Journal club: Why do some children lose their autism diagnosis?

A study published 2 October in JAMA Pediatrics reports that 79 of 213 children who were diagnosed with autism at 12 to 36 months of age no longer met criteria for the condition at 5 to 7 years old. Spectrum asked autism researcher Deborah Fein, Board of Trustees Distinguished Professor Emeritus of Psychological Sciences at the University of Connecticut in Storrs, for her thoughts on the findings. In a 2013 study, Fein and her colleagues described a sample of people who lost their autism diagnosis.

The new paper by Elizabeth Harstad and her colleagues reports that among the toddlers her team considered — all aged 12 to 36 months and diagnosed with autism based on criteria in the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders — 37 percent had lost the diagnosis by age 5 to 7 years. This estimate of “nonpersistent” autism is nearly two to three times higher than that reported in most previous studies, which have estimated that approximately 10 to 20 percent of autistic children lose their autism diagnosis with age. Understanding the reasons for this discrepancy is important not only for predicting a child’s prognosis and recommending services, but also for defining an autism subtype that may differ in both genetic and environmental contributions.

So what might explain the high estimate reported in the new study? The researchers put forth one possibility: The children in their study were older by a year or two at follow-up than the children in the prior studies, giving them more time to lose core signs of the condition. But this cannot fully explain the discrepancy, because several past studies have followed children diagnosed as toddlers into early adulthood and have come up with lower estimates for nonpersistent autism. For example, a 2022 study by Rebecca Elias and Catherine Lord followed a cohort of 155 children diagnosed as toddlers and found that 13 of them, or 19 percent, lost the diagnosis by adulthood. Half lost the diagnosis after age 18.

Harstad and her colleagues report that female sex and relatively high adaptive functioning at baseline significantly contributed to loss of diagnosis, based on regression models that examined multiple factors. Although adaptive functioning alone reached significance as a predictor of persistent versus nonpersistent autism, the data show that the nonpersistent group had higher scores on every cognitive and language measure at both baseline and follow-up. Scores from the Autism Diagnostic Observation Schedule were not available for the initial early diagnostic evaluation; this is unfortunate, because having less obvious signs of autism to begin with is likely to also contribute to the nonpersistence of autism.

And the researchers mentioned the sex difference in the discussion but perhaps underplayed it: 32 percent of the boys in the study (who, of course, predominated in the sample) showed nonpersistence, compared with a whopping 61 percent of the girls.

One other factor that is likely to have been quite important in these outcomes is intervention. The study was done at Boston Children’s Hospital. Massachusetts in general, and Boston in particular, has a wealth of early intervention possibilities for autism. The study investigators don’t make much of the fact that 197 of their 213 participants, or 92 percent, received therapy, primarily applied behavior analysis (ABA), but this seems quite important and quite different from what might be found in other regions of the United States. In fact, I might have concluded that the high prevalence of nonpersistent autism and the high usage of ABA therapy in their sample serves as evidence for the effectiveness of ABA.

One methodological strength of the study is that the investigators looked at demographic differences between participants and eligible nonparticipants. They found that Hispanic families were less likely than families from other racial and ethnic groups to enroll in the study. What’s more, the children who participated had higher baseline cognitive and language scores than nonparticipants, and most of them were white and from affluent families with mothers who had higher educational levels. The fact that many of the participants had to be assessed during the pandemic, when there may have been less interruption of in-person services for white and more affluent families, may also have increased the disparities in accessing effective therapies.

The relationship between autism diagnosis and cognitive functioning in the persistent and nonpersistent groups, though somewhat complex, raises the age-old question of whether autism can span the entire range of intellectual ability, and how this range affects outcome. My colleagues and I studied a group of 2-year-olds with autism and quite low cognitive functioning (all below developmental age equivalence of 12 months) and followed them to age 4. We expected that some of these children would lose the autism diagnosis but retain a diagnosis of intellectual disability, but contrary to expectation, virtually all of them retained a diagnosis of severe autism plus intellectual disability. This is consistent with the Harstad study, which found that none of the nonpersistent children had cognitive scores below average.

So what can one conclude from this interesting and valuable new study? I would suggest that the paper presents possible outcomes for relatively privileged children — those who are white, affluent and have access to early diagnosis and effective early intervention, as well as relatively high cognitive and adaptive functioning at the time of diagnosis. The prevalence of nonpersistent autism in such a sample may therefore be viewed as the upper limit of what is possible given our current state of knowledge, but it may not generalize to broader populations that have more severely affected children, later diagnoses, fewer family resources or minority status, or live in regions with less effective intervention available.