Invited Perspective: Air Pollution and Autism Spectrum Disorder: Are We There Yet?

Autism spectrum disorder (ASD) is a heterogeneous developmental condition with rapidly increasing incidence and prevalence worldwide.¹ This increase could be explained by changes in diagnostic criteria, improved awareness, and reduced stigmatization.^2,3,4,5 Nevertheless, there are tantalizing clues that more than improved recognition and diagnosis may be involved here. Heritability studies strongly suggest that environmental agents are involved in ASD, with and without interactions with genetic factors.^6,7 Indeed, associations of environmental exposures with ASD are often detected in epidemiological studies, and accumulated evidence supports a more decisive contribution of environmental factors than had been assumed.⁸

Initial evidence linking ASD with air pollutants was observed in the United States with an ecological design.⁹ Later studies based their exposure assessment on mandatory emissions reporting,^10,11 and their findings led to even more interest in and studies of the issue. Subsequent studies used more sophisticated methods, including land-use regression or various spatiotemporal models, to assess ambient air pollutants such as particulate matter (PM) and nitrogen oxides.^{12,13,14,15,16,17,18,19,20,21,22,23,24,25,26} Combining this approach with medical administrative data sets enabled researchers to examine the relationships in large populations, but these investigations were usually limited to criteria pollutants. More than a dozen such studies demonstrated associations of criteria pollutants with ASD. Although most studies reported positive associations, a pooled analysis of four European cohorts did not find an association between criteria pollutants and autistic traits,²⁵ and a large Swedish study did not find any associations between criteria pollutants and ASD.²⁴

In their study published in this issue, Rahman et al.²⁷ used a historical population-based cohort to analyze data for almost 300,000 children, including 5,694 diagnosed with ASD, from Kaiser Permanente Southern California hospitals. Using nonlinear distributed lag models, they found that the association of PM with an aerodynamic diameter of $\le 2.5\mathrm{\mu }\mathrm{m}$ (i.e., ${\mathrm{PM}}_{2.5}$) with ASD was strongest during the earliest stages of gestation, decreasing during the last weeks of the pregnancy. Prenatal nitrogen dioxide exposure was not associated with ASD. Ozone exposure at around the end of the second trimester was negatively associated with ASD, whereas exposures in the last weeks of gestation were positively associated with the disorder.

The study makes an essential contribution to the field because it is based on a large population and applies an appropriately sophisticated analytical approach to detect critical windows of susceptibility. On the other hand, its detailed results are inconsistent with most previous studies, especially regarding the heightened vulnerability during early pregnancy and the absence of associations with exposures in later trimesters.^17,18,23

The relationship between air pollutants and ASD presents inherent research challenges that may account for some inconsistency among studies. First, ASD presents a wide range of phenotypes, and it is defined operatively by a set of behaviors resulting from different and still poorly understood biological processes in the brain. Second, criteria pollutants may be only a proxy for the actual agents that promote pathological brain processes. This is especially true for nitrogen oxides and PM measured in mass per volume units, ignoring its sources and chemical components that vary among studies given the different sources. Third, the developmental window for the involvement of the pollutant in the pathological processes is not known. Fourth, ASD is typically diagnosed only several years after birth, long after the exposure occurred. Given the low prevalence of ASD, prospective cohort studies are generally infeasible except in particular cohorts with a high risk of ASD that would lack generalizability.²⁸

In addition, as in any epidemiological study, there are concerns of bias and confounding. For example, live birth bias, which is very difficult to estimate or mitigate,^29,30 may account for the negative association that Rahman et al.²⁷ found with ozone during the end of the second trimester. Possible confounding structures, such as the possible association between spatiotemporally related diagnostic awareness/health care access and spatiotemporally related air pollution levels, may also limit the ability to make causal inferences about this issue. Indeed, mutual adjustment for exposures during several time points would strengthen the causal interpretation.³¹ Finally, animal studies provide relevant findings for possible mechanisms,³² but they can bridge only part of the gap because ASD is not a condition seen in animals. Therefore, despite many positive associations, the relationship with ASD is still not considered in policy discussions of air pollution.

What else can be done to increase the precision of the findings and to produce information that is more directly actionable? First, causal inference can be strengthened by applying methods to assess residual confounding, such as the use of instrumental variables, negative control outcomes, or negative control exposures.³³ In this context, quasi-experimental designs can exploit policy interventions to improve causal estimates.³⁴ Second, other developmental windows of susceptibility should also be examined, as much as statistical power permits, because several studies have already demonstrated associations of early postnatal exposures to air pollutants with ASD, even after adjustment to the prenatal exposure.^17,18,33 Third, reproducing the findings in different settings, where other bias and confounding structures exist, will help clarify whether the association we observe in most studies is causal, even if we cannot estimate its magnitude accurately. For instance, many cities (such as Rome, Italy) have an opposite confounding structure regarding socioeconomic status than the one seen in the United States.³⁵ Finally, promising discoveries of very early ASD biomarkers based on functional brain connectivity and cortical surface area^36,37,38 or metabolic abnormalities³⁹ may enable the study of ASD in prospective cohorts.

There obviously cannot be a randomized controlled trial to prove that air pollution increases the risk for ASD, nor can we expect a perfect animal experiment showing the mechanistic details of such an effect. However, the salience of autism is apparent, and its impact on our society is considerable. The study by Rahman et al.²⁷ is notable for its robust methods, which enabled them to identify modifiable environmental factors, critical developmental windows, and susceptible populations. Accordingly, this paper adds to evidence that can support policymaking. Sometimes, the imperative to act, in this case to protect children, cannot await a complete understanding of the exact details of how an environmental agent causes harm.

Refers to https://doi.org/10.1289/EHP9509