Developmental Trajectories of Autism

The population of interest for this study is autistic persons who reside in California and were born between 1992 and 2016. The Lanterman Act requires the state to provide services for all residents with developmental disabilities, including autism. The California Department of Developmental Services (DDS) coordinates diagnoses and services and has kept digital records of its caseload since 1992; these records were acquired through a data use agreement.

Although DDS enrollment is voluntary and may not include all eligible persons with autism, eligibility is not means-tested, and most children with autistic disorder or ASD (hereafter referred to as ASD) in California are enrolled.¹⁹  In 2014, the DDS adopted the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition diagnostic criteria for ASD, potentially expanding the eligible population.²⁰ 

We matched ASD caseload records from the DDS from 1992 through November 2019 to birth records from the California birth master files (BMF) for 1992 through 2016 on infant names, birthdate, sex, and race and ethnicity using Stata’s user-written dtalink command.²¹  Uncertain matches were manually reviewed; overall, 86.0% of DDS records were matched to a birth record, with match rates increasing over time.¹⁸ 

We then selected the 862 794 annual records of the 122 392 children born between 1992 and 2016 who ever appeared in the DDS caseload with an ASD diagnosis. We excluded evaluations for infants under 1 year of age, those missing evaluative items, duplicates, and individuals with fewer than 4 evaluations, resulting in 71 136 trajectories. Missing data on sociodemographic and community characteristics (Supplemental Table 5A) further reduced the sample to 67 888 individuals for the regression analysis.

DDS clients are evaluated annually using the Client Development Evaluation Report (CDER) to determine service needs. The CDER was revised in 2008,²²  so we created a crosswalk between comparable pre and postrevision items (Supplemental Fig 4).¹⁶  Communication scores evaluating language usage range from 1 (does not use words to communicate) to 5 (uses complete sentences and has a vocabulary of >30 words). Social scores assessing clients’ abilities to engage in and maintain social interactions with others range from 1 (does not engage in 2-way interaction) to 4 (initiates and maintains interaction with others in familiar and unfamiliar settings). Sequences of scores vary in length from 4 to 26 years depending on birth cohort, age at ASD diagnosis, time on the DDS caseload, and frequency of evaluation.

From the BMF we extracted: birth year; maternal education level, race and ethnicity, and birthplace; infant sex; and whether the birth was paid for by Medi-Cal (California’s Medicaid program) as a proxy for socioeconomic status. Race and ethnicity is a social construct included because of its potential impact on children’s treatment, evaluation, and service receipt. Year of DDS entry is the year of the first CDER (with any diagnosis). Age of ASD diagnosis is calculated as the difference between the first date on the DDS ASD caseload and BMF birth date recoded into 5 ordinal categories. A cooccurring intellectual disability (ID) diagnosis is coded as present if an ID diagnosis ever appeared in an individual’s record.

We included 3 characteristics of the 5-digit zip codes in which individuals lived at during entry into the DDS: high inequality (zip code with a Gini coefficient for household income in the top quintile of CA zip codes), low population density (persons per square mile in the bottom quintile); and median home value (in $2000). We derived these measures from the 1990 and 2000 decennial censuses, the 2008 to 2012 5-Year American Community Survey, and the 2014 to 2018 5-Year American Community Survey and linearly interpolated rates for the intervening years.

To identify and describe subgroups with similar developmental trajectories, we used group-based latent trajectory modeling implemented with the lcmm package for R.^23,24  This approach assumes the population is composed of a number of subgroups with similar longitudinal trajectories and uses a polynomial equation to capture the relationship between age and the outcome, which, in our case, are latent variables characterizing individuals’ communication or social functioning. Model selection was based on goodness-of-fit statistics and posterior probability classification tables to confirm group discrimination. We then allocated individuals to their most likely trajectory (Supplemental Table 5B).²³ 

To examine the relationship between trajectory groups, we cross-tabulated group membership and compared each observed cell count to its expected value under an assumption of independence. To understand the sociodemographic compositions of each group, we calculated means and proportions by sociodemographic characteristics and conducted 2-tailed tests for differences from the grand mean and χ² tests of independence for continuous and categorical variables, respectively. Finally, to examine correlates of membership in a social trajectory group characterized by declining functioning in adolescence, we used the GLM function in R to fit a logistic model with standard errors clustered on individuals’ zip codes at entry into the DDS and calculated odds ratios (OR) with 95% confidence intervals (CI)s.²⁵  This study was approved by the Columbia University and California Committee for the Protection of Human Subjects institutional review boards.

Table 1 shows the sample characteristics. Eighty-two percent of the sample is male, almost 23% have an ID diagnosis, and the modal child is diagnosed at age 3 years. Hispanic individuals comprise the largest racial and ethnic group (44%), and 43% of mothers are foreign-born, closely resembling the demographics of California’s general population. Nearly 40% of births are paid for by Medi-Cal, and 27% of mothers have at least a 4-year college degree.

The 6 communication trajectories with 95% confidence intervals are displayed in Fig 1. In 5 of 6 trajectories, there is improvement in evaluated communication skills and little change after about age 15 years. Among these trajectories, groups we label early growth and moderate enter the DDS with moderate communication scores, whereas mid growth, late growth, and limited growth groups enter the DDS with low scores at first evaluation. Whereas members of the early growth, mid growth, and late growth groups – comprising the majority of our sample – reach the score ceiling at varying paces, members of the moderate and limited growth groups improve less and plateau. Finally, the sixth group – labeled “low flat” – includes 8.6% of the sample and is comprised of individuals who never communicate with words at any age.

The 7 social interaction trajectories with 95% confidence intervals are shown in Fig 2. These are markedly more varied than the communication trajectories. Three of the trajectories (high growth, medium-high growth, and moderate growth) exhibit improvement, albeit to different levels; 3 (medium-high flat, medium flat, and low flat) show little change over time and jointly comprise the majority of our sample; and 1 (adolescent decline [AD]) has the highest evaluated social skills at entry to the DDS but exhibits substantial decline in adolescence.

Figure 3 shows the joint distribution of membership in the 2 sets of trajectory groups, with larger circles signifying higher relative frequencies. In general, flat social trajectories tend to be associated with flat or limited improvement communication trajectories, and improving social trajectories tend to be associated with improving communication trajectories. The strongest association is between each dimension’s low flat trajectories, reflecting a subgroup of severely affected individuals whose autism trait levels change little from childhood through adulthood. Meanwhile, the largest cells in Fig 3, each comprising 11% of the sample, are the co-occurrences of the medium-high flat social trajectory, characterized by fairly high, stable social skills, and the early growth or mid growth communication trajectories, which exhibit improved communication during childhood. Notably, more than half of the individuals in the AD group come from the early growth communication group.

Tables 2 and 3 show sociodemographic characteristics stratified by trajectory group. Children with co-occurring ID are more likely to be in the low flat communication and social trajectory groups. Children of white mothers are overrepresented among the early growth communication group, whereas children of Hispanic and Black mothers are underrepresented among this group. Children of Asian mothers are overrepresented among both the low flat communication and social groups. Children of foreign-born mothers are less likely to show early communication growth and more likely to exhibit a flat trajectory. Children of mothers of higher SES (as measured by maternal education or Medi-Cal status) are overrepresented among the early growth and underrepresented among the low flat, limited growth, and late growth communication groups. Social trajectory group membership is less clearly patterned with respect to maternal SES than communication, but children of higher SES mothers are less likely to be in the low flat and moderate growth groups.

Finally, children in the early growth communication and medium-high flat social groups tend to live in communities with, on average, high median home values and low population densities. In contrast, children in the late growth, limited growth, and low flat communication groups live in communities with lower home values. Children in the AD social trajectory group live in communities with both lower median home values and low population densities.

Table 4 displays the results of a multivariable logistic regression model predicting assignment to the AD trajectory relative to all other social trajectories. The results reveal that AD is positively associated with female sex (OR = 1.21, 95% CI = 1.10–1.32). AD children are less likely to be diagnosed either younger or older than age 3. Children of white mothers are more likely to be in the AD group relative to children of Hispanic mothers (OR = 1.13, 95% CI = 1.03–1.25), whereas children of Asian mothers are less likely (OR = 0.73, 95% CI = 0.63–0.85). Those with foreign-born mothers have a slightly lower OR relative to those with US-born mothers (OR = 0.88, 95% CI = 0.81–0.97). Both low and high levels of maternal education are negatively associated with AD relative to high school (HS) graduates. Relative to early growth communication group membership, membership in any other communication trajectory group is negatively associated with an AD social trajectory.

Regarding community-level characteristics, living in a high-inequality zip code is associated with AD (OR = 1.14, 95% CI = 1.04–1.27). Each SD increase in median home value is associated with a decreased odds of AD by about 11% (OR = 0.89, 95% CI = 0.84–0.93), whereas living in a low-density zip code is associated with a 16% increase in odds (OR = 1.16, 95% CI = 1.04–1.29).

Using the largest available database with the longest follow-up of autistic persons, we uncover several typical patterns of ASD chronogeneity and their associations with individual- and community-level demographic characteristics. Although most autistic persons exhibit improved communication as they age, their social functioning tends to remain steadier. Unlike previous studies, we also find a small group whose initial high social evaluations decline rapidly in adolescence.

Like a previous study of this population,¹⁶  we find that children from families with more socioeconomic resources, as indicated by maternal education and private health insurance coverage, tend to exhibit more improvement. This is also true for neighborhood resources, where improvement is more common in zip codes with higher median home values.

We also find disparities by race and ethnicity; children of white mothers are overrepresented among the early growth communication trajectory, whereas children of Hispanic, Black, Asian, and foreign-born mothers are overrepresented among the communication and social trajectories displaying less growth. This analysis does not reveal the source of these disparities, which may be because of some combination of differential ascertainment,^17,18,26  racial bias in evaluations,²⁷  and unequal access to services and support,²⁸  among other potential explanations that require further research. However, these patterns align with broader socioeconomic and racial disparities in health^29,30  and may signify inequities in access to the resources that autistic people need to reach their full potentials.^31,32  Meanwhile, children of foreign-born mothers may experience challenges arising from language access.

Prior research identified distinct trajectory groups that show significant improvement or decline through childhood,^{9,11,14–16,33}  and a few small studies describe children who “regress” into disruptive or aggressive behavior at puberty.^34,35  This study is unique in identifying a subgroup of autistic children whose social trajectory declines in adolescence in a large, diverse cohort.

In the United States, roughly 50 000 autistic adolescents transition to young adulthood each year.³⁶  This transition is accompanied by changes in the healthcare support systems available to autistic persons, including changes in providers and insurance, and loss of educational services, creating obstacles to resource access at a critical time.^36,37  It is important to understand how these changes might co-occur with changes in autistic persons’ trait trajectories.

A typical member of the AD group displays early, rapid improvement in their evaluated communication skills combined with early high social skills. It is possible that, given their high functioning at younger ages, these individuals are exposed to situations and environments, such as mainstream classes, that tax their abilities as they encounter the more complex social interactions of adolescence, resulting in a real or perceived decline in their social functioning. At the same time, the stresses of adolescence or the onset of psychiatric conditions,^38,39  such as depression or anxiety, may trigger a decline in these individuals’ social skills. Adolescence has been found to be a difficult period for autistic persons, when many experience stigma and bullying,⁴⁰  with particular challenges for girls.⁴¹  These explanations are consistent with our results and not mutually exclusive.

Those in the AD group are more likely to be female, diagnosed at age 3, and have white mothers with a HS diploma. Although Medi-Cal receipt (an indicator of low household income) is not associated with AD, greater community-level resources as measured by median home values are protective, whereas living in a high-inequality or low-density community are risk factors.

Challenges to the medical model of ASD by the neurodiversity movement have pointed out that ASD traits, including differences in social and communication functioning, are not necessarily skill deficits, but reflect the range of variation in human interaction.^42,43  Through this lens, we acknowledge that, although it is unclear how the trajectories revealed in our study population might differ from those in the neurotypical population, it is likely that the latter also display significant chronogeneity and socioeconomic disparities. Further, the associations between trajectory groups and community-level variables remind us that evaluations of traits signify not just characteristics of individuals, but interactions between individuals and their environments.

First, both the DDS’ definition of ASD and the items used to evaluate social and communication functioning changed during our study period. To the extent we can investigate given younger cohorts’ shorter observation windows, we do not find evidence of discontinuities in the trajectory groups identified coinciding with these changes. Second, the evaluative items are not clinical assessments and may omit relevant dimensions. In addition, potential ranges of behavior may be truncated by the low levels of interaction represented by the highest scores, especially for older individuals. Finally, the sample is 82% male. Although this is consistent with the gender composition found in most studies of community-identified ASD, it is possible that girls are disproportionately underidentified, and it is unclear how this might affect our results.^41,44 

Areas for future research include: investigating the mechanisms behind racial and ethnic and SES disparities in trajectories and the social decline experienced by the AD group; reproducing these findings using clinically validated measures of social communication; and translating these findings into specific recommendations for clinicians, caregivers, policymakers, and autistic people themselves. In addition, in light of the finding that female sex is associated with the AD group and the under-representation of girls in this dataset, more research is needed on ASD in girls, particularly in adolescence. Finally, many autistic persons do not receive a diagnosis until later in childhood or adulthood,⁴⁵  and research is needed to understand how their trajectories might differ.

Although most individuals diagnosed with ASD show improved communication and social interaction functioning as they age, albeit at varying rates and to varying levels, not all do. Improvement is more common in relation to communication than social functioning, and we uncover a small group that exhibits a decline in their social functioning in adolescence. Trajectory group membership is correlated with SES and race and ethnicity, indicating that unequal access to resources contributes to disparities in individuals’ experiences of ASD’s chronogeneity.

Clinicians can aid in the provision of effective and inclusive services for autistic persons throughout the life course. Clinicians can ensure that all children are screened according to the American Academy of Pediatrics’ recommendations and assist in accessing appropriate resources and support.^46,47  Clinicians should also remain aware that autism traits, social expectations, and needs change as children age. Adolescence can be a time during which autistic children experience stress and may benefit from additional support. Further, the stigma and discrimination faced by autistic children can intersect with that arising from other characteristics, such as race and economic status.⁴⁰  Physicians can provide more effective care to diverse populations by developing cultural competence and awareness of systemic racism and advocating for policies that alleviate disparities.

AD

adolescent decline

ASD

autism spectrum disorder

BMF

birth master files

CDER

Client Development Evaluation Report

DDS

(California) Department of Developmental Services

HS

high school

ID

intellectual disability

OR

odds ratio

SES

socioeconomic status

ZCTA

zip code tabulation area