Modeled Perfluorooctanoic Acid (PFOA) Exposure and Liver Function in a Mid-Ohio Valley Community

Methods for cohort recruitment and data collection have been described in detail elsewhere (Frisbee et al. 2009; Winquist et al. 2013). Most participants were recruited among 40,145 participants in the C8HP who were aged ≥ 20 years and consented to be contacted for further studies. The C8HP cross-sectional survey was conducted in 2005 and 2006 per the terms of a class action lawsuit and included people who were exposed for at least 12 months (at home, work, or school) to water in any of six districts contaminated (to various degrees) by PFOA. The analysis of liver biomarkers was conducted among 30,723 people from the C8HP (including 1,892 people who worked at the chemical plant) with available liver injury biomarker measurements and retrospective serum PFOA estimates. The analysis of incident liver disease included participants in the C8HP as well as additional workers who were recruited from a previously established occupational cohort (Leonard et al. 2008) of 6,026 people who worked at the chemical plant during 1948–2002; there were 32,254 people from these two cohorts who had completed at least one follow-up survey (administered between 2008–2010 and 2010–2011) and had retrospective serum PFOA estimates (3,713 from the occupational cohort and 28,541 community members who had not worked at the plant). An enrollment flowchart is available in the Figure S1. The follow-up surveys covered demographics, residential history, health-related behaviors, and lifetime personal history of various medical diagnoses. The present study was approved by the Emory Institutional Review Board (IRB) as part of the overall approval of the work of the C8 Science Panel, all of which were based on the same IRB-approved informed consent and interview.

Described in detail elsewhere, yearly estimates of PFOA concentrations in local air, surface water, and groundwater were generated using an environmental fate and transport model (Shin et al. 2011a), and these environmental concentrations were combined with self-reported residential history, drinking-water sources, and water consumption rates to estimate yearly PFOA intake rates. These rate estimates were then used as an input in a pharmacokinetic model to generate yearly PFOA serum concentration estimates for each study subject starting in 1952 or the year of birth, whichever occurred later (Shin et al. 2011b). For study participants who were workers at the DuPont plant, job and department-specific yearly PFOA serum concentration estimates were generated using an occupational exposure model based on historical serum PFOA measurements, participants’ work histories, and knowledge of plant operating processes (Woskie et al. 2012). For those with higher residential exposures than occupational exposures, the estimates of PFOA serum concentrations from the residential exposure model were used. For workers whose occupational exposures were higher, serum estimates were decayed 18% per year after a person stopped working at the plant (based on a half-life of 3.5 years) (Olsen et al. 2007), until they reached a level predicted by the residential exposure model. The Spearman’s rank correlation between serum concentration estimates and serum concentrations measured in blood samples in 2005–2006 among 30,303 people was 0.71 (Winquist et al. 2013).

For each follow-up year for each subject, we calculated a measure of cumulative serum PFOA exposure by summing all previous yearly estimates of PFOA serum concentrations (referred to as “cumulative PFOA”) in units of year × ng/mL. Liver disease was analyzed in relation to this measure of cumulative exposure. For liver biomarkers, we assessed this cumulative PFOA exposure measure as well as the estimated serum PFOA concentrations in the year of the participant’s blood draw, because current PFOA serum concentrations might acutely impact markers of liver injury and because associations with estimated serum PFOA, unlike measured serum PFOA, cannot be explained by reverse causation (Watkins et al. 2013).

Liver biomarkers. We examined three markers of liver injury measured in 2005–2006 for the C8HP: alanine aminotransferase (ALT), γ-glutamyltransferase (GGT) and direct bilirubin (sometimes referred to as conjugated bilirubin). ALT enzymes are elevated after liver parenchymal cell injury and serve as a proxy of acute liver damage. Elevation of GGT enzymes is an early marker of cholestatic liver disorders, conditions in which the flow of bile from the liver is slowed or blocked. Elevated direct bilirubin usually signals problems with the liver, bile ducts, or gallbladder. Blood samples were collected from participants and centrifuged, aliquoted, and refrigerated before shipping on dry ice daily from each data collection site to the laboratory (Frisbee et al. 2009). ALT, GGT, and direct bilirubin were measured using a Roche/Hitachi MODULAR automated analyzer (Roche Diagnostics, Indianapolis, IN, USA) at a clinical diagnostic laboratory (LabCorp, Inc., Burlington, NC, USA).

In addition to modeling natural–log-transformed continuous measures of the liver biomarkers, we created dichotomized measures of each biomarker by specifying a threshold for high concentrations using the same cutoff values used in a previously published cross-sectional study in this community (Gallo et al. 2012): 45 IU/L in men and 34 IU/L in women for ALT (Schumann et al. 2002a), 55 IU/L in men and 38 IU/L in women for GGT (Schumann et al. 2002b), and 0.3 mg/dL in both men and women for direct bilirubin (McPherson and Pincus 2007). These cutoff values reflect upper limits of the reference ranges for each assay; measurements above these values would be considered abnormally high, and could prompt additional investigation into liver function in a clinical setting.

Medically validated liver disease. On surveys administered between 2008 and 2011, participants were asked “Have you ever been told by a doctor or other health professional that you had any kind of liver condition such as Hepatitis, Cirrhosis, Fatty Liver, Enlarged Liver, or other liver condition?” A list of reported “other” liver diseases are available in the Supplemental Material, “Medical Records Abstraction.” For each reported condition, participants were asked “How old were you when you were first told that you had [liver condition]?” Participants were then asked to consent to medical record review; medical records were subsequently requested from the identified providers. In lieu of sending in the medical record documentation, physicians could submit an attestation form with the date of diagnosis of the self-reported condition(s), or a statement indicating “To the best of my knowledge, patient does not have this condition.” Trained medical record abstractors manually reviewed the full text of all medical record documentation submitted by providers. We focused on two outcome groupings: a) any medically validated liver disease and b) liver disease restricted to medically validated enlarged liver, fatty liver, or cirrhosis. Our purpose in analyzing the subcategory of fatty and enlarged liver and cirrhosis was to focus on liver conditions that might be expected to result from a toxic chemical exposure, excluding infectious hepatitis or “other liver disease” that includes a number of rare categories such as biliary obstruction and autoimmune liver disease. We excluded subjects whose self-reported liver disease was not medically validated, either because the medical record could not be obtained (n = 170, 9% of reported cases) or because the subject’s obtained medical record did not meet our a priori definition of liver disease (n = 783, 44% of self-reported cases). A list of liver conditions that were abstracted from medical records but did not meet our definition of liver disease (e.g., cysts, hemangiomas, primary or metastatic cancer) is available in the Supplemental Material, “Medical Records Abstraction.” We also excluded any subjects who did not self-report liver disease but whose medical record (obtained because of some other self-reported condition) indicated liver disease (n = 36) to avoid disproportionate inclusion of cases with comorbidities. For the enlarged liver, fatty liver, and cirrhosis outcome, we excluded anyone who reported hepatitis or “other” liver disease.

In linear regression models, natural–log-transformed liver function markers were analyzed in relation to estimated cumulative PFOA serum concentrations through 2005 or 2006 (depending on survey year) and estimated year-specific PFOA serum concentration in 2005 or 2006. Dichotomized measures of the liver function markers as defined above were modeled using logistic regression. Estimated serum PFOA was analyzed as a natural–log-transformed continuous variable and by quintiles; a p-value for trend across quintiles was obtained by including quintile as an ordinal variable in the model. Associations between PFOA and the liver biomarkers were adjusted for a priori covariates measured at baseline in 2005 or 2006 in the C8HP (parameterized as shown in Table 1 except for age): age (included as 5-year age categories), sex, body mass index (BMI), alcohol consumption, race, regular exercise, smoking status, education, household income, fasting status, history of working at DuPont plant, and insulin resistance. Insulin resistance, a predictor of liver damage, was defined using the homeostasis model assessment index (HOMA-IR: the product of basal glucose and insulin levels divided by 2.25) (Matthews et al. 1985). Models were similar to those of Gallo et al. (2012) to maximize comparability. We did not include high cholesterol or hypertension as potential confounders given these conditions may be downstream effects of the exposure (Winquist and Steenland 2014).

Associations between liver disease and estimated serum PFOA were examined in a survival analysis using stratified Cox proportional hazard models with age as the time scale, time-varying cumulative serum PFOA as a predictor, and stratified by birth year to control for any birth cohort trends. Cox model analyses started at the later of age 20 (to restrict to adult disease) or the subject’s age in 1952 (the year after PFOA production started at the plant) and included follow-up until 2011 at the latest, depending on the participant’s birth date, death date and timing of final interview (between 2008 and 2011). We also conducted a prospective subanalysis limited to participants who had not reported liver disease prior to the time of the C8HP in 2005/2006. In all analyses, those participants missing a diagnosis age for liver disease were excluded. We also excluded people born before 1920 (n = 173) because of uncertain reliability of disease self-reporting in this group. For liver disease models, quintiles were defined using the distribution of exposure estimates among cases in the year of diagnosis to maximize power. Cox models controlled for a mix of time-varying and constant covariates identified a priori and measured in the C8HP and follow-up interviews including: sex, years of schooling (constant; < 12 years, high school diploma/GED, some college, or bachelor’s degree or higher), race (white vs. nonwhite), smoking (time-varying; current, former, none), regular alcohol consumption (time-varying; current, former, none), BMI (at time of first study survey; underweight, normal, overweight, obese). To account for induction and latency of liver disease, in secondary analyses we also assessed cumulative PFOA exposure at a 10-year lag in relation to diagnosis of liver disease.

We investigated whether associations between PFOA and liver outcomes differed between a) individuals with and without a history of working at the plant b) men and women c) age < 50 years and age ≥ 50 years. Stratified analyses were conducted so that covariate effects were also estimated separately within each subgroup. Heterogeneity between strata-specific regression coefficients was tested by dividing the difference in coefficients by the square root of the sum of the variances and computing a Z-statistic. All analyses were conducted using SAS version 9.2 (SAS Institute, Cary, NC).

Table S1 shows the results of linear regression models for log-transformed liver biomarkers as a function of log-transformed cumulative serum PFOA for three models containing different nested subsets of covariates. Because results were almost identical between models including and excluding control for household income (missing for approximately 10% of participants), all subsequent models did not include household income as a covariate to maximize use of the data. The other a priori covariates were retained in final models because they had few missing observations and/or had a meaningful impact on estimated associations (see Table S1); only observations with complete data on these covariates were included in final models (n = 28,047). As shown in Table 2, the continuous (natural log transformed) measure of both cumulative and 2005/2006 modeled serum PFOA concentrations were associated with increased ALT and decreased direct bilirubin (p < 0.05), but neither exposure measure was associated with GGT. Assessment of PFOA by quintiles showed a monotonic increase in log-transformed ALT across quintiles of both metrics of PFOA (p < 0.05 for all except quintile 2 vs. quintile 1 for the 2005/2006 PFOA measure); in contrast the negative association between PFOA and log-transformed direct bilirubin was largely isolated to the fifth quintile (compared to the first). Quintile analyses also showed little evidence of an association with GGT; point estimates were in the positive direction but not statistically significant and did not show a monotonic increase across quintiles. Moving from the first to the fifth quintile of cumulative PFOA exposure was associated with an estimated 6% increase in ALT level [calculated as [exp(β)–1] × 100; 95% confidence interval (CI): 4, 8%]; this corresponds to an increase of 1.6 IU/L for an individual starting at the average ALT level of 26 IU/L, or an increase of 3.3 IU/L for an individual starting at 55 IU/L (the 95th percentile of ALT in our data).

As shown in Table 1, 11% of the study population was classified as having above normal ALT levels, 14% had above normal GGT, and 1% had above normal direct bilirubin. Figure 1 shows the results of logistic regression models for above normal ALT, GGT, and direct bilirubin as a function of estimated 2005–2006 cumulative and year-specific PFOA serum concentrations (numerical results presented in Table S2). Both metrics of PFOA were associated with high ALT levels (p < 0.05); the odds ratio (OR) for above normal ALT per unit increase in either cumulative (ln y-ng/mL) or year-specific PFOA (ln ng/mL) was 1.04 (95% CI: 1.01, 1.07). There was evidence of an increasing trend in the odds of having high ALT across PFOA quintiles (p < 0.01), with the highest ORs observed in the fourth quintile compared to the first (cumulative PFOA OR = 1.20, 95% CI: 1.06, 1.35; 2005/2006 PFOA OR = 1.16, 95% CI: 1.03, 1.31). There was little evidence for an association with above normal GGT or direct bilirubin; however, the second and third quintiles of cumulative PFOA exposure had elevated odds of high GGT relative to the first quintile (p < 0.05).

Analyses stratified by sex, age (< 50 and ≥ 50 years) and history of working at the plant provided no evidence of effect modification by these factors (p > 0.05 for interaction, similar point estimates across groups, see Table S3). Notably both metrics of PFOA were associated with higher ALT in all subgroups assessed, but the workers stratum only included 1,681 people and was not statistically significant.

There were 647 medically validated cases of liver disease included in the analysis, including fatty liver (n = 393), hepatitis (n = 157), cirrhosis (n = 66), enlarged liver (n = 44), and other liver disease (n = 48). Numbers of cases included in the enlarged liver, fatty liver, and cirrhosis outcome group were smaller than in the analysis of all liver disease because participants that also were diagnosed with hepatitis or other liver disease were excluded. Average age at diagnosis was 46. Table 3 presents hazard ratios for any liver disease and for liver disease restricted to enlarged liver, fatty liver and cirrhosis for cumulative PFOA with and without a 10-year lag. Table S4 shows results for the same models limited to participants who had no liver disease diagnosis prior to the C8HP and followed prospectively (included 266 cases of all liver disease and 209 cases of enlarged liver, fatty liver and cirrhosis). There was little evidence of increased risk of liver disease with PFOA exposure in any of the analyses; point estimates were mostly below the null and none of the CIs excluded the null.

In Figure 2, we present hazard ratios per natural log y-ng/mL increase in cumulative serum PFOA (with and without a 10-year lag) stratified by sex and history of working at the DuPont plant (numerical results are presented in the Table S5). Consistent with the overall analyses, these stratified analyses showed little evidence of association between PFOA and liver disease. The only elevated hazard ratio (HR) observed was for exposure at a 10-year lag and enlarged liver, fatty liver, and cirrhosis among workers (HR = 1.15; 95% CI: 0.85, 1.55), which was based on only 36 cases.