Holland Bloorview:precision health for kids needs data science the sector hasn't built yet

Pediatric rehabilitation teaching hospital~1,000 staffTorontoClinical, research, neurodevelopment

What they're doing

Holland Bloorview is Canada's largest pediatric rehabilitation hospital, and it has been quietly building one of the most credible pediatric AI research programs in the country. The work runs through the Bloorview Research Institute (BRI), where Dr. Azadeh Kushki, now Associate Chief of Data Science, leads the Neurodiversity and Personalized Health Lab. She holds the Tier 2 Canada Research Chair in Neurodiversity and Personalized Health.

Kushki's recent work includes AI models that predict which neurodivergent children will respond to specific psychotropic medications (stimulants, antidepressants, antipsychotics), trained and validated across four international cohorts. Her group also develops AI-assisted, low-cost technologies for kids, including virtual reality tools to reduce dental anxiety. In 2026, she became the named lead on the flagship project of OBI's NeuroFL federated learning platform, building a multi-modal pipeline to study children with brain-based differences without centralizing their data.

Why it matters

Pediatric AI is the hardest version of health AI. The data is sparser, the populations are smaller, the ethical bar is higher, and the "personalize the care" goal is more important than any single accuracy number. Holland Bloorview is building data science capacity from inside a rehabilitation hospital, which is unusual, and they are doing it around a population (neurodivergent children) where the one-size-fits-all model of care has been failing for decades. That combination, hard population plus precision health ambition, is what makes the work matter beyond the hospital walls.

What's working

A named scientific lead (Kushki) with a CRC and a defined lab, not a scattered cross-functional committee
A publication record that runs across precision medication response, AI-assisted clinical tools, and now federated learning for neurodevelopmental conditions
A purpose-built data science role (Associate Chief of Data Science) that signals BRI takes the infrastructure seriously
Anchor partnership with OBI and the NeuroFL platform, which gives the lab access to multi-site data without standing up its own
Integration with U of T's Institute of Biomedical Engineering, which feeds talent and methodology in both directions

What's harder than it looks

Sample sizes in pediatric neurodevelopment are small even by rare disease standards. Federated learning helps but does not solve the underlying scarcity
The "personalized care" framing has been promised in pediatrics for fifteen years. Funders and families are reasonably tired of the language without deployed tools to point to
Rehabilitation hospitals do not have the same data infrastructure budgets as acute-care academic centres. BRI's data science capacity is real but it is also lean, and growth depends on grants more than operating dollars

Three takeaways for similar orgs

01If you serve a small or specialized population, your AI strategy is a data-pooling strategy. Find or build a federated infrastructure before you fund the model work
02Name a Chief or Associate Chief of Data Science even if your hospital is smaller. The title tells researchers, funders, and partners where to land the work
03Pediatric and rehabilitation orgs are usually under-resourced for AI relative to acute care. Use that as a reason to specialize, not to compete on breadth