Unity Health Toronto:when a hospital startup partnership actually saves lives
What they're doing
Unity Health Toronto runs one of the most measurably successful AI deployments in Canadian healthcare. CHARTwatch is an early warning system built at St. Michael's Hospital that monitors hospitalized patients in real time. It pulls more than 100 inputs from each patient's electronic medical record, watches how those inputs change over time, and alerts clinicians when a patient is at high risk of unexpected death or ICU transfer.
A study published in the Canadian Medical Association Journal in September 2024 showed CHARTwatch reduced unanticipated mortality on the general internal medicine ward by 26%. Non-palliative deaths dropped from 2.1% to 1.6% after the tool went live. Unity Health has more than 40 AI programs live since 2019 and a Data Science and Advanced Analytics team of 25-plus that ships them.
The org also commercialized CHARTwatch through Signal 1, a health AI startup co-founded with Unity Health, backed by TD Bank Group, and now selling the same technology to hospitals worldwide. Separately, Unity Health runs the Health AI Academy with KPMG and Signal 1, funded by the DIGITAL Supercluster, training 200+ health professionals nationally through the Health AI Foundations Course and a smaller Fellowship at St. Michael's.
Why it matters
Unity Health solved two things at once: a real clinical problem (unexpected ward deaths) and a structural problem (how to commercialize hospital-built AI without losing the science). The Signal 1 partnership is the answer to "what do we do with the model after the paper?" Most academic AI dies in the journal. CHARTwatch is paid, deployed, and exporting.
What's working
- A peer-reviewed CMAJ outcome (26% mortality reduction) that gives boards, clinicians, and regulators something to point at
- A commercialization pathway through Signal 1 that keeps the science with the hospital and the scale with the startup
- An education engine (Health AI Academy) that turns the Unity Health playbook into a national capability
- Honest GEMINI-backed data infrastructure underneath, not a one-off pilot
What's harder than it looks
- Hospital-startup commercialization is legally and culturally complicated. Most academic medical centres don't have the IP infrastructure to do this cleanly
- A 26% reduction in mortality is a once-in-a-decade number. Other AI tools will not deliver this and shouldn't be expected to
- Training 200 people a year is meaningful but the gap is in the thousands. National training capacity is still under-built
Three takeaways for similar orgs
- 01The publication is the credential. If your AI tool can't survive peer review, don't deploy it
- 02Build the commercialization path before the tool is finished. Retrofitting IP terms after a research success is the slowest way to get to scale
- 03Training the sector is part of the work. The orgs that train the next layer of practitioners become the named players by default