Study: Two groups of children receive delayed autism diagnoses
Key takeaways:
- Researchers identified two distinct groups of children who were more likely to be diagnosed later.
- One group had milder symptoms, whereas the other had more comorbidities that overshadowed their autism symptoms.
Researchers identified two groups of children with opposing support needs and symptom severity who are more likely to be diagnosed with autism spectrum disorder later, according to an analysis published in JAMA Pediatrics.
“Given the extreme heterogeneity of autism, we hypothesized that multiple distinct subgroups of individuals with delayed diagnosis exist,” Alal Eran, PhD, a computational biologist at Boston Children’s Hospital, told Healio. “However, the composition of these groups was unexpected.”
The study included 23,632 children from the Simons Foundation Powering Autism Research for Knowledge cohort who were diagnosed with autism between 2015 and 2020. They used phenotypic data to group participants based on type and severity of symptoms and used those groupings to determine which children were more likely to receive a delayed diagnosis, which they defined as after age 6 years.
Overall, 17,992 participants received a timely diagnosis before age 6 years, whereas 5,640 participants (23.9%) received a delayed diagnosis. Children who received a timely diagnosis had a mean of 4.34 comorbidities.
The researchers identified two specific groups with opposing support needs and comorbidities who were more likely to be diagnosed later.
The first group (D1) included children with lower support needs and lower comorbidity burden, Eran said. Compared with children who received a timely diagnosis, participants in this group demonstrated fewer repetitive and restrictive behaviors (means ratio = 0.686; 95% CI, 0.674-0.698) and had fewer comorbidities than children who received a timely diagnosis (mean, 3.47).
“We initially assumed that one such group would correspond to individuals previously classified as having Asperger’s syndrome — characterized by mild insistence on sameness, restricted interests, minimal to no speech delay and no significant developmental delays,” Eran told Healio. “Contrary to this expectation, our analysis identified a distinct group (D1) that exhibited fewer autistic traits and lower comorbidity burden, but in some cases, included intellectual disability.”
The second group (D2) included children with higher support needs and more comorbidities like intellectual disabilities, motor delays and congenital disorders. Children in this group had a mean of 8.12 comorbidities — the highest among the cohort. They also were more likely to demonstrate repetitive and restrictive behaviors compared with those who received a timely diagnosis (means ratio = 1.423; 95% CI, 1.388-1.458). The researchers hypothesized that these children’s autism symptoms may have been overlooked by other concurrent conditions.
Eran and colleagues used predictors, including language disorders, autistic traits, motor impairments and psychiatric conditions, to train a random forest classifier to distinguish the two groups, which it achieved with an area under the curve of 0.94. The researchers also replicated the results with an independent cohort of 10,193 participants.
“Prior research often treated individuals with delayed [autism] diagnosis as one group, leading to conflicting findings,” Eran said. “This study reveals two opposing subgroups, explaining why past studies have reported both higher and lower autistic traits in late-diagnosed individuals.”
Eran said these findings indicate the need for better screenings for children at both ends of the autism spectrum and the importance of screening children beyond toddlerhood.
“Early intervention is most effective when started early, but even school-age children can benefit from targeted therapies,” she said. “Providers should ensure that referrals for autism evaluations are made promptly, especially if concerns arise after early childhood.”
Perspective
Back to Top
I think it is particularly important to examine the sample and the groups. Individuals who were diagnosed with autism after age 6 were grouped into D1 or D2 based on level of “support need,” which, in turn, was primarily based on scores on behavioral instruments as well as the presence of other factors as listed in the article’s supplement. From what I could tell in the methods and supplement, the authors used a lot of factors in creating these groups, which runs the risk of overfitting and limits generalizability of the findings outside of the group comprising this sample. All this is to say that although the machine learning technology is interesting and has some really great applications, it must be examined carefully.
With that in mind, I have some reservations about the significance of this study.
I didn’t find this particularly surprising because both the D1 and the D2 groups reflect patients I see in my clinical practice.
Guidelines from the AAP include recommendations for developmental screening up to 30 months old, which means that children with subtle developmental delays who are not identified by that point face additional barriers to diagnosis. In many of these cases, the initial concern for a developmental disability comes from caregiver or teacher identification, or from pediatricians’ less formal developmental surveillance practices. This may partly account for those children in the D1 group, who had lower support needs to start with and may have been more likely to “slip through the cracks” early in life.
Regarding the group with greater support needs (D2), it is likely true that those findings represent, or at least were impacted by, diagnostic overshadowing, as the authors indicated. The question I wonder about with these individuals is what a new diagnosis of autism provides. With multiple developmental disabilities and support needs, does the addition of “autism” to a diagnosis list provide more specific information than the other medical and developmental conditions? Does it help guide treatment, and is that treatment more effective than what these individuals were receiving anyway?
In the United States, an autism diagnosis can open the door for access to resources, like applied behavioral analysis and academic supports. Because these supports are designed to influence issues like behavioral and academic difficulties, they take time to work. This is one reason why earlier initiation is important. There is also evidence that earlier intervention is associated with improved outcomes, particularly when these interventions are designed to address the specific challenges that are often unique for each child.
In my practice, I see mostly children and adolescents who are coming for assessment and management of challenges arising from developmental delays and disabilities. By the time they get to me, these individuals have already been identified as needing assessment for developmental disabilities, if not already diagnosed by other providers or by their schools. I most certainly see patients who fall in both categories of late autism diagnosis vis-à-vis level of support needs, as identified by the authors.
One of my duties is to investigate various elements of their development carefully in order to identify the most specific issues underlying their challenges. When I have a patient who might meet criteria for autism spectrum disorder (ASD), I think it’s particularly important to try to clarify the most specific underlying issue, which may be difficulty with things like cognition, language and communication, movement and/or academic skills like reading. This is especially important when it comes to identifying very specific targets for therapy because the more specific the underlying issue the greater impact therapies are likely to have in addressing the resultant challenges.
Given that a “delayed ASD diagnosis” was one that occurred after age 6 years, a next step in this research may be to examine factors upstream to diagnosis. Current developmental screening recommendations from the AAP include general developmental screenings at 9-, 18-, and 30-month visits, and autism-specific screenings at 18- and 24-month visits. This may represent a barrier to diagnosis for children who are not identified by screenings at these visits. Furthermore, there are lots of factors that may preclude these screenings in the first place, such as transportation and other barriers to routine pediatric primary care, as well as pediatricians’ work burdens. An improved understanding of the limitations of the current workflow may have important implications for policy and guideline development.
A second step I would recommend reflects the authors’ suggestion that temporal data would be enormously beneficial. Both groups had quite a few concurrent conditions, but because this was a retrospective review, it’s impossible to know which diagnoses were given before, after or at the same time as the autism diagnoses.
Moreover, this study may represent an important step toward better classification of autism. It is a well-known phenomenon that autism is highly heterogeneous —“If you have met one child with autism, you’ve met one child with autism”— but this heterogeneity may serve as a barrier to our understanding of the condition on a mechanistic level. By employing machine learning techniques that allow for large-scale sample analysis without sacrificing nuance, I hope that future research will be better able to classify distinct functional impairments that will lead to more specific interventions and greater ability to identify causative mechanisms. It will be important, however, to use this technology judiciously and for readers to be informed about how machine learning works in order to inform their interpretations of these studies, which I suspect will become more and more common.
Collapse