Integrating Shiny with Epic EHR

Transcript#

This transcript was generated automatically and may contain errors.

Hey there, welcome to the Paws at Data Science Hangout. I'm Libby Herron, and this is a recording of our weekly community call that happens every Thursday at 12 p.m. U.S. Eastern Time. If you are not joining us live, you miss out on the amazing chat that's going on. So find the link in the description where you can add our call to your calendar and come hang out with the most supportive, friendly, and funny data community you'll ever experience.

Can't wait to see you there. I am so excited to introduce our featured leader today, Matt Maloney. He's the Director of Applied AI and Data Science at City of Hope. Matt, it's so nice to have you today. How are you doing?

Doing well. Glad to be here today.

Wonderful. It would be great if you could introduce yourself and tell us a little bit about what you do and something you like to do for fun.

Sure. So what do I do? I'm a Director, as Libby said, of Data Science at City of Hope. City of Hope is a large cancer treatment and research organization and has been recognized as one of the nation's top hospitals for cancer care. We have hospitals in a few locations. So Los Angeles, Orange County, Phoenix, Atlanta, and Chicago. And I've been working as a data scientist and now in a leadership position there for about five years. My team specifically focuses on building data solutions that are more operations focused as opposed to research focused. There's a separate group we do work closely with that supports research. But I'd be focused on things like, for example, helping with predicting claim denials or helping manage something for philanthropy and help them find a solution.

So that's a little bit about what I do professionally. Like I said, I've been in health care now for about five years at City of Hope and before that University of Utah Health System, but have a background in finance and econ before that. Something I do for fun. I live in Fort Collins, Colorado with my wife and two small children. And whenever I can, I like to get outside. So climbing, hiking, mountain biking, snowboarding, you know, I really find that stuff helps me recharge. And I love being just outside whenever I can be. It's a beautiful day here with a lot of snow on the ground, actually.

I was just gonna ask, did you get snow? Because my friends in Illinois and Wisconsin certainly did.

Yeah, we got, I don't know, maybe six inches yesterday and it's stuck around because it's been cold. Yeah, well, you live in a great place to get outside. Fort Collins is fun and it's a college town, so it's nice and nice and energetic. Well, I already saw somebody ask in the chat a little bit more about City of Hope would be great. So if you could talk a little bit more about what City of Hope does and what the data science function helps with the kind of problems that you solve, if you could go in a little bit more depth there, that might give us some context.

City of Hope and the data science function

Sure, yeah. So the function of, well, starting with City of Hope, City of Hope is first and foremost cancer research and treatment. So we're not doing any clinical care that is not related to cancer. And then on the research side, cancer is the focus as well. And so what does that cover? That covers fundamental research on the research side. It's all kinds of cancer care for every type of patient and disease type. It's also clinical trials, right? Making sure that we're matching patients with the right clinical trials and getting them enrolled in the best care possible. And I think one of the core missions of the organization is not just providing world-class cancer care, but it's making it as accessible as possible to as many people, right? So it's not just getting the highest quality treatment, but also making it broadly accessible.

So what is the data science role at City of Hope? So data science is a kind of broad term that I know the role varies by organization, even across healthcare organizations. You'll have different data science teams doing different things. So we have a kind of centralized group of data scientists in kind of this larger applied AI and data science department. And we're kind of split into two teams. The first team is focused on research applications. So they tend to work very closely with the academic research teams at City of Hope. And then, sorry, I was distracted by the chat for a second. And then my team is really focused on development and solutions for our internal operational needs. So that could be on the clinical side. You can think of something like, you know, we have a proprietary model for predicting or assessing sepsis risk for patients who have received bone marrow transplants. So that would be more like what we would call clinical operations. We're trying to support an operational workflow with a product we developed and manage. And then more on the business side, you know, my team is also supporting requests around, you know, revenue cycles. So this may be, can we have a model that will help us optimize our billing? One of the earliest projects I work on and still support is a forecasting system for our finance team.

And, you know, trying to think of new applications and other areas. Another example is, recently, we've been tasked with supporting our philanthropy team with doing more, using some LLM technology to help better understand and match gifts with research projects. So those are like a few examples, I think, that can kind of create a vision of what our team is doing. Now, you might be hearing in this that the team I'm in is focused a lot on sort of production solutions that tend to come in the form of like, like software or, you know, something we would put into production that would run either as sort of an endpoint for inference from a model, or it might be something that we would set up as like a daily cron job to deliver productions to some people in a workflow somewhere. Or in the case of the clinical models, you know, signal and alert in one of our EHR systems, right, when, for example, a model has a flag that a patient is at risk for sepsis, for example.

I just wanted to say something about the EHR. And I think that's electronic health record. But I remember our team saying something about how you are integrating POSIT with EPIC, your EHR. And I thought that would be so cool to be able to share with other healthcare people on the call as well. Would you be able to tell us a little bit more about how you're doing that?

Yes. So EPIC is, I'm sure many of you may be familiar with it, but it's the largest organization, or largest vendor for electronic medical record software, right? So when you go to visit your physician or whatever, and you see them working on their computer, there's a good chance that that's the EPIC system they're working in, right? So a lot of the data my team works with is data that originates in the EPIC system. You know, that's where information about patient gets put in. And that's also, importantly, where a lot of the workflows are not just for providers, you know, care providers, physicians, nurses, etc. But also, more and more some of our operational staff, because EPIC doesn't just handle, you know, the medical record side of things for us, but also handles a lot of the billing related work and processes. So we also have what we call our revenue cycle staff working in that environment and with that information.

And so one of the requests that was coming up is to, you know, we would have standalone applications, a lot of them would be living in our PositConnect instance. And so, you know, for example, this is a good, good motivating example, we have a tool that helps our coding team. So this is the team that, you know, a visit happens, and then, you know, basically, it's recorded what treatments and procedures and diagnoses the patients have. And then that moves to then into not only the billing process, but also, you know, just to make sure that we have a rigorous and compliant medical record for the patient. So there's a team that reviews this documentation. And we have a tool that assists them with that documentation. And it was living in PositConnect. And, you know, one of their requests is we do all our workflows in EPIC, this would be a lot lower friction for us if we could embed it.

And EPIC does have a capability of integrating software in a few different ways, right? If you're a, you know, a software vendor that's trying to say develop something that you would want to be available in the marketplace, right, as like a really premier like software solution for other healthcare systems, you would use like a spar on fire framework, you would register as an app developer. But for a lot of our cases, we were hoping there might be a sort of a kind of middle ground to becoming full blown software engineers. And so we've been working with the POSIT team on nesting our applications in EPIC. And, you know, passing a few parameters using what they call, you know, setting it up as an, they call it an FDI record, if anyone cares, I don't actually even know what that acronym means. Maybe somebody else here knows. But you set up this record in their system of where you want this application to live. And then we've been working on some of the details of making sure that the integration goes smoothly. And we're still, you know, we still haven't completed that process, but we're getting pretty close. And it's very promising, because for us, it's a really rapid way to take something we've developed as data scientists, and move it into their preferred environment for their workflows, without necessarily having to engage with our software engineering team, or, you know, get outside of the scope of where we really feel comfortable.

it's a really rapid way to take something we've developed as data scientists, and move it into their preferred environment for their workflows, without necessarily having to engage with our software engineering team

So yeah, yeah, that's, that's what we've been doing, as far as integrating POSIT applications hosted on POSIT Connect with our EHR electronic health record system.

That is awesome. So what I'm hearing is, and we're gonna hop to the slide of questions in just one second, what I'm hearing is you have end users who are medical coders, who are getting to stay in their system, but the information is getting to pass to the applications that your data science team is building, so that they don't have to, like, use a Shiny app, if they're not familiar with it, they can use the applications that they're used to using.

Yeah, so well, they the, you know, the Shiny app is is nested there, but they don't have to, for example, open a separate window, or, you know, type in whatever, you know, arguments are needed, manually, right, some of that information can be passed, which, you know, it's not, for some workflows, it's more critical than others. But but in some of our other projects, where our customers are providers, where their, their workflows and, and time is very, very valuable.

And providers being like doctors and clinicians.

Yeah, yeah, yes, maybe a little bit of jargon. Yeah, we're talking about physicians, you know, physician, physician assistants, nurse practitioners. And the goal here is to eventually take some of the work we're doing, that's more oriented towards providers. And we've already integrated some of our provider facing work through a different EHR integration method, which is like they call it best practice advisory alerts. So we do have other ways of integrating some of our, our work into the EHR system. But this would be another option for us if we can, you know, get it to where we want it to be for for provider workflows.

one thing that I think is often overlooked is how much of a time commitment is required from some kind of subject matter expert like a physician or a clinical documentation specialist to help us through that process of cleaning up the data because we don't have the clinical expertise.

So one of the things I try to do early on in a project is say, okay, you all need a solution for your workflow. Or this is going to look like a partnership where we're going to have to work together and we're going to need some of your time to validate, say, hey, have we cleaned up the lab's data in a way that it's consistent with your understanding of what's clinically relevant, right? And then, you know, sometimes that's a much bigger part of the project than writing the code to make the predictions on top of those data sources.

So, you know, I don't know if I fully answered your question, but I think that we take very, very seriously that whether we're talking about, you know, a traditional predictive model or a generative AI solution, this is one, you know, tool that's usually only a limited part of a complete solution. And one of the things is making sure that the data that goes in is of high quality. And unfortunately, that's still, there's no magic for that. That's still a time consuming process that requires time from us and requires time from the subject matter experts. And that's just the way it is. I think that, you know, we could sit and say that AI is going to solve that problem. But no, there's still a ton of work to be done.

Do the administrators appreciate the time that it takes? Like, the people who run the hospital or the facility at City of Hope include that into their budgetary aspects? Because there's another constraint that we run against, which is the healthcare administrator role and whether or not they recognize that as value added.

Yeah, that's another one I don't have an answer to because I think it's a universal challenge in healthcare. And the only thing, I mean, how I'll speak to that is just we've gotten more mature, as I said, with at the beginning saying this is what's going to be required. And then if administrators can't design some way that, for example, the incentives, you know, because I mean, let me see if this aligns with what you're saying, which is basically we have a situation where the organization wants an improvement and say how the physicians are documenting. Something like this, right? But that's going to require physicians to take time to work with us to validate, you know, the results potentially help with some of the helping us to understand the data. But then they're not given the time or any incentive structure to actually carve out that time. Is that kind of consistent with the problem you're talking about?

OK, OK, so, yeah, so I love this, thinking about this stuff, because my background is economics. That's what I did for grad school, kind of trained as an economist. So I'm always thinking about are the incentives of the people who are needing to be helping us or using these tools aligned with the broader incentives of the organization? And if not, how do we align those incentives? And I am now being very assertive about having that conversation early. So this isn't about clinicians, but a recent example is I found out that some of the clinical documentation specialists who are supposed to be our SMEs for validating some of the work we were doing. They said, you know, we have this system that tracks our time and we don't have any code to log our time for this validation work. And so I immediately was like, well, that's a problem we have to solve before we do any more development work here, because that's an instance where the overall organization wants this problem solved. You have users that are saying they can't log their time against this. We've got to resolve that problem before we can do any more development work, because otherwise the incentives are all not aligned.

Career advice and closing

Well, we'd like to ask everybody if they have a piece of career advice that they have really loved or has been really impactful for them. I don't know if there's something that you have on the top of your head or that you think about.

I think there's a lot of value in sort of finding an industry that you're interested in getting a foot in the door, even if it's not in the ideal role. Because, you know, as a data scientist, I think. One thing that becomes really, really valuable is knowledge of a specific industry or domain knowledge of a specific industry or even domain knowledge around specifically the data that your organization has. And so there's a lot of room once you become like an expert or get your foot in the door in an organization for then defining your role within an organization. At least that's been my experience.

I mean, I don't want to say this applies generally across the board because maybe some organizations have more have more rigorous role definitions than than ours does or has. But, you know, I've seen people come in and they find a problem they're interested in really excel in it. And then all of a sudden, especially because it's a growing area, they've carved out like an entirely different like job definition and career path for themselves. And that hasn't just been in my experience at City of Hope, but it's also been I remember my first job out of out of undergrad. This is going going way back. This is in 2010 job market was still a little bit weak. And I got a job that was like at a small financial firm where I thought I was going to be doing a lot of administrative work. And I like in the first couple of months, you know, they said they had some problems with their spreadsheets or whatever. I learned VBA and, you know, all of a sudden I had a totally different job.

there's a lot of room once you become like an expert or get your foot in the door in an organization for then defining your role within an organization.

But I think that's something that that, you know, I don't exactly want to call it a backdoor, but it's it's organizations. I found the roles aren't as rigorously defined as they might look like from the outside when you see a job posting. And so if you get in the door and you have something you're really passionate about, I think there's often a lot of flexibility to shape your role.

All right, everybody, we are two minutes from the top of the hour, so I think that we will have to wrap it up and say goodbye to Matt and say thank you so much, Matt, for this amazing conversation. I hope that you had a good time.

Yeah, this was this was fun. Thank you. Thank you for organizing and and moderating. And yeah, this is a lot of fun. And yeah, anyone in here, feel free to reach out to me on LinkedIn or whatever method makes sense, because, yeah, I'd love to get to know more people who are interested in similar things or encountering similar challenges in this industry.

Let's do it. Thank you, everybody, for hanging out. If you'd like to save the chat, there's three dots in the top of the chat box that you can click and save the chat, but also join the Hangout channel on our Discord server. So POS.IT slash Discord. The link is only open for the next few hours. So if you're watching this on Discord later, that link's probably not working, but don't worry about it. Join us and have a really, really good time following up. I have tagged some people in our Data Science Hangout channel that I know we're all having a similar conversation. We can get there and follow up. Thank you so much for joining us, everybody. And we will love to see you next week. Next week, we have Isabel Zimmerman and Davis Vaughn, two software engineers from POS.IT that work on all kinds of fun things. Isabel works on Positron . Davis does a lot of work with, like, Tidyverse stuff as well. That's super, super fun to ask questions about. So we look forward to seeing you there.