Doctors warn autism is being over diagnosed as they make crucial new finding using AI
Rapidly rising autism rates could be due to over diagnosis and poor diagnostic criteria, a study suggests.
Using an AI algorithm, researchers in Canada combed through more than 4,000 clinical reports from children being evaluated for autism to measure which criteria was most often used to diagnose autism.
The criteria were sourced from the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5), the gold standard in diagnosing mental conditions.
Diagnostic criteria for autism in the DSM-5 includes behavior like avoiding eye contact, highly limited interests, repetitive movements, and trouble forming friendships or having back and forth conversations, among other signs.
The study found social-related behaviors like nonverbal communication and forming relationships weren’t specific to an Autism diagnosis – they weren’t found much more in people diagnosed with autism than people not diagnosed.
However, repetitive movements – also called ‘stimming’ – and hyperfixations were strongly linked to an autism diagnosis.
The researchers said the findings suggest doctors are over diagnosing autism based on the social-related factors and aren’t spending enough time looking at behaviors like stimming, which are more closely aligned with the condition.
They argued streamlining autism evaluations to focus more on non-social behaviors with AI programs used to evaluate language would make diagnosing autism more ‘effective and efficient,’ as social behaviors are harder to evaluate.
It would also help patients get access to appropriate therapies and treatments faster. While there is no cure for autism, there are some therapies like applied behavioral anaylsis (ABA) and medications thought to improve behaviors.
A new study suggests autism may be over diagnosed due to diagnostic criteria that needs to be re-evaluated (stock image)
Dr Danilo Bzdok, a neuroscientist at the Montreal Neurological Institute-Hospital and Quebec Artificial Intelligence Institute in Canada, said: ‘In the future, large language model technologies may prove instrumental in reconsidering what we call autism today.’
The new study, published Wednesday in the journal Cell, analyzed 4,200 observational clinic reports from 1,080 children in Quebec being evaluated for autism.
The researchers tailored a large language modelling program – a type of AI that processes and understands language – to look through these reports and predict if an autism diagnosis would be given.
Of the 1,080 participants, 429 were diagnosed with autism by doctors. Included children were seven years old on average.
The researchers fed the DSM-5’s seven criteria descriptions into the AI model.
These are: difficulties sharing interests with others or having conversations, difficulties in non-verbal communication like making eye contact, difficulties maintaining relationships with other people, repetitive movements and mimicking, rigid adherence to routines or extreme resistance to change, highly restricted interests, and increased sensitivity to sensory stimulation.
Using the AI model, the researchers argued patients diagnosed with autism were most likely to exhibit non-social behaviors highlighted in the DSM-5’s criteria like repetitive behaviors, echolalia, highly restricted interests, or sensory issues.
The experts suggested that, given the findings, it may be more efficient to focus on non-social behaviors when evaluating a child for autism rather than social ones.
The above graphic shows the behaviors the AI program found was most often linked to an autism diagnosis
They also argued for diagnostic criteria to be re-evaluated to make diagnoses more accurate and it less likely for autism to be over diagnosed.
The team noted there were several limitations, including less data on older children who may exhibit different signs.
The study comes as the US faces a surge in autism diagnoses.
According to the latest CDC data, one in 36 children in the US have autism – just under 2million.
In the early 2000s, this number was closer to seven in 1,000.
Generally, most with the disorder are diagnosed by age five, though some can be tested as young as two.
Research published last year in JAMA Network Open found between 2011 and 2022, autism diagnoses in children between ages five and eight rose 175 percent, from two per 1,000 people to six per 1,000.
However, the biggest increase was among young adults ages 26 to 34, with a 450 percent jump, which suggests they were delayed in getting a diagnosis.
Along with environmental exposures, experts have noted the rise of autism is partly due to doctors getting better at detecting it.
However, autism advocacy groups have said the causes of autism are not fully understood and many experts suggest there isn’t one specific cause.
