Houston Methodist Bringing AI Scribes Into Inpatient, ED Settings

Nine-hospital Houston Methodist has been piloting Ambience Healthcare’s AI note-taking solutions for the emergency department and inpatient settings. The solution is fully integrated with the health systems’ Epic EHR. Jordan Dale, M.D., chief medical information officer and chief health AI officer at Houston Methodist, recently spoke with Healthcare Innovation about why the health system is working to extend AI scribes into more hospital settings. 

Healthcare Innovation: Before we talk about AI scribes, could talk about how you combine the roles of CMIO and chief health AI officer? 

Dale: I think there are natural fits. As a chief medical information officer, I've always felt like I'm kind of a pragmatic person talking about where technology could best be placed and best add value to our patient care and our health system. As artificial intelligence has become a key part of our technology strategy, it lended to that alignment. Also, I was the person who kicked off our AI oversight at Houston Methodist. I'm very used to talking with a mixed group of team members —legal, privacy — and figuring out how we can have constructive conversations about making progress. That's why they formally added that title — just so it was clear who's leading some of the oversight and some aspects of strategy with artificial intelligence. 

HCI: Is there a formal governance structure now in place? And if people within the organization are thinking of deploying something new, do they have to bring it through an approval process?

Dale: Yes, we've had an AI oversight committee since October 2023 and continue to refine it. Anything that's using artificial intelligence, we have an intake process and guiding principles that they should look at. It has been really, impactful for us to have a deeper understanding of this technology.

HCI: We've written quite a bit about deployments of AI scribes in primary care and specialist offices, but none yet in ED and inpatient settings. How did you get started in those settings and how long have you been working on it?

Dale: I am somewhat biased, because I still practice clinically as an inpatient hospitalist. I actually used the technology last week myself. It has a really seamless integration with Epic. We don't want to have a lot of niche use cases with one or two partners. We really want to have platforms that we can use across as many one of our clinicians as possible. 

We knew this had a different workflow and different expectations. We started about four months ago to enable the first few inpatient providers, and that's because we also started a program where we have virtual hospitalists helping in the evenings with the increase in census and volume of admissions. They were naturally at a stable workstation, with a stable video call and very high-fidelity sound. They were good test users because of their willingness, and how it aligned with that workflow. We were able to expand it to our in-person hospitalists across a handful of users as well, just to get some additional feedback. And then we fully integrated it. For me, workflow integration is going to be key to our ability to scale and maintain a solution. And that didn't exist in inpatient for almost any product I'm aware of, until we accomplished that with Ambience.

HCI: What about nursing documentation?

Dale: That’s certainly a frontier that we're exploring with many different ambient listening vendors. I think it's a different use case in terms of user requirements. A lot more of their documentation is discrete and put in flow sheets and really needs to be take a conversation and pull out 10 key things that need to go in 10 different areas. In that realm, physician documentation may be a little bit easier, with just having four or five different areas of our note that we need to summarize. 

We've been doing some early proof-of-concept work trying to understand the workflow. I think with both use cases, what's unique with ED and inpatient nursing is we're going from room to room to room so rapidly that sometimes switching to a different patient on your mobile device, there's a little friction in that experience. 

Houston Methodist has a lot of smart devices in our rooms. We've got cameras and microphones in every room to do our virtual service delivery. It is really apparent to me that for some of those use cases, we need to integrate with some of that technology that exists in our rooms. And we've gotten that feedback from the ED as well. They have been using it for about three months. I think they're up to about 30 users at this point, and they really enjoy it. I mean, similar to inpatient it is high volume, high frequency. You might see four patients in a compressed amount of time, and then you have a little bit of time to try to download your documentation, and then you’ve got to jump back into it. So they like having that assistant to refer back to and build their documentation, and it's performed well. Since we're leading in these integrations, we can also have really impactful feedback sessions with Ambience about what our users want, and where they see the product going in the future. 

HCI: Were there things in the ED that were challenging? I picture that setting as sometimes chaotic, and people moving from one patient to another real quickly, and maybe with a lot of background noise. But are there things that had to be thought through to make it work in that setting?

Dale: Interestingly, I had the same reservations that you shared, especially with the noise and just the number of people who are usually jumping in and out of the room. But we have had no negative feedback.

HCI: What about alerting patients that you're going to be using the technology? The pace is a little faster in the ED and in-patient setting than in ambulatory settings.

Dale: Like I said, I was using it last week. We're fortunate to use a lot of new technology at Houston Methodist, so I think in our standard consents we cover most of those aspects. And I think patients who come here know that we're always pushing and leading in how we're adopting technology. Most of the conversations I had last week, and most that I've heard in the ED and inpatient realm are very short. They already appreciate that it's relieving a burden from the clinician. Also, I think that the learnings from the ambulatory side have helped us where the patients actually recognize their notes are more personalized and tell their story better than me trying to remember five hours later. 

HCI: Are there things you want to measure about either efficiency gains or impact in other ways over time?

Dale: Absolutely. We'll certainly use some of our structure of how we measure value in our ambulatory domain to look at this in the ED and inpatient and it's going to be a mix of subjective factors. Usually, anytime somebody's on these platforms for 30 days, we do a survey of the user. Would you continue to practice without this? Is this relieving burnout and burden? Usually, that's very positive, right? But we have been very transparent about the objective things that we're going to look for, because we need to generate value to our patients and to our health system by adding this technology.

We'll usually look at documentation time and the timeliness of documentation. In the inpatient and ED settings, if you're trying to juggle five different things, that note that might be critical to the next care team may not be something that’s done in a timely fashion. I think the other thing that it could unlock is the ability to do better workload balancing. Nurses could have two patients who do require care, and maybe they should get more credit for that versus patients who are about to be discharged.

We will look to see how this these tools could help us understand the different trends in those care areas. 

HCI: If it does improve the timeliness of the documentation, could that help with the transitions of care between ED and admission or between the inpatient setting and post-acute care or discharge?

Dale: Yes, absolutely. That's what we see more anecdotally right now. We've measured some stuff that's more on chart summarization than ambient listening, but we see significant savings when I can start that from a draft and then add the critical details that are needed on top of that, versus having to start from scratch and try to remember that encounter from five patients ago.

HCI: Are you looking at Ambience in terms of a longer-term vision for their platform and how it might evolve to do other things?

Dale: Yes, we have explored the idea of having AI-generated discharge summaries and things of that nature just to help reduce that burden and to create a higher quality document that's timely for the care transition. I think the next area is to tackle the burden of documentation beyond just the physicians, to support physical therapists, bedside nurses and social workers. Those are certainly conversations that we're having because we don't want to have to add another solution to accomplish this for those providers.

 Also, only part of the audience for my note is directly related to care delivery and the next care team; there are many third-party or secondary uses of that for things like coding and billing. Those are things that our physicians today are trying to process as they're writing their note that if we add that to what this delivers, it can reduce that cognitive burden for them. 

HCI: Do you think it's possible that the EHR vendors like Epic and Oracle will try to develop their own solutions in this space? 

Dale: It’s a great question. It's one we ask ourselves a lot. I have to imagine that they will at some point, to some degree. I’ve seen ambient listening demos from two medical students that, for a certain purpose, did really well, right? So why couldn't Epic or Oracle do this with a few of their developers? But then I think what's been unique with our experience with Ambience is they've gone really deep into the content and refining it for some of our specialists. And refining those models might not be something that those companies want to spend the time on.

HCI: In your role as CMIO, are there other EHR optimization efforts that are top of mind right now?

Dale: We still are balancing two pillars of work. A lot of it is foundational informatics and CMIO work. That could be us looking at nursing flow sheets and making sure that they still serve a purpose, and there's a clear need and make sure there's governance around them. 

But then on the second pillar, it is how can we enable our workforce to be ready for everything that's coming in the future? That's been another rewarding part of these pilots, and we have a few on the nursing side as well — just helping them understand what they need to prepare themselves for and be educated about, so that when this technology is fully matured in their area, we can scale it with them, and not be something we just deliver without their input. For example, one we're doing for nursing is end-of-shift notes that are AI-generated, and summarize what they did during their eight- to 12-hour shift and letting them add any other personal details and signing that note so they can leave their shift in a timely fashion and still have a really high-impact documentation within the chart.