​A team of researchers has harnessed the power of artificial intelligence (AI) to better predict autism spectrum disorder (ASD), by training an AI model to analyze large-scale health administrative and registry data.  

Dr. Christine Armour, investigator at the CHEO Research Institute and clinical geneticist at CHEO, in collaboration with Dr. Kevin Dick, AI Data Scientist at Better Outcomes and Registry Network (BORN) Ontario, and Drs. Mark Walker and Steven Hawken at the Ottawa Hospital Research Institute, published findings from the first of-its-kind study in North America evaluating whether this type of machine learning approach to ASD detection was possible. They discovered it is, with much more research and work ahead.  

“Early diagnosis of ASD would allow us to better predict what the needs are and plan what resources are required, providing the right support to children, youth and their families," said Armour, who is also an Associate Professor of Pediatrics at the University of Ottawa.

ASD is a common neurodevelopmental condition that affects how children communicate, learn, and interact with others. Early identification is important to ensure timely access to care and resources, yet current approaches rely heavily on behavioral observations from parents or providers, which often leads to delayed or missed diagnoses. 

Armour's team demonstrated that it is possible to train an AI model to predict ASD using large-scale health administrative and registry data from BORN, CIHI, and ICES. They used state-of-the-art deep learning methods to model this large, complex, and diverse data – overcoming significant challenges in the field of health AI. Analyses were performed in the Health AI Data Analytics Platform (HAIDAP) at ICES.   By applying AI models to this routinely collected health data in Ontario, they demonstrated it could lead to earlier diagnosis through systematic identification of children at an increased likelihood of receiving an ASD diagnosis by age five. 

This research is supported by the Precision Child and Youth (PCYMH) Collaboratory at the CHEO Research Institute, which aims to transform care by combining research, data, and technology to create personalized mental health interventions that considers the whole child.

The next phase of this project will bring together clinicians, AI scientists, researchers, and individuals with lived experience of ASD to refine and validate the AI model. Through incorporating new high-quality data sources and transformer-based machine learning methods, they hope to improve predictive accuracy and generalizability. They will also explore how various factors influence model performance, assess its utility across diverse populations, and identify key considerations for clinical integration. 

With a bold vision of moving beyond case-by-case identification of ASD to a population-based screening approach, the aim of this research is to support earlier referrals for formal ASD assessment and timely access to resources, ultimately improving mental health outcomes and quality of life for children and families. 

This work represents a foundational step toward a future where AI-enhanced screening can help ensure no child is left behind due to late recognition of ASD.