AI in Medicine: Highlights from Session #2
Hosted by Dr. Trevor Day, our second AMA-ACTT Digital Health Innovation Series session explores how AI tools are already being used in Canadian practice and what this means for clinicians and patients.
Speakers:
- Dr. Muhammad Mamdani – AI & medicine leader and co-creator of Chartwatch
- Dr. Mike Weldon – Emergency physician & electrical engineer, Red Deer
- Dr. Jake Hayward – Emergency physician & clinician scientist, Edmonton, lead for an AI powered scribe tool two-year pilot
Watch now:
Real-World AI in Use Today
AI is fundamentally about using data and math to understand the world. It is already helping with diagnosis, prognosis, treatment planning and communication, especially where cognitive load is high.
- Chartwatch: An early-warning system that predicts ICU transfer or death and pages teams, contributing to a 26% reduction in unexpected mortality.
- Dialysis risk tools: Algorithms flag high-risk hemodialysis patients, supporting targeted interventions, resulting in a 28% reduction of 30-day admissions.
- Clinical chatbots: Tools like the “Felix” post-operative chatbot give patients 24/7 guidance and reduce routine questions while improving the quality of questions. 80% of patients rated it good or excellent.
- Stroke AI in the UK: One of the AI interventions has resulted in 48% of stroke patients discharged with no or slight disability. This is a dramatic improvement from previous rates of 16%.
Jenkins: Alberta’s Homegrown AI Scribe
- Jenkins began as a prototype that listened to doctor-patient conversations and drafted notes.
- The major benefit has been cognitive offload with less mental energy spent typing or dictating, more time and attention for patients.
- Grant funding was available, which enabled the AI scribe to be better developed and hosted within AHS infrastructure, keeping data within the system. There are now a few hundred registered users.
Risks, Errors and Evaluation
- Bias & harm: Historical examples show algorithms directed resources to certain types of patients. Researchers later found that these algorithms were biased, highlighting the need for caution.
- Hallucinations & errors: Most notes need only minor edits, but a small proportion may contain an error that is significant, which reinforces the need for physician review.
- Evaluation is hard: It is labour-intensive to define and measure “clinically significant” errors across thousands of notes.
Consent, Privacy, and Personalization
As the AI Scribe is hosted within Alberta Health Services infrastructure, they can protect the data and can use confidential information securely.
The AHS consent script for patients is short and practical, for example:
“I have a new AHS tool that records our conversation and uses AI to help me write my notes. I review everything before it’s added to your chart. Saying no won’t affect your care. Are you okay if I turn it on?”
Very few patients decline; many explicitly say they’re happy to support anything that makes care easier to provide.
AI Literacy: “Learn and Do”
A KPMG survey ranked Canada 44th of 47 countries for AI literacy. The panel’s advice:
- Learn the basics of AI – key concepts, strengths/limits, and how to interpret performance and safety claims from vendors.
- Use the tools (carefully) – hands-on experience, in and outside medicine, is the best way to build intuition about where AI is trustworthy and where it isn’t.
See other videos in this series.