This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.

John Lee: Instead of doing the same with half the FTEs, you're doing a hundred times more with twice the FTEs.

Bill Russell: My name is Bill Russell. I'm a former health system, CIO, and creator of this Week Health, where our mission is to transform healthcare one connection at a time. Welcome to the 2 29 Podcast where we continue the conversations happening at our events with the leaders who are shaping healthcare.

Let's jump into today's conversation. All right, this is another 229 podcast, and today I'm joined by John Lee, practicing physician, also the founder, and probably lead consultant, if I thought about it, of HITP advisors. Where you it seems like you're really focused on Epic. Is, are you really focused on Epic for the most part?

Yeah,

John Lee: and it's I, I think I have some other competencies, but. My, My I know you have

John Lee: It's my easy place.

Bill Russell: Well, we ran into each other at UGM, and this whole episode is gonna be about UGM we're gonna talk about because you're one of the people who probably posts, I would say, if not the most, a fair amount about UGM and comments on other people's things.

So you've probably read as much as I have you have more access at UGM than I do as a practicing physician. I, of course, they give me a red badge. Right. So everyone knows I'm they know I'm media just by the red badge. And then I haven't escort the whole time I'm on site, which I Okay.

Perfectly understood.

John Lee: Did they wanna put it like a red a on your, I don't, just

Bill Russell: to, when they first gave it to me, I said, is this like the Scarlet Letter? And they looked at me like, that's not funny, man. I'm like, okay. No, actually they're great hosts. They give you somebody to be with you all day.

John Lee: Although it's interesting how it has transformed over the past 10 years. Oh, it's, I remember when I first started going there 10 years ago, it was me and my team and we go to the UGM and then we sit in our hotel and we think, oh, where are we gonna go to dinner to tonight? Maybe watch a movie in our room and now it's not quite HIMSS or ViVE, but it has a lot of those same vibes

Bill Russell: yeah I mean the number of people down in Madison was pretty significant, and that tends to be the place people go. And then invariably you have a dinner with some of your Epic people while you're there as well. I wanna talk to you about that. I mean, this was, I don't know if there was more announcements, but more announcements that people wanted to talk about than any year I can remember.

John Lee: Well, if you know anything about me and what intellectually interests me and what I post a lot about, you're not gonna be surprised that the thing that bugged my eyes out most was all the stuff about Cosmos. Yeah, I agree. I remember talking at UGM in 2022 about Cosmos and back then it was basically, essentially their, what they call their slice of dicey product.

And I talked a little bit about how. I envisioned where it would be going and they are definitely going in that direction and it's mind boggling to me how fast they got there. And it's interesting because last year I thought they were just slowly easing their way into it and then all of a sudden this year it just like hockey sticked up in terms of the things that they are envisioning how it will work.

All the data and what they showcased was pretty interesting. Essentially, almost a partner, if you will, for the clinician as they're doing things to say, Hey, consider this, and, oh, I've looked at this record, or I've looked at this additional information. Perhaps, this is something else you could look at.

I mean, are we looking at the future of medicine, essentially a trained physician or clinician and an AI partner?

And if you look at how we took knowledge in medicine, famously there were a bunch of studies that showed, it took about two decades between discover something and it actually becomes common practice in medicine.

Bill Russell: You weren't pulled enough, and I'm gonna push you to be a little bolder here. So your quote is, maybe we can reduce the adoption lag of new insights to practice to something shorter than two decades. You wanna predict on how much shorter than two decades we're gonna be able to get to, you know,

John Lee: hon?

Honestly I think ultimately where we're gonna get to with this is the limitation is going to be our own mental inertia. I could at least from a technical perspective, I could see something. Being surfaced up in Cosmos, new insight and then getting it into the system operationally in hours or days.

Bill Russell: there was also an announcement about their own foundation model, essentially on top of this data. Right. So developing essentially an LLM on top of this,

John Lee: so it's basically a transformation of that content of the raw atomic data and processing.

Uh,

Bill Russell: it's. Big data at the scale of a country, essentially.

So we used to be able to as a health system, look at bringing all that data together and getting insights and whatnot. But now you're looking at insights across, again, 300 million patients and how many different encounters. And the foundation model becomes a way to tap that Fairly easily, and I assume it's not that hard to layer in research studies, case studies, all sorts of other stuff on top of that. so one of the things that I talked

You did this query wrong or the way that you pulled the data was incorrect, or you forgot to take into consideration this other data feature. And then I ran the query again, and this is the sort of insight that I got. And so what you could then start doing is having this iterative back and forth so that if you think about it, then you can have a knowledge flywheel come up out of this That's

far more nimble than collect a bunch of data over three years. Curate that data, put some, wrap some statistical layers around it. Spend a few months writing stuff into a paper, sending it off to a publisher. Have them do some sort of peer review on it. Have them come back to you several months later.

But then if you have crowdsourcing and layered on top of that, you have all sorts of other mechanisms to. Correct. Refine and polish those sorts of insights and hopefully come up with something even faster and better than what we have right now.

Bill Russell: Again, you had more access, you talked to more people.

I'm curious you're talking about doing queries against the data, which is very deterministic. It's a deterministic way of, but we know that foundation models work a little differently in how they store and retrieve knowledge and whatnot.

And that's what leads to hallucinations and those kind of things. Was there a lot of talk around hallucinations?

Do we really want this? Right? And sometimes people would say both out of the same mouth.

Bill Russell: If anyone's proven that they're not, I mean, they're not an evil empire, it's Epic. I mean, they've been a really good partner to healthcare over the years. But I do understand where that talk comes from.

And, part of that is some of the other AI announcements that came out, and my gosh, you would think they had obliterated abridge and ambience in their business models and whatnot. And I talked to founders of both of those companies while I was up there. And they're like, look, nothing's changed.

they're gonna own the market essentially is right.

John Lee: So, so what in the run up to this, and I'm sure they did this leak on purpose there was this sense that, oh my God, Epic is gonna come up with their own ambient documentation that's gonna put everybody out of business. And what the truth was is that they were still going to do the same thing that they've always planned on doing wrapping

the transactional presentation layer of how to actually deliver healthcare and have some sort of AI or ambient technology underneath to help drive that. Basically what they said is that their favored engine in that EMR chassis is going to be Microsoft,

Bill Russell: right?

And so I'm not saying that it's gonna be completely equivalent, I fully expect that Microsoft is gonna have some most favored nation sort of, access to the content and the tools within Epic. But there is plenty of room for Suki, Ambience, and Abridge.

Bill Russell: I don't remember if it was last year or the year before, but when Satya got up on the stage, I remember leaning over to the person who was my escort at that point.

But we don't have the resources to partner with everybody at this level. It takes a lot of resources for them to work with them and to integrate it and whatnot. But I thought it it signaled, Hey, we are a partnering organization. And clearly for Microsoft, it meant more than that.

I mean, they were able to drive a relationship. I think that is a pretty enviable relationship in the industry.

John Lee: And I think just from my perspective, it's a defensive play both by Microsoft and by Epic against Oracle. Oh, interesting. because I have no direct knowledge of what exactly Oracle is doing, but just from the rhetoric that I hear and just from the WIS and communications that I have with people who are in the or Oracle space.

Like basically like what Epic did now what I, do they have the runway to do that? I don't know.

Bill Russell: but you do know. I mean, it's, I, so I was asked two days and I can do this because, essentially I don't get paid from anybody to do anything except give my opinions. But somebody asked me the other day, it's like, do you think Epic's being defensive against Oracle?

I'm like this battle is over because to switch from one to the other is way too prohibitive. And they just landed Intermountain, northwell and UPMC, I mean, these are the last of the holdouts of the big Cerner organizations. Yeah. And especially in a

John Lee: Intermountain. I mean, they were a development partner of the old Cerner.

And it's only like one or two people left. Are on Cerner. And to be honest with you, certain meetings, we have more Meditech than we have Cerner in the room. Because Meditech has at least a spot that Epic hasn't really gone after yet. Which is a fairly low cost complete system.

Although they're now doing a ERP and whatnot, which is one of the things that. You know that Meditech had going for it. It was a complete system from end to end. From that perspective.

I want to talk about the state of Washington. So that to me was another one of those announcements that I sort of sat back, well, you had the country of Singapore. I think there was another country and then there was the state of Washington, and I sort of sat back and I thought. This is really interesting.

It also addresses fraud, waste, and abuse because you have a lot more transparency and visibility into what's going on. Yeah. You can

John Lee: normalize a ton of data. Right. And

Bill Russell: it could potentially address opioid crisis as well. I mean, there's so many things I like after they announced that I was just sitting there going, that might've been the most compelling announcement to me, and I could see other states going in that direction.

That's where my entire career is around, and I feel the most comfortable in that system. But I really do hope that Meditech, Oracle, they ha still have a substantial presence because we can't just have one.

Bill Russell: I'm with you on that. And I was asked this week, why don't they face more monopoly kind of things.

Why don't they face more of that rhetoric? I said, well, there's two reasons. I'm gonna give you two reasons, but I haven't really thought about it, and I haven't really thought about it much since then. But the two reasons I gave were, number one, they haven't bought anybody. They like grew it organically, like every other monopoly you know of they just bought up the competitors, right?

Like by a long shot. Like, we like dealing with them They're not perfect and, every one of the leaders I work with will tell you where they're not perfect, but they'll say, look, they're the most fair with pricing. They could up their price, they could double their pricing and no one's going anywhere, but they don't do that.

They listen, like when they roll something out and say, Hey, this is what we're gonna charge for ai, and people give them feedback, they go, oh, okay. We see that, we understand that. And they adjust and they listen. They have a good relationship with their clients and this is not an exaggeration to say on a scale of one to 100.

Like Epic is in like 90 plus in that scale of valuable partners and partners they like dealing with. The next closest one is down at like 70. It's just a different dynamic that they've been able to create. So they haven't bought the market and they have nothing but fairly happy clients.

John Lee: but their competitors

Bill Russell: might stop.

John Lee: Well, sure. I mean, and I think, well, no but to, to that point, for instance, at least from what I can tell Oracle probably has some outsized political clout and I think they can probably, if it gets to a certain point, they can start.

Turning some political thumbs hoops, I don't know anything about any inside baseball. I don't. It is just me as an outsider saying this,

Bill Russell: J it is just, it's just two people talking on a Friday afternoon. They do have a lot of government contracts and Epic has, as far as I know, they have, they may have state contracts, but I don't think they have any, no, no federal contracts.

Bill Russell: I agree.

John Lee: And then be able to say, look, if you break this up. This solution that we have for you that is that is saving you billions, if not maybe even trillions of dollars. It's gonna break.

Bill Russell: Yeah. So I want to go down there's a couple paths. I want to close this by asking you all you've walked away from UGM before as a clinical leader.

How do you make sense of all this stuff? How do you determine where you're gonna go? But before I get there, I want to talk about Jevons Paradox. So you have a great line and I used your line in one of my posts because it is such a great line. CFOs are sort of drooling over ai.

They're looking at it going, oh my gosh, I'm gonna be able to cut two thirds of my staff out of rev cycle I remember the first one because it's so funny. It's like the first thing you need to do is get a soul transplant or something to that effect.

What was the second one?

What I would say is actually increase and then take this and make it way higher. If you have not been actually involved in healthcare IT, you don't truly understand that there's probably 80, 90% of the stuff that we want to do that we can't do, right. And that we know can make a huge difference, but we just don't have the bandwidth.

The resources, but the resources end up going up because people see so much more value out of it that they just funnel more, even more resources so that instead of doing the same with half the FTEs, you're doing a hundred times more with twice the FTEs.

Bill Russell: Right? We have a big barrier between here and there, and it's trust, right? So it's clinician trust of AI models. So I've interviewed just a whole bunch of people UCSD, Stanford and others about adoption rates of these AI models. I trust them in Rev RevCycle. I trust, and even in it, I mean, I would be deploying AI all over the place.

John Lee: We are definitely at that point because if you look at some of the generated summaries, that is just enormous. Yeah. Because the time and effort it takes to conceptualize all the stuff that happens during, say, a hospitalization between visits. Come up with some sort of very succinct messaging to the end user, be it a physician or a nurse or whomever, make it so that they can look at one document and get 90% plus of what they need just from that synthesis.

Yeah. Is huge.

Bill Russell: I remember talking to Andrew Rosenberg when the large language models were just coming on the scene. I said, what do you think the killer application is? He said complex patient chart summaries.

John Lee: Yeah.

John Lee: And it's not just the actual effort. It's the emotional angst of Oh God,

Bill Russell: yeah.

John Lee: And now it's just you press a button and it's there.

Maybe have to polish it up a little bit, but it's fantastic. touched on it a bit and I don't think it, it has gotten the hype that it I actually deserves. I personally think the killer app feature of Cosmos is the cosmos, median length of stay,

Bill Russell: All right, so, why, I'm trying to think of the easiest question to get more out of you on that.

John Lee: So right now what ends up happening just at a high level, is that patient comes in, say for pneumonia, you treat them, you send them home. And then let's say there's a whole bunch of other similar pneumonia patients.

Too bad. You suck. You only get two days of reimbursement. so the problem with trying to fix that, to try to fix the processes is that. Instead of getting that length of stay report at, after all the coding is done, you want it early on, right? So that you can actually then adjust some of your processes, operational processes, and if you talk to patient flow people, that has been the Nirvana state and what they've been wanting for like literally decades.

Marker

John Lee: Then you don't have to accept it. You can put in some other number and say you know what? Cosmos, median length stays patient is gonna be three days, but I know that the patient is actually gonna stay for longer. So you can put that in and then your patient flow. Teams can adjust based on that, but I kid you not when I talk to some my inpatient colleagues about this when the providers are forced to put in a length of stay there are

docs that I know who will just passive aggressively put in a length of stay of one day for every single patient that they have. There are others, someone told me that there's somebody he knows who puts in a year length of stay for every single patient that they have. So how do you, how can you operationalize any sort of patient throughput based on that?

So you can look at that and go, oh. Oh, dang, I should have ordered that test or I should have done, because you're gonna have that information at that moment. You had a bridge actually saying real time prior auths, it was their announcement just prior, and so they, they're doing that with Highmark and they're in negotiations with a couple other national players to do real time

prior authorizations. Some of the things that caused significant inefficiencies in the process because we didn't have the information until much farther down. The road are now being moved up to a point where you can actually do something about them.

John Lee: Yeah. So imagine if some future state where you feel confident that the prior law information surrounding the order was accurate.

Why don't you just walk over to the MRI and get your MRI done right now? That's the sort of transformative workflow that I think will start occurring.

Bill Russell: So. Adoption is so key. maybe this is the norm. I'm not used to it. They announced so many new things, so many new features so many AI features alone.

How do you make sense of that? Is, does anything really change or do you still go back to your health system and say, Hey, this stuff was announced, but we're still working this set of problems that we've been working and we're just going to apply these tools to the same problem sets.

It's basically like flying a log switch. You don't have to configure a whole bunch of stuff. You don't have to do a lot of validation. It just works.

Bill Russell: be careful turning it on, because now some of those measurements are based on adoption, not just turning on features.

John Lee: True. And I've, I have my issues with their whole Gold Stars and honor roll it to me starts smelling a little bit like the the Joint commission inspector coming into the hospital and then you cleaning everything up and then everything's pristine or leapfrog coming in and you're turning on all of your medication alerts that everybody ignored.

organizations in trying to accomplish those sorts of things. Yeah. They fully view the, gold stars and honorable as tools to help nudge waiver organizations to hey, help us help you. But ultimately, especially if you're a high performing organization, they tend to begin, I think.

Less relevant.

Bill Russell: How do you make sense of this for the organization that you are sitting in front of?

Versus the other stuff.

Bill Russell: Yeah. I agree. I didn't articulate it well, but that's what I mean by nothing's really changed. I mean, you're still sitting in the executive meeting and they're saying, look, we're at 1% margins. We've got these kinds of pressures. Our quality scores are down in this area.

We have, whatever it happens to be. And we're still solving those things. We just happened to turn over here and all of a sudden now our toolkit not only has screwdrivers and hammers, but it also has lasers and all sorts of other things.

True. And we're going,

These are all the other qualitative stresses on our organization. Spit up the top three things that we need to do over the next quarter and give us some metrics on how to accomplish that.

Bill Russell: I was talking to a former CEO of a large health system this week. And I took the nine 90 from that person's organization for a year and a half ago.

Yeah. And I put it into chat, GPT. And I said identify the stressors on the organization. And it did. And I said, what would be your path forward. And because part of the thing is I was talking to this person about how AI is gonna fundamentally change healthcare. And I said okay, these are the five things that chat GPT recommended.

So what you're describing is very doable.

John Lee: Yeah. I think you just have to figure out how to get those inputs and how to structure. The organizations will interact with it, and quite frankly, if I was a principal at McKinsey or Accenture or any of these other consulting firms, I would be really frightened.

Bill Russell: Yes.

John Lee: And I would be even more frightened if I was someone very junior in those organizations.

Bill Russell: Yes. Oh yeah. No I've paid for those teams of juniors to come out to my site and $300

John Lee: an hour plus. Oh, yeah. To come up with a PowerPoint for you.

And having that kind of knowledge set, I think is distinct because I actually, I think the thing that makes it distinct is. It's about moving people in an organization. It's about educating people. It's about changing hearts and minds and all that stuff. I'm not sure the large language models are changing hearts and minds yet.

John Lee: The other thing that I think is my secret superpower is these large language models are really good at especially within certain domains. Coming up with insights, but connecting things across domains is still a bit of a vulnerability for them.

Kind of the standard recommendation.

Bill Russell: John, I appreciate you coming on the show. I look forward to I don't know, maybe we'll do this again next year. We'll see yeah,

John Lee: absolutely.

Bill Russell: We'll see what we're both doing around that time. I

John Lee: think those UGM conferences are just going to be more and more like what we just saw.

Bill Russell: Yeah there's a lot to digest, A lot to talk about.

John Lee: Yep.

Bill Russell: Thank you for your time. If people want to get a hold of you, HITpeakAdvisors .com. All right. Thanks John. Take care. Thanks. Bye. Thanks for listening to the 2 29 podcast. The best conversations don't end when the event does. They continue here with our community of healthcare leaders. Join us by subscribing at this week health.com/subscribe.