An analysis of digital health records using large language models (LLMs) is challenging a long-held belief about the clinical identifiers of autism.
A new study led by researchers at The Neuro (Montreal Neurological Institute-Hospital) of McGill University and Mila Quebec AI Institute found that social communication factors may not be as important in identifying the condition as previously believed.
This assertion challenges the standard way of diagnosing autism, where clinicians assess individuals based on gold standard manuals like the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5). The DSM-5 criteria for diagnosing autism are split into two categories: one focused on behaviours, sensitivities and interests and the other on social communication and interaction differences.
The study tailored AI modelling to analyze over 4,200 clinical reports about children in Quebec. The analysis found that socialization criteria, such as emotional reciprocity, non-verbal communication and developing relationships, were not highly specific to an autism diagnosis, meaning they were not found much more in individuals diagnosed with autism than those in which a diagnosis was ruled out. Criteria related to repetitive motor movements, highly fixated interests and unusual sensitivity to sensory inputs, however, were strongly linked to an autism diagnosis.
The results, published in Cell, lead the scientists to argue that the medical community may want to reconsider the relative importance of current criteria, and place greater emphasis on repetitive behaviours and special interests.
How AI could improve speed, accuracy of diagnoses
Autism diagnosis currently relies on clinical assessment, as there are no biological tests based on genes, brain imaging or blood analysis. This lengthy process can delay access to essential supports. The researchers suggest that focusing on the more predictive traits could speed up and improve the accuracy of diagnoses. They highlight AI's potential to refine the process.
"In the future, large language model technologies may prove instrumental in reconsidering what we call autism today," said senior author Danilo Bzdok, a scientist at The Neuro and Mila. "Such a data-driven revision of today's brain diseases is a complement to what has historically been done by expert panels and human judgment alone."
This study was supported by the Brain Canada Foundation, Health Canada, National Institutes of Health, the Canadian Institutes of Health Research, the Healthy Brains Healthy Lives initiative, Canada First Research Excellence fund and the Canadian Institute for Advanced Research.
About the study
"Language models deconstruct the clinical intuition behind diagnosing autism" by Jack Stanley, Emmett Rabot, Siva Reddy, Eugene Belilovsky, Laurent Mottron and Danilo Bzdok was published in Cell.
