Introduction
Autism spectrum disorders (ASD) are a complex set of disorders characterized by impairments in social interaction, communication and restricted or stereotyped behaviors (APA
2013). Historically, diagnosed ASD prevalence has been higher among white children than among black or Hispanic peers. Also, black and Hispanic children have been more likely than white children to have severe forms of autism and/or co-occurring intellectual disability (Jarquin et al.
2011; CDC
2018). One explanation for these findings is that autism has been underdiagnosed in some traditionally underserved children, especially those who have milder symptoms (Liptak et al.
2008).
The most recent report of the Autism and Developmental Disabilities Monitoring (ADDM) Network found that overall ASD prevalence was 1.68% among 8 year-olds born in 2006 (CDC
2018). This represented a 15% increase from the ASD prevalence of 1.46% in cohorts born in 2002 and 2004, which, in turn, was more than double the baseline ADDM Network prevalence estimate of 0.67% among cohorts born in 1992 (CDC
2007a,
2014,
2016). The recent ADDM report noted that the prevalence of ASD among black and Hispanic children was approaching the rate identified in white peers (CDC
2018). It was suggested that the narrowing gap between white children and peers of other races might account for some or much of the increase between the birth year 2006 and previous ADDM Network reports.
California Department of Developmental Services (CDDS) data extend further back in time than ADDM to birth years well before 1992 (CDDS
2003). A recent analysis of CDDS data indicated a dramatic increase in U.S. autism prevalence over the last 8 decades to present, by as much as 1000-fold from birth year 1931 to 2012 and 25-fold from 1970 to 2012 (Nevison et al.
2018). The U.S. Department of Education provides additional data on the ASD classification of children, under the auspices of the Individuals with Disabilities Education Act (IDEA) (Gurney et al.
2003; Shattuck
2006). State and local education authorities, acting on behalf of the IDEA, have tracked children ages 3–21 receiving special education services, beginning in 1991, going back to birth cohorts of the early 1970s. The availability of cohort-specific autism counts in both the CDDS and IDEA datasets over multiple, successive, years allows for “constant-age tracking” of ASD prevalence, among specific age groups. The ADDM network also uses a constant-age tracking method, over successive biannual reports, focusing specifically on 8 year-olds (CDC
2007a,
b,
2009a,
b,
2012,
2014,
2016,
2018), though recent work also has tracked prevalence among 4 year-olds (CDC
2019).
In this paper, we use constant-age tracking to describe race-specific trends among black, Hispanic and white children, and examine how race and ethnicity contribute to the total ASD trend across all races. We characterize the time trend in race-specific United States autism prevalence using the two best available datasets, which, in the IDEA dataset, extend most recently to 3 to 5 year-old children, born in 2012–2014. We pose the hypothesis that increased diagnosis of black and Hispanic children is responsible for the continued upward trend in ASD prevalence in recent years and consider what the IDEA 3–5 year-old data might portend for future ADDM reports and for ASD prevalence in the United States.
Discussion
One of the most prominent features of the IDEA 3–5 year-old dataset is the plateau in white ASD prevalence over birth years in the mid 2000s followed by a renewed increase in prevalence after birth year 2007. This plateau might suggest a stabilization of the environmental drivers of ASD in the mid-2000s followed by a new or increasing environmental insult after 2007 (Nevison
2014). Hispanic and black prevalence both increased more or less continuously across birth years 1996–2013, without an obvious plateau, although these groups also had lower prevalence than whites throughout most of this time. It is therefore possible that the mid-2000s plateau is not evident for these races because they were still catching up from historical underascertainment during those years, due to more effective outreach to black and Hispanic communities. However, the finding that black prevalence has exceeded white prevalence in the majority of states since birth year 2009, and that Hispanic prevalence also recently has surpassed white prevalence in a third of all states, suggests that additional factors beyond catch-up and access to services may be involved.
Changes in reporting also may have influenced the observed trends. A major shift in the formatting of the IDEA reports occurred in 2012 (birth year ~ 2008), around the same time that the plateau in white prevalence ended and resumed its rise. To our knowledge, though, the determination of autism classification did not change in 2012 when the IDEA reports were reformatted. Furthermore, the uptick in prevalence around birth year 2007–2008 is also seen in the California DDS dataset (Nevison et al.
2018), which did not undergo a change in reporting protocol at that time. Notably, prevalence did not plateau during birth years in the mid 2000s in the CDDS dataset, but it did grow at a slower rate than during the birth years of the early 2000s. The CDDS trends are not inconsistent with the IDEA data, given that whites comprised only one quarter of California’s kindergarten population over those years, such that a plateau in the white-specific trend may have been overshadowed by ongoing increases among other racial groups.
A recent race-resolved analysis of CDDS cases, age 7 years or older, provides important insight into this issue (Pearl et al.
2019). That analysis indicates that the annual rate in growth of autism prevalence among privately-insured whites indeed slowed down, by more than a factor of 5, to a near plateau between birth years 2000–2010. Hispanic ASD prevalence (both privately or publicly insured) also experienced a near plateau, albeit more briefly, between birth years 2003–2006. However, ASD prevalence among publicly-insured whites (considered “lower income” by the study) increased strongly over birth years 2000–2010, ending at 1.2% in birth year 2010, compared to 0.9% among privately-insured whites. Black ASD prevalence (both privately or publicly insured) also increased strongly over this time frame, ending nearly 40% higher than overall white prevalence by birth year 2010.
The publication of DSM-5 in November, 2013 (APA
2013) may have affected the IDEA trends, although the influence of the new diagnostic guidelines on IDEA is unclear given the discretion afforded to states to determine an autism classification for special education purposes. DSM-5 formally defined the term Autism Spectrum Disorder (ASD), encompassing but no longer distinguishing between milder and more severe subtypes. Assuming DSM-5 was widely adopted by the time of the 2014 IDEA report (birth years 2009–2011), the resumed growth in ASD prevalence, which occurred around birth year 2007, would have predated this diagnostic change. Nevison et al. (
2018) reached a similar conclusion about CDDS trends, i.e., that the renewed uptick in prevalence predated the change to DSM-5.
An additional possibility is that the continuous growth in ASD prevalence among 3–5 year-old black and Hispanic children throughout the 2000–2017 span of IDEA reports reflects a shift toward earlier age of diagnosis among those groups, which can create an artefact upward trend in constant-age tracking data (Hertz-Picciotto
2009). We were not able to assess this possibility directly from the available IDEA data, which do not partition 3, 4 and 5 year-olds separately by race/ethnicity. However, recent ADDM Network findings do not support a specific shift toward earlier diagnosis among blacks and Hispanics. The most recent ADDM report found that the median age of earliest known ASD diagnosis remained consistent over the history of the ADDM Network at 52 months and did not differ significantly by race/ethnicity (CDC
2018).
It is notable that access to health care among blacks and Hispanics improved over the decade that black and Hispanic prevalence caught up to and/or surpassed white prevalence. For children in the 1990s on Medicaid, blacks and Hispanics were shown to receive their first autism diagnosis 1.5–2.5 years later on average than the mean white diagnosis age of 6.3 years (Mandell et al.
2002). Similarly, a 2003/2004 survey found that being black, Latino, or poor was associated with decreased access to services (Liptak et al.
2008). In contrast, a 2013/2014 survey found little inequality between non-Hispanic whites and other races (Augustyn et al.
2019). Much of this improvement may be attributed to the State Children’s Health Insurance Program (CHIP), which has increased access to timely preventive care, specialty care and prescription medications (Liptak et al.
2008). CHIP began in 1998 and was expanded in 2009. CHIP has preferentially boosted insurance coverage for Hispanic and African American populations in particular, leading to increased health services for those populations (Kaiser Family Foundation
2019).
Hispanic autism prevalence has only caught up with white prevalence very recently in the IDEA 3–5 year-old dataset, achieving similar values around birth year 2012 or 2013 in the low, medium and high prevalence states (Fig.
1a–c). In the nationwide mean, Fig.
1d suggests that Hispanic prevalence may now be growing even faster than white or black prevalence, although this may be an artefact of the fact that a disproportionate share of the Hispanic kindergarten population (34%) lives in the six states that report high prevalence of > 1% in the 3–5 year-old age group. Within that high prevalence group itself (Fig.
1a), Hispanic prevalence was substantially lower than black or white prevalence over almost the entire history of the IDEA reports, catching up to whites only in birth year ~ 2013 on average. However, when the states are examined individually, Hispanic prevalence exceeded white prevalence in over a third of states by birth year 2012–2013 (Supplementary Figure S1). Furthermore, a recent study of individuals ages 7 or older found that Hispanic autism prevalence in California surpassed white prevalence in birth year 2010, at which time the Hispanic prevalence rate was 1.14% compared to 1.07% for whites (Pearl et al.
2019).
Black IDEA autism prevalence estimates caught up somewhat earlier to white prevalence, around birth year 2008 (Fig.
1). Subsequently, it has increased at a similar or higher rate than white prevalence, particularly in the high prevalence category states (Fig.
1a), and has exceeded white prevalence in the majority of states since the 2013 IDEA report (birth year ~ 2009). Thus, the IDEA data suggest that prevalence among black children already exceeds white prevalence in the majority of states and that prevalence among Hispanic children may be trending in that direction too. However, the IDEA data also suggest that these recent race/ethnicity trends are likely not yet apparent in available ADDM data, which thus far have extended only through birth year 2006.
The IDEA age 3 to 5 year-old autism dataset, which encompasses all states, and the population-based ADDM 8 year-old ASD dataset are the two primary ongoing sources of race-resolved ASD prevalence information available in the U.S. (IDEA also tracks race-resolved autism counts among 6 to 21 year-olds, but we considered this too broad an age range to provide meaningful constant-age tracking trends.) The two datasets cover overlapping birth year intervals (1996 ± 1 to 2013 ± 1 for IDEA and 1992–2006 for ADDM). IDEA data extend up to 8 years farther into recent time both because they sample a population that is ~ 4 years younger and because there is a 4 year lag in the publication of ADDM data, e.g., the 2014 survey of 8 year-olds born in 2006 was published in 2018 (CDC
2018). Due to their timely annual availability, IDEA autism classification data may provide a bellwether of forthcoming ASD prevalence trends in the ADDM Network. We advance this idea with the caveat that the IDEA autism counts for 3–5 year-olds significantly underrepresent the ADDM ASD estimates for 8 year-olds (CDC
2019).
In addition, some important differences and discrepancies between the IDEA and ADDM datasets should be acknowledged. The core purposes and definitions of the datasets are different, as are the methodologies and the level of detail characterizing the cases. ADDM is population-based while IDEA may be considered universal because it encompasses the vast majority of school age children. The ages of subjects vary, as do the roles of parents and education or health care professionals in instigating the evaluation for ASD. Recent studies generally have not found significant differences among whites and non-whites in health care-based screening rates and referral practices for toddlers (Rea et al.
2019) or in wait time to care, referral and final diagnosis following evaluation by pediatricians for school-age children (Augustyn et al.
2019). Recent studies also generally have found only limited differences between whites and non-whites in the receipt of school-based physical and occupational therapy (Bilaver et al.
2019). However, these researchers all noted the difficulty of separating effects due to socioeconomic status from those due to race/ethnicity and called for further research into lingering race or wealth-based disparities in access to health and education resources.
Over the duration of the ADDM Network, each biannual report has brought shifts in which and how many states were included. In addition, more recently, at some sites with a history of consistent participation, the number of surveyed counties changed across reports, as did access to records (i.e., health-plus-education versus health-only records). These inconsistencies complicate the Network’s ability to track trends in ASD prevalence. As the ADDM investigators commented most recently, “Comparisons with earlier ADDM Network surveillance results should be interpreted cautiously because of changing composition of sites and geographic coverage over time.” (CDC
2018; see also Supplementary File S4).
The inconsistencies of sampling complicate the interpretation not only of the overall ASD trend but also of the race/ethnicity-specific trends. Across the nine most stable ADDM states, white-specific ASD prevalence declined over the most recent three reports in both Arizona and Arkansas, but these were states with inconsistent sampling in which black and/or Hispanic-specific prevalence estimates also decreased. Three other ADDM Network states showed plateaus in white prevalence, but among these, Missouri also showed a sharp decrease in Hispanic prevalence and North Carolina reported a plateau in black ASD prevalence. Only in Georgia did white ASD prevalence stabilize over birth year 2002–2006, while Hispanic and black prevalence contemporaneously continued to increase. However, Hispanic prevalence in birth year 2006 was still 30% lower than white prevalence in Georgia (1.26% compared to 1.79%) (CDC
2018). Indeed, as of birth year 2006, white and Hispanic children still had the highest and lowest, respectively, race-specific rates of ASD in almost all ADDM states, including Georgia, and IDEA data suggest the gap may not close until birth year 2013.
The most recent ADDM Network report (CDC
2018) suggested that white ASD prevalence had largely stabilized and that the 15% increase in prevalence, from 1.46% in birth year 2002–2004 to 1.68% in birth year 2006, was driven by the catch-up of Hispanics and blacks who had been historically underascertained. This interpretation appears substantially based on the ASD prevalence findings from Georgia, which is the only ADDM state whose findings fully support the hypothesis. Notably this hypothesis is not supported by the findings from New Jersey, which, after Georgia, is the ADDM state with the most complete and geographically consistent ascertainment. In New Jersey, white ASD prevalence continued to increase over the birth year period 2002–2006, at a rate comparable to the increases in Hispanic and black prevalence.
The declining white population and increasing Hispanic population may also influence trends in ASD prevalence, both in past and future ADDM reports. Using total population (rather than race-specific) denominators, the white-absolute ASD prevalence declined in five out of nine states (Supplementary Figure S3). In two of these, Arizona and Arkansas, white-specific ASD prevalence also decreased, as discussed above. However, the white-absolute declines in the remaining three states (Georgia, North Carolina and Maryland) resulted from the decreasing white fraction of the school population combined with flat or only slowly increasing white-specific prevalence.
The plateau in white autism prevalence in 3 to 5 year-old in IDEA data from birth years 2004-2007 occurred during roughly the same period as the flattening of white prevalence seen in 8 year-olds in some ADDM sites, although differences in age and likely completeness of ascertainment, as well as inconsistencies in the ADDM protocol, complicate the comparison. The renewed growth in autism prevalence among white children, beginning around birth year 2007 and the ongoing growth in black and Hispanic prevalence observed in the IDEA dataset suggest that similar increases in ASD prevalence across all race and ethnicity groups, including whites, may be observed in the next ADDM report, which will focus on 8 year-olds born in 2008. Indeed, a recent report on 4 year-olds by the Early ADDM Network found an increase of 40% in ASD prevalence among children born in 2014 compared to children born in 2010 (from 2.0 to 2.8%) in New Jersey (CDC
2019). While the other two participating sites, Arizona and Missouri, reported no clear trend among 4 year-olds, these sites also show ambiguous trends in the 8 year-old counts (Fig.
4), possibly due to lack of access to education records (Missouri) and/or frequent changes and inconsistencies in the ascertainment region (Arizona).
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.