Introduction

Autism spectrum disorder

Autism spectrum disorder (ASD) is a developmental disorder characterized by impairments in social interaction, communication, and behavior evident in early development. According to the 2014 surveillance estimate from the Centers for Disease Control and Prevention (CDC), the prevalence of ASD in the USA may be approximately 1 in 45 (Zablotsky et al. 2015). To date, the etiology of ASD has been poorly defined; however, some studies have suggested that ASD may be caused by interactions of susceptible genes with the environment in which environmental triggers may alter gene expression (Volk et al. 2014; Blake et al. 2013; LaSalle 2013; Herbert et al. 2006). Therefore, several investigators have examined the relationships between ASD and exposures to pesticides (Shelton et al. 2012; Roberts et al. 2007), ambient particulate matter (Becerra et al. 2013; Volk et al. 2014; Volk et al. 2011), and heavy metals (Adams et al. 2013; Rahbar et al. 2013; Rahbar et al. 2012; Shelton et al. 2012; Adams et al. 2007; Roberts et al. 2007; Adams et al. 2006; Palmer et al. 2006), but results have been conflicting.

Lead, mercury, and arsenic and neurodevelopment

One hazardous metal of interest is lead (Pb), which can have adverse effects on the health of children, causing behavioral and neurological problems (Ha et al. 2009; Bellinger 2008) and a reduction in IQ scores (Canfield et al. 2003). Numerous studies have investigated the association between ASD and lead exposure (Adams et al. 2006; Adams et al. 2007; Adams et al. 2013; Blaurock-Busch et al. 2012; Fido & Al-Saad 2005; Hertz-Picciotto et al. 2010; Kern et al. 2007; Tian et al. 2011), reporting conflicting results. Mercury (Hg) is a neurotoxic metal found naturally in the earth’s atmosphere in its elemental form, as well as in particulate matter and reactive gaseous compounds, and is the third most frequently released toxic substance from waste facilities, with atmospheric mercury primarily existing in its inorganic form (ATSDR 2007). Results from studies on the associations between mercury and ASD are also conflicting (DeSoto & Hitlan 2007), reporting positive (Fido & Al-Saad 2005; Adams et al. 2007), negative (Holmes et al. 2003), and null (Adams et al. 2006; Adams et al. 2013; Rahbar et al. 2013; Hertz-Picciotto et al. 2010) results in children with ASD in comparison to typically developing (TD) controls. Arsenic, a metalloid that can exist in both organic and inorganic forms, has been found to have harmful effects on the human nervous system (ATSDR 2007; USEPA 2002). Similar to results seen in studies of lead and mercury, results from these studies have been inconsistent, with several reporting no association (Adams et al. 2006; Fido & Al-Saad 2005; Adams et al. 2013; Rahbar et al. 2012; Adams et al. 2013), others reporting higher levels of arsenic in children with ASD (Obrenovich et al. 2011), and another study reporting lower concentrations in children with ASD compared to controls (Kern et al. 2007).

Air pollutants and ASD

In addition to the potential for lead, mercury, and arsenic to influence neurodevelopment and subsequent ASD directly, exposures to ambient air pollutants have also been shown to stimulate inflammation and oxidative stress, which may affect neurologic development (Block & Calderon-Garciduenas 2009; Calderon-Garciduenas et al. 2009). In addition to lead (Jarup 2003; Sanders et al. 2009; Zheng et al. 2003), mercury (Aschner & Aschner 1990; Jarup 2003; Zheng et al. 2003), and arsenic (Jarup 2003) being neurotoxicants that frequently cross the blood-brain barrier affecting neurodevelopment, inflammation from exposure to air pollutants may also contribute to risk of ASD through adverse effects on neurodevelopment (Block & Calderon-Garciduenas 2009; Calderon-Garciduenas et al. 2009; Enstrom et al. 2009; Li et al. 2009). While prenatal exposure to these hazardous metals may be more relevant to the development of autism, assessment of exposure is more frequently conducted at later ages, including time points after the diagnosis has been made. However, an assessment of environmental concentration prenatally is likely to provide stronger evidence of a relationship between heavy metal exposure and ASD than the more frequently used measures in blood, urine, and nails at later ages after the diagnosis of autism has been made (Blanchard et al. 2011). For these reasons, recent studies have investigated the relationship between the geographic distribution of ASD at birth or during childhood and hazardous air pollutants (Becerra et al. 2013; Roberts et al. 2013; Blanchard et al. 2011; Volk et al. 2011; Kalkbrenner et al. 2010; DeSoto 2009; Ming et al. 2008; Windham et al. 2006). Although a recent study of 14 states from the Nurses’ Health Study II reported significantly greater odds of ASD in children born in areas with the highest quintile of ambient lead exposure in comparison to those born in the lowest quintile of exposure (Roberts et al. 2013), a few studies reported no significant association between residential area air concentrations of lead at infancy and early childhood and ASD diagnoses (Windham et al. 2006; Kalkbrenner et al. 2010). Some studies have reported no significant association between ambient mercury compounds and ASD case status (Kalkbrenner et al. 2010), while several others have suggested that risk of ASD is increased with greater ambient air levels of mercury (Blanchard et al. 2011; Palmer et al. 2009; Windham et al. 2006; Roberts et al. 2013), as well as with closer proximity to point sources of mercury (Palmer et al. 2009). When evaluating ambient exposures to arsenic, investigators reported no association between ambient arsenic concentrations and odds of ASD (Windham et al. 2006; Kalkbrenner et al. 2010; Roberts et al. 2013).

For this study, we used an ecological study design and data from the US Environmental Protection Agency’s National-Scale Air Toxics Assessment (US EPA-NATA) to determine if lead, mercury, and arsenic ambient air concentrations in census tracts in 1999 were associated with ASD prevalence among 8-year-old children according to surveillance data from five sites of the Autism and Developmental Disabilities Monitoring (ADDM) Network from 2000 to 2008. Additionally, because ambient toxicant exposures are often co-occurring, this study examined ambient concentrations of each of the aforementioned metals and metalloid, hereafter collectively referred as metals, independently. It also investigated additive and synergistic effects of these metals in combination using this large collection of surveillance data.

Materials and methods

Study sample

The ADDM Network is a multi-state public health surveillance system for ASD and other developmental disabilities established by the CDC in 2000. This system was developed to monitor prevalence trends of ASD and other developmental disabilities in the USA. Records for children with special education classifications and/or relevant diagnostic codes were obtained from school and health sources, including pediatric clinics, hospitals, schools, and diagnostic and treatment centers. ASD status was confirmed through a systematic review of these records by expert ADDM network clinician reviewers using Diagnostic and Statistical Manual of Mental Disorders, 4 th edition, Text Revision (DSM-IV-TR) criteria (American Psychiatric Association 2000). Information on methodology and case ascertainment for this public health surveillance has been previously published (Van Naarden et al. 2007; Rice et al. 2007).

Data for 8-year-old children with a classification of ASD in 2000, 2002, 2004, 2006, and 2008 were obtained from five ADDM surveillance sites including Arizona, Maryland, New Jersey, South Carolina, and Utah. Aggregated data for each census tract were de-identified by each site before provision for this analysis. ADDM sites were included in this analysis if the site investigators expressed interest in the study and were given clearance by their institutional ethics committees to release de-identified data at the tract level. Data were provided for a total of 2558 census tracts; however, considering that the highest estimated prevalence among these sites was 1 in 50 (Blumberg et al. 2013), 69 tracts with populations of less than 40 children were conservatively excluded. These included 17 tracts in which the population of children was rounded to zero. Therefore, 2489 tracts were used for this analysis.

Assessment of exposure variables

The National-Scale Air Toxics Assessment (NATA) is an evaluation of toxic air contaminants developed by the EPA as a screening tool for governing agencies. This assessment uses information about pollutant and emission sources and locations to estimate air quality in various geographic areas using a Gaussian air distribution model. Pollutant sources for these models include mobile sources such as motor vehicles, airplanes, ships, and trains; stationary sources such as industrial facilities, power plants, and gas stations; and indoor sources from activities like renovations and cleaning with chemicals.

For this study, we used 1999 NATA data for air concentrations of inorganic arsenic compounds, lead compounds, and mercury compounds as a proxy for prenatal exposure for children born between 1992 and 2000. We analyzed concentrations only from 1999 under the assumption that relative exposure levels do not change much from year to year. Additionally, Wilcoxon rank-sum tests showed no significant differences in relative air concentration by census tract from 1999 to 2002 or 2002 to 2005. NATA data are based on the Assessment System for Population Exposure Nationwide (ASPEN) model (USEPA 2010), which estimates concentrations of pollutants while accounting for various aspects including but not limited to: location of release, height of release, rate of release, and chemical transformation and deposition rates. Exposure estimates from the model are presented at the state, county, and census tract levels. The average reported annual concentration of each metal (arsenic, lead, and mercury) was downloaded for each census tract and used for analysis.

Statistical analysis

Ambient concentrations of arsenic, lead, and mercury were categorized into quartile cut-points as follows: arsenic Q1 = 0.02 ng/m3, Q2 = 0.06 ng/m3, Q3 = 0.13 ng/m3; lead Q1 = 1.49 ng/m3, Q2 = 2.82 ng/m3, Q3 = 5.90 ng/m3; and mercury Q1 = 1.52 ng/m3, Q2 = 1.55 ng/m3, Q3 = 1.68 ng/m3. To test for additive effects of these metals, we generated a new composite variable by adding the quartile values of each metal (arsenic, lead, and mercury) together, then dividing the sums into quartiles. Because atmospheric concentrations of mercury and lead have decreased dramatically since the 1980s (Butler et al. 2007; Environmental Protection Agency 2015), we split metal concentrations at the 75th percentile mark into binary variables. We also generated an interaction term to assess the possibility of synergistic effects of all three metals.

The number of ASD cases in each census tract was treated as count data for the outcome. After examining the distribution of ASD cases by census tract and determining that the data were over-dispersed (variance > mean), we modeled ASD case counts using negative binomial regression. This model was determined to best fit the data according to the Akaike Information Criterion (AIC). The number of cases for each census tract was entered into a negative binomial regression model as the outcome, with ambient concentrations of metal quartiles as the predictor variables. Because the structure of the data is hierarchical, with census tracts (level 1) nested within counties (level 2), which are further nested into states (level 3), we used multi-level modeling (Bryk & Raudenbush 1992). Due to within and between site differences in census tract population sizes and variations in the total number of years that study sites in the ADDM networks participated in surveillance, we offset each model with the log transformation of an estimated total tract population of children from which cases would have been drawn. This calculation was done by dividing the total tract population of children 0 to 9 years of age, according to the 2000 Census, by ten to account for the number of non-migrant children who were 8 years of age in 2000 or who became 8 years of age in 2002, 2004, 2006, and 2008 (the five surveillance years of data provided for this study). This number was then multiplied by the actual number of surveillance years for each tract as follows: (total population 0 to 9 years of age/10)* number of surveillance years. Separate univariable analyses were used to test the associations between concentrations of each individual metal with prevalence of ASD.

Additionally, we examined potential confounders and effect modifiers including neighborhood SES characteristics such as race (i.e., Black, White, and other) and Hispanic ethnicity distribution, whether the tract was urban or rural, percentage of college-educated residents, percentage of the population below the poverty line, and median household income in the highest quartile. Potential confounders were determined by association with both ASD prevalence, using negative binomial regression, and metal concentrations, using generalized logit models, as evident by p < 0.20 in both models (Szklo & Nieto 2007). Effect modifiers and other potential interactions were determined by a p value of <0.05 for interaction terms in separate models. Final multivariable models were constructed to adjust for potential confounders and stratified to show differences in results based on cut-points at averages of effect modifiers. Variables determined to be highly correlated (i.e., percentage of White residents and percentage of Black residents) were not included as covariates in the same model to prevent multicollinearity. All statistical analyses were done using SAS version 9.3 (SAS Institute Inc. 2011).

Results

Data were provided for a total of 2558 census tracts from five ADDM study sites. This included 644 tracts from Arizona, 631 from Maryland, 588 from New Jersey, 368 from South Carolina, and 327 tracts from Utah. On average, 81.6 % of ASD cases were male, resulting in an approximate male:female ratio of 4:1. Additionally, most (90.3 %) of the census tracts in this analysis were considered urban areas. More information on demographic and descriptive statistics is shown in Table 1.

Table 1 Descriptive statistics indicating characteristics of cases by census tracts and census tract measures included in ADDM Surveillance (n = 2489 tracts)
Full size table

All tract population characteristics were significantly associated with ASD prevalence, as displayed in Table 2. ASD prevalence increased with proportion of White residents (prevalence ratio [PR] for each percentage increase in the proportion of White residents: 1.09, 95 % CI 1.08, 1.11). Additionally, each percentage increase in the proportion of college-educated residents was associated with a 7 % increase in ASD prevalence (PR 1.07, 95 % CI 1.04, 1.10) and median household income within the highest 25th percentile compared to income in the 0 to 75th percentile was associated with a 39 % increase in ASD prevalence (PR 1.39, 95 % CI 1.28, 1.51). Notably, prevalence was lower in areas with a greater proportion of Black residents (PR 0.94, 95 % CI 0.92, 0.96) and other races (PR 0.81, 95 % CI 0.78, 0.84), and Hispanic ethnicity (PR 0.86, 95 % CI 0.84, 0.88). Furthermore, rural geography (PR 0.50, 95 % CI 0.40, 0.63) and proportion of residents below poverty (PR 0.79, 95 % CI 0.75, 0.82) were inversely associated with ASD prevalence.

Table 2 Association between prevalence of ASD and potentially confounding SES factors using multi-level negative binomial regression models
Full size table

In Table 3, we present univariable associations between ambient concentrations of lead, mercury, and arsenic and potential risk factors for exposure. Ambient concentrations of lead, mercury, and arsenic were weakly correlated with each other (all three r 2 < 0.19). When evaluating possible confounding factors, we found that percentages of most tract population factors, modeled as continuous variables, were significantly associated with ambient concentrations of lead, mercury, and arsenic. Specifically, all ambient concentrations of metals of interest decreased for tracts with a greater percentage of White residents (p < 0.01), and median household income in the upper quartile (highest 25th percentile) compared to areas with incomes in the other quartiles (p < 0.01). Tests for trend also indicated a negative trend for proportion of White residents and ambient air concentrations of arsenic, mercury, and summed metal concentrations (p = 0.01). Ambient air pollutant levels increased with increased percentage of residents of other races (p < 0.01), Hispanic ethnicity (p < 0.01), and percent below poverty (p < 0.01). Furthermore, results for the aforementioned trend tests (not displayed in Table 3) also showed a significant positive trend between proportion of Hispanic residents and concentrations of ambient lead and mercury (p = 0.01) as well as percentage of residents living below poverty with all metal concentrations (p = 0.01). In addition, there were significant inverse trends for median household income and all metal concentrations (p < 0.03).

Table 3 Univariable associations between ambient metal concentrations and potential risk factors based on Generalized Logit Models compared to lowest quartile (n = 2489 tracts)
Full size table

In univariable analyses of associations between ambient metal concentrations and ASD prevalence, the observed prevalence was lower for areas with higher concentrations of arsenic, mercury, and summed toxicants and was significantly higher for the 25th to 50th percentile of lead concentrations compared with the lowest 25th percentile. After adjustment for tract population characteristics, including demographics (e.g., percentage of White and Hispanic residents) and SES factors (e.g., percent below poverty and percentage with a college degree), observed ASD prevalence was higher in census tracts with higher ambient lead concentrations. Additionally, results for combined toxicant concentrations showed that in tracts with lower than average percentage of residents below poverty, ASD prevalence was higher in tracts with combined metal concentrations in the 50th to 75th percentile compared to those in the lowest quartile (PR 1.34, 95 % CI 1.12, 1.60). Conversely, the observed prevalence was inconsistent for adjusted analysis of mercury concentrations, controlling for percentage of White and Hispanic residents, percent below poverty, and percentage with a college degree, with increased ASD prevalence in areas with the highest quartile of mercury concentrations but reduced prevalence in the 25th–50th percentile compared to the lowest percentile and significant effect modification by percentage of residents with a college degree (p = 0.02). Table 4 displays other findings from these analyses.

Table 4 Associations between ambient metal concentrations and ASD prevalence based on negative binomial models (n = 2489 tracts)
Full size table

In an effort to assess potential synergistic effects of ambient lead, mercury, and arsenic on ASD prevalence, we compared prevalence of ASD in the presence of eight low/high combinations of lead, mercury, and arsenic. Results in Table 5 show that in univariable analyses, we found no significant associations between synergistically combined toxicant concentrations, as compared to the lowest possible combined concentration of arsenic, mercury, and lead and ASD prevalence. However, after adjusting for tract-level demographic and SES factors modeled as continuous variables, percentage of residents with a college degree, and percent below poverty, we found that areas with a combination of low arsenic and lead concentrations along with high mercury concentrations had a 20 % increased ASD prevalence (PR 1.20, 95 % CI 1.03, 1.40). Additionally, areas with low concentrations of arsenic and high concentrations of both lead and mercury also had a 26 % increased ASD prevalence (PR 1.26, 95 % CI 1.03, 1.54). In tracts with three metal concentrations above the 75th percentile, ASD prevalence was 24 % higher with marginal statistical significance after adjusting for confounding factors, (adjusted PR 1.24, 95 % CI 0.98, 1.58).

Table 5 Interactive effects of ambient metal concentrations on ASD prevalence (n = 2489 tracts)
Full size table

Discussion

For this study, we used surveillance data from three federal agencies, the US Census Bureau, CDC, and EPA. Three prior studies have also used US EPA-NATA data to examine associations between air pollutants and ASD risk in children, including one in San Francisco using California population-based data (Windham et al. 2006), one in North Carolina and West Virginia, which also used surveillance data from two sites of the ADDM study (Kalkbrenner et al. 2010), and another using parent reports from a multi-site study (Roberts et al. 2013). Although our study is not the first multi-state study to explore the relationship between ASD and hazardous air pollutants (Roberts et al. 2013), it is the first to do so using a large number of clinician-reviewed and confirmed surveillance cases from numerous sites.

Ambient lead concentrations and ASD

After adjustment for confounders, results indicated a consistently positive association of 26–36 % increased risk of higher ASD prevalence with increased ambient lead concentrations. We found mixed effects for univariable analysis versus multivariable analysis of arsenic, mercury, and summed metals. Similar to our positive lead results, recent results from the Nurses’ Health Study II reported significantly greater odds of ASD in children born in areas with the highest quintile of ambient lead exposure in comparison to those in the lowest quintile of exposure (OR 1.6, 95 % CI 1.1, 2.3) (Roberts et al. 2013). Conversely, the studies using San Francisco data and data from two sites (North Carolina and West Virginia) found no association between ASD and ambient lead concentrations (Kalkbrenner et al. 2010; Windham et al. 2006). However, the San Francisco study included individuals identified with autism through the California Department of Developmental Services, which may serve a larger portion of children with more severe ASD and various co-existing conditions (Croen et al. 2002; California Department of Developmental Services 2003) where our study used both education and clinical records to confirm ASD status, subsequently ascertaining ASD status for children exhibiting a broader phenotypic array of ASD. The previously mentioned studies also used residential address at birth to measure exposure while tract designation for our study was based on residential address at 8 years old; thus, our results may be biased. For these reasons, we acknowledge that our results may not be entirely comparable to the prior mentioned studies.

Ambient mercury concentrations and ASD

It is surprising that associations based on univariable analysis of different quartiles of ambient mercury levels resulted in 24–37 % reduction in ASD prevalence in areas with mercury concentrations in the 25th to 100th percentile compared to those with the lowest 25th percentile of mercury concentrations. After adjustment for potential confounders, the observed risk of ASD prevalence was 25 % lower for tracts with mercury levels in the 50th to 75th percentile and 13 % higher for tracts with mercury levels in the highest quartile compared to the lowest quartile concentration. The inconsistent results seen in our analyses are complex and highlight the importance of examining other factors associated with exposures to heavy metals. However, the increased prevalence seen in tracts with higher quartiles of mercury versus the lowest after adjustment for confounding factors is in agreement with recently reported results from the Nurses’ Health Study II, which reported odds of parent-reported ASD twice as high for children born in tracts with the highest quintile of airborne mercury (Roberts et al. 2013). Our results are also consistent with those of Windham et al. (2006) who reported increased odds of ASD in children born in neighborhoods with the highest quartile of ambient mercury concentrations after adjustment for maternal age, education, and child’s race (Windham et al. 2006). Kalkbrenner et al. (2010), however, reported no significant associations between ambient mercury concentrations and ASD case status.

We also found evidence of potential synergistic effects of mercury in combination with lead and arsenic on ASD prevalence. Consistent with the aforementioned studies, our study showed no significant associations between ambient arsenic concentrations alone after adjustment for tract-level demographic and SES distributions. Further, the significant influence of high mercury was not apparent in combination with high concentrations of arsenic. Notably, lead concentrations exhibited the most consistent association with ASD prevalence independently. Therefore, although our findings suggest that mercury exposure may be more detrimental in combination with other air pollutants, teasing out the effect of this metal and its complex association with other toxicants will require further investigation to determine if observed results can be replicated in other samples.

Cumulative metal concentrations and ASD

To examine the cumulative effect of all metals of interest, we tested the association of combined metal levels and ASD prevalence. Although in univariable analysis the tracts with the three highest quartiles of combined metals were found to have reduced ASD prevalence compared to tracts with the lowest quartile of exposure, after adjustment for area-based measures of race, ethnicity, and SES factors, tracts with the 50th to 75th percentile of combined metals showed significantly higher prevalence of ASD relative to tracks in the lowest 25th percentile for combined metals. Our results from this analysis are supportive of the positive associations reported by Windham et al. (2006) and Roberts et al. (2013) after adjusting for both individual and tract population characteristics. Both of these studies showed significantly higher odds of ASD in children born in neighborhoods with the highest percentile (quartile and quintile, respectively) concentrations for combined metals (Windham et al. 2006; Roberts et al. 2013). Additionally, although our results for the associations between ambient metal concentrations and ASD prevalence showed no dose-response relationships, results from the Nurses’ Health Study II did report linear relationships between both mercury and lead (Roberts et al. 2013). While other studies have investigated combined effects of multiple toxicant exposures through additive models, our analysis is innovative in that we also investigated synergistic effects using interaction terms. Thus, not only do our results of the additive effects of metals on ASD support those of previous studies, but we strengthened this finding by also demonstrating the possible influence of interactive effects on exposures to multiple toxicants. The observed association with combined metals suggests that co-occurring exposure to toxicants may have a potentially greater effect on ASD prevalence.

SES and ASD prevalence estimates

In addition to the associations seen with ambient lead and mercury concentrations and ASD prevalence, we observed both confounding and significant effect modification by percent below poverty and confounding and effect modification by percentage of college-educated residents for certain metals. The observed effect modifiers are not surprising considering that prior studies have not only reported higher ASD prevalence in areas with higher SES residents (Van Meter et al. 2010; Thomas et al. 2012) but have also reported higher metal exposures in areas with lower SES residents (Mazumdar et al. 2013), which are consistent with the results observed in our analysis. The results of this study build upon those of previous studies on air pollutant exposure and ASD prevalence. Univariable analysis of ambient lead concentrations suggested negative associations with ASD prevalence. Although the relationship between lead exposure and poverty is well established, ASD diagnosis and prevalence has been associated with higher socioeconomic status (SES) in some studies (Bhasin & Schendel 2007; Durkin et al. 2010), while some have shown no association (Pinborough-Zimmerman et al. 2011) and a study from Sweden reported low SES to be associated with increased prevalence of ASD (Rai et al. 2012) Therefore, SES is an important potential confounder to be examined when evaluating the association between lead and ASD.

Limitations

There are several limitations to our analysis. To prevent identification of ASD cases in tracts with small numbers, aggregate data for children aged 8 in 2000, 2002, 2004, 2006, and 2008 were used. Therefore, we could not test for trends in tract prevalence and air pollutant concentrations. As mentioned, assigned tracts for cases were based on residency during surveillance study year at age 8 rather than residency at the time of birth, thus analysis did not account for mobility from the original birth place, individual maternal exposures, or postnatal exposures of the child. Although this likely resulted in some misclassification, model adjustment for residence in the same birth county had no significant effect on results. We also acknowledge that our methods for estimating the total number of 8-year-old children in each tract in biannual years from 2000 to 2008 assume uniform age distributions for each census tract and that this assumption may not be true and might present biased prevalence estimates. This method also does not account for migration in and out of the census tracts in the years following the decennial census in 2000. Furthermore, we excluded 69 tracts with extremely low population sizes due to the rounding of the total number of children in each tract to the nearest 5. However, it should be noted that a sensitivity analysis evaluating the impact of removing these 69 tracts showed that there was no significant influence of these excluded tracts on the analysis.

Regarding limitations of exposure assessment, US EPA-NATA data were only available for 1996, 1999, 2002, and 2005. Because children ascertained from 2000 to 2008 would have been born between 1992 and 2000, we used the reported 1999 NATA estimates as a proxy for prenatal exposures. We did not use 1996 NATA measures, as air pollutant concentrations were reported as underestimated due to modeling based on reported emissions from major and small stationary sources (i.e., EPA Toxics Release Inventory sites) and estimated mobile source exclusive of air monitoring data from federal, state, and local agencies, where 1999 NATA models were inclusive of monitored air quality data from federal, state, and local monitors. However, using Wilcoxon rank-sum tests, we confirmed that tracts with high ambient concentrations of lead, mercury, and arsenic maintained these high relative concentrations from 1999 to 2002 and from 2002 to 2005. This suggests that the air concentrations in census tracts were fairly consistent from 1999 to 2005 and that it is unlikely that we would have found a trend if provided longitudinal data. Additionally, we acknowledge that by categorizing the exposure concentrations into quartiles, we reduced statistical power in our analysis; however, we determined that the sample size of 2489 census tracts provided enough power to sufficiently perform our analyses. We were also unable to assess individual indoor exposures to metals (i.e., household dust), which influences exposure risk based on individual behaviors (Rasmussen et al. 2013). However, previous sources have indicated that air concentrations serve as logical estimates of metal exposures (Payne-Sturges et al. 2004). We also did not have data on maternal occupational exposures, age of homes, nutritional intake, or other possible confounding factors associated with exposures to lead, mercury, and arsenic on the individual level. Previous studies have reported that predicted chronic exposures to air pollutants may be at least 20 % lower than concentrations reported by outdoor air monitors (Ozkaynak et al. 2008), and we recognize that although we only evaluated one potential route of exposure for metals, lead, mercury, and arsenic can be present in other sources including water, food, and various household products. Other co-occurring toxicant exposures, such as pesticides, phthalates, and other metals, may have also influenced our results. The aforementioned exposure limitations could have led to exposure misclassification for tracts, which may also have led to greater misclassification of both additive and interactive results for combined metals.

We adjusted for area-based measures such as demographic characteristics, including percent of White residents and percent with at least a college degree within each tract; however, we are aware that diagnosis and reported ASD prevalence is greater in higher SES areas (Mazumdar et al. 2013), while air concentrations of metals are often greater in lower SES areas (Evans & Kantrowitz 2002). Thus, lower reported prevalence in areas where children possibly have greater exposures to ambient metal concentrations may bias results towards the null. We did not adjust for urbanicity, which may be indicative of greater exposure to air pollutants as well as access to ASD service, as it may have biased our results by possibly being a collider in this relationship. Moreover, due to the age of ascertainment (8 years of age), we cannot confidently determine the temporality of the associations between SES confounding factors (i.e., proportion below poverty and median household income) and ASD status, as caring for a child with ASD may have negatively affected income.

Despite its limitations, our study also has several notable strengths. First, we used reliable surveillance data that were ascertained from multiple sources and that determined ASD status via record review by knowledgeable clinician reviewers. Furthermore, we utilized a large set of data from multiple states. Additionally, assessment of air exposures prior to ASD diagnosis may provide more temporally significant information in comparison to levels in biomarkers obtained at or after diagnosis. While the measurements used in this study were only proxies for potential prenatal exposures, the critical window of exposures for ASD, pre-conception, prenatal, infancy, or early childhood has not yet been defined. Furthermore, studies of individual health factors only may miss important group-level disease determinants including neighborhood characteristics (Diez Roux 2004). Additionally, differences in area-based measures, such as air concentrations within a neighborhood, can be better explained in clusters, as there is no assumption of independence of effects for individuals. The interaction and effect modification observed in our analysis demonstrates the importance of evaluating neighborhood characteristics when assessing area-based measures. More notable is our new findings of synergistic effects of combined ambient metal concentrations. Although our results are consistent with previous studies, they reveal the potential for more influential effect of ambient metal concentrations in combination with each other.

Conclusions

The role of environmental metals in the etiology of ASD has yet to be determined. Many environmental exposures are correlated, and to our knowledge, we are the first to publish results that suggest that synergistic exposure to multiple metals may increase ASD risk through interaction. Specifically, these synergistic effects appear to be highly influenced by the presence of mercury and the absence of arsenic. Notably, although the concentrations of lead and mercury have declined during recent decades, the reported prevalence of ASD has increased in the USA. Therefore, more research should focus on the synergistic effects of combined metals with particular focus on mercury in combination with other heavy metals and toxicants. Furthermore, the strong associations seen with ambient lead concentrations and ASD prevalence suggest future studies should also investigate major sources of lead exposure via outdoor air and subsequent neurodevelopmental outcomes in children residing in areas with frequently and consistently high lead concentrations. Additionally, future research should be done using a large cohort to gather prospective data from mothers during pregnancy, including information on her occupation, SES before and after ASD diagnosis, nutritional intakes, and smoking status, as well as in-home exposures as measured by household dust, water, and soil samples. Data should also be collected from children in early infancy and childhood, including cord-blood and follow-up blood metal concentrations as well as nutritional intake, exposures to second hand smoke, and household dust composition, to better understand the role of metal exposure in the etiology of ASD and to ascertain the critical window of exposure. Inclusion of genetic data in future studies could also contribute to eventually determining the multitude of factors contributing to this highly complicated developmental disorder.