Perinatal Air Pollants and Autism Spectrum Disorder
Perinatal Air Pollants and Autism Spectrum Disorder
We used data from the Nurses' Health Study II, a cohort of 116,430 female nurses from 14 U.S. states that was established in 1989 and has been followed over time with biennial questionnaires. Initially nurses were recruited from 14 U.S. states, but since that time they have moved throughout the United States. Thus, children in the current analyses were born in all 50 U.S. states. In the 2005 questionnaire, respondents were asked "Have any of your children been diagnosed with the following diseases?" with autism, Asperger's syndrome, or other ASD listed as separate responses. In 2007–2008, we sent a questionnaire to the 756 women who had reported having a child with any of these conditions, querying the affected child's sex, birth date, and whether they were adopted. In addition, they were asked "What ASD diagnosis has the child been given?" with autism, Asperger syndrome, and PDD-NOS (PDD not otherwise specified) as possible answers. Women were also asked about other diagnoses (such as attention deficit hyperactivity disorder and obsessive compulsive disorder) (response rate = 84%, n = 636). The Partners Healthcare Institutional Review Board approved this research. Completion and return of questionnaires sent by U.S. mail constituted implied consent.
Cases were excluded for the following overlapping reasons: women reported on the follow-up questionnaire that the child did not have ASD (n = 32); the child was adopted (n = 9); they did not want to participate (n = 20); or they did not report the child's birth year (n = 71). Children reported to have trisomy 18, fragile X, an XXY genotype, or Down's, Angelman's, Jacobsen's, or Rett's syndrome also were excluded (n = 11). Of the remaining children, 329 were born after 1987, when air pollution data were available, but 4 had insufficient address information for geocoding, yielding 325 cases. In this study we refer to children with autism, Asperger's syndrome, or other autism spectrum disorder as cases and use "ASD" to refer to this case definition.
We validated the ASD diagnosis by telephone administration of the Autism Diagnostic Interview–Revised (ADI-R) (Lord et al. 1994) to 50 randomly selected case mothers who indicated willingness to complete the interview (81% of the 636 mothers who responded to the follow up questionnaire were willing to be interviewed). Diagnoses reported by women who were willing versus unwilling to participate in the substudy were similar (25% autism, 51% Asperger's, and 25% PDD-NOS compared with 25% autism, 49% Asperger's, and 23% PDD-NOS, respectively). For the subsample of mothers who completed the ADI-R, 43 children (86%) met full ADI-R criteria for an autism diagnosis (based on minimum scores in all three domains and onset by 3 years of age). The remaining children met the onset criterion and communication domain score, but missed full diagnosis by one point in one domain (n = 5) or had qualifying scores in one or two domains only (n = 2). Thus, all of the children in the subsample exhibited some autistic behaviors and may have been on the autism spectrum.
Controls were children born during 1987–2002 (the years when air pollution data were available) to mothers who indicated that they never had a child with ASD on the 2005 questionnaire and who responded to a supplemental 2001–2002 questionnaire that queried the calendar year and sex of each of their live births and whether they smoked during pregnancy. We randomly selected one child per mother if more than one child was eligible. Of 25,828 potential controls, 3,711 were excluded because of insufficient address information, and 19 were excluded because their mothers reported that they had ASD on the 2009 questionnaire, leaving 22,098 controls. We did not include the 19 new cases because we did not follow up with mothers to confirm their case status.
The mailing address used for the biennial Nurses' Health Study II questionnaire at the approximate time of the index child's birth was geocoded and classified according to state, county, and census tract identifier. Children born from 1987 through 1990 were assigned the geographic location of their mother in 1989 (the first year of study). Children born in 1991 or 1992 were assigned the mother's mailing address in 1991, and births from 1993 through 2002 were assigned the nurses' addresses, updated every other year, in similar manner.
Hazardous air pollutant (HAP) concentrations were assessed by the U.S. EPA National Air Toxics Assessments in 1990, 1996, 1999, and 2002, which uses an inventory of outdoor sources of air pollution, including both stationary sources (e.g., waste incinerators, small businesses) and mobile sources (e.g., traffic) to estimate average ambient concentrations of pollutants for each census tract based on dispersion models (U.S. EPA 2011). Data were downloaded from the U.S. EPA website on 23 June 2010; additional archived data was received on compact disc from the U.S. EPA. Air pollution concentrations were linked to nurses' residential locations at the time of the birth of their child through census tract codes (Neighborhood Change Database 2013). Children were assigned pollution concentrations from the U.S. EPA assessment closest to their year of birth (births 1987–1993 used 1990 concentrations; births 1994–1997 used 1996 concentrations; births 1998–2000 used 1999 concentrations; births 2001–2002 used 2002 concentrations). We categorized each pollutant according to quintiles of concentration in the entire study population.
We examined family and community socioeconomic indicators that may be associated with ASD ascertainment. To characterize community circumstances around the time by which ASD was likely to have been diagnosed, we used two U.S. Census tract variables (linked by mother's mailing address) measured 6 years after the birth of the child: median income and percent of residents with a college education, which we divided into quartiles. We used the maximum of the mother's parents' education during her infancy as a proxy measure of maternal childhood socioeconomic status. The index child's current family income was based on the family income reported by the mother in 2001. The educational attainment of the mother's partner or spouse was reported in 1999. We also examined factors that may be associated with both ASD and air pollutant exposure: smoking, year of birth, maternal age at birth, and air pollution prediction model year. Smoking during the index pregnancy was assessed in 2001. Year of birth was by mother's report. Maternal age at birth was calculated by subtracting the child's birth year from the mother's birth year. The air pollution prediction model year (HAP year), 1990, 1996, 1999, or 2002, was modeled as a categorical variable.
We examined the association of demographic covariates with ASD case status to assess possible confounding. To calculate odds ratios (ORs) for ASD associated with exposure to specific pollutants, we fit separate logistic regression models with ASD case status as the dependent variable and quintiles of each pollutant as the independent variable, both adjusted for child's sex and stratified by sex. To test a linear dose–response relationship of pollutant exposure with ASD while reducing the influence of outliers, we assigned to each child the median pollutant concentration for his or her quintile and conducted logistic regression with these concentrations entered as a continuous independent variable. To test for sex differences in the association of pollutant quintile with ASD, we multiplied a continuous term for pollutant quintile (1–5) by an indicator of male sex and included this term in models with pollutant quintile, male sex, and demographic covariates. To adjust for multiple tests of significance, we calculated p-values adjusted for false discovery rate using the SAS Multtest procedure (SAS Institute Inc., Cary, NC).
Because individual metal concentrations were moderately or highly correlated in preliminary analyses [Pearson correlation coefficient range, 0.13–0.66; see Supplemental Material, Table S1 (http://dx.doi.org/10.1289/ehp.1206187)], we calculated an overall estimate of association with metal exposure by pooling ORs estimated for individual metals (antimony, arsenic, cadmium, chromium, lead, manganese, mercury, nickel), using a random-effects meta-analysis with the SAS Mixed procedure (Higgins et al. 2009). Additionally, we estimated associations between ASD and an overall measure of metal exposure that was derived by summing the quintile category score (1–5, with 1 representing the lowest quintile) for each metal (antimony, arsenic, cadmium, chromium, lead, manganese, mercury, nickel) to create an overall score with values ranging from 8 to 40.
We examined the association of the covariates with the overall metals metric and conducted additional analyses using this metric, examining the effects of state of residence, family income, smoking during pregnancy, HAP model year and urbanicity on the association between this metals metric and ASD [see Supplemental Material, Methods (http://dx.doi.org/10.1289/ehp.1206187)] All models were adjusted for HAP year, tract median income, tract percent college educated, maternal age at birth, child's year of birth, and maternal parents' education. We did not adjust for family income or spouse/partner's education in main analyses because they were measured after the child's birth for most children, and the child's ASD status may have affected income and educational attainment. Additionally, we did not adjust for smoking during pregnancy in the main analyses because 65 cases were missing smoking data (Table 1).
To investigate further whether one or two pollutants were driving the association between correlated pollutants and ASD, we conducted analyses with diesel, lead, manganese, mercury, methylene chloride, and nickel—the pollutants most strongly associated with ASD based on tests of highest versus lowest quintile as well as linear trend—in a single model.
Methods
Selection of Cases and Controls
We used data from the Nurses' Health Study II, a cohort of 116,430 female nurses from 14 U.S. states that was established in 1989 and has been followed over time with biennial questionnaires. Initially nurses were recruited from 14 U.S. states, but since that time they have moved throughout the United States. Thus, children in the current analyses were born in all 50 U.S. states. In the 2005 questionnaire, respondents were asked "Have any of your children been diagnosed with the following diseases?" with autism, Asperger's syndrome, or other ASD listed as separate responses. In 2007–2008, we sent a questionnaire to the 756 women who had reported having a child with any of these conditions, querying the affected child's sex, birth date, and whether they were adopted. In addition, they were asked "What ASD diagnosis has the child been given?" with autism, Asperger syndrome, and PDD-NOS (PDD not otherwise specified) as possible answers. Women were also asked about other diagnoses (such as attention deficit hyperactivity disorder and obsessive compulsive disorder) (response rate = 84%, n = 636). The Partners Healthcare Institutional Review Board approved this research. Completion and return of questionnaires sent by U.S. mail constituted implied consent.
Cases were excluded for the following overlapping reasons: women reported on the follow-up questionnaire that the child did not have ASD (n = 32); the child was adopted (n = 9); they did not want to participate (n = 20); or they did not report the child's birth year (n = 71). Children reported to have trisomy 18, fragile X, an XXY genotype, or Down's, Angelman's, Jacobsen's, or Rett's syndrome also were excluded (n = 11). Of the remaining children, 329 were born after 1987, when air pollution data were available, but 4 had insufficient address information for geocoding, yielding 325 cases. In this study we refer to children with autism, Asperger's syndrome, or other autism spectrum disorder as cases and use "ASD" to refer to this case definition.
We validated the ASD diagnosis by telephone administration of the Autism Diagnostic Interview–Revised (ADI-R) (Lord et al. 1994) to 50 randomly selected case mothers who indicated willingness to complete the interview (81% of the 636 mothers who responded to the follow up questionnaire were willing to be interviewed). Diagnoses reported by women who were willing versus unwilling to participate in the substudy were similar (25% autism, 51% Asperger's, and 25% PDD-NOS compared with 25% autism, 49% Asperger's, and 23% PDD-NOS, respectively). For the subsample of mothers who completed the ADI-R, 43 children (86%) met full ADI-R criteria for an autism diagnosis (based on minimum scores in all three domains and onset by 3 years of age). The remaining children met the onset criterion and communication domain score, but missed full diagnosis by one point in one domain (n = 5) or had qualifying scores in one or two domains only (n = 2). Thus, all of the children in the subsample exhibited some autistic behaviors and may have been on the autism spectrum.
Controls were children born during 1987–2002 (the years when air pollution data were available) to mothers who indicated that they never had a child with ASD on the 2005 questionnaire and who responded to a supplemental 2001–2002 questionnaire that queried the calendar year and sex of each of their live births and whether they smoked during pregnancy. We randomly selected one child per mother if more than one child was eligible. Of 25,828 potential controls, 3,711 were excluded because of insufficient address information, and 19 were excluded because their mothers reported that they had ASD on the 2009 questionnaire, leaving 22,098 controls. We did not include the 19 new cases because we did not follow up with mothers to confirm their case status.
Geocoding
The mailing address used for the biennial Nurses' Health Study II questionnaire at the approximate time of the index child's birth was geocoded and classified according to state, county, and census tract identifier. Children born from 1987 through 1990 were assigned the geographic location of their mother in 1989 (the first year of study). Children born in 1991 or 1992 were assigned the mother's mailing address in 1991, and births from 1993 through 2002 were assigned the nurses' addresses, updated every other year, in similar manner.
Exposure Assessment
Hazardous air pollutant (HAP) concentrations were assessed by the U.S. EPA National Air Toxics Assessments in 1990, 1996, 1999, and 2002, which uses an inventory of outdoor sources of air pollution, including both stationary sources (e.g., waste incinerators, small businesses) and mobile sources (e.g., traffic) to estimate average ambient concentrations of pollutants for each census tract based on dispersion models (U.S. EPA 2011). Data were downloaded from the U.S. EPA website on 23 June 2010; additional archived data was received on compact disc from the U.S. EPA. Air pollution concentrations were linked to nurses' residential locations at the time of the birth of their child through census tract codes (Neighborhood Change Database 2013). Children were assigned pollution concentrations from the U.S. EPA assessment closest to their year of birth (births 1987–1993 used 1990 concentrations; births 1994–1997 used 1996 concentrations; births 1998–2000 used 1999 concentrations; births 2001–2002 used 2002 concentrations). We categorized each pollutant according to quintiles of concentration in the entire study population.
Covariates
We examined family and community socioeconomic indicators that may be associated with ASD ascertainment. To characterize community circumstances around the time by which ASD was likely to have been diagnosed, we used two U.S. Census tract variables (linked by mother's mailing address) measured 6 years after the birth of the child: median income and percent of residents with a college education, which we divided into quartiles. We used the maximum of the mother's parents' education during her infancy as a proxy measure of maternal childhood socioeconomic status. The index child's current family income was based on the family income reported by the mother in 2001. The educational attainment of the mother's partner or spouse was reported in 1999. We also examined factors that may be associated with both ASD and air pollutant exposure: smoking, year of birth, maternal age at birth, and air pollution prediction model year. Smoking during the index pregnancy was assessed in 2001. Year of birth was by mother's report. Maternal age at birth was calculated by subtracting the child's birth year from the mother's birth year. The air pollution prediction model year (HAP year), 1990, 1996, 1999, or 2002, was modeled as a categorical variable.
Analyses
We examined the association of demographic covariates with ASD case status to assess possible confounding. To calculate odds ratios (ORs) for ASD associated with exposure to specific pollutants, we fit separate logistic regression models with ASD case status as the dependent variable and quintiles of each pollutant as the independent variable, both adjusted for child's sex and stratified by sex. To test a linear dose–response relationship of pollutant exposure with ASD while reducing the influence of outliers, we assigned to each child the median pollutant concentration for his or her quintile and conducted logistic regression with these concentrations entered as a continuous independent variable. To test for sex differences in the association of pollutant quintile with ASD, we multiplied a continuous term for pollutant quintile (1–5) by an indicator of male sex and included this term in models with pollutant quintile, male sex, and demographic covariates. To adjust for multiple tests of significance, we calculated p-values adjusted for false discovery rate using the SAS Multtest procedure (SAS Institute Inc., Cary, NC).
Because individual metal concentrations were moderately or highly correlated in preliminary analyses [Pearson correlation coefficient range, 0.13–0.66; see Supplemental Material, Table S1 (http://dx.doi.org/10.1289/ehp.1206187)], we calculated an overall estimate of association with metal exposure by pooling ORs estimated for individual metals (antimony, arsenic, cadmium, chromium, lead, manganese, mercury, nickel), using a random-effects meta-analysis with the SAS Mixed procedure (Higgins et al. 2009). Additionally, we estimated associations between ASD and an overall measure of metal exposure that was derived by summing the quintile category score (1–5, with 1 representing the lowest quintile) for each metal (antimony, arsenic, cadmium, chromium, lead, manganese, mercury, nickel) to create an overall score with values ranging from 8 to 40.
We examined the association of the covariates with the overall metals metric and conducted additional analyses using this metric, examining the effects of state of residence, family income, smoking during pregnancy, HAP model year and urbanicity on the association between this metals metric and ASD [see Supplemental Material, Methods (http://dx.doi.org/10.1289/ehp.1206187)] All models were adjusted for HAP year, tract median income, tract percent college educated, maternal age at birth, child's year of birth, and maternal parents' education. We did not adjust for family income or spouse/partner's education in main analyses because they were measured after the child's birth for most children, and the child's ASD status may have affected income and educational attainment. Additionally, we did not adjust for smoking during pregnancy in the main analyses because 65 cases were missing smoking data (Table 1).
To investigate further whether one or two pollutants were driving the association between correlated pollutants and ASD, we conducted analyses with diesel, lead, manganese, mercury, methylene chloride, and nickel—the pollutants most strongly associated with ASD based on tests of highest versus lowest quintile as well as linear trend—in a single model.
Source...