Clinicians most often diagnose patients with autism when they exhibit repetitive behaviors, special interests, and perception-based behaviors, rather than social deficits, according to a study issued this week in Cell.
“[T]he longstanding experience and expertise of healthcare professionals working alongside people with autism offer a rich resource to unravel the nature of autism,” wrote Jack Stanley, Ph.D., of the Quebec Artificial Intelligence Institute, and colleagues. “In a field that cannot rely upon biological testing methods, breaking down and analyzing subconscious clinical thought and decision-making processes can potentially shed light on opaque facets of the autism phenotype.”
Stanley and colleagues used 4,272 digital health records for 1,080 participants who were assessed for autism at the Autism Spectrum Disorder Assessment Clinic in Montreal (79% male, average age of 7 years, 429 receiving an autism diagnosis). The records included at least one full page of qualitative clinical descriptions, either referral reports and/or assessment reports. The authors built a large language model, a type of artificial intelligence that analyzes human language to identify patterns. Pretrained on hundreds of millions of general language sentences, the model was able to use the digital health records to correctly predict autism diagnoses.
The words that clinicians used most frequently to describe patients they diagnosed with autism involved concepts indicative of repetitive movements and speech, special interests, and sensory-processing and perception-based behavior. For example, the word “flapping” occurred 21 times more often in reports for patients diagnosed with autism compared with reports for children without an autism diagnosis. Additionally, sentences predictive of an autism diagnosis were highly akin to the DSM-5 autism spectrum disorder criteria relating to repetitive behaviors and sensory reactivity. However, there was no overlap between autism-predictive sentences and DSM-5 autism criteria regarding deficits in social communication or social interaction.
“As a consequence of our collective findings, we call into question the heavy focus on social deficits in research and clinical practice, echoed in established diagnostic instruments that are widely used by clinicians,” the authors wrote. “It may in fact be the case that these repetitive, special interest, and perception-based behaviors are much more prototypical of autism than mainstream research and the clinical state-of-the-art suggest.”
For related information, see the Psychiatric News article “More Intensive Early Autism Treatment Not Tied to Better Outcomes.”
(Image: Getty Images/iStock/andreswd)
Don't miss out! To learn about newly posted articles in Psychiatric News, please sign up here.