stratified by alcohol intake, we found that mean serum PFOA and PFHxS levels significantly increased with increasing consumption of alcohol (p<0.05; Table S4). In contrast, the highest levels of PFOS were observed among the light and moderate drinkers (p<0.05). We did not observe a trend in PFNA or PFDA levels by alcohol consumption. Each twofold increase in serum PFOA and PFHxS was positively associated with continuous ALT only in light drinkers (β=1.05 [0.49, 1.62], 0.77 [0.30, 1.25], respectively; Table 4) while the weak positive linear association between serum PFOA, PFHxS and PFNA and continuous TB remained only in heavy drinkers (β=0.033 [0.010, 0.057], 0.039 [0.011, 0.067], and 0.037 [0.013, 0.060], respectively). PFOS, PFNA and PFDA were negatively associated with ALP only in non-drinkers (β=−1.21 [−2.32, −0.09], −1.35 [−2.65, −0.05], −1.53 [−2.59, −0.47], respectively). In logistic models, high GGT was positively associated with PFOA among heavy drinkers (OR_Q4vsQ1=3.08, 95% CI=1.43–6.62; p-trend=0.39; Table S5) and negatively associated with PFDA across quartiles in the same group (p-trend<0.05). We found inverse associations between several PFAS and high levels of ALP among non-, light- and moderate drinkers compared to generally positive associations in the heavy drinkers, although none of these relationships were statistically significant except in continuous models (Tables S6). There were few notable patterns of association between PFAS and high ALT (Table S7), AST (Table S8), or TB (Table S9) levels by alcohol intake. Additional adjustment for coffee intake had minimal influence on any of our observed associations (<5% change in effect estimates; data not shown). Similarly, excluding the small proportion (1.5%) of participants reporting liver conditions had little impact on our results (data not shown).

Table 4.

Linear associations between log-transformed serum PFAS and liver function biomarkers, stratified by alcohol consumption, NHANES 2003–2016 ^a,b,c

		ALT			AST			ALP			GGT			TB
		β	95% CI		β	95% CI		β	95% CI		β	95% CI		β	95% CI
PFOA	Missing	0.93	0.45	1.41	0.71	0.32	1.09	−0.83	−2.43	0.76	−0.23	−1.42	0.96	0.017	0.002	0.031
	Non-drinker	0.41	−0.38	1.21	0.39	−0.11	0.90	−1.07	−2.55	0.41	0.68	−1.62	2.98	0.020	0.006	0.035
	Light	1.05	0.49	1.62	0.85	0.00	1.70	0.21	−0.85	1.28	2.62	−1.38	6.62	0.009	−0.005	0.022
	Moderate	0.92	−0.50	2.34	0.31	−1.04	1.66	−0.74	−2.16	0.68	2.11	−0.65	4.87	0.008	−0.011	0.028
	Heavy	−0.08	−1.96	1.81	0.12	−1.06	1.30	−0.56	−2.23	1.11	1.16	−3.37	5.69	0.033	0.010	0.057
p for interaction			0.73			0.68			0.0026			0.31			0.45
PFOS	Missing	1.00	0.55	1.46	0.56	0.17	0.96	−0.74	−2.12	0.64	−0.38	−1.33	0.57	0.013	0.001	0.024
	Non-drinker	0.27	−0.44	0.97	0.12	−0.31	0.54	−1.21	−2.32	−0.09	0.34	−1.50	2.18	0.006	−0.005	0.017
	Light	0.01	−0.48	0.50	−0.29	−0.82	0.24	−0.43	−1.62	0.75	−1.31	−3.47	0.86	−0.001	−0.013	0.011
	Moderate	−0.75	−2.29	0.79	−1.30	−2.86	0.26	−1.07	−2.41	0.28	−1.08	−3.65	1.50	0.003	−0.015	0.020
	Heavy	0.59	−0.72	1.90	−0.03	−0.85	0.79	−0.41	−2.42	1.60	−4.15	−13.31	5.01	0.014	−0.009	0.037
p for interaction			0.28			0.14			0.07			0.97			0.99
PFHxS	Missing	0.88	0.51	1.25	0.45	0.19	0.71	−1.45	−2.41	−0.49	−0.56	−1.35	0.23	0.012	0.000	0.024
	Non-drinker	0.55	−0.01	1.12	0.37	−0.03	0.77	−0.30	−1.21	0.61	0.31	−0.96	1.58	0.009	−0.005	0.022
	Light	0.77	0.30	1.25	0.45	−0.08	0.99	0.08	−0.57	0.73	0.08	−0.64	0.80	0.008	−0.002	0.018
	Moderate	0.37	−0.41	1.16	−0.19	−0.95	0.58	−0.38	−1.44	0.69	0.28	−1.22	1.78	0.013	0.001	0.025
	Heavy	−0.04	−1.23	1.14	0.26	−0.55	1.08	−1.04	−3.06	0.98	−4.95	−15.38	5.47	0.039	0.011	0.067
p for interaction			0.98			0.96			0.07			0.10			0.0064
PFNA	Missing	0.47	−0.07	1.01	0.23	−0.27	0.72	0.09	−1.22	1.39	−0.39	−1.81	1.03	0.014	0.002	0.026
	Non-drinker	0.37	−0.39	1.14	0.09	−0.33	0.50	−1.35	−2.65	−0.05	−0.02	−1.73	1.69	0.013	−0.001	0.027
	Light	0.61	−0.12	1.34	0.36	−0.46	1.19	0.25	−0.93	1.43	2.20	−0.82	5.21	0.008	−0.004	0.020
	Moderate	0.55	−0.56	1.67	−0.17	−1.18	0.84	−1.31	−2.49	−0.12	1.17	−1.39	3.73	0.000	−0.017	0.016
	Heavy	0.63	−0.72	1.98	−0.39	−1.68	0.90	−0.58	−3.15	1.99	−5.38	−15.58	4.83	0.037	0.013	0.060
p for interaction			0.87			0.87			0.04			0.55			0.80
PFDA	Missing	0.30	−0.22	0.82	0.06	−0.35	0.47	−0.13	−1.50	1.25	0.00	−1.16	1.16	0.010	−0.003	0.023
	Non-drinker	−0.29	−0.87	0.29	−0.07	−0.48	0.34	−1.53	−2.59	−0.47	−0.49	−1.64	0.67	0.008	−0.007	0.022
	Light	0.18	−0.37	0.73	0.03	−0.62	0.67	−0.59	−1.46	0.28	0.77	−1.33	2.86	0.006	−0.008	0.020
	Moderate	0.12	−1.06	1.30	−0.54	−1.61	0.53	−1.66	−3.22	−0.11	−0.20	−3.02	2.61	0.001	−0.015	0.017
	Heavy	−0.48	−1.80	0.84	−1.30	−2.25	−0.36	−0.72	−2.39	0.95	−4.28	−10.63	2.07	0.030	0.011	0.048
p for interaction			0.99			0.55			0.08			0.09			0.95

^a:Perfluorooctanoic acid (PFOA), perfluorooctane sulfonic acid (PFOS), perfluorohexanesulfonic acid (PFHxS), perfluorononanoic acid (PFNA), perfluorodecanoic acid (PFDA)

^b:Alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), gamma-glutamyl transferase (GGT), total bilirubin (TB)

^c:Adjusted for age, age², gender, race/ethnicity, body mass index, poverty income ratio, education level, smoking status and survey cycle

Discussion

In this study, we observed associations between serum levels of several PFAS and liver biomarkers in the U.S. general population, including positive linear associations between most PFAS and ALT and TB, and inverse associations with ALP. Several PFAS were positively associated with high levels (>95^th percentile) of liver biomarkers. Although the interactions were not statistically significant, we observed some suggestion that levels of alcohol intake influenced these associations.

We found that PFOA was associated with increased ALT in these NHANES data 2003–2016, consistent with previous studies using data from different NHANES cycles, including Gleason et al. (Gleason et al., 2015) using data from 2007–2010 and Lin et al. (Lin et al., 2010) using data from the 1999–2000 and 2003–2004 cycles. Serum levels of PFOA in these studies using data from earlier cycles (Gleason: median, 3.7 ng/mL; Lin: mean, 4.6 ng/mL) were slightly higher than in our analysis (median: 2.89 ng/mL, mean: 3.64 ng/mL), reflecting the decline in PFOA serum levels in more recent years due to voluntary discontinuation in its use in the early 2000s. Another analysis using data from NHANES 2011–2014 also observed a positive PFOA-ALT association only among obese participants (geometric mean among obese individuals: 2.2 ng/ml) (Jain and Ducatman, 2019b). We also observed this association, but both in the overall population (geometric mean: 2.8 ng/ml) and among obese participants, where the relationship was slightly stronger, with comparable exposure levels (geometric mean among obese individuals: 2.63 ng/ml). Most other studies investigating the relationship between PFAS and liver biomarkers were conducted within highly exposed populations. A small study in a community in Southeastern Ohio (PFOA median: 354 ng/mL) found that participants with self-reported liver diseases (e.g., cirrhosis of the liver, hepatitis, any other liver condition) had higher levels of measured serum PFOA (527 ng/ml) compared to those without liver disease (441 ng/ml). They likewise did not observe significant relationships between any biomarkers (total bilirubin, ALP, AST, and ALT) and serum PFOA levels, although the study included only 317 participants and 13 cases of liver disease (Emmett et al., 2006). Similar to our observations, the C8 Health Project in a highly exposed community in the Mid-Ohio Valley also found a monotonic increase in associations between measured serum PFOA and PFOS concentrations and ALT serum levels (Gallo et al., 2012). Another study from the C8 Health Project found a significant positive association between modeled cumulative serum PFOA levels and ALT and a negative association with bilirubin (Darrow et al., 2016). They also did not observe evidence of an association with self-reported diagnosis of liver diseases, which we could not evaluate with our data. A longitudinal study conducted among a random sample of the general population in Uppsala, Sweden also found a positive association between changes in plasma PFOA and ALT at levels similar to those observed in our analysis (median: 3.31 ng/ml at age 70, 3.81 ng/ml at age 75, 2.53 ng/ml at age 80) (Salihovic et al., 2018). Two prior NHANES analyses observed a positive association between PFOA and GGT. This association was also found among obese participants in the aforementioned study by Jain and Ducatman (Jain and Ducatman, 2019b). In Lin et al., the association was not only observed in the overall population but also in subgroups of non-Hispanic white individuals and those with higher BMI, lower alcohol consumption, and higher insulin resistance. However, we did not find evidence of a PFOA-GGT association in our analysis, either overall or in obese participants.

We found a positive association between PFHxS and PFNA and ALT, an indicator of liver cell damage or cell death (Giannini et al., 2005; Knudsen et al., 2016), in linear models. A study of residents in Shenyang, China, an industrial city and a major fluoropolymer manufacturing center, reported a strong positive linear association between serum PFNA levels similar to in our study (median: 1.96 ng/ml) and ALT (Nian et al., 2019). Jain and Ducatman also found positive associations between both PFHxS and PFNA and ALT level in NHANES, but only among obese participants (geometric mean among obese: PFHxS: 1.24 ng/ml, PFNA: 0.73 ng/ml) (Jain and Ducatman, 2019b). Gleason et al. likewise found a significant positive association between PFHxS levels and ALT in both linear and logistic models with fewer cycles of NHANES data but with similar exposure levels to our study (median: 1.59 ng/ml vs. 1.4 ng/ml in Gleason et al.) (Gleason et al., 2015). In contrast to that study, we did not find an association between PFHxS and high levels of ALT in our logistic models, nor an association with PFNA. Similarly null associations between PFNA and ALT were also reported in an older occupational study with PFAS exposure levels estimated based on exposure history (Mundt et al., 2007). To our knowledge, no studies on PFHxS have reported its role in alteration in ALT in rats or mice (Butenhoff et al., 2009; Chang et al., 2018; Costello et al., 2022). However, other in vitro or in vivo animal studies have suggested that exposure to both PFHxS and PFNA affected the viability of hepatocytes, increased expression of inflammatory cytokines (Fang et al., 2012), led to altered lipid metabolism and increased lipid accumulation (Das et al., 2017), induced miRNA-mediated hepatotoxicity (Wang et al., 2015), and contributed to liver steatosis (Pfohl et al., 2021). The inconsistencies found in the association between PFHxS or PFNA and ALT in our logistic models compared to prior epidemiologic studies might be due to differences in modeling choices and population exposure distributions. For example, Gleason et al. dichotomized the liver biomarker outcomes according to 75^th percentile levels in 2007–2010 NHANES, and in our study, we defined high levels of liver biomarker based on sex-specific 95^th percentile levels that are higher than the sex-specific reference levels in the study in Shenyang (male: 57 U/L vs 45 U/L in Nian et al., female: 39.40 U/L vs 34 U/L) (Nian et al., 2019). Other factors could also affect the level of ALT and vary across study populations, including medication use, BMI, and lifestyle factors such as diet, smoking and sleep (Giannini et al., 2005; Jang et al., 2012; Kettner et al., 2016; Malakouti et al., 2017).

We found that ALP was negatively associated with PFOS, PFHxS and PFDA in linear models and the association with high levels of ALP remained consistent in logistic models. Elevated ALP may indicate liver damage, liver disease, or bone diseases (Chalasani et al., 2018). In contrast to our findings, most prior NHANES studies have noted positive or null associations between PFAS and ALP. Gleason et al. found a positive association between PFOA and ALP in their linear model, and high levels of ALP (>75^th percentile) were non-significantly and negatively associated with PFOA and PFNA (Gleason et al., 2015). Jain and Ducatman found no linear associations between PFAS and ALP except for PFDA among non-obese participants (Jain and Ducatman, 2019b). Our findings may differ from these studies because we included data from the most recent NHANES cycles, which included lower serum levels of legacy PFAS. To our knowledge, there is no animal or human evidence to support a biologic mechanism for a protective effect of PFAS on ALP levels. However, inverse associations for ALP and other liver biomarkers have been observed in other settings, including the aforementioned association with bilirubin reported in Darrow et al. and Salihovic et al. A study of participants from Canadian Health Measures Survey (CHMS) that included children as young as 3 years as well as adults also found a negative linear association between plasma PFOS and ALP (Cakmak et al., 2022). However, another study from the CHMS that included only the adult participants >20 years found positive associations between PFAS and ALP (Borghese et al., 2022). We speculate that these inverse associations in our study might be due to residual confounding, since ALP is widely found throughout the body and lower levels could be driven by a number of other factors, including oral contraceptives (Naz et al., 2016) and coffee intake, which is also a PFAS exposure source (Jang et al., 2012; Xiao et al., 2014), although we found that adjustment for the latter did not change our observed associations.

Like our findings, another NHANES study also found a positive association between PFOS and TB, although serum PFOS levels in that study were higher than ours (median: 10.1 ng/mL vs 11.3 ng/mL) (Gleason et al., 2015). However, another NHANES-based study did not find an association between PFOS exposure and total bilirubin (Lin et al., 2010). In the CHMS, a positive association was found between PFOS and TB although their exposure levels were lower than ours (geometric mean: 9.5 ng/mL vs 5.3 μg/L) (Cakmak et al., 2022); no association was found in analyses limited to participants aged >20 years (Borghese et al., 2022). A study in the general elderly population in Sweden found that changes in plasma PFAS levels, including PFOS, were negatively associated with TB. This might be due to the drop in plasma PFOS levels observed with age (median: 13.2 ng/ml at age 70, 12.6 ng/ml at age 75 and 0.57 at age 80) or because of the older age of their population in general (Salihovic et al., 2018).

Previous studies that evaluated relationships of PFAS and liver diseases usually excluded heavy drinkers or assessed categorized drinking history (non-, former, or current drinkers) as a confounder in their models. However, studies have increasingly shown that light-to-moderate drinking could also have effects on liver health (Ajmera et al., 2018; Sookoian et al., 2016). In analyses of high biomarker levels, we found a positive association among the heavy drinkers between PFOA and high GGT and an inverse association between PFDA and high GGT. The positive association between PFOA and high GGT among heavy drinkers could be explained by the “two-hit” hypothesis of the pathogenesis of fatty liver disease, which suggests that environmental exposures may undermine protective responses of the liver against risk factors (Wahlang et al., 2019). High serum GGT is a predictor of cellular antioxidant inadequacy and biomarkers of several diseases, including liver disease (Koenig and Seneff, 2015). Serum GGT involves in glutathione metabolism and is a well-established and more sensitive biomarker for heavy alcohol intake compared with other liver enzymes (Strid and Litten, 2003; Whitfield, 2001). In vitro studies have shown that PFAS may reduce intracellular GSH levels and increase oxidative stress, with stronger effects for PFAS with longer carbon chains (Ojo et al., 2020; Pan et al., 2018). To better understand the mechanism of PFAS on liver disease and interaction with alcohol, future studies should explore the relationship between biomarkers of oxidative stress such as serum glutathione in the context of exposure to PFAS and co-exposure with high levels of alcohol. We did not find strong evidence that the association between most PFAS and ALP varied by alcohol intake. A study using NHANES 24h-dietary recall data (2001–2010) found a negative dose-response association between alcohol intake and serum ALP (Agarwal et al., 2016). Another cross-sectional study in Korea found an association between current alcohol consumption, heavy lifetime alcohol intake and adjusted mean ALP levels. (Jang et al., 2012). Residual confounding by lifestyle factors, biliary obstruction and cholestasis could have distorted the associations we observed. We found some linear associations for liver biomarkers with multiple PFAS, mostly in the heavy alcohol consumption group. Our observation of a positive linear association between TB levels and PFHxS, PFNA and PFDA among heavy drinkers may suggest a synergistic association between these three PFAS and heavy alcohol consumption. The finding of a positive association between the highest levels of PFNA and PFDA and high TB levels in heavy drinkers is consistent with several other studies (O’Malley et al., 2015; Puri et al., 2020; Tanaka et al., 2013). One possible explanation is that alcohol interferes with the process of bilirubin conjugation (Hossain et al., 2018; O’Malley et al., 2015) but the effect of PFAS on bilirubin elimination under co-exposure to alcohol remains unclear.

This study has several strengths. First, because PFAS were only measured in one-third of samples in each NHANES cycle, we included seven cycles to maximize our sample size (NHANES 2003–2016). NHANES sampled participants across the U.S. and represent the exposure pattern of the U.S. general population. Second, we evaluated established biomarkers for liver injury rather than self-reported history of liver disease to avoid reporting bias. For the first time, we also assessed the role of alcohol consumption in this association based on detailed drinking histories. Despite these strengths, there remain several limitations to this study. The cross-sectional NHANES design limited our ability to evaluate the causality between PFAS exposure and liver function biomarkers. Reverse causation cannot be ruled out since liver disease could lead to kidney dysfunction and reduce the excretion of PFAS, thus abnormal liver function could increase the accumulation of PFAS rather than vice versa (Byrne and Targher, 2015; Musso et al., 2014). Increased alcohol intake could also increase the risk of kidney conditions and thus, influence the accumulation of PFAS (Park et al., 2021). However, the positive associations with ALT observed in our data have also been observed in longitudinal studies (Salihovic et al. 2018). Some liver enzymes have short-term within-person variability (Dufour et al., 2000) and one-time measurement may be insufficient to identify the early stages of reversible liver damage (Eslam et al., 2020; Sanyal, 2019). However, most of the PFAS we evaluated have estimated serum half-lives of several years (Jian et al., 2018). There may also be outcome misclassification as we defined ‘high’ levels of liver biomarkers as the 95^th percentile or lower rather than using liver biopsy or liver ultrasound images to evaluate the degree of liver damage and liver function. Finally, the measurement of total bilirubin in the NHANES does not distinguish between direct (conjugated) and indirect (unconjugated) bilirubin, which is bound to serum albumin. Direct bilirubin may be a more important indicator of PFAS-related liver toxicity because it is a byproduct of the breakdown of hemoglobin and is strongly associated with hepatocyte injury.

Conclusion

In summary, we found that increasing serum levels of PFOA, PFHxS and PFNA were associated with high levels of ALT, and we observed weak positive associations between several PFAS and TB. Although serum concentrations of PFOA have decreased in the U.S. population over the last 20 years, its association with ALT remained consistent with the inclusion of the most recent PFAS data from NHANES, indicating that the lower levels of PFOA exposure now experienced by the general population could still be toxic to the liver. Although the interactions with alcohol consumption were not statistically significant, our analyses also suggest that heavy alcohol consumption might interact with PFAS on these relationships.

Data availability

NHANES is publicly available data on the CDC website.

Table 1.

Characterizing serum PFAS levels by demographic features: median and quartiles (ng/mL), NHANES 2003–2016

	PFAS (median, Q1, Q3) a
	N	PFOA	PFOS	PFHxS	PFNA	PFDA
Cycle
2003–2004	1529	4.13 (2.81, 5.84)	21.71 (14.87, 30.58)	1.79 (1.08, 3.07)	0.93 (0.61, 1.49)	0.17 (0.14, 0.32)
2005–2006	1504	4.13 (2.65, 6.27)	17.75 (11.45, 27.38)	1.71 (0.95, 3.08)	1.02 (0.69, 1.64)	0.29 (0.16, 0.49)
2007–2008	1798	4.28 (2.96, 6.06)	13.94 (8.73, 21.48)	1.92 (1.06, 3.34)	1.20 (0.84, 1.71)	0.24 (0.14, 0.39)
2009–2010	1892	3.29 (2.17, 4.63)	10.06 (6.16, 15.50)	1.66 (0.94, 2.84)	1.20 (0.83, 1.75)	0.24 (0.14, 0.38)
2011–2012	1603	2.16 (1.49, 3.14)	7.02 (4.32, 10.80)	1.29 (0.74, 2.89)	0.88 (0.61, 1.32)	0.20 (0.12, 0.31)
2013–2014	1801	2.06 (1.32, 3.09)	5.51 (3.22, 9.02)	1.37 (0.76, 2.46)	0.65 (0.40, 1.00)	0.45 (0.08, 0.26)
2015–2016	1667	1.65 (1.06, 2.47)	5.21 (2.93, 8.63)	1.21 (0.69, 2.09)	0.54 (0.35, 0.89)	0.10 (0.07, 0.22)
Sex
Male	5805	3.38 (2.16, 5.18)	12.42 (6.89, 21.65)	2.00 (1.29, 3.29)	1.00 (0.69, 1.56)	0.21 (0.18, 0.40)
Female	5989	2.50 (1.49, 4.09)	8.15 (4.13, 15.01)	1.19 (0.69, 2.19)	0.89 (0.55, 1.35)	0.20 (0.14, 0.39)
Age Category
18–29	2578	2.68 (1.66, 4.39)	8.29 (4.32, 15.03)	1.40 (0.80, 2.80)	0.82 (0.58, 1.29)	0.19 (0.14, 0.30)
30–39	1807	2.60 (1.50, 4.40)	8.47 (4.18, 15.87)	1.37 (0.70, 2.38)	0.90 (0.57, 1.40)	0.20 (0.14, 0.39)
40–49	1903	2.69 (1.67, 4.44)	9.28 (5.06, 17.43)	1.40 (0.79, 2.46)	0.90 (0.59, 1.47)	0.20 (0.14, 0.39)
50–59	1685	3.09 (1.89, 4.91)	11.11 (6.00, 19.58)	1.59 (0.96, 2.78)	0.99 (0.67, 1.50)	0.20 (0.14, 0.40)
60–69	1834	3.19 (2.08, 4.79)	11.99 (6.80, 21.13)	1.89 (1.13, 2.99)	1.00 (0.69, 1.59)	0.27 (0.19, 0.40)
≥70	1987	3.19 (2.10, 4.90)	14.24 (7.81, 24.59)	2.00 (1.23, 3.28)	0.99 (0.60, 1.50)	0.20 (0.14, 0.40)
Race/Ethnicity
NH-White	5156	3.18 (2.00, 4.99)	10.76 (5.84, 19.06)	1.70 (1.00, 2.92)	0.97 (0.60, 1.43)	0.20 (0.14, 0.39)
NH-Black	2481	2.56 (1.46, 4.10)	10.92 (5.16, 21.41)	1.40 (0.79, 2.64)	0.65 (1.00, 1.60)	0.18 (0.29, 0.45)
Mexican
American	2047	2.10 (1.29, 3.49)	7.01 (3.79, 12.85)	1.20 (0.69, 2.06)	0.79 (0.50, 1.19)	0.19 (0.13, 0.30)
Other	2110	2.29 (1.47, 3.77)	7.80 (4.17, 14.58)	1.29 (0.69, 2.18)	0.93 (0.59, 1.56)	0.21 (0.14, 0.49)
Body mass index
Underweight and normal	3623	2.80 (1.80, 4.49)	9.90 (5.39, 17.53)	1.50 (0.89, 2.79)	0.90 (0.59, 1.46)	0.20 (0.15, 0.40)
Overweight	3828	3.17 (1.96, 4.99)	11.12 (5.89, 19.89)	1.70 (1.00, 2.89)	0.99 (0.65, 1.55)	0.20 (0.18, 0.40)
Obese	4182	2.76 (1.66, 4.40)	9.40 (4.89, 17.72)	1.49 (0.87, 2.59)	0.59 (0.90, 1.39)	0.20 (0.13, 0.32)
Ratio of family income to poverty
Missing	992	2.68 (1.73, 4.29)	9.14 (4.99, 17.71)	1.52 (0.79, 2.60)	0.99 (0.65, 1.46)	0.20 (0.16, 0.39)
≤1	2360	2.39 (1.37, 3.89)	7.88 (3.99, 14.33)	1.30 (0.69, 2.39)	0.81 (0.50, 1.29)	0.20 (0.10, 0.30)
>1	8442	3.00 (1.87, 4.78)	10.52 (5.59, 19.01)	1.60 (0.99, 2.80)	0.98 (0.60, 1.48)	0.20 (0.14, 0.39)
Education
Less than high school	3252	2.59 (1.61, 4.31)	9.80 (5.09, 18.27)	1.49 (0.88, 2.59)	0.90 (0.59, 1.40)	0.20 (0.14, 0.39)
High school or equivalent	2945	3.08 (1.87, 4.83)	10.90 (5.39, 19.67)	1.60 (0.91, 2.79)	0.98 (0.60, 1.48)	0.19 (0.14, 0.39)
At least some college	5581	2.90 (1.80, 4.64)	9.90 (5.30, 17.71)	1.59 (0.90, 2.80)	0.93 (0.60, 1.46)	0.20 (0.14, 0.39)
Smoking
Non-smokers	6187	2.76 (1.69, 4.48)	9.70 (5.00, 17.53)	1.49 (0.80, 2.60)	0.90 (0.60, 1.40)	0.20 (0.14, 0.39)
Former smokers	2732	3.10 (1.99, 4.87)	11.54 (6.38, 20.47)	1.79 (1.06, 2.95)	0.98 (0.65, 1.55)	0.20 (0.19, 0.40)
Current smokers	2273	2.90 (1.78, 4.69)	9.40 (4.99, 18.07)	1.60 (0.96, 2.70)	0.90 (0.59, 1.40)	0.20 (0.13, 0.39)
Alcohol consumption
Missing	1995	2.90 (1.80, 4.59)	10.71 (5.76, 20.02)	1.50 (0.90, 2.59)	0.90 (0.59, 1.40)	0.20 (0.14, 0.39)
Non-drinkers	1602	2.47 (1.46, 4.18)	9.27 (4.63, 17.90)	1.29 (0.70, 2.50)	0.82 (0.50, 1.39)	0.20 (0.13, 0.20)
1–2 drinks/day	4223	2.90 (1.80, 4.69)	10.33 (5.40, 18.40)	1.60 (0.90, 2.79)	0.98 (0.60, 1.47)	0.20 (0.15, 0.39)
2–4 drinks/day	1442	2.90 (1.87, 4.61)	9.36 (5.11, 17.40)	1.68 (0.95, 2.88)	0.95 (0.60, 1.43)	0.20 (0.14, 0.39)
>4 drinks/day	1054	3.18 (1.99, 5.06)	9.20 (5.35, 17.82)	1.78 (1.08, 3.00)	0.97 (0.60, 1.47)	0.20 (0.14, 0.38)

^a:Perfluorooctanoic acid (PFOA), perfluorooctane sulfonic acid (PFOS), perfluorohexanesulfonic acid (PFHxS), perfluorononanoic acid (PFNA), perfluorodecanoic acid (PFDA)

Highlights:

• PFOA, PFHxS, and PFNA were positively associated with high alanine transferase.

• Several PFAS were positively associated with elevated total bilirubin.

• Positive associations in heavy drinkers suggest that further study of interactive effects is needed.