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

The 229 Podcast: Exploring How Epic Cosmos is Changing Care with Phil Lindemann

Bill Russell: - [00:00:00] Today on the 2 29 podcast.

Phil Lindemann: how can I instantly learn from what the million other physicians have done in similar situations and get the insights on that, right, in my workflow.

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. It's the 2 29 podcast and today we are joined by Phil Lindeman, VP of Data and Research at Epic. And recently returning from the Packers Second Victory up at Lambo Field. Congratulations on that victory, by the way.

Phil Lindemann: Thank you. Hard earned. Yes.

Bill Russell: so, which was better The, UGM hearing all the stuff that Cosmos is doing, which is something that you're intimately involved in or the Lambo victory. Is it hard to [00:01:00] really compare the two?

Phil Lindemann: is pretty hard. Like, I got to bring my daughter who. knew very little about football, but I was like, this is an experience that's like going to a rock concert.

And she, I, she loved it. It was great. All like, Vince barely slept. It was awesome.

Bill Russell: Well, how old is she, you think? You

Phil Lindemann: think she's she's 12 and I'm not gonna say like. The rules of football were, we were pretty rudimentary when we were when we got there. There's offense, there's defense.

We started there,

Bill Russell: but it's yeah, but it's every father's dream to, you know, be going to games for the next let's use 12, you know. She comes home from college and says, dad, can we go to a game that's, gonna get in trouble for that, by the way. because one of my daughters actually is my producer.

So she's gonna listen to that and go, I didn't know you wanted me to go to a football game with you and anyway, get in trouble. So, Phil, VP of Data and Research. Leads, epic's Cosmos platform, 250 million patient records, and spearheading life sciences initiatives for clinical trial innovation. You've been [00:02:00] at Epic for 21 years.

When you started, did you start right outta college? Oh,

Phil Lindemann: yes. I had never heard of the place. They hired me and I thought, this is how old this was. I thought I was gonna take CDs and put them in computers at a hospital, and I was gonna go from computer to computer and just run and maybe set up a couple settings.

And I was like, I could do this

Bill Russell: what does the progression look like at Epic? We are gonna, we're gonna dive into Cosmos in a deep way. Sure. But what does the progression look like? You start as a college student and you know. Very little. I mean, as a college graduate, you're okay, I know a computer, I have some skills.

And Epic sort of takes you under their wing and says, all right, you're gonna, you're gonna learn healthcare a little bit at a time here.

Phil Lindemann: I got lucky. I had the right mentors at Epic who gave me opportunities right away. I don't know if I was bored and they said I'll give him this challenge, but one of the first projects they put me on was at a big health system in Minneapolis and they said, oh, we [00:03:00] need a fresh set of eyes.

And it turns out it was the largest go live of multiple products or different departmental things that had ever gone live in history. And to me, I came in here thinking, ah, they've got this all figured out. It's software. There's no bugs. Everything must work perfectly together. So it was like this really rude awakening, but it also was an endless amount of problems to solve.

And to solve the problem you first had to understand the clinical context or the workflow behind it, and I just, I loved it So. Kind of things took off from there.

Bill Russell: the transition to analytics. So you work in a lot of different let's just call them modules within the, right. The, The hospital system.

And then you you kept complaining about the data and you got that phone call from Judy talk, talk a little bit about that.

Phil Lindemann: Yeah, I so that first exposure you got to. Sort of see how the obstetric department worked in the ICU department and the OR and the lab and how all these things fit together or rather [00:04:00] didn't fit together.

And as I went from module to module, as you're saying, I said, why is it so hard to get a report? Like I just want a report that tells me the most of this thing happened. It's still like seemingly simple things, and it would be really difficult for me. And then one day. I got this call and says, Hey, you're gonna go over to analytics and you're gonna figure out who was reporting.

It was called at that time. And then I was like, oh my gosh, this is so difficult. This is so hard. And it wasn't a technical thing necessarily. Some was technical, some was just administrative. So it. It really, I think I have this unique perspective of, I know how all these products feel about like you have to make the software work for your users and that's like top priority.

And then the analytics team is sitting there saying, we're gonna build a platform so we can build all these reports for everybody. And I can sort of see the pain of both of those worlds and help make them work together better. And. I think that will probably never be done, Lisa. It's like flossing your teeth.

Bill Russell: So it's a never ending process. Yes,

Phil Lindemann: there's, it is a constant process improvement [00:05:00] of how do you get analytics in the hands of every person who needs them and make sure they understand what they're asking for.

Bill Russell: Former CIOI was never an epic customer but I will tell you, doing the best of breed thing, data and analytics.

It's like bringing that data, build out a clinically integrated network, and I'd look at them and be like. There's 45 different EHRs, like, yeah. And we're gonna measure the clinically integrated network on this standard set of, you know, things. I'm like, yeah, but we don't collect the data the same way.

Like you have to start at such a rudimentary level, like we're gonna, we're gonna collect the data this way. And then you look at the data structure and how EHR stores the data and it gets. Messy, and then you have to bring it in, and then you have to clean it up, and then you have to I, this sort of leads me to Cosmos and saying, yeah this was, no, I mean, people just keep throwing this number around like it's no big deal.

Like, Hey, you know, 250 million patient records. Well, it's one thing to have them, it's another thing to get them into a form and a structure that you can actually do something with it. I'm not sure what the question is, like how did we get [00:06:00] from there to here? I mean, how did we get from 21 years ago to where we're today?

Where I'm watching at UGM, where the doctors essentially using Cosmos as a as a tool, as a almost another set of eyes on some of the work that they're doing.

Phil Lindemann: Yeah, it's how can I instantly learn from what the million other physicians have done in similar situations and get the insights on that, right, in my workflow.

So it's it's like the point, it's what we've always dreamed that this would happen. So even when I started 21 years ago, you know, this idea that we could learn from each other if all of this data could be uniform and mapped together. So I think. You're right. It's just there was a lot of hard work that led to this, and now you're sort of seeing the fruits of it and it's it's assumed that it was obvious like this.

This was just going to happen. It was all the same and it was and still is. There are really painful things about how to work through this. There's like [00:07:00] people governance, there's the technical attributes of it. It's. Working with each health system and getting them to agree on a single set of definitions when they've done something a different way for 30 years.

I mean, it's all the same stuff that you'd see at any analytics discussion. Like I've been at industry conferences and like, you know, capital One and Delta Airlines have the same problems. Like it's just there. There is a similar theme to how these things are solved here. So yeah, building Cosmos was not an overnight process.

There were. There were even little attempts, there were little like practice projects that we had done with the community. And then about, about the time I moved to analytics is when we sort of said, okay, this has to happen and build the infrastructure. That's like a year and a half. And then about 2019 is when the first groups started joining and taking off from there.

And,

Bill Russell: and so, when a new health system, because you're still signing on health systems that are saying, okay, we're ready to. To participate with Cosmos. I mean, does that, when you see, hey, somebody just signed on the dotted line, does that still represent a significant [00:08:00] amount of work?

I think some people assume that it's just, oh, yeah, Cosmos, click the button and boom, all the data's in there, and now we're participating.

Phil Lindemann: Yeah, well, you kind of have to think about the work that was done leading up to it. It's like everyone has stood on the shoulders of the previous person, so now we're all on the hundredth floor thinking, oh, this is easy.

But it was like foundationally, a whole bunch of stuff had to be built before, but like if a new group signs up and we do them in waves, there's like four waves a year where. A dozen groups will go live together and there's an eight week onboarding. Doesn't take that much time and they map a couple things, but it's, it is relatively straightforward for a new group to add now.

But you know, long time ago it took longer. Because nothing was mapped, there weren't a lot of standards. So like there's a lot of standard ways to map data that didn't exist in those times when you were trying to get 45 oh EMRs talking like I think standards have improved. Our ability to move data has improved our we have a [00:09:00] foundational system that all groups start off with and you know, that's been available for over a decade.

So, you know, we've got a decade of organizations that have sort of started cleanly and you know, some of the older ones we work through that, but it. It's, it is it is a challenge.

Bill Russell: So I'm sitting at UGM I'm watching. Yeah. You know, you guys do the, I don't know what you call that, sort of the role playing it's called

Phil Lindemann: The skit.

Yeah, the skit. It's a fun thing.

Bill Russell: So you're doing the skit and I'm watching as the doctors having a conversation and Cosmos is essentially saying, Hey, consider this. Look at this or those kinds of things. Which is really the promise of ai. AI is the assistant that sits there and says, Hey I've looked through the clinical documentation, I've looked through other patients, I've looked through all these things, and responds back and this is it feels like we're just scratching the surface.

But I was blown away. And I talked to two doctors who were in the room once A CIO, who's a doctor, and another was a CMIO and I said, you know, I just wanna check what I feel like I just saw, and I [00:10:00] realize adoption's gonna be a big thing here, but Yeah. It feels to me like as a patient, I would want a doctor who's using something like this that has 250 million patients, that gets narrowed down to the patients who are like me and looks at outcomes of that patient group and says, yeah, hey, you know, you may wanna also order this test.

And they both looked at me and said, no, you're right. That's. That's a, an amazingly powerful tool. And you're right, adoption is gonna be, is gonna be the challenge, but just to have that tool is very powerful from their perspective.

Phil Lindemann: Yeah. John Lee, I don't know if you know him, he's a physician.

I'm sure you've run across him, but you know, he, this was years ago, he said, you know, Phil, someday. Patients are gonna start asking. He's like, they already asked this question. Do they have MyChart? Does my doctor have MyChart? And he said, then they're gonna start asking, does my doctor have Cosmos? And you're right about that.

So the beauty of the function that you were seeing there, it's called Best care Choices. [00:11:00] And what it does is it essentially uses Cosmos. So if I'm in the office and you're the doctor. It will use Cosmos and automatically build this precision cohort of patients that look like me and then say, okay, well of all the other patients who had high blood pressure, who the first drug that they tried did not work, these are the drugs that were tried second, that worked the best, they actually had the best performance.

And you can even, you know, we have this sort of. Moment where the physician can share that right in the exam room with the patient and explain it to them and say, you know, we have a couple options for medications here. And really the physician said, you know, what's cool about this is, you know, patients ask you all the time, well, what worked for patients like me?

Like, what do you think will work? And now it's getting at the point where you could say, well, it worked for patients like me.

Bill Russell: And it's not just that physician's experience, it's the experience of physicians across the entire network. Right. I think from an a data standpoint, I think [00:12:00] is this based on outcomes, right?

So you have the medical record. Is it based on outcomes? Like, do we have that kind of data in Cosmos where it's saying, look, this was effective?

Phil Lindemann: That's the whole point, is. How do you say something is better than something else if you can't measure it? So when we look at something like high blood pressure, we obviously measure blood pressure.

It's a pretty common variable, but really the reason why you don't want to have high blood pressure is so that you don't have a heart attack or you don't have a stroke. So those are the endpoints that we're looking at. And because Cosmos has these longitudinal records, in some cases, over 20 years of longitudinal data on a single patient.

We apply a whole series of data quality checks, so before that screen that you saw at UGM, before that screen even shows to a doctor a thousand quality checks, literally a thousand plus data quality checks passed through to say. Ensure that we had at least three years of readings on that patient. So when we do a three year heart attack risk, it's truly looking through for three years that they [00:13:00] did show up in the office at year three and they haven't had a heart attack yet.

Well, we know, okay, that one, there's no heart attack. So that's the types of things that we have to think through that are sort of like under the covers, but to the physician, ideally, it's a very clean, straightforward experience That's.\ Essentially crowdsourced by their peers and saying, this is what has worked.

It's not what was the most popular drug at the time, or which one was ordered the most? It's outcomes driven. It is actually looking at which drugs work the best which interventions or treatments work the best independent of how many times it was ordered or how many times it was given. then because of the size of Cosmos, which I should, it grew a little bit.

It's 300 million not two 50, but whatever. Well,

Bill Russell: and by the time this airs, it could be up to three 20. Sure.

Phil Lindemann: But anyway, so it's using that as the cohort to it.

Bill Russell: we talk about AI a lot and Cosmos sort of came to be, in the consciousness around this time of generative ai, but it, as I hear [00:14:00] you describe it, it represents an awful lot of different AI technologies, I would imagine.

Phil Lindemann: I think of Cosmos almost as, it's almost like a living entity. It is the community that participates in it. The thousands of researchers that have access to Cosmos, and then it's this whole set of tools. That we build on top of it that essentially poke into the medical record.

So the one I just talked about best care choices is one we, something very simple. We do growth charts. So, you know, you look at your kids' growth charts, well, there are children with rare diseases. That affect their growth. And oftentimes there is not enough data or the growth chart is very old that exists.

We can use Cosmos to generate a growth chart for that rare disease cohort. And we've got a handful that are now available where if you were looking at a CDC growth chart, the child might be not even on the growth. But now you can actually look and say, for my patients like me, this is how I'm doing. So these things don't have to be [00:15:00] world changing, but they're just going to start showing up in doctors and patients' experiences where everyone is learning from.

What happened to patients like me and sort of infusing that throughout the medical record. So when we designed Cosmos, that was really the North Star is how can we box up the learnings, the insights, and then just make that integrated into the fabric of Epic so that every doctor can benefit researchers benefit.

I love to talk about the research side of it, but. The focus is how do we make the physician and patient experience better, which I think you saw a lot of at UGM on stage.

Bill Russell: I was gonna go into the clinical trials and the research perspective. Talk about how it's being utilized, like what types of organizations are utilizing it, what kind of research are they doing and what kind of outcomes are we seeing?

Phil Lindemann: The scale of research that is possible 300 million patients. One of the first ones that a lot of groups look at is rare disease. Basically, if you want to in a specific domain, if you read a research paper, it always says the [00:16:00] estimated prevalence of this disease is x. With Cosmos, you are at a place where you could say, this is the prevalence.

So it, it's rewriting the first line in a lot of research for every one of these rare diseases that someone wants to do. So those are kind of quick wins. You could go and do a paper on that. But right now there is. Something like 2000 plus researchers that have access to Cosmos all around the world.

500 of them are data scientists, so they're actually writing code in Cosmos and having access to that. And right now we're seeing about two peer reviewed journals coming outta Cosmos a week that the community is building. So those are other health systems that are in there actively doing research.

And one of the neat things that I like the most about it is. Cosmos isn't just what meds did you have and what diagnoses? There's social drivers, data, how much do you drink? Do you smoke? Do you have transportation issues other information that gives us a more holistic picture of a patient's health.

When you think about all the things that can [00:17:00] influence it. So you're seeing a lot of people study those types of things because they're historically. Has never been a database that could tie those two things together, those two worlds. And we make it, in some cases fairly trivial to go in there in a few minutes and have an understanding of how a population looks.

So I would say research, I don't wanna say it's booming, but it is, it's like the rocket ship has launched in the last year as we're watching what the community's doing. And it's just great to see because it, it was a trickle at first, but now. Community has published more journal articles this year than they have every year combined of Cosmos.

And that was by August 1st, so that was a little bit of delayed data. So the research community is really strong in Cosmos and just getting stronger.

Bill Russell: I mean, do you guys bring that community in and of itself together, sort of like you bring

Phil Lindemann: the providers? We do. We there's two ways we do it.

There's one at Epic and there's one where we go to the customer. So we have a symposium here where we bring [00:18:00] more of the, kind of the hardcore data scientists and the people who really want to dig in. And we have a discussion about how we can improve. because we treat the research aspect of it as a product.

What statistical libraries do you need? Do you need more GPUs? What types of things do you wanna do? But then the most fun is, and this is a core tenet of Epic, is if you wanna learn. How your users are gonna use your software. You gotta go observe them. You have to go immerse yourself in their environment.

And in research, we need to go immerse ourselves with researchers. So we have data hons all over the country where our research and development teams, our researchers, will go out. And hang out, you know, in a large cafeteria or an auditorium. And everyone will have their laptops and have their they're logged into Cosmos and they're just ripping through queries and trying to get ideas for papers.

So those datat thoughts are something that we do a lot of to really spark interest and expose people to what's possible because you can really expedite the amount of research you're doing versus sort of saying, well, first let's build a database, then let's [00:19:00] collect data from all these sites, and then let's apply data quality rules.

So make sure it's all mapped. Let's make sure we have access provisioned. Essentially with Cosmos, they just say, I want that person to have access. They're eligible for X, Y, and Z, and away they go and they've, they're off dipping into a large data set with a whole set of statistical libraries to apply to it.

So it's, it's something we've all wanted for a really long time.

Bill Russell: I mean, it would be Christmas for a data scientist. I mean, exactly the first time they get in there, they're just like, oh my gosh. Look at I've, right. I have access to. Is there a data set that you sort of covet that you look at and say, man, if we could bring that data set into this, or do people like just take Cosmos as one of the data sets and then go out and get a whole bunch of others?

Phil Lindemann: Well, so, because Cosmos is de-identified. You can't bring more data into it, sort of, you can't side load data because it would redo the de-identification algorithm and we can't do that. But what's important to know is, you were talking, you know, in the old days, how did we bring all these systems together?

There [00:20:00] is one data model, code base deployment of Epic for the entire world. And, you know, everyone might be on a slightly different upgrade cycle within a month or a year, but. Rip off the cover and the tables and the columns are identical and there's 170,000 plus data elements. That are eligible to be brought into Cosmos.

So that's the data set that I want is we've only brought a portion of that together, a high value portion, but there is so much more work to do in the depths of very specific subspecialties and different social driver data that we can continue to bring in. So. There's a never ending stream of updated data types that are getting into Cosmos.

And sometimes we do fun things like

Bill Russell: you just wanna get all the epic data

Phil Lindemann: into Cosmos. Well, let's start with that. That is the easiest one to normalize and bring together.

Bill Russell: I was kind of surprised that the social determinants data, but if I thought about it, health systems have been collecting this for quite some time.

I just

Phil Lindemann: like a decade. Yeah. [00:21:00] And it's not. It is not evenly recorded, like some groups do a really good job of it and see it as a core tenant. Others are saying, yeah, we have it a little bit. But that's kind of one of those things where it's like 10 years ago we had to have the foresight to say, if you want to do a predictive model.

Off social driver data, you have to have social driver data. So we gotta start collecting it now and we have to make sure it's embedded into Epic in a meaningful way that everyone's not sort of just inventing this themselves. So we looked at what was then Institute of Medicine's. 10 sort of drivers of health, social drivers of health, and then modeled it after that.

And we've since expanded it. But it meant that everyone in the world who wanted to collect this data was collecting it in the same format, mapped to SNOMED concepts, if that, you know, for a couple listeners who are gonna speak ontologies. But that was a core part of it. And then monitoring that, it's being used, getting implementation programs to say, how do you get to talk to a patient about these things?

Making sure it's available in MyChart. So it's all hard work. Like at [00:22:00] the end of the day, unfortunately, there was no like, magic, like, we're gonna build it and this'll, this'll work. But it was, you know, in some cases it's quite a slog to get this stuff going, but as you can see, it's starting to pay off and people are seeing the benefit of it.

Bill Russell: Yeah. I was listening to the acquired pod podcasts on Epic, which has made its round. I've talked to a lot of people that have listened to it And somebody from outside the industry was saying, you know, what's the most impressive thing about Epic? And I said, you know, people are gonna talk about a lot of different things, but for me it's one thing that's been applied differently over the years and that one thing is it used to be really hard to get people within the health system to agree on things.

And Epic came in with a very prescriptive model and said, Hey, this is what it takes to do a successful EHR implementation. And they got everyone sort of working together. And while they did that in a hospital, which as a former CIO, I know how hard that is. Yeah. Especially as a third party, like as a third party, they were able to.

Help that to happen. And I said, you know, I think the most impressive thing is they've scaled that they now do that as an [00:23:00] industry. It's like, as an industry, sometimes we struggle to, oh, should we share data? Should we not share data? What data should we share? And you guys have created these things all along the way that's like, Hey, look, you don't have to share your data.

We're not gonna force you to share your data, but if you do share your data, your doctors get access to this and your researchers get access to this. And it incents the behavior that as a patient we all want. And I don't know, you can't really comment on that, but I just sort of wanna throw that out.

I mean, it's just one of the, one of the impressive things. I mean, Cosmos is very hard to do as a single system. And it takes it takes someone at that at Core. I want to talk to you about ai so, a lot's been said about AI and Epic and UGM, the awful lot of announcements about AI and there was like 160 some odd things last year.

And so it's just it feels like AI has taken hold and that we see a future where AI is [00:24:00] really going to be. Involved, if not a significant presence in the future of care. I'm curious what that looks like or how you see that playing out. What does it feel like to be a patient?

What does it feel like to be a practicing clinician in the next couple of years with AI sort of coming along?

Phil Lindemann: there's a lot to, to unwrap there. You know, I think as we're looking at how do we implement AI around the company, part of it is getting people to rethink how they approach a problem. There were things that were just, I don't wanna say they're too hard, but they were not worth the squeeze to optimize versus just.

Saying We're gonna work on something else because we can ensure the outcome we want. With ai, we can implement some things more quickly and automate steps of a process that you wouldn't have been able to if then else code it. And it's getting people to look at their project lists and reprioritize [00:25:00] where things are now possible that worked before and that's kind of, so it doesn't really matter any technology you're going to do that.

Like when I was talking about the growth charts before in Cosmos. That was something somebody asked for 10 years ago and said, oh, if we could just bring data together from our whole community, we could build growth charts for rare disease. And then literally like someone just forwards me this thing from 10 years ago and says, Hey, can we do this now?

And I was like yes we can. And sent it off to our RD team and in six months they had the thing in production. So. It's some of those things where we have to reevaluate where we can have the biggest impact. Now that we know AI is one of these tools in our tool belt.

Bill Russell: Are your coders, I would assume are using AI pretty extensively at this point.

They're not hand coding as much as they used to. Oh,

Phil Lindemann: absolutely. Yes. Yes, it is essential if you want to,

Bill Russell: so when we see more things being released as the, is the velocity higher?

Phil Lindemann: I think any one of our customers would say the velocity is breakneck right now. Yes. Things are coming out at a [00:26:00] rate faster than they can be implemented.

So we have to sort of take an approach of can we get it as a feature that can be turned on with zero implementation, where it's just obvious like, okay, that's how that should have worked or is it worth the cost of what it's going to? The cost of what it's gonna take to implement this. So some things you're going to have to explain to a doctor, or they're going to have to, you know, do a little bit of education on it, or it's something that they're only gonna use in certain situations.

So we have to be thoughtful about when we roll something out, what is an obvious win? Like, this is just gonna be a great thing. It's gonna work every time. And that's a home run, that's, let's do that versus things where it's like, well that'll be kind of interesting in a couple situations, but.

You know, maybe we don't spend our time on that.

Bill Russell: What about your old stomping grounds? The analytics side? I mean, is AI transforming how we interact with the data?

Phil Lindemann: Ooh, I am. I don't wanna say mixed opinions because I am 100% onboard that. AI [00:27:00] will be the way that analytics are written. It is.

It is. Like, I think SQL Report writing in a few years I think will be sort of this you know, classic cathartic experiment that people do for fun. Like I,

Bill Russell: so you think we're gonna be looking at the enterprise, the Star Trek, like talk to the computer and it's going to write the SQL queries, write the R, whatever, and it's all of a sudden it's just gonna pop back some data for us.

Phil Lindemann: That. So yes, it will get there. What the timeline is, I think is what everyone will debate. The problem is literacy still exists, right? That is the big issue, is like we have tools. Right now we have this one called Sidekick, and Sidekick is a way to ask questions of the data. Not as a lay person, but like basically as a departmental expert.

Like you know the language and you know how to ask for things. And depending on how you ask the question, it might give you a completely different answer, a correct answer based on what you asked. But did you really know [00:28:00] what you asked for? So I think literacy is this thing that we need to either encode in the system or figure out.

How we can make people more literate about how data's going to give you answers. It's like when you ask a report writer for something they know, like what you're kind of getting at. How do we turn that into something that can be codified? And until we get there, there's no magic box that's spitting out answers.

The literacy gap is too big. So it's interesting because we have made tools that are for end users, like, who don't have a lot of data literacy that are sort of almost having queries written for them. And what I'm finding is. That's good. It'll work for something like show me the top 10 times things this happen.

It's pretty solid at that. But when you start to get into more complex things where the user might not understand like how the underlying data exists, that tool ends up being more powerful in the hands of an analytics person. So the analytics person isn't going anywhere, but they might be able to [00:29:00] get more done with the AI tool, which is sort of a common thing that you would hear.

But that's where I'm, there's like this duality of, we're always trying in analytics. To push things from the analytics department to the frontline, like ideally, an end user who has the question can just get the answer with no interventions in between. I don't see that. As it's gonna change overnight until we close that literacy gap.

Bill Russell: I was talking to a CIO I'm sort of cheating here a little bit, but I'm bringing, like, the conversations I've had just this week I was talking to a CIO and he was saying, you know, they're doing really cool things with AI for the for the clinicians. They're doing really cool things for the patient.

I mean, there's some good announcements there. On the analytics side and the researchers, he's like, you know, but I've got this team that's just inundated with stuff. I'm like, can they make my analysts and my other people, my builders, what can they give them a set of tools that's gonna make them more effective?

because they're kind of overwhelmed with the velocity at this point. I'm curious, you know, at what point are you looking at that group [00:30:00] saying, Hey, we can, we can really help them.

Phil Lindemann: yeah, nonstop. There's an entire team working on those types of initiatives. So there is you know, in, in analytics, when you think about how people do things, they say, well, is there a report for this?

They kind of do a search, right? Is there a report? And if they don't find a report, well, can I build a report with like a report building tool? Okay, I can't build a report. Can someone code me a report? So there's like this continuum of complexity and we've put an agent on top of that essentially you can ask a question to and it'll say, oh, is there a report already made?

Can I make one? Or can I go write one for this user? And building that tool is something that we think will supercharge analytics teams who we're gonna go through that process anyway and try and do that. So that's. On the analytics front for the analysts, we're also looking at ways that build can be automated in a similar way.

So like when you think about the, there is infinite complexity in Epic for configuration and different things that you could do if you could dream it up, right? It's designed for that. That's how, you know, some of these really [00:31:00] major academic groups have. You know, really come to Epic and said, okay, I can do all the things I could dream about.

In some cases I'm just using configuration standardly in Epic. But sometimes those things could be easier to say, okay, build me a rule that says, well, this and this is true. So we're building an AI that you can ask at a question and it will start to draft a rule for you that you can basically have like logic being built for you, and then you can plug those logic into standard places in the system.

So a little complex, but yeah, there's the stuff where it's like the analyst knows exactly what they wanna do. Can we just have it talk to the computer? Computer does it. You okay it. You bring it to change control and away it goes.

Bill Russell: So I was after UGMI was I wrote a couple of articles and I monitored the various social media things.

The two biggest concerns, and you probably hear these all the time, one of them being data privacy on a large scale data set like that. The second being the use of the data and a concept called Drift, how it could. [00:32:00] Potentially. But I think that depends on what AI model's being used against the dataset and how that's being applied.

I'm curious with those two things as the backdrop of people's concerns, I'm sure you get asked about those things all the time. I mean, what's the

Phil Lindemann: Yeah.

Bill Russell: What's the response?

Phil Lindemann: Yeah, and I think the, from the privacy, I think some people look at us and they're like. They're almost a little perplexed because usually when people are trying to bring data together, it's to sell it, right?

They're trying to build something that's an asset. Since day one of building Cosmos, the Cosmos data can't be sold. It can't be sold by Epic. It can't be sold by any member of it. So that immediately. Provides a safer haven that people feel comfortable knowing that the only place their data will ever exist is within the confines of Cosmos.

And the charter of Cosmos is. You can do things in Cosmos as long as they're creating generalizable medical knowledge. And that's a vast group of things, but essentially you can use that to, to create new [00:33:00] knowledge. So step one is that's what the data's for, period. It is not to be sold. But the other thing that we did that we had to.

Get both sites and patients and things like that on board with is we did things like support, local laws and rules. So we went above and beyond HIPAA and standard privacy laws. So like New York State, HIV results can't be transmitted to these, to something like this that's just in their laws.

So we had to be able to support those types of things. And then we had to add things like a patient level opt out. So patient doesn't wanna be in Cosmos. If they can let their doctor know and they'll be out. So we wanted to make sure that the site had control but we didn't sort of. Tear down the data so much that it was unusable.

So we had to make sure that through safety and controls, we could have a high value data set that people could find really useful, but was still safe. So that I could talk your ear off about all the things that we do here, but it's, it is really something [00:34:00] that's never done and we're always trying to fine tune how we think about it and how we work through all the crazy situations that come up.

Bill Russell: The fact that the data can't be sold and it's not intended to be sold. I think that always throws people for a loop. It's like, why are they doing it? It definitely

Phil Lindemann: does.

Bill Russell: Yeah. Like, yeah. Yeah, because there, there are epic customers that have formed their own consortium to put that data together for the purpose of selling it.

I mean, yeah. But, and that's, you know, as long as they follow those rules it's acceptable.

Phil Lindemann: Yeah. That's their prerogative. I mean, it's something that's gone on for decades and there's hundreds of businesses that do things like that. And everyone has to differentiate somehow. Ours was to say, our goal is to build tools that are gonna help the doctors that made the data in the first place.

So it's really, we see it as the closed loop system. So the value for Epic is we're gonna make the thing we make for doctors better.

Bill Russell: I'm an optimist. I'm a technology optimist. I believe that technology is gonna make healthcare better, but you're gonna give me some ammunition here. So, alright.

[00:35:00] Epic has more healthcare data than anyone. I believe that's a true statement. It feels like it's a true statement. But you know, we, especially during election years, we'll hear all this talk about, you know, outcomes aren't getting any better in the United States and so forth. I do wanna recognize that social determinants is 80% of outcomes and 20% is actually healthcare.

But with that being said, when do you think we will start to see some material improvement just based on the utilization of ai, the utilization of data in those things? I thought it was interesting this year. For the first time in I know this is pharmaceutical, but the first time in decades, we saw the obesity rate go down and it's GLP ones, right?

So it's just like this one thing transformed a. A trend that had been going on for decades. I'm curious when we can maybe anticipate data having that same thing, maybe new outcomes, new advances, new or maybe we're already seeing it. I don't know.

Phil Lindemann: Yeah, I think [00:36:00] we're definitely seeing it and it's you know, you take something like GLP ones and obesity and the, you know, the impact on the population is massive.

But I think our own research department, epic Research has done. Some studies on lifespan of patients with like cystic fibrosis and different things, and shown how it just continues to go up and up until it almost becomes the standard for human life. So I think within certain diseases you see incredible progress that is going to be in the data and it's being able to have a data set that can focus in on those microcosms If I have to sort of, what's the.

The call to action here. I think we as a society need to be better about getting patients to be part of that loop where they're recording pros in their phone or patient reported outcomes. When you're on a brand new drug that's new on the market. Every month they should be asking you how it's going. Are you having side effects?

What are the things happening? And then just like we're doing with [00:37:00] these Cosmos tools, we should be sharing those things back with the patient. Here's the profile of what this looks like and it's right in your MyChart and things like that. So, and if you make a great drug, you should be rewarded for that.

So I think there is an opportunity for transparency with data like we've never seen before, knowing that every player in the health system now can have a connected device into that one ecosystem. Patients have MyChart, epic users in health systems. There is payers through payer platform.

We're trying to get the life sciences companies on board. We have aura for diagnostic testing companies, so, you know, with the patient at the center and then their care team and clinicians. Our goal is that this whole ecosystem could be communicating with data to make sure the best and the brightest are coming to the top.

Bill Russell: The small-mindedness person in me was most excited when I saw the single login to MyChart. because it was like, oh my gosh, I've, I have four of those. That's awesome. There's gonna be one. Yeah. I don't know it. And I don't know what the practical application of that [00:38:00] yet is for me, but it's, it feels right. It just feels like it. That's how it should be, not four different applications on my phone. And I think that is one step towards simplifying the experience for the patient so that they go, oh, I know where to go to give that feedback.

Phil Lindemann: Yeah, the next thing. So like, that's just, that was almost a mechanical joining, but the next thing for us is.

As much as we want to have this be true, the way that we've designed the flow of information and work in health systems and things like that doesn't always happen perfectly, right? People write things on notes and they put things here and there, and how do we catch the things that drop in between these handoffs?

The handoffs are the most critical, potentially have the highest risk for danger. So like we, we did something, the Christ Hospital, they're in Ohio. They were one of the first to use a I'll call it incidental OMA detection. So oftentimes you just get a normal chest x-ray and there'll be some lung nodules, or they'll have some findings on there, and [00:39:00] then in a little free text at the end of the radiology result, they'll say, you should follow up on this in six to 12 months.

It's just subtly there and it might get missed. It gets passed on to a couple different doctors. Now we use ai. To scan that radiology result. And if it sees something that says, oh, they need a follow up, it makes a discreet follow up in our system, that goes to a queue that someone can monitor and it doesn't get lost.

And Christ Hospital, when they implemented this, found that their, I don't remember the exact percentage, but you know, they are finding a higher percentage, like 20% higher. Early stage lung cancer than the rest of the country because they're picking these things up and then calling these patients and having them in.

So actually 46% was the average. So Christ is at 69%, so 25% higher, so is pretty impressive. That's just like one little AI feature. So I love it when there's these AI things that are essentially pretty easy to turn on, but they have incredible outcomes that no users really needed [00:40:00] to do more be trained on.

Like that queue of follow ups already existed, but certain things didn't get put on it. Now AI is helping. Nothing falls through those cracks and puts those things in there. I had a physician tell me there are hundreds of those types of situations in medicine that we're gonna be uncovering those rocks for the next few years.

And just saying, attack that, figure out that handoff build something to help that, where we've already built the substrate, the infrastructure that has the queue and has the reminder and has the alert, but things fall out of that system. How do we pick them up and put them back on the conveyor belt to that system?

That's, to me, that's pretty exciting. Although it's. You know, you're picking up nickels there, but they're nickels that could save someone's life.

Bill Russell: Oh, absolutely. Uh, Last and most important question packer Super Bowl.

Phil Lindemann: Of course every

Bill Russell: year course for every year. doesn't matter what the, it's like, yeah.

You gotta be a diehard Hope springs eternal at the beginning of the season. It's a beautiful thing. That's right. Well, you got, you definitely have a good team and it looks kind of, kind of exciting. Hey Phil, really appreciate the time and sharing [00:41:00] the progress that's going on over at Epic.

Really appreciate it.

Phil Lindemann: Excellent. All right. Great talking

Bill Russell: to you.

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.

If you have a conversation, that's too good not to share. Reach out. Also, check out our events on the 2 29 project.com website. Share this episode with a peer. It's how we grow our network, increase our collective knowledge and transform healthcare together. Thanks for listening. That's all for now.