Autism-predicting model developed on ‘readily available’ data of child’s behaviour

New Delhi, August 20

Age of a child when they smile first, put together their first short sentence, along with difficulties experienced while eating, could be strong predictors of autism, according to researchers who developed an Artificial Intelligence (AI) model for the purpose.

Developed using readily available information related to a child’s behaviour, the model showed an accuracy of almost 80 per cent for children under the age of two years, the researchers, including those at Karolinska Institutet in Sweden, said.

The AI model was trained on a US database of about 30,000 individuals, with and without autism. Using machine learning, the model analyses 28 different features of a child’s behaviour before identifying them as being on the spectrum or not, they said. Machine learning algorithms can discern patterns in data to make predictions.

The child’s behavioural information for the AI model can be obtained without extensive assessments and medical tests before 24 months of age, the authors said in the study published in The Journal of the American Medical Association Network Open.

While autism can be diagnosed in a child at any age, symptoms such as not responding to their name, not smiling and avoiding eye contact, usually start to show within the first two years. The neurodevelopmental disorder is marked by repetitive behaviours and affected social behaviour.

However, early detection is fraught with challenges, including a lack of awareness among parents and guardians regarding early signs, leading to a delayed consultation with a professional.

Therefore, the study’s findings are “significant because they show that it is possible to identify individuals who are likely to have autism from relatively limited and readily available information,” said first author Shyam Rajagopalan, an affiliated researcher at Karolinska Institute.

The model’s performance was tested on a separate set of close to 12,000 participants.

“When testing the model for 11,936 participants, including 10,476 in the autism spectrum disorder (ASD) group and 1,460 in the non-ASD group, we correctly identified 9,417 participants with or without ASD (78.9 per cent). Among the children with ASD, the model correctly identified 8,262 (78.9 per cent) individuals,” the authors wrote.

They also analysed the child’s behavioural features most influential in predicting autism.

“Features like problems with eating foods, age at first use of short phrases or sentences including an action word, age at first construction of longer sentences, age at achieving bowel training, and age at first smile emerge as the most significant predictors,” the authors wrote.

The model identified autism in individuals with more severe symptoms and more general developmental issues, they said.

Further improvements to and validation of the autism-predicting model are planned and work to include genetic information in the model is underway, the authors said.