The idea of “classic” autism has existed for many years. In the fourth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV), published in 1994, for example, researchers and clinicians considered “autistic disorder” the classic autism syndrome, whereas “pervasive developmental disorder-not otherwise specified” described people who had autism traits but did not meet the full criteria for autistic disorder.
In the DSM-5, published in 2013, the broader “autism spectrum disorder” has become even more heterogeneous and has not, as a whole, mapped well onto any genetic or other etiological factors. Perhaps because of that, the idea of “classic,” “frank” or “prototypical” autism has resurfaced.
An emerging argument posits that starting with samples of research participants who fit the category of “prototypical autism” could optimize the chances of uncovering underlying etiological factors.
Proponents suggest that researchers compartmentalize cases by age, sex and outcome, as well as other key descriptors (e.g., degree of language delay or presence of intellectual disability) — a framework put forth in 2021 by Laurent Mottron at the University of Montreal. Then, autistic people within each compartment would be included or excluded depending on their phenotypic distance from the prototypical cases in that compartment, as well as the sample size needed for the specific study.
How would such “prototypical” cases be identified? Advocates suggest the clinical intuition of experts would determine selection — in combination with machine-learning algorithms and quantitative scores derived from diagnostic assessment tools.
Typicality would be determined by how similar cases are to the experts’ view of classic autism, how quickly this judgment can be made and how exemplary a teaching case they would be. But this proposal, in our view, hinges upon several questionable assumptions.
The first concern is the idea that experts will agree on which cases are prototypical. A small number of studies on inter-rater agreement on diagnoses of young autistic children show good results, but we could not find any reliability studies using prototypicality as the criterion. If independent experts cannot reach good agreement about prototypical cases, this would seem to invalidate all further studies of such cases.
The second assumption is that arriving at a more homogeneous set of cases will maximize the possibility of uncovering etiological commonalities. Although this is potentially true, neither neuroanatomical nor genetic studies have mapped well onto severe or similar cases to date.
The third assumption is that common co-occurring conditions (for example, anxiety, sleep disruption and intellectual disability) are unlikely to share etiologies with the prototypical autism characteristics. Instead, they likely have their own causes, which influence how autism manifests.
But once again the available evidence contradicts this notion. A few defined genetic syndromes have a phenotype that meets criteria for autism, such as Phelan-McDermid syndrome.
The underlying genetic variant in these syndromes clearly leads to autism and intellectual disability, as well as other distinctive characteristics of each syndrome. This occurs through pleiotropy, wherein a single gene affects two or more traits.
For that matter, it stretches credulity to imagine that the autism in a person with an IQ of 30 shares etiology with the autism in a person with an IQ of 130. Even monozygotic twins are not necessarily concordant for autism; one can have autism and the other not, and twins concordant for autism may still differ in their observed behavior.
The interaction between the core features of autism and the co-occurring conditions also affects how autism develops over time (behaviorally and possibly biologically).
There are still other practical issues to consider in the use of prototypical cases: What would qualify someone as an “expert”? Is there an objective way to define these clinicians or researchers? Who would decide?
Another contention relates to the compartments within autism. Characteristics and domains of functioning, such as cognitive level and language delay, that would inform these groupings are not necessarily easy to assess and are more dynamic than static.
Global outcome, for example, is hard to predict from early childhood functioning, and it might require another 5 to 10 years to ascertain. As much as 37 percent of early-diagnosed children lose their autism diagnosis by school age three to five years later, according to a 2023 study.
In addition, these criteria could easily multiply into an impossible number of compartments (combining, for example, two levels of sex, plus several levels of nonverbal IQ, language, age, overall trait severity, etc.). Each would require its own prototypical case at the center of the compartment.
At an even more fundamental level, the argument for studying prototypical autism cases assumes that autism exists as a valid syndrome. In the past 10 years, despite at least six decades of increasing literature on autism as a syndrome, this assumption continues to come under question.
A 2014 special issue of Autism was devoted to just this question, with interesting positions taken on both sides of the argument. Whereas some advocate taking autism apart and not assuming the validity of the autism syndrome, others — including us — have concluded that to jettison the syndrome would be to increase the heterogeneity of conditions under study even more broadly.
Acritical look at the accuracy of clinicians’ rapid diagnostic judgments is also essential. A survey of 151 clinicians found that 97 percent reported familiarity with the idea of “frank autism,” and, on average, respondents estimated that about 40 percent of autistic people fall into this category and could be diagnosed in about 10 minutes. That work, by Ashley de Marchena and Judith Miller, both then at the Children’s Hospital of Philadelphia, reintroduced the concept of frank autism in the research literature in 2017. Several recent studies showed that initial impressions of autism (in the first 5 to 10 minutes) tend to be correct but that initial impressions of non-autism miss a significant number of people with the condition; the latter might be assumed to be less prototypical. For instance, these judgments missed 37 percent of people with autism in a 2024 study whose authors included de Marchena and one of us (Fein).
To the extent that prototypical autism is defined by correct initial impressions, one might conclude that such cases do exist and that their features could be made explicit through additional research. But given the wide range of incorrect initial diagnoses in the published literature, more investigation is clearly needed to establish how reliable expert identification of prototypical cases truly is.
Of the existing measures used in autism research, the Calibrated Severity Score (CSS) on the Autism Diagnostic Observation Schedule (ADOS) may come closest to prototypicality. The CSS is a score of autism trait severity that accounts for a person’s age and language functioning level. Although that metric is conceptually distinct from prototypicality, people with high CSS scores might well be those whom clinicians judge to be prototypical.
But combining prototypical cases across ADOS modules could result in a highly heterogeneous set of cases. For instance, there could be a highly verbal adult with awkward social interactions alongside a nonverbal toddler with severe developmental delays in the same group. How likely is it that these people share etiologies, pathophysiologies, outcomes or effective treatments?
To combine them into a homogenous “prototypical” group, other biological, developmental and co-occurring factors must be taken into account, which are likely to be outside the defining criteria for prototypical autism.
If prototypical case research is a dead end, what are the alternatives? A comprehensive review of such suggestions is beyond the scope of this essay, but several possibilities can be mentioned. A genetics-first approach would begin with identifying a genetic syndrome and then exploring the range of phenotypes. Another option is studying children in the first 18 to 24 months of life, when the features of autism may be in a prodromal phase and treatment not yet begun.
Yet another possibility is agreeing on a set of characteristics across research sites to produce a large sample of well-described cases in which one could find clusters of cases. Or researchers could focus on features that might be less susceptible to environmental inputs, such as apparently high pain thresholds in some autistic people.
The concept of prototypical cases is appealing, especially to experienced clinicians, the large majority of whom believe they can identify such cases. This belief may, however, be an artifact of human cognition, where a great deal of varied input can produce spurious phenotypic categories that will not map onto biological variables.
Human cognition tends to reify concepts — that is, to treat something that is abstract, such as an idea, as real or concrete. Peter Zachar of Auburn University has contended that this tendency applies to psychiatric and neurodevelopmental classifications. Those arguing against the assumption that autism is a coherent syndrome would claim that 60 years of research has led us to reify a syndrome that does not in fact exist. This does not seem likely to us.
But if one wanted to pursue prototypicality, we think that the next step would be to assess the degree of agreement among experienced clinicians at different sites, who have not trained together. Moreover, this approach would then need to be used in a series of research efforts to determine if it actually produces more homogeneous outcomes.
We have argued that taking the prototypicality approach further into the realm of etiology depends on several unwarranted assumptions and therefore is unlikely to advance research productivity.