darwin-lifestyle

Methods

Surveillance Sites and Procedures

For 2020, the ADDM Network included 11 sites (Arizona, Arkansas, California, Georgia, Maryland, Minnesota, Missouri, New Jersey, Tennessee, Utah, and Wisconsin) that monitored ASD prevalence. Each site selected a geographic area of its state to monitor ASD among children aged 8 years (Table 1). Children included in this report were born in 2012 and lived in surveillance areas of the 11 sites during 2020. Sites were competitively funded and functioned as public health authorities under the Health Insurance Portability and Accountability Act of 1996 Privacy Rule and met applicable local institutional review board, privacy, and confidentiality requirements under 45 CFR 46 (14).

Case Ascertainment and Surveillance Case Definition

The ADDM Network conducts active surveillance of ASD by using multiple sources of information within a community (Table 1). The methods for collecting information and the case definition were unchanged from the 2018 surveillance year (12) and were modeled after those developed by CDC’s Metropolitan Atlanta Developmental Disabilities Surveillance Program (3). Sites request records from community medical, education, and service providers containing specific billing codes from the International Classification of Diseases, Ninth Revision (ICD-9) or International Classification of Diseases, Tenth Revision (ICD-10) or special education classification. The protocol allowed each site to select the ICD codes that necessitate record review if those codes closely aligned with program-recommended ICD codes (11). All ADDM Network sites used records from medical service providers that evaluated children with developmental disabilities and, for the first time, all sites had at least partial access to public school education records (Table 1). ADDM Network sites received information (including demographic data and ICD codes or special education classifications) for children with one or more of the requested codes or classifications, and ADDM staff manually reviewed the contents of associated (electronic and paper-based) records when available. If any part of the child’s record contained information meeting the case definition, ADDM staff abstracted information from the child’s developmental evaluations, special education plans, and other documents (e.g., cognitive or IQ tests) from all data sources. At certain sites, full record review could not be completed for all records because of the COVID-19 pandemic or other restrictions on physically accessing the location where records were stored (Table 1).

Children met the ASD case definition if they were aged 8 years in 2020 (born in 2012), lived in the surveillance area for at least 1 day during 2020, and had documentation in their records that they ever received 1) a written ASD diagnostic statement from a qualified professional, 2) a special education classification of autism (either primary exceptionality of ASD or an evaluation reporting criterion 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, Rett syndrome) obtained from administrative or billing information. Five children had an ICD code for Rett syndrome (F84.2) and no other indicators of ASD and did not meet the ASD case definition. ASD-related diagnostic conclusions (including suspected ASD or ruled out ASD) were collected verbatim from evaluations and were reviewed and classified by ADDM Network staff with clinical expertise at each site.

Additional Data Sources and Variable Definitions

Population denominators were obtained from the U.S. Census vintage 2021 county-level single-year-of-age postcensal population estimates for 2020 (15). In this report, the Asian and Native Hawaiian or other Pacific Islander categories were combined into Asian or Pacific Islander because current systems often combine these categories or are not explicit whether “Asian” at a given data source includes “Native Hawaiian or other Pacific Islander.” Population denominators include categories for American Indian or Alaska Native (AI/AN), Asian or Pacific Islander (A/PI), Black, White, two or more races, and Hispanic ethnicity. In previous ADDM Network reports, the denominator data were based on the National Center for Health Statistics postcensal bridged race estimates (also produced by the Census Bureau) (16); the bridged race data set did not include a category for two or more races, which increased counts in the other categories.

Surveillance areas at three sites (Arizona, California, and Minnesota) comprised subcounty school districts. For these sites, county population estimates were adjusted using the National Center for Education Statistics public school enrollment counts and the American Community Survey tract-level ages 5–9 years population estimates described previously (12). The primary race and ethnicity and sex information came from medical or education records and, when missing, was augmented by birth certificate linkages (among children born in the state of their residence at age 8 years), administrative, or billing information. Children with missing or unknown race or ethnicity information were excluded from race- and ethnicity-specific prevalence estimates.

Census tract-level median household income (MHI) was measured using the 2020 American Community Survey 5-year estimates (17). Population counts of children aged 8 years were estimated by dividing the number of children aged 5–9 years by five for each census tract. The tracts included in the surveillance areas were classified into three approximately equal-sized population groups (i.e., tertiles) of low, medium, and high MHI by using data from all sites. Children meeting the ADDM Network case definition for ASD were geocoded and assigned to a socioeconomic status (SES) group corresponding to their 2020 address. Census tract information was available for 96.0% of children; the remainder could not be linked to a census tract but had service receipt or school attendance that indicated study area residence.

A child was classified as having intellectual disability if they had an IQ score ≤70 on their most recent cognitive test or intellectual disability was indicated in a statement in a developmental evaluation from a qualified professional. Children were classified in the borderline range for IQ if the score on their most recent test was 71–85, and in the average or higher range with most recent IQ score >85 or with a statement their IQ was in the average range without a specific score. Age at first developmental evaluation was limited to children with information on the earliest collected or historically reported evaluation. Age at first ASD diagnosis was based on the earliest documented age when a qualified professional diagnosed ASD.

Analytic Methods

Prevalence was calculated as the number of children with ASD per 1,000 children in the defined population or group. Overall ASD prevalence estimates included all children with ASD from all 11 sites. Prevalence also was stratified by sex and by race and ethnicity using both the U.S. Census postcensal population estimates as well as the National Center for Health Statistics postcensal bridged race denominators. The Wilson score method was used to calculate 95% CIs. Pearson chi-square tests were used to compare proportions, and the Mantel-Haenszel (Woolf) test of homogeneity compared prevalence ratios across sites. Permutation tests were conducted to test differences in medians. Cochran Armitage tests were used to detect trends in prevalence across SES tertiles. Prevalence estimates with a relative SE >30% (and ratios calculated from those estimates) were considered to have limited statistical precision and were suppressed. Statistical tests with p values <0.05 and prevalence ratio 95% CIs that excluded 1.0 were considered statistically significant. R software (version 4.2; R Foundation) and additional packages were used to conduct analyses (12).

Results

ASD Prevalence

The overall ASD prevalence per 1,000 children aged 8 years was 27.6 (one in 36) and ranged from 23.1 in Maryland to 44.9 in California (Table 2). The overall male-to-female prevalence ratio was 3.8, with overall ASD prevalence of 43.0 among boys and 11.4 among girls. The same sites conducted ASD surveillance in 2018 and reported a combined prevalence of 23.0; however, certain sites changed their geographic areas or access to data sources for the current reporting period (Supplementary Table 1, https://stacks.cdc.gov/view/cdc/124397). The two sites with the largest relative changes (Missouri [48.5%] and Wisconsin [49.5%]) from 2018 to 2020 had increased access to education records in 2020 but no change in the geographic areas.

Overall, ASD prevalence per 1,000 children aged 8 years differed by racial and ethnic groups (Table 3); prevalence among White children (24.3) was lower than prevalence among Black, Hispanic, or A/PI children (29.3, 31.6, and 33.4, respectively). Among AI/AN children, ASD prevalence was 26.5 overall and was similar to other groups, but estimates met the 30% relative SE threshold for statistical precision in just one site (Arizona). ASD prevalence among children of two or more races was 22.9, which was not different than among White children but was lower than prevalence among AP/I, Black, and Hispanic children. Missouri was the only site in which White children had higher ASD prevalence than another racial or ethnic group (White compared with two or more races). Additional prevalence ratios comparing racial and ethnic groups are available (Supplementary Table 2, https://stacks.cdc.gov/view/cdc/124397). Prevalence calculations using the bridged-race denominator racial and ethnic categories used in previous reports (Supplementary Table 3, https://stacks.cdc.gov/view/cdc/124397) yielded similar findings of lower ASD prevalence among White children compared with that among Asian, Black, and Hispanic children.

In eight sites, ASD prevalence was not associated with census tract-level MHI, but in three sites (Arizona, New Jersey, and Utah), lower ASD prevalence was observed among children living in census tracts with higher MHI (Figure 1). When all sites were combined, prevalence of ASD was lower among census tracts with higher MHI; however, ASD prevalences for the low, medium, and high SES tertiles were all between 23.0–27.2.

ASD Identification

The percentage of children with diagnostic statements, special education classifications, and ICD codes varied by site (Table 4). Across sites, the percentage of children with ASD who had a documented ASD diagnostic statement was 74.7% overall (range = 60.9% in Wisconsin to 94.7% in New Jersey). ASD prevalence per 1,000 children aged 8 years based exclusively on documented ASD diagnostic statements was 20.6 overall (range = 17.1 in Wisconsin to 35.4 in California) (Figure 2). The overall percentage of children with ASD who had a documented ASD special education classification was 65.2% (range = 44.9% in Utah to 84.9% in Minnesota) (Table 4). The percentage of children with ASD who had a documented ICD code was 71.6% (range = 51.9% in Minnesota to 82.7% in California). A majority of (74.2%) children with ASD had at least two of the three types of ASD identification documented in their records (i.e., ASD diagnostic statement, special education classification, and ASD ICD code) and 37.4% had all three (Figure 3). A majority of children with an ICD code (89.5% of 4,472 children) also had a documented ASD diagnostic statement or ASD special education classification; among all children with ASD, few (7.5% of 6,245 children) met the case definition through having an ICD code only. A majority of children with documents indicating an ASD diagnosis or ASD special education classification had these mentioned multiple times in their records (overall median number of diagnoses documented: two; range: one in Tennessee to six in New Jersey; overall median special education classifications documented: four, site-specific medians ranging from two in Wisconsin and Tennessee to six in California) (Supplementary Table 4, https://stacks.cdc.gov/view/cdc/124397).

Among children with ASD, 37.4% ever had an evaluation report noting that ASD was suspected but not confirmed (Table 4). Overall, 11.6% of children with ASD had an ASD diagnosis or special education eligibility ruled out (range = 4.3% in Georgia to 29.3% in California). For a majority of children, ASD was confirmed after ASD had previously been ruled out; however, 3.9% (range = 0.2% in New Jersey to 12.8% in California) of all children with ASD had an evaluation ruling out ASD more recently than one confirming ASD.

Cognitive Ability Among Children with ASD

Data on cognitive ability were available for 4,165 (66.7%) children aged 8 years with ASD (range: 39.7% in Wisconsin to 91.2% in Arkansas) (Table 5). Among children with data on cognitive ability, the median age of the most recent cognitive test or examiner impression was 67 months (interquartile range: 51–81 months) (Supplementary Table 5, https://stacks.cdc.gov/view/cdc/124397). Girls with ASD were less likely than boys with ASD to have data on cognitive ability (64.4% versus 67.3%). Similar percentages of Black and White children had data on cognitive ability (66.8% and 65.0%, respectively), but Hispanic children (68.8%) were more likely to have cognitive data than White children. AI/AN (79.3%) and A/PI (71.2%) children and those of two or more races (73.9%) all had cognitive data at least as often as the other groups.

Among children aged 8 years with ASD who had data on cognitive ability, 37.9% were classified as having intellectual disability at their most recent test or examination, 23.5% were classified in the borderline range (IQ 71–85), and 38.6% were classified in the average or higher range (IQ >85) (Table 5). The percentage of children classified as having intellectual disability varied widely among sites (range = 21.7% in California to 51.0% in Tennessee). The median age of most recent test also varied by site (range = 55 months in Wisconsin to 79 months in Arizona) (Supplementary Table 5, https://stacks.cdc.gov/view/cdc/124397). Overall, girls with ASD were more likely to be classified as having an intellectual disability than boys with ASD (42.1% versus 36.9%), and Black children were more likely than Hispanic and White children to be classified as having intellectual disability (50.8%, 34.9%, and 31.8%, respectively). The percentage of children with ASD and intellectual disability among A/PI, two or more races, or AI/AN children was 41.3%, 37.8%, and 34.8%, respectively.

Age at First Evaluation and ASD Diagnosis

Among 5,744 children aged 8 years with ASD and recorded evaluations, 49% were evaluated by age 36 months (range = 38.5% in Utah to 59.5% in Maryland) (Table 6). The median age at first recorded evaluation ranged from 32 months in California to 44 months in Utah. Children with ASD with an intellectual disability were more likely to be evaluated by age 36 months compared with children with ASD without an intellectual disability (61.8% versus 46.0%).

Among the 4,663 children aged 8 years with ASD who had an evaluation containing an ASD diagnostic statement, the median age at earliest known diagnosis was 49 months (range = 36 months in California to 59 months in Minnesota) (Table 7). Children with ASD and intellectual disability had a lower median age at diagnosis (43 months) than children without an intellectual disability (53 months). When special education classifications of autism were considered with ASD diagnoses for earliest identification, 5,579 children with ASD were identified with a median age of 52 months (range = 39 months in California and New Jersey to 60 months in Arizona).

Discussion

For 2020, the prevalence estimate of ASD per 1,000 children aged 8 years was 27.6 (range: 23.1 in Maryland to 44.9 in California), which is higher than previous estimates from the ADDM Network. For the first time, the overall ASD prevalence for girls was >1% (11.4); in contrast, the prevalence among boys had already been noted to be higher (11.5) in the first ADDM Network report in 2002 (4). The continued variability in prevalence across ADDM sites, as well as the shifting in differences between demographic groups, highlight an ongoing need to better understand the systems and practices that contribute to this variability.

In its earliest years, the ADDM Network consistently reported lower overall ASD prevalence among Black and Hispanic versus White children aged 8 years. The White-Black gap in ASD prevalence narrowed in 2014, and there was no overall difference in ASD prevalence in 2016 or 2018 (Supplementary Figure 1, https://stacks.cdc.gov/view/cdc/124397). ASD prevalence among Asian, Black, and Hispanic children was at least 30% higher in 2020 than 2018, and ASD prevalence among White children was 14.6% higher than in 2018. Although this was the first time the ADDM Network reported lower ASD prevalence among White children than among other groups for children aged 8 years, a similar pattern was observed among children aged 4 years in 2018 (18). In addition, similar patterns were reported in analyses of national special education data and of California Developmental Services data, illustrating the prevalence of ASD classifications among Black and Hispanic children catching up and eclipsing that of White children over time (19,20). These patterns might reflect improved screening, awareness, and access to services among historically underserved groups. ASD prevalence in 2020 also was associated with lower SES, the opposite of what was observed previously (13), further supporting progress in identifying children regardless of race and ethnic group. As evidence grows of increased access to identification, attention might shift to what factors, such as social determinants of health, could lead to higher rates of disability among certain populations.

Even with higher ASD prevalence among Black compared with White children, Black children with ASD remained more likely to have co-occurring intellectual disability than White children, a finding that has been observed over multiple ADDM Network surveillance reports and among Black compared with White children without ASD in the United States (21). If Black children with ASD have less access to services than White children with ASD, as has been previously reported, the disproportionality in co-occurring intellectual disability might indicate an underascertainment of ASD among Black children without intellectual disability. Continued monitoring of trends is warranted, and it might be appropriate to re-examine potential risk or protective factors that were previously studied when the demographic composition of ASD was different.

During this period of changing demographic differences in ASD prevalence, the ADDM Network implemented two methodological changes. First, a new ASD case definition was adopted for the 2018 surveillance year. The previous case definition relied on reviewing written descriptions of ASD symptoms that were documented in comprehensive developmental evaluations. It could classify children without any formal ASD identification as ASD cases and could exclude children who had ASD diagnosed but lacked sufficient corroborating details in their records. An analysis found that non-White children were more likely to have incomplete records, which could lead to underascertainment of ASD compared with White children (22). However, a retroactive application of the current case definition to the 2014 and 2016 surveillance years indicates similar prevalence ratios by race and ethnicity as the previous case definition (23). The second change, implemented in 2020, is using population denominators with standardized racial and ethnic categories. The most important difference from the previous (bridged-race) denominators is the inclusion of a category for two or more races, which reduces the size of the denominators among the other racial groups. Nevertheless, prevalence estimates based on the previous bridged-race denominators produced a similar pattern of lower ASD prevalence among White children compared with the other groups (Supplementary Table 3, https://stacks.cdc.gov/view/cdc/124397). Thus, there were qualitatively similar patterns when consistent case definitions and denominator data sets were applied during 2014–2020.

Although ASD can be identified by age 1 year in certain cases (24,25), as described in this report, a majority of children aged 8 years living in ADDM communities were not identified until they were several years older. The reported median age of identification has not changed much over the years of ADDM Network surveillance, but it does not necessarily indicate a lack of progress in community early identification efforts. In a recent analysis of ADDM Network data during 2002–2016, the median age of diagnosis might mask progress in early detection if more children are identified (i.e., prevalence increases) at all ages and does not include children who might have ASD diagnosed after age 8 years (26,27). Therefore, the ADDM Network now reports the cumulative incidence of ASD by age 48 months as a measure of early identification and compares children aged 4 years and 8 years living in the same communities as a measure of progress (28,29). The 2020 report on early identification of ASD found more children were identified at early ages than in the past, but many are still not identified until they are school-aged (30).

CDC maintains a list of peer-reviewed autism prevalence studies with similar metrics to ADDM surveillance reports (https://www.cdc.gov/ncbddd/autism/data/autism-data-table.html). Other federal programs reporting ASD prevalence information in the United States include the National Survey for Children’s Health (NSCH) and the National Health Interview Survey. The ASD prevalence estimate based on the 2020 and 2021 NSCH was 2.9% and the 95% CI (2.7%–3.1%) included the 2020 ADDM Network ASD prevalence estimate (2.76%) (31). These surveys aim to produce nationally representative estimates among children aged 3–17 years old and ascertain information about ASD through parental report, whereas the ADDM Network estimates are not intended to be nationally representative and are generated from empirical data collected from multiple sources among participating communities. The active surveillance approach used by the ADDM Network allows reporting of when and where children are identified with ASD and affords comparisons between and within diverse U.S. communities and is not dependent on parental survey participation and ASD reporting. To facilitate comparisons between different data sources, CDC maintains an interactive website that presents U.S. state-based ASD prevalence data from four data systems (ADDM Network, NSCH, Medicaid, and special education) (https://www.cdc.gov/ncbddd/autism/data/index.html).

Limitations

The findings in this report are subject to at least seven limitations. First, the methods rely on the availability, quality, and completeness of existing information and records to ascertain ASD cases and other indicators. Although all sites had access to special education classification data, certain sites did not have access to education records for their entire population, limiting the ability to identify children with ASD exclusively identified and served through their schools. Sites requested records from public school special education programs but did not review private school education records. Incomplete information could lead to misclassifying children’s cognitive ability, overestimating the age when they were first evaluated or when ASD was diagnosed, or failing to ascertain that the children were identified as having ASD. Sex information reflects what is represented in children’s records and might not reflect their gender identity. Second, the case definition for intellectual disability was measured using a child’s latest cognitive test or examiner statement of a child’s cognitive ability. Diagnostic and special education eligibility criteria for intellectual disability requires concurrent adaptive functioning deficits (32). IQ scores are not necessarily stable measures of intellectual ability over time, can increase among children with ASD in response to intensive early therapeutic interventions (33), and might be unstable during early childhood (34). The age at which children had their most recent test or examiner impression of cognitive ability varied by site. Third, the ADDM Network sites are not intended to be representative of the states in which the sites are located. ADDM Network sites are selected through an objective and competitive process, and findings do not necessarily generalize to all children aged 8 years in the United States. Interpretations of temporal trends can be complicated by changing surveillance areas, case definitions, data source access, and diagnostic practices. Fourth, small numbers result in imprecise estimates for certain sites and subgroups, and estimates falling below the selected threshold for statistical precision were suppressed. Fifth, the surveillance data system does not collect the number of ASD ICD codes a child received at a specific source, limiting comparability to analyses of claims/billing databases that consider number of ICD codes received. Sixth, the COVID–19 pandemic resulted in reduced access to records from some sources at certain sites; it was often possible to electronically obtain some data elements from these sources but not manually review the full contents of records. Disruptions in services and school closures during 2020 might have resulted in less documentation of ASD in records, which could decrease ASD ascertainment by ADDM sites. Finally, the prevalence of undetected ASD in each community as well as false-positive ASD diagnoses and classifications are unknown.

Future Directions

For the 2022 and 2024 surveillance years, the ADDM Network will continue to monitor ASD prevalence among children aged 8 years; progress in early ASD identification among children aged 4 years; and the health status of, needs of, and planning for adolescents with ASD as they prepare to transition to adulthood. The 2020 early identification ADDM Network report documents the impact of the COVID-19 pandemic on early evaluation and detection of ASD; the effects of the pandemic on ASD identification also will be examined among children aged 4 and 8 years in future years of surveillance. Additional analyses are needed to better understand changing patterns in ASD prevalence and differences between groups; for example, changes between 2010 (when higher income was associated with higher ASD prevalence) to the present findings of higher prevalence among lower-SES neighborhoods are comparable to studies from France and Sweden (35,36). In the future, it might be possible to link the Social Vulnerability Index to children ascertained through the ADDM Network to better describe disparities within communities.