Epic has built such a large user community that it makes sense that they will have one of the most used Artificial Intelligence plays in the industry. Seth Hain, VP of R&D for Epic discusses how they are thinking about the space and where they are going. Hope you enjoy.
Bill Russell: 00:07 Welcome to this week in health it influence where we discuss the influence of technology on health with people who are making it happen. We are the fastest growing podcasts in the health it space. My name is Bill Russell. Recovering healthcare CIO and creator of this week in health, a set of podcasts and videos dedicated to developing the next generation of health it leaders. This podcast is brought to you by health lyrics. If you’re looking for an executive coach that’s been there and done that, run a $250 million budget within a $7 billion health system around it. Let’s talk visit healthlyrics.com to schedule your free consultation. Uh, if you’re enjoying the show and want to support our mission to develop the next generation of health leaders, there are five easy ways you can do that. Share it with peer, share it on social media, follow our social accounts, linkedin, Twitter, Youtube.
Bill Russell: 00:52 Send me feedback and questions. I love the feedback and questions. In fact, this episode is a result of one of those, uh, one of those emails. Uh, you could do that [email protected] and you can hit the website and subscribe to our new newsletter. We have one last thing that I want to make people aware of. We launched a new service this week. Uh, you know, here’s the question. Do you, do you have anyone who’s investing in your career? Uh, for many people, the, the answer to that question is no, and that’s really not okay. Uh, we just launched a service this week health daily to address that gap and to invest in your career. These are daily videos that highlight conversations with industry leaders where we make them pragmatic by answering the question of, so what, what does this mean for you today in your role, in your career?
Bill Russell: 01:36 So check it out at this week, health.com/daily and sign up today. So as I said, I love your feedback. Today’s guests is really result of this. Um, you know, to our listeners, I agree. It’s hard to have a show about health it where you don’t talk to epic about what they’re doing in the innovation space. So, uh, you know, I had a conversation with Judy, just wanted to name drop there and she made the uh, uh, you know, the connections. And, uh, today I’m happy to be joined by Seth Hain, the VP of r and d at epic. Good Morning Seth, and welcome to the show.
Seth Hain: 02:11 Good morning. Thanks for inviting me.
Bill Russell: 02:13 You know, it’s, it, it, I, I just had to drop Judy’s name cause everybody in the industry drops Judy’s name because she’s so approachable and easy to talk to. Um, and I appreciate her making this, this introduction. You have a, you’ve been there, you’ve been at epic for a while. Tell us a little bit about your, uh, how you got to epic and your career at epic.
Seth Hain: 02:32 Yeah, sure. So I’ve been here about 14 years and before coming here I was studying mathematics and as I transitioned into a role here at the company, my focus was really on backend engineering, database architectures, performance and scalability of the system. And about five, six years ago, um, I started combining that previous history and with the architecture, um, experience and focused on building out our machine learning platform and data science efforts. Um, these days I, in addition to focusing on machine learning, I’m also working with our business intelligence tools and our data warehouse products. So it kind of full analytics and AI is my area of focus
Bill Russell: 03:22 these days. So I, I’m looking forward to the show. I, I love having, I like, I like geeking out so I love having CTOs on it and people with technical backgrounds cause we can, we can really dive in, uh, to those things. So we go back and forth between talking about leadership and the industry and, uh, I, I have to confess these are, these are my favorite episodes, so thank thanks for coming on the show. I’m looking forward to it. So, um, yeah, let’s, let’s start with the open ended question, which is, is there anything you’re working on today that you’re really excited about that you can share with us?
Seth Hain: 03:56 Yeah. I think that the key is continuing to help organizations fully adopt machine learning into their practice. It’s something that we’re already seeing tremendous results from and is frankly pretty widespread across the community. So helping folks fully realize that I think is the key right now. And there’s so many different areas that can be applied. I’m certainly the first people tend to think about is the acute space as well as population health. I think a an exciting area of opportunity for machine learning and where we’ve got to a number of use cases and it already is continuing to improve physician efficiency and user efficiency by introducing machine learning into workflows as well.
Bill Russell: 04:44 So, so what gives an idea for those of us who aren’t familiar, give us an idea of the uh, the tools, the epic machine learning and AI tool set that you guys are working with. There’s really
Seth Hain: 04:56 two focuses that we have. The first one is on a library of machine learning or data science models that organizations can implement out of the box. And that tends to be the starting place that folks go to. The second one is the platform behind the scenes that allows machine learning to run in real time on data as it’s flowing into the operational system and embed those results transparently back into workflows. That backend platform is where both the model library that I referenced runs that library includes things like I just referenced acute models, population health, those workflow models. But the platform also allows, um, new and kind of innovating organizations to embed their own machine learning models into their workflows as well. So really those two areas, the library for organizations to just kind of out of the box began using machine learning as well as the underlying platform for folks that want to take that next step.
Bill Russell: 06:02 Cool. So have you, so one of the strengths of epic is obviously your user community. You have some, some, uh, some users that are, uh, very large systems, very innovative, doing some exciting things. Um, you know, what are ways that, that the whole community benefits from each other in this area of advanced analytics and AI if they’re on an epic platform?
Seth Hain: 06:25 I think one of the, the keys to being successful with do you call it AI or machine learning and in really similar terminology here is to not just think of it as a model, as a piece of content, but think about the holistic picture, the clinical program that you’re implementing, be it fall risk or understanding patient flow throughout the hospital in an operational sense and think of that programmatic aspect of how you’re going to train users to use those predictions, how that implements into the system that you already have in many cases. And so through events like our user group meeting coming up here in the fall are expert group meeting in the spring as well as a number of other um, opportunities that we have to interact and help the community share their clinical practices. That’s where I think the real value comes is seeing that bigger picture of how people are using machine learning in their practice.
Bill Russell: 07:29 Does. So is that progression, really analytics, then advanced analytics, then I’m at data science and then machine learning and then AI. I mean is that sort of the progression to do systems really have to be good at analytics and really be good at data science before they start to venture into this area? I think that there’s,
Seth Hain: 07:50 they’re related but I think there’s kind of two when we’re talking about analytics and AI and I think it’s probably worth just diving into the terms for a moment here. While they’re related, we tend to see folks using them in slightly different ways. So on the analytics side, what I’m thinking of is things like self service tools where a physician might want to dive into a large population data set using something like our slicer dicer tool to explore and identify, say their high risk diabetes patients and then be able to do bulk outreach. So starting with a population and sorting through it if you will, in a visual user interface. So that’s kind of on the analytics side and often you want to use that then to monitor building dashboards, those sorts of things. Which you can do with the Code Obi Suite on the artificial intelligence or machine learning side.
Seth Hain: 08:43 And what I keep using kind of both of those terms, machine learning essentially being a subcomponent of the broader AI term, which encompasses things like robotics as well. For example. Um, in that context it’s often say the informaticist team or others working directly in workflows that are introducing models in that context. So a simple example of this might be a care manager workflow today where they’re trying to do outreach to CHF patients and they might be looking at a particular lab score on which they sort those patients for outreach. Instead, they can simply swap in a predictive model of their longterm CHF risk, increase the effectiveness of their outreach, but keep the same workflow, not even really needing to worry about the implementation details behind the machine learning. Of course, having the ability to hover over and understand what’s behind it if they want, but not needing to to use it in the system
Bill Russell: 09:48 now. So I, so that’s an interesting use case. Do you have, um, you know, maybe as a platform to go off of here, do you have like a one or two client success stories that you share with us and you can feel free to shout out to the, the organizations that have come up with these things, but are there any success stories that, that uh, that you guys like to highlight around around the use of, uh, AI Ml,
Seth Hain: 10:11 models? Yeah. A couple come to mind. Um, the first one is north oaks an organization down in Louisiana and they’re actually one of the organizations or community hospital down there that has at this point adopted over 10 models for our machine learning library. They saw a 40% reduction in codes outside of the ICU by implementing deterioration models, which really are a way, you know, if I’m thinking about it from the patient perspective, I can be confident when I’m either, you know, maybe unfortunately I’m in a med search on the med surge floor or a family member next to somebody that’s at the hospital. I can know that as those monitors at the bedside are collecting information, the system is running in real time analysis of that information along with my longitudinal chart to understand my risk of say hospital acquired infections, deterioration in the context of north oaks here, early onset of Sepsis, and then alert individuals maybe say in the ICU that I need some intervention, somebody to come check on me. Um, so north oak certainly stands out as one example of another.
Bill Russell: 11:29 Let’s stop on North oaks real quick cause that kind of surprised me that you started off with a community hospital in Louisiana. You know, you would think that the, the, the, the organizations using this are in, you know, in Seattle or la or San Francisco or Chicago. Uh, but you’re saying that, you know, even the, the small hospitals have access and the benefit of utilizing these models is that because you’re, you’re packaging them up in a way that they can just implement them out of the box.
Seth Hain: 12:01 Yeah, and that’s the key to those two approaches that I spoke of earlier. Right. The, the library enables organizations to quickly begin implementing in using those models in their existing workflows. And the underlying platform in, um, provides them the options to do that cost effectively. It’s cloud based. So as they need resources to localize or retrain models to their particular populations, they can spin up those resources in a public cloud, train those models, and then implement them directly back into workflows without needing to bring all that infrastructure in house. So, yeah, our goal is really to help organizations of all sizes. I grew up in a small town outside of Lincoln, Nebraska, uh, with the community hospital, right. Help organizations like the community I grew up and be able to implement machine learning for all patients.
Bill Russell: 12:59 Yeah. So all right. So that, that’s small, uh, as a smaller organization and that’s exciting. What about the larger organization that has the resources and has, uh, the, the ability to hire staff, um, you know, are there success stories around those kinds of organizations that, that can build out models really that can, that can go across, uh, the entire health ecosystem?
Seth Hain: 13:23 Yeah. A couple of organizations immediately come to mind. Um, Ochsner also happening to be in Louisiana, an academic organization down there. Yeah. You’re making, you’re making Louisiana sounded like the new silicon valley. I am with my second example, I’ll get back up here into the west. Um, auctioneer is implemented seven different models they have built in their system and operationalize them. Um, an example recently of something they did was reducing hospital acquired infections by around 35% after implementing a model. And the interesting thing is that these machine learning models work outside of the context of providing direct care as well.
Seth Hain: 14:06 So if you think about an organization like rush down in Chicago closer by here to Madison and Verona, um, they wanted to reduce the number of individuals leaving their ed without being seen and build out a model that identified those that were at high risk of walking away without first having been, um, having touched base with a physician and based on those risk scores could then send folks around but just touch base with them and let them know their status and they saw a reduction of 50% in individuals leaving the Ed without being seen. So we end up seeing folks use the platform after having identified a specific concern within their organization or opportunity and then building out machine learning and implementing back and workflow in real time, um, to be able to achieve those outcomes.
Bill Russell: 14:59 So if, if is doing that, does that become something that they can share with the larger health community and something that people can just tap into?
Seth Hain: 15:08 Yeah. Yeah. So no, that’s a great question. And there’s share in a couple of different ways, right? So one, yes, rush presented at our expert group meeting earlier this spring on the model that I just talked about in their workflow around helping folks understand how to operationalize something like that. The platform also enables folks to share models between organizations so that it’s not just a, the bigger picture workflow, but as, as there is interest, the ability to transport those models as well. We implement that using something called docker to container technology behind the scenes, which allows folks to package up their models and then be able to distribute them. Same thing we’re using for our library of models as well. Interesting.
Bill Russell: 16:02 So you know, so talk about the landscape. We have, um, you know, different health organizations are doing different partnerships at this point. Uh, you know, you have a, you know, Google, Amazon, Microsoft and others and uh, you know, more CIOs that I’m talking to are sort of dabbling in that area. Um, you know, what are, you know, what are the strengths of each and where, where do they fit into healthcare? And if you could put sorta your platform in, in sort of that framework as well, that’d be great.
Seth Hain: 16:34 Yeah. So I spoke earlier about how there is a need to be able to quickly spin up in down resources in some context because these types of machine learning algorithms can be quite intensive on occasion, right? There’s, there’s significant math behind the scenes of these and calculations that need to take place is the data’s crunched to identify those patterns in public clouds. Um, be it Azure, AWS, or Google cloud provide an effective means to do that. And so what we do with our platform is be able to efficiently extend that. So even if you’ve got your operational system on premises in your own data centers, there’s still opportunities for you to leverage the cloud cost-effectively in those contexts.
Bill Russell: 17:25 And our platform extends out in that way. That’s interesting. So, so let’s talk about, you know, your, you have a platform, but you know, we know that social determinants make up a huge amount of the outcome in terms of people’s overall health. Uh, how is, how is epic really working to advance the work in, uh, in population health, around social determinants? I mean, you gave us these example of, of rush. I mean, how are you bringing in that, that other data and, and is that data all just flowing into your system and being stored in some aspect of the EHR and then being utilized by your, uh, by your tools?
Seth Hain: 18:06 So it’s, there’s two aspects to that question. The first one is around data flow that the piece you spoke of. Um, and certainly social determinants of health is one aspect of that. And folks are capturing that as part of workflow, say in an ambulatory setting today, um, or bringing it in through other sources into the system is they want to make sure that that information is present as, as part of the conversation around chronic conditions.
Seth Hain: 18:35 But in other cases that, you know, might not be within the kind of epic infrastructure ecosystem. And so in those contexts, say a pacs image, um, we may call out as part of a prediction to incorporate that data alongside data that’s been captured within the operational workflows. In an epic and incorporate it into the prediction. The second aspect that I think is important here is regardless of the, kind of all of the variety of sources of data, having that prediction in front of the whole care management team around population health, there’s a number of folks that come into contact with that individual, you know, certainly the nurses and providers in an ambulatory setting. Um, but for my kids, it might also be the nurse at the school. And so making sure that those predictions are available to all of the users participating in the management of, of those chronic conditions and the population more broadly, um, is, is a key aspect of how we built the platform. So that’s embedded back into the workflow.
Bill Russell: 19:44 So, so let’s talk about two things here. One, um, I’m going to use two different personas here. One, my daughter who’s getting ready to go off to college. And the second being a clinician who’s been doing this for maybe five to 10 years, are there, what would you say to those people if they say, look, I want to be relevant in the, in the, uh, AI machine learning world that is, is emerging, um, you know, what skillsets would you recommend that they sort of develop or courses that they would take or path that they would get on so that they can really be an active participant in this movement?
Seth Hain: 20:20 Yeah, I, I think that there are, I tend to think of a three legged stool of kind of skillsets that oftentimes not one person embodies all three. And so it’s a team effort. Um, but understanding and being able to communicate between those teams is important. And that really is around one, the domain expertise, understanding be it the clinical context in which the model will be implemented or say the healthcare administration context that you might be using if you’re thinking about understanding patient flow as an example. So domain expertise being the first and critical. Um, the second one being deep data science, understanding, um, and kind of mathematics background around it. And then the third understanding how to implement it into that core system that, uh, the workflow system that you have embedded throughout your healthcare organization. And I think those are kind of the three keys and being able to have x deep expertise in one, but then the ability to extend it into the other two is important. Um, it might be one place to start.
Seth Hain: 21:29 Um, that I think is worth quickly mentioning. The National Academy of Medicine in DC recently formed a working group around artificial intelligence in healthcare, um, and they’re developing a publication that will be made available later this year. Um, it’s really targeted at those three groups that I just highlighted. And while it has some of the history around machine learning and its place in health care, a key focus is around implementation and understanding how to get it, put it into practice and then after it’s live, how to continue monitoring its use in the system so that you know, you’re getting the results you want. So I point folks, excuse me, um, to resources like that.
Bill Russell: 22:15 So, um, so my daughter going off to college and she says, all right here, I want to do this. I mean, do you recommend she, you know, go into analytics, track a mathematics track at data science track, let’s assume she’s not going to go into clinical track and learn that she’ll learn that when she gets out, but if she really wants to, uh, to, uh, advance in this space or is there one path that’s better than another?
Seth Hain: 22:42 I don’t think there is. Um, I don’t think there is. I would generally see folks starting in a computer science path if they’re truly interested in this kind of deep technical piece. Um, and then branching into the mathematics aspects of it that particularly sets folks up for having an impact with those models. Right. How do you put them into practice in software so that people are getting value from them? That’s what we tend to look at is we look at candidates here at epic is making sure that they’re driven towards helping customers be successful with this type of use of machine learning. Um, if they’re more interested on kind of the deep theory, um, starting in mathematics tends to make sense. So little bit of a difference of what type of flavor she’s interested in. But I think both of those are important.
Bill Russell: 23:35 You like how to do that. Like how I get free, uh, career advice for my daughter by doing the podcast. Um, hopefully other people are getting a getting benefits out of it, but it’s, Eh, I it, it, I think one of the things that I hear over and over again and I’m hearing from you as well, it’s, um, eh, this isn’t going to be in one brain, in one person. This is going to be a team effort. You’re going to bring, uh, clinicians and uh, and a computer scientists and data analysts and uh, and workflow expertise. You’re going to bring a whole group of people into a room and you’re going to build the models together. Uh, as really as a team. And that the, those probably are the characteristics of the most effective systems, uh, in terms of implementing this. I, that’s what I’m hearing from you is that, is that what you have found?
Seth Hain: 24:25 Yeah, it’s absolutely accurate. I think if you, if you consider it’s interesting, it mirrors what’s happening on the technical side, right? A deep understanding of the data, where it came from and how it can be useful is really aligns with that domain expertise, the mathematics to understand patterns in it and then the computer science background to put that into practice in the software. So you kind of need all to be clear on the computer science side. In many cases, these are folks that understand how to configure and set up the software. It’s not that they need to know how to program, um, it might be even physician builders. Um, of course we have here at epic for folks that have deep clinical background but also want to get into the system, configuring these into their workflows. So it’s really an understanding of how to build and configure the software.
Bill Russell: 25:18 Um, that hits at that top level. Yeah. Side. You know, I, I really have two questions to close up here. One, one’s going to be a really focusing on a health system and uh, uh, you know, so a health system is looking to do this, this kind of work. Um, you know, we’re, where do you, where do you recommend that they start? Do they, they start really at the establishing a program with as objectives? Um, I mean, or is there, is there a different path that people have taken in terms of really getting quick wins and then sort of a, yeah, I don’t know, making the, making the program visible and, and then really establishing it. I mean, what, what has worked, what’s worked really well?
Seth Hain: 26:02 I tend to see folks have a lot of success starting with specific problems they want to solve and then as they approach those, thinking about building out a foundation for them to repeat that approach over and over or around new problems. So you might start with an early onset of sepsis model and implement a program around that. Moved to deterioration, moved to fall risk branch out into the ambulatory space and the operational models. Right. Um, but I think that the core as you’re working on those initial projects is to build out a foundation that allows you to scale. And I think I see there is not needing to worry
Seth Hain: 26:44 about shipping data off to a bunch of different systems and bouncing users between different workflows, being able to bring the intelligence directly into where the users are working, regardless of where that intelligence might’ve come from. It might’ve come from data scientists down in at rush or Ochsner or out on the west coast, but being able to put that directly into their workflows so they can make the most of it with that foundation.
Bill Russell: 27:13 No, it’s interesting. I mean, one of the, you know, with, with epic, you are the, the center really, or the operational world within an acute care setting. But as you know, and as our listeners know, uh, this setting is getting larger and larger and larger, uh, and you have other data sets and, and, and new datasets, um, that are all out there. I mean, how, how do you play? Um, I, I know it would be nice if the whole world was on epic. I guess from your perspective it would be nice, but actually from a client, from a patient perspective, if, if everything was on a single EMR, it would be kind of Nice cause you would solve that interoperability problem and those kinds of things. But that’s not the world we live in. So it’s, it’s a lot of different sets of a lot of different EMR, a lot of different things can, can we still do these kinds of models on your toolset with, um, these disparate systems, uh, in a community?
Seth Hain: 28:11 Yeah. And I think that gets to a kind of couple of different pieces that I mentioned earlier, but kind of pulling those threads together. Um, the first one is the variety of different data sets out there. Uh, you’re absolutely right. Not all of that lives within the epic ecosystem. Um, and it’s, but it’s important to be able to incorporate it into these predictions that are able to look at tens, if not hundreds of different data points around an individual or an organization. And so as we build out the platforms, we’re always making sure that there’s the ability to reach out from an analytics perspective or an AI perspective to those other systems and be able to, you using APIs, incorporate those into predictions. The other aspect is using common industry standard approaches to incorporating that intelligence or machine learning into the workflows so that folks that are using something like a container based approach or say an open source algorithm like tensorflow to build their model, still have the opportunity to map it into how that data might be captured within the operational system and use that package directly within their workflows. So I think there’s a couple of different strategies depending on, uh, which piece you’re looking at here. But it’s important to kind of stick to and helping encourage those industry standard approaches, be it interoperability or open source technologies for bringing machine learning and those sorts of things.
Bill Russell: 29:43 So I’ll throw out One last question and you answer it. Don’t answer it. I just, I appreciate, uh, the conversation this far. So here’s the interesting thing for me. So if you’re a startup, if you and I are doing a startup, um, one of the things we want to avoid is we don’t want to compete with epic cause you and I are startup. We don’t have a lot of cash. Um, but where, where, where might you tell people, hey, you might want to focus a, again, we’re through, we’re staying in the AI machine learning space, where, where would you say people, hey, there’s, there’s gaps here. There’s places for you to play that you can really, uh, benefit the overall ecosystem. Uh, that epic isn’t necessarily solving that problem today and there’s a, some runway for a startup to really make it make an impact.
Seth Hain: 30:32 Yeah. I, I tend to think of some of the kind of very deep, um, specialty workflows as an example where it might make sense, not, not the workflow per se, but there might be some specific expertise you could bring around building a machine learning model or in, and then being able to integrate that in through API is like I just talked about into an organization system. So there’s a number of different ways throughout our operational software to be able to incorporate external knowledge, if you will, in, and so thinking about an area of expertise that an organ that a company might have around, uh, a deep understanding of medicine, being able to build out that content and then integrate it into the workflow, I think is one interesting example where there is opportunity.
Bill Russell: 31:29 Yeah. Well that’s exciting. It’s exciting that you guys are, uh, giving people the opportunity to get that information back in the workflow because we know that workflow matters and we hear that over and over again from clinicians. Um, you know, Seth, thanks for coming on the show. A great discussion. I really appreciate it. Uh, is there anything else you want to leave with our listeners or a way to follow you or resources that you think are valuable?
Seth Hain: 31:55 Uh, I, I really enjoyed the conversation. Uh, appreciate the opportunity to sit down and chat and have a great, great morning.
Bill Russell: 32:04 Well, thanks I and a, you know, please come back every Friday for more great interviews with influencers. And don’t forget every Tuesday we take a look at the news that is impacting health, it, this show is production of this week in health it for more great content You check out our website @thisweekinhealthit.com or the youtube channel, which you can get to from our website site. Thanks for listening. That’s all for now.
Amy Li founded Dance4Healing months after she conquered Stage IV cancer. Amy is full of energy and enthusiasm about what is possible with exponential technologies and design thinking to support those facing similar challenges. Great conversation, I hope you enjoy.
We caught up with Anne Wellington the Managing Director of the Cedars Sinai Accelerator to discuss how entrepreneurs can find their way in the healthcare maze. Hope you enjoy.
Albert Chan, MD delivered the keynote for the provider’s section of the Health 2.0 conference on digital health innovation. The team at Sutter has a very practical lens with which they approach innovation. Hope you enjoy.
Dr. Michael Pfeffer says clinician burnout is multi-factorial. Admin burden, regulatory challenges, malpractice concerns, and the rise of consumerism. We caught up with him at Health 2.0 for a quick discussion. Hope you enjoy.
Winjie Miao, EVP and Chief Experience Officer for Texas Health Resources discusses with us how they organized to deliver experiences.
Conversational technologies are everywhere and now you see them emerging in healthcare. We caught up with Kristi Ebong from Orbita to discuss enabling platforms for conversations. Hope you enjoy.