Autism spectrum disorder (ASD) is a developmental disability characterized by the occurrence of persistent impairments in social interaction and the presence of restricted, repetitive patterns of behaviors, interests, or activities [1]. The Autism and Developmental Disabilities Monitoring (ADDM) Network is a population-based surveillance system that estimates the prevalence of ASD and other developmental disabilities among children in selected communities throughout the U.S. [2]. ADDM is modeled after the work of the Metropolitan Atlanta Developmental Disabilities Surveillance Program (MADDSP), which has tracked ASD in selected counties within metropolitan Atlanta, Georgia since 1996 [3]. The most recent estimate of ASD prevalence among 8-year-old children in MADDSP was 21.8/1000 (1 in 46 8-year-old children) in 2018, similar to the 23.0/1000 (1 in 44) prevalence seen among all ADDM Network sites in 2018 [4]. Approximately a third of children with ASD in ADDM also had intellectual disability (ID).

Using single census variable measures of socioeconomic status (SES) (i.e., family poverty, college education, and median household income), ADDM Network reports based on data from 2002 to 2010 showed a higher prevalence of ASD in areas of higher SES [5–7]. While this association has been consistent over time for MADDSP, recently there has been an inverse association (lower prevalence in areas of higher SES) among other ADDM Network sites [4]. The Centers for Disease Control and Prevention developed the Social Vulnerability Index (SVI) to help public health officials and emergency response planners meet the needs of socially vulnerable populations in emergency response and recovery efforts [8]. Use of the SVI has been expanded beyond disaster relief efforts and has been applied to analyses of public health surveillance data by examining associations with COVID-19 vaccination rates [9] and stay-at-home orders [10], and emergency department visits for firearm injuries [11]. The SVI includes 15 census-based indicators, grouped into four themes (SES, household composition and disability, racial and ethnic minority status and language, and housing type and transportation) and an overall composite measure.

We applied SVI indicators to autism surveillance data from MADDSP to evaluate the application of a comprehensive SVI measure to describe patterns of ASD prevalence across communities and assess whether the prevalence of autism and co-occurring ID vary by SVI measure overall, by sex, and by race or ethnicity.

MADDSP conducts surveillance for 8-year-old children with ASD every 2 years. To identify children, staff request records from educational sources for children who receive special education services and from medical sources that evaluate and treat children with developmental disabilities based on select International Classification of Diseases, Ninth Revision (ICD-9) or International Classification of Diseases, Tenth Revision (ICD-10) codes for developmental disabilities. MADDSP staff then review records and abstract information for children who reside in the surveillance area. MADDSP conducts its work as a designated agent of the Georgia Department of Public Health and through memorandums of understanding with education sources.

Our primary analysis focused on data collected in Georgia’s DeKalb and Gwinnett Counties for 2018. Children were included in the analysis if they were 8 years old (born in 2010), lived in the surveillance area for at least 1 day during the 2018 surveillance year and had an address that could be geocoded to a census tract, and met the ASD case definition. Children met ASD case status if they had documentation in health, service, or education records that they had ever received any of the following: 1) a written statement from a qualified professional diagnosing autism, 2) a special education classification of autism (either primary exceptionality of ASD or an evaluation concluding criteria for autism eligibility was met) in public school, or 3) an ASD ICD code (ICD-9 codes between 299.00 and 299.99 or ICD-10 codes in the F84 range except for F84.2) obtained from administrative or billing information. A child was classified as having ID based on scores of ≤70 on their most recent test of cognitive ability or a statement from a qualified professional about a child’s cognitive ability in a developmental evaluation [12]. We also abstracted sex and race or ethnicity for each child. We excluded 56/514 (10%) of ASD cases that were included in the 2018 MADDSP surveillance area based on county of residence but could not be linked to the census tract.

We obtained SVI variables for 2018 by census tract for DeKalb and Gwinnett counties from the Centers for Disease Control and Prevention’s publicly available SVI data [13]. We estimated the population of 8-year-old children by census tract using the American Community Survey’s 5–9-year-old population estimates [14] and dividing by 5. We linked SVI data to 8-year-old ASD cases and population estimates by census tract. Next, we grouped census tracts into tertiles, classified as areas of high, medium, and low social vulnerability for each SVI measure (overall, 4 themes, and 15 individual indicators). We refer to the SVI measures as follows: “Overall” (overall SVI measure), “SES” (theme one: socioeconomic status), “Household” (theme two: household composition and disability), “Minority” (theme three: racial or ethnic minority status and language), and “Transportation” (theme four: housing type and transportation). We summed ASD cases and population estimates by these areas and calculated prevalence per 1000 (i.e., [number of 8-year-old children with ASD/8-year-old population] × 1000) and 95% confidence intervals (95% CI) overall, by sex, and by race or ethnicity. We calculated prevalence ratios (PR) to compare the prevalence between low, medium, and high SVI tertiles; differences were considered statistically significant if CIs did not include 1.00. We also calculated the prevalence for each census tract, grouped prevalence into high, medium, and low SVI areas, and mapped census tracts using ArcGIS Pro [Environmental Systems Research Institute, Inc., Redlands, CA]. Analyses were conducted using SAS v9.4 [SAS Institute Inc., Cary, NC]. Due to small sample sizes, we do not present prevalence among Asian children within each theme.

To examine differences over time, we conducted a secondary analysis focused on the association of ASD prevalence with SVI overall and for the four themes for 2014 and 2016. This analysis used the same methods and surveillance area as described previously; however, the case definition for ASD in 2014 involved an additional step of clinician review. A detailed description of methods in classifying ASD cases and how they changed between years has been previously described [15].

We identified 458 8-year-old children in 2018 who met the case definition for ASD. Eighty-three percent of the children were male; 43% of children were Black, 28% White, 14% Hispanic, 7% Asian, and 8% other or unknown race or ethnicity (Table A1). The population of 8-year-old children in the surveillance area was 23,136 yielding an overall ASD prevalence of 19.8 (95% CI 18.1–21.7) per 1000 children.

ASD prevalence varied across SVI themes (Table 1). Prevalence was higher in areas of low compared to high vulnerability for the SES (PR = 1.3 [95% CI 1.03–1.66]) and Transportation (PR = 1.6 [95% CI 1.24–1.98]) themes, and higher in areas of medium compared to high vulnerability for all themes. We further explored prevalence by sex and race or ethnicity (Table 1). Prevalence was higher among males than females across all tertiles for all themes. Associations between low versus high and medium versus high vulnerability among males were generally the same as those seen overall. Females had lower prevalence in areas of low compared to high vulnerability for the Minority theme (PR = 0.5 [95% CI 0.31–0.97]). Black children had higher ASD prevalence than Hispanic and White children across all tertiles for all themes (Table 1). Among Black children, the prevalence was significantly higher in areas of low compared to high vulnerability for the SES (PR = 1.5 [95% CI 1.03–2.31]) and Household (PR = 1.6 [95% CI 1.06–2.30]) themes, and in areas of medium compared to high vulnerability for the SES (PR = 1.6 [95% CI 1.13–2.13]), Household (PR = 1.5 [95% CI 1.07–2.01]), and Transportation (PR = 1.5 [95% CI 1.03–2.07]) themes. Among White children, the prevalence was higher in areas of low compared to high vulnerability for the Transportation theme (PR = 2.1 [95% CI 1.26–3.64]), and in areas of medium compared to high vulnerability for the Minority theme (PR = 1.8 [95% CI 1.15–2.98]). Among Hispanic children, the prevalence was significantly higher in areas of low compared to high vulnerability only for the SES theme (PR = 2.0 [95% CI 1.09–3.84]).

We also examined prevalence among the 352 children with information available on ID (Table 2). For children with ASD and no ID, we saw higher prevalence in areas with low compared to high vulnerability overall (PR = 1.8 [1.26–2.53]), and for the SES (PR = 2.15 [95% CI 1.47–3.10]), and Transportation themes (PR = 2.3 [1.61–3.36]), and in areas of medium compared to high vulnerability overall (PR = 1.7 [95% CI 1.22–2.47]), and for the SES (PR = 2.0 [95% CI 1.37–2.86]), Household (PR = 1.7 [95% CI 1.22–2.34]) and Transportation themes (PR = 1.9 [95% CI 1.28–2.78]). For children with both ASD and ID, we saw higher prevalence in areas of low compared to high vulnerability for the Transportation theme (PR = 1.5 [95% CI 1.02–2.28]).

Figure 1 illustrates prevalence by tertile for each individual SVI indicator. ASD prevalence was higher in areas with low compared to high vulnerability for the SVI indicator of persons with incomes below the poverty threshold (PR = 1.4 [95% CI 1.10–1.74]), not having a vehicle (PR = 1.5 [95% CI 1.17–1.87]) and living in multi-unit structures (PR = 1.5 [95% CI 1.20–1.91]). Prevalence was higher in areas of medium compared to high vulnerability for all SES and Minority theme variables, civilian with a disability (PR = 1.3 [95% CI 1.03–1.66]), and not having a vehicle (PR = 1.4 [95% CI 1.06–1.75]).

We visualized the spatial distribution of social vulnerability and ASD prevalence in our surveillance area (Fig. 2). We saw higher ASD prevalence throughout most of Gwinnett County with a mix of high, medium, and lower vulnerability. In DeKalb County, we saw lower prevalence and high vulnerability in the south and low vulnerability with a range of prevalence in the northern part of the county.

Finally, we compared 2018 ASD prevalence by tertile of SVI theme to prevalence in 2014 and 2016 (Table 3). While case counts and prevalence were similar between years, we saw differences in prevalence between tertiles by year. ASD prevalence was significantly higher in areas of low compared to high vulnerability for the SES (PR = 1.3 [95% CI 1.03–1.66]) and Transportation themes (PR = 1.6 [95% CI 1.24–1.98]) in 2018, and higher in areas of medium compared to high vulnerability for all themes in 2018 and the SES (PR = 1.6 [95% CI 1.28–2.02]) and Household themes (PR = 1.3 [95% CI 1.04–1.61]) in 2016. We saw no significant differences between areas for any theme in 2014.

Disparities in the prevalence of ASD by race or ethnicity and SES have been frequently observed. These longstanding disparities appear to be changing, and more comprehensive measures can help understand these changes. The SVI includes more encompassing measures than the SES indicators used in previous ADDM analyses. Using these SVI indicators, we found that ASD prevalence was generally higher in areas of low or medium vulnerability compared to areas of high vulnerability. However, the opposite pattern (higher prevalence in areas of high vulnerability) was seen among females and Black children for the Minority theme. Spatially, we saw variable patterns across our surveillance area reinforcing the importance of examining these associations at the smallest geographical area possible to better understand local-level community differences.

While it may be possible that children in highly vulnerable communities are less affected by ASD, it is more likely that families with higher social vulnerability lack the resources needed to access developmental screenings which could result in an ASD diagnosis. Poverty, unemployment, per capita income, and living in multi-unit structures may indicate fewer financial resources, including health insurance, which can impact access to care. Limited English proficiency or not having a high school diploma may make understanding the processes necessary to access developmental evaluations or awareness of ASD symptoms more challenging, particularly when symptoms are mild. Increased access to resources that educate parents on child development, such as Learn the Signs. Act Early. [16], may help to fill this gap. Finally, not owning a vehicle or being a civilian with a disability may speak to physical barriers to the ability to access needed services.

The higher ASD prevalence observed among females and Black children in areas of high vulnerability for the Minority theme might indicate a process toward better identification of ASD; however, given long-standing disparities in identification among these groups [17,18], more remains to be done. Associations observed between SVI and ASD prevalence in 2016 and 2018, but not in 2014 could indicate shifts in the residency of children who are the most socially vulnerable or a widening of disparities in access to screening and diagnoses. Applying SVI to future MADDSP data as well as other communities within ADDM could help to further explain variations between years.

Using a composite measure (the SVI) to assess associations between ASD prevalence and social vulnerability has several advantages. In previous ADDM network analyses, the association of SES and ASD prevalence was examined using individual census variables [4,6]. These variables were linked to ASD case data by census tract and tracts were ranked within and between ADDM sites. This method allowed for comparisons between ADDM sites but could not be applied to communities outside of the network. Because SVI measures are ranked nationally, applying SVI to ADDM surveillance allows for broader comparisons between communities and potentially also other diseases or conditions. In addition, the SVI expands beyond SES to examine a broader set of social factors that may play a role in explaining disparities in the identification of ASD within a community. While our findings related to ASD prevalence and SES are consistent with previous ADDM Network findings for Georgia [4,6], associations with the other SVI themes (Transportation, Minority, Housing) have not been previously described.

Our results are subject to some limitations. Data collected relies on the availability and completeness of existing records to ascertain ASD cases. Educational data are only collected from public schools; information for private or homeschooled students is not included. Sample sizes for some groups (females, Asian, and Hispanic children) were small, thus limiting our ability to detect statistical differences between tertiles. Records for 106 (23%) of children with ASD were missing information on cognitive ability; characteristics of these children may differ from children without this information. Our analysis is focused on separate groups of 8-year-old children each year, so we cannot describe trends over time. Our surveillance area is composed of one part of one metropolitan area and findings may differ when applied to other communities. Lastly, features associated with a census tract may not accurately represent the experiences of everyone living within that tract.

Linking SVI information to public health surveillance data can extend understanding of the associations between social determinants of health and other diseases or conditions. Experience gained from linking MADDSP data to SVI indicators will be applied to inform analyses in other ADDM sites. Results from this study can be used to improve understanding of inequities among children with ASD in racial and ethnic minority groups or those living in low-resource settings. Enhanced resources within these communities can lead to earlier identification and provision of services to improve outcomes.