University of Southern California Dissertations and Theses

Susceptibility to autism likelihood associated with prenatal ambient air pollution exposures: identifying vulnerable populations

(USC Thesis Other)

Susceptibility to autism likelihood associated with prenatal ambient air pollution exposures: identifying vulnerable populations

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Susceptibility to Autism Likelihood Associated with Prenatal Ambient Air Pollution Exposures: Identifying Vulnerable Populations by Xin Yu A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (POPULATION, HEALTH AND PLACE) May 2024 Copyright © 2024 Xin Yu ii Dedication To my beloved family, and the autism community iii Acknowledgements I begin by extending my deepest gratitude to my co-advisors, Dr. Rob McConnell and Dr. Daniel Hackman. Thank you for your guidance, patience, and invaluable support, and for allowing me the freedom to translate my fanciful ideas into practical academic outcomes. Your time and efforts devoted to helping me achieve success during my PhD journey are greatly appreciated. I also wish to express my gratitude to my esteemed committee members: Dr. Anny Xiang, Dr. Darren Ruddell, and Dr. Jonathan Tarbox. Dr. Xiang, thank you for your support regarding the study population and datasets, as well as your insightful suggestions to strengthen the research project. Dr. Ruddell, your patience with my questions, no matter how trivial, has been invaluable. And Dr. Tarbox, thank you for introducing me to the concept of neurodiversity; let's continue working together to support the neurodiversity community. My thanks also extend to all collaborators on the APAR project: Dr. Md Mostafijur Rahman, Frederick Lurmann, Dr. Joel Schwartz, Dr. Mike Kleeman, Dr. Sandrah Eckel, Dr. Zhanghua Chen, and Dr. Jiu-Chiuan Chen, as well as the Kaiser Permanente research team (Mayra Martinez, Jane Lin, Ting Chow, and Ziming Zhuang). Thank you for your supports on this project and all the constructive advice on my research topics. A special acknowledgment goes to Dr. Erika Garcia for her captivating course on causal inference, which broadened my horizons and shaped my third empirical study. I also appreciate the Northwestern team (Dr. William Funk, Dr. Yeunook Bae, Dr. Fariba Tayyari, and Cathelin Huang) for providing an opportunity to explore biomarkers, laying the groundwork for my future research interests. Furthermore, I'd like to express gratitude to the Spatial Science Institute for the opportunity to be part of this exciting PhD program and all the supports for the smooth and productive PhD life as an international student. Thanks to the faculty and staff for their iv assistance over the years. To my cohort friends (Dr. Michelle Livings, Dr. Yingyi Lin, Mengya Xu, Leo Lerner, and Rachel Wilkie), I miss our in-class discussions and the stimulating ideas we shared, not to mention the gossip. And to the previous PHP program cohort, thank you for your guidance. Above all, I am deeply thankful to my supportive and beloved parents, Ping Long and Xiao Yu, for being my strongest support and the sanctuary of my soul. The COVID period was exceptionally challenging for my family in Wuhan, China, and I am grateful for your strength, bravery, and unwavering support. I miss you and hope for a reunion soon. To my wonderful and loving boyfriend, Linghao Zhao (future Dr. Zhao), thank you for your patience, love, and kindness. Thank you for tolerating all my little quirks. Thank you for believing in me and my work. I love you with my whole heart! And to my three adorable guinea pigs (San-San, Tu-Tu, and Oreo), your companionship is cherished. Treat is ready for you. v Table of Contents Dedication....................................................................................................................................... ii Acknowledgements........................................................................................................................iii List of Tables ...............................................................................................................................viii List of Figures................................................................................................................................. x Abbreviations................................................................................................................................ xii Abstract........................................................................................................................................ xiv Chapter 1 Introduction .................................................................................................................... 1 1.1. Risk factors of ASD in literature ........................................................................................1 1.2. Conceptual models of risk factors on ASD.........................................................................6 1.3. Challenges of examining the interactive effects of multiple co-existing factors..............10 1.4. Study settings....................................................................................................................11 1.5. Research outline................................................................................................................13 Chapter 2 Evidence of Susceptibility to Autism Risks Associated with Early Life Ambient Air pollution: A Systematic Review............................................................................................ 22 2.1. Abstract.............................................................................................................................22 2.2. Introduction.......................................................................................................................23 2.3. Methods.............................................................................................................................26 2.4. Results...............................................................................................................................29 2.5. Discussion.........................................................................................................................35 2.6. Conclusions.......................................................................................................................42 2.7. Tables................................................................................................................................43 2.8. Figures...............................................................................................................................51 2.9. Supplemental Material......................................................................................................52 vi Chapter 3 Prenatal Air Pollution, Maternal Immune Activation, and Autism Spectrum Disorder ............................................................................................................................................... 66 3.1. Abstract.............................................................................................................................66 3.2. Introduction.......................................................................................................................67 3.3. Methods.............................................................................................................................70 3.4. Results...............................................................................................................................74 3.5. Discussion.........................................................................................................................76 3.6. Conclusion ........................................................................................................................79 3.7. Tables................................................................................................................................80 3.8. Figures...............................................................................................................................83 3.9. Supplemental Material......................................................................................................84 Chapter 4 Neighborhood Disadvantage and Autism Spectrum Disorder in a Population with Health Insurance.................................................................................................................... 93 4.1. Abstract.............................................................................................................................93 4.2. Introduction.......................................................................................................................95 4.3. Methods.............................................................................................................................96 4.4. Results...............................................................................................................................97 4.5. Discussion.........................................................................................................................98 4.6. Conclusion ........................................................................................................................99 4.7. Tables..............................................................................................................................101 4.8. Figures.............................................................................................................................103 4.9. Supplemental Material....................................................................................................104 Chapter 5 The Potential Effects of Hypothetical PM2.5 Interventions on Childhood Autism in Different Neighborhood Socioeconomic Contexts ............................................................. 117 5.1. Abstract...........................................................................................................................117 5.2. Introduction.....................................................................................................................118 vii 5.3. Methods...........................................................................................................................120 5.4. Results.............................................................................................................................126 5.5. Discussion.......................................................................................................................128 5.6. Conclusion ......................................................................................................................133 5.7. Tables..............................................................................................................................134 5.8. Figures.............................................................................................................................136 5.9. Supplemental Material....................................................................................................139 Chapter 6 Summary and Suggestions for Future Research ........................................................ 160 6.1. Summary of Findings......................................................................................................161 6.2. Contribution to Knowledge.............................................................................................162 6.3. Limitations......................................................................................................................165 6.4. Suggestions for Future Research ....................................................................................169 References................................................................................................................................... 173 viii List of Tables Table 2.1 Modification of regulated air pollutant (PM2.5, PM10, NO, NO2, SO2) associations with ASD by sex.................................................................................................................... 43 Table 2.2 Modification of air pollutant (PM2.5, PM10, NO, NO2, CO, O3, traffic-related pollutants, airborne mercury) exposure associations with ASD by education level ............. 45 Table 2.3 Modification of air pollutant (PM2.5, PM10, NO, NO2) exposure associations with ASD by neighborhood deprivation level............................................................................... 47 Table 2.4 Modification of air pollutant (PM2.5,PM10, NO2, O3, near-roadway pollution) exposure associations with ASD by maternal factors ........................................................... 48 Table 2.5 Modification of regulated air pollutant (PM2.5,PM10, NO2, O3, traffic-related pollution) exposure effects of ASD by children’s genotypes................................................ 50 Table 3.1 Characteristics of children, with and without autism spectrum disorder (ASD).......... 80 Table 3.2 Adjusted hazard ratios and 95% confidence intervals for ASD associated with each air pollutant during pregnancy (in single pollutant models) and with each MIA (in single condition models)........................................................................................................ 81 Table 3.3 The interaction between MIA-related conditions and pregnancy PM2.5 on risk of ASD in children..................................................................................................................... 82 Table 4.1 Characteristics of children, with and without autism spectrum disorder (ASD)........ 101 Table 4.2 Associations between neighborhood disadvantage and ASD diagnosis in models adjusting for varying sets of covariates............................................................................... 102 Table 5.1 Characteristics of mother-child pairs.......................................................................... 134 Table 5.2 Estimated effects of hypothetical PM2.5 interventions on the risk of ASD cases by age 5..................................................................................................................................... 135 Supplemental Table 2.1 PubMed search syntax ........................................................................... 52 Supplemental Table 2.2 Embase search syntax ............................................................................ 52 Supplemental Table 2.3 Detailed Criteria for the qualitative assessment of risk of bias ............. 53 Supplemental Table 2.4 An example of comparing two estimated hazard ratios......................... 53 Supplemental Table 2.5 Excluded Publications on air pollution and autism due to the lack of any effect modifiers............................................................................................................... 54 Supplemental Table 2.6 Characteristics of included studies, grouped by study design and alphabetically ordered by author surname for each design ................................................... 55 Supplemental Table 2.7 Assessment of risk of bias ..................................................................... 63 Supplemental Table 2.8 Modification of air toxics associations with ASD by sex...................... 64 Supplemental Table 3.1 MIA and ICD-9 codes............................................................................ 84 ix Supplemental Table 3.2 Adjusted HRs and 95% CIs of ASD risk for MIA and PM2.5 in a coadjusted model....................................................................................................................... 86 Supplemental Table 3.3 Adjusted hazard ratios and 95% confidence intervals for ASD associated with each air pollutant during pregnancy (in single pollutant models) and with each MIA (in single condition models) including all follow up until December 31, 2019 without censoring at age 5 (a total of 6,366 ASD cases).............................................. 86 Supplemental Table 3.4 The interaction between MIA-related conditions and pregnancy PM2.5 on risk of ASD including all follow up until December 31, 2019 without censoring at age 5 (a total of 6,366 ASD cases).................................................................... 87 Supplemental Table 4.1 ICD codes and diagnostic description for ASD ascertainment............ 108 Supplemental Table 4.2 Factor loadings of the primary neighborhood disadvantage index and indices of neighborhood deprivation and advantage .................................................... 108 Supplemental Table 4.3 Crude associations between neighborhood disadvantage index and ASD..................................................................................................................................... 109 Supplemental Table 4.4 Associations between neighborhood disadvantage components, neighborhood disadvantage and advantage indices, and ASD in models adjusting for varying sets of covariates .................................................................................................... 109 Supplemental Table 4.5 Multivariable-adjusted associations between race and ethnicity, maternal education, and ASD.............................................................................................. 110 Supplemental Table 4.6 Stratified analyses: Associations between neighborhood characteristics and ASD by maternal race and ethnicity..................................................... 111 Supplemental Table 4.7 Sensitivity analyses: Adjusted associationsa between neighborhood disadvantage index and ASD............................................................................................... 112 Supplemental Table 4.8 Associations between neighborhood disadvantage index and ASD by race and ethnicity based on the interaction model.......................................................... 112 Supplemental Table 5.1 ICD codes and diagnostic description for ASD ascertainment............ 140 Supplemental Table 5.2 Loadings of the neighborhood disadvantage index ............................. 140 Supplemental Table 5.3 Outcome Cox Regression Model Selection......................................... 141 Supplemental Table 5.3 Characteristics by neighborhood disadvantage index deciles.............. 142 Supplemental Table 5.5 Estimated effects of hypothetical PM2.5 interventions on the risk of ASD cases by age 5 based on the parsimonious linear Cox regression model ................... 144 Supplemental Table 5.6 Estimated effects of hypothetical PM2.5 interventions on the risk of ASD cases by age 5 based on non-movers.......................................................................... 145 Supplemental Table 5.7 Estimated effects of hypothetical PM2.5 interventions on the risk of ASD cases by age 5 based Cox regression models with different baseline hazard functions by neighborhood disadvantage deciles................................................................ 146 Supplemental Table 5.8 Estimated effects of hypothetical PM2.5 interventions on the risk of ASD cases by age 5 with confidence intervals based on family-clustered bootstrapping .. 147 x List of Figures Figure 1.1 Conceptual model of risk factors of ASD and potential pathways ............................. 16 Figure 1.2 Susceptibility model of environmental pollution-associated ASD likelihood ............ 16 Figure 1.3 Map of Kaiser Permanente Southern California study catchment area....................... 17 Figure 1.4 Cumulative incidence of ASD by age 5 in the study cohort by census tracts............. 18 Figure 1.5 Distribution of air pollutant concentrations during pregnancy across birth year 2001-2014.............................................................................................................................. 19 Figure 1.6 Neighborhood disadvantage index in Southern California by census tracts............... 20 Figure 1.7 Study aims of this dissertation..................................................................................... 21 Figure 2.1 Screening of studies for inclusion based on PRISMA criteria .................................... 51 Figure 3.1 ASD risk associations with interactions between MIA indicators and the concentration of PM2.5 components during pregnancy.......................................................... 83 Figure 4.1 Associations between neighborhood disadvantage, maternal race and ethnicity and ASD diagnosis.............................................................................................................. 103 Figure 5.1 Distribution of pregnancy average PM2.5 concentrations of hypothetical interventions by neighborhood disadvantage deciles.......................................................... 136 Figure 5.2 Estimated effects of hypothetical PM2.5 interventions (percentage reduction) on the cumulative incidence of ASD children by neighborhood disadvantage deciles ........... 137 Figure 5.3 Estimated effects of hypothetical PM2.5 interventions (standard thresholds) on the cumulative incidence of ASD by neighborhood deprivation deciles.................................. 138 Supplemental Figure 3.1 Derivation of study sample................................................................... 88 Supplemental Figure 3.2 Distribution of air pollutant concentrations during pregnancy across birth year 2001-2014 .................................................................................................. 89 Supplemental Figure 3.3 ASD risk associations with interactions between MIA indicators and the concentration of PM2.5 components during the first trimester .................................. 90 Supplemental Figure 3.4 ASD risk associations with interactions between MIA indicators and the concentration of PM2.5 components during the second trimester.............................. 91 Supplemental Figure 3.5 ASD risk associations with interactions between MIA indicators and the concentration of PM2.5 components during the third trimester................................. 92 Supplemental Figure 4.1 Derivation of study sample................................................................. 113 Supplemental Figure 4.2 The distribution of neighborhood SES measures from the three census sources among all census tracts in Southern California .......................................... 114 Supplemental Figure 4.3 The distribution of neighborhood disadvantage index by maternal and children demographic characteristics ........................................................................... 115 xi Supplemental Figure 4.4 The main and interactive associations between dichotomized neighborhood disadvantage (quintiles 4-5 vs quintiles 1-3), maternal race and ethnicity and ASD .............................................................................................................................. 116 Supplemental Figure 5.1 Derivation of study population........................................................... 148 Supplemental Figure 5.2 DAG ................................................................................................... 149 Supplemental Figure 5.3 The splines of pregnancy PM2.5 and neighborhood disadvantage, and their tensor interaction in the best-fitted outcome model ............................................. 150 Supplemental Figure 5.4 Estimated effects of hypothetical PM2.5 interventions (percentage reduction) on the cumulative incidence of ASD by neighborhood deprivation deciles based the parsimonious outcome model.............................................................................. 151 Supplemental Figure 5.5 Estimated effects of hypothetical PM2.5 interventions (standard thresholds) on the cumulative incidence of ASD by neighborhood deprivation deciles based the parsimonious outcome model.............................................................................. 152 Supplemental Figure 5.6 Estimated effects of hypothetical PM2.5 interventions (percentage reduction) on the cumulative incidence of ASD by neighborhood deprivation deciles based non-movers................................................................................................................ 153 Supplemental Figure 5.7 Estimated effects of hypothetical PM2.5 interventions (standard thresholds) on the cumulative incidence of ASD by neighborhood deprivation deciles based non-movers................................................................................................................ 154 Supplemental Figure 5.8 Estimated effects of hypothetical PM2.5 interventions (percentage reduction) on the cumulative incidence of ASD by neighborhood deprivation deciles based on Cox regression models with different baseline hazard functions by neighborhood disadvantage deciles..................................................................................... 155 Supplemental Figure 5.9 Estimated effects of hypothetical PM2.5 interventions (standard thresholds) on the cumulative incidence of ASD by neighborhood deprivation deciles based on Cox regression models with different baseline hazard functions by neighborhood disadvantage deciles..................................................................................... 156 Supplemental Figure 5.10 Estimated effects of hypothetical PM2.5 interventions (percentage reduction) on the cumulative incidence of ASD by neighborhood deprivation deciles with confidence intervals based on family-clustered bootstrapping ................................... 157 Supplemental Figure 5.11 Estimated effects of hypothetical PM2.5 interventions (standard thresholds) on the cumulative incidence of ASD by neighborhood deprivation deciles with confidence intervals based on family-clustered bootstrapping ................................... 158 Supplemental Figure 5.12 Comparison of the predicted cumulative incidence of ASD at age 5 per 10,000 children between G-computation and Kaplan-Meier estimates..................... 159 xii Abbreviations ACS American Community Survey AIC Akaike information criterion ASD Autism Spectrum Disorder BC Black Carbon CI Confidence Interval EMM Effect Measure Modification (Modifier) EMR Electronic Medical Record HR Hazard Ratio ICD International Classification of Diseases IQR Interquartile Range KPSC Kaiser Permanente Southern California MIA Maternal Immune Activation NO2 Nitrogen Dioxide NO3 - Nitrate O3 Ozone OM Organic Matter OR Odds Ratio PM Particulate Matters PM10 Particulate matter with aerodynamic diameter < 10 µm PM2.5 Particulate matter with aerodynamic diameter < 2.5 µm RERI Relative Excess Risk due to Interaction SES Socioeconomic Status xiii SO4 2- Sulfate xiv Abstract Autism spectrum disorder (ASD) is a complex set of developmental disorders characterized by deficits in social interactions and communication and the presence of restricted, repetitive, and stereotyped patterns of behaviors. The estimated prevalence of ASD in the United States increased from 0.66% in 2002 to 2.78% in 2020. The etiology of autism is unclear but likely to be multifactorial. While research has emphasized a strong genetic component in ASD, accumulative evidence suggests that environmental pollution, maternal health conditions during pregnancy, as well as social context or chronic social stress may also alter neurodevelopment. Challenges remain to explore the interactive or joint effects of these co-existing environmental factors. Given the potentially shared pathways such as oxidative stress and neuroinflammation among these co-occurring factors, the “multiple- exposure” hypothesis of ASD deserves further research. In this dissertation, I conducted one systematic review and three quantitative epidemiological studies based on a large retrospective birth cohort born in Southern California Kaiser Permanente between 2001 and 2014. In the review article (Chapter 2), child’s sex, maternal nutrition or diabetes, socioeconomic factors, and child risk genotypes have been reported to modify the effect of early-life air pollutants on ASD likelihood in the epidemiologic literature. However, the sparsity of studies on corresponding modifying hypotheses precludes conclusive findings. In the first empirical study (Chapter 3), I hypothesized that co-exposure of prenatal PM2.5 or PM2.5 components and maternal immune activation (MIA)-related conditions may have interactive association on the subsequent likelihood of ASD. However, the empirical evidence did not support this hypothesis, as no statistically significant interactions between MIA conditions and prenatal PM2.5 exposure with ASD likelihood were found. In the second empirical xv study (Chapter 4), I found that children residing in more disadvantaged neighborhoods at birth had a higher likelihood of ASD diagnosis among a population with health insurance and thus relatively free of bias in ascertainment due to limited access to care. In the third study (Chapter 5), I applied an innovative window-based G-computation approach to quantify the changes in cumulative incidence of ASD under hypothetical PM2.5 interventions in the entire cohort and by neighborhood disadvantage deciles. I found that all communities would have similar benefit from hypothetical PM2.5 interventions regardless of neighborhood disadvantage levels. These results warrant replication study on other population from different regions. Although the studies did not find interactive associations between prenatal air pollution and MIA or prenatal air pollution and neighborhood disadvantage on ASD likelihood, the “multipleexposure” framework may nevertheless be useful for future research in understanding the complex interplay between other risk factors for autism, for example genetic–environmental interactions or nutrition-related susceptibility. Further investigation is needed to identify the potential biological pathways for the co-existing factors on ASD likelihood and to provide more plausible hypotheses on the interactive associations. 1 Chapter 1 Introduction Autism spectrum disorder (ASD) is a complex set of developmental disorders characterized by deficits in social interactions and communication and the presence of restricted, repetitive, and stereotyped patterns of behaviors (American Psychiatric Association 2013b).. The estimated prevalence of ASD in the United States increased from 0.66% in 2002 to 2.78% in 2020 (Centers for Disease Control and Prevention 2007; Maenner et al. 2023). This upward trend in prevalence can be attributed in part to the expanded definition of autism as a spectrum, which now includes milder cases since 2008. Additionally, more extensive screening of children has played a role in this increase (Hertz-Picciotto and Delwiche 2009; Blumberg et al. 2013). The noticeable skewed sex ratio of autism has been recognized since the first report of autism in the 1960s (Kanner 1968) with boys being four times more likely to receive an autism diagnosis than girls (Maenner et al. 2023). ASD imposes lifetime social and emotional hardship on children (Kuhlthau et al. 2010) and their families (Rao and Beidel 2009), as well as economic burdens on society (Buescher et al. 2014). 1.1. Risk factors of ASD in literature The etiology of autism is unclear but likely to be multifactorial. No single causes or biomarkers of ASD have been identified, but rather, the combination of genetic, maternal physical health, environmental, and social factors may form the landscape of autism etiology. The intricate interaction between these various factors is under the ongoing scientific investigation. 2 1.1.1. Genetics The estimated heritability of ASD in twin studies, ranging from 50% to 90%, suggests a genetic component in the origin of ASD (Bai et al. 2019; Tick et al. 2016). Various genes associated with ASD have been identified, including those arising from spontaneous de novo mutations as well as inherited variations from parents. Children with ASD have been reported to have significantly higher burdens of rare de novo copy-number variants, coding-sequence variants, and loss-of-function mutations compared to their unaffected siblings and the general population (Gaugler et al. 2014; Huguet, Ey, and Bourgeron 2013). While single common inherited variations, such as met proto-oncogene (hepatocyte growth factor receptor; MET) has been associated with ASD risk, emerging genome-wide research emphasizes the more substantial influence of multiple polygenic traits (Gaugler et al. 2014). While studies have shown a substantial heritability in ASD (Tick et al. 2016; Tchaconas and Adesman 2013), it's worth noting that the concordance rate in monozygotic twins has not reached 100%, indicating that environmental factors are present alongside genetics. Given that the diagnoses of ASD can start around age 2 and the subtle social and communication impairment may be present earlier (Bacon et al. 2018; Charman and Baird 2002), prenatal and early postnatal periods represent a critical window for investigating the potential impact of environmental factors on ASD. 1.1.2. Maternal health during pregnancy Population-based epidemiologic studies have shown associations of ASD or other neurodevelopmental disorders with a broad spectrum of maternal health conditions during pregnancy, including asthma (Theoharides et al. 2016), infection (Zerbo et al. 2013; Jiang et al. 2016), diabetes (Xiang et al. 2015), hypertension (Maher et al. 2018), autoimmune disease (K. 3 Lyall et al. 2014), and allergy (L. A. Croen et al. 2005). The concept of maternal immune activation has been proposed as a potential framework for understanding ASD etiology (Meldrum et al. 2013). The prenatal phase is a critical period for brain development, encompassing the growth, multiplication, and migration of neurons during the first and second trimesters, as well as synapse formation, pruning, and reorganization in the third trimester (Herschkowitz 1988). The maternal immune system is actively interacting with fetal immune cells to create a prenatal environment to support pregnancy (Racicot et al. 2014). However, infections and other health issues may change the cytokine profile of mothers with higher levels of pro-inflammatory cytokines, which can alter the equilibrium of immune function and potentially hinder fetal brain development (Ravaccia and Ghafourian 2020). Unhealthy behaviors such as smoking during pregnancy have also been shown to be related to adverse birth outcomes (Ion and Bernal 2015) and children’s cognitive development (Clifford, Lang, and Chen 2012; Huizink and Mulder 2006). Maternal smoking has been considered as a risk factor for ASD, but existing studies have yielded equivocal results. While one meta-analysis reported nonsignificant overall effects of smoking during pregnancy on ASD, there was notable variability among individual studies (Rosen et al. 2015). One recent study with a large cohort reported that heavy prenatal smoking (20+ cigarettes per day) significantly increased the likelihood of ASD in offspring (von Ehrenstein et al. 2020). Several studies have shown that adequate antioxidative nutrition, such as folic acid, polyunsaturated fatty acid, and vitamin D, intake during pregnancy may serve as potential protective factors for normal neurodevelopment (Morse 2012). Narrowed down to ASD, maternal dietary features including folic acid, vitamin D, and multivitamins were associated with reduced likelihood of ASD (Zhong et al. 2020). 4 1.1.3. Environmental pollution Research on the association between environmental exposures and ASD is a rapidly expanding field, with a particular focus on early-life air pollution exposure. Evidence on the association between early-life air pollution exposure and ASD is accumulating. Since the last decade, more than 20 population-based observational studies have examined associations between ASD and early-life exposure to regulated air pollutants [e.g. particulate matter <2.5 µm (PM2.5) and <10 µm (PM10) in aerodynamic diameter, nitrogen dioxide (NO2), ozone (O3), carbon oxide (CO)], the near-roadway air pollution mixture, and air toxics. The relatively consistent associations between PM2.5 and higher likelihood of ASD has been reported by many North American studies (Becerra et al. 2013; Jo, Eckel, Wang, et al. 2019; Pagalan et al. 2019). A recent meta-analysis has shown that the association between PM2.5 exposure and increased likelihood of ASD was observed in all three trimesters but the association during the second trimester was not significant. The time-specific evidence for other pollutants including PM10, traffic-related air pollution and toxics was not obvious (Dutheil et al. 2021). Other chemical exposures such as pesticides, organochlorine compounds, phthalates, and heavy metals have been shown as risk factors to the brain development of fetuses and children, as well as ASD (Cheng et al. 2019). Possible mechanisms through which prenatal environmental pollutants affect fetal neurodevelopment include maternal systemic oxidative stress(Leni, Künzi, and Geiser 2020) and proinflammatory cytokine production(Xu et al. 2012). These factors can lead to placental and endothelial dysfunction, resulting in increased fetal oxidative stress that may disrupt the differentiation and organization of the fetal brain(Block and CalderonGarciduenas 2009). 5 1.1.4. Socioeconomic characteristics Socioeconomic status (SES) has been regarded as distal causes of disease (Link and Phelan 1995), with lower SES often associated with a higher risk of health issues. However, findings on the associations between SES and ASD were mixed. Studies conducted in countries with free and universal healthcare systems, such as Sweden and Finland, have reported a consistent negative relationship between SES and ASD (Rai et al. 2012; Lehti et al. 2015). In contrast, studies in the United States have generally shown a positive gradient that the likelihood of autism is higher among populations with higher SES (Bhasin and Schendel 2007; Durkin et al. 2010). Since we lack biomarkers for autism, the diagnoses are mainly based on the observation by clinicians, thus, countries like the United States without universal early childhood developmental screening may experience disparities in the resources of autism screening and diagnosis. This potential underdiagnosis of ASD may confound the true biological pathways between SES and ASD. One recent study in California observed a reversal in the social pattern of ASD diagnosis after 2010. One possible reason for this reverse pattern was the reduced gaps in accessibility to ASD screening since the American Academy of Pediatrics promoted the pervasive developmental screenings among children in 2006. (Winter et al. 2020). Numeric potential pathways may exist between social factors and the diagnosis of ASD. Women with lower SES are more likely to have chronic health issues such as obesity and diabetes, insufficient nutrition intake, and unhealthy behaviors, all of which have been shown to be related to higher likelihood of ASD. Furthermore, as the previous section described, emerging evidence has shown associations between environmental pollutants including air pollution and autism. Given that individuals with lower SES often experience higher exposure of environmental pollution, differential exposure levels may contribute to the increased likelihood of autism within this population. Additionally, the psychological stress related to the low SES at 6 both the individual level and the contextual level may also be associated with dysfunctions in the immune system of mothers, making the fetus more vulnerable to the co-exposed environmental risk factors of neurodevelopment. 1.2. Conceptual models of risk factors on ASD 1.2.1. Conceptual model of potential biological pathways Epidemiological studies have shown that genetics, maternal health during pregnancy, environmental pollution, and social factors are associated with ASD. Some key biological pathways including oxidative stress and neuroinflammation have been proposed (James et al. 2006; Onore, Careaga, and Ashwood 2012) and summarized into a conceptual model of biological pathways shown in Figure 1.1. The theoretical pathways of air pollution and autism may have subtle differences before and after birth. In the prenatal period, mothers serve as intermediaries between air pollution and fetuses. High air pollution exposure during pregnancy may be related to maternal systemic oxidative stress (Leni, Künzi, and Geiser 2020) and proinflammatory cytokine production (Xu et al. 2012), resulting in placental and endothelial dysfunction. This, in turn, may lead to heightened fetal oxidative stress and such imbalance may disrupt the differentiation and organization of the fetal brain (Block and Calderon-Garciduenas 2009; Block et al. 2012). For the early postnatal period, children can directly breathe the air. The inhaled air pollutants may directly disturb the redox balance and chronic inflammation on the children, which may have harmful effects on the still developing brain. Similarly, SES factors can have a long-lasting influence on children’s development. During pregnancy, low SES at both the individual and contextual level may induce psychological stress in mothers, which can be associated with increased oxidative stress and systematic chronic 7 inflammation. These chronic imbalances in metabolism may increase fetal oxidative stress and neuroinflammation (Schoots et al. 2018). For the early postnatal period, young children can interact with their parents, caregivers, and developing social relationships. Consequently, unhealthy social relationships can directly influence children’s response to stress. Stressful experiences in early childhood may also introduce chronic inflammation in children, potentially having harmful effects on their developing brains. Regarding neighborhood contextual SES, another important pathway for U.S. studies are the disparities in access to health care and ASD diagnosis (Thomas et al. 2012). Children with low SES may lack timely diagnosis of developmental disorders and thus miss the optimal window for intervention (MacDonald et al. 2014). The disparities in the diagnosis of developmental disorders may even enhance the inequity of health and quality of life for low SES populations. 1.2.2. Conceptual model of susceptibility Given the shared pathways among these co-occurring factors and the absence of a single identifiable cause or biomarker for ASD, the “multiple- exposure” hypothesis of ASD warrants further research. It is possible that the co-existence of more than one condition can have synergistic effects on ASD. For example, disturbance in the maternal immune system may make the mothers more sensitive to environmental pollution, including air pollution and some chemical exposure. Sufficient intake of antioxidant vitamins may buffer the effect of imbalanced oxidative stress, thus alleviating the effect of prenatal air pollution (Goodrich et al. 2018). The potential mechanisms of co-exposed environmental pollution and adverse SES are complicated. In this dissertation, I focused on the interactive effects of three external factors on ASD: prenatal air pollution, maternal immune activation (MIA), and neighborhood socioeconomic status (SES) context. 8 A conceptual framework of ASD risk associated with environmental pollution is shown in Figure 1.2. The right side of this model represents the processes from the emission of pollution to the final health outcome (ASD in this study). Two mechanisms (shown in the middle part of Figure 1.2) of health disparity may occur upstream in the paradigm (from emission to internal doses) and downstream (from internal doses to the development of health conditions). The first mechanism is called differential exposure, meaning that people can be exposed to different levels of environmental pollution. People may reside in areas closer to factories and roadways, or they may be involved in activities at locations where more pollutants are concentrated, so they are exposed to higher levels of ambient pollution. Given that they may have higher ambient environmental pollution, the internal doses also tend to be higher. The differential exposure to environmental hazards such as air pollution can put them at a higher risk of diseases. Second, adverse SES may also elevate vulnerability to the toxic effect of environmental contaminants, known as differential susceptibility. Susceptibility to environmental pollutants can vary from person to person and can be influenced by a range of factors, including genetics, age, sex, preexisting health conditions, psychological stress, and lifestyle choices. Those factors might influence the biological response (e.g., immune function, oxidative stress) of mothers and fetus to the internal exposure to pollutants and ultimately modify the likelihood of diseases. In the context of public health, SES is often considered as a distal factor of health conditions (Link and Phelan 1995). Individuals are nested in the social ecology from the macro level (laws and regional policy) to the community level (neighborhoods) and finally down to the micro level (interpersonal and individual). Social contexts at different levels (for example, neighborhood context and individual/family characteristics) are correlated and they may have synergistic effects on both the determining factors of differential exposure and susceptibility 9 (shown in the left-hand side of Figure 1.2). In the United States, populations with lower SES are more likely to live or work in areas with higher environmental pollution. Therefore, they are experiencing higher accumulative doses of pollutants, which may be harmful to their health. For differential susceptibility, lower SES is related to several biological determining factors of susceptibility, including poor health status and higher rates of chronic conditions such as obesity and diabetes (McLaren 2007; Robbins et al. 2005) and unhealthy behaviors like smoking (Hiscock et al. 2012). Also, people living in disadvantaged neighborhoods may lack access to healthy food (Walker, Keane, and Burke 2010), and therefore the poor nutrition status and insufficient intake of antioxidant vitamins can also make them more sensitive to environmental exposure. Low SES is often associated with higher chronic stress. Studies have shown that people under high chronic stress tend to have increased respiration (Widjaja et al. 2013), allowing higher absorption of environmental pollution into their bodies. It is common that people with low SES can experience both higher exposure levels to pollution and higher susceptibility to environmental pollutants. This “double jeopardy” situation may intensify existing health disparities (Morello-Frosch and Shenassa 2006). In summary, prenatal air pollution, adverse social experiences, and maternal health conditions may share similar pathways of maternal and fetal imbalanced oxidative stress and inflammation on prenatal brain development and finally lead to autistic traits in children. Thus, it is possible that these prenatal risk factors can interact with each other and form sufficient causal pies of ASD. However, most studies only focus on the main individual association of each prenatal factor on ASD. There is a lack of studies on the synergistic effects of multiple risk factors together. 10 1.3. Challenges of examining the interactive effects of multiple co-existing factors The limited number of studies examining the interactive association of multiple coexisting factors on ASD may be attributed to several challenges and complexities involved in conducting such research. First, the complex interplay of multiple factors, including genetics, environment, lifestyle, and socioeconomic status, on ASD can be logistically and theoretically challenging. Although the Rothman causal pie model has been put forward for decades (Rothman 1976), only recently has maternal immune activation been raised as a conceptual framework to explore “multiple-exposure” during pregnancy on neurodevelopmental disorders. Research on the interactive associations is needed to investigate how and why the co-existing factors can reinforce the other on fetus neurodevelopment. Second, access to large and high-quality cohort datasets can ensure robust analyses of complex interactions. Examining the interactive effects of multiple factors typically requires sufficient statistical power. Moreover, understanding the long-term or chronic effects of multiple interacting factors on health often necessitates longitudinal studies that track individuals over extended periods. Identifying a time sequence of exposures preceding health outcome can provide stronger evidence for potential causality. However, collecting comprehensive longitudinal data on multiple co-existing factors can be practically challenging. Third, environmental exposures are often measured at different scales. For example, maternal health conditions and behaviors can be measured at individual levels, whereas ambient air pollutants are typically represented as raster layers with different spatial and temporal resolution. Additionally, socioeconomic status encompasses a spectrum, ranging from the individual/family level to the community level, as well as the macro policy environment. 11 Therefore, combining exposures with different scales requires sophisticated statistical techniques such as multilevel regression models. Finally, not only do researchers need data from different sources, but also domain knowledges from different fields to study the interactive effects of multiple co-existing factors on health conditions. Therefore, investigating multiple co-existing factors often requires interdisciplinary collaboration among experts in various fields (e.g., epidemiology, environmental science, and social sciences). In this dissertation, I put forward a conceptual model of susceptibility of ASD likelihood associated with prenatal ambient air pollution exposures. Based on the conceptual model, I examined two hypotheses on the interaction between prenatal PM2.5 and MIA, as well as the interaction between prenatal PM2.5 and neighborhood disadvantage on ASD likelihood in offspring. All the analyses were conducted on a large retrospective cohort with more than 300,000 mother-child pairs born between 2001 and 2014 in Southern California, which ensure the statistical power to detect small interactive associations. The external environmental data, including ambient PM2.5 and neighborhood SES measures were assigned to the cohort based on the residential addresses. Last, a group of experts in various domains (epidemiology, environmental health, neurodevelopment, neighborhood characteristics, and biostatistics) have assisted me on the empirical research presented in this dissertation. 1.4. Study settings This dissertation focused on populations residing in Southern California. According to the CDC ADDM resilient network, the estimated prevalence of ASD in California is twice the national average (Maenner et al. 2023). Additionally, air pollution in California is higher than the national average, potentially due to high population density and traffic, as well as wildfires. 12 Therefore, southern California is a good place to conduct epidemiological studies to examine the environmental influence on ASD. This dissertation utilized a population-based retrospective pregnancy cohort that included more than 300,00 mothers with singleton deliveries at Kaiser Permanente Southern California (KPSC) hospitals (Figure 1.3) between January 1, 2001 and December 31, 2014. KPSC is a large integrated healthcare system with over 4.5 million members across Southern California. KPSC membership reflects the diverse socioeconomic demographics in the study’s region (Koebnick et al. 2012). In this cohort, more than 6,000 children got ASD diagnosis during the full follow-up period and more than 4,000 got the diagnosis before age 5. The distribution of Kaplan-Meier (Bland and Altman 1998) estimated incidence of ASD by census tracts (assigned based on residential addresses at birth) is shown in Figure 1.4. The Moran’s I tests (Dubin 1998) were not statistically significant with varying definitions of neighbors to examine spatial autocorrelation, and thus no spatial clusters were observed for census tract-level ASD incidence in this cohort. The detailed residential addresses are not shown due to privacy concerns. Mothers’ health conditions, ASD diagnosis in children, and demographic information were extracted from high-quality integrated electronic medical records (EMR) maintained by KPSC. Maternal address history was also extracted from EMR. Addresses were geocoded using ArcGIS (ArcGIS 2021). Addresses based only on street name, 5-digit postal code, locality, or administrative unit were considered too uncertain to be geolocated into the correct grid used for exposure assignments. Monthly estimates of PM2.5 and four major PM2.5 chemical components [BC, OM, NO3 - , and SO4 2- ] with a 1 km spatial resolution were estimated by a hybrid model (version V4.NA.02) that integrates chemical transport model outputs, satellite observations, and ground-based 13 measurements as developed by the Atmospheric Composition Analysis Group at Washington University in St. Louis (Van Donkelaar et al. 2019). Exposures to PM2.5 and these selected components were assigned to maternal address during each of three trimesters. Exposures were time-weighted to account for changes in maternal addresses during pregnancy. The pregnancy average PM2.5 exposure by birth year is shown in Figure 1.5. PM2.5 mass, NO3 - , SO4 2- and BC during the study period decreased over time, while OM was relatively stable. Neighborhood was defined as the US census tract of the residential address at the child’s birth. Data were extracted from the 2000 US decennial census for births from 2001-2004, and 2005-2009 and 2010-2014 American Community Survey (ACS) 5-year estimates for corresponding birth years. Neighborhood disadvantage was defined as the first principal component of seven census tract indicators(B.L. Levy, Owens, and Sampson 2019): poverty, unemployment, female-headed households with children, public assistance, less than a high school education, bachelor’s degree or greater, and professional occupation. The spatial distribution of neighborhood disadvantage index generated based on the 2005-2009 ACS data is shown in Figure 1.6. The Moran’s I tests (Dubin 1998) were statistically significant with all three definitions of neighbors to examine spatial autocorrelation, indicating the. Spatial autocorrelation in neighborhood disadvantage index. The spatial distributions for the disadvantage index based on 2000 US decennial census and 2010-2014 ACS data were similar to the index generated from 2005-2009 ACS data. Thus, I only chose one as an example. 1.5. Research outline This dissertation aims to investigate the potential effect modifiers of the association between prenatal air pollution and ASD. A systematic review (Chapter 2) and three empirical 14 studies (Chapter 3-5) were conducted and the three aims are presented in Figure 1.7. Chapter 6 is the conclusion chapter with future directions. The systematic review presented in Chapter 2 summarizes the existing evidence on the susceptibility to autism risks associated with early life ambient air pollution. A systematic literature search in PubMed and Embase was conducted. A total of 19 publications examined modifiers of the associations between early life air pollution exposures and ASD. Child’s sex, maternal nutrition or diabetes, socioeconomic factors, and child risk genotypes were reported to modify the effect of early-life air pollutants on ASD risk in the epidemiologic literature. However, the sparsity of studies on comparable modifying hypotheses precludes conclusive findings. Further research is needed to identify susceptible populations and potential targets for preventive intervention. Chapter 3 is an empirical study examining interactions between MIA-related conditions and prenatal PM2.5 or major PM2.5 components on ASD likelihood. I hypothesized that children whose mothers experienced conditions causing immune activation, such as infection, allergy, autoimmune, and hypertension, were more vulnerable to air pollution-associated autism likelihood. In the birth cohort described in section 1.4, MIA-related conditions were extracted from the mothers’ EMR, and air pollutants were assigned based on residential addresses during pregnancy. I estimated the ASD risk associated with MIA-related conditions, air pollution, and their interactions, using Cox regression models adjusted for covariates. Chapter 4 examines the association between neighborhood disadvantage and ASD likelihood, with potential effect modification by maternal and child demographic characteristics in a population with health insurance. I hypothesized that when disparities in access to screening and diagnostic services are reduced, neighborhood disadvantage is associated with higher 15 likelihood of ASD diagnosis. Socioeconomic disadvantage at the neighborhood level was assigned based on the residential address at birth. Associations between neighborhood disadvantage and ASD diagnosis were determined by hazard ratios (HRs) from Cox regression models adjusted for covariates. Effect modification by maternal race and ethnicity, maternal education, and child sex was assessed. Chapter 5 presents a study using G-computation to examine the potential effects of hypothetical air pollution interventions on the cumulative incidence of childhood ASD and whether these intervention effects vary by neighborhood disadvantage levels. Two sets of intervention strategies were examined: 1) an overall percentage reduction in PM2.5 concentration; 2) using PM2.5 regularity standards as thresholds. An innovative window-based G-computation model based on Cox regression framework was applied to assess effect modification by neighborhood disadvantage. In Chapter 6, I summarize the findings from the three empirical studies. I also discuss the strengths and limitations, and provide some insights for future research. 16 Figure 1.1 Conceptual model of risk factors of ASD and potential pathways Figure 1.2 Susceptibility model of environmental pollution-associated ASD likelihood 17 Figure 1.3 Map of Kaiser Permanente Southern California study catchment area Notes: This map is adapted from Carter, S. A., M. M. Rahman, J. C. Lin, Y. H. Shu, T. Chow, X. Yu, M. P. Martinez, S. P. Eckel, J. C. Chen, Z. Chen, J. Schwartz, N. Pavlovic, F. W. Lurmann, R. McConnell, and A. H. Xiang. 2022. "In utero exposure to nearroadway air pollution and autism spectrum disorder in children." Environ Int 158: 106898. Supplemental material eFigure 1, https://doi.org/10.1016/j.envint.2021.106898. 18 Figure 1.4 Cumulative incidence of ASD by age 5 in the study cohort by census tracts Note: Moran’s I test was conducted using ArcGIS. 19 Figure 1.5 Distribution of air pollutant concentrations during pregnancy across birth year 2001- 2014 20 Figure 1.6 Neighborhood disadvantage index in Southern California by census tracts Note: Moran’s I test was conducted using ArcGIS. 21 Figure 1.7 Study aims of this dissertation 22 Chapter 2 Evidence of Susceptibility to Autism Risks Associated with Early Life Ambient Air pollution: A Systematic Review This chapter has been published as Yu, X., M. M. Rahman, Z. Wang, S. A. Carter, J. Schwartz, Z. Chen, S. P. Eckel, D. Hackman, J. C. Chen, A. H. Xiang, and R. McConnell. 2022. "Evidence of susceptibility to autism risks associated with early life ambient air pollution: A systematic review." Environ Res 208: 112590. 2.1. Abstract Background: Many studies have found associations between early life air pollution exposure and subsequent onset of autism spectrum disorder (ASD). However, characteristics that affect susceptibility remain unclear. Objective: This systematic review examined epidemiologic studies on the modifying roles of social, child, genetic and maternal characteristics in associations between prenatal and early postnatal air pollution exposure and ASD. Methods: A systematic literature search in PubMed and Embase was conducted. Studies that examined modifiers of the association between air pollution and ASD were included. Results: A total of 19 publications examined modifiers of the associations between early life air pollution exposures and ASD. In general, estimates of effects on risk of ASD in boys were larger than in girls (based on 11 studies). Results from studies of effects of family education (2 studies) and neighborhood deprivation (2 studies) on air pollution-ASD associations were inconsistent. Limited data (1 study) suggest pregnant women with insufficient folic acid intake might be more susceptible to ambient particulate matter less than 2.5 µm (PM2.5) and 10 µm (PM10) in aerodynamic diameter, and to nitrogen dioxide (NO2). Children of mothers with 23 gestational diabetes had increased risk of ozone-associated ASD (1 study). Two genetic studies reported that copy number variations may amplify the effect of ozone, and MET rs1858830 CC genotype may augment effects of PM and near-roadway pollutants on ASD. Conclusions: Child’s sex, maternal nutrition or diabetes, socioeconomic factors, and child risk genotypes were reported to modify the effect of early-life air pollutants on ASD risk in the epidemiologic literature. However, the sparsity of studies on comparable modifying hypotheses precludes conclusive findings. Further research is needed to identify susceptible populations and potential targets for preventive intervention. 2.2. Introduction Autism spectrum disorder (ASD) is a complex developmental disorder characterized by impairments in social interactions and communication and the presence of restricted, repetitive, and stereotyped patterns of behaviors(American Psychiatric Association 2013b). It is associated with lifetime social and emotional hardship for children (Kuhlthau et al. 2010) and their families (Rao and Beidel 2009), as well as economic burdens on families and society (Buescher et al. 2014). The estimated prevalence of ASD in the United States increased from 0.66% in 2002 to 2.78% in 2020 (Centers for Disease Control and Prevention 2007; Maenner et al. 2023), which is only partly explained by more widespread screening of children and an expanded definition of ASD since 2008 to include more mild cases (Hertz-Picciotto and Delwiche 2009; Blumberg et al. 2013). ASD can be diagnosed starting around age 2, but subtle social and communication impairment may be present earlier (Bacon et al. 2018; Charman and Baird 2002), suggesting determinants occur in utero or in early life. Although studies have shown a substantial heritability in ASD (Tick et al. 2016; Tchaconas and Adesman 2013), environmental factors are likely to play an important role in causation. 24 In the last decade, many studies have examined associations between ASD and early-life exposure to ambient air pollutants [e.g. particulate matter <2.5 µm (PM2.5) and <10 µm (PM10) in aerodynamic diameter, nitrogen dioxide (NO2), ozone (O3), carbon monoxide (CO)], the nearroadway air pollution mixture, and, in the U.S., air toxics derived from the National-Scale Air Toxics Assessment. However, results were not uniformly consistent. For example, several studies in the United States showed that prenatal exposure to PM2.5 was associated with increased risk of ASD (Becerra et al. 2013; Kalkbrenner et al. 2015; McGuinn et al. 2020; Raz et al. 2015; Kaufman et al. 2019; Jo, Eckel, Wang, et al. 2019; Talbott, Arena, et al. 2015; Volk et al. 2013). In contrast, studies conducted in Europe have reported no association (Guxens et al. 2016; Gong et al. 2014; Gong et al. 2017). Some studies of air toxics have found metals (mercury, cadmium, nickel, lead, and others), aromatic solvents, and chlorinated solvents to be associated with a higher risk of autism (Blanchard, Palmer, and Stein 2011; Kalkbrenner et al. 2018; Lewandowski et al. 2009; Roberts et al. 2013; von Ehrenstein et al. 2014; Talbott, Marshall, et al. 2015; Windham et al. 2006), but these findings also have not been consistent. Three high-quality systematic reviews and meta-analyses described these inconsistencies and proposed some plausible explanations for differences between studies, including different levels of air pollutant exposures, diverse ASD ascertainment methods, and different ASD phenotypes (Flores-Pajot et al. 2016; Lam et al. 2016; Chun et al. 2020). However, none focused on possible differences in effects due to risk or protective factors that may have increased or decreased the susceptibility of some populations. ASD likely has a multifactorial etiology (Hertz-Picciotto, Schmidt, and Krakowiak 2018), and in recent years a “second hit” hypothesis that exposures such as air pollution would require additional risk factor(s) to cause disease has been proposed (Estes and McAllister 2016; Bilbo et al. 2018). Thus, susceptibility to air 25 pollution might occur in children with other risk factors for ASD, such as male sex; maternal infections, diabetes mellitus, hypertension and other maternal inflammatory conditions, so-called “maternal immune activators” (K. Lyall, Schmidt, and Hertz-Picciotto 2014; Ornoy, WeinsteinFudim, and Ergaz 2015); in children of mothers with low nutritional intake of folic acid (Levine et al. 2018), medication for neuro- or neuropsychological disorders (Lisa A Croen et al. 2011); or in children from marginalized communities with low socioeconomic status (Mathiarasan and Hüls 2021). Child’s Genotypic variation increasing susceptibility to air pollution may help identify biological pathways (Huguet, Ey, and Bourgeron 2013). Synergistic associations of other risk factors could markedly increase the attributable burden of air pollution in concert with these other exposures. In addition, understanding patterns of susceptibility can help identify sensitive subpopulations that may benefit most from air pollution emission controls. The Clean Air Act in the United States requires that the regulatory standard protects everybody, including the most vulnerable, from adverse impacts of air pollution (O'Neill et al. 2003). Exposure reductions can also reduce environmental health disparities (J.I. Levy, Greco, and Spengler 2002), as low socioeconomic status (SES) communities often experience greater air pollution exposure and larger effects for some outcomes (Evans and Kantrowitz 2002; Hajat, Hsia, and O'Neill 2015; Fuller et al. 2017). The purpose of this review was to examine how ASD susceptibility to air pollution has varied among vulnerable subpopulations in studies to date. We identified knowledge gaps in our understanding of ASD susceptibility to air pollution and suggest possible approaches for future studies to address these gaps. 26 2.3. Methods 2.3.1. Eligibility criteria This review followed the Preferred Reporting Items for Systematic Reviews and MetaAnalyses (PRISMA) checklist (Moher et al. 2009). To be included in this review, a study met the following criteria: 1) It was published in a peer-reviewed journal; 2) The exposure was ambient criteria (regulated) air pollution or airborne toxics, thus, indoor airborne chemical and secondhand smoking exposures were not included; 3) The health outcome was ASD or related subphenotypes, 4) It was empirical, which excluded reviews, comments, replies, protocols, conceptual studies, and letters to the editor; 5) The study was in humans, thus, animal studies were not included; 6) The study was published in English; and 7) Effect modification was evaluated in stratified analyses or in interactions. 2.3.2. Search strategy and study selection We identified articles published in peer-reviewed journals from PubMed and Embase database. The database search was conducted on May 1st, 2021. The search keywords were a combination of terms for exposures and health outcomes. For exposures, keywords were “air pollutants”, “air pollution”, “particulate matter”, “PM”, “nitrogen dioxide”, “NO2”, “nitrogen oxide”, “NO”, “ozone”, “O3”, “Sulfur/Sulphur dioxide”, “SO2”, “carbon monoxide”, “CO”, “vehicle emissions”, and “traffic”. For outcomes, key terms were “autism spectrum disorder”, “ASD”, “autistic”, “pervasive developmental disorder”, “PDD”, “PDD-NOS”, and “Asperger Syndrome”. The search syntaxes for the two databases were shown in Supplemental Table 2.1 and Supplemental Table 2.2. We also reviewed references cited in eligible studies to identify additional articles for inclusion; no additional eligible publications were added from the references. 27 The study selection was conducted in two steps. Step 1 involved screening titles and abstracts to identify human studies of ambient air pollution and ASD. Studies were identified sequentially according to the first six eligibility criteria (Section 2.3.1). If a study failed to meet an eligibility criterion, there was no further review of other inclusion criteria. For Step 2, full-text was reviewed to select articles reporting effect modification or interactions of air pollution with other factors in association with ASD. In this step, all 7 inclusion criteria were evaluated. Because there were few studies with comparable exposures and effect modifiers, we did not conduct a meta-analysis. 2.3.3. Data extraction Two major categories of data were extracted from studies selected. The first included study characteristics including year, study design, study population, sample size, exposure measurement, outcome ascertainment, modifiers, and covariates. The second focused on effect sizes for effect modification. For stratified analyses, the estimated effect size and corresponding confidence interval from each modifying category were recorded. When interactions were tested, results of additive and multiplicative tests were also extracted. 2.3.4. Assessment of risk of bias For each study, the assessment of risk of bias was based on the nine-point NewcastleOttawa Quality Assessment Scale (NOS) (Wells et al. 2014). For case-control studies, the scale consists of three evaluation sections corresponding to (a) the possibility of bias in selection of cases and controls (4 points), (b) comparability of cases and controls and treatment of confounders on the basis of study design or data analysis (2 points), and (c) the quality and completeness of exposure assessment (3 points). For cohort studies, the three sections correspond to (a) the assessment of possible bias in the selection of both exposed and nonexposed 28 participants (4 points), (b) the comparability between exposed and nonexposed participants and treatment of confounders on the basis of study design or data analysis (2 points), and (c) the evaluation of potential bias in assessment of the outcome and in follow-up (3 points). A higher score represents a higher quality of study. A common cut-off score for an acceptable quality study is 7 (Chun et al. 2020; Flores-Pajot et al. 2016). Additionally, we assessed the risk of bias for each study according to the qualitative guidelines developed by Lam et al. (Lam et al. 2016). These guidelines are based on the GRADE principles (Guyatt et al. 2008). Five items from the guidelines were applicable for this study, including the source population representativeness, the misclassification of exposures and outcomes, the comprehensiveness of potential confounders, and the selective reporting of results. We added an evaluation of misclassification of effect modifiers (the sixth item), using similar rules to those for misclassification of pollution exposure. Each item was rated with 4 risk levels (low, probably low, probably high, and high) or not applicable (insufficient information to rate). Detailed criteria for the assessment of the six items are listed in Supplemental Table 2.3. Database search, data extraction and assessment of risk of bias were conducted by two authors (XY and ZW) independently. Disagreements were resolved by a third author (MR). 2.3.5. Analytical approach Effect modifiers evaluated based on the literature review included (1) child’s sex, (2) markers for socioeconomic status (maternal education, race, and neighborhood deprivation and other characteristics from census data), (3) maternal factors (nutrition, maternal immune activators, such as diabetes), and (4) genetics. We summarized the results of effect modification from each study. 29 Additive or multiplicative interaction effects between the modifier and air pollution exposure were reported, if available. For studies that showed only results from stratified analyses by susceptibility factors, evaluation of effect modification was based on comparison of each subgroup's effect size, as described by Altman and Bland (2003). We calculated the original coefficients and standard errors from the reported odd ratios or hazard ratios and corresponding 95% confidence intervals reported in most studies. The difference between the original regression coefficients was assessed with the Wald test using the pooled standard error from the two strata. An example is presented in Supplemental Table 2.4. 2.4. Results 2.4.1. Study selection Studies identified and then excluded at each step of selection are shown in the PRISMA flow chart (Figure 2.1). A total of 1272 publications (N=618 from PubMed and N=654 from Embase) were identified through the database search, of which 296 were duplicates. The remaining 976 articles were screened by title and abstract and 937 publications that were not empirical human studies on the association between air pollution and ASD were excluded. A total of 39 articles studied the association between ambient air pollution and ASD were identified. In the full-text assessment, another 20 studies were excluded because they did not examine any effect modifiers (Supplemental Table 2.5). A total of 19 publications were included in this review. 2.4.2. Study characteristics A description of the 19 included studies is shown in Supplemental Table 2.6. All studies that evaluated effect modification were published since 2010. The majority (N=15) were conducted in the United States, including 7 based on populations in California. There were four 30 studies from other countries, one each from Sweden, Israel, Denmark, and Canada. Studies from other countries were not included because they did not examine effect modifiers. There were 13 case-control studies, 4 cohort studies and 2 ecological studies. The most widely examined air pollutant was particulate matter (PM); 11 publications studied the effect of PM2.5 and 10 studied PM10. Nitrogen oxides (NOx, NO, and NO2) were examined in 12 articles, ozone (O3) in 6 studies, sulfur dioxide (SO2) in 1 study, and carbon monoxide (CO) in 1 study. Nine studies examined more than one regulated air pollutant. In addition, three studies examined effects of the near-roadway air pollution mixture, and 5 examined effects of airborne toxicants. Except for two ecological studies, all studies examined exposure during pregnancy (N=17) and 10 of these assessed trimester-specific effects. Five studies assessed pre-pregnancy effects, and 7 examined effects of postnatal exposure. Methods for ASD ascertainment were varied, including detailed clinical and neuropsychological evaluations, extraction from electronic medical records, and questionnaire surveys. Six studies further divided ASD into sub-phenotypes based on comorbidities. 2.4.3. Risk of bias Both the NOS and the navigation guide from Lam et al. (2016) provided criteria for assessing bias in individual level observational studies. Therefore, two ecological studies (potentially subject to ecological fallacy) were excluded from the assessment of bias. NOS scores (the last column of Supplemental Table 2.6) for all the 17 observational studies with individual level data passed the cut-off score of 7 for acceptable quality. The assessment of risk of bias corresponding to the guide from Lam is shown in Supplemental Table 2.7. Most studies were rated as “low” or “probably low” on all 6 criteria for bias, with some exceptions. For example, the source population representativeness had “probably 31 high” bias in Kalkbrenner et al. (2018), because the study population was volunteers. Three studies of air toxics were rated as “probably high” risk of exposure misclassification bias because this dataset has high uncertainty in the exposure estimates (Xue and Jia 2019; George et al. 2011). Among the effect modifiers, the measure of folic acid intake in Goodrich et al. (2018) had “probably high” bias, because the assessment of folic acid intake was based on retrospective recall in a telephone interview. The potential misclassification of urbanicity in Kalkbrenner et al. (2010) was classified as “probably high” because there was no description of how the “urbanicity” modifier was defined or obtained. 2.4.4. Effect measure modifiers 2.4.4.1. Social and demographic features A total of 11 papers (7 of regulated pollutants and 4 of air toxics) examined effect modification by child’s sex; 3 examined modification by maternal or family education level. Two studies assessed variation in air pollution effects by neighborhood deprivation and one by urbanicity. Effect modification by race and income were only evaluated in the two ecological studies. • Child’s sex: Most studies reported that the effect estimates of early life exposures to both PM10 and PM2.5 were stronger in boys than in girls (Table 2.1). Only Jo, Eckel, Wang, et al. (2019) found that the hazard ratio in boys for the first trimester average PM2.5 [HR= 1.18 (95% CI: 1.08, 1.27)] was statistically different from that in girls [HR=0.90 (95% CI: 0.76, 1.07)]; p-interaction=0.03. Larger but not significantly different PM2.5 effects were observed in boys in the third trimester, for the entire pregnancy average exposure, and during the first year of life. 32 ASD-associated sex interactions with other regulated pollutants were not statistically significant (Table 2.1). One study found no association of NO2 during pregnancy with ASD risk in girls [OR=1.00 (95% CI: 0.86, 1.16)] and a positive effect in boys [OR=1.07(95% CI: 1.00, 1.13)], but interaction was not significant (Pagalan et al. 2019). NO exposure during pregnancy had a similar pattern of interactions with sex in the same study. Another study found average NO2 exposure during pregnancy was associated with increased risk in girls [OR=1.46 (95% CI: 0.61, 3.46)] but decreased risk in boys [OR= 0.51 (95% CI: 0.35, 0.76)] and the difference in the effect estimates was significant (p=0.03) (Raz et al. 2018). One ecological study reported no significant effect modification by sex in the association between exposure to unhealthy air (air quality index>100) and ASD prevalence (Al-Hamdan et al. 2018). Four studies assessed the interaction of child’s sex with air toxics exposure. More than 100 air toxics were studied. Therefore, we show only those for which there were significant interactions (Supplemental Table 2.8). With exceptions of ethylidene dichloride and polyaromatic hydrocarbons (von Ehrenstein et al. 2014), there was a consistent pattern of larger air toxics effects sizes for boys than girls. Two studies reported airborne lead to be a risk factor for ASD in boys but not in girls (Roberts et al. 2013; von Ehrenstein et al. 2014). Other metals (such as antimony, cadmium, and nickel) and volatile organic compounds were also reported as risk factors for ASD in boys not for girls (Roberts et al. 2013; Kalkbrenner et al. 2018). • Education Three studies (2 case-control and 1 ecological design) demonstrated no consistent pattern of effect modification of the air pollution-ASD association by education level (Table 2.2). Becerra et al. (2013) found that the effect of traffic-related NO2 during pregnancy on ASD was statistically significantly higher in children whose mothers did not graduate from high school 33 than in children whose mothers graduated high school. However, in the same study, stronger associations of CO and NO2 exposure during pregnancy were observed in children with maternal education level greater than high school compared with those with less than high school, and these differences in point estimates were statistically significant. A Swedish study reported that associations of traffic-related NOx and PM10 exposure during pregnancy and the first year of life were stronger in the more educated subgroup (with more than 12 years of education), but neither main effects nor interactions were statistically significant (Gong et al. 2017). An ecological study on air toxics found that the association of mercury with ASD risk was significantly higher in census tracts with below-average percentage of college educated households (Dickerson et al. 2016). • Neighborhood characteristics Two case-control studies examined the modifying effect of neighborhood deprivation, based on an index including income, education, unemployment and other socioeconomic characteristics in small socioeconomically homogeneous areas (Table 2.3). One U.S. study found that children living in the most deprived neighborhoods had stronger ASD associations with PM2.5 (dichotomized at 12 ug/m3 ) during the first year of life [OR=2.17 (95%CI: 1.14, 4.15); interaction p-value 0.08] (McGuinn et al. 2019). When modeling PM2.5 as a continuous variable, the interaction between neighborhood deprivation level also was not significant in the same study (Table 2.3). In contrast, a Swedish study found that traffic-related NOx and PM10 exposure during the first year of life were more strongly associated with ASD in less deprived neighborhoods (p-interaction=0.09) (Gong et al. 2017). Kalkbrenner et al. (2010) reported that there were no modifying effects of census tract urbanicity (100% rural, mixed, and 100% urban) on the association of air toxics with ASD. Dickerson et al. (2016) found that children from 34 census tracts with higher poverty rates had stronger associations of a combined metals index with ASD. • Race Racial differences in effects were examined in only one ecological study, in which there was an association of ASD with unhealthy air quality (AQI > 100) only among Asians, which was not significant (OR= 3.59; 95%CI: 0.87, 14.8; p-value=0.07) after adjustment for confounders (Al-Hamdan et al. 2018). 2.4.4.2. Maternal factors Two studies reported effect modification by maternal factors (one for folic acid intake and one for maternal diabetes). In a case-control study, Goodrich et al. (2018) found that children of mothers with low folic acid intake during the first trimester (but not in other time periods) had stronger associations of ASD with NO2 than mothers with sufficient folic acid intake [p-values for additive and multiplicative interaction<0.05] (Table 2.4). Jo, Eckel, Chen, et al. (2019) reported increased ASD risk associated with first trimester O3 exposure among mothers with gestational diabetes mellitus (GDM) diagnosed < 24 weeks' gestation [adjusted HR 1.50 per 15.7 ppb O3 (95% CI: 1.08, 2.09)], compared to mothers without diabetes. No O3 associations with ASD were observed among children of mothers with later onset GDM or with pre-existing diabetes. 2.4.4.3. Genetic factors Two studies of interactions with genetic factors were identified (Table 2.5). Kim et al. (2017) examined 15 interactions between 3 types of copy number variation (CNV) burden (duplication, deletion, and total) and 5 air pollutants (NO2, O3, PM2.5 and PM10, and nearroadway pollution). Only interactions of O3 exposures during pregnancy (per 6.2 ppb) with CNV 35 duplication burden per increase of 1,356,513 base pairs [OR=1.55 (95% CI: 1.09, 2.21) and with total CNV burden OR=1.36 (95% CI: 1.01, 1.81) were significant. In another study, ASD risk was increased in association with an interaction of exposure to high PM10, NO2, and dispersionmodeled near-roadway pollution (dichotomized at the 75th percentile of each distribution) with the MET receptor tyrosine kinase rs1858830 CC genotype (compared to MET CG or GG genotype) (Volk et al. 2014). 2.5. Discussion Since Windham et al. (2006) published the first study of the association between prenatal air toxics and the risk of ASD in 2006, more than 30 papers have examined air pollution-ASD association in diverse populations and regions. In the last decade, studies have begun to examine susceptibility to air pollution. This approach has the potential to identify susceptible populations and potentially to explain some inconsistency in the main effects of air pollution. However, we found only limited evidence that male sex, low education, maternal folate deficiency and gestational diabetes, and child’s copy number and MET receptor tyrosine kinase genotypic variation may increase risk of ASD associated with air pollution. There were few studies that examined each susceptibility characteristic, except sex of the child, and little consistency of findings across studies. 2.5.1. Susceptibility by sex The most widely examined effect modifier was child’s sex. In general, epidemiological studies found stronger effects of early life PM2.5 and air toxics exposure on the risk of ASD in boys than girls. These findings are consistent with toxicological studies showing that the activation of microglial cells induced by neuroinflammation may lead to synapse dysfunction involved in ASD (Hammond, Robinton, and Stevens 2018; Bolton et al. 2017). Males may be 36 more sensitive to PM or other inflammatory insults because they have more microglia than females during gestational and early postnatal periods (Lenz and McCarthy 2015). An alternative explanation is that higher prenatal testosterone levels during fetal brain development activate microglia, making males more vulnerable to air pollutants or other inflammatory insults (Auyeung, Lombardo, and Baron-Cohen 2013; McCarthy 2016). In human studies, the small sample of girls in many studies of air pollution and ASD limited ability to detect small differences between effects in boys and girls. For example, one study with 441 ASD cases in girls reported larger effect estimates in boys [boys (HR=1.18; 95% CI, 1.08, 1.27) than in girls (HR=0.90; 95% CI, 0.76, 1.07); p-interaction=0.03)] (Jo, Eckel, Wang, et al. 2019). In another study, a considerably larger PM2.5 effect in boys (OR=1.73; 95% CI, 1.29, 2.31) than in girls (OR=1.12; 95% CI, 0.59, 2.12)] was not significantly different (p-interaction=0.17), but there were only 23 girls with ASD (Raz et al. 2015). 2.5.2. Susceptibility by socioeconomic characteristics Sparse and inconsistent findings limit our ability to draw conclusions about how SES modifies air pollution effects. For example, Becerra et al. showed that the associations of NO2 with ASD were stronger for children of mothers without high school degrees, but for CO stronger associations were observed among children of better educated women. Reasons for these findings are unclear. However, in general, larger effects among lower educated mothers might be expected if low-educated mothers were more likely to be at home (Cohn, Livingston, and Wang 2014) and so residential exposure estimates during pregnancy were more accurate. More educated mothers may be more likely to work in indoor office jobs and in buildings that filter ambient pollution and therefore have little exposure. Highly educated mothers may be more likely to have diets supplemented with folate or rich in antioxidants that could diminish the effect 37 of air pollution (Lim et al. 2019; Croft et al. 2018). One potential explanation for why smaller effects of ambient air pollution might be observed in children of less educated mothers comes from the cumulative risk model, based on a hypothesis that large effects of co-occurring other relatively high adverse exposures or social stressors may result in little additional risk from ambient air pollution (O'Neill et al. 2003). One challenge in assessing effects of SES is that SES can have independent effects at both the individual level and the area level (O'Neill et al., 2003). Neighborhood SES markers such as average education, income, employment, rates of poverty or home values in the census tract have sometimes been found to more strongly modify effects of air pollution on cardiovascular outcomes than individual level education, occupation and income (Chi et al., 2016). According to the “double jeopardy” conceptual model (O'Neill et al., 2003), in addition to the generally greater concentrations of air pollution associated with neighborhood SES, deprived neighborhoods may be lacking in features such as healthy food availability and green space, contributing to a higher risk of chronic conditions, such as diabetes and obesity, which may increase susceptibility to ambient pollutants. 2.5.3. Susceptibility by maternal health conditions and nutrition intake One study reported that increased ASD risk was associated with first trimester O3 among mothers with gestational diabetes mellitus (GDM) diagnosed before 24 weeks' gestation (Jo, Eckel, Chen, et al. 2019). Diabetes is one example of many conditions that cause maternal immune activation (MIA), including viral and bacterial infections (Zerbo et al. 2013; Jiang et al. 2016), asthma (L. A. Croen et al. 2005), and preeclampsia (Dachew et al. 2018) that have been associated with higher risk of ASD. Common biological pathways of effects of MIA, O3 and PM2.5 include systemic inflammation and oxidative stress that are associated with increased 38 reactive oxygen species in the placenta and fetus (Brown et al. 2001; Chen and Lippmann 2009; Bonini and Sargis 2018; Cuffe, Xu, and Perkins 2017). Thus, it is biologically plausible to hypothesize that the presence of MIA during critical time windows may have synergistic effects with early-life air pollution on ASD, a “second hit” resulting in disease (Estes and McAllister 2016; Bilbo et al. 2018). This hypothesis has the potential to identify new approaches to prevention based either on treatment, for example dietary approaches to reducing gestational diabetes, in addition to regulation of pollutant levels. Parenthetically, in the United States, regulated air pollutant exposure has been declining during the period of increasing rates of ASD (Sullivan et al. 2018). Therefore, it is not plausible that these air pollutant exposures alone are increasing the risk of ASD. However, increasing prevalence of maternal characteristics such as gestational diabetes that are known risk factors for ASD (Xiang et al. 2015) have the potential to increase the fraction of ASD attributable to air pollution even in the context of stable or decreasing regional pollutant levels. We acknowledge that some MIA triggers, such as diabetes, may also be potential mediators in the association between air pollution and ASD. One recent study found little evidence of mediation effects of maternal immune biomarkers in the association between prenatal NO2 exposure and ASD in children (Volk et al. 2020). However, the study did not examine interactions. Future research might examine both mediating and modifying roles of MIA in air pollution-associated ASD risk. Because diet can be modified, dietary susceptibility to air pollution effects could have major public health significance (J.B. Adams et al. 2018). Sufficient folic acid and vitamin intake during pregnancy has been associated with reduced ASD risks in children (Levine et al. 2018; Schmidt 2013; Schmidt et al. 2019). However, only one study examined the role of folic acid 39 intake as a factor protecting against air pollution effects on autism (Goodrich et al. 2018). The timing of the largest protective effect of folic acid, during first trimester, suggests a window of vulnerability to pollution. Epigenetic dysregulation caused by air pollution is one pathway that may explain such effects (Marsit 2015; Lin, Baccarelli, and Burris 2016). Antioxidants in maternal diets can reduce the oxidative stress induced by air pollution (Kannan et al. 2006; Block et al. 2012). Other nutritional factors including vitamin D (Stubbs, Henley, and Green 2016) and polyunsaturated fatty acids are also antioxidants and may improve autism symptoms in children with ASD (Kristen Lyall et al. 2013). These also merit investigation as potential modifiers of air pollution effects. 2.5.4. Genetic Susceptibility to Air pollution-associated ASD There is substantial heritability in ASD (Tick et al. 2016; Tchaconas and Adesman 2013). Hundreds of genes have been identified in biological pathways that may contribute to ASD (Chaste and Leboyer 2012; Wisniowiecka-Kowalnik and Nowakowska 2019). However, we identified only two studies that examined gene-environment interaction (air pollution with CNVs and with MET CC genotype) on the risk of ASD. CNV duplication and deletion burden increased the prenatal ozone exposure association with ASD risk in children (Kim et al. 2017). A few animal studies have provided evidence for ozone-induced oxidative stress and autism-like behavior (Block et al. 2012; Bignami et al. 1994; Sorace et al. 2001). In humans, oxidative stress is a key feature of the ASD phenotype (James et al. 2006; Yui et al. 2016). It is possible that CNVs contribute to genes altering reactive oxygen species (Kushima et al. 2018). Thus, CNVs and prenatal air pollution exposure can potentially share common biological pathways leading to ASD. Another study found interactions of a specific functional polymorphism, MET receptor tyrosine kinase rs1858830 CC genotype, with prenatal near-roadway air pollution and regional 40 PM2.5, PM10, and NO2. Genetic studies have shown that children with MET CC genotype had decreased expression of MET protein in brain, which is associated with ASD (Campbell et al. 2006; Campbell et al. 2008). Animal experiments have also shown that prenatal exposure to PAH (a component of traffic-related air pollution and PM) causes both decreased MET protein expression in the brain and autism-like behavior in mouse offspring (Sheng et al. 2010). Therefore, it is plausible that decreased MET protein expression is a biological pathway common to air pollution exposures and ASD, and MET genotype may play a role in the air pollution association with ASD. Future studies may investigate the interaction between air pollution and ASD polygenic risk. One reason for the paucity of research on the interaction between genes and air pollution is that obtaining both genetic and environmental data for the same study population is expensive and time consuming (Kim et al. 2017). Opportunities exist for assigning early-life air pollution exposures to existing genetic studies that could advance our understanding of air pollution-gene interaction, including ASD polygenic risk, and biological pathways involved in ASD. 2.5.5. Quality of included studies Overall, most studies included in this review were rated as “low” or “probably low” in risk of bias, especially for the relatively low bias in the measurements of air pollution and of effect modifiers, lending credibility to the effects observed. However, some limitations of these studies emerged during the review. Large sample size is required to observe interaction effects for relatively uncommon diseases like ASD. Studies of air pollution-sex interaction generally did not identify statistically significant effects, potentially due to the small proportion of ASD cases in girls reducing the power to identify interactions. In the U.S., additional studies outside California could make the 41 results more representative. Future reviews might also include studies of continuous measures of autistic traits, which may allow for detection of subtle effect modification in smaller sample sizes and reduce potential bias due to disparities in disease diagnosis by sociodemographic characteristics. The role of effect modification was usually not the main research hypothesis for the studies reviewed; thus, the reported interactions may reflect selective reporting bias. The absence of a standard approach to reporting results of effect modification is another limitation. Some studies only reported the results of stratified analyses. Others reported the p-values of multiplicative interaction terms from statistical models. Only a few articles assessed interaction on both the additive and multiplicative scale. Reporting risk measures (relative risks, odds ratios, hazard ratios) for each stratum of exposure and the interaction on both additive and multiplicative scales is good practice (Knol and VanderWeele 2012). The studies by Goodrich et al. (2018) and McGuinn et al. (2019) are examples of reporting additive and multiplicative interactions. Finally, most studies do not explicitly describe the theoretical or conceptual frameworks supporting the study hypothesis and design for assessing effect modification. 2.5.6. Strengths and limitations of this review To our knowledge, this is the first systematic review summarizing the evidence on effect modifiers of the association between air pollution and the risk of ASD. The review followed the PRISMA checklist. The literature search was based on two most widely used databases, PubMed and Embase. The data extraction was thorough, including complementary results reported in supplementary materials. Rigorous assessment of risk of bias was conducted considering population representativeness, the potential for misclassification of exposure, ASD outcome, effect modifiers and the potential for selective reporting bias. 42 Because there were relatively few studies and results were not uniformly consistent, we did not attempt to conduct meta-analyses, and there were few definitive conclusions that could be drawn about how the association between air pollution and ASD was modified by the factors studied. We only included peer-reviewed studies published in English. It is possible that additional relevant information might be available in book chapters, non-English and grey literature. 2.6. Conclusions This review synthesized studies of potential effect modifiers of the association between ambient air pollution and the risk of ASD. Maternal gestational diabetes, insufficient folic acid intake during pregnancy, low maternal education, child male sex and high CNV burden or MET rs1858830 CC genotype may increase susceptibility to the effect of prenatal air pollution exposure on ASD risk. However, the limited number of studies on each modifier precludes definitive conclusions. Additional investigation of these susceptibility characteristics would benefit from consistent definitions across studies with sample sizes sufficient to identify interactions and a priori hypotheses, for example that exposure to maternal immune activators would increase effects of air pollutants on ASD. 43 2.7. Tables Table 2.1 Modification of regulated air pollutant (PM2.5, PM10, NO, NO2, SO2) associations with ASD by sex Studya and modifier Exposure Effect size by sex P Interactionb Effect size comparisonc Kalkbrenner et al. (2015) Boys (#case:828) Girls (#case: 151) PM10 prepregnancy per 10 �g/m3 Boys: OR=0.91 (0.78, 1.06) Girls: OR=0.99 (0.71, 1.38) 0.96 OR(boys)/OR(girls): 0.92 (0.64, 1.33) (p=0.65) PM10 1st trimester per 10 �g/m Boys: OR=0.89 (0.76, 1.04) Girls: OR=0.74 (0.50, 1.08) 0.12 OR(boys)/OR(girls): 1.20 (0.79, 1.82) (p=0.38) PM10 2nd trimester per 10 �g/m3 Boys: OR=1.00 (0.83, 1.19) Girls: OR=0.95 (0.63, 1.45) 0.81 OR(boys)/OR(girls): 1.05 (0.67, 1.66) (p=0.82) PM10 3rd trimester per 10 �g/m3 Boys: OR=1.33 (1.09, 1.63) Girls: OR=1.46 (0.92, 2.32) 0.21 OR(boys)/OR(girls): 0.91 (0.55, 1.51) (p=0.72) PM10 postpregnancy (0-3 months) per 10 �g/m3 Boys: OR=1.01 (0.82, 1.26) Girls: OR=1.52 (0.95, 2.44) 0.42 OR(boys)/OR(girls): 0.66 (0.40, 1.12) (p=0.12) PM10 postpregnancy (3-6 months) per 10 �g/m3 Boys: OR=0.75 (0.60, 0.93) Girls: OR=0.63 (0.39, 1.03) 0.19 OR(boys)/OR(girls): 1.19 (0.70, 2.03) (p=0.52) PM10 postpregnancy (6-9 months) per 10 �g/m3 Boys: OR=0.87 (0.70, 1.08) Girls: OR=0.85 (0.53, 1.38) 0.65 OR(boys)/OR(girls): 1.02 (0.61, 1.73) (p=0.93) PM10 postpregnancy (9-12 months) per 10 �g/m3 Boys: OR=1.14 (0.93, 1.41) Girls: OR=1.52 (0.95, 2.45) 0.11 OR(boys)/OR(girls): 0.75 (0.45, 1.26) (p=0.28) Raz et al. (2015) Boys (#case:137) Girls (#case: 23) PM2.5 pregnancy per 4.4 �g/m3 Boys: OR=1.73 (1.29, 2.31) Girls: OR=1.12 (0.59, 2.12) 0.17 OR(boys)/OR(girls): 1.54 (0.76, 3.12) (p=0.23) Raz et al. (2018) Boys (#case:1435) Girls (#case: 286) NO2 prepregnancy (9 months) per 5.85 ppb Boys: OR=1.22 (0.88, 1.67) Girls: OR= 0.75 (0.37, 1.53) Not reported OR(boys)/OR(girls): 1.63 (0.75, 3.54) (p=0.22) NO2 pregnancy per 5.85 ppb Boys: OR=0.51 (0.35, 0.76) Girls: OR= 1.46 (0.61, 3.46) Not reported OR(boys)/OR(girls): 0.35 (0.14, 0.90) (p=0.03) NO2 postpregnancy (9 months) per 5.85 ppb Boys: OR=1.63 (1.17, 2.27) Girls: OR=1.17 (0.57, 2.41) Not reported OR(boys)/OR(girls): 1.39 (0.63, 3.08) (p=0.41) Ritz et al. (2018) Boys (#case:11853) Girls (#case: 3534) NO2 postpregnancy (9 months) per 11.41 �g/m3 Boys: OR=1.06 (0.99, 1,14) Girls: OR=1.12 (0.99, 1.26) Not reported OR(boys)/OR(girls): 0.95 (0.82, 1.09) (p=0.44) SO2 postpregnancy (9 months) per 2.80 �g/m3 Boys: OR=1.25 (1.16, 1.35) Girls: OR=1.11 (0.97, 1.27) Not reported OR(boys)/OR(girls): 1.13 (0.96, 1.31) (p=0.13) PM10 postpregnancy (9 months) per 3.80 �g/m3 Boys: OR=1.05 (1.00, 1.10) Girls: OR=1.01 (0.93, 1.11) Not reported OR(boys)/OR(girls): 1.04 (0.94, 1.15) (p=0.45) PM2.5 postpregnancy (9 months) per 3.61 �g/m3 Boys: OR=1.07 (1.01, 1.13) Girls: OR=1.02 (0.92, 1.12) Not reported OR(boys)/OR(girls): 1.05 (0.94, 1.17) (p=0.41) 44 Table 2.1 cont. Jo, Eckel, Wang, et al. (2019)d Boys (#case:2030) Girls (#case: 441) PM2.5 pregnancy per 6.5 �g/m3 Boys: HR=1.25 Girls: HR=1.02 Not significant HR(boys)/HR(girls): 1.23 PM2.5 1st trimester per 6.5 �g/m3 Boys: HR=1.18 (1.08, 1.27) Girls: HR=0.90 (0.76, 1.07) 0.03 HR(boys)/HR(girls): 1.31 (1.09, 1.58) (p=0.005) PM2.5 2nd trimester per 6.5 �g/m3 Boys: HR=1.07 Girls: HR=1.12 Not significant HR(boys)/HR(girls): 0.96 PM2.5 3rd trimester per 6.5 �g/m3 Boys: HR=1.12 Girls: HR=1.04 Not significant HR(boys)/HR(girls): 1.08 PM2.5 postpregnancy (1 year) per 6.5 �g/m3 Boys: HR=1.25 Girls: HR=1.19 Not significant HR(boys)/HR(girls): 1.05 Pagalan et al. (2019)e Boys (#case:1091) Girls (#case: 216) PM2.5 pregnancy per 1.5 �g/m3 Boys: OR=1.04 (0.98, 1.10) Girls: OR=1.03 (0.90, 1.18) Not significant OR(boys)/OR(girls): 1.01 (0.87, 1.17) p=0.90) NO pregnancy per 10.7 ppb Boys: OR=1.09 (1.02, 1.15) Girls: OR=0.98 (0.83, 1.13) Not significant OR(boys)/OR(girls): 1.11 (0.94, 1.31) (p=0.21) NO2 pregnancy per 4.8 ppb Boys: OR=1.07 (1.00, 1.13) Girls: OR=1.00 (0.86, 1.16) Not significant OR(boys)/OR(girls): 1.07 (0.91, 1.26) (p=0.41) Al-Hamdan et al. (2018) The percentages of days with unhealthy AQI (AQI>100) per 10% Boys: OR=1.10 (0.65, 1.87) Girls: OR=1.06 (0.63, 1.78) Not reported OR(boys)/OR(girls): 1.04 (0.50, 2.18) (p=0.92) Abbreviations: HR, hazard ratio; OR, odds ratio a For additional details on study population, exposure measurement, outcome ascertainment, and covariates, see Supplemental Table 2.6. b P interaction column lists the qualitative or quantitative test results of interaction terms reported in each study. “Not reported” indicates the authors did not report the results either qualitatively or quantitatively. “Not significant” indicates the authors noted the results of interaction tests qualitatively or in figures, but the exact p-values were not reported. c Based on method described by Altman and Bland (2013). An example of calculation is shown in Supplemental Table 2.4. d Confidence intervals for the effect size of average PM2.5 during pregnancy and the 2nd and 3rd trimester were reflected in a figure. e Confidence intervals for the effect size of PM2.5, NO, and NO2 for each trimester were reflected in a figure. 45 Table 2.2 Modification of air pollutant (PM2.5, PM10, NO, NO2, CO, O3, traffic-related pollutants, airborne mercury) exposure associations with ASD by education level Studya and modifier Exposure Effect size by education level P Interactionb Effect size comparisonc Becerra et al. (2013) less than high school (#case: 1,725) high school (#case: 1,861) higher than high school (#case: 3,926) Traffic NO pregnancy per 9.40 ppb <high school: OR=1.11 (1.05, 1.18) =high school: OR=1.03 (0.97, 1.09) > high school: OR=0.99 (0.95, 1.03) not reported OR(low)/OR(moderate): 1.08 (0.99, 1.17) (p=0.08) OR(low)/OR(high): 1.12 (1.04, 1.20) (p=0.0016) Traffic NO2 pregnancy per 5.41 ppb < high school: OR=1.17 (1.10, 1.25) = high school: OR=1.06 (1.00, 1.13) > high school: OR=1.03 (0.99, 1.07) not reported OR(low)/OR(moderate): 1.10 (1.01, 1.21) (p=0.03) OR(low)/OR(high): 1.14 (1.05, 1.22) (p=0.0008) CO pregnancy per 0.55 ppm < high school: OR=0.90 (0.85, 0.96) = high school: OR=1.03 (0.97, 1.09) > high school: OR=1.09 (1.04, 1.14) not reported OR(low)/OR(moderate): 0.87 (0.80, 0.95) (p=0.0017) OR(low)/OR(high): 0.83 (0.77, 0.89) (p<0.0001) NO pregnancy per 29.67 ppb < high school: OR=0.96 (0.89,1.03) = high school: OR=1.02 (0.95, 1.09) > high school: OR=1.04 (0.99, 1.10) not reported OR(low)/OR(moderate): 0.94 (0.85, 1.04) (p=0.24) OR(low)/OR(high): 0.92 (0.84, 1.01) (p=0.08) NO2 pregnancy per 10.47 ppb < high school: OR=0.97 (0.90, 1.04) = high school: OR=1.08 (1.01, 1.16) > high school: OR=1.07 (1.02, 1.12) not reported OR(low)/OR(moderate): 0.90 (0.81, 0.99) (p=0.04) OR(low)/OR(high): 0.91 (0.83, 0.99) (p=0.03) Ozone pregnancy per 11.54 ppb < high school: OR=1.09 (1.02, 1.16) = high school: OR=1.07 (1.01, 1.14) > high school: OR=1.04 (0.99, 1.09) not reported OR(low)/OR(moderate): 1.02 (0.93, 1.11) (p=0.68) OR(low)/OR(high): 1.05 (0.97, 1.14) (p=0.25) PM2.5 pregnancy per 8.25 �g/m3 < high school: OR=1.04 (0.96, 1.12) = high school: OR=1.09 (1.01, 1.17) > high school: OR=1.06 (1.00, 1.12) not reported OR(low)/OR(moderate): 0.95 (0.86, 1.06) (p=0.39) OR(low)/OR(high): 0.98 (0.89, 1.08) (p=0.70) PM10 pregnancy per 4.68 �g/m3 < high school: OR=0.97 (0.91, 1.04) = high school: OR=1.08 (1.01, 1.16) > high school: OR=1.02 (0.97, 1.07) not reported OR(low)/OR(moderate): 0.90 (0.82, 0.99) (p=0.03) OR(low)/OR(high): 0.95 (0.88, 1.03) (p=0.23) Gong et al. (2017) ≤12 years of education (#case: 2,564) >12 years of education (#case: 2,571) Traffic NOx pregnancy per 10 �g/m3 ≤12 years: OR=0.99 (0.88, 1.12) >12years: OR=1.02 (0.94, 1.12) 0.65 OR(low)/OR(high): 0.97 (0.84, 1.13) (p=0.69) Traffic NOx postpregnancy (1st year of life) per 10 �g/m3 ≤12 years: OR=1.04 (0.91, 1.19) >12 years: OR=1.04 (0.94, 1.15) 0.99 OR(low)/OR(high): 1.00 (0.85, 1.18) (p=1.00) Traffic PM10 pregnancy per 20 �g/m3 ≤12 years: OR=0.89 (0.73, 1.08) >12 years: OR=1.04 (0.88, 1.23) 0.13 OR(low)/OR(high): 0.84 (0.66, 1.11) (p=0.24) Traffic PM10 postpregnancy (1st year of life) per 20 �g/m3 ≤12 years: OR=0.94 (0.76, 1.16) >12 years: OR=1.07 (0.89, 1.28) 0.24 OR(low)/OR(high): 0.88 (0.66, 1.16) (p=0.36) 46 Table 2.2 cont. Dickerson et al. (2016) Tracts below the average percent of college education (n=1459) Tracts above the average percent of college education (n=1030) Mercury 4th quantile vs. 1st quantile < average: RR=1.37 (0.75, 2.49) >average: RR=0.91 (0.50, 1.66) 0.02 OR(low)/OR(high): 1.51 (0.64, 3.52) (p=0.34) Abbreviations: HR, hazard ratio; OR, odds ratio a For additional details on study population, exposure measurement, outcome ascertainment, and covariates, see Supplemental Table 2.6. b P interaction column lists the qualitative or quantitative test results of interaction terms reported in each study. “Not reported” indicates the authors did not report interaction Pvalue either qualitatively or quantitatively. c Based on method described by Altman and Bland (2013). An example of calculation is shown in Supplemental Table 2.4. 47 Table 2.3 Modification of air pollutant (PM2.5, PM10, NO, NO2) exposure associations with ASD by neighborhood deprivation level Studya and modifier Exposure Effect size by neighborhood deprivation P Interactionb Effect size comparisonc Gong et al. (2017) high (#case: 2,030) moderate (#case: 1,643) low (#case: 1,463) Traffic NOx pregnancy per 10 �g/m3 high: OR=0.87 (0.72, 1.06) moderate: OR=1.02 (0.93, 1.12) low: OR=1.08 (0.96, 1.23) 0.16 OR(high)/OR(moderate): 0.85 (0.69, 1.06) (p=0.15) OR(high)/OR(low): 0.81 (0.64, 1.01) (p=0.07) Traffic NOx postpregnancy (1st year of life) per 10 �g/m3 high: OR=0.85 (0.67, 1.07) moderate: OR=1.05 (0.94, 1.16) low: OR=1.13 (0.97, 1.32) 0.09 OR(high)/OR(moderate): 0.81 (0.63, 1.05) (p=0.11) OR(high)/OR(low): 0.75 (0.57, 1.00) (p=0.05) Traffic PM10 pregnancy per 20 �g/m3 high: OR=0.86 (0.66, 1.11) moderate: OR=1.01 (0.84, 1.23) low: OR=1.13 (0.91, 1.42) 0.24 OR(high)/OR(moderate): 0.85 (0.62, 1.18) (p=0.33) OR(high)/OR(low): 0.76 (0.54, 1.07) (p=0.12) Traffic PM10 postpregnancy (1st year of life) per 20 �g/m3 high: OR=0.88 (0.67, 1.15) moderate: OR=1.06 (0.86, 1.29) low: OR=1.20 (0.93, 1.55) 0.20 OR(high)/OR(moderate): 0.83 (0.59, 1.16) (p=0.28) OR(high)/OR(low): 0.73 (0.51, 1.06) (p=0.10) McGuinn et al. (2019)d high (#case: 187) moderate (#case: 235) low (#case: 252) PM2.5 pregnancy per 5 �g/m3 high: OR=1.16 (0.63, 2.16) moderate: OR=0.97 (0.55, 1.71) low: OR=0.98 (0.58, 1.66) 0.79 OR(high)/OR(moderate): 1.20 (0.52, 2.76) (p=0.68) OR(high)/OR(low): 1.18 (0.53, 2.66) (p=0.68) PM2.5 postpregnancy (1st year of life) per 5 �g/m3 high: OR=2.45 (1.08, 5.56) moderate: OR=1.88 (0.83, 4.25) low: OR=1.83 (0.90, 3.70) 0.57 OR(high)/OR(moderate): 1.30 (0.41, 4.14) (p=0.65) OR(high)/OR(low): 1.34 (0.45, 3.95) (p=0.60) Abbreviations: HR, hazard ratio; OR, odds ratio. a For additional details on study population, exposure measurement, outcome ascertainment, and covariates, see Supplemental Table 2.6. b P interaction column lists the qualitative or quantitative test results of interaction terms reported in each study. c Based on method described by Altman and Bland (2013). An example of calculation is shown in Supplemental Table 2.4. d Additive interactions were examined based on dichotomized pregnancy or postpregnancy PM2.5 concentrations (≤ 12.0 �g/m3 or > 12.0 �g/m3 ). No significant additive interactions were found at 0.05 alpha level. 48 Table 2.4 Modification of air pollutant (PM2.5,PM10, NO2, O3, near-roadway pollution) exposure associations with ASD by maternal factors Studya and modifier Exposure Effect size P Interactionb Effect size comparisonc Goodrich et al. (2018)d Folic acid intake low: <800 ug (#case: 166) high: >800 ug (#case: 180) NRP* 1st trimester Dichotomous: <=15.58 ppb; >15.58 ppb Continuous: per 29.4 ppb low: OR=1.57 (0.92, 2.70) high: OR=0.92 (0.59, 1.45) 0.09 OR(low)/OR(high): 1.71 (0.85, 3.44) (p=0.14) NO2 1st trimester Dichotomous: <=14.18 ppb; >14.18 ppb Continuous: per 9.5 ppb low: OR=1.53 (0.91, 2.56) high: OR=0.74 (0.46, 1.19) 0.01 OR(low)/OR(high): 2.07 (1.02, 4.17) (p=0.04) PM10 1st trimester Dichotomous: <=22.88 �g/m3 ; >22.88 �g/m3 Continuous: per 13.32 �g/m3 low: OR=1.33 (0.81, 2.19) high: OR=0.94 (0.59, 1.49) 0.32 OR(low)/OR(high): 1.41 (0.72, 2.79) (p=0.32) PM2.5 1st trimester Dichotomous: <=12.40 �g/m3 ; >12.40 �g/m3 Continuous: per 10.02 �g/m3 low: OR=1.13 (0.70, 1.83) high: OR=0.97 (0.60, 1.59) 0.74 OR(low)/OR(high): 1.16 (0.59, 2.31) (p=0.66) Ozone 1st trimester Dichotomous: <=33.41 �g/m3 ; >33.41 �g/m3 Continuous: per 22.88 �g/m3 low: OR=1.07 (0.66, 1.73) high: OR=1.14 (0.71, 1.82) 0.92 OR(low)/OR(high): 0.94 (0.48, 1.84) (p=0.85) Jo, Eckel, Chen, et al. (2019) none (#case: 2,167), GDM* diagnosed < 24 weeks' gestation (#case: 73) GDM* diagnosed ≥ 24 weeks' gestation (#case: 160) pre-existing type 2 diabetes (#case: 71) Ozone preconception (12 weeks) per 15.7 ppb None: HR=0.98 (0.92, 1.03) GDM<24 weeks: HR=1.14 (0.76, 1.72) GDM≥24 weeks: HR=0.95 (0.77, 1.17) Pre-existing: HR=1.09 (0.85, 1.38) 0.71 HR (GDM<24)/HR(None): 1.16 (0.77, 1.76) (p=0.47) HR (GDM>=24)/HR(None): 0.97 (0.78, 1.20) (p=0.78) HR (pre-existing)/HR(None): 1.11 (0.87, 1.43) (p=0.40) Ozone pregnancy per 15.7 ppb None: HR=1.10 (0.95, 1.27) GDM<24 weeks: HR=1.96 (1.24, 3.11) GDM≥24 weeks: HR=0.83 (0.53, 1.29) Pre-existing: HR=1.14 (0.53, 2.46) 0.07 HR (GDM<24)/HR(None): 1.78 (1.10, 2.89) (p=0.02) HR (GDM>=24)/HR(None): 0.75 (0.47, 1.20) (p=0.24) HR (pre-existing)/HR(None): 1.04 (0.47, 2.26) (p=0.93) Ozone 1st trimester per 15.7 ppb None: HR=0.95 (0.90, 1.01) GDM<24 weeks: HR=1.50 (1.08, 2.09) GDM≥24 weeks: HR=0.90 (0.69, 1.18) Pre-existing: HR=1.07 (0.69, 1.64) 0.05 HR (GDM<24)/HR(None): 1.58 (1.13, 2.21) (p=0.01) HR (GDM>=24)/HR(None): 0.95 (0.72, 1.25) (p=0.70) HR (pre-existing)/HR(None): 1.13 (0.73, 1.74) (p=0.59) 49 Table 2.4 cont. Ozone 2nd trimester per 15.7 ppb None: HR=1.04 (0.98, 1.11) GDM<24 weeks: HR=1.29 (0.95, 1.76) GDM≥24 weeks: HR=0.86 (0.72, 1.03) Pre-existing: HR=1.05 (0.66, 1.65) 0.17 HR (GDM<24)/HR(None): 1.24 (0.91, 1.70) (p=0.18) HR (GDM>=24)/HR(None): 0.83 (0.68, 1.00) (p=0.05) HR (pre-existing)/HR(None): 1.01 (0.64, 1.60) (p=0.97) Ozone 3rd trimester per 15.7 ppb None: HR=1.06 (0.99, 1.12) GDM<24 weeks: HR=0.97 (0.73, 1.30) GDM≥24 weeks: HR=1.03 (0.78, 1.35) Pre-existing: HR=0.99 (0.73, 1.35) 0.85 HR (GDM<24)/HR(None): 0.92 (0.68, 1.23) (p=0.56) HR (GDM>=24)/HR(None):0.97 (0.73, 1.29) (p=0.84) HR (pre-existing)/HR(None): 0.93 (0.68, 1.28) (p=0.67) Ozone postpregnancy (1st year of life) per 15.7 ppb None: HR=0.93 (0.78, 1.10) GDM<24 weeks: HR=2.01 (0.67, 6.07) GDM≥24 weeks: HR=0.72 (0.50, 1.02) Pre-existing: HR=1.17 (0.63, 2.17) 0.01 HR (GDM<24)/HR(None): 2.16 (0.71, 6.59) (p=0.18) HR (GDM>=24)/HR(None):0.77 (0.52, 1.15) (p=0.20) HR (pre-existing)/HR(None): 1.26 (0.66, 2.39) (p=0.48) * Abbreviations in alphabetical order: GDM, gestational diabetes mellitus; HR, hazard ratio; NRP, near-roadway pollution; OR, odds ratio. a For additional details on study population, exposure measurement, outcome ascertainment, and covariates, see Supplemental Table 2.6. b P interaction column lists the quantitative test results of interaction terms reported in each study. c Based on method described by Altman and Bland (2013). An example of calculation is shown in Supplemental Table 2.4. d Additive interactions were examined based on dichotomized air pollutant concentrations at the median values. The cut off points for NRP, NO2, PM10, PM2.5, and ozone were 15.58 ppb, 14.18 ppb, 22.88 �g/m3 , 12.40 �g/m3 , and 33.41 �g/m3 , respectively. A significant additive interaction was found only between NO2 and folic acid. 50 Table 2.5 Modification of regulated air pollutant (PM2.5,PM10, NO2, O3, traffic-related pollution) exposure effects of ASD by children’s genotypes Studya Air pollutants Genotypes Interaction Effect Size (95% CI) P interactionb Kim et al. (2017)c TRP per 17.7 ppb CNV burdend : Duplications OR=0.95 (0.73, 1.23) Not reported NO2 per 5.7 ppb CNV burden: Duplications OR=0.94 (0.77, 1.15) Not reported Ozone per 6.2 ppb CNV burden: Duplications OR=1.55 (1.09, 2.21) Not reported PM10 per 6.2 �g/m3 CNV burden: Duplications OR=0.93 (0.73, 1.19) Not reported PM2.5 per 3.7 �g/m3 CNV burden: Duplications OR=0.95 (0.72, 1.24) Not reported TRP per 17.7 ppb CNV burden: Deletions OR=0.81 (0.59, 1.11) Not reported NO2 per 5.7 ppb CNV burden: Deletions OR=0.96 (0.73, 1.26) Not reported Ozone per 6.2 ppb CNV burden: Deletions OR=1.06 (0.83, 1.36) Not reported PM10 per 6.2 �g/m3 CNV burden: Deletions OR=0.98 (0.73, 1.33) Not reported PM2.5 per 3.7 �g/m3 CNV burden: Deletions OR=0.83 (0.61, 1.13) Not reported TRP per 17.7 ppb CNV burden: Total OR=0.88 (0.68, 1.15) Not reported NO2 per 5.7 ppb CNV burden: Total OR=0.92 (0.75, 1.14) Not reported Ozone per 6.2 ppb CNV burden: Total OR=1.36 (1.01, 1.81) Not reported PM10 per 6.2 �g/m3 CNV burden: Total OR=0.87 (0.68, 1.11) Not reported PM2.5 per 3.7 �g/m3 CNV burden: Total OR=0.84 (0.65, 1.08) Not reported Volk et al. (2014)e TRP >=30.2 ppb MET CC OR=2.9 (1.0, 10.6) 0.09 TRP < 30.2 ppb MET CG/GG OR=1.0 (reference) NO2 >=17.5 ppb MET CC OR=3.6 (1.3, 12.7) 0.03 NO2 <17.5 ppb MET CG/GG OR=1.0 (reference) Ozone >=41.8 ppb MET CC OR=0.95 (0.42, 2.2) Not reported Ozone <41.8 ppb MET CG/GG OR=1.0 (reference) PM10>=29.2 �g/m3 MET CC OR=3.2 (1.3, 9.1) Not reported PM10<29.2 �g/m3 MET CG/GG OR=1.0 (reference) PM2.5>=16.0 �g/m3 MET CC OR=2.1 (0.92, 5.4) Not reported PM2.5<16.0 �g /m3 MET CG/GG OR=1.0 (reference) Abbreviations: CNV, copy number variation; OR, odds ratio a For additional details on study population, exposure measurement, outcome ascertainment, and covariates, see Supplemental Table 2.6. b P interaction column lists the qualitative or quantitative test results of interaction terms reported in each study. “Not reported” indicates the authors did not report the results either qualitatively or quantitatively. c Each air pollutant was measured as an average from pregnancy to child’s first 2 years of life in Kim et al. (2017). d CNV burden was evaluated as a continuous variable. The OR for the interaction was per increase of 1356513 base pairs of CNV burden. e Each air pollution was measured as pregnancy average in Volk et al. (2014). 51 2.8. Figures Figure 2.1 Screening of studies for inclusion based on PRISMA criteria 52 2.9. Supplemental Material Supplemental Table 2.1 PubMed search syntax PubMed ("Air Pollutants"[MeSH Terms] OR "Air Pollution"[MeSH Terms] OR ("air"[All Fields] AND "pollu*"[All Fields]) OR ("Particulate Matter"[MeSH Terms] OR "Particulate Matter"[All Fields] OR "PM"[All Fields] OR "PM10"[All Fields] OR "PM2.5"[All Fields] OR "PM1"[All Fields] OR "PM0.1"[All Fields]) OR ("Nitrogen Dioxide"[MeSH Terms] OR "Nitrogen Oxides"[MeSH Terms] OR "Nitrogen Dioxide"[All Fields] OR "Nitrogen Oxides"[All Fields] OR "NO2"[All Fields] OR "NO"[All Fields]) OR ("Ozone"[MeSH Terms] OR "Ozone"[All Fields] OR "O3"[All Fields]) OR ("Sulfur Dioxide"[MeSH Terms] OR "Sulfur Dioxide"[All Fields] OR "sulphur dioxide"[All Fields] OR "SO2"[All Fields]) OR ("Carbon Monoxide"[MeSH Terms] OR "Carbon Monoxide"[All Fields]) OR ("Vehicle Emissions"[MeSH Terms] OR ("traffic"[All Fields] AND ("air"[All Fields] AND "pollu*"[All Fields])) OR (("vehicle"[All Fields] OR "traffic"[All Fields] OR "automobile"[All Fields]) AND ("emissions"[All Fields] OR "emission"[All Fields])))) AND ("Autism Spectrum Disorder"[MeSH Terms] OR "Asperger Syndrome"[MeSH Terms] OR "Autistic Disorder"[MeSH Terms] OR "child development disorders, pervasive"[MeSH Terms] OR "autis*"[All Fields] OR "ASD"[All Fields] OR "Autism Spectrum Disorder"[All Fields] OR "Asperger Syndrome"[All Fields] OR "Autistic Disorder"[All Fields] OR "child development disorders pervasive"[All Fields] OR "PDD"[All Fields] OR "PDDNOS"[All Fields] OR "PDD-NOS"[All Fields]) Supplemental Table 2.2 Embase search syntax No. Query Results Date #10 #8 AND #9 654 1-May-21 #9 'autism'/exp OR 'asperger syndrome'/exp OR 'pervasive developmental disorder not otherwise specified'/exp OR autis* OR 'asd' OR 'autism spectrum disorder' OR asperger* OR 'autistic disorder' OR 'pervasive development disorders' OR 'pdd' OR 'pddnos' OR 'pdd-nos' 109937 1-May-21 #8 #1 OR #2 OR #3 OR #4 OR #5 OR #6 OR #7 416869 1-May-21 #7 'exhaust gas'/exp OR (traffic AND air AND pollu*) OR ((vehicle OR traffic OR automobile) AND (emissions OR emission)) 33282 1-May-21 #6 'carbon monoxide'/exp OR 'carbon monoxide' 56552 1-May-21 #5 'sulfur dioxide'/exp OR 'sulfur dioxide' OR 'sulphur dioxide' OR 'so2' 20768 1-May-21 #4 'ozone'/exp OR ozone OR 'o3' 45303 1-May-21 #3 'nitrogen dioxide'/exp OR 'nitrogen oxide'/exp OR 'nitrogen dioxide' OR 'nitrogen oxide' OR 'no2' 31688 1-May-21 #2 'particulate matter'/exp OR 'particulate matter' OR 'pm' 146357 1-May-21 #1 'air pollutant'/exp OR 'air pollutant' OR 'air pollution'/exp OR 'air pollution' OR (('air'/exp OR 'air' OR 'air'/exp OR air) AND pollu*) 213508 1-May-21 53 Supplemental Table 2.3 Detailed Criteria for the qualitative assessment of risk of bias Criteria and considerations Population representativeness • Volunteer recruitment • Hospital-based or birth records Comparability between included and excluded population Air pollution exposure measurements • Accuracy of the air pollution modelling approach • Spatial and temporal resolution • Accounted for changes in addresses during pregnancy Whether the exposure concentrations were extrapolated to times beyond the available observation data Outcome ascertainment • Cases identified by questionnaire or extracted from medical records or detailed clinical evaluation • Case ascertainment validated Sensitivity analysis was conducted to account for potential misclassification of cases Identification of modifiers • Quality of the data sources to generate the measures of the effect modifiers Clear criteria for dichotomizing measures of effect modifiers Confounders • Adjusted, at a minimum, for maternal age and the calendar year of birth • For studies not examining sociodemographic characteristics as effect modifiers, these factors were considered as covariates Whether additional included covariates were appropriate for adjustment Selective outcome reporting Whether the results of effect modification were reported with sufficient detail Note: These criteria were generated based on the Instructions for making risk of bias determinations proposed by Lam et al. (2016) Supplemental Table 2.4 An example of comparing two estimated hazard ratios Group 1 (Boys) Group 2 (Girls) HR 1.18 0.90 Log HR 0.1655 -0.1054 95% CI for HR 1.08, 1.27 0.76, 1.07 95% CI for log HR 0.0770, 0.2390 -0.2744, 0.0677 Width of CI 0.1620 0.3421 SE=width/(2*1.96) 0.0413 0.0873 Difference between log odds ratios d=E1-E2 0.1655 – (-0.1054) = 0.2709 SE(d) $0.0413! + 0.0873! = 0.0966 CI(d) 0.2709 ± 1.96*0.0966 0.0816, 0.4602 Test of interaction Z=0.2709 / 0.0966=2.8043 (p=0.005) Ratio of hazard ratios (RHR) RHR=exp(d) exp(0.2709)=1.31 CI(RHR) exp(0.0816), exp(0.4602) 1.09, 1.58 Notes: Adapted from Altman and Bland (2003). The data in this example are effect sizes of PM2.5 during the first trimester of ASD risks among boys and girls reported by Jo, Eckel, Wang, et al. (2019). The calculations in this table show how to compare two hazard ratios through point estimations and confidence intervals. 54 Supplemental Table 2.5 Excluded Publications on air pollution and autism due to the lack of any effect modifiers 1 Blanchard KS, Palmer RF, Stein Z. 2011. The value of ecologic studies: Mercury concentration in ambient air and the risk of autism. Rev Environ Health 26:111-118. 2 Chen G, Jin Z, Li S, Jin X, Tong S, Liu S, et al. 2018. Early life exposure to particulate matter air pollution (PM(1), PM(2.5) and PM(10)) and autism in shanghai, china: A case-control study. Environ Int 121:1121-1127. 3 Dickerson AS, Rahbar MH, Han I, Bakian AV, Bilder DA, Harrington RA, et al. 2015. Autism spectrum disorder prevalence and proximity to industrial facilities releasing arsenic, lead or mercury. Sci Total Environ 536:245-251. 4 Frye RE, Cakir J, Rose S, Delhey L, Bennuri SC, Tippett M, et al. 2020. Prenatal air pollution influences neurodevelopment and behavior in autism spectrum disorder by modulating mitochondrial physiology. Mol Psychiatry. 5 Geng R, Fang S, Li G. 2019. The association between particulate matter 2.5 exposure and children with autism spectrum disorder. Int J Dev Neurosci 75:59-63. 6 Gong T, Almqvist C, Bölte S, Lichtenstein P, Anckarsäter H, Lind T, et al. 2014. Exposure to air pollution from traffic and neurodevelopmental disorders in Swedish twins. Twin Res Hum Genet 17:553-562. 7 Guxens M, Ghassabian A, Gong T, Garcia-Esteban R, Porta D, Giorgis-Allemand L, et al. 2016. Air pollution exposure during pregnancy and childhood autistic traits in four European population-based cohort studies: The ESCAPE project. Environ Health Perspect 124:133-140. 8 Jung CR, Lin YT, Hwang BF. 2013. Air pollution and newly diagnostic autism spectrum disorders: A populationbased cohort study in Taiwan. PLoS ONE 8. 9 Kalkbrenner AE, Windham GC, Serre ML, Akita Y, Wang X, Thayer BP, et al. 2013. Exposure to ambient course particulate matter by gestational period and autism spectrum disorders in North Carolina and California. Am J Epidemiol 177:S156. 10 Kaufman JA, Wright JM, Rice G, Connolly N, Bowers K, Anixt J. 2019. Ambient ozone and fine particulate matter exposures and autism spectrum disorder in metropolitan Cincinnati, Ohio. Environ Res 171:218-227. 11 Kerin T, Volk H, Li W, Lurmann F, Eckel S, McConnell R, et al. 2018. Association between air pollution exposure, cognitive and adaptive function, and ASD severity among children with autism spectrum disorder. J Autism Dev Disord 48:137-150. 12 McGuinn LA, Windham GC, Kalkbrenner AE, Bradley C, Di Q, Croen LA, et al. 2020. Early life exposure to air pollution and autism spectrum disorder: Findings from a multisite case-control study. Epidemiology 31:103-114. 13 Oudin A, Frondelius K, Haglund N, Källén K, Forsberg B, Gustafsson P, et al. 2019. Prenatal exposure to air pollution as a potential risk factor for autism and ADHD. Environ Int 133:105149. 14 Talbott EO, Arena VC, Rager JR, Clougherty JE, Michanowicz DR, Sharma RK, et al. 2015. Fine particulate matter and the risk of autism spectrum disorder. Environ Res 140:414-420. 15 Talbott EO, Marshall LP, Rager JR, Arena VC, Sharma RK, Stacy SL. 2015. Air toxics and the risk of autism spectrum disorder: The results of a population based case-control study in southwestern Pennsylvania. Environ Health 14:80. 16 Volk HE, Hertz-Picciotto I, Delwiche L, Lurmann F, McConnell R. 2011. Residential proximity to freeways and autism in the CHARGE study. Environ Health Perspect 119:873-877. 17 Volk HE, Lurmann F, Penfold B, Hertz-Picciotto I, McConnell R. 2013. Traffic-related air pollution, particulate matter, and autism. JAMA Psychiatry 70:71-77. 18 Volk HE, Park B, Hollingue C, Jones KL, Ashwood P, Windham GC, et al. 2020. Maternal immune response and air pollution exposure during pregnancy: Insights from the early markers for autism (ema) study. J Neurodev Disord 12:42. 19 Windham GC, Zhang L, Gunier R, Croen LA, Grether JK. 2006. Autism spectrum disorders in relation to distribution of hazardous air pollutants in the San Francisco bay area. Environ Health Perspect 114:1438-1444. 20 Yousefian F, Mahvi AH, Yunesian M, Hassanvand MS, Kashani H, Amini H. 2018. Long-term exposure to ambient air pollution and autism spectrum disorder in children: A case-control study in Tehran, Iran. Sci Total Environ 643:1216-1222. 55 Supplemental Table 2.6 Characteristics of included studies, grouped by study design and alphabetically ordered by author surname for each design Study Study Design Study Population Sample Size Exposure Outcome Modifier Covariates NOS scorea Becerra et al. (2013) Casecontrol Cases were from Los Angeles County, California born in 1995-2006 from the state DDS* system. Controls were from birth records; cases and controls matched on birth year and sex. cases: 7,594 controls: 75,635 Residence CO, NO2, NO, O3, PM10, PM2.5 from nearest monitoring station; NO and NO2 also estimated by LUR* models based on ground measurements during 2006-2007; average for each trimester and the entire pregnancy based on daily average concentrations. Autistic disorder Linked to California DDS* services records (diagnosis based on DSM-IVR*) Validation shown in other studies Maternal education (less than high school, high school, and higher than high school) from birth records maternal age, maternal place of birth, race/ethnicity; type of birth (single, multiple), parity; insurance type; and gestational age at birth (weeks). S=4 C=2 E=2 Total: 8 Gong et al. (2017) Casecontrol The source population for both cases and controls are children born and living in Stockholm County, Sweden between 1993 and 2007 from the Stockholm Youth Cohort cases: 5,136 controls: 18,237 Annual residence PM10 and NOx for the entire pregnancy period and the first year after birth from a Gaussian air quality dispersion model. PM10 based on data in 2004. NOx based on data in 1990, 1995, 2000, 2002- 2004. ASD (with or without intellectual disability) Linked to records from the National Patient Register (NPR), the Clinical Database for Child and Adolescent Psychiatry in Stockholm (PASTILL), the Habilitation Register (HAB), and the Stockholm Regional Health Care Data warehouse (diagnosis based on DSM-IV*) Parental Education (>12 years of education and <=12 years of education) from longitudinal database, Statistics Sweden Neighborhood deprivation (least deprived, intermediate deprived, and most deprived), an index defined by small-area market statistics considering a series of SES* data sex, birth month, sibling order, maternal age, paternal age, mother’s marital status, parents’ birth countries, mother’s education, father’s education, mother’s employment, father’s employment, disposable income within household, neighborhood deprivation, calendar year of birth, and municipality of birth S=4 C=2 E=2 Total: 8 56 Supplemental Table 2.6 cont. Goodrich et al. (2018) Casecontrol Cases identified from children 2-5 years of age born in California between 1997- 2008 from CHARGE* study. Controls identified from California Vital Statistics birth records. Cases and controls are matched on sex, age, and geographic region. cases: 346 controls: 260 Historical residential locations' near roadway air pollution from CALINE4* between 1995 and 2000. PM2.5, PM10, Ozone, and NO2 from nearby EPA* air quality stations between 1997 to 2009, average for three trimesters and the entire pregnancy based on monthly average concentrations. ASD SCQ* for case screening and ADOS* and ADI-R* for case ascertainment Linked to California DDS* services records Folic acid intake (high: >800 ug and low: <800 ug) from telephone interviews self-reported financial hardship between 3 months before pregnancy to time of interview (yes/no), child’s year of birth, vitamin A and zinc intake during the first month of pregnancy S=4 C=2 E=2 Total: 8 Kalkbrenner et al. (2010) Casecontrol Cases identified from children aged 8 years from ADDM* sites in North Carolina (born in 1994 and 1996) and West Virginia (born in 1992 and 1994). Controls identified from children aged 8 years with speech and language impairment in the surveillance system in North Carolina and West Virginia .Cases and controls matched on access to developmental evaluations. cases: 383 controls: 2,829 (children with speech and language impairment) Perinatal annual average exposure of 35 hazardous air pollutants from NATA* model in 1996 based on residential census tracts. ASD (with or without intellectual disability) Linked to ADDM* network records (diagnosis based on DSM-IV-TR*) Census tract level of urbanicity (rural, mixed, urban), no description of data source and classification criteria Child's sex from birth certificate data race, maternal education, maternal age, smoking in pregnancy, marital status, census tract median household income, and urbanicity S=3 C=2 E=2 Total: 7 57 Supplemental Table 2.6 cont. Kalkbrenner et al. (2015) Casecontrol Both cases and controls are identified from children aged 8 years from North Carolina (born in 1994, 1996, 1998, and 2000) and Bay Area, California (born in 1996) cases: 979 (North Carolina: 645, California: 334) controls: 14,666 (North Carolina: 12,434, California: 2,232) Residence PM10 from 1 year before the delivery to 1 year after the birth from a kriging geostatistical method based on EPA* monitoring stations' measurements, average for preconception period, the entire pregnancy, each trimester, and 4 postnatal quarters based on daily concentrations. ASD (with or without intellectual disability; AD* and ASDNOS*) Linked to ADDM* network records (diagnosis based on DSM-IV-TR*) Child's sex from birth certificate data Season of birth, race/ethnicity, maternal education, maternal age, census 2000 block group median household income, census 2000 block group urbanization, and a nonparametric term for the calendar week of the child’s birth S=3 C=2 E=2 Total: 7 Kalkbrenner et al. (2018) Casecontrol Both cases and controls are identified from children born between 1994 and 2007 from AGRE* program, a volunteer repository of families across the U.S. cases: 1,540 controls: 477 Historical residential locations' perinatal exposure of 155 hazardous air pollutants’ annual average concentrations from NATA* model in 1996, 1999, 2002, and 2005 ASD Screening based on parent interview tool, ADI-R* Quantitative measure of autistic traits by SRS* Severity evaluated by CSS* Child's sex from AGRE* records individual level: year of birth, race/ ethnicity, number of siblings in the family, and maternal age Census block group: population density, percentage with a high school education or more, and median rent S=3 C=2 E=2 Total: 7 Kim et al. (2017) Casecontrol Cases identified from children 2-5 years of age born in California between 1997- 2008 from CHARGE* study. Controls identified from California Vital Statistics birth records. Cases and controls are matched on sex, age, and geographic region. cases: 158 controls: 147 Historical residential locations' near roadway air pollution from CALINE4* between 1995 and 2000. PM2.5, PM10, Ozone, and NO2 from nearby EPA* air quality stations between 1997 to 2009, average for the entire pregnancy and child’s first and second year of life based on daily average concentrations. ASD SCQ* for case screening and ADOS* and ADI-R* for case ascertainment Linked to California DDS* services records Copy Number Variation (CNV) burden (duplication, deletion, or collectively), detected by hotspot array from whole blood maximum education level of parents, child’s sex and child’s ethnicity S=4 C=2 E=2 Total: 8 58 Supplemental Table 2.6 cont. McGuinn et al. (2019) Casecontrol Cases identified from children 30- 68 months of age born in California, Colorado, Georgia, North Carolina, Pennsylvania, and Maryland, USA born in 2003 to 2006 selected from SEED*. Controls identified from state birth records within the same catchment area. cases: 674 controls: 855 Residence PM2.5 from a hybrid prediction model incorporated satellite imagery, monitoring stations, land use, etc., from 2002 to 2007, average for the entire pregnancy and the 1st year of life based on daily average concentrations. ASD SCQ* for case screening ADOS* and ADI-R* for case ascertainment Linked to SEED* records Neighborhood (census tract) deprivation (low, moderate, and high), categorized on tertile cut points of an index based on 8 tract-level SES* parameters. study site, year of birth, month of birth, maternal age, maternal race/ethnicity, maternal education, and maternal smoking S=4 C=2 E=2 Total: 8 Raz et al. (2015) Casecontrol Both cases and controls are children from the Nurses’ Health Study II Cohort born between 1990 and 2002 in the US cases: 245 controls: 1,522 Non-movers: cases: 160 controls: 986 Residence PM2.5, PM10, PM10-2.5 from nationwide spatiotemporal models based on data from EPA monitors, average for 9 months before conception, the entire pregnancy, each trimester, and 9 months after birth based on monthly average concentrations. ASD Reported by mothers Validation based on ADI-R* and SRS* to 50 sample cases Child's sex from maternal self-report. child’s birth year, birth month, and sex, maternal age at birth, paternal age at birth, and median census tract income in the birth year S=3 C=2 E=2 Total:7 59 Supplemental Table 2.6 cont. Raz et al. (2018) Casecontrol Both cases and controls are children born in the central coastal area of Israel between 2005 and 2009 from the National Insurance Institute of Israel cases: 2,098 controls: 54,191 The small statistical area level or city-level residence NO2 from the optimized dispersion model based on air pollution monitoring data and meteorological records, average for 9 months before conception, the entire pregnancy, and 9 months after birth based on weekly average concentrations. ASD Linked to National Insurance Institute records (diagnosis based on DSM-IV*) Child's sex from the National Insurance Institute of Israel year of birth, calendar month of birth, population group, paternal age and census poverty index S=4 C=2 E=2 Total: 8 Ritz et al. (2018) Casecontrol Both cases and controls are children born in Denmark between 1989 and 2013 from the Danish medical registers. cases: 15,387 controls: 68,139 Residence NO2, PM2.5, PM10, and SO2 from a GIS*-based air pollution and human exposure modeling system (AirGIS), average for 9 months before conception, the entire pregnancy, each trimester, and 9 months after birth based on hourly concentrations. ASD (AD, Asperger, and PDD-NOS*) Linked to the Danish National Patient Register and the Danish Psychiatric Central Register (diagnosis based on ICD-10*) Child's sex from the Danish Medical Birth Registry maternal age, paternal age, location of birth, neighborhood SES employment, neighborhood SES housing, and maternal smoking during pregnancy S=4 C=2 E=2 Total: 8 Roberts et al. (2013) Casecontrol Both cases and controls are children from the Nurses’ Health Study II Cohort born in U.S. between 1987 and 2002 cases: 325 controls:22,101 Census tract level of residence 14 kinds of hazardous air pollutants’ annual average concentrations from the NATA in 1990, 1996, 1999, and 2002, average for the entire pregnancy. ASD Reported by mothers Validation based on ADI-R* and SRS* to 50 sample cases Child's sex from maternal self-report. maternal age at birth, year of birth, maternal parents’ education, Census tract median income, Census tract percent college educated, and HAP model year S=3 C=2 E=2 Total: 7 60 Supplemental Table 2.6 cont. Volk et al. (2014) Casecontrol Both cases and controls are children from the CHARGE* study who aged 24 to 60 months born in California cases: 251 controls: 156 Historical residential locations' traffic-related pollution from CALINE4* model and ambient PM2.5, PM10, ozone, and NO2 from EPA* Air Quality System data, average for the entire pregnancy based on monthly average concentrations. ASD SCQ*, MSEL*, and VABS* for case screening and ADOS* and ADI-R* for case ascertainment The MET receptor tyrosine kinase (MET) genotype (CC, CG/GG) from blood samples child’s sex, child’s ethnicity, maternal education, maternal age, and home ownership S=4 C=2 E=2 Total: 8 Jo, Eckel, Chen, et al. (2019) Cohort Singleton children 5 years of age born in Kaiser Permanente Southern California (KPSC) hospitals from 1999 to 2009 in 14 service areas across Southern California, with 5- year follow up total:246,420 cases: 2,471 controls: 243,949 Residential PM2.5, PM10, NO2, and O3 for 12weeks before pregnancy, each trimester, the entire pregnancy and the first year of life based on monthly average concentrations from four closest stations in the EPA* regional air quality monitoring network. ASD Screening based on CHAT* and diagnosis from pediatric developmental specialist Identification from EMR* ICD-9* code Maternal diabetes (none, Gestational diabetes mellitus (GDM) diagnosed < 24 weeks' gestation, GDM diagnosed >= 24 weeks' gestation, preexisting type 2 diabetes) from EMR* birth year, KPSC medical center service areas, maternal age, parity, maternal race/ethnicity, maternal education, census tract median household income, maternal history of comorbidities before pregnancy, child sex, and family specified as a random effect S=3 C=2 O=3 Total: 8 61 Supplemental Table 2.6 cont. Jo, Eckel, Wang, et al. (2019) Cohort Singleton children aged 5 years born in Kaiser Permanente Southern California (KPSC) hospitals from 1999 to 2009 in 14 service areas across Southern California, with 5- year follow up total: 246,420 cases: 2,471 controls: 243,949 Residence PM2.5, PM10, NO2, and O3 for each trimester, the entire pregnancy and the first year of life based on monthly average concentrations from four closest stations in the EPA* regional air quality monitoring network. ASD Screening based on CHAT* and diagnosis from pediatric developmental specialist Identification from EMR* ICD-9* code Children's sex from EMR* birth year, KPSC medical center service areas, maternal age, parity, maternal race/ethnicity, maternal education, census tract median household income, maternal history of comorbidities before pregnancy, child sex, and family specified as a random effect S=3 C=2 O=3 Total: 8 Pagalan et al. (2019) Cohort Children with mothers registered with the provincial health insurance plan born in Metro Vancouver, British Columbia, Canada between 2004 and 2009, with 5-year minimum followup total: 132,256 cases: 1,307 controls: 130,949 Zip code level residence PM2.5, NO, and NO2 from a LUR* model with station monitoring data from Metro Vancouver’s Air Quality Monitoring Network, average for each trimester and the entire pregnancy based on monthly average concentrations. ASD Linked to the British Columbia Autism Assessment Network (BCAAN) (diagnosis based on ADOS* and ADI-R*) Child's sex from birth records child sex, birth month, birth year, maternal age, maternal birthplace, and neighborhoodlevel urbanicity and income band S=3 C=2 O=3 Total: 8 von Ehrenstein et al. (2014) Cohort Children born in Los Angeles County, California between 1995 and 2006 whose mother resided in 5-km buffer around monitoring stations cases: 768 controls: 147,954 Residence 24 toxic air pollutants from the nearest monitoring station, average for each trimester and the entire pregnancy based on monthly average concentrations. ASD (with or without language disability) Linked to California DDS* services records (diagnosis based on DSM-IV*). Validation shown in other studies Child's sex from birth records birth year, maternal race/ethnicity, maternal age and education, type of insurance, place of birth mother, child sex, parity S=3 C=2 O=3 Total: 8 62 Supplemental Table 2.6 cont. Al-Hamdan et al. (2018) Ecological Children aged 8 years from 15 ADDM* sites, USA, born between 2000- 2012 15 ADDM* sites 22,153 ASD cases 2,097,188 general population Percentage of days with unhealthy air quality index (AQI) for each site ASD prevalence per 1000. Biennial ASD cases data for 15 sites within the USA for the years 2000 to 2012 obtained from the ADDM* Network of CDC* Child’s sex race (Whites, Blacks, Hispanics, Asians) Average cigarette smoking rate for each site County-level household income, sunlight, heat index, PM2.5 Dickerson et al. (2016) Ecological Children aged 8 years born in Arizona, Maryland, New Jersey, South Carolina, and Utah, USA in 2000, 2002, 2004, 2006, and 2008 2,489 census tracts 4,486 ASD cases Annual average air concentrations of inorganic arsenic, lead, mercury and combined by census tract from NATA* ASD prevalence by census tract Biennial ASD cases data by census tract for 5 states (Arizona, Maryland, New Jersey, South Carolina, and Utah) within the USA from years 2000 to 2008 obtained from the ADDM Network Poverty (% poverty below average, % poverty above average) Education (% college educated below average, % college educated above average) percentage of White, percentage of Hispanic, percent below poverty, and percentage with a college degree * Abbreviations in alphabetical order: AD, Autistic Disorder; ADDM, Autism and Developmental Disabilities Monitoring Network; ADI-R, Autism Diagnostic Interview-Revised; ADOS, Autism Diagnostic Observation Schedule; AGRE, Autism Genetic Resource Exchange; ASD-NOS, Autism Spectrum Disorder-Not Otherwise Specified; ASD, Autism Spectrum Disorder; CHARGE, Childhood Autism Risk from Genetics and the Environment; CHAT, Checklist for Autism in Toddlers; CO, carbon monoxide; CSS, Calibrated Severity Score; DDS, Department of Developmental Services; DSM, Diagnostic and Statistical Manual of; Mental Disorders; EMR, Electronic medical records; EPA, Environmental Protection Agency; GIS, Geographic Information System; ICD, International Classification of Diseases; LUR, Land-use Regression; MSEL, Mullen Scales of Early Learning; NATA, National-Scale Air Toxics Assessment; NO, nitric oxide; NO2, nitrogen dioxide; NOS, Newcastle-Ottawa Quality Assessment Scale; NOx, nitrogen oxides; O3, ozone; PDD-NOS, Pervasive Developmental Disorder – Not Otherwise Specified; PM, particulate matter; PM10 , PM ≤ 10 µm in aerodynamic diameter; PM10–2.5 , PM between 10 and 2.5 µm in aerodynamic diameter; PM2.5 , PM ≤ 2.5 µm in aerodynamic diameter; SCQ, Social Communication Questionnaire; SEED, Study to Explore Early Development; SES, Socioeconomic Status; SO2, sulfur dioxide; SRS, Social Responsiveness Scale; VABS, Vineland Adaptive Behavior Scale. a Newcastle-Ottawa Quality Assessment Scale (NOS) score column records scores for each section and the total score (see Section 2.3). 63 Supplemental Table 2.7 Assessment of risk of bias Study Population representativeness Air pollution Exposures Outcomes Modifiers Confounders Selective outcome reporting Becerra et al. (2013) probably low probably low probably low Maternal education: probably low low Probably low Gong et al. (2017) low probably low probably low Parental education: probably low Neighborhood deprivation: probably low low low Goodrich et al. (2018) probably low probably low low Folic acid intake: probably high low low Kalkbrenner et al. (2010) low probably high probably low Urbanicity: probably high Child's sex: low low probably low Kalkbrenner et al. (2015) low probably low probably low Child's sex: low low low Kalkbrenner et al. (2018) probably high probably high probably low Child's sex: low low low Kim et al. (2017) probably low probably low low copy number variations: low probably low low McGuinn et al. (2019) probably low probably low low Neighborhood deprivation: probably low low low Raz et al. (2015) probably low probably low probably low Child's sex: low low probably low Raz et al. (2018) low probably low probably low Child's sex: low probably low probably low Ritz et al. (2018) low Low probably low Child's sex: low low Probably low Roberts et al. (2013) probably low Probably high probably low Child's sex: low low low Volk et al. (2014) low probably low low MET: low probably low low Jo, Eckel, Chen, et al. (2019) probably low probably low low Maternal diabetes: probably low low low Jo, Eckel, Wang, et al. (2019) probably low probably low low Child's sex: low low low Pagalan et al. (2019) low probably low low Child's sex: low low low von Ehrenstein et al. (2014) probably low probably low probably low Child's sex: low low low 64 Supplemental Table 2.8 Modification of air toxics associations with ASD by sex Studya Exposure Case number Effect size P Interactionb Effect size comparisonc Roberts et al. (2013) Case-control Overall metalsd Quitile 5 vs Quintile 1 Boys (#case: 279) Girls (#case: 46) Boys: OR=1.7 (1.1, 2.6) Girls: OR=0.8 (0.2, 2.4) <0.01 OR(boys)/OR(girls): 2.13 (0.57, 7.91) (p=0.26) Antimony Quitile 5 vs Quintile 1 Boys (#case: 209) Girls (#case: 35) Boys: OR=1.7 (1.1, 2.7) Girls: OR=0.3 (0.1, 1.5) <0.01 OR(boys)/OR(girls): 5.67 (1.36, 23.60) (p=0.02) Cadmium Quitile 5 vs Quintile 1 Boys (#case: 279) Girls (#case: 46) Boys: OR=1.6 (1.1, 2.4) Girls: OR=0.7 (0.3, 2.1) <0.10 OR(boys)/OR(girls): 2.29 (0.80, 6.52) (p=0.12) Lead Quitile 5 vs Quintile 1 Boys (#case: 279) Girls (#case: 46) Boys: OR=1.7 (1.2, 2.6) Girls: OR=0.9 (0.4, 2.4) <0.10 OR(boys)/OR(girls): 1.89 (0.71, 5.01) (p=0.20) Nickel Quitile 5 vs Quintile 1 Boys (#case: 279) Girls (#case: 46) Boys: OR=1.9 (1.2, 2.9) Girls: OR=0.7 (0.2, 2.2) <0.05 OR(boys)/OR(girls): 2.71 (0.76, 9.74) (p=0.13) Trichloroethylene Quitile 5 vs Quintile 1 Boys (#case: 279) Girls (#case: 46) Boys: OR=1.4 (1.1, 2.3) Girls: OR=0.6 (0.1, 2.7) <0.10 OR(boys)/OR(girls): 2.33 (0.43, 12.63) (p=0.33) Kalkbrenner et al. (2018) Case-control 1,2,4-Trichlorobenzene Quartile 4 vs Quartile 1 Boys (#case: 1210) Girls (#case: 330) Boys: OR=1.47 (1.08, 2.00) Girls: OR=1.23 (0.92, 1.65) 0.08 OR(boys)/OR(girls): 1.20 (0.78, 1.83) (p=0.41) 2,4,6-Trichlorophenol Quartile 4 vs Quartile 1 Boys (#case: 1210) Girls (#case: 330) Boys: OR=3.27 (0.9, 10.82) Girls: OR=1.56 (0.49, 4.95) 0.06 OR(boys)/OR(girls): 2.10 (0.40, 11.06) (p=0.38) 2,4-Toluene diisocyanate Quartile 4 vs Quartile 1 Boys (#case: 1210) Girls (#case: 330) Boys: OR=1.80 (1.12, 2.91) Girls: OR=1.26 (0.82, 1.94) 0.06 OR(boys)/OR(girls): 1.43 (0.75, 2.72) (p=0.28) Dibutylphthalate Quartile 4 vs Quartile 1 Boys (#case: 1210) Girls (#case: 330) Boys: OR=1.16 (0.80, 1.67) Girls: OR=0.89 (0.62, 1.27) 0.05 OR(boys)/OR(girls): 1.30 (0.78, 2.18) (p=0.31) Diesel Particulate Matter Quartile 4 vs Quartile 1 Boys (#case: 1210) Girls (#case: 330) Boys: OR=1.48 (1.04, 2.09) Girls: OR=1.20 (0.85, 1.70) 0.08 OR(boys)/OR(girls): 1.23 (0.75, 2.02) (p=0.40) Ethylidene dichloride (1,1- Dichloroethane) Quartile 4 vs Quartile 1 Boys (#case: 1210) Girls (#case: 330) Boys: OR=0.46 (0.30, 0.72) Girls: OR=0.75 (0.50, 1.11) 0.01 OR(boys)/OR(girls): 0.61 (0.34, 1.11) (p=0.11) 65 Supplemental Table 2.8 cont. Hexachlorobenzene Quartile 4 vs Quartile 1 Boys (#case: 1210) Girls (#case: 330) Boys: OR=1.17 (0.95, 1.44) Girls: OR=0.97 (0.79, 1.19) 0.02 OR(boys)/OR(girls): 1.21 (0.90, 1.62) (p=0.21) Hydrochloric acid Quartile 4 vs Quartile 1 Boys (#case: 1210) Girls (#case: 330) Boys: OR=1.24 (0.80, 1.92) Girls: OR=0.96 (0.62, 1.49) 0.09 OR(boys)/OR(girls): 1.29 (0.70, 2.40) (p=0.42) Pentachlorophenol Quartile 4 vs Quartile 1 Boys (#case: 1210) Girls (#case: 330) Boys: OR=0.97 (0.37, 2.50) Girls: OR=0.53 (0.21, 1.36) 0.05 OR(boys)/OR(girls): 1.83 (0.48, 6.96) (p=0.38) Phenol Quartile 4 vs Quartile 1 Boys (#case: 1210) Girls (#case: 330) Boys: OR=1.00 (0.60, 1.66) Girls: OR=0.75 (0.46, 1.21) 0.08 OR(boys)/OR(girls): 1.33 (0.66, 2.69) (p=0.42) Xylenes Quartile 4 vs Quartile 1 Boys (#case: 1210) Girls (#case: 330) Boys: OR=1.59 (1.07, 2.39) Girls: OR=1.27 (0.85, 1.89) 0.09 OR(boys)/OR(girls): 1.25 (0.71, 2.21) (p=0.44) von Ehrenstein et al. (2014) Cohort Lead per 13.64 ng/m3 Boys (#case: 291) Girls (#case: 57) Boys: OR=1.63 (1.31, 2.01) Girls: OR=0.98 (0.60, 1.61) not reported OR(boys)/OR(girls): 1.66 (0.97, 2.85) (p=0.06) PAHs per 0.79 ppb Boys (#case: 453) Girls (#case: 104) Boys: OR=0.94 (0.75, 1.17) Girls: OR=1.62 (0.97, 2.68) not reported OR(boys)/OR(girls): 0.58 (0.33, 1.01) (p=0.05) Note: OR: odds ratio; PAH, Polycyclic aromatic hydrocarbons a Kalkbrenner et al. (2010) reported the results from stratified analysis by gender for butadiene, mercury compounds, styrene, and tetrachloroethane, but none of the differences in effect size were significantly different between boys and girls; thus their results were not tabulated. b P interaction column lists quantitative test results of interaction terms reported in each study; “Not reported” indicates the authors did not report the results either qualitatively or quantitatively. c Based on method described by Altman and Bland (2013). d Overall score for metal exposure 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). 66 Chapter 3 Prenatal Air Pollution, Maternal Immune Activation, and Autism Spectrum Disorder This chapter has been published as Yu, X., M. M. Rahman, S. A. Carter, J. C. Lin, Z. Zhuang, T. Chow, F. W. Lurmann, M. J. Kleeman, M. P. Martinez, A. van Donkelaar, R. V. Martin, S. P. Eckel, Z. Chen, P. Levitt, J. Schwartz, D. Hackman, J. C. Chen, R. McConnell, and A. H. Xiang. 2023. "Prenatal air pollution, maternal immune activation, and autism spectrum disorder." Environ Int 179: 108148. 3.1. Abstract Background: Autism Spectrum Disorder (ASD) risk is highly heritable, with potential additional non-genetic factors, such as prenatal exposure to ambient particulate matter with aerodynamic diameter< 2.5 µm (PM2.5) and maternal immune activation (MIA) conditions. Because these exposures may share common biological effect pathways, we hypothesized synergistic associations of prenatal air pollution and MIA-related conditions would increase ASD risk in children. Objectives: This study examined interactions between MIA-related conditions and prenatal PM2.5 or major PM2.5 components on ASD risk. Methods: In a population-based pregnancy cohort of children born between 2001-2014 in Southern California, 318,751 mother-child pairs were followed through electronic medical records (EMR); 4,559 children were diagnosed with ASD before age 5. Four broad categories of MIA-related conditions were classified, including infection, hypertension, maternal asthma, and autoimmune conditions. Average exposures to PM2.5 and four PM2.5 components, black carbon (BC), organic matter (OM), nitrate (NO3 ⁻ ), and sulfate (SO4 2⁻), were estimated at maternal 67 residential addresses during pregnancy. We estimated the ASD risk associated with MIA-related conditions, air pollution, and their interactions, using Cox regression models to adjust for covariates. Results: ASD risk was associated with MIA-related conditions [infection (hazard ratio 1.11; 95% confidence interval 1.05-1.18), hypertension (1.30; 1.19-1.42), maternal asthma (1.22; 1.08-1.38), autoimmune disease (1.19; 1.09-1.30)], with higher pregnancy PM2.5 [1.07;1.03-1.12 per interquartile (3.73 μg/m3 ) increase] and with all four PM2.5 components. However, there were no interactions of each category of MIA-related conditions with PM2.5 or its components on either multiplicative or additive scales. Conclusions: MIA-related conditions and pregnancy PM2.5 were independently associations with ASD risk. There were no statistically significant interactions of MIA conditions and prenatal PM2.5 exposure with ASD risk. 3.2. Introduction Autism spectrum disorder (ASD) is characterized by deficits in social communication and interaction, and restricted, repetitive and stereotyped behaviors (American Psychiatric Association 2013b). The estimated prevalence of ASD is 1 in 36 among 8-year-old children in the United States (Maenner et al. 2023). Although ASD risk is heritable, only approximately 15- 20% of diagnoses are due to spontaneous single gene or chromosomal mutations (Bai et al. 2019; Rylaarsdam and Guemez-Gamboa 2019); other factors that may contribute to increased risk or clinical severity, are not well understood and are likely multifactorial. Diagnoses of ASD can be made starting around age 2, and subtle social and communication impairment may be present earlier (Bacon et al. 2018; Charman and Baird 2002). Therefore, prenatal and early life environment has been the focus of autism environmental epidemiology. 68 Accumulating evidence indicates that prenatal ambient air pollutants, especially particulate matter (PM) with aerodynamic diameter < 2.5 µm (PM2.5), are modifiable environmental risk factors for ASD (Jo, Eckel, Wang, et al. 2019; Chun et al. 2020; Lam et al. 2016; Rahman et al. 2022). PM2.5 is a complex mixture of solid and liquid particles with varying sizes, chemical composition, and toxicity (K. Adams et al. 2015; Fong et al. 2019). A previous study from our group found that prenatal exposure to some key components of PM2.5 such as black carbon (BC), organic matter (OM), nitrate (NO3 - ), and sulfate (SO4 2- ) were associated with increased risk of ASD (Rahman, Carter, Lin, Chow, Yu, Martinez, Chen, et al. 2023). Understanding the effects of different PM2.5 components can help to better develop sourcespecific ambient air quality standards and prevention strategies. Maternal infection (Zerbo et al. 2013; Jiang et al. 2016), hypertension (Maher et al. 2018), asthma (Theoharides et al. 2016), and autoimmune disease (K. Lyall et al. 2014) during pregnancy have been shown consistently to be associated with ASD and other neuropsychiatric or neurodevelopmental disorders in children (Estes and McAllister 2016; Malkova et al. 2012). These maternal health conditions may disturb the immune function of mothers and trigger systemic inflammation during pregnancy (Simoes et al. 2018; Patterson 2011). Therefore, they have been proposed to be proxies for maternal immune activation (O’Connor and Ciesla 2022; Estes and McAllister 2016). Maternal immune activation is a potentially useful framework for understanding ASD neurobiology (Meldrum et al. 2013), but most children exposed to health conditions that are proxies for maternal immune activators (MIA) do not develop ASD. It has been hypothesized that subsequent additional exposures are required for ASD symptoms to occur, and co-existing risk factors sharing similar or complementary biological pathways may provide sufficient cause 69 (Estes and McAllister 2016; Bilbo et al. 2018). Several studies have shown separate associations between prenatal environmental exposures including ambient air pollution (Rahman et al. 2022), nutritional intake (M. Li et al. 2019), social environment (Bhasin and Schendel 2007), and maternal health conditions (Zerbo et al. 2013; K. Lyall et al. 2014) during pregnancy on ASD risk, but there has been little examination of the interactive effects among these risk factors on ASD (Yu et al. 2022). Animal models have shown that PM and MIA conditions are both associated with impaired social communication in offspring (Carlezon et al., 2019; Jones et al., 2020; Klocke et al., 2018; Klocke et al., 2017; Zhang et al., 2018), and with increased oxidative stress and inflammatory cytokine levels (Leni, Künzi, and Geiser 2020; S.E. Smith et al. 2007). Therefore, we hypothesized that the co-exposure of prenatal PM2.5 or PM2.5 components and MIA-related conditions may have interactive effects on the subsequent risk of ASD. Synergies between MIA-related conditions and air pollution-induced ASD would have important implications for a large co-burden of disease from these exposures and for the development of clinical and public health interventions to reduce ASD risk. We tested this hypothesis using data from a large, population-based Southern California pregnancy cohort, because a large sample size is required to detect interactions between coexposures. Information on ASD and diagnoses of maternal health conditions, proxies of MIA, were extracted from the electronic medical record (EMR). The large Southern California gradients in air pollution exposure during pregnancy were assessed using high-resolution spatiotemporal hybrid models (Van Donkelaar et al. 2019). 70 3.3. Methods 3.3.1. Study population This study utilized a population-based retrospective pregnancy cohort that included mothers with singleton deliveries (n=370,723) at Kaiser Permanente Southern California (KPSC) hospitals between January 1, 2001 and December 31, 2014. KPSC is a large integrated healthcare system with over 4.5 million members across Southern California. KPSC membership reflects the diverse socioeconomic demographics in the study region (Koebnick et al. 2012). Information related to the mothers, including maternal address history, and to the children were extracted from high-quality integrated electronic medical records (EMR) maintained by KPSC. Addresses were geocoded using ArcGIS (ArcGIS 2021). Addresses based only on street name, 5-digit postal code, locality, or administrative unit were considered too uncertain to be geolocated into the correct grid used for exposure assignments. A total of 51,152 births were excluded due to 1) missing gender, maternal race/ethnicity and age at delivery, implausible age of delivery or birth weight (n=666); 2) maternal age at delivery (n=159); 3) incomplete maternal residential address history in pregnancy or geocodes not suitable for exposure assignment (n=51,147). The final data analysis included 318,751 mother-child pairs with complete data on residential estimates of PM2.5 composition exposures (Supplemental Figure 3.1). Both KPSC and the University of Southern California Institutional Review Boards approved this study with waiver of individual subject consent. 3.3.2. ASD ascertainment The outcome was ASD diagnosis before age 5. Children were followed from birth through EMR until clinical diagnosis of ASD, loss to follow-up, or age 5, whichever came first, as described previously (Coleman et al. 2015; Jo, Eckel, Wang, et al. 2019; Xiang et al. 2018; 71 Xiang et al. 2015). The presence or absence of ASD in children was identified by International Classification of Diseases, Ninth Revision (ICD- 9) codes 299.0, 299.1, 299.8, 299.9 from the EMR records before October 1, 2015 (the date of KPSC implementation of ICD-10 codes) and subsequently ICD-10 codes F84.0, F84.5, F84.9 F84.0, F84.3, F84.5, F84.8, F84.9. 3.3.3. Air pollution exposure assessment Monthly estimates of PM2.5 and four major PM2.5 chemical components [BC, OM, NO3 - , and SO4 2- ] with a 1 km spatial resolution were estimated by a hybrid model (version V4.NA.02) that integrates chemical transport model outputs, satellite observations, and ground-based measurements as developed by the Atmospheric Composition Analysis Group at Washington University in St. Louis. The exposure model (Van Donkelaar et al. 2019) and its application to this cohort (Rahman, Carter, Lin, Chow, Yu, Martinez, Chen, et al. 2023), are described elsewhere. Exposures to PM2.5 and these selected components were assigned to maternal address during the entire pregnancy and each of three trimesters. Exposures were time-weighted to account for changes in maternal addresses during pregnancy. 3.3.4. Conditions triggering maternal immune activation We considered four broad categories of MIA-related conditions during pregnancy: infections, gestational hypertension, asthma, and autoimmune diseases. Any occurrence (incident or recurring) during pregnancy of each category of conditions was operationalized as a yes/no variable. The pregnancy period was defined as the time between the last menstrual period and the date of delivery. Autoimmune conditions were identified from 1 year before pregnancy and the pregnancy period, considering that most autoimmune diseases are chronic with long-lasting effects. 72 Since the cohort included births from 2001 to 2014, prior to KPSC adoption of ICD-10 codes (in 2015), only ICD-9 codes were used to identify each of the MIA-related conditions diagnoses during pregnancy (Supplemental Table 3.1). The ICD-9 codes for maternal infection during pregnancy were based on Zerbo et al. (2013). The ICD-9 codes for hypertension were from Savitz et al. (2014). Asthma was categorized based on the ICD-9 code 493.xx and the usage of medication for asthma during pregnancy (Martinez et al. 2020). Maternal autoimmune disease classification was based on L. A. Croen et al. (2005), to which we added antiphospholipid antibody syndrome, aplastic anemia, dermatitis herpetiformis, giant cell arteritis, hemolytic anemia, Kawasaki’s disease, and Sydenham chorea [from K. Lyall et al. (2014)]. 3.3.5. Covariates We included the priori known potential risk factors for ASD including child sex, maternal age at delivery, parity, and maternal history of severe comorbidities [>1 diagnosis of heart, lung, kidney, liver disease, or cancer] (Xiang et al. 2015). We also adjusted for birth years as a nonlinear term with 4 degrees of freedom to account for the non-linear increasing trend of ASD prevalence. An indicator variable for the season at conception (dry from April-October; wet from November-March) was used to adjust for the potential air pollution seasonality. In a sensitivity analysis adjusting for month of birth (instead of season), associations of air pollutants and of MIA-related conditions with ASD risk were similar, so only the adjustment for season was included in the final analyses. Maternal race/ethnicity, maternal education [≤high school, some college, and ≥ college] and neighborhood disadvantage index (B.L. Levy, Owens, and Sampson 2019) at birth have been shown to be associated with ASD in this cohort (Yu et al. 2024) and were adjusted as socioeconomic covariates in this study. A missing indicator variable was used for missing values in categorical covariates (parity [n = 17,860], education [n = 3,024]). We used 73 the same set of covariates for models assessing ASD associations with MIA-related conditions during pregnancy, air pollution, and interactions to ensure that the estimated associations were comparable under the same modeling strategy. 3.3.6. Statistical analyses The main associations of ASD with average air pollution exposures, any MIA (the presence of at least one category of MIA-related conditions) and 4 types of MIA-related conditions during pregnancy and each trimester were estimated as hazard ratios (HRs) using multivariable Cox proportional hazard models, adjusting for covariates described above. Children were followed from birth through the EMR until clinical diagnosis of ASD, loss to follow-up, or age 5, whichever came first. Standard errors were estimated using robust sandwich estimators to control for potential correlation within families. Air pollution exposures were modeled as continuous variables with linear effects, as modeling them using spline functions did not show evidence of non-linearity. Pregnancy air pollution exposures were scaled to the interquartile range (IQR) of each air pollutant to reflect the distribution of air pollutants in this cohort. Each category of MIA-related conditions (e.g., any infection) was represented as a binary indicator (1=present; 0=absent). The proportional hazards assumption of Cox proportional hazard models was assessed using the Schoenfeld residual plot (Schoenfeld 1982). No clear nonrandom patterns with follow-up time were observed. We tested the interactions between each indicator of MIA-related conditions (the presence of any MIA during pregnancy or 4 indicators for each category) and PM2.5 and its components during pregnancy on ASD risk. The multiplicative interaction was examined by adding a multiplicative interaction term between the pollutant and MIA-related conditions to the Cox proportional hazard model, adjusting for the main effects and covariates. The associations 74 between air pollutants and ASD for each MIA condition strata (present/absent) were estimated based on the multiplicative interaction models by reparametrizing the indicator of MIA. The additive interaction was estimated by post-hoc relative excess risk due to interaction (RERI) between each indicator MIA-related conditions variable and air pollution (continuous variables per IQR increase) using the method described in VanderWeele and Knol (2014) based on estimated coefficients from interaction models. Sensitivity analysis was conducted including all follow up until December 31, 2019 without censoring at age 5. Two-sided statistical tests were applied at an alpha level of 0.05 and the uncertainty in estimates was reported by 95% confidence intervals (CIs). Data analyses were performed in R, version 4.2. 3.4. Results There were 4,559 children (1.4%) diagnosed with ASD before age 5 (Table 3.1). The median age at diagnosis was 3.53 years. Children with ASD were 4.3 times more likely to be boys (n=3,703) than girls (n=856). Mothers of children with ASD were slightly older at delivery (31.3 years, IQR=27.5-35.3) than mothers of children without ASD (30.4 years, IQR=26.2-34.3). Mothers of children with ASD were more likely to have more than high school education. The prevalence of gestational infection, hypertension, asthma, autoimmune diseases, and any MIA condition was 48.5%, 9.7%, 7.2%, 11.2%, and 59.4% respectively for the entire cohort. The prevalence of all five MIA indicators was generally higher in mothers of children with ASD than mothers of children without ASD. PM2.5 concentrations, NO3 - , SO4 2- and BC during the study period decreased over time, while OM was relatively stable (Supplemental Figure 3.2). The median levels of PM2.5, BC, OM, NO3 - , and SO4 2- , during pregnancy were 15.1 micrograms per meter-cubed (μg/m3 ) (IQR=13.1- 75 16.9), 1.83 μg/m3 (IQR=1.35-2.20), 6.40 μg/m3 (IQR=5.41-7.39), 2.79μg/m3 (IQR=2.33-3.49), and 1.47 μg/m3 (IQR=1.23-1.73), respectively. Pregnancy average PM2.5 was associated with ASD risk [HR=1.07 (95% CI 1.03-1.12) per IQR=3.73 μg/m3 ] after adjustment for child sex, maternal race/ethnicity, maternal age at delivery, parity, maternal education, maternal history of pre-pregnancy severe comorbidities, neighborhood disadvantage index, birth year, and season at conception (Table 3.2), consistent with previous analysis in a subset of this cohort (Jo, Eckel, Wang, et al. 2019). The presence of MIA-related conditions during pregnancy was associated with higher risk of ASD [infection: HR=1.11 (95% CI 1.05-1.18); hypertension: HR=1.30 (95% CI 1.19-1.42); asthma: HR=1.22 (95% CI 1.08-1.38); autoimmune conditions: HR=1.19 (95% CI 1.09-1.30); any MIA: HR=1.17 (95% CI 1.10-1.25), respectively after adjusting for covariates. No statistically significant interactions were observed between the five indicators of MIA-related conditions (4 categories and any MIA) and pregnancy average PM2.5 on either multiplicative or additive scales (Table 3.3). The interactions between MIA-related conditions and pregnancy-average exposure of each PM2.5 component exposure were also not statistically significant, except for the interaction between SO4 2- and autoimmune disease indicator (pinteraction=0.002) (Figure 3.1). For interactions between trimester specific PM2.5 components exposures and MIA indicators (Supplemental Figure 3.3, Supplemental Figure 3.4, and Supplemental Figure 3.5), only the interaction between the second trimester average SO4 2- and the autoimmune disease indicator was significant (p-interaction=0.002). However, if these interaction p-values were adjusted for false discovery rate for 100 comparisons [for 5 MIA indicators, 4 components and 4 exposure windows (all pregnancy and 3 trimester exposures)] the 76 threshold p-value for significance would be 0.0005 after the Bonferroni correction, so the isolated SO4 2- interactions cannot be considered statistically significant. We also examined the independent association of ASD risk with pregnancy PM2.5 exposure and all four categories of MIA-related conditions in one co-adjusted model. The associations of ASD risk with pregnancy PM2.5 exposure, each MIA-related conditions indicator and ASD were attenuated but remained statistically significant in the co-adjusted model (Supplemental Table 3.2). Results were similar in sensitivity analysis with 6,366 ASD cases during follow up without censoring at age 5 (Supplemental Table 3.3 and Supplemental Table 3.4). 3.5. Discussion Higher levels of average PM2.5 exposures during pregnancy and the presence of each of the four broad categories of conditions that are maternal immune activators were associated with ASD risk. However, we did not observe associations of ASD risk in children with interactions between prenatal PM2.5 or its components and MIA-related conditions either on the multiplicative or additive scales. The conceptual framework of MIA has been proposed in recent years for studying the effects of co-occurring prenatal environmental exposures with fetal neurodevelopment (O’Connor and Ciesla 2022; Estes and McAllister 2016). However, to our knowledge, this is the first large population-based cohort study explicitly examining the interactions of prenatal particulate air pollution exposure with a broad spectrum of maternal clinical conditions during pregnancy that are proxies of maternal immune activation on the risk of ASD in offspring. A previous cohort study from our group reported interactions between gestational diabetes mellitus and O3 exposure on ASD risk (Jo, Eckel, Chen, et al. 2019). One case-control study showed that 77 maternal immune biomarkers mediated associations between prenatal air pollution exposure and ASD risk (Volk et al. 2020). However, interactions between immune biomarkers or proxies of MIA and air pollution were not examined in that case-control study. PM2.5 is comprised of a heterogeneous mixture of solid and liquid particles with varying chemical composition that may have different toxicity. We have previously found that prenatal exposure to PM2.5 components BC, OM, NO3 - , SO4 2⁻were associated with increased ASD risk (Rahman, Carter, Lin, Chow, Yu, Martinez, Chen, et al. 2023). Studies have shown that the association between prenatal PM2.5 and fetal neurodevelopment may vary by exposure time windows (Pagalan et al. 2019; Raz et al. 2015). In the cohort used for this study, our previous research reported that PM2.5 exposure during the first and second trimesters was associated with higher ASD risk (Rahman et al. 2022). Therefore, we also examined the interactions between trimester specific PM2.5 exposure and MIA-related conditions. Similar patterns of associations were found as with the pregnancy average PM2.5. Overall, no effect modification by MIA-related conditions on the association between PM components and ASD risk was found in our study. Combinations of different component factors, such as PM2.5 and MIA-related conditions, may be required for disease development (Rothman 1976). Thus, it has been hypothesized that multiple risk factors sharing similar or complementary biological mechanisms are required for ASD to occur (Estes and McAllister 2016; Bilbo et al. 2018). Evidence from animal and in vitro studies has shown that both proxies of MIA and prenatal PM exposure may induce neuroinflammation of glial cells and oxidative stress, leading to impaired neurodevelopment (Baines et al. 2020; Morris et al. 2021; Di Domenico et al. 2020). Children with ASD also exhibit metabolic disruption, with increased oxidative stress (James et al. 2004; James et al. 2006; Gorrindo et al. 2013) and chronic neuroimmune activation in the central nervous system 78 (Onore, Careaga, and Ashwood 2012; Vargas et al. 2005; Voineagu et al. 2011). Thus, it has been hypothesized that co-existing MIA-related conditions and high PM exposure may have synergistic effects on ASD risk. Although our results did not support this hypothesis, this “multiple-exposure” framework may be useful for identifying other multifactorial risk factors for developing ASD. The present study has several strengths. The large study population provided statistical power to assess even modest associations of ASD risk with interactions between MIA-related conditions and prenatal air pollution, had these effects been present. Mother-child pairs were followed through the EMR in a single integrated healthcare system with standard diagnostic criteria, which helped avoid screening and ascertainment bias. The high-quality EMR also provided relevant confounders in an ethnically diverse sample of children. Current state-of-theart high spatiotemporal air pollution exposure models were used to assign prenatal ambient exposures accounting for residential mobility. We also acknowledge some limitations. Beyond clinical diagnoses of maternal immune activators, we lack biomarkers for MIA, which would be expensive to collect for a large cohort. Clinical symptoms and diagnoses of MIA-related conditions can often indicate immune dysregulation in mothers (Mor and Cardenas 2010), which may influence fetal development (Simoes et al. 2018; Patterson 2011). Thus, using the diagnoses of clinical conditions as proxies of MIA in a large population can also provide insights into the potential etiology of ASD. We also acknowledge the limitation of using ICD codes to extract information on maternal health conditions during pregnancy. We lack information on the severity of MIA-related conditions and may misclassify MIA-related conditions if mothers did not come for care if they had mild symptoms. Other epidemiological studies have examined interactions of air pollution with 79 nutritional and genetic risks (Goodrich et al. 2018; Volk et al. 2014; Kim et al. 2017; Yu et al. 2022), but information on these risks was not available in this large cohort. Changes in residential addresses during pregnancy were accounted for in the pollutant exposure assignment, but the exposures away from home at work or elsewhere were unavailable. Since our main hypothesis was about interactions among MIA-related conditions and prenatal air pollution during pregnancy on ASD, we did not assess mediation of air pollution effects by MIA-related conditions. In the co-adjusted model with four categories of clinical conditions and pregnancy PM2.5 (Supplemental Table 3.2) the estimated effect sizes were similar to the single exposure models (Table 3.2), suggesting that these effects of MIA-related conditions and PM2.5 were independent. 3.6. Conclusion In conclusion, our findings do not support the hypothesis that ASD risk in children is associated with multiplicative or additive interaction in analyses of a broad spectrum of MIArelated conditions during pregnancy and prenatal air pollution exposure in this large populationbased cohort. However, PM2.5 and four broad categories of conditions related to MIA were independently associated with ASD risk. 80 3.7. Tables Table 3.1 Characteristics of children, with and without autism spectrum disorder (ASD) Children, No. (%) or median [interquartile range] Characteristics Overall With ASD Without ASD (n =318,751) (n= 4,559) (n= 314,192) Sex = Male (%) 163,182 (51.2) 3,703 (81.2) 159,479 (50.8) Maternal age at delivery, 30.4 [26.3, 34.3] 31.3 [27.5, 35.3] 30.4 [26.2, 34.3] median [IQR*], years Maternal immune activation (MIA) conditions; N(%) Infection 154,894 (48.6) 2,396 (52.6) 152,498 (48.5) Hypertension 30,762 (9.7) 616 (13.5) 30,146 (9.6) Asthma 22,860 (7.2) 429 (9.4) 22,431 (7.1) Autoimmune diseases 35,669 (11.2) 651 (14.3) 35,018 (11.1) Any MIA 189,373 (59.4) 2,977 (65.3) 186,396 (59.3) Parity; N (%) 0 111,981 (35.1) 1,844 (40.4) 110,137 (35.1) 1 104,561 (32.8) 1,495 (32.8) 103,066 (32.8) >=2 84,176 (26.4) 903 (19.8) 83,273 (26.5) Unknown 18,033 (5.7) 317 (7.0) 17,716 (5.6) Maternal Education; N(%) High school or lower 112,096 (35.2) 1,335 (29.3) 110,761 (35.3) Some college 94,525 (29.7) 1,477 (32.4) 93,048 (29.6) College graduate or higher 109,087 (34.2) 1,713 (37.6) 107,374 (34.2) Unknown 3,043 (1.0) 34 (0.7) 3,009 (1.0) Neighborhood disadvantage index [IQR]a 0.07 [-1.22, 1.48] 0.07 [-1.31, 1.53] 0.07 [-1.21, 1.48] Race/ethnicity; N (%) Non-Hispanic white 81,050 (25.4) 956 (21.0) 80,094 (25.5) Non-Hispanic black 29,773 (9.3) 447 (9.8) 29,326 (9.3) Hispanic 161,415 (50.6) 2,300 (50.4) 159,115 (50.6) Asian/Pacific Islander 39,974 (12.5) 744 (16.3) 39,230 (12.5) Other 6,539 (2.1) 112 (2.5) 6,427 (2.0) Any history of maternal severe comorbiditiesb ; N (%) 46,717 (14.7) 839 (18.4) 45,878 (14.6) Year of birth; N (%) 2001-2005 78,257 (24.6) 818 (17.9) 77,439 (24.6) 2006-2010 111,174 (34.9) 1,308 (28.7) 109,866 (36.0) 2011-2014 129,320 (40.6) 2,433 (53.4) 126,887 (40.4) *Abbreviations: IQR, interquartile range. a Census tract level neighborhood disadvantage index. Higher values represent more disadvantaged neighborhoods b >=1 diagnosis of heart, lung, kidney, or liver disease; cancer. 81 Table 3.2 Adjusted hazard ratios and 95% confidence intervals for ASD associated with each air pollutant during pregnancy (in single pollutant models) and with each MIA (in single condition models) HRa (95% CI) Pregnancy Air Pollution Exposuresb PM2.5 per 3.73 μg/m3 1.07 (1.03, 1.12) BC per 0.84 μg/m3 1.06 (1.02, 1.10) OM per 1.98 μg/m3 1.09 (1.04, 1.14) NO3 - per 1.15 μg/m3 1.05 (1.00, 1.10) SO4 2- per 0.50 μg/m3 1.08 (1.03, 1.15) Maternal Immune Activation Infection 1.11 (1.05, 1.18) Hypertension 1.30 (1.19, 1.42) Asthma 1.22 (1.08, 1.38) Autoimmunec 1.19 (1.09, 1.30) Any MIAd 1.17 (1.10, 1.25) Abbreviations: CI, confidence interval; HR, hazard ratio a Adjusted for child sex, maternal race/ethnicity, maternal age at delivery, parity, education, maternal history of severe comorbidities, neighborhood disadvantage index, birth year, and season. b Each air pollutant was scaled to its interquartile range. c Any autoimmune disease within 1 year prior or during pregnancy. The other three categories of MIA-related conditions were restricted to pregnancy. d Any MIA represents the presence of at least one of the categories of MIA-related conditions. 82 Table 3.3 The interaction between MIA-related conditions and pregnancy PM2.5 on risk of ASD in children PM2.5-associated risk HRa (95% CI) multiplicative p-interaction additive RERI (95% CI) Infection No 1.07 (1.01, 1.13) 0.92 0.01 (-0.06, 0.08) Yes 1.07 (1.02, 1.13) Hypertension No 1.08 (1.04, 1.13) 0.17 -0.07 (-0.20, 0.06) Yes 1.01 (0.92, 1.11) Asthma No 1.07 (1.02, 1.11) 0.52 0.07 (-0.07, 0.20) Yes 1.11 (0.99, 1.25) Autoimmuneb No 1.06 (1.02, 1.11) 0.20 0.09 (-0.03, 0.22) Yes 1.13 (1.03, 1.25) Any MIAc No 1.07 (1.01, 1.14) 0.99 0.01 (-0.06, 0.08) Yes 1.07 (1.02, 1.12) Abbreviations: CI, confidence interval; HR, hazard ratio; RERI, relative excess risk due to interaction a Adjusted for child sex, maternal race/ethnicity, maternal age at delivery, parity, education, maternal history of severe comorbidities, neighborhood disadvantage index, birth year, and season. Results were scaled per interquartile (3.73 μg/m3 ) increase in PM2.5 exposure. b Any autoimmune disease within 1 year prior or during pregnancy. The other three categories of MIA-related conditions were restricted to pregnancy. c Any MIA represents the presence of at least one of the categories of MIA-related conditions. 83 3.8. Figures Figure 3.1 ASD risk associations with interactions between MIA indicators and the concentration of PM2.5 components during pregnancy Notes: Models were adjusted for child sex, maternal race/ethnicity, maternal age at delivery, parity, education, maternal history of severe comorbidities, neighborhood disadvantage index, birth year, and season. Air pollutants were scaled to their interquartile range. 84 3.9. Supplemental Material Supplemental Table 3.1 MIA and ICD-9 codes Infection Infection categories ICD-9 codes Organism-specific infection categories Bacterial infections 001-005, 008.0-008.5, 010-018, 020-027, 030-041, 079.9, 079.98, 080- 083, 088.0, 090-098, 320-322, 373.11, 424, 461, 481, 482, 483, 537.9, 540, 542, 567, 574.00, 575.1, 590, 595, 597, 599.0, 614, 616, 646.5, 647.0, 647.1, 647.3, 658.4, 659.31, 670.02, 681, 682, 684, 686, 795.5, 919.5 Mycoses 110-118 Parasitic infections 006, 007, 084-087, 088.8, 120-129, 130-136, 647.4, 647.8 Viral infections 008.6, 008.0, 042, 045-049, 050-057, 060-066, 070-077, 079.0-079.8 079.99, 460, 466, 480, 487, 571.4, 573.1-573.3 Unknown organism 099.9, 350.2, 351.0, 370.20, 371.00, 372.00, 372.3, 372.30, 373.00, 380.10, 380.23, 381.3, 381.4, 382.9, 462, 463, 464.0, 465.9, 466.0, 472.0, 473.9, 486, 490, 491.9, 528.0, 530.1, 530.10, 533.90, 535.0, 535.00, 535.50, 575.8, 575.9, 597.80, 599.7, 616.0, 616.10, 646.61-646.63, 647.23, 647.81, 647.83, 647.91, 686.9, 730.90 Organ-specific infection categories Cardiovascular 421, 422.0, 424 Skin 035, 053, 078.0, 078.1, 110, 111, 117.1, 117.2, 133, 680-682, 684-686 Ear 380-382 Eye 054.4, 076, 370-373 Gastrointestinal 001-009, 040.2, 041.4, 070, 072, 112.0, 123.0, 123.2-123.9, 528.0, 528.1, 528.3, 530.1, 530.2, 531.9, 531-533, 535, 537.9, 540, 542, 570-573, 574.00, 575.0, 575.1, 575.8, 575.9, 577.0, 577.1 Genitourinary 041.6, 054, 078.1, 079.9, 079.98, 098.0-098.3, 112.1, 112.2, 131.00- 131.02, 131.09, 132.2, 590, 599, 614, 616, 646.6 Lower respiratory 010-012, 115, 466, 480-487, 490, 491, 795.5 Upper respiratory 032, 033, 040.1, 041.5, 460-465, 472, 473, 476 Unknown organ 038.0, 038.11, 047.9, 041.04, 041.10, 041.11, 041.19, 041.85, 041.89, 041.9, 052.9, 054.8, 054.9, 079.89, 079.99, 097.1, 099.9, 112.9, 134.9, 350.2, 351.0, 567.2, 647.01, 647.23, 647.61, 647.63, 647.51, 647.81, 647.83, 647.91, 658.41, 659.31, 670.02, 730.90, 919.5, 998.5 Hypertension Hypertension categories ICD-9 Codes Chronic hypertension with/without superimposed preeclampsia 401.0, 401.1, 401.9, 405.01, 405.09, 405.11, 405.19, 405.91, 405.99, 416.0, 459.30, 459.31, 459.32, 459.33, 459.39, 642.00-642.04, 642.10- 642.14, 642.20-642.24, 642.70-642.74 Eclampsia or severe preeclampsia 642.50-642.54, 642.60-642.64 85 Supplemental Table 3.1 cont. Mild preeclampsia 642.40-642.44 Gestational hypertension 642.30-642.34, 642.90-642.94 Asthmaa 493 Autoimmune Autoimmune categories ICD-9 codes Autoimmune thyroid disease 242.00, 242.01, 242.9, 244.9, 245.2, 245.9 Idiopathic thrombocytopenic purpura 287.3, 287.4, 287.5 Inflammatory bowel disease 555.0, 555.1, 555.9, 556.1, 556.5, 556.9, 558, 558.9 All other autoimmune disease 255.4, 704.00,704.01, 720.0,720.9, 289.81, 284, 283.0, 279.4, 573.3,571.49, 136.1,711.20, 579, 710.9,756.9,446.20, 273.2, 694.0, 695.4, 446.5, 446.21,583.81, 357, 283.0, 583.89, 516.3, 364.00,714, 446.1, 697, 386.00, 340, 358, 377.30,377.39, 694.4,694.5,694.9, 733.99, 725, 710.3,710.4, 571.6, 696.0,696.1, 443, 099.3,711.10,711.90,364.00, 714.0,714.9, 390,391,392,393,394,395,396,397,398, 135,517.8,583.81, 710.2, 695.1, 714.3, 392, 583.81,695.4,710.0, 710.1, 446.6, 360.12,363.20,364.3, 287.0, 403.90, 403.91, 437.4, 446.0, 446.20, 446.4, 446.5, 446.7, 447.6, 447.8, 447.9, 583.81, 709.01 a Asthma is identified based on both ICD-9 codes and medication records. 86 Supplemental Table 3.2 Adjusted HRs and 95% CIs of ASD risk for MIA and PM2.5 in a coadjusted model Co-adjusted model a HR (95% CI) Infection 1.12 (1.07, 1.32) Hypertension 1.28 (1.19, 1.38) Asthma 1.19 (1.09, 1.32) Autoimmuneb 1.16 (1.08, 1.25) PM2.5 per IQR=3.73 μg/m3 1.06 (1.02, 1.09) Abbreviations: CI, confidence interval; HR, hazard ratio; IQR, interquartile range. a Risk factors adjusted for each other and for child sex, maternal race/ethnicity, maternal age at delivery, parity, education, maternal history of severe comorbidities, neighborhood disadvantage index, birth year, and season. b Any autoimmune disease within 1 year prior or during pregnancy. The other three MIA categories were restricted to pregnancy. Supplemental Table 3.3 Adjusted hazard ratios and 95% confidence intervals for ASD associated with each air pollutant during pregnancy (in single pollutant models) and with each MIA (in single condition models) including all follow up until December 31, 2019 without censoring at age 5 (a total of 6,366 ASD cases) HRa (95% CI) Pregnancy Air Pollution Exposuresb PM2.5 per 3.73 μg/m3 1.06 (1.02, 1.09) BC per 0.84 μg/m3 1.05 (1.02, 1.08) OM per 1.98 μg/m3 1.07 (1.03, 1.12) NO3 - per 1.15 μg/m3 1.03 (1.00, 1.07) SO4 2- per 0.50 μg/m3 1.06 (1.01, 1.11) Maternal Immune Activation Infection 1.14 (1.08, 1.20) Hypertension 1.30 (1.20, 1.40) Asthma 1.20 (1.08, 1.33) Autoimmunec 1.18 (1.10, 1.27) Any MIAd 1.20 (1.14, 1.27) Abbreviations: CI, confidence interval; HR, hazard ratio a Adjusted for child sex, maternal race/ethnicity, maternal age at delivery, parity, education, maternal history of severe comorbidities, neighborhood disadvantage index, birth year, and season. b Each air pollutant was scaled to its interquartile range. c Any autoimmune disease within 1 year prior or during pregnancy. The other three MIA categories were restricted to pregnancy. d Any MIA represents the presence of at least one of the MIA categories. 87 Supplemental Table 3.4 The interaction between MIA-related conditions and pregnancy PM2.5 on risk of ASD including all follow up until December 31, 2019 without censoring at age 5 (a total of 6,366 ASD cases) PM2.5-associated risk HRa (95% CI) multiplicative pinteraction additive RERI (95% CI) Infection No 1.04 (1.00, 1.09) 0.41 0.03 (-0.02, 0.09) Yes 1.07 (1.02, 1.11) Hypertension No 1.06 (1.02, 1.10) 0.51 -0.02 (-0.12, 0.08) Yes 1.03 (0.96, 1.12) Asthma No 1.06 (1.02, 1.10) 0.85 0.00 (-0.11, 0.11) Yes 1.05 (0.95, 1.15) Autoimmuneb No 1.05 (1.01, 1.09) 0.49 0.04 (-0.06, 0.14) Yes 1.08 (1.00, 1.17) Any MIAc No 1.05 (1.00, 1.10) 0.66 0.02 (-0.03, 0.08) Yes 1.06 (1.02, 1.10) Abbreviations: CI, confidence interval; HR, hazard ratio; RERI, relative excess risk due to interaction a Adjusted for child sex, maternal race/ethnicity, maternal age at delivery, parity, education, maternal history of severe comorbidities, neighborhood disadvantage index, birth year, and season. Results were scaled per interquartile (3.73 μg/m3 ) increase in PM2.5 exposure. b Any autoimmune disease within 1 year prior or during pregnancy. The other three categories of MIA-related conditions were restricted to pregnancy. c Any MIA represents the presence of at least one of the categories of MIA-related conditions. 88 Supplemental Figure 3.1 Derivation of study sample 370,723 singleton births at KPSC hospitals between 1/01/2001 and 31/12/2014 with KPSC membership at age 1 and alive at age 1 318,751 included in primary analysis 51,972 Excluded: 666* missing or errors in covariates (90 birth weight; 476 gender; 96 maternal race-ethnicity; 44 maternal age at delivery; 40 missing or errors in 2 covariates) 159 maternal age at delivery < 15 yrs or > 55 yrs 51,147 Missing/incomplete addresses, non-California addresses, or addresses not suitable for assigning exposure data 89 Supplemental Figure 3.2 Distribution of air pollutant concentrations during pregnancy across birth year 2001-2014 90 Supplemental Figure 3.3 ASD risk associations with interactions between MIA indicators and the concentration of PM2.5 components during the first trimester Notes: Models were adjusted for child sex, maternal race/ethnicity, maternal age at delivery, parity, education, maternal history of severe comorbidities, neighborhood disadvantage index, birth year, and season. Air pollutants were scaled to their interquartile ranges. 91 Supplemental Figure 3.4 ASD risk associations with interactions between MIA indicators and the concentration of PM2.5 components during the second trimester Notes: Models were adjusted for child sex, maternal race/ethnicity, maternal age at delivery, parity, education, maternal history of severe comorbidities, neighborhood disadvantage index, birth year, and season. Air pollutants were scaled to their interquartile ranges. 92 Supplemental Figure 3.5 ASD risk associations with interactions between MIA indicators and the concentration of PM2.5 components during the third trimester Notes: Models were adjusted for child sex, maternal race/ethnicity, maternal age at delivery, parity, education, maternal history of severe comorbidities, neighborhood disadvantage index, birth year, and season. Air pollutants were scaled to their interquartile ranges. 93 Chapter 4 Neighborhood Disadvantage and Autism Spectrum Disorder in a Population with Health Insurance This chapter has been published as Yu, X., M. M. Rahman, S. A. Carter, J. C. Lin, T. Chow, F. W. Lurmann, J. C. Chen, M. P. Martinez, J. Schwartz, S. P. Eckel, Z. Chen, R. McConnell, A. H. Xiang, and D. A. Hackman. 2024. "Neighborhood Disadvantage and Autism Spectrum Disorder in a Population With Health Insurance." JAMA Psychiatry 81 (2): 209-213. 4.1. Abstract Importance: Family socioeconomic status has been associated with autism spectrum disorder (ASD) diagnoses. Less is known regarding the role of neighborhood disadvantage in the United States, particularly when children have similar access to health insurance. Objective: To evaluate the association between neighborhood disadvantage and the diagnosis of ASD and potential effect modification by maternal and child demographic characteristics. Design, Setting, and Participants: This cohort study examined a retrospective birth cohort from Kaiser Permanente Southern California (KPSC), an integrated health care system. Children born in 2001 to 2014 at KPSC were followed up through KPSC membership records. Electronic medical records were used to obtain an ASD diagnosis up to December 31, 2019, or the last follow-up. Data were analyzed from February 2022 to September 2023. Exposure: Socioeconomic disadvantage at the neighborhood level, an index derived from 7 US census tract characteristics using principal component analysis. Main Outcomes and Measures: Clinical ASD diagnosis based on electronic medical records. Associations between neighborhood disadvantage and ASD diagnosis were determined 94 by hazard ratios (HRs) from Cox regression models adjusted for birth year, child sex, maternal age at delivery, parity, severe pre-pregnancy health conditions, maternal race and ethnicity, and maternal education. Effect modification by maternal race and ethnicity, maternal education, and child sex was assessed. Results: Among 318 372 mothers with singleton deliveries during the study period, 6357 children had ASD diagnoses during follow-up; their median age at diagnosis was 3.53 years (IQR, 2.57-5.34 years). Neighborhood disadvantage was associated with a higher likelihood of ASD diagnosis (HR, 1.07; 95% CI, 1.02-1.11, per IQR = 2.70 increase). Children of mothers from minoritized racial and ethnic groups (African American or Black, Asian or Pacific Islander, Hispanic or Latinx groups) had increased likelihood of ASD diagnosis compared with children of White mothers. There was an interaction between maternal race and ethnicity and neighborhood disadvantage (difference in log-likelihood = 21.88; P < .001 for interaction under χ2 4); neighborhood disadvantage was only associated with ASD among children of White mothers (HR, 1.17; 95% CI, 1.09-1.26, per IQR = 2.00 increase). Maternal education and child sex did not significantly modify the neighborhood-ASD association. Conclusions and Relevance: In this study, children residing in more disadvantaged neighborhoods at birth had higher likelihood of ASD diagnosis among a population with health insurance. Future research is warranted to investigate the mechanisms behind the neighborhoodrelated disparities in ASD diagnosis, alongside efforts to provide resources for early intervention and family support in communities with a higher likelihood of ASD. 95 4.2. Introduction Autism spectrum disorder (ASD) affects 2.8% of US children (Maenner et al. 2023) and is associated with hardships (Kuhlthau et al. 2010) and economic burdens (Buescher et al. 2014) for children and families. While family-level socioeconomic status (SES) has been associated with ASD (Bhasin and Schendel 2007; Durkin et al. 2010), less is known about neighborhood disadvantage, which relates to health independent of family SES (Minh et al. 2017; Ncube et al. 2016) and is associated with environmental and contextual factors that may impact maternal health during pregnancy and child development. Only three studies considered individual ASD diagnoses with family- and neighborhood-level SES simultaneously, with mixed findings (X. Li et al. 2014; Kelly et al. 2019; King and Bearman 2011). Higher SES was associated with lower ASD likelihood in countries with free and universal healthcare (Rai et al. 2012). While the opposite pattern has been found in the US (King and Bearman 2011), it reversed after 2010, with lower SES related to higher ASD diagnosis rates more recently (Winter et al. 2020). This may be due to changes in access to developmental screening, as inequities in screening may mask underlying disparities. Consequently, studies are needed to determine if neighborhood disadvantage is associated with increased ASD diagnoses when reducing inequities in access to screening/diagnostic services, and if such disparities vary based on child and maternal sociodemographic characteristics. To address these questions, we utilized a birth cohort in Kaiser Permanente Southern California (KPSC) hospitals for whom integrated insurance and care ensures greater access to, and reduced disparities in, diagnostic screening for ASD. 96 4.3. Methods 4.3.1. Study Population This retrospective cohort included mothers with singleton deliveries (n=318,372) at KPSC hospitals between 2001-2014 (derivation of study sample in Section 4.9.1 eMethods and Supplemental Figure 4.1). Characteristics and covariates (Section 4.9.1 eMethods) for mothers and children were extracted from electronic medical records (EMRs). ASD diagnosis was ascertained based on at least two clinical diagnoses using International Classification of Diseases, Ninth Revision (ICD-9) and ICD-10 codes (Section 4.9.1 eMethods and Supplemental Table 2.1) (Coleman et al. 2015). Children were followed from age 1 until the first ASD diagnosis, the last date of KPSC membership, death, or December 31, 2019, whichever occurred first. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines. Institutional Review Boards at KPSC and the University of Southern California approved this study with a waiver of individual participant consent. 4.3.2. Neighborhood disadvantage Neighborhood was defined as the US census tract of the residential address at the child’s birth. Data were extracted from the 2000 US decennial census for births from 2001-2004, and 2005-2009 and 2010-2014 American Community Survey (ACS) 5-year estimates for corresponding birth years. Neighborhood disadvantage was defined as the first principal component of seven census tract indicators (B.L. Levy, Owens, and Sampson 2019): poverty, unemployment, female-headed households with children, public assistance, less than a high school education, bachelor’s degree or greater, and professional occupation (Section 4.9.1 eMethods, Supplemental Figure 4.2, and Supplemental Table 4.2). 97 4.3.3. Statistical Analysis We utilized Cox proportional hazard regression models with census tracts modelled as random intercepts adjusting for covariates. Neighborhood disadvantage was treated as a continuous variable and the linear association estimated as a hazard ratio (HR) scaled to its interquartile range (IQR). Neighborhood disadvantage was also categorized into quintiles to examine non-linearity. Multiplicative interaction tests were used to examine effect modification by maternal race and ethnicity, maternal education, and child sex in fully adjusted models. If the global interaction was significant, stratification analysis was conducted, with neighborhood disadvantage scaled to a stratum-specific IQR. Two-sided statistical tests were used (a = .05) and 95% confidence intervals (CIs) were reported. Sensitivity analyses are described in Section 4.9.1 eMethods. Analyses were performed in R (version 4.2). 4.4. Results There were 6,357 children (2.00%) diagnosed with ASD at a median age of 3.53 years (IQR=2.57-5.34). Males (n=5,164) were 4.3 times more likely than females (n=1,193) to have ASD diagnoses (Table 4.1). The neighborhood disadvantage index explained 66.1% of the variance in the neighborhood SES indicators. The index exhibited variability by maternal race and ethnicity, maternal education, and child sex (Supplemental Figure 4.3). Higher neighborhood disadvantage at birth was associated with higher likelihood of ASD diagnosis [HR=1.09 (95% CI 1.05-1.13)] per IQR increase in disadvantage score, adjusted for birth year, maternal age, parity, maternal comorbidity, and medical center (Table 4.2, Model 1; crude associations in Supplemental Table 4.3). While further adjustment for maternal race and ethnicity, maternal education and child sex attenuated the point estimate, the association was still 98 significant [HR=1.07 (95% CI 1.02-1.11) per IQR=2.7 increase] (Table 4.2, Model 2 and Figure 4.1A). Residence in neighborhoods in the 4th and 5th highest quintiles of disadvantage was associated with higher ASD likelihood (Table 4.2 and Supplemental Figure 4.4A). Similar associations were found for each separate neighborhood indicator and alternative neighborhood composites (4.9.1 eMethods and Supplemental Table 4.4). Children of mothers from minoritized racial and ethnic groups had higher likelihood of ASD diagnosis than children of White mothers (Figure 4.1B and Supplemental Table 4.5). A significant interaction was observed between maternal race and ethnicity and neighborhood disadvantage (p-interaction<0.001). Neighborhood disadvantage was significantly associated with ASD diagnosis only for children of White mothers [HR=1.17 (95% CI 1.09-1.26) per IQR=2.0 increase in White] (Figure 4.1C). Similar patterns were observed for alternative neighborhood SES measures (Supplemental Table 4.6) and a dichotomized indicator of high disadvantage at the 4-5th quintile (Supplemental Figure 4.4C). There was no significant effect modification by maternal education (p-interaction=0.07) or child sex (p-interaction=0.09). Sensitivity analyses produced similar results (Section 4.9.2 eResults Sensitivity Analyses, Supplemental Table 4.7, and Supplemental Table 4.8). 4.5. Discussion Neighborhood disadvantage was associated with higher likelihood of ASD diagnosis, independent of maternal education, in a large birth cohort with access to health insurance. This result is consistent with studies in Sweden, where screening and healthcare are universally accessible (X. Li et al. 2014). Thus, when disparities in access to screening and diagnostic services are reduced, neighborhood disadvantage is associated with higher likelihood of ASD diagnosis. 99 While children of mothers from minoritized racial and ethnic groups had higher likelihood of ASD diagnosis, the association between neighborhood disadvantage and ASD diagnosis was only observed for children of White mothers, a pattern similar to the stronger associations between neighborhood disadvantage and preterm birth and low birthweight in White populations (Ncube et al. 2016). The mechanisms behind these patterns are unknown and merit future research. However, they may be due to systemic racism, discrimination, and their impact on maternal health during pregnancy (Colen et al. 2018). Thus, it is possible that families from minoritized racial and ethnic groups may not experience similar health improvements with increasing SES as White populations, due to discrimination and related stressors and barriers experienced in more affluent neighborhoods (Colen et al. 2018). The study has several strengths, including the use of an integrated system with standard diagnostic procedures, a large study population, and ample statistical power. In addition, results were robust to sensitivity analyses and the use of alternative neighborhood measures. With respect to limitations, a lack of information on fathers and socioeconomic indices such as income may have led to unmeasured family-level confounding. Second, although the KPSC membership covers both commercial insurance and Medi-Cal, results need replication in other health service settings. Third, even within an integrated health system, access to care may vary by factors such as co-pays. Additionally, the prevalence of ASD may not be accurately estimated based on clinical diagnoses in electronic medical records. 4.6. Conclusion Neighborhood disadvantage is associated with higher likelihood of ASD diagnosis when differences in access to care are reduced by providing similar health insurance. However, the benefits of reduced neighborhood disadvantage were only observed in White populations while 100 minoritized racial and ethnic groups had higher likelihood of ASD diagnosis. Future research is needed to identify the mechanisms underlying these complex associations. Investment in early intervention and family support for communities and populations with higher diagnosis rates, along with universal access to screening and diagnosis, is warranted. 101 4.7. Tables Table 4.1 Characteristics of children, with and without autism spectrum disorder (ASD) Children, No. (%) or median [interquartile range] Characteristics Overall With ASD Without ASD (n =318,372) (n= 6,357) (n= 312,015) Individual and family level characteristics Sex = Male, No. (%) 162,986 (51.2) 5,164 (81.2) 157,822 (50.6) Maternal age at delivery, 30.4 [26.3, 34.3] 31.3 [27.5, 35.2] 30.4 [26.2, 34.3] median [IQR*], year Parity, No. (%) 0 111,854 (35.1) 2,656 (41.8) 109,198 (35.0) 1 104,428 (32.8) 2,042 (32.1) 102,386 (32.8) >=2 84,091 (26.4) 1,265 (19.9) 82,826 (26.5) Unknown 17,999 (5.7) 394 (6.2) 17,605 (5.6) Maternal Education, No. (%) High school or lower 112,045 (35.2) 1,911 (30.1) 110,134 (35.3) Some college 94,418 (29.7) 2,036 (32.0) 92,382 (29.6) College graduate or higher 108,871 (34.2) 2,362 (37.2) 106,509 (34.1) Unknown 3,038 (1.0) 48 (0.8) 2,990 (1.0) Race and ethnicity, No. (%) African American/ Black 29,752 (9.4) 645 (10.1) 29,107 (9.3) Asian/ Pacific Islander 39,854 (12.5) 929 (14.6) 38,925 (12.5) Hispanic/ Latinx 161,319 (50.7) 3,181 (50.0) 158,138 (50.7) White 80,909 (25.4) 1,445 (22.7) 79,464 (25.5) Othera 6,538 (2.1) 157 (2.5) 6,381 (2.0) History of maternal comorbidityb , No. (%) 46,647 (14.7) 1,150 (18.1) 45,497 (14.6) Year of birth, No. (%) 2001-2004 78,236 (24.6) 1,439 (22.6) 76,797 (24.6) 2005-2009 111,064 (34.9) 2,003 (31.5) 109,061 (35.0) 2010-2014 129,072 (40.5) 2,915 (45.9) 126,157 (40.4) Neighborhood SES measuresc , median [IQR*] Povertyd 9.5 [4.7, 17.3] 10.0 [4.8, 17.8] 9.5 [4.7, 17.3] Unemploymentd 8.8 [6.2, 11.9] 8.9 [6.2, 11.9] 8.8 [6.2, 11.9] Female-headed householdsd 17.5 [11.7, 24.6] 17.8 [12.1, 25.3] 17.5 [11.7, 24.6] Public assistanced 3.3 [1.6, 6.3] 3.4 [1.7, 6.4] 3.3 [1.6, 6.3] Less than high schoold 21.2 [10.6, 35.6] 21.6 [10.3, 35.9] 21.2 [10.6, 35.6] Bachelor and post-graduated 19.3 [10.5, 32.5] 19.70 [10.8, 33.5] 19.3 [10.5, 32.5] Professional occupationd 28.9 [19.1, 40.8] 28.60 [19.1, 41.5] 28.9 [19.1, 40.8] Disadvantage index 0.07 [-1.22, 1.48] 0.12 [-1.25, 1.54] 0.07 [-1.22, 1.48] Abbreviations: ASD, autism spectrum disorder; IQR, interquartile range; SES, socioeconomic status. a Other race and ethnicity includes patients who identified as “other” or “multiple” races. b ≥1 diagnosis of heart, lung, kidney, or liver disease; cancer c Based on census tract at birth address d Percentage of each neighborhood SES census tract indicator, for example the percentage of families below the poverty line across all Census tracts. 102 Table 4.2 Associations between neighborhood disadvantage and ASD diagnosis in models adjusting for varying sets of covariates Model 1 (adjusted covariates)a Model 2 (adjusted covariates and maternal race and education)b HR (95% CI) HR (95% CI) Disadvantage index (per IQR=2.70) 1.09 (1.05, 1.13) 1.07 (1.02, 1.11) Disadvantage index quintiles 1st quintile reference reference 2nd quintile 0.98 (0.91, 1.06) 0.97 (0.89, 1.05) 3rd quintile 1.03 (0.95, 1.12) 1.01 (0.93, 1.09) 4th quintile 1.17 (1.08, 1.27) 1.14 (1.05, 1.24) 5th quintile 1.16 (1.07, 1.26) 1.12 (1.02, 1.22) Abbreviations: CI, confidence interval; HR, hazard ratio; IQR, interquartile range a Model 1 adjusted for birth year, maternal age, parity, comorbidity, and medical center. Census tracts are modelled as random intercepts. b Model 2 adjusted for covariates in Model 1, plus maternal race and ethnicity, maternal education and child sex at birth. Census tracts are modelled as random intercepts. 103 4.8. Figures Figure 4.1 Associations between neighborhood disadvantage, maternal race and ethnicity and ASD diagnosis Abbreviations: ASD, autism spectrum disorder; HR, hazard ratio Notes: Figure (A) shows the main association between neighborhood disadvantage (per IQR=2.70 increase) and ASD in the entire study population from the fully adjusted Model 2. Figure (B) shows the main association between maternal race/ethnicity and ASD in the entire study population from the fully adjusted Model 2 (reference group: White). Figure (C) shows the association between neighborhood disadvantage and ASD in analyses fully stratified by maternal race and ethnicity scaled to the stratum-specific IQR, adjusted for birth year, maternal age, parity, comorbidity, medical center, maternal education and child sex at birth. IQR for African American/ Black, Asian/ Pacific Islander, Hispanic/ Latinx, White, and Other were 2.71, 2.25, 2.45, 2.00, and 2.36 respectively. Similar results are found when the coefficient for each race/ethnicity group is extracted from the interaction model (Supplemental Table 4.8). Other race/ethnicity includes patients who identified as “other” or “multiple” races. Census tracts are modelled as random intercepts. The interpretation of the coefficient is an effect corresponding to the range of exposure of the middle 50% of the population. Disadvantage (A) ASD diagnosis associated with neighborhood disadvantage (main association) Other White (Reference) Hispanic/ Latinx Asian/ Pacific Islander African American/ Black Race and ethnicity (B) ASD diagnosis associated with maternal race and ethnicity (main association) Other White Hispanic/ Latinx Asian/ Pacific Islander African American/ Black 0.8 1.0 1.2 1.4 1.6 HR Race and ethnicity (C) ASD diagnosis associated with neighborhood disadvantage stratified by maternal race and ethnicity 104 4.9. Supplemental Material 4.9.1. eMethods • Study population This retrospective cohort study included mothers with singleton deliveries (n=370,723) at Kaiser Permanente Southern California (KPSC) hospitals between January 1, 2001, and December 31, 2014. KPSC is a large integrated healthcare system with over 4.5 million members across 10 Southern California counties (Imperial, Kern, Los Angeles, Orange, Riverside, San Bernardino, San Diego, Santa Barbara, San Luis Obispo and Ventura), with diverse membership similar in socioeconomic characteristics to the region(Koebnick et al. 2012). The KPSC member birth cohort in this study included mothers and their children covered by employer-sponsor coverage as well as Medicaid, as insurance type was not a criterion for exclusion. In general, 9% of Kaiser members are covered by Medicaid or the Children's Health Insurance Program (CHIP). Characteristics for mothers and children were extracted from KPSC electronic medical records (EMR). A total of 52,351 births were excluded due to (1) missing gender, maternal race and ethnicity and age at delivery (n=666); (2) maternal age <15 or >55 at delivery (n=159); (3) incomplete maternal residential address at birth or geocodes not suitable for neighborhood SES assignment (n=51,526) (Supplemental Figure 4.1). Eligible children (n=318,372) were followed until a clinical diagnosis of ASD, the last date of KPSC health plan membership, death, or December 31, 2019, whichever occurred first. Institutional Review Boards at KPSC and the University of Southern California approved this study with a waiver of individual participant consent. • ASD Ascertainment 105 ASD diagnosis in children was coded as a single variable (presence/absence) from EMR records, by International Classification of Diseases, Ninth Revision (ICD-9) codes 299.0, 299.1, 299.8, 299.9 for records before October 1, 2015 (the date of KPSC implementation of ICD-10) and ICD-10 codes F84.0, F84.3, F84.5, F84.8, F84.9 thereafter (see description of ICD codes in Supplemental Figure 5.1). For this report, a coded ASD diagnosis was required from at least two separate visits on different dates. This methodology for ascertaining ASD diagnoses has been previously validated in the KPSC member population, with a positive predictive value of 87% (Coleman et al. 2015), and has been employed in previous studies (Xiang et al. 2018; Xiang et al. 2015). • Neighborhood disadvantage index and neighborhood deprivation/advantage indices Neighborhood was defined as the US census tract of the mother’s residential address at the time of the child’s birth, geocoded using ArcGIS 12. Data from the 2000 US decennial census and the American Community Survey (ACS) 5-year estimates for 2005-2009 and 2010- 2014 were assigned to children born between 2001-2004, 2005-2009, and 2010-2014, respectively. ACS data based on 2010 census tract boundaries (2010-2014 estimates) were matched to the 2000 census tract boundaries using crosswalks from the Longitudinal Tract Data Base (Logan, Xu, and Stults 2014). Neighborhood disadvantage was based on seven census tract indicators(B.L. Levy, Owens, and Sampson 2019): the percentage of (1) families below the poverty line; (2) unemployment in the civilian labor force; (3) female-headed households with children; (4) households receiving public assistance income; (5) population 25 years and over with less than high school education; (6) population 25 years and over with Bachelor’s degree or greater; and (7) the population employed in professional or managerial occupations in the civilian labor force. 106 Indices of neighborhood deprivation and advantage(B.L. Levy, Owens, and Sampson 2019) were generated based on the seven census indicators using confirmatory factor analysis with oblique promax rotation on two latent factors. For each latent factor, indicators with absolute loadings greater than 0.4 were selected (Howard 2016). A constraint was added that the loadings for raw indicators should be consistent across three census datasets via confirmatory factor analysis (CFA), ensuring latent factors were comparable across the three census data sources. Census tracts with an estimated population of less than 100 or with missing values in the seven raw indicators were excluded from the principal component analysis and the factor analysis. The loadings for the first principal component of neighborhood disadvantage, our primary neighborhood variables, and the supplementary indices of neighborhood deprivation and advantage are shown in Supplemental Figure 4.2. R codes for generating contextual SES indices are available on GitHub (https://github.com/lindayu0408/SES_compositional_indices). CFA was implemented using lavaan package, multilevel Cox regression models were implemented using coxme package. The first principal component, represented as the disadvantage index, was conducted via the princomp function in stats package. • Effect modifiers and covariates Key covariates and potential effect modifiers were derived from the EMR (see Table 1). Covariates were selected due to known associations with ASD including maternal age at delivery, parity, maternal history of comorbidity [>=1 diagnosis of heart, lung, kidney, liver disease, or cancer], medical center (to address the potential differences in the practice of screening and diagnosis) and birth year to account for the increasing trend of ASD prevalence (Xiang et al. 2015). 107 Maternal race and ethnicity based on self-report included in the EMR and was recorded in five categories: African American/ Black, Asian/ Pacific Islander, Hispanic/ Latinx, White, and other (patients identified as “other” or “multiple” races). Maternal educational attainment was classified into three levels: high school or less, some college, and college graduate or higher. Child sex assigned at birth was coded as male and female. A missing indicator variable was used for missing values in categorical covariates (parity [n = 17,999], maternal education [n=3,038]). • Sensitivity analysis We conducted four sensitivity analyses: (1) we examined associations between alternative neighborhood SES measures (disadvantage, advantage indices and seven individual census measures) and ASD; (2) we censored follow-up time up to age 5 to ensure the same length of follow-up for children born in different years; (3) we adjusted birth year as three periods (2001-2004, 2005-2009, and 2010-2014) to control the potential differences among the methodologies in three census data sources; (4) we limited the population to mother-child pairs who did not move during the pregnancy to control for potential misalignment of neighborhoods; and (5) we estimated the association between neighborhood disadvantage index and ASD diagnosis by racial and ethnic groups based on the interaction model, which assumed baseline hazard function and coefficients of covariates were the same across racial and ethnic groups. 4.9.2. eResults Sensitivity Analyses The proportional hazards assumption was assessed using the Schoenfeld residual plot. No clear non-random patterns with follow-up time were observed. Sensitivity analysis censoring follow-up time at age 5 identified substantively similar associations between neighborhood SES measures and ASD but the associations were no longer significant. Analyses utilizing birth year as three time periods (2001-2004, 2005-2009, and 2010-2014) for adjustment and sensitivity 108 analysis restricted to those who did not move during pregnancy produced similar results to the primary analysis that neighborhood disadvantage was associated with higher likelihood of ASD diagnosis. All sensitivity analyses are outlined in Supplemental Table 4.6-4.8. Supplemental Table 4.1 ICD codes and diagnostic description for ASD ascertainment ICD-9 codes Description 299.0 Autistic disorder 299.1 Childhood disintegrative disorder 299.8 Other specified pervasive developmental disorders 299.9 Unspecified pervasive developmental disorder ICD-10 codes F84.0 Autistic disorder F84.3 Other childhood disintegrative disorder F84.5 Asperger's syndrome F84.8 Other pervasive developmental disorders F84.9 Pervasive developmental disorder, unspecified Supplemental Table 4.2 Factor loadings of the primary neighborhood disadvantage index and indices of neighborhood deprivation and advantage Disadvantage indexa Deprivation indexb Advantage indexb Poverty 0.407 0.918 Unemployed 0.332 0.689 Female-headed households 0.260 0.517 Public assistance 0.382 0.838 Less than high school education 0.415 -0.934 Bachelor and post-graduate degree -0.403 0.900 Occupation: Professional -0.418 0.955 a The first principal component generated based on all seven census indicators. b For the 2 latent confirmatory factor analysis, the deprivation index is generated based on % poverty, % unemployed, % femaleheaded households, and % public assistance; the advantage index is generated based on % less than high school, % bachelor and post, and % professional occupation. The selection of indicators is based on exploratory factor analysis with a loading cutoff point at 0.4 (see Section 4.9.1). 109 Supplemental Table 4.3 Crude associations between neighborhood disadvantage index and ASD Crude Association HR (95% CI) Disadvantage index (per IQR=2.70) 0.97 (0.94, 1.01) Disadvantage index quintiles 1st quintile reference 2nd quintile 0.88 (0.81, 0.95) 3rd quintile 0.87 (0.80, 0.94) 4th quintile 0.95 (0.87, 1.03) 5th quintile 0.93 (0.86, 1.01) Abbreviations: CI, confidence interval; HR, hazard ratio; IQR, interquartile range Supplemental Table 4.4 Associations between neighborhood disadvantage components, neighborhood disadvantage and advantage indices, and ASD in models adjusting for varying sets of covariates Model 1a (adjusted covariates)b Model 2a (adjusted covariates, maternal race and ethnicity, education and child sex)c HR (95% CI) HR (95% CI) Individual neighborhood measures Poverty (IQR=12.6) 1.08 (1.05, 1.12) 1.07 (1.03, 1.11) Unemployment (IQR=5.7) 1.07 (1.04, 1.11) 1.06 (1.02, 1.10) Female-headed households (IQR=12.9) 1.05 (1.01, 1.08) 1.04 (1.00, 1.07) Public assistant (IQR=4.7) 1.06 (1.03, 1.09) 1.05 (1.01, 1.08) Less than high school (IQR=25.0) 1.08 (1.04, 1.12) 1.06 (1.01, 1.11) Bachelor and post (IQR=22.0) 0.93 (0.90, 0.97) 0.95 (0.91, 0.99) Professional occupation (IQR=21.7) 0.93 (0.89, 0.96) 0.94 (0.90, 0.98) Alternative neighborhood composite indices Deprivation index (IQR=1.12) 1.08 (1.04, 1.12) 1.06 (1.02, 1.10) Advantage index (IQR=1.25) 0.93 (0.89, 0.96) 0.95 (0.91, 0.99) Abbreviations: CI, confidence interval; HR, hazard ratio; IQR, interquartile range a Census tracts are modelled as random intercepts in all three models. b Model 1 adjusted for birth year, maternal age, parity, comorbidity, medical center. c Model 2 adjusted for covariates in Model 2, maternal race and ethnicity, maternal education and child sex. 110 Supplemental Table 4.5 Multivariable-adjusted associations between race and ethnicity, maternal education, and ASD Model 1a (adjusted covariates)b Model 2a (adjusted covariates and neighborhood disadvantage index)c Model 3a (mutually adjusted and disadvantage index)d HR (95% CI) HR (95% CI) HR (95% CI) Maternal race and ethnicity African American/ Black 1.20 (1.09, 1.32) 1.15 (1.04, 1.27) 1.13 (1.02, 1.25) Asian/ Pacific Islander 1.10 (1.01, 1.20) 1.09 (1.00, 1.19) 1.11 (1.02, 1.20) Hispanic/ Latinx 1.13 (1.06, 1.20) 1.09 (1.02, 1.17) 1.07 (1.00, 1.15) White reference reference reference Othere 1.31 (1.11, 1.54) 1.29 (1.10, 1.53) 1.29 (1.09, 1.52) Maternal education <=high school reference reference reference some college 1.09 (1.03, 1.17) 1.11 (1.04, 1.18) 1.10 (1.04, 1.18) college & post 0.91 (0.85, 0.97) 0.94 (0.88, 1.01) 0.94 (0.88, 1.01) Abbreviations: CI, confidence interval; HR, hazard ratio a Census tracts are modelled as random intercepts in all three models. b Model 1 adjusted for birth year, maternal age, parity, comorbidity, medical center, and child sex. c Model 2 adjusted for covariates in Model 1 and the neighborhood disadvantage index. d Model 3 mutually adjusted for maternal race and ethnicity, maternal education, covariates, and neighborhood disadvantage index. e Other race and ethnicity includes patients who identified as “other” or “multiple” races. 111 Supplemental Table 4.6 Stratified analyses: Associations between neighborhood characteristics and ASD by maternal race and ethnicity African American/ Black (9.4%) Asian/ Pacific Islander (12.5%) Hispanic/ Latinx (50.7%) White (25.4%) Otherb (2.1%) HRa (95%CI) HRa (95%CI) HRa (95%CI) HRa (95%CI) HRa (95%CI) p-interactionc Individual neighborhood indicators Povertyd 1.07 (0.94, 1.21) 1.05 (0.97, 1.13) 1.04 (0.99, 1.10) 1.15 (1.08, 1.21) 1.03 (0.84, 1.26) 0.011 Unemploymentd 0.99 (0.88, 1.11) 1.06 (0.97, 1.17) 1.06 (1.01, 1.11) 1.11 (1.03, 1.20) 1.00 (0.79, 1.27) 0.146 Female-headed householdsd 0.90 (0.79, 1.02) 1.11 (1.02, 1.20) 1.02 (0.97, 1.07) 1.05 (0.99, 1.12) 1.24 (1.00, 1.53) 0.016 Public assistantd 0.96 (0.85, 1.08) 1.04 (0.97, 1.11) 1.04 (0.99, 1.09) 1.14 (1.08, 1.20) 1.03 (0.84, 1.26) <0.001 Less than high schoold 1.06 (0.93, 1.20) 1.04 (0.95, 1.15) 1.02 (0.96, 1.08) 1.15 (1.08, 1.23) 1.01 (0.79, 1.28) <0.001 Bachelor and postd 0.96 (0.87, 1.06) 0.94 (0.84, 1.04) 0.99 (0.95, 1.04) 0.89 (0.81, 0.97) 1.06 (0.81, 1.38) 0.003 Professional occupationd 0.96 (0.87, 1.07) 0.96 (0.86, 1.06) 0.98 (0.93, 1.03) 0.89 (0.81, 0.96) 0.94 (0.73, 1.21) 0.004 Neighborhood composite indices Disadvantage indexd 1.01 (0.89, 1.15) 1.05 (0.97, 1.14) 1.03 (0.98, 1.09) 1.16 (1.09, 1.23) 1.01 (0.80, 1.27) <0.001 Advantage indexd 0.96 (0.86, 1.08) 0.95 (0.86, 1.06) 0.99 (0.93, 1.04) 0.87 (0.80, 0.94) 1.00 (0.77, 1.29) 0.001 Abbreviations: CI, confidence interval; HR, hazard ratio a All models adjusted for birth year, maternal age, parity, maternal education, comorbidity, medical center, child sex. Census tracts are modelled as random intercepts. b Other race and ethnicity includes patients who identified as “other” or “multiple” races. c Global p-interaction is generated by likelihood ratio test. d Neighborhood SES measures were scaled to their stratum-specific IQR. 112 Supplemental Table 4.7 Sensitivity analyses: Adjusted associationsa between neighborhood disadvantage index and ASD Model 1 (censored at age 5) Model 2 (adjust birth year as 2001-2004, 2005-2009, and 2010-2014)a Model 3 (non-movers only)b HR (95% CI) HR (95% CI) HR (95% CI) Disadvantage index (IQR=2.70) 1.05 (0.997, 1.10) 1.07 (1.02, 1.11) 1.06 (1.00, 1.12) Disadvantage index quintiles 1st quintile reference reference reference 2nd quintile 0.95 (0.86, 1.04) 0.98 (0.90, 1.06) 0.97 (0.87, 1.08) 3rd quintile 0.93 (0.84, 1.02) 1.01 (0.93, 1.10) 0.94 (0.84, 1.06) 4th quintile 1.06 (0.96, 1.17) 1.15 (1.06, 1.25) 1.08 (0.96, 1.21) 5th quintile 1.10 (0.99, 1.22) 1.12 (1.02, 1.22) 1.15 (1.03, 1.30) Abbreviations: CI, confidence interval; HR, hazard ratio; IQR, interquartile range a All models adjusted for birth year, maternal age, parity, maternal race and ethnicity, maternal education, comorbidity, medical center, child sex. Census tracts are modelled as random intercepts. Supplemental Table 4.8 Associations between neighborhood disadvantage index and ASD by race and ethnicity based on the interaction model HR (95% CI) African American/ Black 0.94 (0.84, 1.05) Asian/ Pacific Islander 1.04 (0.94, 1.16) Hispanic/ Latinx 1.04 (0.98, 1.10) White 1.29 (1.17, 1.41) Other 0.99 (0.77, 1.28) Note: The interaction model was adjusted for birth year, maternal age, parity, maternal education, comorbidity, medical center, child sex. Census tracts are modelled as random intercepts. For each racial and ethnic group, the neighborhood disadvantage index was scaled to the global IQR=2.70 of the entire cohort. 113 Supplemental Figure 4.1 Derivation of study sample 370,723 singleton births at KPSC hospitals between 1/01/2001 and 31/12/2014 with KPSC membership at age 1 and alive at age 1 318,372 included in primary analysis 51,351 Excluded: 666 missing or errors in covariates (90 birth weight; 476 gender; 96 maternal race-ethnicity; 44 maternal age at delivery; 40 missing or errors in 2 covariates) 159 maternal age at delivery < 15 yrs or > 55 yrs 51,526 Missing/incomplete addresses, non-California addresses, or addresses not suitable for assigning census data 114 Supplemental Figure 4.2 The distribution of neighborhood SES measures from the three census sources among all census tracts in Southern California 115 Supplemental Figure 4.3 The distribution of neighborhood disadvantage index by maternal and children demographic characteristics 116 Supplemental Figure 4.4 The main and interactive associations between dichotomized neighborhood disadvantage (quintiles 4-5 vs quintiles 1-3), maternal race and ethnicity and ASD Abbreviations: ASD, autism spectrum disorder; HR, hazard ratio Notes: Figure (A) shows the main association between dichotomized indicator of high disadvantage at the 4th quintile (quintiles 4-5 vs quintiles 1-3) and ASD in the entire study population from the fully adjusted Model 2. Figure (B) shows the main association between maternal race and ethnicity and ASD in the entire study population from the fully adjusted Model 2 (reference group: non-Hispanic White). Figure (C) shows the association between the dichotomized indicator of high disadvantage at the 4th quintile and ASD in each maternal race and ethnicity stratum adjusted for birth year, maternal age, parity, comorbidity, medical center, maternal education, and child sex at birth. Census tracts are modelled as random intercepts. Disadvantage (A) ASD diagnosis associated with neighborhood disadvantage (main association) Other White (Reference) Hispanic/ Latinx Asian/ Pacific Islander African American/ Black Race and ethnicity (B) ASD diagnosis associated with maternal race/ethnicity (main association) Other White Hispanic/ Latinx Asian/ Pacific Islander African American/ Black 0.8 1.0 1.2 1.4 1.6 1.8 2.0 HR Race and ethnicity (C) ASD diagnosis associated with neighborhood disadvantage stratified by maternal race and ethnicity 117 Chapter 5 The Potential Effects of Hypothetical PM2.5 Interventions on Childhood Autism in Different Neighborhood Socioeconomic Contexts 5.1. Abstract Background: Particle air pollution is associated with autism spectrum disorder (ASD). Residing in disadvantaged neighborhoods may increase vulnerability to pollution due to stress or other social determinants of health. Understanding the variability of impacts of air pollution interventions to reduce ASD incidence across disadvantage levels can inform policy to protect vulnerable populations. Methods: In a cohort of 318,298 children born between 2001-2014 in Southern California, pregnancy average PM2.5 exposure and neighborhood disadvantage were assigned to maternal residential addresses. Two sets of hypothetical PM2.5 interventions were simulated: (1) percentage reduction; (2) regulatory standards as thresholds. G-computation was applied to quantify changes in cumulative incidence of ASD by age 5 under hypothetical interventions compared to observed exposures. Cumulative incidence of ASD was modeled using Cox regression with smooth splines of PM2.5, neighborhood disadvantage, and their interaction, adjusted for covariates. Findings: A total of 4,548 children were diagnosed with ASD. Had the pregnancy average PM2.5 decreased by 10% or below 12 µg/m3 , 4.6 (95% CI, 2.1-7.5) and 9.8 (95% CI, 4.0- 16.5) ASD cases by age 5 would have been prevented per 10,000 children, respectively. If we further decreased the standard to 10 or 8 µg/m3 , 11.9 (95% CI, 4.1, 21.3) and 12.9 (95% CI, -0.4, 27.3) ASD cases would have been prevented per 10,000 children. Decreases in ASD incidence under hypothetical PM2.5 interventions were similar across neighborhood disadvantage levels. 118 Interpretation: Hypothetical PM2.5 interventions may reduce ASD incidence similarly across levels of neighborhood disadvantage. Our findings indicate that actions to decrease ambient PM2.5 levels to meet or surpass current standards for PM2.5 will help prevent ASD. 5.2. Introduction Autism spectrum disorder (ASD) is characterized by impairments in social interactions and communication and the presence of restricted, repetitive, and stereotyped patterns of behaviors (American Psychiatric Association 2013a). The estimated prevalence of ASD in the United States has increased from 0.66% in 2002 (Centers for Disease Control and Prevention 2007) to 2.76% in 2020 (Maenner et al. 2023). ASD imposes lifetime social and emotional hardship on children (Kuhlthau et al. 2010) and families (Rao and Beidel 2009), as well as economic burdens on families and society (Buescher et al. 2014). Accumulating evidence indicates that prenatal ambient air pollutants especially particulate matter with aerodynamic diameter < 2.5 µm (PM2.5) are modifiable environmental risk factors for ASD (Jo, Eckel, Wang, et al. 2019; Chun et al. 2020; Lam et al. 2016; Rahman et al. 2022). Our previous study has also reported that neighborhood disadvantage is associated with increased ASD likelihood (Yu et al. 2024). It is possible that people living in disadvantaged neighborhoods may be more vulnerable to environmental pollution due to higher social stress (Morello-Frosch and Shenassa 2006) and other social determinants of health which are associated with chronic inflammation and immune system dysregulation (Cohen, Doyle, and Baum 2006). Some studies have shown stronger associations between air pollution and cardiovascular diseases (Fuller et al. 2017) or cognitive decline (Christensen et al. 2022) in people living in more disadvantaged neighborhoods while mixed findings were reported for birth outcomes (Heo, Fong, and Bell 2019). One study in the U.S. focused on the interaction between 119 air pollution and neighborhood disadvantage on ASD and reported marginally significant interaction and stronger associations between PM2.5 and ASD for people living in more disadvantaged neighborhoods (McGuinn et al. 2019). Assessing the interaction between neighborhood disadvantage and air pollution can help identify vulnerable communities that can benefit most from environmental policies. While association analyses have provided evidence on the impacts of air pollution on health, intervention analyses may provide support to environmental policy and regulations (Bind 2019). In recent decades, causal inference framework has provided methodologies to estimate rigorously, under multiple assumptions, causal effects related to counterfactuals, i.e., scenarios describing what would have happened under different, possibly counter-to-the-fact, conditions (Glass et al. 2013). Emerging studies have applied the causal inference framework to evaluate the effects of PM2.5 exposure on mortality (Schwartz et al. 2015; Wang et al. 2016) and respiratory outcomes (Moore et al. 2008; Garcia et al. 2019). G-computation (Naimi, Cole, and Kennedy 2017; Hernán and Robins 2010; Robins 1986), one of the causal inference methods, estimates the standardized mean outcome given the exposure and covariate distributions. It can be applied to quantify the impacts of hypothetical exposure interventions in a counterfactual population with the same covariates distributions (Hernán and Robins 2010). Stratum-specific Gcomputation has been applied to estimate the potential effect modification by categorical variables such as gender and race/ethnicity (Josey et al. 2023). It remains a challenge for evaluate the effect modification by continuous variables using G-computation. Thus, we put forward an innovative window-based G-computation method to estimate the potential effect modification by continuous variables such as neighborhood disadvantage. 120 Based on a large birth cohort of children born in Kaiser Permanente Southern California (KPSC) hospitals from 2001-2014, we applied G-computation to estimate the potential causal effects of hypothetical air pollution interventions on the cumulative incidence of ASD by age 5. Further, we examined the modification by neighborhood disadvantage levels of the effects of hypothetical air pollution interventions on ASD incidence using the window-based Gcomputation method. 5.3. Methods 5.3.1. Study Population The retrospective birth cohort includes mothers with singleton deliveries (n=370,723) at Kaiser Permanente Southern California (KPSC) hospitals between 2001 and 2014 (see also Section 5.9.1 eMethods). A total of 52,425 births were excluded due to (1) missing gender, maternal race and ethnicity and age at delivery (n=666); (2) maternal age <15 or >55 at delivery (n=159); (3) incomplete maternal residential address at birth or geocodes not able to be assigned a US census tract (n=51,600) (Supplemental Figure 5.1). Eligible children (n=318,298) were followed until a clinical diagnosis of ASD, the last date of KPSC health plan membership, death, or age 5, whichever occurred first. Information related to mothers and children was extracted from the electronic medical records (EMR) maintained by KPSC. Institutional Review Boards at KPSC and the University of Southern California approved this study with a waiver of individual participant consent. 5.3.2. ASD ascertainment The outcome of interest was ASD diagnoses before age 5. The presence or absence of ASD diagnosis was ascertained based on at least two clinical diagnoses using International Classification of Diseases, Ninth Revision (ICD-9) codes of 299 and ICD-10 codes of F84 121 (Supplemental Table 5.1). The age of diagnosis was calculated based on the date of the first visit. This methodology for ascertaining ASD diagnoses has been previously validated in the KPSC member population, with a positive predictive value of 87% (Coleman et al. 2015), and has been employed in previous studies (Xiang et al. 2018; Xiang et al. 2015). 5.3.3. Air pollution exposure assessment PM2.5 with a 1 km spatial resolution was estimated by a hybrid model (version V4.NA.02) that integrates chemical transport model outputs, satellite observations, and groundbased measurements as developed by the Atmospheric Composition Analysis Group at Washington University in St. Louis. This modelling framework integrates satellite observations of aerosol optical depth from multiple satellite products (MISR, MODIS Dark Target, MODIS and SeaWiFS, Deep Blue, and MODIS MAIAC) and PM2.5 simulated by GEOS-Chem (http://geos-chem.org) chemical transport model to estimate ground-level mass concentrations of PM2.5, as described elsewhere (Van Donkelaar et al. 2019). This model produced good crossvalidated R2 of 0.7 for PM2.5 (Van Donkelaar et al. 2019) and has been applied to this cohort previously (Rahman, Carter, Lin, Chow, Yu, Martinez, Chen, et al. 2023). Pregnancy average PM2.5 exposures were time-weighted based on maternal residential addresses to account for changes in maternal addresses during pregnancy. 5.3.4. Neighborhood-level disadvantage index Neighborhood was defined as the US census tract of the residential address at the child’s birth. Data were extracted from the 2000 US decennial census for births from 2001-2004, and 2005-2009 and 2010-2014 American Community Survey (ACS) 5-year estimates for corresponding birth years. Neighborhood disadvantage was defined as the first principal component of seven census tract indicators (B.L. Levy, Owens, and Sampson 2019): poverty, 122 unemployment, female-headed households with children, public assistance, less than a high school education, bachelor’s degree or greater, and professional occupation (Section 5.9.1 eMethods and Supplemental Table 5.2), which has been applied to this cohort previously (Yu et al. 2024). 5.3.5. Covariates Covariates were selected a priori based on literature on ASD and air pollution(Lam et al. 2016; Rahman et al. 2022), expert knowledge and a directed acyclic graph (DAG)(Greenland, Pearl, and Robins 1999) (Supplemental Figure 5.2), including child sex at birth, parity, maternal education and race/ethnicity, maternal history of comorbidity [>=1 diagnosis of heart, lung, kidney, or liver disease; cancer], maternal age at delivery (Xiang et al. 2015). We also adjusted for birth years as a natural spline with 4 degrees of freedom to account for the non-linear increasing trend of ASD prevalence as our previous studies of this cohort (Carter et al. 2022; Rahman, Carter, Lin, Chow, Yu, Martinez, Chen, et al. 2023). An indicator variable for the season at conception (dry from April-October; wet from November-March) was used to adjust for air pollution seasonality. A missing indicator variable was used for missing values in categorical covariates (parity [n = 17,992; 5.7%], maternal education [n = 3,038; 1.0%]). 5.3.6. Hypothetical interventions Three main types of hypothetical interventions have been used in G-computation studies: setting the exposure to a fixed value; shifting the exposure down by percentage reduction; and setting a ceiling for the exposure akin to a regulatory standard (T.J.S. Smith, Keil, and Buckley 2023). Given that it is unrealistic that everyone experiences the exact same level of air pollution exposures, we focused on the latter two intervention strategies: (1) an overall percentage reduction in PM2.5 concentration by 10%, 20% and 30%; (2) reducing PM2.5 to the previous 123 standard (12 µg/m3 ) as well as more ambitious goals for reduced exposure to 10, and 8 µg/m3 whereby concentrations above the threshold were reduced to the threshold and those below the threshold remained the same. These two strategies both examined what would have happened to the cumulative incidence of ASD by age 5 if air pollution were lower than the observed exposures (the natural course). The percentage reduction strategy imposed plausible hypothetical PM2.5 exposures considering the observed distributed exposure across neighborhoods. The standard thresholds strategy can provide intuitive evidence for establishing regularity standards which should be designed to protect the most vulnerable population. 5.3.7. Statistical analysis • G-computation for Cox regression outcome models We applied G-computation with Cox proportional hazard regression outcome models (Chatton et al. 2022) to estimate the marginal effects of air pollution interventions on the cumulative incidence of ASD by age 5. G-computation involves two steps. First, Cox regression model was generated to estimate the marginal mean survival curves in the study population based on observed pregnancy average PM2.5 concentration, neighborhood disadvantage index, and covariates. We considered the nonlinearity of continuous variables, the interaction between PM2.5 and neighborhood disadvantage as we hypothesized, the interaction between PM2.5 and child sex as we previously observed in this cohort (Rahman et al. 2022), as well as the potential 2-way interactions among covariates. The selection of the best-fitted model was based on the Akaike information criterion (AIC) (Supplemental Table 5.3) and the comparison between the predicted cumulative incidence at age 5 and the Kaplan-Meier estimate. The best-fitted model included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines and their tensor interaction, an interaction between the PM2.5 cubic splines and child sex, natural splines of birth 124 year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (maternal history of severe comorbidity and maternal race/ethnicity, maternal history of severe comorbidity, and maternal age at delivery and parity). In the second step of G-computation, we used the best-fitting Cox regression model to estimate the marginal cumulative incidence of ASD by age 5 per 10,000 children under the different scenarios. For the natural course, we used observed PM2.5 concentrations, neighborhood disadvantage, and all the observed values of covariates as inputs to the outcome model. For scenarios under hypothetical interventions, we replaced the observed PM2.5 concentrations by the counterfactual PM2.5 concentrations while keeping all the observed values for neighborhood disadvantage and covariates in the outcome model. The effects of intervention strategies were estimated as the differences in the predicted cumulative incidence of ASD by age 5 between the counterfactual scenarios and the natural course. This difference can be interpreted as the number of ASD cases by age 5 per 10,000 children that would have been prevented by the hypothetical air pollution intervention, given that the necessary assumptions hold. Confidence intervals (CIs) were estimated by bootstrapping with 1,000 times replication (Efron and LePage 1992). • Window-based G-computation Effect modification by neighborhood disadvantage was examined using an innovative window-based G-computation model. In this study, windows were defined as neighborhood disadvantage deciles. For the subpopulation under each window (neighborhood disadvantage decile), the counterfactual population was that same subpopulation in each window with PM2.5 exposure set to the intervened level (all covariates remained the same as observed). Based on the 125 Cox regression model of ASD cumulative incidence estimated using the entire study population considering the nonlinearity of PM2.5, neighborhood disadvantage and their tensor interaction, we estimated the effects of air pollution interventions in each window (neighborhood disadvantage decile) as the difference in cumulative incidence between the natural course of window-based subpopulation and their counterfactual. The effect modification by neighborhood disadvantage was evaluated by the p-value for the interaction term in the Cox regression model, as well as comparing the 95% confidence intervals of the estimated effects of air pollution interventions (i.e. the differences in the cumulative incidence of ASD by age 5 between intervened PM2.5 exposure and the natural course) across neighborhood disadvantage deciles. • G-computation model assumptions To interpret results as causal effects, several assumptions are required(Naimi, Cole, and Kennedy 2017; Hernán and Robins 2010; Robins 1986). The conditional exchangeability assumes that given observed covariates, the exposed and unexposed groups are exchangeable. In the context of G-computation, it assumes that we have adequate control for confounding factors in the outcome model. Counterfactual consistency assumes that the potential outcome under intervention is consistent and well-defined. In other words, for each intervention, we have a clear definition without varying versions that might impact an intervention’s effect. The positivity assumption requires that all individuals have a non-zero probability of receiving any level of treatment (in this study, any level of counterfactual PM2.5 exposures), preventing exclusion of certain groups from the analysis and facilitating causal inference across the entire population. Last, for G-computation there is an additional assumption of correct model specification, because it is crucial that the outcome regression model—which is the basis for estimating the 126 counterfactual marginal effects— adequately capture the associations between exposure, covariates and the health outcome to obtain unbiased effect estimates. 5.3.8. Sensitivity analyses We conducted several sensitivity analyses: (1) we applied a parsimonious outcome regression model with linear associations of pregnancy PM2.5, neighborhood disadvantage, and their interaction on ASD to avoid overfitting in the outcome model; (2) we limited the population to mother-child pairs who did not move during the pregnancy to control for potential measurement errors in pregnancy average PM2.5 exposure and neighborhood disadvantage index at birth; (3) we assumed different baseline hazard functions by neighborhood disadvantage deciles in the outcome Cox regression to acknowledge that the nonparametric baseline hazard function may vary by neighborhood disadvantage deciles; (4) we estimated 95% CI using clustered bootstrapping by family to address the nested structure of children with the same mothers. Two-sided statistical tests were applied at an alpha level of 0.05. All statistical analyses were performed in R (version 4.2) (R Core Team 2023). Cox regression models were fitted by generalized additive models with REML estimation from the “mgcv” library (Wood 2011) and the bootstrapping processes were implemented by the “boot” library (Canty and Ripley 2017). 5.4. Results A total of 4,548 children (1.4 %) had an ASD diagnosis by age 5 (Table 5.1). The Kaplan-Meier estimated cumulative incidence of ASD by age 5 was 156.0 (95% CI 151.5, 160.5) per 10,000 children. Children with ASD were 4.3 times more likely to be boys (n=3,694) than girls (n=854). Mothers of children with ASD were slightly older at delivery (31.32 years, IQR=27.49-35.25) than mothers of children without ASD (30.36 years, IQR=26.23-34.28). 127 The distributions of observed and counterfactual pregnancy average PM2.5 exposure concentrations across neighborhood disadvantage deciles are shown in Figure 5.1. The observed PM2.5 exposure during pregnancy slightly increased along increasing neighborhood disadvantage deciles. The mean values were 14.02 µg/m3 and 15.54 µg/m3 for the least (decile 1) and the most (decile 10) disadvantaged neighborhoods, respectively (Supplemental Table 5.4). The observed PM2.5 exposures showed similar variances across neighborhood disadvantage deciles. Estimated effects of the two sets of hypothetical PM2.5 interventions on the cumulative incidence of ASD by age 5 are displayed in Table 5.2. Based on the best-fitted Cox regression, and under the assumptions necessary for causal interpretation, we estimated that had pregnancy PM2.5 concentrations decreased by 10%, 4.6 (95% CI 2.1-7.5) ASD cases by age 5 would have been prevented per 10,000 children. Projected to this cohort, 146.8 (95% CI 68.1-238.9) cases out of 4,548 would have been prevented. If pregnancy PM2.5 had decreased by 30%, 10.6 (95% CI 3.6-19.2) ASD cases by age 5 would have been prevented per 10,000 children, corresponding to 338.4 (95% CI 115.2-611.6) cases in this cohort. Had pregnancy average PM2.5 been no higher than the previous standard (12 µg/m3 ), 9.8 (95% CI 4.0-16.5) ASD cases by age 5 would have been prevented per 10,000 children. This means in this cohort 310.3 (95% CI 125.9-526.6) cases would have been prevented. If we further decreased the standard to 10 or 8 µg/m3 , 378.2 (95% CI 129.9-679.1) or 410.1 (95% CI -12.9-870.4) ASD cases would have been prevented in this cohort, respectively. In the best-fitted Cox regression, the nonlinear tensor interaction of PM2.5 and neighborhood disadvantage index was not significant (p-interaction=0.49) (Supplemental Figure 5.3). The effect modification by neighborhood disadvantage of the effects of hypothetical interventions is shown in Figure 5.2 for percentage reduction interventions and in Figure 5.3 for 128 standard thresholds. Although the point estimates of PM2.5 intervention effects on ASD were slightly higher in neighborhoods with low disadvantage levels (deciles 1-2), the estimated effects of all hypothetical PM2.5 interventions were not statistically significant different across neighborhood disadvantage deciles, with confidence intervals largely overlapping. Sensitivity analyses results showed similar patterns of the effects of hypothetical PM2.5 interventions across neighborhood disadvantage deciles to the main analyses when using the parsimonious linear model (Supplemental Table 5.5 and Supplemental Figure 5.4-5.5), based on non-movers (240,760 mother-child pairs) (Supplemental Table 5.6 and Supplemental Figure 5.6- 5.7), assuming different baseline hazard functions by neighborhood disadvantage deciles in the outcome model (Supplemental Table 5.7 and Supplemental Figure 5.8-5.9), and using clustered bootstrapping by family (Supplemental Table 5.8 and Supplemental Figure 5.10-5.11). 5.5. Discussion Accumulating evidence has shown associations between prenatal air pollution exposure and ASD in offspring (Jo, Eckel, Wang, et al. 2019; Chun et al. 2020; Lam et al. 2016; Rahman et al. 2022). This is the first study that applied G-computation, a causal inference method, to estimate the effects of counterfactual air pollution interventions on the cumulative incidence of ASD. Under several assumptions, discussed below, we found that both percentage reductions in pregnancy PM2.5 or restricting PM2.5 under previous and more strict standard thresholds may reduce ASD incidence by age 5. Our findings provide support to the recent revision of the primary annual PM2.5 standard by the U.S. Environmental Protection Agency to 9 µg/m3 falling within the range of stringent standards (10 and 8 µg/m3 ) examined in our study (U.S. Environmental Protection Agency 2024). However, we did not observe statistically significant 129 differences in the effects of hypothetical PM2.5 interventions across neighborhood disadvantage levels. Emerging studies have applied causal inference framework to quantify the potential public health benefits from hypothetical air pollution interventions in respiratory conditions (Garcia et al. 2019; Urman et al. 2020), dementia (Letellier et al. 2022), and birth outcomes (Keil, Buckley, and Kalkbrenner 2021). This is the first study that applied G-computation to answer the question “What would have happened to ASD incidence in children by age 5 if their prenatal air pollution exposure had been decreased?”. The two sets of hypothetical interventions we examined may give direct support to relevant policies. The percentage reduction provides practical goals considering the current observed levels of PM2.5 while the standard threshold strategies give evidence to the reevaluation of PM2.5 regulatory standards. In addition, assessing the differences in the health effects of hypothetical environmental interventions across subpopulations may help target the most vulnerable groups. One recent study has applied stratum-specific G-method to address the differences in the hypothetical PM2.5 interventions on mortality by race and income-level (Josey et al. 2023). It remains challenging to address the effect modification by a continuous variable using the causal inference framework. The innovative window-based G-computation extends the stratum-specific G method by categorical variables (such as gender and race/ethnicity) to the evaluation of effect modifiers with a continuous nature (for example, the neighborhood disadvantage index in this study). The subpopulation in each window (i.e., neighborhood disadvantage deciles in this study) served as their own counterfactuals. This novel modeling strategy can be applied to other scenarios involving hypotheses about the effects of interaction between two continuous variables. 130 Given the psychological stress and related social determinants of health (Morello-Frosch and Shenassa 2006), we initially hypothesized that people living in more disadvantaged neighborhoods are more vulnerable to air pollution-related ASD risk. However, our analyses in a large birth cohort in Southern California did not find significant interactions between prenatal PM2.5 and neighborhood disadvantages on ASD. Only a few studies have examined the interaction between air pollution and neighborhood SES on ASD. A U.S-based study reported stronger association for the joint effect between high neighborhood disadvantage and PM2.5 levels above 12.0 μg/m3 (McGuinn et al. 2019), while another study in Sweden found no significant interactions between traffic-related air pollution or PM10 and neighborhood disadvantage on ASD (Gong et al. 2017). Recent research on the susceptible time-window of air pollution on ASD risk (Rahman et al. 2022), as well as the varying associations by PM2.5 components (Rahman, Carter, Lin, Chow, Yu, Martinez, Levitt, et al. 2023; O'Sharkey et al. 2024), suggest that we may not capture the variance in the vulnerability by neighborhood disadvantage when focusing only on pregnancy average PM2.5 mass. To interpret G-computation results as causal effects, several assumptions are required (Naimi, Cole, and Kennedy 2017; Hernán and Robins 2010; Robins 1986). First, conditional exchangeability assumes adequate control for confounding factors. The Cox regression model of ASD cumulative incidence included important confounding factors identified by experts and a DAG based on current literature on air pollution and health, as well as potential risk factors of ASD likelihood (Lam et al. 2016; Rahman et al. 2022). Our regression model also included nonlinearities and interaction to account for covariates relations and reduce likelihood of residual confounding. Some other factors related to ASD risk such as genetic variants (Gaugler et al. 2014) and nutritional intake (Morse 2012) were not available in EMRs. Nonetheless, as we 131 controlled upstream SES factors in the model, they may not be strong confounders (Supplemental Figure 5.2). Given the ongoing investigation into the etiology of ASD, the inclusion of new risk factors in the outcome model, as they are identified, could enhance its comprehensiveness. Counterfactual consistency assumes a well-defined PM2.5 intervention under each examined scenario. In this study, two sets of counterfactual PM2.5 interventions (percentage reduction and standard thresholds) were clearly defined. However, the potential execution of these hypothetical interventions may vary, potentially violating this assumption. For example, suppose there are two policy strategies to reduce PM2.5; one promoting public transit to reduce on-road exhaust and another enforcing new regulation on port emissions. Although both policies may help achieve our proposed intervention goals for PM2.5 concentration, their varied approaches may lead to differences in the composition and toxicological profiles of PM2.5. Therefore, health outcomes under the same intervention goal may exhibit inconsistency, violating the assumption of counterfactual consistency. Future research could simulate scenarios with different versions of action-oriented intervention approaches and examine their potential effects on health. Positivity assumption requires that the exposure value can be experienced within all covariates combination subgroups. In this study, we examined neighborhood disadvantage levels as the key modifier. Both the least and most disadvantaged communities had small proportions of individuals whose pregnancy average PM2.5 was below 8 µg/m3 (the most ambitious intervention goal in this study) (Figure 5.1). We acknowledge that in our study sample certain subgroups of covariates combinations may have few to no people who experienced this low PM2.5 exposure concentrations; thus, the positivity assumption maybe be violated due to random 132 sampling. Given that everyone has the right and possibility to experience lower air pollution and EPA’s recent decision to lower the PM2.5 standard (U.S. Environmental Protection Agency 2024), we argue that the lower levels of 8 µg/m3 we proposed in this study are achievable and that there is no violation to the structural positivity assumption. G-computation method also requires a correctly specified model to avoid biased estimates. In the main analysis, the outcome regression model—which had the lowest AIC, an indicator of model fit, among the several models assessed—considered both the nonlinearity of PM2.5, neighborhood disadvantage index, and their nonlinear interaction. Results from sensitivity analyses with different model specifications were not markedly different. The estimated cumulative incidence of ASD at age 5 by neighborhood disadvantage deciles from the KaplanMeier estimation and the G-computation outcome regression model with tensor interaction are comparable (Supplemental Figure 5.12), indicating good fit with the observed data. Other assumptions include no interference among observations and lack of measurement error. Causal methods assume that participants are independent of others. Since ASD has relatively high heritability (Bai et al. 2019; Rylaarsdam and Guemez-Gamboa 2019), including siblings from the same family in a study may violate this assumption. In a sensitivity analysis, we addressed the nested structure by family using clustered bootstrapping. The estimated confidence intervals were comparable to the main analysis. Regarding exposure measurement error, in a sensitivity analysis including only mothers who did not move during pregnancy to reduce the potential misalignment of both neighborhood disadvantage and air pollution exposures, estimated effects were similar to the main results. For the potential misclassification in ASD diagnosis, the methodology for ascertaining ASD diagnoses applied in this study has been previously validated in the KPSC member population, with a positive predictive value of 133 87% (Coleman et al. 2015). In addition to limitations regarding interpretations of findings under assumptions, given that Cox regression is a semi-parametric model with nonparametric baseline hazard estimated based on the study cohort, the estimated effects of hypothetical interventions may not be generalizable to other populations that may have different baseline hazard. 5.6. Conclusion Based on a large birth cohort in Southern California, this study demonstrated the hypothetical interventions of ambient PM2.5 may reduce the incidence of ASD by age 5 using the G-computation causal inference method. The estimated decreases in ASD incidence were similar across communities with varying levels of neighborhood disadvantage. Findings based on both the previous standard threshold and more ambitious thresholds showed substantial decreases in ASD incidence, indicating the significance of concrete actions to actively attain the recent lower standard for PM2.5 and perhaps the reevaluation of even lower PM2.5 standard. 134 5.7. Tables Table 5.1 Characteristics of mother-child pairs Children, No. (%) or median [interquartile range] Characteristics Overall With ASD Without ASD (n = 318,298) (n = 4,548) (n = 313,750) Pregnancy average PM2.5 (median [IQR]) 15.09 [13.15, 16.89] 14.79 [13.06, 16.28] 15.09 [13.15, 16.90] Neighborhood disadvantage index (median [IQR]) 0.07 [-1.22, 1.48] 0.07 [-1.31, 1.53] 0.07 [-1.21, 1.48] Povertya 9.50 [4.70, 17.30] 10.10 [4.90, 18.10] 9.50 [4.70, 17.30] Unemploymenta 8.80 [6.20, 11.90] 9.15 [6.40, 12.20] 8.80 [6.20, 11.88] Female-headed householdsa 17.53 [11.74, 24.58] 17.81 [12.09, 25.33] 17.52 [11.74, 24.58] Public assistancea 3.30 [1.60, 6.30] 3.32 [1.64, 6.40] 3.30 [1.60, 6.30] Less than high schoola 21.20 [10.60, 35.60] 21.10 [9.79, 35.80] 21.23 [10.60, 35.60] Bachelor and post-graduatea 19.30 [10.50, 32.50] 20.40 [11.02, 34.60] 19.26 [10.50, 32.50] Professional occupationa 28.90 [19.10, 40.80] 29.20 [19.30, 42.57] 28.90 [19.10, 40.77] Sex = Male, No. (%) 162,944 (51.2) 3,694 (81.2) 159,250 (50.8) Maternal age at delivery, (median [IQR]), years 30.38 [26.25, 34.29] 31.32 [27.49, 35.25] 30.36 [26.23, 34.28] Parity, No. (%) 1 111,816 (35.1) 1,842 (40.5) 109,974 (35.1) 2 104,405 (32.8) 1,490 (32.8) 102,915 (32.8) >2 84,085 (26.4) 901 (19.8) 83,184 (26.5) Unknown 17,992 ( 5.7) 315 ( 6.9) 17,677 ( 5.6) Maternal education, No. (%) High school or lower 112,029 (35.2) 1,334 (29.3) 110,695 (35.3) Some college 94,406 (29.7) 1,472 (32.4) 92,934 (29.6) College graduate or higher 108,825 (34.2) 1,708 (37.6) 107,117 (34.1) Unknown 3,038 ( 1.0) 34 ( 0.7) 3,004 ( 1.0) Maternal race/ethnicity, No. (%) Non-Hispanic White 80,865 (25.4) 953 (21.0) 79,912 (25.5) Non-Hispanic Black 29,752 ( 9.3) 446 ( 9.8) 29,306 ( 9.3) Hispanic/Latino 161,304 (50.7) 2,299 (50.5) 159,005 (50.7) Asian/Pacific Islander 39,845 (12.5) 738 (16.2) 39,107 (12.5) Other 6,532 ( 2.1) 112 ( 2.5) 6,375 ( 2.0) History of maternal comorbiditya , No. (%) 46,635 (14.7) 1,150 (18.1) 6,420 ( 2.0) Year of birth, No. (%) 2001-2004 78,213 (24.6) 817 (18.0) 77,396 (24.7) 2005-2009 111,039 (34.9) 1,306 (28.7) 109,733 (35.0) 2010-2014 129,046 (40.5) 2,425 (53.3) 126,621 (40.4) Abbreviations: ASD, autism spectrum disorder; IQR, interquartile range; PM2.5, particulate matter with aerodynamic diameter < 2.5 µm a ≥Percentage of each neighborhood SES census tract indicator, for example the percentage of families below the poverty line across all Census tracts. b ≥1 diagnosis of heart, lung, kidney, or liver disease; cancer. 135 Table 5.2 Estimated effects of hypothetical PM2.5 interventions on the risk of ASD cases by age 5 Intervention Estimated cumulative incidence by age 5 per 10,000 (95% CI) Decrease in cumulative incidence by age 5 per 10,000 (95% CI) Cases would have been prevented in the study population (95% CI) Natural course 155.4 (150.9, 159.9) reference reference Percentage reduction 10% decrease 150.8 (145.5, 155.7) 4.6 (2.1, 7.5) 146.8 (68.1, 238.9) 20% decrease 147.3 (140.1, 153.3) 8.1 (3.6, 13.8) 258.3 (114.2, 440.7) 30% decrease 144.8 (135.2, 152.8) 10.6 (3.6, 19.2) 338.4 (115.2, 611.6) Standard thresholds 12 µg/m3 standard 145.6 (137.6, 152.7) 9.8 (4.0, 16.5) 310.3 (125.9, 526.6) 10 µg/m3 standard 143.5 (133.2, 152.2) 11.9 (4.1, 21.3) 378.2 (129.9, 679.1) 8 µg/m3 standard 142.5 (127.5, 156.4) 12.9 (-0.4, 27.3) 410.1 (-12.9, 870.4) Abbreviations: ASD, autism spectrum disorder; CI, confidence interval; PM2.5, particulate matter with aerodynamic diameter < 2.5 µm Notes: Results were estimated based on the Cox regression model of ASD cumulative incidence by age 5 with the lowest AIC. The model included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 136 5.8. Figures Figure 5.1 Distribution of pregnancy average PM2.5 concentrations of hypothetical interventions by neighborhood disadvantage deciles Notes: For x-axis, “1” represents the least disadvantaged neighborhoods and “10” represents the most disadvantaged neighborhoods. 137 Figure 5.2 Estimated effects of hypothetical PM2.5 interventions (percentage reduction) on the cumulative incidence of ASD children by neighborhood disadvantage deciles Notes: On x-axis, “1” represents the least disadvantaged neighborhoods and “10” represents the most disadvantaged neighborhoods. Results were estimated based on the window-based G-computation where windows are defined as neighborhood disadvantage deciles. The Cox regression model of ASD incidence included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 138 Figure 5.3 Estimated effects of hypothetical PM2.5 interventions (standard thresholds) on the cumulative incidence of ASD by neighborhood deprivation deciles Notes: On x-axis, “1” represents the least disadvantaged neighborhoods and “10” represents the most disadvantaged neighborhoods. Results were estimated based on the window-based G-computation where windows are defined as neighborhood disadvantage deciles. The Cox regression model of ASD incidence included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 139 5.9. Supplemental Material 5.9.1. eMethods • Study population This retrospective cohort study included mothers with singleton deliveries (n=370,723) at Kaiser Permanente Southern California (KPSC) hospitals between January 1, 2001, and December 31, 2014. KPSC is a large integrated healthcare system with over 4.5 million members across 10 Southern California counties (Imperial, Kern, Los Angeles, Orange, Riverside, San Bernardino, San Diego, Santa Barbara, San Luis Obispo and Ventura), with diverse membership similar in socioeconomic characteristics to the region(Koebnick et al. 2012). The KPSC member birth cohort in this study included mothers and their children covered by employer-sponsor coverage as well as Medicaid, as insurance type was not a criterion for exclusion. In general, 9% of Kaiser members are covered by Medicaid or the Children's Health Insurance Program (CHIP). • Neighborhood disadvantage index Neighborhood was defined as the US census tract of the mother’s residential address at the time of the child’s birth, geocoded using ArcGIS 12. Data from the 2000 US decennial census and the American Community Survey (ACS) 5-year estimates for 2005-2009 and 2010- 2014 were assigned to children born between 2001-2004, 2005-2009, and 2010-2014, respectively. ACS data based on 2010 census tract boundaries (2010-2014 estimates) were matched to the 2000 census tract boundaries using crosswalks from the Longitudinal Tract Data Base (Logan, Xu, and Stults 2014). Neighborhood disadvantage was based on seven census tract indicators(B.L. Levy, Owens, and Sampson 2019), as described previously(Yu et al. 2024) : the percentage of (1) families below the poverty line; (2) unemployment in the civilian labor force; (3) female-headed 140 households with children; (4) households receiving public assistance income; (5) population 25 years and over with less than high school education; (6) population 25 years and over with Bachelor’s degree or greater; and (7) the population employed in professional or managerial occupations in the civilian labor force. The first principal component was conducted via the princomp function in stats package and the loadings for seven indicators are shown in Supplemental Table 5.2. Supplemental Table 5.1 ICD codes and diagnostic description for ASD ascertainment ICD-9 codes Description 299.0 Autistic disorder 299.1 Childhood disintegrative disorder 299.8 Other specified pervasive developmental disorders 299.9 Unspecified pervasive developmental disorder ICD-10 codes F84.0 Autistic disorder F84.3 Other childhood disintegrative disorder F84.5 Asperger's syndrome F84.8 Other pervasive developmental disorders F84.9 Pervasive developmental disorder, unspecified Supplemental Table 5.2 Loadings of the neighborhood disadvantage index Disadvantage indexa Poverty 0.407 Unemployed 0.332 Female-headed households 0.260 Public assistance 0.382 Less than high school education 0.415 Bachelor and post-graduate degree -0.403 Occupation: Professional -0.418 141 Supplemental Table 5.3 Outcome Cox Regression Model Selection Model Description AIC Model 1 • Pregnancy PM2.5 as linear term • Neighborhood disadvantage as linear term • The interaction between linear pregnancy PM2.5 and neighborhood disadvantage • Adjusted birth year as natural splines with 4 degrees of freedom and other covariates 111733.6 Model 2 • Pregnancy PM2.5 as smooth cubic splines • Neighborhood disadvantage as smooth cubic splines • The interaction between smooth cubic splines of pregnancy PM2.5 and smooth cubic splines of neighborhood disadvantage • Adjusted birth year as natural splines with 4 degrees of freedom and other covariates 111719.8 Model 3 • Pregnancy PM2.5 as smooth cubic splines • Neighborhood disadvantage as smooth cubic splines • The interaction between smooth cubic splines of pregnancy PM2.5 and smooth cubic splines of neighborhood disadvantage • The interaction between smooth cubic splines of pregnancy PM2.5 and child’s sex • Adjusted birth year as natural splines with 4 degrees of freedom and other covariates 111715.3 Model 4 • Pregnancy PM2.5 as smooth cubic splines • Neighborhood disadvantage as smooth cubic splines • The interaction between smooth cubic splines of pregnancy PM2.5 and smooth cubic splines of neighborhood disadvantage • The interaction between smooth cubic splines of pregnancy PM2.5 and child’s sex • Adjusted birth year as natural splines with 4 degrees of freedom, delivery age as natural splines with 3 degrees of freedom and other covariates 111708.5 Model 5 • Pregnancy PM2.5 as smooth cubic splines • Neighborhood disadvantage as smooth cubic splines • The interaction between smooth cubic splines of pregnancy PM2.5 and smooth cubic splines of neighborhood disadvantage • The interaction between smooth cubic splines of pregnancy PM2.5 and child’s sex • Adjusted birth year as natural splines with 4 degrees of freedom, delivery age as natural splines with 3 degrees of freedom and other covariates • Adjusted for 2-way interactions among covariates (prepregnancy comorbidity*maternal race/ethnicity, natural splines of delivery age and parity) 111682.5 Model 6 • Pregnancy PM2.5 as smooth cubic splines • Neighborhood disadvantage as smooth cubic splines • The 3-way interaction between smooth cubic splines of pregnancy PM2.5, smooth cubic splines of neighborhood disadvantage and child’ sex • Adjusted birth year as natural splines with 4 degrees of freedom, delivery age as natural splines with 3 degrees of freedom and other covariates • Adjusted for 2-way interactions among covariates (prepregnancy comorbidity*maternal race/ethnicity, natural splines of delivery age and parity) 111689 142 Supplemental Table 5.4 Characteristics by neighborhood disadvantage index deciles Neighborhood Disadvantage Decile 1 (lowest) 2 3 4 5 6 7 8 9 10 (highest) Population, No. 31925 31744 31844 31826 31831 31860 31785 31869 31800 31814 ASD case, No. (%) 517 ( 1.6) 488 ( 1.5) 435 ( 1.4) 421 ( 1.3) 417 ( 1.3) 407 ( 1.3) 489 ( 1.5) 419 ( 1.3) 487 ( 1.5) 468 ( 1.5) Pregnancy average PM2.5 (median [IQR]) 14.02 [12.37, 15.95] 14.45 [12.56, 16.28] 13.99 [11.49, 16.01] 14.84 [12.65, 16.65] 15.15 [13.36, 16.93] 15.51 [13.55, 17.52] 15.31 [13.37, 17.00] 15.63 [13.75, 17.72] 15.63 [14.02, 17.62] 15.54 [13.80, 17.59] Sex = Male, No. (%) 16472 (51.6) 16327 (51.4) 16327 (51.3) 16296 (51.2) 16203 (50.9) 16244 (51.0) 16181 (50.9) 16289 (51.1) 16306 (51.3) 16299 (51.2) Maternal age at delivery, (median [IQR]), years 32.93 [29.64, 36.19] 31.85 [28.35, 35.28] 31.25 [27.62, 34.82] 30.68 [26.83, 34.43] 30.26 [26.31, 34.12] 29.84 [25.78, 33.73] 29.50 [25.43, 33.56] 29.09 [24.87, 33.25] 28.84 [24.63, 33.06] 28.47 [24.19, 32.77] Parity, No. (%) 1 12470 (39.1) 11978 (37.7) 12070 (37.9) 11668 (36.7) 11397 (35.8) 11179 (35.1) 10670 (33.6) 10551 (33.1) 10175 (32.0) 9658 (30.4) 2 11272 (35.3) 11266 (35.5) 10669 (33.5) 10637 (33.4) 10355 (32.5) 10353 (32.5) 10163 (32.0) 10011 (31.4) 9906 (31.2) 9773 (30.7) >2 6249 (19.6) 6640 (20.9) 7166 (22.5) 7745 (24.3) 8110 (25.5) 8706 (27.3) 9094 (28.6) 9663 (30.3) 9967 (31.3) 10745 (33.8) Unknown 1934 ( 6.1) 1860 ( 5.9) 1939 ( 6.1) 1776 ( 5.6) 1969 ( 6.2) 1622 ( 5.1) 1858 ( 5.8) 1644 ( 5.2) 1752 ( 5.5) 1638 ( 5.1) Maternal education, No. (%) High school or lower 3940 (12.3) 6140 (19.3) 7710 (24.2) 9597 (30.2) 10578 (33.2) 12514 (39.3) 13447 (42.3) 15020 (47.1) 15890 (50.0) 17193 (54.0) Some college 6430 (20.1) 8345 (26.3) 8983 (28.2) 9744 (30.6) 10256 (32.2) 10267 (32.2) 10313 (32.4) 10089 (31.7) 10146 (31.9) 9833 (30.9) College graduate or higher 21211 (66.4) 16966 (53.4) 14868 (46.7) 12249 (38.5) 10714 (33.7) 8799 (27.6) 7755 (24.4) 6452 (20.2) 5416 (17.0) 4395 (13.8) Unknown 344 ( 1.1) 293 ( 0.9) 283 ( 0.9) 236 ( 0.7) 283 ( 0.9) 280 ( 0.9) 270 ( 0.8) 308 ( 1.0) 348 ( 1.1) 393 ( 1.2) Maternal race/ethnicity, No. (%) Non-Hispanic White 15329 (48.0) 13485 (42.5) 12768 (40.1) 10504 (33.0) 8008 (25.2) 6785 (21.3) 5578 (17.5) 4058 (12.7) 2554 ( 8.0) 1796 ( 5.6) Non-Hispanic Black 1168 ( 3.7) 1552 ( 4.9) 1797 ( 5.6) 2370 ( 7.4) 2566 ( 8.1) 2807 ( 8.8) 3215 (10.1) 3967 (12.4) 4459 (14.0) 5851 (18.4) Hispanic/Latino 6269 (19.6) 9475 (29.8) 11205 (35.2) 13731 (43.1) 16372 (51.4) 18243 (57.3) 19623 (61.7) 21090 (66.2) 22607 (71.1) 22689 (71.3) Asian/Pacific Islander 8320 (26.1) 6351 (20.0) 5256 (16.5) 4476 (14.1) 4158 (13.1) 3410 (10.7) 2773 ( 8.7) 2229 ( 7.0) 1738 ( 5.5) 1134 ( 3.6) Other 839 ( 2.6) 881 ( 2.8) 818 ( 2.6) 745 ( 2.3) 727 ( 2.3) 615 ( 1.9) 596 ( 1.9) 525 ( 1.6) 442 ( 1.4) 344 ( 1.1) History of maternal comorbiditya , No. (%) 4388 (13.7) 4677 (14.7) 4888 (15.3) 4815 (15.1) 4914 (15.4) 4665 (14.6) 4669 (14.7) 4613 (14.5) 4490 (14.1) 4516 (14.2) Povertyb 2.60 [1.40, 4.40] 3.60 [2.20, 5.70] 4.50 [2.80, 6.80] 6.10 [4.10, 8.40] 8.30 [5.90, 10.90] 10.30 [7.90, 12.80] 12.65 [10.00, 16.10] 16.10 [12.70, 19.98] 20.90 [17.60, 24.90] 30.24 [25.50, 35.90] Unemploymentb 5.00 [3.50, 6.80] 6.18 [4.43, 8.30] 7.00 [5.10, 9.55] 7.40 [5.77, 9.60] 8.31 [6.50, 10.80] 9.00 [7.10, 11.76] 9.83 [7.70, 12.35] 11.10 [8.70, 13.80] 11.85 [9.70, 14.75] 14.47 [11.17, 17.90] Female-headed householdsb 9.67 [6.21, 14.31] 12.40 [8.37, 17.53] 13.64 [9.96, 19.33] 15.36 [11.26, 20.71] 17.20 [12.53, 22.78] 17.69 [13.02, 23.96] 18.58 [14.50, 24.83] 21.22 [15.17, 27.84] 23.14 [17.59, 29.27] 28.84 [22.44, 35.52] Public assistanceb 0.90 [0.40, 1.60] 1.40 [0.70, 2.10] 1.71 [1.00, 2.60] 2.30 [1.40, 3.30] 2.90 [1.85, 4.30] 3.79 [2.50, 5.20] 4.70 [3.10, 6.50] 6.00 [4.20, 7.90] 7.70 [5.60, 9.70] 11.30 [8.70, 14.70] 143 Supplemental Table 5.4 cont. Abbreviations: ASD, autism spectrum disorder; IQR, interquartile range; PM2.5, particulate matter with aerodynamic diameter < 2.5 µm a ≥1 diagnosis of heart, lung, kidney, or liver disease; cancer. b ≥Percentage of each neighborhood SES census tract indicator, for example the percentage of families below the poverty line across all Census tracts. Less than high schoolb 4.30 [2.70, 6.20] 7.30 [5.70, 9.90] 11.10 [8.40, 13.90] 14.60 [11.40, 19.20] 19.70 [15.40, 24.80] 24.60 [19.10, 30.40] 29.70 [24.20, 36.40] 35.30 [28.50, 43.50] 42.65 [35.50, 51.40] 51.80 [42.80, 60.35] Bachelor and postgraduateb 54.80 [48.00, 61.70] 38.80 [33.90, 44.70] 31.70 [26.80, 35.80] 24.60 [20.43, 29.90] 21.20 [16.50, 25.40] 16.10 [13.00, 21.20] 13.70 [10.20, 17.40] 11.00 [7.60, 15.00] 8.00 [5.60, 10.80] 5.40 [3.60, 7.70] Professional occupationb 57.20 [52.30, 62.90] 45.70 [42.30, 50.40] 39.90 [36.00, 44.10] 34.80 [30.90, 38.80] 30.70 [26.90, 34.28] 25.70 [22.25, 30.00] 22.60 [19.30, 26.84] 19.00 [15.60, 23.40] 15.60 [12.40, 19.45] 11.73 [9.30, 15.20] 144 Supplemental Table 5.5 Estimated effects of hypothetical PM2.5 interventions on the risk of ASD cases by age 5 based on the parsimonious linear Cox regression model Intervention Estimated cumulative incidence by age 5 per 10,000 (95% CI) Decrease in cumulative incidence by age 5 per 10,000 (95% CI) Cases would have been prevented in the study population (95% CI) Natural course 155.4 (150.9, 159.9) reference reference Percentage reduction 10% decrease 151.2 (146.2, 156.2) 4.2 (1.7, 6.7) 133.4 (55.6, 213.1) 20% decrease 147.1 (140.7, 153.4) 8.2 (3.5, 13.1) 261.8 (111.5, 417.7) 30% decrease 143.2 (135.0, 151.2) 12.1 (5.3, 19.3) 385.5 (167.3, 614.2) Standard thresholds 12 µg/m3 standard 146.7 (140.1, 153.2) 8.6 (3.7, 13.8) 275.2 (116.4, 440.2) 10 µg/m3 standard 142.2 (133.4, 150.6) 13.2 (5.7, 21.0) 419.9 (182.4, 669.9) 8 µg/m3 standard 137.4 (126.2, 148.1) 17.9 (8.0, 28.5) 570.1 (253.9, 907.7) Abbreviations: ASD, autism spectrum disorder; CI, confidence interval; PM2.5, particulate matter with aerodynamic diameter < 2.5 µm Notes: Results were estimated based on the parsimonious Cox regression model of ASD cumulative incidence by age 5. The model included pregnancy PM2.5 and neighborhood disadvantage as linear terms, their linear interaction, interaction between the PM2.5 and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and twoway interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 145 Supplemental Table 5.6 Estimated effects of hypothetical PM2.5 interventions on the risk of ASD cases by age 5 based on non-movers Intervention Estimated cumulative incidence by age 5 per 10,000 (95% CI) Decrease in cumulative incidence by age 5 per 10,000 (95% CI) Cases would have been prevented in the study population (95% CI) Natural course 156.5 (151.3, 161.7) reference reference Percentage reduction 10% decrease 151.1 (144.6, 156.6) 5.3 (2.7, 9.2) 128.6 (65.2, 220.9) 20% decrease 147.0 (137.7, 153.8) 9.5 (4.5, 17.0) 229.2 (107.5, 410.3) 30% decrease 143.6 (132.5, 152.9) 12.9 (4.9, 22.8) 310.7 (119.1, 548.0) Standard thresholds 12 µg/m3 standard 145.7 (134.0, 153.3) 10.8 (4.5, 21.2) 258.8 (108.9, 511.2) 10 µg/m3 standard 142.5 (128.8, 152.7) 14.0 (4.9, 26.6) 336.3 (118.0, 641.2) 8 µg/m3 standard 139.6 (124.6, 156.3) 16.9 (1.1, 31.1) 406.6 (25.8, 748.6) Abbreviations: ASD, autism spectrum disorder; CI, confidence interval; PM2.5, particulate matter with aerodynamic diameter < 2.5 µm Notes: Results were estimated based on the Cox regression model of ASD cumulative incidence by age 5 with the lowest AIC, generated based on non-movers. The model included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 146 Supplemental Table 5.7 Estimated effects of hypothetical PM2.5 interventions on the risk of ASD cases by age 5 based Cox regression models with different baseline hazard functions by neighborhood disadvantage deciles Intervention Estimated cumulative incidence by age 5 per 10,000 (95% CI) Decrease in cumulative incidence by age 5 per 10,000 (95% CI) Cases would have been prevented in the study population (95% CI) Natural course 155.4 (150.9, 159.9) reference reference Percentage reduction 10% decrease 150.7 (145.4, 155.6) 4.7 (2.2, 7.6) 148.3 (68.9, 240.8) 20% decrease 147.1 (139.9, 153.2) 8.2 (3.7, 14.0) 261.8 (116.4, 445.9) 30% decrease 144.5 (134.9, 152.6) 10.8 (3.8, 19.5) 344.4 (119.9, 620.3) Standard thresholds 12 µg/m3 standard 145.5 (137.5, 152.7) 9.8 (4.0, 16.6) 312.6 (127.0, 529.9) 10 µg/m3 standard 143.3 (132.9, 152.1) 12.1 (4.2, 21.6) 384.0 (134.9, 688.4) 8 µg/m3 standard 142.1 (127.1, 155.9) 13.2 (0.0, 27.8) 421.4 (0.0, 884.9) Abbreviations: ASD, autism spectrum disorder; CI, confidence interval; PM2.5, particulate matter with aerodynamic diameter < 2.5 µm Notes: Results were estimated based on the Cox regression model of ASD cumulative incidence by age 5 assuming different baseline hazard functions by neighborhood disadvantage deciles. The model included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 147 Supplemental Table 5.8 Estimated effects of hypothetical PM2.5 interventions on the risk of ASD cases by age 5 with confidence intervals based on family-clustered bootstrapping Intervention Estimated cumulative incidence by age 5 per 10,000 (95% CI) Decrease in cumulative incidence by age 5 per 10,000 (95% CI) Cases would have been prevented in the study population (95% CI) Natural course 155.4 (150.6, 160.1) reference reference Percentage reduction 10% decrease 150.8 (145.1, 156.0) 4.6 (2.1, 7.5) 146.8 (65.3, 237.6) 20% decrease 147.3 (139.8, 153.8) 8.1 (3.4, 13.8) 258.3 (106.8, 438.4) 30% decrease 144.8 (134.9, 153.4) 10.6 (3.2, 19.2) 338.4 (102.0, 610.7) Standard thresholds 12 µg/m3 standard 145.6 (137.4, 153.1) 9.8 (3.8, 16.3) 310.3 (120.0, 520.2) 10 µg/m3 standard 143.5 (133.0, 152.9) 11.9 (3.6, 21.2) 378.2 (113.6, 674.6) 8 µg/m3 standard 142.5 (127.3, 157.3) 12.9 (-1.2, 27.4) 410.1 (-38.7, 872.1) Abbreviations: ASD, autism spectrum disorder; CI, confidence interval; PM2.5, particulate matter with aerodynamic diameter < 2.5 µm Notes: Results were estimated based on the Cox regression model of ASD cumulative incidence by age 5 with the lowest AIC. The model included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). The confidence intervals were estimated by family-clustered bootstrapping. 148 Supplemental Figure 5.1 Derivation of study population 370,723 singleton births in KPSC hospitals between 1/01/2001 and 31/12/2014 with KPSC membership at age 1 and alive at age 1 318,298 included in primary analysis 52,425 Excluded: 666* missing or errors in covariates (90 birth weight; 476 gender; 96 maternal race-ethnicity; 44 maternal age at delivery; 40 missing or errors in 2 covariates) 159 maternal age at delivery < 15 yrs or > 55 yrs 51,600 Missing/incomplete addresses, non-California addresses, or addresses not suitable for assigning census data 149 Supplemental Figure 5.2 DAG Abbreviations: ASD, autism spectrum disorder; PM2.5, particulate matter with aerodynamic diameter < 2.5 µm; SES, socioeconomic status. Notes: Green nodes represent the key exposures of interest in this study; Blue node is the health outcome of interest; White nodes are the adjusted covariates in the outcome Cox regression model; Gray nodes represent the unmeasured variables on the potential pathways; Squared nodes are the population selection criteria. In this study, we used population with health insurance and good health care access. 150 Supplemental Figure 5.3 The splines of pregnancy PM2.5 and neighborhood disadvantage, and their tensor interaction in the best-fitted outcome model Notes: The p-values for splines were calculated by a Wald-type test (Wood 2013). Results were estimated based on the Cox regression model of ASD cumulative incidence by age 5 with the lowest AIC. The model included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 151 Supplemental Figure 5.4 Estimated effects of hypothetical PM2.5 interventions (percentage reduction) on the cumulative incidence of ASD by neighborhood deprivation deciles based the parsimonious outcome model Notes: On x-axis, “1” represents the least disadvantaged neighborhoods and “10” represents the most disadvantaged neighborhoods. Results were estimated based on the parsimonious Cox regression model of ASD cumulative incidence by age 5. The model included pregnancy PM2.5 and neighborhood disadvantage as linear term, their linear interaction, interaction between the PM2.5 and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 152 Supplemental Figure 5.5 Estimated effects of hypothetical PM2.5 interventions (standard thresholds) on the cumulative incidence of ASD by neighborhood deprivation deciles based the parsimonious outcome model Notes: On x-axis, “1” represents the least disadvantaged neighborhoods and “10” represents the most disadvantaged neighborhoods. Results were estimated based on the parsimonious Cox regression model of ASD cumulative incidence by age 5. The model included pregnancy PM2.5 and neighborhood disadvantage as linear term, their linear interaction, interaction between the PM2.5 and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 153 Supplemental Figure 5.6 Estimated effects of hypothetical PM2.5 interventions (percentage reduction) on the cumulative incidence of ASD by neighborhood deprivation deciles based non-movers Notes: On x-axis, “1” represents the least disadvantaged neighborhoods and “10” represents the most disadvantaged neighborhoods. Results were estimated based on the Cox regression model of ASD cumulative incidence by age 5 with the lowest AIC, generated based on non-movers. The model included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 154 Supplemental Figure 5.7 Estimated effects of hypothetical PM2.5 interventions (standard thresholds) on the cumulative incidence of ASD by neighborhood deprivation deciles based non-movers Notes: On x-axis, “1” represents the least disadvantaged neighborhoods and “10” represents the most disadvantaged neighborhoods. Results were estimated based on the Cox regression model of ASD cumulative incidence by age 5 with the lowest AIC, generated based on non-movers. The model included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 155 Supplemental Figure 5.8 Estimated effects of hypothetical PM2.5 interventions (percentage reduction) on the cumulative incidence of ASD by neighborhood deprivation deciles based on Cox regression models with different baseline hazard functions by neighborhood disadvantage deciles Notes: On x-axis, “1” represents the least disadvantaged neighborhoods and “10” represents the most disadvantaged neighborhoods. Results were estimated based on the Cox regression model of ASD cumulative incidence by age 5 assuming different baseline hazard functions by neighborhood disadvantage deciles. The model included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 156 Supplemental Figure 5.9 Estimated effects of hypothetical PM2.5 interventions (standard thresholds) on the cumulative incidence of ASD by neighborhood deprivation deciles based on Cox regression models with different baseline hazard functions by neighborhood disadvantage deciles Notes: On x-axis, “1” represents the least disadvantaged neighborhoods and “10” represents the most disadvantaged neighborhoods. Results were estimated based on the Cox regression model of ASD cumulative incidence by age 5 assuming different baseline hazard functions by neighborhood disadvantage deciles. The model included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 157 Supplemental Figure 5.10 Estimated effects of hypothetical PM2.5 interventions (percentage reduction) on the cumulative incidence of ASD by neighborhood deprivation deciles with confidence intervals based on family-clustered bootstrapping Notes: On x-axis, “1” represents the least disadvantaged neighborhoods and “10” represents the most disadvantaged neighborhoods. Results were estimated based on the Cox regression model of ASD cumulative incidence by age 5 with the lowest AIC. The model included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). The confidence intervals were estimated by family-clustered bootstrapping. 158 Supplemental Figure 5.11 Estimated effects of hypothetical PM2.5 interventions (standard thresholds) on the cumulative incidence of ASD by neighborhood deprivation deciles with confidence intervals based on family-clustered bootstrapping Notes: On x-axis, “1” represents the least disadvantaged neighborhoods and “10” represents the most disadvantaged neighborhoods. Results were estimated based on the Cox regression model of ASD cumulative incidence by age 5 with the lowest AIC. The model included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). The confidence intervals were estimated by family-clustered bootstrapping. 159 Supplemental Figure 5.12 Comparison of the predicted cumulative incidence of ASD at age 5 per 10,000 children between G-computation and Kaplan-Meier estimates Notes: On x-axis, “1” represents the least disadvantaged neighborhoods and “10” represents the most disadvantaged neighborhoods. Results were estimated based on the window-based G-computation where windows are defined as neighborhood disadvantage deciles. The Cox regression model of ASD incidence included pregnancy PM2.5 and neighborhood disadvantage index as cubic splines, their tensor interaction, interaction between the PM2.5 splines and child sex, natural splines of birth year with 4 degrees of freedom, natural splines of maternal age at delivery with 3 degrees of freedom, parity, maternal education, maternal race/ethnicity, maternal history of severe comorbidity, season and two-way interactions among covariates (history of maternal comorbidity and maternal race/ethnicity, history of maternal comorbidity and maternal education, and maternal age at delivery and parity). 160 Chapter 6 Summary and Suggestions for Future Research Autism spectrum disorder (ASD) characterized by deficits in communication and stereotyped repeated behaviors, affects 1 in 36 of children in the United States in 2020 (Maenner et al. 2023). Despite ongoing investigation, the precise etiology of ASD remains elusive. Twin studies estimating the heritability of ASD have yielded a wide range, from 50% to 90% (Bai et al. 2019; Tick et al. 2016), indicating that beyond genetics, environmental factors may also contribute to its development. Studies have reported several potential environmental factors associated with ASD, including early-life air pollution (Flores-Pajot et al. 2016; Lam et al. 2016; Chun et al. 2020), maternal health conditions (Zerbo et al. 2013; Jiang et al. 2016; Xiang et al. 2015) and health behaviors (von Ehrenstein et al. 2020), and SES characteristics (Rai et al. 2012; Lehti et al. 2015). However, the potential interactive associations among these co-existing factors on ASD likelihood is understudied. Prenatal air pollution, particularly PM exposures, is consistently reported to be associated with higher likelihood of ASD (Flores-Pajot et al. 2016; Lam et al. 2016; Chun et al. 2020). As a modifiable environmental factor, understanding the variability in the susceptibility of ASD likelihood associated with air pollution holds potential to support vulnerable populations. Therefore, a key goal of this dissertation is to explore the potential effect modifiers of the association between prenatal air pollution and ASD. Two primary hypotheses were examined: (1) The maternal immune activation may modify the association between prenatal PM2.5 and ASD; (2) Population living more disadvantaged neighborhoods may be more vulnerable to air pollution-related ASD risk. Additionally, the main association between neighborhood disadvantage and ASD likelihood was reported, addressing the divergent findings in existing literature (Yu et al. 2024). This chapter provides a synthesis of the key findings from Chapters 2- 161 5. Subsequently, the dissertation's contributions and limitations are discussed, followed by recommendations for future research directions that can build upon these findings. 6.1. Summary of Findings Chapter 2 is a systematic review on the effect modifiers of the association between earlylife air pollution and ASD likelihood (Yu et al. 2022). Among the 39 papers examining the association between air pollution and ASD, 19 of them examined potential effect modifiers, which can be grouped into three broad categories: social and demographic characteristics, maternal health conditions and nutritional intake during pregnancy, and genetics. Maternal gestational diabetes, insufficient folic acid intake during pregnancy, low maternal education, child male sex and high CNV burden or MET rs1858830 CC genotype may increase susceptibility to the effect of prenatal air pollution exposure on ASD risk. However, the limited number of studies on each modifier precludes definitive conclusions. Chapter 3 presents empirical research which examined the potential interaction between conditions of maternal immune activation and prenatal air pollution exposures on the likelihood of ASD (Yu et al. 2023). Based on a large retrospective birth cohort in KPSC between 2001 and 2014, I found that MIA-related conditions (infection, hypertension, asthma, and autoimmune diseases) during pregnancy and prenatal PM2.5 (or its components) were independently associations with ASD likelihood. However, there were no statistically significant interactions of MIA conditions and prenatal PM2.5 exposure with ASD likelihood. Chapter 4 aimed to examine the main association between neighborhood disadvantage and ASD diagnosis (Yu et al. 2022). Given the mixed previous findings on the association between SES and ASD reported in studies from different regions with varying healthcare policies, it is possible that the disparity in the access to neurodevelopmental disorder screening 162 and diagnosis may mask the underlying gradient between neighborhood SES and ASD. In this study, I utilized a large study population with health insurance and good access to the screening and diagnostic services of ASD, and found that living in more disadvantaged neighborhoods was associated with higher likelihood of ASD diagnosis in a population with health insurance. Chapter 5 focused on the potential interaction between neighborhood disadvantage and prenatal air pollution on ASD likelihood. It aimed to answer an intervention-oriented question: “How would hypothetical air pollution interventions affect the cumulative incidence of ASD in children before age 5 under different neighborhood socioeconomic context?”. Two sets of interventions were examined: (1) percentage reduction (10%, 20%, and 30% decrease) and (2) standard threshold of PM2.5 (12, 10, and 8 μg/m3 ). Using the G-computation method, I found that these hypothetical interventions of ambient PM2.5 may reduce the incidence of ASD. The estimated decreases in ASD incidence were similar across communities with varying levels of neighborhood disadvantage. 6.2. Contribution to Knowledge The contribution of this dissertation can be summarized into two aspects: (1) I confirmed the main associations between prenatal PM2.5 exposures, conditions of MIA during pregnancy, and neighborhood disadvantage on ASD likelihood in a large birth cohort with reduced chances of selection bias and measurement errors. (2) I examined two hypotheses on the susceptibility of ASD likelihood associated with prenatal air pollution. However, the interaction between PM2.5 and MIA, or the interaction between PM2.5 and neighborhood disadvantage were not significantly associated with ASD. In Chapter 3, I focused on MIA as an individual-level variable with the potential to modify the air pollution-ASD association. Should future studies identify individual-level health 163 conditions that influence susceptibility to ASD likelihood, personalized or precision medicine approaches could be developed to prevent the development of ASD in fetuses. In Chapter 4, I found that in a population with reduced disparity in the access to healthcare, living in neighborhoods with higher disadvantage levels were associated with ASD. This find indicated that investment in early intervention and family support for communities and populations with higher diagnosis rates, along with universal access to screening and diagnosis, is warranted. In Chapter 5, I aimed to answer an intervention-oriented question about how air pollution intervention policy would reduce the incidence of ASD and whether the effects would vary by neighborhood disadvantage levels. Research on this intervention-oriented question may provide direct support for policies or regulations aimed at allocating more resources to vulnerable communities, thereby reducing health inequities. While the empirical findings in Chapter 3 and Chapter 5 did not support the hypothesis that maternal immune activation (MIA) or neighborhood disadvantage may increase susceptibility to autism likelihood associated with prenatal environmental pollution, these hypotheses warrant further replication. However, the “multiple-exposure” hypothesis and the susceptibility model can also be applied to study the interaction between other environmental exposure (such as nutrition intake, health behaviors), or gene-environment interaction on ASD likelihood or other early childhood health issues. Combinations of different component factors are required for disease development (Rothman 1976). Thus, investigating the synergistic effects of co-existing may help shed lights on the complex etiology of ASD or other neurodevelopmental disorders. As outlined in Section 1.3, several challenges exist in examining the interactive association among co-existing factors on ASD likelihood. The first challenge lies in the 164 complexity of the conceptual model required to explore the “multiple-exposure” hypothesis. In this dissertation, I proposed a conceptual model of susceptibility along the pathway between prenatal environment pollution and health outcome in offspring (Figure 1.2). Based on this model, I hypothesized that population with lower SES may face the “double jeopardy” situation where they experience both higher exposure levels and increased vulnerability to pollution. This vulnerability may stem from various factors such as genetics, sex, maternal immune activation, maternal health behaviors, and nutritional intake(Yu et al. 2022). The three original research chapters (Chapter 3-5) provide empirical findings on the three potential mechanisms in this conceptual model. Second, large and high-quality population data are required to conduct analysis on the interaction among risk factors of ASD. In this dissertation, I applied a high quality intergenerational dataset of more than 300,000 mother-child pairs born in Southern California between 2001 and 2014. The substantial sample size ensures that adequate statistical power to detect the potential small interactive effects. Additionally, the retrospective cohort design also gives an opportunity to examine multiple exposures because the exposure contrast is not preassigned and no sampling was conducted based on a specific type of exposure. Therefore, I can examine three different exposures (prenatal PM2.5, maternal immune activation, and neighborhood disadvantage) and their interactions on ASD likelihood. This cohort also provides the longitudinal data to examine the temporal relationship between exposures and outcomes. Given that all the exposures of interest occurred prior to the diagnosis of ASD, which can provide valuable insights into causality and temporal sequence (Hill 1965). The third challenge involves integrating exposure data from diverse sources, which may differ in format and spatial resolution. In this dissertation, I used the EMR maintained by KPSC 165 to extract ASD diagnosis and conditions of MIA during pregnancy using ICD codes. The EMR not only contains high-quality data sources on individual health history, but also provides residential addresses, facilitating the linkage of individual-level data to environmental exposure models. In this dissertation, ambient PM2.5 exposures were represented as raster layers with 1 km spatial resolution estimated by a hybrid spatiotemporal model (Van Donkelaar et al. 2019), and the neighborhood SES measures (B.L. Levy, Owens, and Sampson 2019) were based on the census tracts. Both the raster layers and areal vector layers can be mapped to individual data using the geocoded residential address. Therefore, the data acquisition and management paradigm used in this dissertation can serve as a practical example for future study aiming at the synergistic effects of multiple environmental exposures from both the ecological scale to individual-level data. Last but not least, interdisciplinary knowledge and collaboration are necessary for conducting research with multiple exposures on health outcomes. I led the systematic review (Chapter 2) and three empirical studies (Chapter 3-5) with the assistance of experts from various domains including environmental exposure modeling, neurodevelopment psychiatry, statistics, environmental epidemiology, and neighborhood research. This interdisciplinary collaboration is essential for breaking down knowledge silos that have historically hindered the exploration of the "multiple-exposure" hypothesis in human health research. 6.3. Limitations 6.3.1. Lack of other potential key covariates I acknowledge the limitations of this dissertation. First, information on other potential key covariates including maternal nutrition intake during pregnancy and information regarding fathers’ sociodemographic characteristics and health behaviors were not available from EMR. 166 Nutritional intake during pregnancy has been associated with fetal neurodevelopment (Morse 2012). Given that there is limited research on the association between nutrition intake and air pollution, as well as the control for upstream SES factors in the statistical models, nutrition intake may not be a strong confounder. In Chapter 4 which focused on the main association between neighborhood disadvantage and ASD, the absence of paternal SES features may result in uncontrolled confounding by family-level SES. Adjusting for fathers' SES features could potentially confound the association between neighborhood disadvantage and ASD. In addition, I also lack data on meteorologic factors such as heat in this dissertation. Studies have shown that heat exposure during pregnancy has been associated with adverse birth outcomes including preterm birth, low birth weight, still birth (Bekkar et al. 2020), and congenital heart defects(Auger et al. 2017). However, few studies have examined the association between climate change and ASD. As ongoing investigations into the etiology of ASD continue, a clearer understanding of the environmental impacts on ASD will be achieved. 6.3.2. Lack of genetic data to study genetic susceptibility I did not have access to genetic data in this dissertation research given that it would be expensive to get genetic information in such a large birth cohort. Studies have shown a substantial heritability in ASD (Tick et al. 2016; Tchaconas and Adesman 2013). However, population studies of twins indicate less than 100% concordance, indicating that environmental factors are present alongside genetics. In addition, most twin pairs shared the same environment, so effects attributed to genetics may actually be caused by common environment. The interaction between genetics and environmental factors is increasingly recognized as critical in understanding the etiology of ASD. For other study cohort with available genetics information, 167 the external environmental data can be integrated to study genetic susceptibility to environmental pollution related ASD likelihood. 6.3.3. Measurement errors in environmental exposures I acknowledge the potential presence of measurement errors in the environmental exposures assessed in this dissertation. PM2.5 exposures extracted from the hybrid spatiotemporal model may contain uncertainty in their estimates in the ambient concentration (Van Donkelaar et al. 2019). Moreover, we did not account for activity patterns (the time spent at home, at work, and during transit) of the mothers during pregnancy. Therefore, the modeled ambient air pollution may be different from the actual individual exposures. However, these measurement errors are likely to be non-differential among children with and without ASD and may bias the associations between PM2.5 and ASD to null. For MIA conditions in Chapter 3, beyond clinical diagnoses of maternal immune activators, I lack biomarkers for MIA, which would be expensive to collect for a large cohort. Clinical diagnoses of MIA-related conditions can often indicate immune dysregulation in mothers (Mor and Cardenas 2010), which may influence fetal development (Simoes et al. 2018; Patterson 2011). Thus, utilizing diagnoses of clinical conditions as proxies of MIA in a large population can also provide insights into the potential etiology of ASD. Additionally, I lack information on the severity of MIA-related conditions and may misclassify MIA-related conditions if mothers did not come for care if they had mild symptoms. In Chapter 4 and Chapter 5 neighborhoods were defined as US census tracts considering the availability of required SES indicators measures to generate the disadvantage index for the time period spanning from 2001 to 2014. The concept of Uncertain Geographic Context Problem (UGCoP) highlights the issue that findings on associations between area-based attributes, such as 168 SES context, and individual outcomes could be influenced by how contextual units or neighborhoods are geographically delineated (Kwan 2012). It is possible that the association between SES at the census tract level and ASD may vary under other definitions of neighborhoods. Nonetheless, census tracts are typically designed to represent relatively homogeneous areas in terms of population size, allowing researchers to capture variations in neighborhood characteristics that may influence individual behaviors, health outcomes, or social disparities. The use of census tracts provides a practical and widely accepted method for studying the neighborhood context in research and policymaking endeavors. 6.3.4. Measurement errors in ASD diagnosis In this dissertation, the heath outcome of interest is the ASD diagnosis identified by ICD9 and ICD-10 codes from EMR. To address potential misclassification of ASD, the methodology for ascertaining ASD diagnoses applied in this study has been previously validated in the KPSC member population, with a positive predictive value of 87% (Coleman et al. 2015). However, in addition to ICD codes, I lack information on the detailed evaluation items of ASD and the severity. For all the empirical studies (Chapter 3-5), I examined ASD diagnosis before age 5 in the main analysis or as a sensitivity analysis. It is possible that the early diagnosis may capture the most obvious symptoms that presented at young age. Thus, the misclassification rate in the early ASD diagnosis should be even lower. 6.3.5. Generalizability This study may face generalizability issues. In this dissertation, I applied a birth cohort maintained by KPSC. A previous study has shown that KPSC membership reflects the diverse socioeconomic demographics in the study region (Koebnick et al. 2012). However, caution is warranted when generalizing findings from this dissertation to other populations from different 169 regions. In Chapter 4, I made the assumption that in this KPSC birth cohort the disparity in the access the neurodevelopmental disorder screening and diagnostic services were reduced. Although the KPSC membership covers both commercial insurance and Medi-Cal, the co-pays may still vary, differences that may influence access to care. Consequently, the replication of results in other healthcare settings is necessary, and their applicability to populations without health insurance may be limited. 6.4. Suggestions for Future Research In this dissertation, I put forward a conceptual model for susceptibility to examine the interactive associations between multiple co-existing factors in early life on ASD likelihood. I found that prenatal PM2.5, MIA, and neighborhood disadvantage were individually associated with ASD likelihood, but the interaction between MIA and prenatal PM2.5 or the interaction between neighborhood disadvantage and prenatal PM2.5 were not statistically significant. These findings warrant future replication studies across different populations and study regions. Beyond replications, I also foresee two potential applications of the susceptibility framework in future research. One is the investigation of the etiology of ASD and the other is the potential to identify interventions for other early childhood health outcomes. 6.4.1. Suggestions for research on the etiology of ASD to understand the susceptibility In addition to replication research, further investigation can focus on the sub-phenotypes of ASD. ASD is a set of neurodevelopmental disorders which encompasses a wide range of symptoms and characteristics, leading to variability in presentation and severity among individuals diagnosed with the disorder. In current literature, intellectual disability (ID) is often considered as a prominent comorbidity of ASD (Maenner et al. 2023). I did not include this information in this large sample derived from EMR due to some overlapping criteria between 170 ASD and ID and clinicians’ hesitance to label a child with ID (Thurm et al. 2019), making the electronic medical records a suboptimal source of ID information in cases of ASD. Future study may consider sub-phenotyping ASD by language ability, cognitive functioning sensory sensitivities, and comorbid conditions based on clinical descriptions. It is possible that different phenotypes may have varying etiologic pathways. Although I did not find empirical evidence on the interaction between MIA and prenatal PM2.5 or the interaction between neighborhood disadvantage and prenatal PM2.5, the susceptibility related to other potential risk factors such as genetics (Gaugler et al. 2014) and nutritional intake (Morse 2012) need future research. Genetic factors, for instance, may play a significant role in ASD susceptibility, albeit obtaining genetic data can be costly. Similarly, assessing nutrition intake during pregnancy often requires detailed survey data, adding another layer to research efforts. Leveraging existing cohorts focused on genetics and nutrition to assign environmental exposures, such as prenatal PM2.5 based on geocoded residential addresses, could offer a more efficient approach to studying their combined associations on ASD likelihood. By integrating data from various sources and employing the susceptibility framework, future studies can elucidate the intricate interplay between genetic predisposition, maternal health, nutrition intake, as well as, environmental exposures, and ASD likelihood, ultimately advancing our understanding of this complex neurodevelopmental disorder. In Chapter 3, I hypothesized that conditions of MIA may increase the fetus susceptibility of ASD likelihood associated with prenatal air pollution. This hypothesis finds support in animal studies where exposure to PM and MIA conditions has been linked to impaired social communication in offspring (Carlezon et al., 2019; Jones et al., 2020; Klocke et al., 2018; Klocke et al., 2017; Zhang et al., 2018), as well as elevated levels of oxidative stress and 171 inflammatory cytokines (Leni, Künzi, and Geiser 2020; S.E. Smith et al. 2007). Moving forward, future research may focus on the potential biomarkers of maternal immune disturbance to study the variance in the susceptibility due to immune functions during pregnancy and fetal neurodevelopment. The foundation of the "multiple-exposure" hypothesis and the susceptibility framework lies in the idea that co-existing factors may share similar or complementary biological pathways (Meldrum et al. 2013). To investigate the potential biological pathways of environmental impacts on ASD, researchers are increasingly designing and implementing innovative approaches such as multi-omics analysis (Hasin, Seldin, and Lusis 2017). This integrative approach combines data from various "omics", including genomics, transcriptomics, proteomics, metabolomics, and epigenomics, to comprehensively examine the molecular mechanisms underlying complex diseases. By employing multi-omics analyses, researchers can uncover molecular signatures associated with environmental exposures and ASD likelihood, providing valuable insights into the combined associations between genetic predisposition, environmental factors, and neurodevelopmental outcomes. 6.4.2. Suggestions for research on interventions and potential policy regulations The susceptibility framework presents a versatile tool not only for studying combined associations among co-existing risk factors with other early childhood health outcomes but also for informing precision health interventions for individuals and supporting community-based environmental regulations. The susceptibility associated with MIA, as explored in Chapter 3, may provide insights into precision medicine interventions for early childhood health concerns. Some clinical conditions during pregnancy are modifiable through medication and healthy behaviors 172 interventions. If future research could find evidence on how maternal immune dysregulation during pregnancy influences individual susceptibility to environmental factors, then personalized interventions or medication study can be designed to evaluate their efficacy in reducing the risk of early childhood health issues. Co-existing factors can manifest at individual levels, such as maternal immune activation (MIA), as well as at ecological or population levels, such as prenatal air pollution and neighborhood socioeconomic status (SES). In Chapter 5, I applied the G-computation to examine the potential effects of hypothetical air pollution interventions on ASD incidence. Future studies can examine more action-oriented hypothetical interventions alongside source-oriented air pollution exposure models. Examples of such interventions may include the reduction of on-road traffic-related air pollution or the regulation of emissions from heavy industrial factories. 173 References Adams, J. B., T. Audhya, E. Geis, E. Gehn, V. Fimbres, E. L. Pollard, J. Mitchell, J. Ingram, R. Hellmers, D. Laake, J. S. Matthews, K. Li, J. C. Naviaux, R. K. Naviaux, R. L. Adams, D. M. Coleman, and D. W. Quig. 2018. "Comprehensive Nutritional and Dietary Intervention for Autism Spectrum Disorder-A Randomized, Controlled 12-Month Trial." Nutrients 10 (3). https://doi.org/10.3390/nu10030369. Adams, K., D. S. Greenbaum, R. Shaikh, A. M. van Erp, and A. G. Russell. 2015. "Particulate matter components, sources, and health: Systematic approaches to testing effects." J Air Waste Manag Assoc 65 (5): 544-58. https://doi.org/10.1080/10962247.2014.1001884. https://www.ncbi.nlm.nih.gov/pubmed/25947313. Al-Hamdan, A. Z., P. P. Preetha, R. N. Albashaireh, M. Z. Al-Hamdan, and W. L. Crosson. 2018. "Investigating the effects of environmental factors on autism spectrum disorder in the USA using remotely sensed data." Environ Sci Pollut Res Int 25 (8): 7924-7936. https://doi.org/10.1007/s11356-017-1114-8. Altman, D. G., and J. M. Bland. 2003. "Interaction revisited: the difference between two estimates." BMJ 326 (7382): 219. https://doi.org/10.1136/bmj.326.7382.219. https://www.ncbi.nlm.nih.gov/pubmed/12543843. American Psychiatric Association. 2013a. Diagnostic and statistical manual of mental disorders. 5th ed.Am Psychiatric Assoc. ---. 2013b. Diagnostic and statistical manual of mental disorders (DSM-5®). American Psychiatric Pub. ArcGIS. 2021. "The geocoding process." Accessed Oct 28. https://desktop.arcgis.com/en/arcmap/latest/manage-data/geocoding/the-geocodingprocess.htm. Auger, N., W. D. Fraser, R. Sauve, M. Bilodeau-Bertrand, and T. Kosatsky. 2017. "Risk of Congenital Heart Defects after Ambient Heat Exposure Early in Pregnancy." Environ Health Perspect 125 (1): 8-14. https://doi.org/10.1289/EHP171. https://www.ncbi.nlm.nih.gov/pubmed/27494594. Auyeung, B., M. V. Lombardo, and S. Baron-Cohen. 2013. "Prenatal and postnatal hormone effects on the human brain and cognition." Pflugers Arch 465 (5): 557-71. https://doi.org/10.1007/s00424-013-1268-2. https://www.ncbi.nlm.nih.gov/pubmed/23588379. Bacon, E. C., E. Courchesne, C. C. Barnes, D. Cha, S. Pence, L. Schreibman, A. C. Stahmer, and K. Pierce. 2018. "Rethinking the idea of late autism spectrum disorder onset." Dev Psychopathol 30 (2): 553-569. https://doi.org/10.1017/S0954579417001067. https://www.ncbi.nlm.nih.gov/pubmed/28803559. Bai, D., B. H. K. Yip, G. C. Windham, A. Sourander, R. Francis, R. Yoffe, E. Glasson, B. Mahjani, A. Suominen, H. Leonard, M. Gissler, J. D. Buxbaum, K. Wong, D. Schendel, A. Kodesh, M. Breshnahan, S. Z. Levine, E. T. Parner, S. N. Hansen, C. Hultman, A. Reichenberg, and S. Sandin. 2019. "Association of Genetic and Environmental Factors With Autism in a 5-Country Cohort." JAMA Psychiatry 76 (10): 1035-1043. 174 https://doi.org/10.1001/jamapsychiatry.2019.1411. https://www.ncbi.nlm.nih.gov/pubmed/31314057. Baines, Kelly J, Dendra M Hillier, Faraj L Haddad, Nagalingam Rajakumar, Susanne Schmid, and Stephen J Renaud. 2020. "Maternal immune activation alters fetal brain development and enhances proliferation of neural precursor cells in rats." Frontiers in Immunology 11: 1145. Becerra, T. A., M. Wilhelm, J. Olsen, M. Cockburn, and B. Ritz. 2013. "Ambient air pollution and autism in Los Angeles county, California." Environ Health Perspect 121 (3): 380-6. https://doi.org/10.1289/ehp.1205827. Bekkar, B., S. Pacheco, R. Basu, and N. DeNicola. 2020. "Association of Air Pollution and Heat Exposure With Preterm Birth, Low Birth Weight, and Stillbirth in the US: A Systematic Review." JAMA Netw Open 3 (6): e208243. https://doi.org/10.1001/jamanetworkopen.2020.8243. https://www.ncbi.nlm.nih.gov/pubmed/32556259. Bhasin, T. K., and D. Schendel. 2007. "Sociodemographic risk factors for autism in a US metropolitan area." J Autism Dev Disord 37 (4): 667-77. https://doi.org/10.1007/s10803- 006-0194-y. https://www.ncbi.nlm.nih.gov/pubmed/16951989. Bignami, G., B. Musi, G. Dell'Omo, G. Laviola, and E. Alleva. 1994. "Limited effects of ozone exposure during pregnancy on physical and neurobehavioral development of CD-1 mice." Toxicol Appl Pharmacol 129 (2): 264-71. https://doi.org/10.1006/taap.1994.1251. https://www.ncbi.nlm.nih.gov/pubmed/7992316. Bilbo, S. D., C. L. Block, J. L. Bolton, R. Hanamsagar, and P. K. Tran. 2018. "Beyond infection - Maternal immune activation by environmental factors, microglial development, and relevance for autism spectrum disorders." Exp Neurol 299 (Pt A): 241-251. https://doi.org/10.1016/j.expneurol.2017.07.002. https://www.ncbi.nlm.nih.gov/pubmed/28698032. Bind, M. A. 2019. "Causal Modeling in Environmental Health." Annu Rev Public Health 40: 23- 43. https://doi.org/10.1146/annurev-publhealth-040218-044048. https://www.ncbi.nlm.nih.gov/pubmed/30633715. Blanchard, K. S., R. F. Palmer, and Z. Stein. 2011. "The value of ecologic studies: mercury concentration in ambient air and the risk of autism." Rev Environ Health 26 (2): 111-8. https://doi.org/10.1515/reveh.2011.015. Bland, J Martin, and Douglas G Altman. 1998. "Survival probabilities (the Kaplan-Meier method)." Bmj 317 (7172): 1572-1580. Block, M. L., and L. Calderon-Garciduenas. 2009. "Air pollution: mechanisms of neuroinflammation and CNS disease." Trends Neurosci 32 (9): 506-16. https://doi.org/10.1016/j.tins.2009.05.009. https://www.ncbi.nlm.nih.gov/pubmed/19716187. Block, M. L., A. Elder, R. L. Auten, S. D. Bilbo, H. Chen, J. C. Chen, D. A. Cory-Slechta, D. Costa, D. Diaz-Sanchez, D. C. Dorman, D. R. Gold, K. Gray, H. A. Jeng, J. D. Kaufman, M. T. Kleinman, A. Kirshner, C. Lawler, D. S. Miller, S. S. Nadadur, B. Ritz, E. O. Semmens, L. H. Tonelli, B. Veronesi, R. O. Wright, and R. J. Wright. 2012. "The 175 outdoor air pollution and brain health workshop." Neurotoxicology 33 (5): 972-84. https://doi.org/10.1016/j.neuro.2012.08.014. https://www.ncbi.nlm.nih.gov/pubmed/22981845. Blumberg, S. J., M. D. Bramlett, M. D. Kogan, L. A. Schieve, J. R. Jones, and M. C. Lu. 2013. "Changes in prevalence of parent-reported autism spectrum disorder in school-aged U.S. children: 2007 to 2011-2012." Natl Health Stat Report (65): 1-11, 1 p following 11. https://www.ncbi.nlm.nih.gov/pubmed/24988818. Bolton, J. L., S. Marinero, T. Hassanzadeh, D. Natesan, D. Le, C. Belliveau, S. N. Mason, R. L. Auten, and S. D. Bilbo. 2017. "Gestational Exposure to Air Pollution Alters Cortical Volume, Microglial Morphology, and Microglia-Neuron Interactions in a Sex-Specific Manner." Front Synaptic Neurosci 9: 10. https://doi.org/10.3389/fnsyn.2017.00010. https://www.ncbi.nlm.nih.gov/pubmed/28620294. Bonini, M. G., and R. M. Sargis. 2018. "Environmental Toxicant Exposures and Type 2 Diabetes Mellitus: Two Interrelated Public Health Problems on the Rise." Curr Opin Toxicol 7: 52-59. https://doi.org/10.1016/j.cotox.2017.09.003. https://www.ncbi.nlm.nih.gov/pubmed/29392186. Brown, D. M., M. R. Wilson, W. MacNee, V. Stone, and K. Donaldson. 2001. "Size-dependent proinflammatory effects of ultrafine polystyrene particles: a role for surface area and oxidative stress in the enhanced activity of ultrafines." Toxicol Appl Pharmacol 175 (3): 191-9. https://doi.org/10.1006/taap.2001.9240. https://www.ncbi.nlm.nih.gov/pubmed/11559017. Buescher, A. V., Z. Cidav, M. Knapp, and D. S. Mandell. 2014. "Costs of autism spectrum disorders in the United Kingdom and the United States." JAMA Pediatr 168 (8): 721-8. https://doi.org/10.1001/jamapediatrics.2014.210. https://www.ncbi.nlm.nih.gov/pubmed/24911948. Campbell, D. B., C. Li, J. S. Sutcliffe, A. M. Persico, and P. Levitt. 2008. "Genetic evidence implicating multiple genes in the MET receptor tyrosine kinase pathway in autism spectrum disorder." Autism Res 1 (3): 159-68. https://doi.org/10.1002/aur.27. https://www.ncbi.nlm.nih.gov/pubmed/19360663. Campbell, D. B., J. S. Sutcliffe, P. J. Ebert, R. Militerni, C. Bravaccio, S. Trillo, M. Elia, C. Schneider, R. Melmed, R. Sacco, A. M. Persico, and P. Levitt. 2006. "A genetic variant that disrupts MET transcription is associated with autism." Proceedings of the National Academy of Sciences 103 (45): 16834-9. https://doi.org/10.1073/pnas.0605296103. https://www.ncbi.nlm.nih.gov/pubmed/17053076. Canty, Angelo, and Brian Ripley. 2017. "Package ‘boot’." Bootstrap Functions. CRAN R Proj. Carter, S. A., M. M. Rahman, J. C. Lin, Y. H. Shu, T. Chow, X. Yu, M. P. Martinez, S. P. Eckel, J. C. Chen, Z. Chen, J. Schwartz, N. Pavlovic, F. W. Lurmann, R. McConnell, and A. H. Xiang. 2022. "In utero exposure to near-roadway air pollution and autism spectrum disorder in children." Environ Int 158: 106898. https://doi.org/10.1016/j.envint.2021.106898. https://www.ncbi.nlm.nih.gov/pubmed/34627014. 176 Centers for Disease Control and Prevention. 2007. "Prevalence of autism spectrum disorders-- autism and developmental disabilities monitoring network, 14 sites, United States, 2002." MMWR Surveill Summ 56 (1): 12-28. Charman, T., and G. Baird. 2002. "Practitioner review: Diagnosis of autism spectrum disorder in 2- and 3-year-old children." J Child Psychol Psychiatry 43 (3): 289-305. https://doi.org/10.1111/1469-7610.00022. https://www.ncbi.nlm.nih.gov/pubmed/11944873. Chaste, P., and M. Leboyer. 2012. "Autism risk factors: genes, environment, and geneenvironment interactions." Dialogues Clin Neurosci 14 (3): 281-92. https://www.ncbi.nlm.nih.gov/pubmed/23226953. Chatton, Arthur, Florent Le Borgne, Clémence Leyrat, and Yohann Foucher. 2022. "Gcomputation and doubly robust standardisation for continuous-time data: A comparison with inverse probability weighting." Statistical Methods in Medical Research 31 (4): 706- 718. Chen, L. C., and M. Lippmann. 2009. "Effects of metals within ambient air particulate matter (PM) on human health." Inhal Toxicol 21 (1): 1-31. https://doi.org/10.1080/08958370802105405. https://www.ncbi.nlm.nih.gov/pubmed/18803063. Cheng, J., B. Eskenazi, F. Widjaja, J. F. Cordero, and R. L. Hendren. 2019. "Improving autism perinatal risk factors: A systematic review." Med Hypotheses 127: 26-33. https://doi.org/10.1016/j.mehy.2019.03.012. https://www.ncbi.nlm.nih.gov/pubmed/31088644. Christensen, Grace M, Zhenjiang Li, John Pearce, Michele Marcus, James J Lah, Lance A Waller, Stefanie Ebelt, and Anke Hüls. 2022. "The complex relationship of air pollution and neighborhood socioeconomic status and their association with cognitive decline." Environment international 167: 107416. Chun, H., C. Leung, S. W. Wen, J. McDonald, and H. H. Shin. 2020. "Maternal exposure to air pollution and risk of autism in children: A systematic review and meta-analysis." Environ Pollut 256: 113307. https://doi.org/10.1016/j.envpol.2019.113307. https://www.ncbi.nlm.nih.gov/pubmed/31733973. Clifford, A., L. Lang, and R. Chen. 2012. "Effects of maternal cigarette smoking during pregnancy on cognitive parameters of children and young adults: a literature review." Neurotoxicol Teratol 34 (6): 560-70. https://doi.org/10.1016/j.ntt.2012.09.004. https://www.ncbi.nlm.nih.gov/pubmed/23022448. Cohen, Sheldon, William J Doyle, and Andrew Baum. 2006. "Socioeconomic status is associated with stress hormones." Psychosomatic medicine 68 (3): 414-420. Cohn, D'Vera, Gretchen Livingston, and Wendy Wang. 2014. After decades of decline, a rise in stay-at-home mothers. Pew Research Center, Social & Demographic Trends Project. Coleman, K. J., M. A. Lutsky, V. Yau, Y. Qian, M. E. Pomichowski, P. M. Crawford, F. L. Lynch, J. M. Madden, A. Owen-Smith, J. A. Pearson, K. A. Pearson, D. Rusinak, V. P. Quinn, and L. A. Croen. 2015. "Validation of Autism Spectrum Disorder Diagnoses in Large Healthcare Systems with Electronic Medical Records." J Autism Dev Disord 45 177 (7): 1989-96. https://doi.org/10.1007/s10803-015-2358-0. https://www.ncbi.nlm.nih.gov/pubmed/25641003. Colen, Cynthia G, David M Ramey, Elizabeth C Cooksey, and David R Williams. 2018. "Racial disparities in health among nonpoor African Americans and Hispanics: The role of acute and chronic discrimination." Social science & medicine 199: 167-180. Croen, L. A., J. K. Grether, C. K. Yoshida, R. Odouli, and J. Van de Water. 2005. "Maternal autoimmune diseases, asthma and allergies, and childhood autism spectrum disorders: a case-control study." Arch Pediatr Adolesc Med 159 (2): 151-7. https://doi.org/10.1001/archpedi.159.2.151. https://www.ncbi.nlm.nih.gov/pubmed/15699309. Croen, Lisa A, Judith K Grether, Cathleen K Yoshida, Roxana Odouli, and Victoria Hendrick. 2011. "Antidepressant use during pregnancy and childhood autism spectrum disorders." Archives of general psychiatry 68 (11): 1104-1112. Croft, D., R. Block, S. J. Cameron, K. Evans, C. J. Lowenstein, F. Ling, W. Zareba, P. K. Hopke, M. J. Utell, S. W. Thurston, K. Thevenet-Morrison, and D. Q. Rich. 2018. "Do elevated blood levels of omega-3 fatty acids modify effects of particulate air pollutants on fibrinogen?" Air Qual Atmos Health 11 (7): 791-799. https://doi.org/10.1007/s11869- 018-0586-0. Cuffe, J. S., Z. C. Xu, and A. V. Perkins. 2017. "Biomarkers of oxidative stress in pregnancy complications." Biomark Med 11 (3): 295-306. https://doi.org/10.2217/bmm-2016-0250. https://www.ncbi.nlm.nih.gov/pubmed/28157383. Dachew, B. A., A. Mamun, J. C. Maravilla, and R. Alati. 2018. "Pre-eclampsia and the risk of autism-spectrum disorder in offspring: meta-analysis." Br J Psychiatry 212 (3): 142-147. https://doi.org/10.1192/bjp.2017.27. https://www.ncbi.nlm.nih.gov/pubmed/29436313. Di Domenico, M., S. G. M. Benevenuto, P. P. Tomasini, V. Y. Yariwake, N. de Oliveira Alves, F. L. Rahmeier, M. da Cruz Fernandes, D. J. Moura, P. H. Nascimento Saldiva, and M. M. Veras. 2020. "Concentrated ambient fine particulate matter (PM2.5) exposure induce brain damage in pre and postnatal exposed mice." Neurotoxicology 79: 127-141. https://doi.org/10.1016/j.neuro.2020.05.004. https://www.ncbi.nlm.nih.gov/pubmed/32450181. Dickerson, A. S., M. H. Rahbar, A. V. Bakian, D. A. Bilder, R. A. Harrington, S. Pettygrove, R. S. Kirby, M. S. Durkin, I. Han, L. A. Moyé, 3rd, D. A. Pearson, M. S. Wingate, and W. M. Zahorodny. 2016. "Autism spectrum disorder prevalence and associations with air concentrations of lead, mercury, and arsenic." Environ Monit Assess 188 (7): 407. https://doi.org/10.1007/s10661-016-5405-1. Dubin, Robin A. 1998. "Spatial autocorrelation: a primer." Journal of housing economics 7 (4): 304-327. Durkin, M. S., M. J. Maenner, F. J. Meaney, S. E. Levy, C. DiGuiseppi, J. S. Nicholas, R. S. Kirby, J. A. Pinto-Martin, and L. A. Schieve. 2010. "Socioeconomic inequality in the prevalence of autism spectrum disorder: evidence from a U.S. cross-sectional study." PLoS One 5 (7): e11551. https://doi.org/10.1371/journal.pone.0011551. https://www.ncbi.nlm.nih.gov/pubmed/20634960. 178 Dutheil, F., A. Comptour, R. Morlon, M. Mermillod, B. Pereira, J. S. Baker, M. Charkhabi, M. Clinchamps, and N. Bourdel. 2021. "Autism spectrum disorder and air pollution: A systematic review and meta-analysis." Environ Pollut 278: 116856. https://doi.org/10.1016/j.envpol.2021.116856. Efron, Bradley, and Raoul LePage. 1992. Introduction to bootstrap. Wiley & Sons, New York. Estes, M. L., and A. K. McAllister. 2016. "Maternal immune activation: Implications for neuropsychiatric disorders." Science 353 (6301): 772-7. https://doi.org/10.1126/science.aag3194. https://www.ncbi.nlm.nih.gov/pubmed/27540164. Evans, G. W., and E. Kantrowitz. 2002. "Socioeconomic status and health: the potential role of environmental risk exposure." Annu Rev Public Health 23: 303-31. https://doi.org/10.1146/annurev.publhealth.23.112001.112349. https://www.ncbi.nlm.nih.gov/pubmed/11910065. Flores-Pajot, M. C., M. Ofner, M. T. Do, E. Lavigne, and P. J. Villeneuve. 2016. "Childhood autism spectrum disorders and exposure to nitrogen dioxide, and particulate matter air pollution: A review and meta-analysis." Environ Res 151: 763-776. https://doi.org/10.1016/j.envres.2016.07.030. https://www.ncbi.nlm.nih.gov/pubmed/27609410. Fong, K. C., Q. Di, I. Kloog, F. Laden, B. A. Coull, P. Koutrakis, and J. D. Schwartz. 2019. "Relative toxicities of major particulate matter constituents on birthweight in Massachusetts." Environ Epidemiol 3 (3): e047. https://doi.org/10.1097/EE9.0000000000000047. https://www.ncbi.nlm.nih.gov/pubmed/31342007. Fuller, C. H., K. R. Feeser, J. A. Sarnat, and M. S. O'Neill. 2017. "Air pollution, cardiovascular endpoints and susceptibility by stress and material resources: a systematic review of the evidence." Environ Health 16 (1): 58. https://doi.org/10.1186/s12940-017-0270-0. https://www.ncbi.nlm.nih.gov/pubmed/28615066. Garcia, E., R. Urman, K. Berhane, R. McConnell, and F. Gilliland. 2019. "Effects of policydriven hypothetical air pollutant interventions on childhood asthma incidence in southern California." Proc Natl Acad Sci U S A 116 (32): 15883-15888. https://doi.org/10.1073/pnas.1815678116. https://www.ncbi.nlm.nih.gov/pubmed/31332016. Gaugler, Trent, Lambertus Klei, Stephan J Sanders, Corneliu A Bodea, Arthur P Goldberg, Ann B Lee, Milind Mahajan, Dina Manaa, Yudi Pawitan, and Jennifer Reichert. 2014. "Most genetic risk for autism resides with common variation." Nature genetics 46 (8): 881-885. George, Barbara Jane, Bradley D Schultz, Ted Palma, Alan F Vette, Donald A Whitaker, and Ronald W Williams. 2011. "An evaluation of EPA’s National-Scale Air Toxics Assessment (NATA): comparison with benzene measurements in Detroit, Michigan." Atmospheric Environment 45 (19): 3301-3308. Glass, T. A., S. N. Goodman, M. A. Hernan, and J. M. Samet. 2013. "Causal inference in public health." Annu Rev Public Health 34: 61-75. https://doi.org/10.1146/annurev-publhealth031811-124606. https://www.ncbi.nlm.nih.gov/pubmed/23297653. 179 Gong, T., C. Almqvist, S. Bölte, P. Lichtenstein, H. Anckarsäter, T. Lind, C. Lundholm, and G. Pershagen. 2014. "Exposure to air pollution from traffic and neurodevelopmental disorders in Swedish twins." Twin Res Hum Genet 17 (6): 553-62. https://doi.org/10.1017/thg.2014.58. Gong, T., C. Dalman, S. Wicks, H. Dal, C. Magnusson, C. Lundholm, C. Almqvist, and G. Pershagen. 2017. "Perinatal Exposure to Traffic-Related Air Pollution and Autism Spectrum Disorders." Environ Health Perspect 125 (1): 119-126. https://doi.org/10.1289/ehp118. Goodrich, A. J., H. E. Volk, D. J. Tancredi, R. McConnell, F. W. Lurmann, R. L. Hansen, and R. J. Schmidt. 2018. "Joint effects of prenatal air pollutant exposure and maternal folic acid supplementation on risk of autism spectrum disorder." Autism Res 11 (1): 69-80. https://doi.org/10.1002/aur.1885. Gorrindo, P., C. J. Lane, E. B. Lee, B. McLaughlin, and P. Levitt. 2013. "Enrichment of elevated plasma F2t-isoprostane levels in individuals with autism who are stratified by presence of gastrointestinal dysfunction." PLoS One 8 (7): e68444. https://doi.org/10.1371/journal.pone.0068444. https://www.ncbi.nlm.nih.gov/pubmed/23844202. Greenland, S., J. Pearl, and J. M. Robins. 1999. "Causal diagrams for epidemiologic research." Epidemiology 10 (1): 37-48. https://www.ncbi.nlm.nih.gov/pubmed/9888278. Guxens, M., A. Ghassabian, T. Gong, R. Garcia-Esteban, D. Porta, L. Giorgis-Allemand, C. Almqvist, A. Aranbarri, R. Beelen, C. Badaloni, G. Cesaroni, A. de Nazelle, M. Estarlich, F. Forastiere, J. Forns, U. Gehring, J. Ibarluzea, V. W. Jaddoe, M. Korek, P. Lichtenstein, M. J. Nieuwenhuijsen, M. Rebagliato, R. Slama, H. Tiemeier, F. C. Verhulst, H. E. Volk, G. Pershagen, B. Brunekreef, and J. Sunyer. 2016. "Air Pollution Exposure during Pregnancy and Childhood Autistic Traits in Four European Population-Based Cohort Studies: The ESCAPE Project." Environ Health Perspect 124 (1): 133-40. https://doi.org/10.1289/ehp.1408483. Guyatt, G. H., A. D. Oxman, G. E. Vist, R. Kunz, Y. Falck-Ytter, P. Alonso-Coello, H. J. Schunemann, and Grade Working Group. 2008. "GRADE: an emerging consensus on rating quality of evidence and strength of recommendations." BMJ 336 (7650): 924-6. https://doi.org/10.1136/bmj.39489.470347.AD. https://www.ncbi.nlm.nih.gov/pubmed/18436948. Hajat, A., C. Hsia, and M. S. O'Neill. 2015. "Socioeconomic Disparities and Air Pollution Exposure: a Global Review." Curr Environ Health Rep 2 (4): 440-50. https://doi.org/10.1007/s40572-015-0069-5. https://www.ncbi.nlm.nih.gov/pubmed/26381684. Hammond, T. R., D. Robinton, and B. Stevens. 2018. "Microglia and the Brain: Complementary Partners in Development and Disease." Annu Rev Cell Dev Biol 34: 523-544. https://doi.org/10.1146/annurev-cellbio-100616-060509. https://www.ncbi.nlm.nih.gov/pubmed/30089221. Hasin, Yehudit, Marcus Seldin, and Aldons Lusis. 2017. "Multi-omics approaches to disease." Genome biology 18 (1): 1-15. 180 Heo, S., K. C. Fong, and M. L. Bell. 2019. "Risk of particulate matter on birth outcomes in relation to maternal socio-economic factors: a systematic review." Environ Res Lett 14 (12). https://doi.org/10.1088/1748-9326/ab4cd0. https://www.ncbi.nlm.nih.gov/pubmed/34108997. Hernán, Miguel A, and James M Robins. 2010. "Causal inference." In Causal Inference: What If, 161-170. CRC Boca Raton, FL. Herschkowitz, N. 1988. "Brain development in the fetus, neonate and infant." Biol Neonate 54 (1): 1-19. https://doi.org/10.1159/000242818. https://www.ncbi.nlm.nih.gov/pubmed/3061475. Hertz-Picciotto, I., and L. Delwiche. 2009. "The rise in autism and the role of age at diagnosis." Epidemiology 20 (1): 84-90. https://doi.org/10.1097/EDE.0b013e3181902d15. Hertz-Picciotto, I., R. J. Schmidt, and P. Krakowiak. 2018. "Understanding environmental contributions to autism: Causal concepts and the state of science." Autism Res 11 (4): 554-586. https://doi.org/10.1002/aur.1938. https://www.ncbi.nlm.nih.gov/pubmed/29573218. Hill, Austin Bradford. 1965. The environment and disease: association or causation? : Sage Publications. Hiscock, R., L. Bauld, A. Amos, J. A. Fidler, and M. Munafo. 2012. "Socioeconomic status and smoking: a review." Ann N Y Acad Sci 1248: 107-23. https://doi.org/10.1111/j.1749- 6632.2011.06202.x. https://www.ncbi.nlm.nih.gov/pubmed/22092035. Howard, Matt C. 2016. "A review of exploratory factor analysis decisions and overview of current practices: What we are doing and how can we improve?" International Journal of Human-Computer Interaction 32 (1): 51-62. Huguet, G., E. Ey, and T. Bourgeron. 2013. "The genetic landscapes of autism spectrum disorders." Annu Rev Genomics Hum Genet 14: 191-213. https://doi.org/10.1146/annurev-genom-091212-153431. https://www.ncbi.nlm.nih.gov/pubmed/23875794. Huizink, Anja C, and Eduard JH Mulder. 2006. "Maternal smoking, drinking or cannabis use during pregnancy and neurobehavioral and cognitive functioning in human offspring." Neuroscience & Biobehavioral Reviews 30 (1): 24-41. Ion, R., and A. L. Bernal. 2015. "Smoking and Preterm Birth." Reprod Sci 22 (8): 918-26. https://doi.org/10.1177/1933719114556486. https://www.ncbi.nlm.nih.gov/pubmed/25394641. James, S. J., P. Cutler, S. Melnyk, S. Jernigan, L. Janak, D. W. Gaylor, and J. A. Neubrander. 2004. "Metabolic biomarkers of increased oxidative stress and impaired methylation capacity in children with autism." Am J Clin Nutr 80 (6): 1611-7. https://doi.org/10.1093/ajcn/80.6.1611. https://www.ncbi.nlm.nih.gov/pubmed/15585776. James, S. J., S. Melnyk, S. Jernigan, M. A. Cleves, C. H. Halsted, D. H. Wong, P. Cutler, K. Bock, M. Boris, J. J. Bradstreet, S. M. Baker, and D. W. Gaylor. 2006. "Metabolic endophenotype and related genotypes are associated with oxidative stress in children with 181 autism." Am J Med Genet B Neuropsychiatr Genet 141B (8): 947-56. https://doi.org/10.1002/ajmg.b.30366. https://www.ncbi.nlm.nih.gov/pubmed/16917939. Jiang, H. Y., L. L. Xu, L. Shao, R. M. Xia, Z. H. Yu, Z. X. Ling, F. Yang, M. Deng, and B. Ruan. 2016. "Maternal infection during pregnancy and risk of autism spectrum disorders: A systematic review and meta-analysis." Brain Behav Immun 58: 165-172. https://doi.org/10.1016/j.bbi.2016.06.005. https://www.ncbi.nlm.nih.gov/pubmed/27287966. Jo, H., S. P. Eckel, J. C. Chen, M. Cockburn, M. P. Martinez, T. Chow, F. W. Lurmann, W. E. Funk, A. H. Xiang, and R. McConnell. 2019. "Gestational diabetes mellitus, prenatal air pollution exposure, and autism spectrum disorder." Environ Int 133 (Pt A): 105110. https://doi.org/10.1016/j.envint.2019.105110. Jo, H., S. P. Eckel, X. Wang, J. C. Chen, M. Cockburn, M. P. Martinez, T. Chow, N. Molshatzki, F. W. Lurmann, W. E. Funk, A. H. Xiang, and R. McConnell. 2019. "Sex-specific associations of autism spectrum disorder with residential air pollution exposure in a large Southern California pregnancy cohort." Environ Pollut 254 (Pt A): 113010. https://doi.org/10.1016/j.envpol.2019.113010. Josey, Kevin P, Scott W Delaney, Xiao Wu, Rachel C Nethery, Priyanka DeSouza, Danielle Braun, and Francesca Dominici. 2023. "Air Pollution and Mortality at the Intersection of Race and Social Class." New England Journal of Medicine 388 (15): 1396-1404. Kalkbrenner, A. E., J. L. Daniels, J. C. Chen, C. Poole, M. Emch, and J. Morrissey. 2010. "Perinatal exposure to hazardous air pollutants and autism spectrum disorders at age 8." Epidemiology 21 (5): 631-41. https://doi.org/10.1097/EDE.0b013e3181e65d76. Kalkbrenner, A. E., G. C. Windham, M. L. Serre, Y. Akita, X. Wang, K. Hoffman, B. P. Thayer, and J. L. Daniels. 2015. "Particulate matter exposure, prenatal and postnatal windows of susceptibility, and autism spectrum disorders." Epidemiology 26 (1): 30-42. https://doi.org/10.1097/ede.0000000000000173. Kalkbrenner, A. E., G. C. Windham, C. Zheng, R. McConnell, N. L. Lee, J. J. Schauer, B. Thayer, J. Pandey, and H. E. Volk. 2018. "Air Toxics in Relation to Autism Diagnosis, Phenotype, and Severity in a U.S. Family-Based Study." Environ Health Perspect 126 (3): 037004. https://doi.org/10.1289/ehp1867. Kannan, S., D. P. Misra, J. T. Dvonch, and A. Krishnakumar. 2006. "Exposures to airborne particulate matter and adverse perinatal outcomes: a biologically plausible mechanistic framework for exploring potential effect modification by nutrition." Environ Health Perspect 114 (11): 1636-42. https://doi.org/10.1289/ehp.9081. https://www.ncbi.nlm.nih.gov/pubmed/17107846. Kanner, L. 1968. "Autistic disturbances of affective contact." Acta Paedopsychiatr 35 (4): 100- 36. https://www.ncbi.nlm.nih.gov/pubmed/4880460. Kaufman, J. A., J. M. Wright, G. Rice, N. Connolly, K. Bowers, and J. Anixt. 2019. "Ambient ozone and fine particulate matter exposures and autism spectrum disorder in metropolitan Cincinnati, Ohio." Environ Res 171: 218-227. https://doi.org/10.1016/j.envres.2019.01.013. 182 Keil, A. P., J. P. Buckley, and A. E. Kalkbrenner. 2021. "Bayesian G-Computation for Estimating Impacts of Interventions on Exposure Mixtures: Demonstration With Metals From Coal-Fired Power Plants and Birth Weight." Am J Epidemiol 190 (12): 2647-2657. https://doi.org/10.1093/aje/kwab053. https://www.ncbi.nlm.nih.gov/pubmed/33751055. Kelly, Brian, Stefan Williams, Sylvie Collins, Faisal Mushtaq, Mark Mon-Williams, Barry Wright, Dan Mason, and John Wright. 2019. "The association between socioeconomic status and autism diagnosis in the United Kingdom for children aged 5–8 years of age: Findings from the Born in Bradford cohort." Autism 23 (1): 131-140. Kim, D., H. Volk, S. Girirajan, S. Pendergrass, M. A. Hall, S. S. Verma, R. J. Schmidt, R. L. Hansen, D. Ghosh, Y. Ludena-Rodriguez, K. Kim, M. D. Ritchie, I. Hertz-Picciotto, and S. B. Selleck. 2017. "The joint effect of air pollution exposure and copy number variation on risk for autism." Autism Res 10 (9): 1470-1480. https://doi.org/10.1002/aur.1799. King, M. D., and P. S. Bearman. 2011. "Socioeconomic Status and the Increased Prevalence of Autism in California." Am Sociol Rev 76 (2): 320-346. https://doi.org/10.1177/0003122411399389. https://www.ncbi.nlm.nih.gov/pubmed/21547238. Knol, M. J., and T. J. VanderWeele. 2012. "Recommendations for presenting analyses of effect modification and interaction." Int J Epidemiol 41 (2): 514-20. https://doi.org/10.1093/ije/dyr218. https://www.ncbi.nlm.nih.gov/pubmed/22253321. Koebnick, C., A. M. Langer-Gould, M. K. Gould, C. R. Chao, R. L. Iyer, N. Smith, W. Chen, and S. J. Jacobsen. 2012. "Sociodemographic characteristics of members of a large, integrated health care system: comparison with US Census Bureau data." Perm J 16 (3): 37-41. https://doi.org/10.7812/tpp/12-031. https://www.ncbi.nlm.nih.gov/pubmed/23012597. Kuhlthau, K., F. Orlich, T. A. Hall, D. Sikora, E. A. Kovacs, J. Delahaye, and T. E. Clemons. 2010. "Health-Related Quality of Life in children with autism spectrum disorders: results from the autism treatment network." J Autism Dev Disord 40 (6): 721-9. https://doi.org/10.1007/s10803-009-0921-2. https://www.ncbi.nlm.nih.gov/pubmed/20033762. Kushima, I., B. Aleksic, M. Nakatochi, T. Shimamura, T. Okada, Y. Uno, M. Morikawa, K. Ishizuka, T. Shiino, H. Kimura, Y. Arioka, A. Yoshimi, Y. Takasaki, Y. Yu, Y. Nakamura, M. Yamamoto, T. Iidaka, S. Iritani, T. Inada, N. Ogawa, E. Shishido, Y. Torii, N. Kawano, Y. Omura, T. Yoshikawa, T. Uchiyama, T. Yamamoto, M. Ikeda, R. Hashimoto, H. Yamamori, Y. Yasuda, T. Someya, Y. Watanabe, J. Egawa, A. Nunokawa, M. Itokawa, M. Arai, M. Miyashita, A. Kobori, M. Suzuki, T. Takahashi, M. Usami, M. Kodaira, K. Watanabe, T. Sasaki, H. Kuwabara, M. Tochigi, F. Nishimura, H. Yamasue, Y. Eriguchi, S. Benner, M. Kojima, W. Yassin, T. Munesue, S. Yokoyama, R. Kimura, Y. Funabiki, H. Kosaka, M. Ishitobi, T. Ohmori, S. Numata, T. Yoshikawa, T. Toyota, K. Yamakawa, T. Suzuki, Y. Inoue, K. Nakaoka, Y. I. Goto, M. Inagaki, N. Hashimoto, I. Kusumi, S. Son, T. Murai, T. Ikegame, N. Okada, K. Kasai, S. Kunimoto, D. Mori, N. Iwata, and N. Ozaki. 2018. "Comparative Analyses of Copy-Number Variation in Autism Spectrum Disorder and Schizophrenia Reveal Etiological Overlap and Biological Insights." Cell Rep 24 (11): 2838-2856. 183 https://doi.org/10.1016/j.celrep.2018.08.022. https://www.ncbi.nlm.nih.gov/pubmed/30208311. Kwan, Mei-Po. 2012. "The uncertain geographic context problem." Annals of the Association of American Geographers 102 (5): 958-968. Lam, J., P. Sutton, A. Kalkbrenner, G. Windham, A. Halladay, E. Koustas, C. Lawler, L. Davidson, N. Daniels, C. Newschaffer, and T. Woodruff. 2016. "A Systematic Review and Meta-Analysis of Multiple Airborne Pollutants and Autism Spectrum Disorder." PLoS One 11 (9): e0161851. https://doi.org/10.1371/journal.pone.0161851. https://www.ncbi.nlm.nih.gov/pubmed/27653281. Lehti, V. , S. Hinkka-Yli-Salomaki, K. Cheslack-Postava, M. Gissler, A. S. Brown, and A. Sourander. 2015. "Maternal socio-economic status based on occupation and autism spectrum disorders: a national case-control study." Nord J Psychiatry 69 (7): 523-30. https://doi.org/10.3109/08039488.2015.1011692. https://www.ncbi.nlm.nih.gov/pubmed/25732938. Leni, Zaira, Lisa Künzi, and Marianne Geiser. 2020. "Air pollution causing oxidative stress." Current opinion in toxicology 20: 1-8. Lenz, Kathryn M, and Margaret M McCarthy. 2015. "A starring role for microglia in brain sex differences." The Neuroscientist 21 (3): 306-321. Letellier, N., L. A. Gutierrez, J. Duchesne, C. Chen, S. Ilango, C. Helmer, C. Berr, M. Mortamais, and T. Benmarhnia. 2022. "Air quality improvement and incident dementia: Effects of observed and hypothetical reductions in air pollutant using parametric gcomputation." Alzheimers Dement 18 (12): 2509-2517. https://doi.org/10.1002/alz.12606. https://www.ncbi.nlm.nih.gov/pubmed/35142029. Levine, S. Z., A. Kodesh, A. Viktorin, L. Smith, R. Uher, A. Reichenberg, and S. Sandin. 2018. "Association of Maternal Use of Folic Acid and Multivitamin Supplements in the Periods Before and During Pregnancy With the Risk of Autism Spectrum Disorder in Offspring." JAMA Psychiatry 75 (2): 176-184. https://doi.org/10.1001/jamapsychiatry.2017.4050. https://www.ncbi.nlm.nih.gov/pubmed/29299606. Levy, Brian L, Ann Owens, and Robert J Sampson. 2019. "The varying effects of neighborhood disadvantage on college graduation: Moderating and mediating mechanisms." Sociology of Education 92 (3): 269-292. Levy, J. I., S. L. Greco, and J. D. Spengler. 2002. "The importance of population susceptibility for air pollution risk assessment: a case study of power plants near Washington, DC." Environ Health Perspect 110 (12): 1253-60. https://doi.org/10.1289/ehp.021101253. https://www.ncbi.nlm.nih.gov/pubmed/12460806. Lewandowski, T. A., S. M. Bartell, J. W. Yager, and L. Levin. 2009. "An evaluation of surrogate chemical exposure measures and autism prevalence in Texas." J Toxicol Environ Health A 72 (24): 1592-603. https://doi.org/10.1080/15287390903232483. Li, M., E. Francis, S. N. Hinkle, A. S. Ajjarapu, and C. Zhang. 2019. "Preconception and Prenatal Nutrition and Neurodevelopmental Disorders: A Systematic Review and MetaAnalysis." Nutrients 11 (7). https://doi.org/10.3390/nu11071628. 184 Li, X., C. Sjostedt, K. Sundquist, B. Zoller, and J. Sundquist. 2014. "Neighborhood deprivation and childhood autism: a nationwide study from Sweden." J Psychiatr Res 53: 187-92. https://doi.org/10.1016/j.jpsychires.2014.02.011. https://www.ncbi.nlm.nih.gov/pubmed/24613033. Lim, C. C., R. B. Hayes, J. Ahn, Y. Shao, D. T. Silverman, R. R. Jones, and G. D. Thurston. 2019. "Mediterranean Diet and the Association Between Air Pollution and Cardiovascular Disease Mortality Risk." Circulation 139 (15): 1766-1775. https://doi.org/10.1161/circulationaha.118.035742. Lin, V. W., A. A. Baccarelli, and H. H. Burris. 2016. "Epigenetics-a potential mediator between air pollution and preterm birth." Environ Epigenet 2 (1). https://doi.org/10.1093/eep/dvv008. Link, B. G., and J. Phelan. 1995. "Social conditions as fundamental causes of disease." J Health Soc Behav Spec No: 80-94. https://www.ncbi.nlm.nih.gov/pubmed/7560851. Logan, John R, Zengwang Xu, and Brian J Stults. 2014. "Interpolating US decennial census tract data from as early as 1970 to 2010: A longitudinal tract database." The Professional Geographer 66 (3): 412-420. Lyall, K., P. Ashwood, J. Van de Water, and I. Hertz-Picciotto. 2014. "Maternal immunemediated conditions, autism spectrum disorders, and developmental delay." J Autism Dev Disord 44 (7): 1546-55. https://doi.org/10.1007/s10803-013-2017-2. https://www.ncbi.nlm.nih.gov/pubmed/24337796. Lyall, K., R. J. Schmidt, and I. Hertz-Picciotto. 2014. "Maternal lifestyle and environmental risk factors for autism spectrum disorders." Int J Epidemiol 43 (2): 443-64. https://doi.org/10.1093/ije/dyt282. https://www.ncbi.nlm.nih.gov/pubmed/24518932. Lyall, Kristen, Kassandra L Munger, Éilis J O'Reilly, Susan L Santangelo, and Alberto Ascherio. 2013. "Maternal dietary fat intake in association with autism spectrum disorders." American journal of epidemiology 178 (2): 209-220. MacDonald, R., D. Parry-Cruwys, S. Dupere, and W. Ahearn. 2014. "Assessing progress and outcome of early intensive behavioral intervention for toddlers with autism." Res Dev Disabil 35 (12): 3632-44. https://doi.org/10.1016/j.ridd.2014.08.036. https://www.ncbi.nlm.nih.gov/pubmed/25241118. Maenner, Matthew J, Zachary Warren, Ashley Robinson Williams, Esther Amoakohene, Amanda V Bakian, Deborah A Bilder, Maureen S Durkin, Robert T Fitzgerald, Sarah M Furnier, and Michelle M Hughes. 2023. "Prevalence and characteristics of autism spectrum disorder among children aged 8 years—Autism and Developmental Disabilities Monitoring Network, 11 sites, United States, 2020." MMWR Surveillance Summaries 72 (2): 1. Maher, G. M., G. W. O'Keeffe, P. M. Kearney, L. C. Kenny, T. G. Dinan, M. Mattsson, and A. S. Khashan. 2018. "Association of Hypertensive Disorders of Pregnancy With Risk of Neurodevelopmental Disorders in Offspring: A Systematic Review and Meta-analysis." JAMA Psychiatry 75 (8): 809-819. https://doi.org/10.1001/jamapsychiatry.2018.0854. https://www.ncbi.nlm.nih.gov/pubmed/29874359. 185 Malkova, N. V., C. Z. Yu, E. Y. Hsiao, M. J. Moore, and P. H. Patterson. 2012. "Maternal immune activation yields offspring displaying mouse versions of the three core symptoms of autism." Brain Behav Immun 26 (4): 607-16. https://doi.org/10.1016/j.bbi.2012.01.011. https://www.ncbi.nlm.nih.gov/pubmed/22310922. Marsit, C. J. 2015. "Influence of environmental exposure on human epigenetic regulation." J Exp Biol 218 (Pt 1): 71-9. https://doi.org/10.1242/jeb.106971. https://www.ncbi.nlm.nih.gov/pubmed/25568453. Martinez, M. P., J. Lin, T. Chow, J. Chung, X. Wang, and A. H. Xiang. 2020. "Maternal Gestational Diabetes and Type 2 Diabetes During Pregnancy and Risk of Childhood Asthma in Offspring." J Pediatr 219: 173-179 e1. https://doi.org/10.1016/j.jpeds.2019.12.053. https://www.ncbi.nlm.nih.gov/pubmed/31987655. Mathiarasan, Sahana, and Anke Hüls. 2021. "Impact of Environmental Injustice on Children’s Health—Interaction between Air Pollution and Socioeconomic Status." International Journal of Environmental Research and Public Health 18 (2): 795. McCarthy, M. M. 2016. "Sex differences in the developing brain as a source of inherent risk." Dialogues Clin Neurosci 18 (4): 361-372. https://www.ncbi.nlm.nih.gov/pubmed/28179808. McGuinn, L. A., G. C. Windham, A. E. Kalkbrenner, C. Bradley, Q. Di, L. A. Croen, M. D. Fallin, K. Hoffman, C. Ladd-Acosta, J. Schwartz, A. G. Rappold, D. B. Richardson, L. M. Neas, M. D. Gammon, L. A. Schieve, and J. L. Daniels. 2020. "Early Life Exposure to Air Pollution and Autism Spectrum Disorder: Findings from a Multisite Case-Control Study." Epidemiology 31 (1): 103-114. https://doi.org/10.1097/ede.0000000000001109. McGuinn, L. A., G. C. Windham, L. C. Messer, Q. Di, J. Schwartz, L. A. Croen, E. J. Moody, A. G. Rappold, D. B. Richardson, L. M. Neas, M. D. Gammon, L. A. Schieve, and J. L. Daniels. 2019. "Air pollution, neighborhood deprivation, and autism spectrum disorder in the Study to Explore Early Development." Environ Epidemiol 3 (5). https://doi.org/10.1097/ee9.0000000000000067. McLaren, L. 2007. "Socioeconomic status and obesity." Epidemiol Rev 29: 29-48. https://doi.org/10.1093/epirev/mxm001. https://www.ncbi.nlm.nih.gov/pubmed/17478442. Meldrum, Suzanne Jacqueline, Tobias Strunk, Andrew Currie, Susan Lynne Prescott, Karen Simmer, and Andrew JO Whitehouse. 2013. "Autism spectrum disorder in children born preterm—role of exposure to perinatal inflammation." Frontiers in neuroscience 7: 123. Minh, Anita, Nazeem Muhajarine, Magdalena Janus, Marni Brownell, and Martin Guhn. 2017. "A review of neighborhood effects and early child development: How, where, and for whom, do neighborhoods matter?" Health & place 46: 155-174. Moher, D., A. Liberati, J. Tetzlaff, D. G. Altman, and Prisma Group. 2009. "Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement." BMJ 339: b2535. https://doi.org/10.1136/bmj.b2535. https://www.ncbi.nlm.nih.gov/pubmed/19622551. 186 Moore, K., R. Neugebauer, F. Lurmann, J. Hall, V. Brajer, S. Alcorn, and I. Tager. 2008. "Ambient ozone concentrations cause increased hospitalizations for asthma in children: an 18-year study in Southern California." Environ Health Perspect 116 (8): 1063-70. https://doi.org/10.1289/ehp.10497. https://www.ncbi.nlm.nih.gov/pubmed/18709165. Mor, G., and I. Cardenas. 2010. "The immune system in pregnancy: a unique complexity." Am J Reprod Immunol 63 (6): 425-33. https://doi.org/10.1111/j.1600-0897.2010.00836.x. https://www.ncbi.nlm.nih.gov/pubmed/20367629. Morello-Frosch, R., and E. D. Shenassa. 2006. "The environmental "riskscape" and social inequality: implications for explaining maternal and child health disparities." Environ Health Perspect 114 (8): 1150-3. https://doi.org/10.1289/ehp.8930. https://www.ncbi.nlm.nih.gov/pubmed/16882517. Morris, R. H., S. J. Counsell, I. M. McGonnell, and C. Thornton. 2021. "Early life exposure to air pollution impacts neuronal and glial cell function leading to impaired neurodevelopment." Bioessays: e2000288. https://doi.org/10.1002/bies.202000288. https://www.ncbi.nlm.nih.gov/pubmed/33751627. Morse, N. L. 2012. "Benefits of docosahexaenoic acid, folic acid, vitamin D and iodine on foetal and infant brain development and function following maternal supplementation during pregnancy and lactation." Nutrients 4 (7): 799-840. https://doi.org/10.3390/nu4070799. https://www.ncbi.nlm.nih.gov/pubmed/22852064. Naimi, Ashley I, Stephen R Cole, and Edward H Kennedy. 2017. "An introduction to g methods." International journal of epidemiology 46 (2): 756-762. Ncube, C. N., D. A. Enquobahrie, S. M. Albert, A. L. Herrick, and J. G. Burke. 2016. "Association of neighborhood context with offspring risk of preterm birth and low birthweight: A systematic review and meta-analysis of population-based studies." Soc Sci Med 153: 156-64. https://doi.org/10.1016/j.socscimed.2016.02.014. https://www.ncbi.nlm.nih.gov/pubmed/26900890. O’Connor, Thomas G, and Allison A Ciesla. 2022. "Maternal immune activation hypotheses for human neurodevelopment: Some outstanding questions." Biological Psychiatry: Cognitive Neuroscience and Neuroimaging 7 (5): 471-479. O'Neill, M. S., M. Jerrett, I. Kawachi, J. I. Levy, A. J. Cohen, N. Gouveia, P. Wilkinson, T. Fletcher, L. Cifuentes, J. Schwartz, Pollution Workshop on Air, and Conditions Socioeconomic. 2003. "Health, wealth, and air pollution: advancing theory and methods." Environ Health Perspect 111 (16): 1861-70. https://doi.org/10.1289/ehp.6334. https://www.ncbi.nlm.nih.gov/pubmed/14644658. O'Sharkey, Karl, Qi Meng, Sanjali Mitra, Seung-a Paik, Jonathan Liu, Jiaqi Shen, Laura K Thompson, Ting Chow, Jason Su, and Myles Cockburn. 2024. "Associations between brake and tire wear-related PM2.5 metal components, particulate oxidative stress potential, and autism spectrum disorder in Southern California." Environment International: 108573. Onore, C., M. Careaga, and P. Ashwood. 2012. "The role of immune dysfunction in the pathophysiology of autism." Brain Behav Immun 26 (3): 383-92. 187 https://doi.org/10.1016/j.bbi.2011.08.007. https://www.ncbi.nlm.nih.gov/pubmed/21906670. Ornoy, A., L. Weinstein-Fudim, and Z. Ergaz. 2015. "Prenatal factors associated with autism spectrum disorder (ASD)." Reprod Toxicol 56: 155-69. https://doi.org/10.1016/j.reprotox.2015.05.007. https://www.ncbi.nlm.nih.gov/pubmed/26021712. Pagalan, L., C. Bickford, W. Weikum, B. Lanphear, M. Brauer, N. Lanphear, G. E. Hanley, T. F. Oberlander, and M. Winters. 2019. "Association of Prenatal Exposure to Air Pollution With Autism Spectrum Disorder." JAMA Pediatr 173 (1): 86-92. https://doi.org/10.1001/jamapediatrics.2018.3101. Patterson, P. H. 2011. "Maternal infection and immune involvement in autism." Trends Mol Med 17 (7): 389-94. https://doi.org/10.1016/j.molmed.2011.03.001. https://www.ncbi.nlm.nih.gov/pubmed/21482187. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Racicot, K., J. Y. Kwon, P. Aldo, M. Silasi, and G. Mor. 2014. "Understanding the complexity of the immune system during pregnancy." Am J Reprod Immunol 72 (2): 107-16. https://doi.org/10.1111/aji.12289. https://www.ncbi.nlm.nih.gov/pubmed/24995526. Rahman, M. M., S. A. Carter, J. C. Lin, T. Chow, X. Yu, M. P. Martinez, Z. Chen, J. C. Chen, D. Rud, J. P. Lewinger, A. van Donkelaar, R. V. Martin, S. P. Eckel, J. Schwartz, F. Lurmann, M. J. Kleeman, R. McConnell, and A. H. Xiang. 2023. "Associations of Autism Spectrum Disorder with PM(2.5) Components: A Comparative Study Using Two Different Exposure Models." Environ Sci Technol 57 (1): 405-414. https://doi.org/10.1021/acs.est.2c05197. https://www.ncbi.nlm.nih.gov/pubmed/36548990. Rahman, M. M., S. A. Carter, J. C. Lin, T. Chow, X. Yu, M. P. Martinez, P. Levitt, Z. Chen, J. C. Chen, D. Rud, J. P. Lewinger, S. P. Eckel, J. Schwartz, F. W. Lurmann, M. J. Kleeman, R. McConnell, and A. H. Xiang. 2023. "Prenatal exposure to tailpipe and nontailpipe tracers of particulate matter pollution and autism spectrum disorders." Environ Int 171: 107736. https://doi.org/10.1016/j.envint.2023.107736. https://www.ncbi.nlm.nih.gov/pubmed/36623380. Rahman, M. M., Y. H. Shu, T. Chow, F. W. Lurmann, X. Yu, M. P. Martinez, S. A. Carter, S. P. Eckel, J. C. Chen, Z. Chen, P. Levitt, J. Schwartz, R. McConnell, and A. H. Xiang. 2022. "Prenatal Exposure to Air Pollution and Autism Spectrum Disorder: Sensitive Windows of Exposure and Sex Differences." Environ Health Perspect 130 (1): 17008. https://doi.org/10.1289/EHP9509. https://www.ncbi.nlm.nih.gov/pubmed/35040691. Rai, D., G. Lewis, M. Lundberg, R. Araya, A. Svensson, C. Dalman, P. Carpenter, and C. Magnusson. 2012. "Parental socioeconomic status and risk of offspring autism spectrum disorders in a Swedish population-based study." J Am Acad Child Adolesc Psychiatry 51 (5): 467-476.e6. https://doi.org/10.1016/j.jaac.2012.02.012. Rao, P. A., and D. C. Beidel. 2009. "The impact of children with high-functioning autism on parental stress, sibling adjustment, and family functioning." Behav Modif 33 (4): 437-51. 188 https://doi.org/10.1177/0145445509336427. https://www.ncbi.nlm.nih.gov/pubmed/19436073. Ravaccia, D., and T. Ghafourian. 2020. "Critical Role of the Maternal Immune System in the Pathogenesis of Autism Spectrum Disorder." Biomedicines 8 (12). https://doi.org/10.3390/biomedicines8120557. https://www.ncbi.nlm.nih.gov/pubmed/33271759. Raz, R., H. Levine, O. Pinto, D. M. Broday, Yuval, and M. G. Weisskopf. 2018. "Traffic-Related Air Pollution and Autism Spectrum Disorder: A Population-Based Nested Case-Control Study in Israel." Am J Epidemiol 187 (4): 717-725. https://doi.org/10.1093/aje/kwx294. Raz, R., A. L. Roberts, K. Lyall, J. E. Hart, A. C. Just, F. Laden, and M. G. Weisskopf. 2015. "Autism spectrum disorder and particulate matter air pollution before, during, and after pregnancy: a nested case-control analysis within the Nurses' Health Study II Cohort." Environ Health Perspect 123 (3): 264-70. https://doi.org/10.1289/ehp.1408133. Ritz, B., Z. Liew, Q. Yan, X. Cui, J. Virk, M. Ketzel, and O. Raaschou-Nielsen. 2018. "Air pollution and Autism in Denmark." Environmental Epidemiology (Philadelphia, Pa.) 2 (4): e028. https://doi.org/10.1097/ee9.0000000000000028. Robbins, J. M., V. Vaccarino, H. Zhang, and S. V. Kasl. 2005. "Socioeconomic status and diagnosed diabetes incidence." Diabetes Res Clin Pract 68 (3): 230-6. https://doi.org/10.1016/j.diabres.2004.09.007. https://www.ncbi.nlm.nih.gov/pubmed/15936465. Roberts, A. L., K. Lyall, J. E. Hart, F. Laden, A. C. Just, J. F. Bobb, K. C. Koenen, A. Ascherio, and M. G. Weisskopf. 2013. "Perinatal air pollutant exposures and autism spectrum disorder in the children of Nurses' Health Study II participants." Environ Health Perspect 121 (8): 978-84. https://doi.org/10.1289/ehp.1206187. Robins, James. 1986. "A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect." Mathematical modelling 7 (9-12): 1393-1512. Rosen, B. N., B. K. Lee, N. L. Lee, Y. Yang, and I. Burstyn. 2015. "Maternal Smoking and Autism Spectrum Disorder: A Meta-analysis." J Autism Dev Disord 45 (6): 1689-98. https://doi.org/10.1007/s10803-014-2327-z. https://www.ncbi.nlm.nih.gov/pubmed/25432101. Rothman, K. J. 1976. "Causes." Am J Epidemiol 104 (6): 587-92. https://doi.org/10.1093/oxfordjournals.aje.a112335. Rylaarsdam, L., and A. Guemez-Gamboa. 2019. "Genetic Causes and Modifiers of Autism Spectrum Disorder." Front Cell Neurosci 13: 385. https://doi.org/10.3389/fncel.2019.00385. https://www.ncbi.nlm.nih.gov/pubmed/31481879. Savitz, D. A., V. A. Danilack, S. M. Engel, B. Elston, and H. S. Lipkind. 2014. "Descriptive epidemiology of chronic hypertension, gestational hypertension, and preeclampsia in New York State, 1995-2004." Matern Child Health J 18 (4): 829-38. https://doi.org/10.1007/s10995-013-1307-9. https://www.ncbi.nlm.nih.gov/pubmed/23793484. 189 Schmidt, R. J. 2013. "Maternal folic acid supplements associated with reduced autism risk in the child." Evid Based Med 18 (6): e53. https://doi.org/10.1136/eb-2013-101311. https://www.ncbi.nlm.nih.gov/pubmed/23635843. Schmidt, R. J., A. M. Iosif, E. Guerrero Angel, and S. Ozonoff. 2019. "Association of Maternal Prenatal Vitamin Use With Risk for Autism Spectrum Disorder Recurrence in Young Siblings." JAMA Psychiatry 76 (4): 391-398. https://doi.org/10.1001/jamapsychiatry.2018.3901. Schoenfeld, David. 1982. "Partial residuals for the proportional hazards regression model." Biometrika 69 (1): 239-241. Schoots, M. H., S. J. Gordijn, S. A. Scherjon, H. van Goor, and J. L. Hillebrands. 2018. "Oxidative stress in placental pathology." Placenta 69: 153-161. https://doi.org/10.1016/j.placenta.2018.03.003. https://www.ncbi.nlm.nih.gov/pubmed/29622278. Schwartz, J., E. Austin, M. A. Bind, A. Zanobetti, and P. Koutrakis. 2015. "Estimating Causal Associations of Fine Particles With Daily Deaths in Boston." Am J Epidemiol 182 (7): 644-50. https://doi.org/10.1093/aje/kwv101. https://www.ncbi.nlm.nih.gov/pubmed/26346544. Sheng, L., X. Ding, M. Ferguson, M. McCallister, R. Rhoades, M. Maguire, A. Ramesh, M. Aschner, D. Campbell, P. Levitt, and D. B. Hood. 2010. "Prenatal polycyclic aromatic hydrocarbon exposure leads to behavioral deficits and downregulation of receptor tyrosine kinase, MET." Toxicol Sci 118 (2): 625-34. https://doi.org/10.1093/toxsci/kfq304. Simoes, L. R., G. Sangiogo, M. H. Tashiro, J. S. Generoso, C. J. Faller, D. Dominguini, G. A. Mastella, G. Scaini, V. V. Giridharan, M. Michels, D. Florentino, F. Petronilho, G. Z. Reus, F. Dal-Pizzol, A. I. Zugno, and T. Barichello. 2018. "Maternal immune activation induced by lipopolysaccharide triggers immune response in pregnant mother and fetus, and induces behavioral impairment in adult rats." J Psychiatr Res 100: 71-83. https://doi.org/10.1016/j.jpsychires.2018.02.007. https://www.ncbi.nlm.nih.gov/pubmed/29494891. Smith, Stephen EP, Jennifer Li, Krassimira Garbett, Karoly Mirnics, and Paul H Patterson. 2007. "Maternal immune activation alters fetal brain development through interleukin-6." Journal of Neuroscience 27 (40): 10695-10702. Smith, T. J. S., A. P. Keil, and J. P. Buckley. 2023. "Estimating Causal Effects of Interventions on Early-life Environmental Exposures Using Observational Data." Curr Environ Health Rep 10 (1): 12-21. https://doi.org/10.1007/s40572-022-00388-y. https://www.ncbi.nlm.nih.gov/pubmed/36418665. Sorace, A., L. de Acetis, E. Alleva, and D. Santucci. 2001. "Prolonged exposure to low doses of ozone: short- and long-term changes in behavioral performance in mice." Environ Res 85 (2): 122-34. https://doi.org/10.1006/enrs.2000.4097. https://www.ncbi.nlm.nih.gov/pubmed/11161662. 190 Stubbs, G, K Henley, and J Green. 2016. "Autism: Will vitamin D supplementation during pregnancy and early childhood reduce the recurrence rate of autism in newborn siblings?" Medical hypotheses 88: 74-78. Sullivan, Timothy J., Charles T. Driscoll, Colin M. Beier, Dallas Burtraw, Ivan J. Fernandez, James N. Galloway, David A. Gay, Christine L. Goodale, Gene E. Likens, Gary M. Lovett, and Shaun A. Watmough. 2018. "Air pollution success stories in the United States: The value of long-term observations." Environmental Science & Policy 84: 69-73. https://doi.org/https://doi.org/10.1016/j.envsci.2018.02.016. http://www.sciencedirect.com/science/article/pii/S1462901117312352. Talbott, E. O., V. C. Arena, J. R. Rager, J. E. Clougherty, D. R. Michanowicz, R. K. Sharma, and S. L. Stacy. 2015. "Fine particulate matter and the risk of autism spectrum disorder." Environ Res 140: 414-20. https://doi.org/10.1016/j.envres.2015.04.021. Talbott, E. O., L. P. Marshall, J. R. Rager, V. C. Arena, R. K. Sharma, and S. L. Stacy. 2015. "Air toxics and the risk of autism spectrum disorder: the results of a population based case-control study in southwestern Pennsylvania." Environ Health 14: 80. https://doi.org/10.1186/s12940-015-0064-1. Tchaconas, A., and A. Adesman. 2013. "Autism spectrum disorders: a pediatric overview and update." Curr Opin Pediatr 25 (1): 130-44. https://doi.org/10.1097/MOP.0b013e32835c2b70. https://www.ncbi.nlm.nih.gov/pubmed/23274432. Theoharides, T. C., I. Tsilioni, A. B. Patel, and R. Doyle. 2016. "Atopic diseases and inflammation of the brain in the pathogenesis of autism spectrum disorders." Transl Psychiatry 6 (6): e844. https://doi.org/10.1038/tp.2016.77. https://www.ncbi.nlm.nih.gov/pubmed/27351598. Thomas, Pauline, Walter Zahorodny, Bo Peng, Soyeon Kim, Nisha Jani, William Halperin, and Michael Brimacombe. 2012. "The association of autism diagnosis with socioeconomic status." Autism 16 (2): 201-213. Thurm, A., C. Farmer, E. Salzman, C. Lord, and S. Bishop. 2019. "State of the Field: Differentiating Intellectual Disability From Autism Spectrum Disorder." Front Psychiatry 10: 526. https://doi.org/10.3389/fpsyt.2019.00526. https://www.ncbi.nlm.nih.gov/pubmed/31417436. Tick, B., P. Bolton, F. Happe, M. Rutter, and F. Rijsdijk. 2016. "Heritability of autism spectrum disorders: a meta-analysis of twin studies." J Child Psychol Psychiatry 57 (5): 585-95. https://doi.org/10.1111/jcpp.12499. https://www.ncbi.nlm.nih.gov/pubmed/26709141. U.S. Environmental Protection Agency. 2024. "Final Rule to Strengthen the National Air Quality Health Standard for Particulate Matter ", 2024. https://www.epa.gov/system/files/documents/2024-02/pm-naaqs-overview.pdf. Urman, R., E. Garcia, K. Berhane, R. McConnell, W. J. Gauderman, and F. Gilliland. 2020. "The Potential Effects of Policy-driven Air Pollution Interventions on Childhood Lung Development." Am J Respir Crit Care Med 201 (4): 438-444. https://doi.org/10.1164/rccm.201903-0670OC. https://www.ncbi.nlm.nih.gov/pubmed/31644884. 191 Van Donkelaar, Aaron, Randall V Martin, Chi Li, and Richard T Burnett. 2019. "Regional estimates of chemical composition of fine particulate matter using a combined geoscience-statistical method with information from satellites, models, and monitors." Environmental science & technology 53 (5): 2595-2611. VanderWeele, Tyler J, and Mirjam J Knol. 2014. "A tutorial on interaction." Epidemiologic methods 3 (1): 33-72. Vargas, D. L., C. Nascimbene, C. Krishnan, A. W. Zimmerman, and C. A. Pardo. 2005. "Neuroglial activation and neuroinflammation in the brain of patients with autism." Ann Neurol 57 (1): 67-81. https://doi.org/10.1002/ana.20315. https://www.ncbi.nlm.nih.gov/pubmed/15546155. Voineagu, I., X. Wang, P. Johnston, J. K. Lowe, Y. Tian, S. Horvath, J. Mill, R. M. Cantor, B. J. Blencowe, and D. H. Geschwind. 2011. "Transcriptomic analysis of autistic brain reveals convergent molecular pathology." Nature 474 (7351): 380-4. https://doi.org/10.1038/nature10110. https://www.ncbi.nlm.nih.gov/pubmed/21614001. Volk, H. E., T. Kerin, F. Lurmann, I. Hertz-Picciotto, R. McConnell, and D. B. Campbell. 2014. "Autism spectrum disorder: interaction of air pollution with the MET receptor tyrosine kinase gene." Epidemiology 25 (1): 44-7. https://doi.org/10.1097/ede.0000000000000030. Volk, H. E., F. Lurmann, B. Penfold, I. Hertz-Picciotto, and R. McConnell. 2013. "Trafficrelated air pollution, particulate matter, and autism." JAMA Psychiatry 70 (1): 71-7. https://doi.org/10.1001/jamapsychiatry.2013.266. Volk, H. E., B. Park, C. Hollingue, K. L. Jones, P. Ashwood, G. C. Windham, F. Lurman, S. E. Alexeeff, M. Kharrazi, M. Pearl, J. Van de Water, and L. A. Croen. 2020. "Maternal immune response and air pollution exposure during pregnancy: insights from the Early Markers for Autism (EMA) study." J Neurodev Disord 12 (1): 42. https://doi.org/10.1186/s11689-020-09343-0. https://www.ncbi.nlm.nih.gov/pubmed/33327930. von Ehrenstein, O. S., H. Aralis, M. Cockburn, and B. Ritz. 2014. "In utero exposure to toxic air pollutants and risk of childhood autism." Epidemiology 25 (6): 851-8. https://doi.org/10.1097/ede.0000000000000150. von Ehrenstein, O. S., X. Cui, Q. Yan, H. Aralis, and B. Ritz. 2020. "Maternal Prenatal Smoking and Autism Spectrum Disorder in Offspring: a California Statewide Cohort and Sibling Study." Am J Epidemiol. https://doi.org/10.1093/aje/kwaa182. https://www.ncbi.nlm.nih.gov/pubmed/32830844. Walker, Renee E, Christopher R Keane, and Jessica G Burke. 2010. "Disparities and access to healthy food in the United States: A review of food deserts literature." Health & place 16 (5): 876-884. Wang, Y., I. Kloog, B. A. Coull, A. Kosheleva, A. Zanobetti, and J. D. Schwartz. 2016. "Estimating Causal Effects of Long-Term PM2.5 Exposure on Mortality in New Jersey." Environ Health Perspect 124 (8): 1182-8. https://doi.org/10.1289/ehp.1409671. https://www.ncbi.nlm.nih.gov/pubmed/27082965. 192 Wells, G.A. , B. Shea, D. O'Connell, J. Peterson, V. Welch, M. Losos, and P. Tugwell. 2014. "The Newcastle- Ottawa Scale (NOS) for Assessing the Quality of Nonrandomised Studies in Meta-Analyses.". Accessed 1 May. http://www.ohri.ca/programs/clinical_epidemiology/oxford.asp. Widjaja, D., M. Orini, E. Vlemincx, and S. Van Huffel. 2013. "Cardiorespiratory dynamic response to mental stress: a multivariate time-frequency analysis." Comput Math Methods Med 2013: 451857. https://doi.org/10.1155/2013/451857. https://www.ncbi.nlm.nih.gov/pubmed/24386006. Windham, G. C., L. Zhang, R. Gunier, L. A. Croen, and J. K. Grether. 2006. "Autism spectrum disorders in relation to distribution of hazardous air pollutants in the san francisco bay area." Environ Health Perspect 114 (9): 1438-44. https://doi.org/10.1289/ehp.9120. Winter, A. S., C. Fountain, K. Cheslack-Postava, and P. S. Bearman. 2020. "The social patterning of autism diagnoses reversed in California between 1992 and 2018." Proc Natl Acad Sci U S A 117 (48): 30295-30302. https://doi.org/10.1073/pnas.2015762117. https://www.ncbi.nlm.nih.gov/pubmed/33199592. Wisniowiecka-Kowalnik, B., and B. A. Nowakowska. 2019. "Genetics and epigenetics of autism spectrum disorder-current evidence in the field." J Appl Genet 60 (1): 37-47. https://doi.org/10.1007/s13353-018-00480-w. https://www.ncbi.nlm.nih.gov/pubmed/30627967. Wood, S. N. 2011. "Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models." Journal of the Royal Statistical Society Series B: Statistical Methodology 73 (1): 3-36. ---. 2013. "On p-values for smooth components of an extended generalized additive model." Biometrika 100 (1): 221-228. Xiang, A. H., X. Wang, M. P. Martinez, K. Page, T. A. Buchanan, and R. K. Feldman. 2018. "Maternal Type 1 Diabetes and Risk of Autism in Offspring." JAMA 320 (1): 89-91. https://doi.org/10.1001/jama.2018.7614. https://www.ncbi.nlm.nih.gov/pubmed/29936530. Xiang, A. H., X. Wang, M. P. Martinez, J. C. Walthall, E. S. Curry, K. Page, T. A. Buchanan, K. J. Coleman, and D. Getahun. 2015. "Association of maternal diabetes with autism in offspring." JAMA 313 (14): 1425-34. https://doi.org/10.1001/jama.2015.2707. https://www.ncbi.nlm.nih.gov/pubmed/25871668. Xu, X., F. Deng, X. Guo, P. Lv, M. Zhong, C. Liu, A. Wang, K. Tzan, S. Y. Jiang, M. Lippmann, S. Rajagopalan, Q. Qu, L. C. Chen, and Q. Sun. 2012. "Association of systemic inflammation with marked changes in particulate air pollution in Beijing in 2008." Toxicol Lett 212 (2): 147-56. https://doi.org/10.1016/j.toxlet.2012.05.014. https://www.ncbi.nlm.nih.gov/pubmed/22617750. Xue, Z., and C. Jia. 2019. "A Model-to-Monitor Evaluation of 2011 National-Scale Air Toxics Assessment (NATA)." Toxics 7 (1). https://doi.org/10.3390/toxics7010013. https://www.ncbi.nlm.nih.gov/pubmed/30857354. Yu, X., M. Mostafijur Rahman, S. A. Carter, J. C. Lin, Z. Zhuang, T. Chow, F. W. Lurmann, M. J. Kleeman, M. P. Martinez, A. van Donkelaar, R. V. Martin, S. P. Eckel, Z. Chen, P. 193 Levitt, J. Schwartz, D. Hackman, J. C. Chen, R. McConnell, and A. H. Xiang. 2023. "Prenatal air pollution, maternal immune activation, and autism spectrum disorder." Environ Int 179: 108148. https://doi.org/10.1016/j.envint.2023.108148. https://www.ncbi.nlm.nih.gov/pubmed/37595536. Yu, X., M. M. Rahman, S. A. Carter, J. C. Lin, T. Chow, F. W. Lurmann, J. C. Chen, M. P. Martinez, J. Schwartz, S. P. Eckel, Z. Chen, R. McConnell, A. H. Xiang, and D. A. Hackman. 2024. "Neighborhood Disadvantage and Autism Spectrum Disorder in a Population With Health Insurance." JAMA Psychiatry 81 (2): 209-213. https://doi.org/10.1001/jamapsychiatry.2023.4347. https://www.ncbi.nlm.nih.gov/pubmed/37966844. Yu, X., M. M. Rahman, Z. Wang, S. A. Carter, J. Schwartz, Z. Chen, S. P. Eckel, D. Hackman, J. C. Chen, A. H. Xiang, and R. McConnell. 2022. "Evidence of susceptibility to autism risks associated with early life ambient air pollution: A systematic review." Environ Res 208: 112590. https://doi.org/10.1016/j.envres.2021.112590. https://www.ncbi.nlm.nih.gov/pubmed/34929192. Yui, K., Y. Kawasaki, H. Yamada, and S. Ogawa. 2016. "Oxidative Stress and Nitric Oxide in Autism Spectrum Disorder and Other Neuropsychiatric Disorders." CNS Neurol Disord Drug Targets 15 (5): 587-96. https://doi.org/10.2174/1871527315666160413121751. https://www.ncbi.nlm.nih.gov/pubmed/27071787. Zerbo, O., Y. Qian, C. Yoshida, J. K. Grether, J. Van de Water, and L. A. Croen. 2013. "Maternal Infection During Pregnancy and Autism Spectrum Disorders." J Autism Dev Disord 45 (12): 4015-25. https://doi.org/10.1007/s10803-013-2016-3. https://www.ncbi.nlm.nih.gov/pubmed/24366406. Zhong, C., J. Tessing, B. K. Lee, and K. Lyall. 2020. "Maternal Dietary Factors and the Risk of Autism Spectrum Disorders: A Systematic Review of Existing Evidence." Autism Res 13 (10): 1634-1658. https://doi.org/10.1002/aur.2402. https://www.ncbi.nlm.nih.gov/pubmed/33015977.

Abstract (if available)

Abstract Autism spectrum disorder (ASD) is a complex set of developmental disorders characterized by deficits in social interactions and communication and the presence of restricted, repetitive, and stereotyped patterns of behaviors. The estimated prevalence of ASD in the United States increased from 0.66% in 2002 to 2.78% in 2020. The etiology of autism is unclear but likely to be multifactorial. While research has emphasized a strong genetic component in ASD, accumulative evidence suggests that environmental pollution, maternal health conditions during pregnancy, as well as social context or chronic social stress may also alter neurodevelopment. Challenges remain to explore the interactive or joint effects of these co-existing environmental factors. Given the potentially shared pathways such as oxidative stress and neuroinflammation among these co-occurring factors, the “multiple- exposure” hypothesis of ASD deserves further research.

In this dissertation, I conducted one systematic review and three quantitative epidemiological studies based on a large retrospective birth cohort born in Southern California Kaiser Permanente between 2001 and 2014. In the review article (Chapter 2), child’s sex, maternal nutrition or diabetes, socioeconomic factors, and child risk genotypes have been reported to modify the effect of early-life air pollutants on ASD likelihood in the epidemiologic literature. However, the sparsity of studies on corresponding modifying hypotheses precludes conclusive findings. In the first empirical study (Chapter 3), I hypothesized that co-exposure of prenatal PM2.5 or PM2.5 components and maternal immune activation (MIA)-related conditions may have interactive association on the subsequent likelihood of ASD. However, the empirical evidence did not support this hypothesis, as no statistically significant interactions between MIA conditions and prenatal PM2.5 exposure with ASD likelihood were found. In the second empirical study (Chapter 4), I found that children residing in more disadvantaged neighborhoods at birth had a higher likelihood of ASD diagnosis among a population with health insurance and thus relatively free of bias in ascertainment due to limited access to care. In the third study (Chapter 5), I applied an innovative window-based G-computation approach to quantify the changes in cumulative incidence of ASD under hypothetical PM2.5 interventions in the entire cohort and by neighborhood disadvantage deciles. I found that all communities would have similar benefit from hypothetical PM2.5 interventions regardless of neighborhood disadvantage levels.

These results warrant replication study on other population from different regions. Although the studies did not find interactive associations between prenatal air pollution and MIA or prenatal air pollution and neighborhood disadvantage on ASD likelihood, the “multiple-exposure” framework may nevertheless be useful for future research in understanding the complex interplay between other risk factors for autism, for example genetic–environmental interactions or nutrition-related susceptibility. Further investigation is needed to identify the potential biological pathways for the co-existing factors on ASD likelihood and to provide more plausible hypotheses on the interactive associations.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Early life air pollution exposure, gestational diabetes mellitus, and autism spectrum disorder

PDF

Examining exposure to extreme heat and air pollution and its effects on all-cause, cardiovascular, and respiratory mortality in California: effect modification by the social deprivation index

PDF

Personal PM2.5 exposure during pregnancy in an environmental health disparities population

PDF

Associations between ambient air pollution and hypertensive disorders of pregnancy

PDF

Evaluation of new methods for estimating exposure to traffic-related pollution and early health effects for large population epidemiological studies

PDF

Associations of ambient air pollution exposures with perceived stress in the MADRES cohort

PDF

Effects of near-roadway air pollution exposure on obesity, obesity-related behavior, and neurobehavioral deficits during peripuberty

PDF

Disparities in exposure to traffic-related pollution sources by self-identified and ancestral Hispanic descent in participants of the USC Children’s Health Study

PDF

Air pollution, smoking, and multigenerational DNA methylation Signatures: a study of two southern California cohorts

PDF

Prenatal air pollution exposure, newborn DNA methylation, and childhood respiratory health

PDF

Prenatal environmental exposures and fetal growth in the MADRES cohort

PDF

The impacts of the COVID-19 pandemic on therapy utilization among racially/ethnically and socio-economically diverse children with autism spectrum disorder

PDF

Linking air pollution to integrative gene and metabolites networks in young adult with asthma

PDF

Investigating daily effects of activity space-based built environment exposures on physical activity behaviors in Hispanic women during pregnancy and early postpartum

PDF

Impact of urban source emissions on ambient particulate matter (PM) composition and associated toxicity in local and regional scales using source appointment models

PDF

Assessment of the mortality burden associated with ambient air pollution in rural and urban areas of India

PDF

Developmental trajectories of sensory patterns in young children with and without autism spectrum disorder: a longitudinal population-based study from infancy to school age

PDF

The environmental and genetic determinants of cleft lip and palate in the global setting

PDF

Behavioral and neural influences of interoception and alexithymia on emotional empathy in autism spectrum disorder

PDF

Air pollution, mitochondrial function, and growth in children

Asset Metadata

Creator Yu, Xin (author)

Core Title Susceptibility to autism likelihood associated with prenatal ambient air pollution exposures: identifying vulnerable populations

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Health and Place,Population

Degree Conferral Date 2024-05

Publication Date 04/01/2024

Defense Date 03/20/2024

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag Air pollution,autism,interaction,maternal immune activation,neighborhood,OAI-PMH Harvest

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Hackman, Daniel A. (committee chair), McConnell, Rob Scot (committee chair), Ruddell, Darren M. (committee member), Tarbox, Jonathan (committee member), Xiang, Anny Hui (committee member)

Creator Email lindayu950408@gmail.com,xyu059@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113865296

Unique identifier UC113865296

Identifier etd-YuXin-12746.pdf (filename)

Legacy Identifier etd-YuXin-12746

Document Type Dissertation

Format theses (aat)

Rights Yu, Xin

Internet Media Type application/pdf

Type texts

Source 20240403-usctheses-batch-1134 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu