Attention Deficit/Hyperactivity Disorder and Childhood Autism in Association with Prenatal Exposure to Perfluoroalkyl Substances: A Nested Case–Control Study in the Danish National Birth Cohort

1Perfluoroalkyl substances (PFASs) are a group of human-made fluorine-containing compounds with unique properties that make materials resistant to stains, oil, and water. PFASs have been used widely in commercial products since the 1950s, and they are persistent in the environment and in living organisms throughout the globe. Human exposure routes include contamination of food from packaging, bioaccumulation in the food chain, and household dust. Perfluorooctane sulfonate (PFOS) and perfluorooctanoic acid (PFOA) are the two most frequently used PFASs; they have estimated biological half-lives in humans between 4 and 5 years. PFOS and PFOA concentrations in humans were reported to be decreasing in some countries following a drop in production since 2000, but exposure to other short-chain compounds such as perfluorobutane sulfonate (PFBS) and perfluorohexane sulfonate (PFHxS) and long-chain perfluorononanoic acid (PFNA) and perfluorodecanoic acid (PFDA) are reported to be increasing.

2PFASs can cross the placental barrier and expose the fetus during the most vulnerable period of development. Experimental data suggest that PFASs may be developmental neurotoxicants that can affect neuronal cell development, alter cognitive function, and reduce habituation and learning ability in mice. PFASs also have endocrine-disruptive properties and might interfere with thyroid hormone function, which is essential in regulating fetal brain development.

3Attention deficit/hyperactivity disorder (ADHD) is considered one of the most common neurobehavioral disorders worldwide, and is characterized by inattention, hyperactivity, increased impulsivity, and motivational/emotional dysregulation. Autism is a neurodevelopmental disorder characterized by impairments in communication and reciprocal social interaction, coupled with repetitive behavior. The incidence of ADHD and autism has increased over the past decades, and it has been suggested that the rise is not attributable solely to changes in diagnostic practices or parental awareness. The etiologies are not well understood, but both environmental and genetic factors are thought to contribute to ADHD and autism. ADHD and autism disproportionately affect boys, and studies suggest that prenatal exposure to endocrine-disrupting chemicals may be associated with the occurrence of both diseases.

4A limited number of epidemiologic studies have evaluated the potential neurobehavioral or neurocognitive impact of PFASs and findings were inconclusive. Several cross-sectional studies have reported positive associations between serum levels of some PFASs with impulsivity and ADHD in children. Reverse causality, however, is a concern for studies that measure PFAS levels in children already diagnosed with ADHD at time of blood draw. Little evidence of associations was found for prenatal exposures to PFOS or PFOA and behavioral problems in 7-year-old children assessed with the Strengths and Difficulties Questionnaires in the prospective Danish birth cohort. A study conducted in a community with high long-term exposure to PFOA in contaminated drinking water reported that in utero PFOA levels were associated with higher Full-Scale IQ and decreased ADHD characteristics among children 6–12 years of age. However, prenatal PFOA exposures were estimated based on exposure modeling. A recent study examined the associations between several endocrine-disrupting chemicals, including PFASs, and autistic behaviors in children but no conclusive evidence was found, perhaps due to small sample size (175 mother–child pairs) and low statistical power.

5We conducted a nested case–control study within the framework of the Danish National Birth Cohort (DNBC) to examine whether prenatal exposure to PFASs is associated with ADHD or autism in children.

6The DNBC is a nationwide cohort study of pregnancies and health-related outcomes in the children; details have been described elsewhere. Briefly, pregnant women were recruited through their general practitioners during early gestation (weeks 6–12) from 1996 to 2002. About 50% of all general practitioners in Denmark participated in the study, and 60% of the women invited agreed to participate. Women were ineligible if they did not speak sufficient Danish for interviews or intended not to carry their pregnancy to term. Information was collected during four computer-assisted telephone interviews (twice during pregnancy and twice postpartum). Two prenatal maternal blood samples were collected and stored, one each in the first and second trimester. English versions of questionnaires are available online.

7Written informed consent was obtained from all participants at recruitment. Study procedures have been approved by the Danish Data Protection Agency and the Institutional Review Board at University of California, Los Angeles.

8Source population. The source population for this study consisted of live-born singletons along with mothers who participated in the first telephone interview, conducted approximately during the 12th gestational week, and had provided a blood sample at least once either during the first or second pregnancy trimesters. This resulted in 83,389 mother–child pairs, with 42,737 boys and 40,652 girls; we excluded from the original DNBC those with an unsuccessful pregnancy (n = 6,207), non-singleton births (n = 2,080), births with unknown birth outcomes (n = 25) or missing dates of birth (n = 99), mothers who emigrated (n = 51) or died (n = 3), and women who did not participate in the first telephone interview (n = 4,578) or did not provide a prenatal blood sample (n = 4,609).

9Selection of cases and controls. We identified children who were diagnosed with ADHD and autism, respectively, by linking DNBC records to the Danish National Hospital Registry that contains the nationwide data for all admissions for somatic illnesses, and also to the Danish Psychiatric Central Registry, which covers admissions to all psychiatric hospitals in Denmark. The record linkage relied on the unique civil registration numbers given to all Danish citizens at birth. All diagnoses are based on the International Classification of Diseases, 10th Revision (ICD-10 codes F90.0 for ADHD and F84.0 for childhood autism) and included inpatient and outpatient records. A total of 890 ADHD cases and 301 autism cases were identified in the cohort during an average of 10.7 years of follow-up (record linkage was conducted on 1 August 2011). Because of the high costs of measuring PFASs, we randomly selected 220 cases of ADHD and autism each for inclusion in this study.

10We used a case-cohort sampling strategy and randomly selected 550 children (440 males and 110 females) as controls from the source population, frequency matched to cases by sex. The flow chart of subject selection and sampling fractions of cases and controls is shown in Figure 1.

11PFAS measurements. Details about analytic methods for PFASs have been described previously. Briefly, the collected maternal blood samples were sent by mail to Statens Serum Institut in Copenhagen, and separated and stored in freezers at –20°C or –80°C or in liquid nitrogen. We used 0.1 mL stored maternal plasma, and the samples were analyzed at the Department of Environmental Science at Aarhus University. Most of the samples (87%) for both cases and controls were collected during the first trimester; if the first sample was not available, we used the second sample collected in the second trimester instead. Solid-phase extraction technique was used for extraction and purification, and PFAS concentrations were measured by liquid chromatography–tandem mass spectrometry. Measurements were performed in a random sequence for cases and controls by laboratory personnel blinded to diagnoses and any other participant information. Seventeen maternal samples (5 ADHD, 7 autism, 5 controls) were either not available from the biobank or failed the PFAS extraction and purification process, and hence were excluded. For 21 samples included in our current study, PFOA and PFOS values previously had been analyzed at the 3M Toxicology Laboratory for an earlier study in the DNBC. For quality control we compared the PFOA and PFOS value measured from the two laboratories in these samples and found high correlations (Pearson r = 0.94 for PFOS and r = 0.95 for PFOA).

12Of the 16 different PFASs detected in maternal plasma, we focused on the six PFASs for which at least 90% of all samples were above the lower limit of quantitation (LLOQ): PFOS, 100%; PFOA, 100%; PFHxS, 98%; PFHpS (perfluoroheptane sulfonate), 96%; PFNA, 92%; and PFDA, 90%. The LLOQ for the six PFASs are as follows: PFOS, 0.28 ng/mL; PFOA, 0.20 ng/mL; PFHxS, 0.08 ng/mL; PFHpS, 0.11 ng/mL; PFNA, 0.27 ng/mL; and PFDA, 0.09 ng/mL. The full panel for the LLOQ and distribution of all PFASs was reported previously.

13Statistical analysis. We used generalized linear models and accounted for the sampling fractions of cases and controls to estimate risk ratios (RR) and 95% confidence intervals (CIs) for prenatal PFAS exposures and ADHD or autism. We also performed unconditional logistic regression to estimate odds ratios (ORs) without applying the sampling weights. PFAS concentrations were first analyzed as continuous variables (per natural-log unit increase). The PFAS values were natural-log transformed to reduce the influence of outliers, to improve the model fit, and to make interpretation simpler and more consistent across different PFASs that vary in their ranges of concentration. We also categorized PFAS values into quartiles according to the distribution among controls, using the lowest quartile as the reference group. Moreover, we fitted generalized additive models with a smoothing function of natural-log PFAS values to examine potential nonlinear relations. Five knots were set as the upper limit of number of degrees of freedom, and we compared model fit and visually inspected plots of the smoothed data. We did not find evidence for nonlinearity between natural-log PFAS values and ADHD or autism (data not shown).

14Potential confounders were chosen a priori considering variables that may influence PFAS exposures and previously suggested risk factors of ADHD or autism. We included maternal age at delivery (≤ 24, 25–29, 30–34, ≥ 35 years), parity (1, > 1), socioeconomic status (low/medium or high), maternal smoking (never, ≤ 9 cigarette/day, > 9 cigarettes/day) and alcohol drinking (yes, no) during pregnancy, mother’s self-reported psychiatric illnesses (yes, no), gestational week of blood draw (4–8, > 8 week), child’s birth year (1998–2000, 2001–2003), and the matching factor child’s sex in the final model. Socioeconomic status was determined based on self-reported maternal and paternal education and occupation using three categories (high, medium, and low): Higher education (4 years beyond high school) or work in management were classified as high; skilled workers and middle-range education as medium; and unskilled workers and unemployed as low. To determine maternal psychiatric illnesses, we asked women to report whether they had ever seen a doctor or psychologist because of depression, anxiety, childhood psychiatric disorder, family problems/life crisis, or other mental health problems. Additionally, other potential confounders such as father’s age at child’s birth, mother’s prepregnancy body mass index, whether the pregnancy was planned, and season of conception were evaluated but not included in final models because they changed effect estimates of interest minimally (< 1%).

15To account for PFAS values below the LLOQ when PFASs were analyzed as continuous variables, we used multiple imputation with the procedure PROC MI in SAS (SAS Institute Inc.) with all six PFASs and all covariates included in the model. Ten simulated complete data sets were generated via imputation, and we employed standard analytical procedures to combine the results.

16A Pearson correlation matrix for the considered PFASs is presented in Supplemental Material, Table S1. We constructed a “multiple PFAS” model where we simultaneously included all PFASs in one model to examine whether any single PFAS may be of particular importance. We also evaluated potential effect measure modification by child’s sex; we compared the sex-stratified estimates and examined the p-value for the PFAS–sex interaction term. For ADHD we also conducted analyses in which we excluded children born after 2000 because the duration of follow-up may not have been long enough to identify children with this diagnosis. For these stratified analyses we used logistic regression without applying sampling weights because the weighted analyses may underestimate uncertainty in our data when the number of actual measured samples is small. Finally, in sensitivity analyses we excluded PFAS values that were greater than three times the 75th percentile (n = 2 PFOA, n = 7 PFHxS, n = 1 PFNA, n = 2 PFHpS, n = 1 PFDA) to ensure that individuals with extreme exposure values did not disproportionately influence our results. Analyses were performed using SAS version 9.2.

17Table 1 presents the demographic characteristics of cases and controls. Table 2 shows the median and interquartile range distribution of maternal PFAS concentrations during pregnancy in cases and controls.

18We generally found no association between ADHD or autism in children and PFAS levels in maternal plasma (modeled as natural-log units) (Table 3). We did not detect apparent effect modification by child’s sex (all PFASs and sex interaction p-values ≥ 0.25), but because both diagnoses were more prevalent in boys, estimates for girls were less precise (see Supplemental Material, Table S2).

19When we categorized PFAS values, mothers in the highest quartile of PFOS, PFHxS, PFHpS, and PFDA were less likely to have a child diagnosed with ADHD than mothers in the lowest quartile, after adjustment for potential confounders (Table 4). When all PFASs were simultaneously entered into the model, PFOA and PFNA levels were positively associated with ADHD, whereas negative associations with the other compounds persisted, with most showing monotonic trends. There was some evidence of a positive association between PFHxS and autism, though RRs for the highest quartile were closer to the null than RRs for the second and third quartiles. Similar patterns were found with lower precision of the estimates when we used logistic regression without applying sampling weights (see Supplemental Material, Table S3).

20Results were similar to those from the primary models when we performed additional sensitivity analyses restricting the analyses to children born before 2001 (see Supplemental Material, Table S4), and excluding extreme PFAS values (results not shown).

21Overall, our results do not suggest that prenatal exposure to PFASs increases the risk of ADHD or childhood autism in children. We observed some inverse associations between several PFASs and ADHD after controlling for potential confounders. In the “multiple PFAS” model, we found some positive as well as negative associations between PFASs and ADHD but these might be subject to multicollinearity or sparse data bias. Results were mostly close to null for autism in both single and multiple PFAS models.

22Toxicology studies have raised concerns that PFASs are neurotoxic and hormone disruptive and can impair fetal brain development. However, some neurotoxic effects in rats were observed at doses several orders of magnitude higher than the PFAS levels found in the U.S. and Danish general populations. Several epidemiologic studies have investigated associations between PFASs and hyperactivity or behavioral problems in children, but the findings have been inconclusive. A previous study based on a subset of children from the Danish National Birth Cohort found some inverse associations between prenatal PFOA, but not PFOS, and behavioral problems in 7-year-old children measured by (parent-reported) items in the Strength and Difficulty Questionnaire. Another study also suggested a lower prevalence of ADHD characteristics in children associated with higher estimated in utero PFOA exposures based on the Clinical Confidence Index. There is, however, no biologic explanation for PFASs protecting the developing brain from ADHD, and potential biases such as uncontrolled confounding or selection bias might have driven these unexpected findings. No apparent associations were found between PFASs and autism in current and a previous small study.

23Because several PFASs are moderately to highly correlated, it is difficult to disentangle mixture effects from compound-specific effects. A recent in vitro assay reported an additive or more than additive antagonistic effect for a mixture of compounds (PFHxS, PFOS, PFOA, PFNA, and PFDA) on androgen receptor function. Unfortunately, our sample is too small to allow for interaction analyses between different PFASs. Further experimental studies are needed to determine mechanisms of action for PFAS mixtures on biologic targets that could better inform our population-based studies in terms of the most biologically relevant exposure model to be employed.

24It has previously been shown that prenatal exposure to PFASs can increase the incidence of fetal resorption and neonatal deaths in animal models. PFASs may interfere with sex and thyroid hormone homeostasis, and it has been suggested that higher PFAS levels are associated with reduced fecundity in women and with an increased risk for miscarriage. It is therefore possible that PFAS exposure at a level that reduces fetal or neonatal survival, especially in high-risk fetuses and infants susceptible to neurological disorders such as ADHD and autism, could appear to have null or even protective effects on adverse neurobehavioral outcomes in children based on observational studies, because only live-born children can be followed up and examined.

25There are several strengths in our study. First, the PFAS measures were obtained from maternal plasma samples collected in pregnancy before the assessment of the outcomes in the children. Previous studies have shown that PFASs are stable in human serum, and measurements obtained from serum or plasma samples gave comparable results. High correlations between maternal and cord blood PFAS measures were also reported, and these suggested that PFASs measured in maternal plasma can be used as a reasonable surrogate for fetal exposure levels throughout gestations. Furthermore, the maternal PFAS levels in our study are similar to those previously measured during the same time period in the U.S. general population. Study participants were selected from a well-defined nationwide pregnancy cohort with an average of 10.7 years of follow-up, sufficiently long to assess the outcomes of interest. The outcome measures were clinical diagnoses using standardized diagnostic criteria from both the general and psychiatric hospital registries in Denmark, a country with high-quality health care and universal coverage for its population. Diagnoses of childhood autism recorded in the psychiatric registry have previously been shown to have high validity: A study extracted and reviewed the medical records of 499 childhood autism cases from the registry and confirmed the diagnoses for 94% of the cases. Follow-up was conducted through record linkage that did not require subjects’ responses, thus minimizing chances for selection bias due to subject’s nonresponse.

26Our study also has some limitations. Both ADHD and autism are about four times more prevalent in boys, and because of cost limitations we were required to sample no more than 220 cases for each diagnostic group, resulting in few female cases (n = 41 with ADHD; n = 33 with autism). Thus, our subgroup analyses by sex were relatively imprecise for girls, resulting in effect estimates with wide CIs. For autism, the cases were limited to children diagnosed with childhood autism. Although this is the most severe disorder of the autism spectrum, it constitutes only a part of autistic spectrum disorders; specifically children with Asperger’s syndrome and other pervasive development disorders were not studied. Moreover, we have no data for other endocrine-disrupting chemicals, preventing us from evaluating possible correlations or interactions of PFASs with other ubiquitous environmental chemicals with these properties such as polychlorinated biphenyls (PCBs), organophosphates, bisphenol A, and phthalates. Further, our blood samples had to be transported to the laboratory by ordinary mail before being processed, which may have induced some random measurement errors.

27In summary, we found no consistent evidence that prenatal PFAS exposures were associated with ADHD or childhood autism in children in the Danish National Birth Cohort. Both weak negative associations as well as some positive associations between PFASs and ADHD that we observed in multiple PFAS models should be further explored. We recommend that future studies analyze a larger sample, consider both prenatal exposure and exposure during first year of life, assess the potential mixture effects of exposures to different co-occurring endocrine disruptors, and examine more sensitive indicators such as neuropsychological functioning in children.