S3E4 Personalized Medicine with GenAI (ft. Justin Brueck, Endeavor Health)

Justin (Guest)   1:00
Yeah. Thanks Jordan. So you know, as a community based hospital we we we exist in a very unique setting I think and I bring that up to say that you know in a lot of academic organizations, they're really focused on kind of bringing ideas from bench to bedside through their own. Research protocols, right? I think what's very unique about endeavor is we we kind of have this idea that we we're not the experts everywhere. We want to learn from people that are doing things in the industry. I you know that. These podcasts are very important. It's it's great to learn and hear from others what they're some of their pain points and priorities are, but I think really as I consider innovation in our Community Hospital setting, it's solving for the last mile, the last quarter mile and honestly like the last three feet and. So as we've been looking at our innovation agenda for this last year, a lot of things are like just how do you actually get it into scale. How do you get out of this whole, you know pilotitis type thing? And and you actually have the ability to get something. To scale and so our team has had some really incredible wins this year. We went live with a bridge, which is our ambient solution that we're using right now.And you want to talk about impact? We went from having a pilot that was intended to go for, you know, six months or something like that. Three months into the pilot, we said we have to scale this just the outcomes from our providers in terms of the way. That they were talking about this relative to their own work life balance, actually getting out of the pajama time. You know, cycles that they're all in. And interestingly, we went from having you know, our first group of 50, we're now at 500 and I don't know if I've ever seen a technology necessarily that has had that quick of uptake and it's actually staying power. We're seeing our clinicians using this 8090% of the time for their notes and so really excited to be able to bring that technology forward. One of the byproducts of that actually has been that we're actually seeing even increases in our rbu productivity. Most recently, we did a a review and based on our pilot. Providers before and after we're actually seeing almost a 14% gain in their productivity from using this technology, so a huge, huge win for us on that one.

Jordan Cooper   3:11
I've heard a lot about ambient listening technology. I think there's a few different flavors out there. How is it? Well, you know what? I think that's been that's been something that addresses a variety of things. After COVID, you had a lot of provider retention issues. I think now that you have llm's large language models, there's more capacity to kind of help with a provider documentation. I've also heard that. It helps with provider notes. And then there's talk in the future about integrating discrete data elements directly into the EHR. Is that something that you've seen on the road map?

Justin (Guest)   3:52
Yeah, yeah. So actually we we are we are integrating this discretely into the note right now. What we haven't done yet and and it is on a future road map, is actually make it so that when the provider is having the conversation, it can actually tee up those orders and even begin to kind of pen those for them to just simply sign off on. I think that's going to be a really interesting space because now you're able to actually move into a world where you really don't even need the keyboard and the office. So I I I do see that being a huge area of opportunity. Other thing is. What's next? Best I think there's more technology coming out that's actually intending to in real time, almost provide the whisper into the physician's ear of hey, based off of what you're talking about, it looks like this may be blank. ABC condition. Maybe ask a few more probing questions around this and I and I think that's gonna really demonstrate how you're bringing clinical decision support into the real time conversation. So really excited to see what that does in terms of trying to, you know, help providers diagnose more accurately.But also as I think about it, it's on my research hat. It's the ability to say how do you take that real time conversation that's happening and identify that the patient might actually be eligible for a clinical trial. And now that you have these large language models that are able to understand all the criteria, inclusion, exclusion criteria you then? Have the ability to whisper to the physician. Hey, this trial may be something that's of value to that patient. That's something that you wouldn't have been able to do before. And I think because of how these technologies are saving time. For our physicians, I think it actually proves that's that opportunity to fill in the appointment with something that could be more value added like enrolling in a in a research day or a clinical trial.

Jordan Cooper   5:38
So you you mentioned something just now? You said, what if there was and this is a future potential capability of something like a bridge or ambient listening AI. What if the LLM in real time could go through the patient's EHR? Understand what sort of clinical trials are running at Northwestern and see if that. Patient would be eligible for clinical trials in real time. All the providers standing there but you also mentioned. The value of podcasts and having. Everybody else peers around the country interacting with each other. So my question is as fire is a standard AHL 7 or is a standard for sharing in different health systems around the country. What I like to ask is what if that patient is eligible for a clinical trial at a different healthcare delivery system? Is there any discuss?About populating a fire to pipeline about standardizing data formats about not just seeing if this patient is eligible for a clinical trial. Aisle and and and enrolling and engaging in research at northwestern medicine. But perhaps 1000 miles away, a listener to this very podcast says, hey, at my institution there's a clinical trial.How about in real time the patient get information? The provider get information about his eligibility.

Justin (Guest)   6:55
Yeah. So first of all, northwestern. Competitor to ours. But very yeah, you're good.

Jordan Cooper   7:01
Price it in Denver? Yes, endeavor.

Justin (Guest)   7:02
I'm good. But but what I would say, you know, at the end of the day, we want to have trials that meet the patients where they are, right. We we want them to be available because part of the issue that you have with clinical trials honestly is that if they're not available in the the patient's backyard, they're not going to have access to it. And a lot of these require follow up visits and things. Like that, but to your point about, you know. Patients. Should be able to access clinical trials regardless of if it's in our organization or not. I mean, at the end of the day, there are not going to be enough individuals in our organization to make, you know, every trial available. It doesn't make any sense for us to do that. So the question becomes, I think an important one is how do you allow that data portability so that that patient, if they are actually going to do that, they have the ability to take their data with you and to your point around fire? Agree with you that is definitely. A way that we need to begin thinking about that. I mean, a lot of this is going to be predicated on where, you know the the large EMR vendors ultimately position these types of of programs. But you know, for me, there's no benefit in us preventing a patient from being able to access what could be a lifesaving drug. And so whatever we can do to help facilitate it, we need to do it.

Jordan Cooper   8:13
So we're on a topic of AI. I'd like to pivot for a minute. I believe you said that you're setting up your own Gemini tenant. Would you tell us more about that?

Justin (Guest)   8:20
Yeah, so really exciting. We're working on setting up our own internal kind of large language model that's going to be focused on. Initially, operational issues and processes and things like that. So we've had a lot of use cases brought forward from our HR team about how do you create, you know, memos and also these different things based on all our policies or even our staff being able to query and say, hey, I've had an employee who has been. Absent three or four times this, what could be potential next steps that I would explore and have it actually surface those? Resources. So they're not trying to find it within a, you know, a busy website or internal Internet that I think is gonna be a very, very early early use case because it doesn't include Phi and it's really focused on just surfacing and summarizing policies. And you know if. You wanna put it into a poem? Go for it. But I I I think the next generation of this really is gonna get into the business insights and then ultimately how are you pulling this all together? So you're actually saying. The entirety of the clinical picture and physicians will have the ability to ask those big questions. So right now, like I said, we're in the early stages. I actually am getting my login later this week, so I'll be able to actually start playing around with the tech, but it'll be very I think it'll be helpful for us to be able to play with this and start getting it in the hands of others so they. Can understand and and help us design those next use cases, but really excited to have that up and running as a demo right now and our hope is that we'll be able to scale it into the clinical space towards the end of this year or early next.

Jordan Cooper   9:49
It's fascinating. We started this conversation by saying how do we go the last three feet when we have a successful pilot? How do we scale? I think a lot of listeners to this show are listening are wondering right now. Yeah, we have a bunch of interest in AI. We have lots of different experiments and pilots going on. How do we kind of accelerate that and move our own AI model training on on our own data, move it into production quickly and not get bogged down in committees? And I think starting with non clinical data without pH is a good way to just kind of get. Running as processes, IRB or whatever is viewing it in the background.

Justin (Guest)   10:24
Yeah.

Jordan Cooper   10:25
Also on AI I think that another kind of early win, I think you're working on digital pathology.

Justin (Guest)   10:32
Yep.

Jordan Cooper   10:32
Would you mind sharing with our listeners what you're doing there?

Justin (Guest)   10:35
Yeah. We just announced last week a partnership that we've established with Google and A and a variety of other vendors that is really setting a new standard for. Dicom and imaging for metadata surrounding digital pathology. And so for me this is. I mean, if you think about radiology, Can you imagine a world where you're still walking around with films? It's just unbelievable that we do that in pathology, though, and we do it with these little pieces of glass and you know with tissue on it. And so through a lot of leadership and a gentleman named Doctor John Groth, who is our lead of digital pathology here, we've actually created this first in the world. HL 7 compliant. All these different things standard that we hope that will enable us to begin offering this service. Not only to our existing providers, but hopefully extend it beyond so that we can, you know, essentially reduce the amount of waste that we have in our system and duplication of services because we don't have. The ability to offer pathology services at all of our location. So I'm actually very, very excited. This is a big investment. To be clear, the ROI on this is probably 10 years and so being able to have the support of the organization to take this bigger leap is is incredible. But now you're starting to see the benefit. I mean, if you take a look, there are studies left and right showing how AI performs in radiology. And how it performs even better at certain types of imaging and I think. We're going to get to that point in pathology very quickly. And so we've already done a couple test case using prostate AI, our AI that's focused on prostate cancer and identification of what would be maybe a a lower grade prostate cancer versus higher grade and the tools really. Do work and so I'm excited to see how this continues to evolve. And again, like now you're starting to train. The data on images in a way that I think is gonna be truly groundbreaking and for the entire. Industry and John is a is a thought leader. Also when it comes to how do you use this information to make it more personable for patients? And Can you imagine like you actually have the ability to do a teleconsole with your pathologist to explain your slides and your? Diagnosis again. You won't be able to do that for every single patient, but if you have a very complex being able to show those images to a patient, imagine what that does in terms of giving them the confidence. And the care that you're providing, but also allows them to ask questions of the experts. The person who actually is making the diagnosis, oftentimes in pathology and radiology, is sitting behind the scenes. They now will have the ability to do that live and I think we'll usher in a new way that we think about diagnostics in totality.

Jordan Cooper   13:11
Mm hmm. So you are the VP of Innovation. And I think I've heard a lot of other a few other guests on Healthy Data podcast speak about how there's one to 2% margins in most healthcare delivery systems and they're looking for new revenue streams, right. And some healthcare delivery systems are thinking about, well, maybe we can start developing our own products and productizing the software and selling it to other organizations. So I know you said big investment ROI in its digital path could be 10 years. But I wonder if you've thought about internal ROI versus external ROI selling at the other health systems versus just making it open source. What are your thoughts on that?

Justin (Guest)   13:57
Yeah, you don't. Actually, ARPA has a very unique program right now that they're launching that's around kind of creating a data Commons that would actually. Create a marketplace for data commercialization and it's it's in early stages right now, but we are absolutely looking at that because as we're generating all of this data, how do we put it to better use in terms of being able to train these models and ultimately improve those? Models. So there absolutely is an understanding that that is a dual mandate.

Jordan Cooper   14:22
Mm hmm.

Justin (Guest)   14:27
That we have is to think about not only what is the the impact, because to your point like this is a 10 year ROI. How can we kind of shorten that ROI by being able to leverage those assets in a responsible way? To drive, you know, new discovery and or new curation of these models. So it's, it's on our, it's on our Rd. Map it's definitely conversations that we're having right now. And like I mentioned, ARPA, I think it's called index is the is a program under ARPA is looking at creating this data Commons that would. Almost allow for a free capital market and setting up pricing on data and things like that, which is I think very they're very novel. I've never really had the chance to dive into it a lot, but if you think about it, it's almost like a a Stock Exchange for data and the the market is pricing the database off of that value of the data in real time. So a lot more to unpack with that. But if folks are interested, check out Arpa's website. It talks a little bit about it.

Jordan Cooper   15:22
I think our listeners can be fascinated in that we've touched upon partnerships with large organizations like Google before and many different episodes. We've talked about commercialization of data. We've talked about AI in general and but we haven't. This is a new one on healthy data podcast, so I think it's as fascinating about a means of generating new revenue for other health systems around the country. I think listeners will be interested. I do want to touch upon the topic of genomics. I know that's a topic that is of interest. To you, we haven't yet hit it today, so please enlighten our listeners. What's going on at Endeavor Health with genomics?

Justin (Guest)   15:58
Yeah. So you know, genomics is something that is near and dear to me. I've had the privilege of serving as the administrative lead alongside Peter Hulick and Mark Dennenberger, who are clinician directors and we have done I think. Significant amount of work to help educate about the value of genomics, but we still have a lot of work to do. I mean, if you think about it, 20% of your of your health is based off of your genetics. And yet we don't talk about it. And and I guarantee you that most of your listeners do not have their genetics done. They have never had a had a single gene test, let alone pharmacogenomics or whole genome sequencing. And so for our system, we're really trying to think about how do we break down some of the traditional barriers that were established by honestly, a lot of it was done by our payers and for good reason. Genetic tests used to cost $10,000 per test, right? Now you can get a genome for $100. And So what? Our system is really focused on that. Trying to figure out what the plan is to bring that to patients at scale. Right now, we've tested about 50,000 patients as part of primary care, which to be honest with you is is the largest program that's based in clinical genomics in the US as it relates to really offering clinical genomics as part of primary care. But we have 1.4 million patients and it will take us a very long time unless we add a zero to the number of patients that were testing every week. And so to me, the opportunity in front of us is how are we beginning to leverage all of the new technologies, the genetic AI to help us with consenting patients to explain results, to activate patients into the follow on care. Because the reality is that in genetics there. A wonderful thing that happens. You're doing the right thing for the patient, and you're doing it earlier, which means there's likely to be less costs associated with it. And honestly, there's better outcomes if you identify stage 1 cancer versus stage 4 cancer, the costs are dramatically different than the outcomes are dramatically different as well. And so for us, we're really focusing on moving from a reactive state of disease into something that is more proactive and really excited. Actually this last year in partnership with a company called Genomic MD, we actually created the first commercially available clinical grade polygenic risk score. And you know, polygenic risk score is A is a fascinating area and then can provide a little bit more perspective. On that if you're interested. But the reality is that all of us have apologetic risk score for every disease out there, every single disease. Knowing your polygenic risk score ultimately will allow us to be able to triage and take care of patients based off of their own genomics. And if you think about. It you know, I'm actually my mother, is adopted. And being that she's adopted, I don't know that entire side of my family history. But if I had my genetic screening done and I'm getting these polygenic risk scores and all these other things. It's giving me insight to something that I don't see today. And so you know, I I think this is a fascinating area that's gonna continue to grow. I was interesting. I was just looking at something. I think it was Deloitte was listing their top innovations and for the last several years it's been AI, you know, is #1. Genomics actually went back to #1, and I think that is beginning to show the maturity of genetics, but it's also demonstrating the value because I think, you know, AI has been a little bit hyped. But being able to actually triage and manage patients. Based off of their genetic risk, is the next frontier of medicine.

Jordan Cooper   19:35
Well, Justin, it sounds like Endeavor Health is all about the triple aim of improving outcomes, improving value, reducing cost, improving quality. So I'd like to offer you an opportunity to say a parting message to our listeners. We've covered a lot of ground from ambient listening, early wins with Gen. AI, talking about standardizing data, making Gen. AI. Patient visits, providing real time feedback from the EHR using. Using llm's to the provider at the point of care, we've covered the value of genomics and how to transition from being reactive to proactive in providing care. And we've covered AI with digital path, even an opportunity potentially to create a Stock Exchange for data. Any parting thought? To anybody who is titillated by this laundry list of innovative ideas.

Justin (Guest)   20:26
Yeah. Yeah. I mean, the thing that I will say is there is so much opportunity and potential for us to change the way that we deliver care. And you know, I I view health systems as being in a very unique position to make an impact on care delivery, no doubt. But I think we're also at a unique position to start thinking about our role as truly an orchestrator of care. And understanding that you know, I I joke around this. I have six kids. So like we go grocery shopping and we shop at Costco. So we shop at, you know, jewel. We shop at Trader Joe's. We go everywhere, right? And in healthcare, we're very hesitant to embrace a world where it's OK for me to get certain services at different locations and through different companies. And I think for health systems, the next iteration of this is how do we create those partnerships that allow for the interoperability of. The data so that we're not creating any waste in the system. But we start getting patients where they need to be seen with the right modality. You know, I recently was looking at getting a colonoscopy and looking at how long it takes to get into the to have a colonoscopy. The reality is, like with my genetics being I have a low polygenic risk score for colorectal cancer. I don't have any of the inherited disease factors related to monogenic. Could we actually begin thinking about positioning Justin as? An individual and not going down the traditional route of colonoscopy, but instead maybe doing something like .Like multi cancer, early detection, testing or fit testing, or Cologuard. And I think that's really the next iteration of this for health systems is being willing and comfortable to look at what's in the best interest of the patients and understanding all the different partners that we can bring to the table to give them that best outcome.

Jordan Cooper   22:16
Well, yeah. Did it Justin you added yet another topic. Personalized medicine based on genomics. I think this has been a fascinating conversation for our listeners. This has been Justin Brick, the system vice president of innovation and research at Endeavor Health. Justin, I'd like to thank you very much for joining us today.

Jordan Cooper stopped transcription