The work began in 2020–21, first in a research context and is now embedded into clinical workflow. It has a clear aim: to shorten the diagnosis journey for the 1 in 12 people living with a rare disease by better equipping frontline providers and moving appropriate cases to genetics faster. “We were able to make this happen because we got the right people in the room — genetics and data — to share what we knew,” said Ivan Terekhov, the project’s Data Architect and Director of Research Informatics, AI and Technology at the CHEO Research Institute.
World-first AI algorithm at CHEO speeds rare-disease diagnoses for kids
The CHEO Research Institute has developed a rules-based AI algorithm, ThinkRare, that uses routinely collected clinical information from around 300,000 patient charts to flag children and youth who may have an undiagnosed rare genetic disorder. When the criteria are met, clinicians are prompted to consider referral and genetic testing, helping families get to informed care sooner.
“We were able to make this happen because we got the right people in the room — genetics and data — to share what we knew.”
Ivan Terekhov
ThinkRare has already contributed to 17 new diagnoses to date, with a diagnostic rate of 71%, compared with about one third for patients referred through a conventional pathway in recent work. One recent case reached a confirmed diagnosis at five months of age. That meant earlier clarity for the family and a quicker path to informed care.
About 1 to 3% of children live with a rare genetic disorder, so earlier identification matters for care and for daily life. Families describe the impact in everyday terms: “This investigation definitely gave us peace of mind.” Another parent said, “Less questions, less stress, less blaming ourselves.” Earlier answers can enable access to specialized clinics or trials and help families obtain school supports, even when no specific therapy exists.
ThinkRare was designed for breadth and equity. While initiatives like this exist elsewhere, they only focus on a single disease and tend to be supported by pharmaceutical funding tied to those treatments. The CHEO algorithm takes a broader, condition-agnostic approach, treated and untreated conditions alike, so that fewer children are missed and more families get clarity.
In practice, the team reports flagging about 5 to 10 patients per month. “As a result, one specialist may be able to follow a child instead of three, four or more, which is ultimately better for families and for an already-stretched healthcare system,” said Alexandre White-Brown, the project manager and genetic counsellor.
“As a result, one specialist may be able to follow a child instead of three, four or more, which is ultimately better for families and for an already-stretched healthcare system.”
Alexandre White-Brown
The ThinkRare project ended up being a useful catalyst for CHEO and the CHEO Research Institute to develop its AI@CHEO principles and framework to help translate AI principles into practice, guide how CHEO incorporates AI with privacy, monitoring and governance in place, and makes sure they stay up to date with rapidly changing tools and technologies. The initiative is a prime example that shows how using existing patient data enables innovative AI work that impacts patients directly. The goal now is to make ThinkRare available across Canada and even internationally through the creation of an app or API to make the algorithm adaptable to multiple Electronic Medical Record systems, so that the benefits reach more families and hospitals. Broader deployment would, however, require funding. But with targeted investment, ThinkRare is a Canadian innovation that could scale from a single-centre success to a national asset.