The number of health systems choosing to work with Pittsburgh-based Abridge on enterprise-wide roll-outs of generative AI-based clinical documentation tools continues to grow. Just in the past week, Abridge announced deals with Mayo Clinic, Duke Health, and Johns Hopkins. Other customers include Yale New Haven Health System, Emory Healthcare, Kaiser Permanente, the University of Vermont Health Network, and UChicago Medicine. Eric Poon, M.D., chief health information officer for Duke Health, spoke this week with Healthcare Innovation about the deployment starting there.
The Abridge ambient AI platform will be available to 5,000 Duke Health clinicians at more than 150 primary and specialty clinics. The platform will be used throughout Duke Health Integrated Practice and Duke Primary Care clinics, which serve communities in the Triangle and other locations in North Carolina.
Healthcare Innovation: There are several solutions on the market for capturing clinical notes, but it seems like Abridge is winning lots of customers in the academic medical center space. Is there something particular about their approach that was appealing?
Poon: We started trialing two products back in October, and while our providers were very happy with both of them, we did have to make a choice of one, and I think there were factors related to their overall roadmap and their overarching philosophy that made them more favorable than one of the competitors.
HCI: Has Duke Health piloted with one specific group of users before considering expanding it?
Poon: We deployed the two products across 160 of our providers, after we did a smaller scale safety pilot. We wanted to make sure that the products we were deploying were ready for clinical use, but after that, we deployed it to another 160 providers, and that’s where we actually collected a lot of very helpful feedback from providers of various specialties.
HCI: Have you heard initial feedback from the doctors that it helps them to feel more present and focus on the patient in the interaction rather than having to be thinking about documenting during that time?
Poon: Oh, absolutely. And you’re talking to one of them. Before using this technology, I didn’t realize, how much of my brain during a clinical visit I was devoting to being a core transcriptionist and typing up some cryptic notes in the electronic medical record as I talked to patients. Once I could disconnect myself from the keyboard, I was able to have much more natural conversations with patients. The encounters actually went faster for me. I was able to finish my clinics visits more or less on time, which has never happened before. I started using this technology, and, of course, the notes were far easier to clean up afterwards.
HCI: So has the widespread rollout begun yet, and is it fully optional for the clinicians to either use it or not?
Poon: We will certainly not making it mandatory. We’re still in the first week of the massive rollout. Our clinicians are very excited about this. We got lots of positive feedback during the 160-provider trial. I would say that we haven’t heard this much positive feedback on technology for a long time, so that really made the decision to deploy the solution much easier for our leadership.
HCI: Does this have the potential to work in a setting like an emergency department as well?
Poon: Yes, it does, and that’s one of the things we’re going to be trying out very soon. They haven’t done it yet.
HCI: I was intrigued to read that in addition to implementing this clinical notes platform, Duke Health was exploring opportunities with Abridge to co-develop other clinical applications that use ambient AI, and I was wondering if you might have some examples of areas where that kind of technology could have an impact the way it’s having with clinical documentation.
Poon: I think we are certainly very interested in deploying this technology in other clinical settings, and to work with Abridge to think about how you could use it in other aspects upstream and downstream from the patient encounter.
HCI: Concerning how AI is impacting clinical care in general, do you see opportunities for things like better clinical decision support or improved communication with patients through the portal?
Poon: Yes. I think we’re just beginning to see the introduction of generative AI into clinical practice. I think ambient technology has come onto the scene very quickly. We also — separate from Abridge — are looking at how we can use AI to draft MyChart messages, something that other organizations have also been doing. We’re doing a very thoughtful evaluation to think about where it adds value and where it’s going to be used broadly. I think there are opportunities to use it for some administrative functions, to help with various aspects of revenue cycle, for coding, for chart reviews. I do think that the technology could summarize the complex information in the chart. It can really help bring clinicians closer to the bedside, so that clinicians return to what led them into the profession in the first place.
HCI: I understand that you’re at the very front end of the wider roll-out.
But are there some obvious research questions that the use of this technology opens up that you’d be interested in following over the next year or so?
Poon: Yes, certainly we are evaluating across the board what is the impact of the introduction of this technology on clinician self-reported burnout, whether there is satisfaction with the documentation practices, whether clinician satisfaction overall will improve, and whether we will make improvements in our recruitment and retention and productivity. We are also actively exploring what would be the right way to introduce this technology to clinicians in training. That is still uncharted territory. I certainly believe that clinicians in training should be exposed to this technology before they are done with training and start practicing independently, because this technology is going to become more and more ubiquitous. But when that should start is still an open question.
HCI: Is there a governance framework at Duke Health around the deployments of AI tools? Do things have to kind of go through a vetting process before deployment?
Poon: Yes. We actually have given this a lot of thought over the last few years. Even before generative AI became the hot thing to pursue, we actually stood up a governance process. We call it ABCDS as which stands for “algorithm-based clinical decision support” oversight. It’s a multi-disciplinary group. We work with the clinical groups that are interested in deploying AI in various use cases. We ask them to present the evidence for exploring the technology, and we ask them to spell out how they will evaluate it.
Once they have tried it out, they will present the evidence that this is actually making an impact before we will give them the green light to continue to deploy the technology. So in some ways, it is a just-in-time approach to give some guidance. We don’t ask folks to prove to the nth degree that something is successful before they try it. But this is a way for us to encourage innovation while still keeping safety and efficacy and equity in in our minds as we use AI; it’s really part of our responsible AI strategy.
