C8 Health Project population.
The C8 Health Project enrolled participants between August 2005 and July 2006. The project’s purpose was to collect health data from members of the class action lawsuit through questionnaires and blood tests, including measurement of PFCs. This community was highly exposed to PFOA, but exposure to other PFCs reflects typical background levels. Individuals were eligible to participate in the C8 Health Project if they could provide documentation proving they had consumed water for at least 1 year between 1950 and 3 December 2004 in a water district supplied by Little Hocking Water Association of Ohio; City of Belpre, Ohio; Tupper Plains–Chester District of Ohio; Village of Pomeroy, Ohio; Lubeck Public Service District of West Virginia; Mason County Public Service District of West Virginia; or private water sources within areas of documented PFOA contamination. Participants provided informed consent and were compensated $400 for completing the questionnaire and providing blood. At the time of enrollment, participants did not know their individual exposure to PFOA. The C8 Health Project collected data on 69,030 people. The total number of people eligible to join the class action lawsuit is unknown. Participation rates based on U.S. Census counts of current residents of the eligible water districts are estimated at around 80% and in some ZIP codes within eligible water districts appear close to 100% (Frisbee et al. 2009
). The overall participation rate is likely < 80% because former residents were also eligible to participate. In this population, the strongest predictor of PFOA serum level was current residence in a contaminated water district, with distance to the plant directly affecting PFOA levels (Steenland et al. 2009
). Because plant emissions varied over the 50-year contamination period, the relative PFOA exposure levels of the water districts remained the same, such that high-exposure water districts were always more exposed than low-exposure water districts, regardless of the absolute exposure levels. Exposure to other PFCs is not determined by residence in a PFOA-contaminated water district and reflects typical background levels, which likely occur through dietary and consumer product exposures (Vestergren and Cousins 2009
). Of the 69,030 C8 Health Project participants, 12,016 were 5–18 years of age at enrollment. Of these children, 11,046 (92%) had serum PFC measurements.
Laboratory analysis (Exygen Research Inc., State College, PA) of PFCs used automated solid-phase extraction combined with reverse-phase high-performance liquid chromatography (Kuklenyik et al. 2004
). Four PFCs were detectable in 100% of samples (PFOA, PFOS, PFHxS, PFNA); we included these four in our analyses. We examined the association between PFCs and ADHD using restricted cubic splines (Desquilbet and Mariotti 2010
) and determined that quartiles of exposure best captured the nonlinear nature of the PFC–outcome associations. All analyses categorized PFCs into quartiles.
We examined the relation between PFC serum concentrations and a diagnosis of ADHD as reported in the C8 Health Project questionnaire. Participants were asked “Has a doctor or health professional ever told you that you have/had ‘Attention Deficit Disorder’ (ADD or ADHD)?” We additionally constructed a second, more sensitive ADHD definition by combining report of ADHD diagnosis with current use of a medication commonly used to treat ADHD (Braun et al. 2006
). Participants were asked to list all current prescription and over-the-counter medications they were currently taking for any reason. Based on guidance from clinical experts, medications considered treatment for ADHD were methylphenidate, dextroamphetamine, mixed amphetamine salts, lisdexamfetamine, dexmethylphenidate, atomoxetine, clonidine, guanfacine, imipramine, nortriptyline, bupropion, and carbamazepine. We included several medications that are prescribed off-label for complicated and recalcitrant cases of ADHD. This decision yielded 23 children who reported a diagnosis with ADHD and use of off-label medications only that would not have been included in the analysis of ADHD with medication had we restricted the category solely to those taking licensed drugs. These children tended to be older and male. Finally, we examined report of learning problems based on the question “Has a representative from a school or a health professional ever told you that you have/had a learning problem?” Among the 60% of the C8 Health Project population that gave permission to use identifying information allowing us to determine precisely who completed the questionnaire, 10.7% of children reported completing the questionnaires for themselves. Parents or legal guardians accounted for 98.2% of nonchild responders. As a sensitivity analysis, among the 60% of children where the respondent was identified, we compared the results for all children versus children where the parent or legal guardian completed the survey. Restriction to the subset with the parent/guardian as the named respondent had no effect on the pattern of results. The vast majority of the 682 children responding for themselves were old enough to be in high school and were likely adequate reporters of whether they had been diagnosed with ADHD. It is also possible that these children provided the preliminary information on the questionnaire themselves (name, birth date, address, and identification of who was completing the survey) and then passed the survey to a parent to answer the more involved questions on health outcomes. This change in respondent would not have been noted on the questionnaire.
Other covariates available for analysis included age (modeled as quintiles), sex, race/ethnicity (non-Hispanic white vs. other), body mass index (BMI) z
-score based on the 2000 CDC growth charts of BMI for age (CDC 2009
), and average household income (≤ $30,000 vs. > $30,000). Institutional review board approval was granted from the Mount Sinai Program for the Protection of Human Subjects.
Analysis was performed using SAS software (version 9.2; SAS Institute Inc., Cary, NC). Given the concern that PFCs may be hormonally active (White et al. 2011
), we first examined the potential for sex to modify the association between exposure and outcome by comparing the effect estimates of stratified and unstratified models and by examining the p
-value for the PFC–sex interaction term (Rothman and Greenland 1998
). There was little evidence that sex modified the exposure–outcome association, although for PFOA there was a modest, imprecise, suggestion of a stronger association among females; we report unstratified models. We assessed the potential for age, sex, race/ethnicity, BMI z
-score, and average household income to act as confounders by looking for associations between these covariates and the exposure, and these covariates and the outcome (Rothman and Greenland 1998
). Neither BMI z
-score nor race/ethnicity met the criteria for confounding, although we restricted analyses to 10,546 non-Hispanic white children (95.0%) to facilitate comparisons with other studies requiring adjustment for race/ethnicity. Because average household income was missing for 21.3% of participants, we compared odds ratios (ORs) adjusted for age, sex, and income with ORs adjusted for age and sex only. There was a < 10% change in ORs between the fully and partially adjusted models, so we excluded average household income from analysis, allowing us to retain a larger study population. We ran logistic regression models adjusted for age and sex to calculate the OR and 95% confidence intervals (CIs) for each PFC–outcome combination.
We also performed several secondary analyses: a
) For all 10,546 children, we ran each model, simultaneously adjusting all PFCs for one another to account for confounding. For instance, in the model for PFOA, we adjusted for age, sex, PFOS, PFHxS, and PFNA. b
) Among the 6,523 children permitting use of identifying information, we restricted analysis to the 2,437 children who lived in the same PFOA-exposed water district their entire life. Because PFOA exposure is directly related to residential location, this residential restriction is more likely to maintain the same relative ranking of PFOA exposure distribution over time, potentially reducing exposure misclassification, particularly if there is an early-life critical developmental window of susceptibility to PFOA. c
) To facilitate comparisons with the NHANES population (Hoffman et al. 2010
), we ran two additional analyses. We restricted analysis to the 3,571 children 12–15 years of age so that our age range was comparable to that of NHANES. We also examined the 5,262 children with PFOA exposures below the median (range, 0.6 ng/mL to < 28.2 ng/mL) to align our exposure range more closely to that of NHANES.