The Problem with Criterion B: Did we contribute to Autism's heterogeneity?

Autism is routinely described as extraordinarily “heterogeneous”.

The same diagnosis can describe a person who is non-speaking, has significant intellectual disability and requires lifelong, around-the-clock care, and another who lives independently, has a family, and works in a senior professional position.

These are profoundly different clinical presentations.

Historically, this variation has been considered something that we have observed about autism over time, informed by research and lived experience.

However, in recent years the number of people receiving an autism diagnosis has risen substantially. In the UK, this growth - alongside increases in ADHD diagnosis and demand for mental health services - has prompted a major government review examining the factors that may be contributing to these trends.

The review does not extend to examining the diagnostic criteria themselves.

Yet the criteria may also be relevant to understanding the increasing heterogeneity of autism. Changes made to the diagnosis over time - and particularly the structure of Criterion B - may themselves have contributed to the breadth of presentations that now fall within the autism spectrum.

This is worth examining more closely.

As I explored in Redefining Autism, the DSM-5 changed autism in two important ways. It changed what counted as autism, substantially reorganising and broadening its symptom architecture. And it changed how an autism diagnosis could be established, allowing retrospective diagnosis when early developmental features were considered to have been overlooked or masked by learned strategies.

Criterion B is an interesting example of the first change. It now brings together the restricted and repetitive behaviours (RRBs) historically associated with autism, the newly introduced sensory symptoms, and remnants of the former Communication domain.

The result is four symptom groups containing some remarkably different phenomena:

1. Stereotyped or repetitive motor movements, use of objects, or speech.

2. Insistence on sameness, inflexible adherence to routines, or ritualised patterns of verbal or nonverbal behaviour.

3. Highly restricted, fixated interests that are abnormal in intensity or focus.

4. Hyper- or hyporeactivity to sensory input, or unusual interest in sensory aspects of the environment.

Only two of these four symptom groups are required to meet Criterion B.

The structure of Criterion B itself creates considerable variability. Each criterion bundles together quite different phenomena, often linked by a simple “or”: repetitive movement or speech; inflexible routines or rigid thinking; interests unusual in focus or intensity; sensory hypersensitivity, hyposensitivity or sensory seeking. Only two of the four criteria are required. These can then combine with the Social Communication criteria in numerous ways, producing remarkably different presentations that nevertheless meet the same diagnostic threshold.

Kamp-Becker (2024) made a similar point in relation to ICD-11, whose diagnostic architecture broadly parallels DSM-5. Depending upon how its diagnostic features are combined, more than 300 different symptom constellations can potentially result in an ASD diagnosis.

As the criteria are currently constellated, a verbally fluent person with good cognitive ability could potentially satisfy Criterion B through sensory hypersensitivity and rigid thinking, without repetitive motor behaviour, repetitive language or restricted interests.

That is a very different presentation from the autism phenotype historically associated with the diagnosis — and one that overlaps considerably with phenomena seen elsewhere in adult mental health.

Sensory differences had long been observed in autistic children, and there were calls for their formal recognition in the diagnostic criteria. DSM-5 included them. But their placement within the RRB domain was based on expert consensus rather than empirical evidence that sensory phenomena formed part of the same underlying construct.

Looking more closely at how Criterion B was assembled, some curious anomalies emerge.

Under DSM-IV, “stereotyped and repetitive use of language or idiosyncratic language” belonged to the Communication domain, while repetitive motor mannerisms — such as hand or finger flapping, twisting, or complex whole-body movements — belonged to Restricted and Repetitive Behaviour. DSM-5 combined them. Criterion B1 now groups “stereotyped or repetitive motor movements, use of objects, or speech” together, with echolalia included as an example.

The new diagnostic architecture brought repetitive speech, repetitive motor movements, rigid thinking, restricted interests and sensory sensitivity together within the Criterion B domain without clear evidence that these phenomena formed a single coherent construct.

Spackman and colleagues (2026) recently undertook an analysis testing whether the different phenomena grouped together under Criterion B actually belonged together.

Using data from autistic children and adolescents, they compared the four-part structure created by DSM-5 with alternative ways of grouping the same behaviours.

The DSM-5 structure performed worst.

A better-fitting structure separated repetitive motor behaviour from repetitive language, and restricted interests from unusual interests.

Sensory features were also separated into three distinct dimensions: hypersensitivity, hyposensitivity and sensory seeking.

Compulsions and self-injurious behaviour - commonly included in measures of repetitive behaviour but not represented as separate Criterion B categories - also improved the model when they were included.

The problem, then, was not simply that different people could meet Criterion B in different ways. It was that Criterion B itself may not accurately represent the phenomena it was designed to capture.

The analysis also revealed some subtle distinctions that had not previously been considered: Two behaviours classified as repetitive motor behaviours — watching things spin and fascination with the sight or sound of running water — actually fitted better with sensory seeking. Although the behaviour appeared to be ‘repetitive’, what was driving it may be sensory.

The implications here are not merely statistical. The alternative structure was much more closely related to meaningful differences between children - including age, IQ, anxiety, social communication and, particularly, language ability.

And here there is some historical irony. DSM-IV treated repetitive language and repetitive motor behaviour as distinct phenomena, placing the former within Communication and the latter within Restricted and Repetitive Behaviour. DSM-5 combined them within B1. Yet when Spackman and colleagues separated them again, the distinction proved meaningful: the resulting structure captured around a quarter of the variation in children’s language ability, while the broader RRB and sensory models captured virtually none. This is particularly striking given that difficulties in language production sat at the heart of the communication impairments described in the original autism construct.

Spackman and colleagues are explicit about why this matters:

“Reliance on overly broad diagnostic domains that collapse distinct subdomains has wide-ranging implications across research and clinical practice…”

They warn that grouping distinct phenomena together can produce erroneous findings in correlational and neurobiological research, obscure signals in intervention studies, and contribute to missed or misclassified presentations in clinical practice.

Their conclusion deserves attention: a general RRB/sensory domain “may not be empirically valid.”

It is possible therefore that Criterion B is not simply revealing the heterogeneity of autism. It may be contributing to it.

Implications for diagnosis in adulthood

Spackman’s study examined children and adolescents, whose current behaviour was reported by their caregivers. However DSM-5 also allows autism to be diagnosed retrospectively in adulthood.

If Criterion B is already this heterogeneous when diagnosis is undertaken in childhood, based on behavioural observations in real time, what happens when those same criteria are reconstructed from memory, by exploring an individuals’ subjective experience, twenty or thirty years later?

In 2018, Arvidsson and colleagues reported that Swedish children diagnosed with autism between the ages of seven and twelve had, by 2014, around half the autism symptom scores of children receiving the diagnosis a decade earlier.

No corresponding decline occurred among children diagnosed before age seven.

In other words, what qualified as autism in clinical practice was changing.

Eight years later, LaBianca and colleagues (2026) examined genetic data from a large Danish study, including more than 17,000 people diagnosed with ASD. They found a compatible pattern: as the years passed, people receiving an autism diagnosis had progressively lower polygenic scores for autism.

But LaBianca and colleagues did not simply report the decline and speculate about the reason for their findings. They tested different explanations, running simulated scenarios to see which one best reproduced the pattern in their data.

The pattern was most consistent with lowering of the diagnostic threshold.

It was not consistent with better detection, or diagnostic substitution.

Different country. Different dataset. Different methodology. Different kind of evidence.

Strikingly compatible result.

LaBianca et al. cannot tell us which changes to the autism construct produced this pattern. But their findings make the architecture of that construct considerably more important.

When a diagnostic category permits hundreds of different symptom configurations; when its constituent behaviours do not form the coherent construct its architecture assumes; and when the genetic liability of the population receiving that diagnosis is declining over time, surely we need to consider the possibility that the boundaries of the construct itself have become too broad?

Contemporary prevalence increases have been extraordinary. In recent studies, some of the steepest increases have occurred from adolescence onward, particularly among females (Fyfe et al., 2026; Messias et al., 2026). The sex ratio and clinical characteristics of the diagnosed population are changing. Recent genetic evidence adds another layer: Zhang and colleagues (2025) demonstrated different polygenic profiles for early- and later-diagnosed autism, with the later-diagnosed profile showing stronger genetic correlations with ADHD, depression and PTSD.

These are not trivial observations.

Diagnostic changes do not remain inside academic journals. They change how people understand themselves, how clinicians conceptualise distress, how advocacy groups respond, and where increasingly strained health and disability resources are directed.

Yet these studies of rising prevalence repeatedly return to the same explanatory framework:

Better recognition.

Previously missed cases.

We are finally finding the girls.

The clinical and sociological consequences of how we interpret research findings can be substantial. Yet surprisingly little attention has been given to a fundamental measurement question:

Has the thing we are counting changed?

Instead, findings from some of the world’s most prestigious universities and research consortiums are continually interpreted through the same epistemological lens: under-recognition.

Perhaps it is time to change the lens.

Because this is the problem hidden in Criterion B.

_________________________________________

Arvidsson O, Gillberg C, Lichtenstein P, Lundström S. (2018) Secular changes in the symptom level of clinically diagnosed autism. J Child Psychol Psychiatry. Jul;59(7):744-751.

Fyfe C, Winnell B, Dougherty M, et al (2026) Time trends in the male to female ratio for autism incidence: population based, prospectively collected, birth cohort study. BMJ. 392:e084164.

Kamp-Becker I (2024) Autism spectrum disorder in ICD-11-a critical reflection of its possible impact on clinical practice and research. Mol Psychiatry. Mar;29(3):633-638.

LaBianca S, Lousdal ML, Dybdahl Krebs M, (2026) Changes in Genetic Contributions to ASD and ADHD by Year of Diagnosis. iPSYCH Study Consortium and Autism Spectrum Disorder Working Group of the Psychiatric Genomics Consortium JAMA Psychiatry. Online ahead of print.

Messias E, Casanova E, Huang R, (2026) Autism Spectrum Disorder Incidence by Age and Sex in a US Health Care System, 2016 to 2024. JAMA Network Open. 2026; 9(7). Published online July 22, 2026.

Spackman E, Chetcuit L, Leekam S, et al (2026) Are Sensory Features a Subdomain of Restricted and Repetitive Behaviors? Evaluating Empirical Support for the DSM-5 Autism Journal of the American Academy of Child & Adolescent Psychiatry (2026)

Zhang, X., Grove, J., Gu, Y. et al. (2025) Polygenic and developmental profiles of autism differ by age at diagnosis. Nature 646, 1146–1155