Deploying AI in healthcare is not just about the tech. UCHealth and the University of Colorado ran a rigorous pilot study of Abridge's ambient documentation tool across 250 clinicians. Through the rare combination of operational leadership, physician informatics leaders and a nimble research team, we could study a live deployment without breaking it.
When people talk about AI in healthcare, the conversation usually jumps straight to the technology. What model. What accuracy. What vendor. But after being involved with UCHealth and the University of Colorado’s deployment of Abridge’s ambient documentation tool across 250 physicians and APPs in the ambulatory setting, I continue to believe the technology is the easy part.
What made this work was everything around it.
The Partnership
UCHealth’s relationship with CU is genuinely unusual. It’s not a typical academic-operational handshake. What exists here is a tight integration between world-class data scientists, medical and operational leadership, and a physician informatics team that knows how to translate between the world of research and the world of real clinical workflows.
Yes, the combo sounds obvious, but it rarely comes together like this.
Studying a Live Deployment — Without Breaking It
The ACCORDS team took on something hard: using rigorous study methods inside a real-world deployment that was actively evolving. Operational decisions were being made in real time — some of which directly affected randomization and study validity.
As a result, the ACCORDS team pivoted. They adapted the study design to accommodate the changes without losing the integrity of the study. The result was a quality improvement project grounded in actual Epic EHR log data, combined with physician and APP satisfaction data, that produced data we could act on.
The ability to do rigorous science inside a messy, moving deployment — is increasingly a must-have capability. Organizations that want to be data-driven about AI adoption can’t wait for perfect.
From 250 to 3,000
What started with 250 clinicians have since scaled up to 3,000 physicians and APPs. We integrated new tools into physician workflows — starting with careful pilot testing and expanding with coaching, standard note design, and support by trainers rounding in clinics.
There’s a complex dance required to move a project like this through go-live: the research enterprise, operational leadership, and the medical informatics team all have to move together. I am grateful to work for an organization that can do this well.
Links:
- Data-Driven Results and AI Take Center Stage at State of ACCORDS(article)
- Watch the presentation on YouTube (Dr. Katy Trinkley, 7 minutes)