Making AI Safe for Healthcare w/ Dr. Holly Urban

Generative AI is transforming the way clinicians interact with technology. In this episode, Dr. Holly Urban, VP of Business Development at Wolters Kluwer, joins John Driscoll to discuss how AI, ambient listening tools, and trusted medical content like UpToDate are improving clinical workflows, easing documentation burdens, and making healthcare delivery more precise, efficient, and human-centered.

🎙️⚕️ABOUT DR. HOLLY URBAN
After practicing as a primary care pediatrician, Dr. Urban worked for several EHR technology and evidence-based content companies, and has served in healthcare IT leadership roles for over fifteen years. With this experience, she brings unique insight into how evidence and workflow can optimally merge to benefit clinical users and healthcare teams. Her experience with both provider and payer customers allows her to holistically understand the healthcare ecosystem with all the complicated drivers that influence how care is provided.

Dr. Urban believes generative AI holds a tremendous opportunity to improve healthcare for customers, through improved clinician efficiency and patient contextualized clinical decision support. However, clinical GenAI needs to be built on trust, both in terms of data consumed by the AI model and by clinical validation of the output.

Dr. Urban also holds a BA from Stanford University, an MD from the University of Colorado School of Medicine, and an MBA from Regis University. Before joining Wolters Kluwer Health, she served as CMIO at Oracle Cerner, Director of Product Management at MCG Health, and VP of Product Management at McKesson Horizon Clinicals.

🎙️⚕️ABOUT CARETALK
CareTalk is a weekly podcast that provides an incisive, no B.S. view of the US healthcare industry. Join co-hosts John Driscoll (President U.S. Healthcare and EVP, Walgreens Boots Alliance) and David Williams (President, Health Business Group) as they debate the latest in US healthcare news, business and policy. 

GET IN TOUCH
Follow CareTalk on LinkedIn
Become a CareTalk sponsor
Guest appearance requests
Visit us on the web
Subscribe to the CareTalk Newsletter

Transcript

John Driscoll: 0:07

Welcome to Care Talk America's home for incisive debate on healthcare, business, and politics. I'm John Driscoll, the chairman of Yukon Health, and I am delighted this morning to. bring Holly Urban, Dr. Holly Urban to the show to tell us a little bit about the practical applications of Gen AI in practice today in healthcare. Welcome, Holly.

Dr. Holly Urban: 0:31

Thank you so much. It's an absolute pleasure to be here.

John Driscoll: 0:36

Holly, maybe you could tell us a little bit about your journey to, in healthcare and to Walters clear what got you to this point. Then we'd like to get into the practical applications of what's going on with AI and, and the practice of medicine today.

Dr. Holly Urban: 0:51

Okay, great. Yeah, so, um, I'm a little bit unique in that I'm a third generation physician. My grandfather was a gp, my dad was an orthopedic surgeon, and, um, I was trained as a pediatrician and I, um, was practicing in pediatrics and working in quality. And, um, this was about the time that Joint Commission rolled out. Um, medication reconciliation and at the hospital I was working at, they had an IT shop that sort of. Designed a whole, uh, electronic system to do medication reconciliation, um, which is where you compare a patient's medications that they're on at home to, against what they're on in the hospital, and make sure that there's no contraindications or that everything's, you know, that they're continuing the medications they need. It's a, it's a big patient safety effort. Well, our IT department developed this system with no clinician input. They threw it over the fence to quality. They said, Dr. Urban, can you please go roll this out to the hospital? It didn't go particularly well and I kept saying what I, I recognized the power of using technology for patient safety and better patient outcomes, but it had to be done with the clinician involved. And so why didn't they design this or why didn't they do it this way? And it just sort of ignited in me this passion to be more involved in designing software. So I left clinical practice and started working at an EHR vendor about. Little over 15 years ago, and since then, I've, uh, stayed at, um, working in healthcare IT at EHR vendors and now at Walters Klu

John Driscoll: 2:23

Med reconciliation thing is quite a big deal. If you remember the, I think it was the, um, the academies report to UR is human. Which indicated that we were losing as many people to, uh, death and injury, to unmanaged, reconciled meds of as many as the equivalent of 2 7 40 sevens crashing a week. That's right. And we, we really were, that that was a massive issue and I think people don't realize that there without some sort of. Digital overlay and informed connection, connectivity to the doctor at the point of prescription. Getting the meds your, your per your, your, your beds. Right. Particularly for anyone with any complex condition who may be on four or five different meds or anyone over the age of 70, typically on five or six different meds. That's a, that's a, that's a massively important task. And you're saying that they basically, the IT department. Designed it, you know, sort of took its own shot at it and then, and yeah. And threw it at the doctors.

Dr. Holly Urban: 3:24

Right. And sort of did it in a way that didn't reflect their workflow, that didn't reflect, you know, how they're interacting with the system that didn't interact, that didn't, um, was it, um, integrated with their current systems for ordering medications or prescribing medications. So, you know, I think it's a good lesson that's. It's not a new issue where you're talking about technology. It has to work in the workflow. It has to work and be integrated with your current systems in order for it to work. It's not enough to have the next greatest shiniest technology if you don't have those things. Um, also taken under account.

John Driscoll: 4:00

Well, I think what's, what's exciting about your approach is that workflow isn't just the digital workflow, it's actually the way patients interact from the point. That they try to make an appointment to show up in the office to actually kind of go through the system. And it's everything doctors do, most of which are, are physical and manual assisted by digital. Mm-hmm. And, and so in integrate having a doctor sort of at the center of solving what's the doctor's problem, it's who to make a lot of sense. Uh, and so how did you get. From there to, you said you joined one of the EHR companies. Obviously EHRs are the electronic health records that sort of are the, that, that, that, that, that, uh, I guess define us as patients or I guess at it's at their best and, and define this as bills at their worst. Um, what, what, which, which, which health record company did you join?

Dr. Holly Urban: 4:55

So I started working at a company called McKesson on their Horizon clinical solution. Um, at the time they were up there and competing with Epic and Cerner. Um, they sort of have fallen off and, um, they actually shut down that whole horizon clinicals line. But from there I went to two evidence-based content companies, um, zinc's Health, which does evidence-based order sets and MCG Health, which does, uh, evidence-based medical necessity guidelines. Then from there I went to Oracle, Cerner, where I was the CMIO for their federal business. So,

John Driscoll: 5:28

which it's, it's, it's fascinating that you went from one big challenge, which is this, the re reconciling meds and automating that, which really did save a lot of lives to evidence-based pathways because I think what's what might shock a lot of patients. Or non practitioners is how often the ev the, the, the, the guidelines that are informed by data, evidence-based guidelines aren't really integrated into workflow. What was that experience like and how effective were you?

Dr. Holly Urban: 6:02

Yeah, I mean, it, it, it, it's one of the reasons why coming to Walters CLO as a dream job for me because, um, of the up-to-date product, it's the most commonly used clinical decision support, you know, we have. 600 million views of our content a year. Um. But in order for it to really be at its best is when it's actuated within the workflow. So it's, you know, it's one thing for you to be ha go out to a referential site where you're gonna go and read a website. It's much more impactful when it's actually within your workflow. And that can mean, you know, within the workflow, the electronic health, health record. Really excitingly some of the new work we're doing with the ambient vendors, um, such as a bridge where we're actually gonna have that up-to-date content within that ambient workflow. Um, so that it's absolutely enabling that clinician to not only use these ambient vendors, which what typically what these ambient vendors do is they're recording the conversation between the patient and the physician, and then they're using generative AI to draft a note while. Since they're the physician's already in that workflow where they're drafting their note, it's a good opportunity to say, how can we leverage some clinical decision support to make sure that they're leveraging the most recent evidence-based recommendations, that they're making good decisions for their patients so that they can have better patient outcomes. Yeah, I, I think, and

John Driscoll: 7:27

just to. Could contextualize this, Holly, I think it's about 17 years between the time that yes, a academics have have determined that there is a particular guideline or or form of therapy and it becomes general practice. So with all of and, and. And if anything, the amount of knowledge around what works and what doesn't is actually increasing at a time when doctors have less and less time to absorb it, particularly in the context of the patient visit, which in many cases is, you know, I don't know, seven to eight minutes. And so all of the prep work and. Data work that, that, that happens to a patient going into a visit, um, is not really informed by that clinical insight that has to be really provided by the, the, the NP or the doctor or the specialist at the time of that relatively short visit. And I think the exciting thing about the. Clinical insights that you could, you can actually make that visit more intelligent, more impactful by leveraging the best information. I think that's, that I think is why, you know, as I understand it, what Walter SLU is doing with UpToDate is really creating accessible, digestible insights that are constantly being updated around what actually should. As part of practice to make the doctor's job a little bit easier. Is that the right way to think about up to date?

Dr. Holly Urban: 8:54

Yeah, absolutely. Yeah. You, you hit the nail on the head. I mean, medical knowledge has grown exponentially and, and the complexity is getting to the level where there's no way one individual physician can carry around all this information. So not only just keeping on top of the medical literature, but then also being able to. Remember all the complexity and that's why the technology is so exciting to me because if I can say, if I can help break down sort of what are these patient specific contextual details that are gonna impact your decision and then offer you like. Here's the most recommended therapy. Here's the recommended medication in a way that it's in the workflow where you're making those decisions. I mean, that is so powerful. Again, the whole goal is to make sure that patients have better outcomes, um, and that they're getting the most recent therapy to, to achieve those good outcomes.

John Driscoll: 9:48

Well, and, and, and then to your point, unless it's provided digitally, you can't update it in real time for what actually is happening at the NIH or at a particular cancer center or at a, or a particular insight into how a drug might interact with different populations in different ways. Um, so how do you may maybe now as we, as we, as we step into that next phase of ambient listing, I'm not sure people realize how common. These ambient listener listening pieces of software like Nuance, a bridge, Suki. But maybe the, we could, we could, we could contextualize this by saying that in, in some of the better practices historically. Doctors would actually employ scribes people. That's right. Almost along the lines of, of the ancient Greece where the doctor would be trying to do the work and someone would be literally typing or writing it down to help substantiate and inform the patient record. Now that we've got these wonderful, this wonderful ability to, to, to, to, to, to listen. Technology, what is that? Just maybe describe that, that ambient, what, what is ambient listening technology? What, just describe that and then we can get into a little bit of how, how that connects to Walters clear up to date.

Dr. Holly Urban: 11:10

Okay. Sounds great. So what the te at that, at the heart of the ambient scribe technology is using generative AI to listen to the patient visit. So. As the patient's in the room, as the physician's having the conversation with the patient, it's listening and it's capturing those details. And then it's the generative AI technology. Then drafts a note. In current state electronic health records. Physicians take an enormous amount of time. Writing notes, it's, it's, there's a huge administrative burden associated with it. Oftentimes, they're having to do it at night. They call this pajama time. And what these ambient scribe vendors have found is that by listening to the conversation with the patient, I. Using general to draft a note, which is then reviewed by the physician to, and then signed off. It decreases the amount of time and related to documentation burden, and it's really addressing burnout for physicians as well, because they're not having that huge administrative burden of manually writing out all these notes.

John Driscoll: 12:15

I think the other thing, Holly, that's, that, that's important is it returns us back to where a doctor can actually look into the eyes of a patient. That's right. And have some level of emo you know, it's where technology actually can improve the human interaction. Yeah. And, and it's to go,

Dr. Holly Urban: 12:31

yeah. We've seen that with, with session. Like saying, thank you for letting me like. Look eye to eye with my patient and not be sitting and looking at a computer while my patient's across the room. So it's very exciting.

John Driscoll: 12:42

Yeah. My, my, my 91-year-old mother always notices whether someone's making eye contact, and it's a huge, it's a huge difference to her. And after she's tr trying to figure out where people did their training, she really wants to know that they're actually connecting to her. I,

Dr. Holly Urban: 12:56

I think we all do as patients, we want our physicians to see us. And so if they're not. Being forced to type or work on an electronic health road, but can actually look at us and talk to us and not worry about that. Um, it's really, I think, very powerful for physicians and patients alike.

John Driscoll: 13:11

So you've got, you know, you, I think you've got alliances now with a bridge. Mm-hmm. Which is the epic is the dominant, um, academic health center, electronic health record. Um, Zuki has got, I, I think at least a, a lot of, uh, more, um, ambulatory

Dr. Holly Urban: 13:29

mm-hmm. Uh,

John Driscoll: 13:29

folks. And these are all technologies that provide that listening layer that can then translate it into a note that's connected. And I guess how does, how does. How does, how does UpToDate interact with that in an automated way?

Dr. Holly Urban: 13:44

Yeah, so it's, it's something that we, um, have been working on very closely with them because one of the things about these ambient scribes is there's a lot of them. So they're looking to say, how can we differentiate ourselves in the market? And one of the ways is by providing advanced clinical decision support within the workflow. So the ideal. The state, ideal state is that as the physician either reviews a chart prior to the visit, they would then tap into the UpToDate knowledge base, our very deep, very rich knowledge base, and we could say, Hey, maybe you need to this. I can see that the blood pressure's been high over the past three visits. Maybe you need to consider a second line therapy for blood pressure, and then here's some ones you might wanna suggest. So it's all about taking both the contextual information about the patient and then providing sort of guidance within the workflow to make sure that they're making the right decisions.

John Driscoll: 14:40

So I, is that something that interacts, were the agents interacting with the data and then providing the doctor effectively and a, a, a backup intelligent alternative?

Dr. Holly Urban: 14:51

Yep, that's exactly right. Yeah. And it, and it's grounded, right? It's all based, it's not the sort of the generalized knowledge that you're gonna find in a, in a standard LLM um, solution. It's all grounded on UpToDate so that you know where it's coming from. You know that, that. Trusted source that physicians have trusted for 30 years. Um, it's has a very, the basis and trust so that they can really believe in the recommendations.

John Driscoll: 15:17

Well, and I, and I think you raise a really important point that sometimes folks don't, don't necessarily don't tie into that the actual. Um, contextual sources for the large language models are not, are, are not always clear and in fact are almost impossible to back propagate because of the nature of the way deep neural networks, which are the, the, the reasoning, um, machinery inside the large language models. That's how they work. They take an enormous amount of information. Um, we, we assume they get it mostly right, but in healthcare it needs to be absolutely right. There's also a probabilistic element in the LLMs that you, that you, you really have to control for, because in healthcare you really want that next correct. Second or third line therapy tied to the facts of, of the, of the research and the work that we've done. And so I, I actually think, you know, you're, well, it's agentic at one level, Holly, what you've got with up to date is something that, you know, doctors can actually lean into. And that, that's why I think the, the power of what you're doing is that you can actually. Go right to the sources and doctors can then test whether the recommendations are things that they can, they, that are based on science and, and, and the best, best, best current practice, which is what I think is very powerful about Walters Klu being connected here. Now, how does, how, how does Walters Klu get paid? I mean, I know that the, for the ambient engines, you know, you, you, those are subscriptions, um mm-hmm. That are either bought through an ER, like. Connected to like a bridge in the case with, with mm-hmm. Epic or, uh, Suki on a one-off basis. Um, like how, how do you guys get compensated for being connected to this?

Dr. Holly Urban: 17:06

Well, we, you know, we have relationships with large health systems, um, that we call our enterprise customers. And so some of this were just. Doing to reinforce their existing subscriptions. Um, so to make sure that they're getting continuing as, as this technology evolves, as people's workflows adapt, especially as they're adopting, um, new solutions like Ambient, um, that they're able to leverage that, uh, clinical decision support that they're already paying for within the workflow. We're working out the, the very small details of the, of the licensing, but um, in general, we wanna make sure that they're able as they move forward, to continue to leverage. This information source.

John Driscoll: 17:46

So a, a hospital system would already have Walters core, effectively a subscription to UpToDate, um, to get access to the best clinical information. And then this would be, you know, this would be part of that in some way, shape, or fashion.

Dr. Holly Urban: 18:01

Correct.

John Driscoll: 18:03

And so is your goal to be connected to all of the ambient listening technologies out there?

Dr. Holly Urban: 18:08

Well, that's a good question. I don't, you know, we have announced partnerships and I think we really wanna make sure that, um, clinicians and their patients are getting the value from our current partnerships. Um, we are working on ways that we can, um. You know, create a solution that's more scalable so that we could work with multiple ambient vendors. Um, again, just to make sure that no matter where you are as a clinician, you're able to leverage that, um, important clinical decision support solution that's, that is up to date.

John Driscoll: 18:41

Where do you see, you know, the, the Agen ai, you know, you're, you're, you're a doctor. You've, you've practiced, you've, you've seen the, the pluses and the minuses of the integration or disintegration of technology in healthcare. Where do you think, we're stepping back a little bit, where do you, how, how important do you think these LLMs and healthcare AI agents are going to be to the actual practice of medicine going forward and.

Dr. Holly Urban: 19:07

So I, I think you said it earlier, I think that to some degrees we're too excited, but in other group degrees, maybe we're not excited enough. I mean, as just example, this whole discussion of the ambient scribes of the, of the notes technology, I mean, this is great, but it's really just fixing kind of. One point solution. You know, why, why are we not thinking more broadly about how we can use AI to improve the overall workflow? I mean, one of the reasons why physicians struggle so much writing notes is because there's such a heavy dependency on downstream, on collecting quality measures, on billing, on coding, um, on, you know, authorization requests. I mean, there's just such a, a, um. An administrative burden associated with these notes, and that's why there's such a struggle with them. So why wouldn't we use AI to try and improve some of these administrative burdens so that note doesn't become sort of this. You know, monolith of, of, of administration requirements. You know, if we can let AI listen to the patient doctor conversation and capture what happens, what if that all could just flow then to a bill and it doesn't, wouldn't require all these manual. Steps of a CDI solution and coding, et cetera. So when we're talking about really using technology to fix healthcare, why aren't we doing it in a way that would lessen the administrative burden and let clinicians focus on that patient and focus on their clinical tasks? I

John Driscoll: 20:36

mean that, that, that is the ultimate promise here.

Dr. Holly Urban: 20:40

Mm-hmm. That

John Driscoll: 20:40

doctors return to just being doctors and not, you know, the patients aren't, aren't, I mean, aren't, aren't looked at as a bill to be generated. Right. And data isn't there just to populate and then be argued about by the managed care folks. Yes. And the billing agents aren't on both sides, but I think that what you're doing is really a great first step. A, to improve clinical practice. Yes. And b, to potentially accelerate us to get to that point where, where, uh, AI is seen as a partner, not just a a, just a, a new or impediment to workflow. So, um, I'm really glad that you, we got a chance to talk about what you're doing right now to improve clinical workflow in the doc doctor's lives.

Dr. Holly Urban: 21:21

Well, great. Well, thank you so much for having me. I've enjoyed the conversation.

John Driscoll: 21:26

That's it for another episode of Care Talk. If you liked what you heard or you didn't, we'd love you to subscribe on your favorite service. And thank you, Holly, for joining us and we're excited about, uh, how, how all of us are gonna navigate this journey of AI to improve healthcare going forward.

Dr. Holly Urban: 21:41

Great. Thank you. It was, it was great to talk to you.