CHEO research shows how AI could reduce emergency department visits for mental health care

Feb. 4, 2026, Ottawa, ON - CHEO researchers have developed a promising new way to improve mental health care by using artificial intelligence (AI) and electronic health records to better predict which young people are most at risk of returning to the CHEO emergency department (ED) for mental health concerns. 

When a child or youth comes to the ED for mental health care, nearly half will return to the ED within six months. These return visits signal ongoing challenges and unmet needs. Reducing these repeat visits can mean earlier supports for families, more stability for young people, and less strain on the health care system. 

“This is the first step to planning more effective emergency mental health care for kids who come to CHEO. By working directly with clinicians, we have been able to gain insights into what the data means and ensuring it stays grounded in real-world experience, leading to more accurate treatment and overall, better outcomes,” said Dr. Kathleen Pajer, Director of the CHEO Research Institute’s Precision Child and Youth Mental Health (PCYMH) Collaboratory, which integrates biological, social, environmental, and clinical factors to create a more complete picture of each young person’s mental health needs.

In a new study published in BMC Medical Informatics and Decision Making, the PCYMH Collaboratory’s AI Data Scientist, Navjot Bains and team, analyzed more than 12,700 CHEO emergency department encounters involving 8,696 children and youth over a six-year period to help identify risk earlier, personalize care plans, and intervene before a crisis escalates.

Using information already collected during routine care, the team developed a machine learning model that could help clinicians identify children and youth who may be more likely to return to the ED within 30 days after their initial visit. The model consistently performed better than predictions based on clinical judgment alone, showing the potential for AI to support decision‑making in a meaningful way.

Importantly, the study found that many of the factors highlighted by the algorithm – such as past ED visits or previous mental health care use – were the same factors clinicians identified as meaningful. This alignment helps ensure innovative technology, like AI, remains grounded in real‑world experience and expertise and stays practical and usable for care teams. 

By relying on routinely collected clinical and demographic data, this hospital-enabled research shows how the predictive models can be embedded directly into hospital workflows, allowing clinicians to receive real-time insights, without adding new burdens. 

Ultimately, with better prediction of a young person’s likelihood of returning to the ED for mental health reasons, clinicians can reach out sooner to support families and plan more targeted follow-ups, while hospitals can better allocate mental health and emergency resources where the greatest needs emerge.