Lessons from a Physician-CIO on AI Governance with Dr. Stacey Johnston

Welcome and thank you for joining us on today's AI Explained. I'm Krishna Gadeh and the founder and CEO of Fiddler AI. I'll be your host today. You can put your questions in the Q&A box anytime in the fire search chat. Today's session will also be recorded and sent to all the attendees after the session. We have a very special guest today on AI Explained and that is Dr. Stacey Johnstone, CIO and Digital Execution Officer at Beacon Health System. And today we are exploring with her on AI in healthcare and ROI, governance, controls, observability, all of those fun things. Welcome Dr. Johnstone. All right. Thank you so much. Johnstone, John Beacon in October 2024. She most recently served as the Vice President and Chief Applications Officer for Baptist Health in Jacksonville, Florida. Prior to that, she was the Chief Medical Information Officer at Beaufort Memorial Hospital in South Carolina. During her time in these roles, she was always maintained and active medical staff privileges, worked as a hospitalist and she classed to continue this as she feels it gives her a unique insight into how technology and care intersect. And prior to this, she received her Bachelor of Degree from Washington College and Dr. Ruffman is in degree numbering. So great to, you know, great background. It's great to see, you know, technologists coming from, you know, different backgrounds and very excited to host you here Dr. Johnstone. Well, again, thanks for having me and now my camera is working. So we're good now. Awesome. So I guess let's start with that, right? Like you started as a clinician. What moment made you realize that healthcare needed systems level change and not just, you know, bedside care? Yeah. So honestly, a lot of this actually started straight out of residency. So, you know, back in the day, we were still on paper, paper records and then, you know, we were switching over to electronic health records and residency and, of course, at that time, many attendings were not super excited about it. So of course, they gave it to the residents to try to figure it out. And even the residents were raising concerns and saying, you know, one saying I've heard a lot over my tenure in this role is, you know, we are the highest grade secretaries out there. And so this idea of putting in our orders, typing them in, you know, which just went against everything a lot of people were trained, you know, to do at that time. And so, and so what I realized was either I could buck the system. I mean, this was coming whether we wanted it to or not at this point in time as a regulatory requirement. And so, you know, I could either buck it or embrace it. And so what I realized was if I embraced it, it made my life easier. You know, the more I embraced it, the more time I spent and developing my documentation templates or my order sets just kind of made my workflow easier and therefore made my colleagues work flow easier. So at that time, I was a family medicine and then became an attending as hospitalist and interestingly enough, I went back to paper. And so then would let a second round of implementation of paper to an electronic health record by that time. I was the only person in that health system at Bupert that had been on any sort of CPOE. So next thing I know I became the chief medical information officer and what it, you know, that was that mindset of, okay, if we spend time in the beginning really evaluating our workflows, trying to make them better, you know, by using the systems, how can it, you know, how can it save us time and the long run? And so versus if we don't put what it called care and feeding into the EHR, then it's just going to, you know, all that pajama time, all of that, you know, that was happening in real life. And so I really spent a lot of time on the hospital's workflows and on the order sets and, you know, automation and using Dragon Medical One and the templates. And so really building up our templates, really building up auto text. And so the more we could, and that, that was the version of automating back then, you know, was templates. And even the Dragon Medical One and then we went to the DMO mobile app. I mean, we're like, oh my gosh, this is amazing. And like AI wasn't even, you know, a thing that thought, yeah, I mean, we know and even thought that that was even possible. But what I realized was, you know, taking the time to make it better for the physicians ultimately ended up making it better for patients, you know. And decreasing time in the EHR obviously is, you know, that EHR can contribute to burnout and burnout obviously leads to a lesser quality care. And so we really, you know, initially that triple aim that you might have heard about is, you know, good quality care, reduced cost and, you know, patient, more about the patient. And then adding in the quadruple aim was really talking about enhancing the caregiver experience. And then we really took that to heart and, you know, making the caregiver experience better. And then now they're talking about the print up will aim, which is equity and care. So, you know, I still think we have a lot of room to go, especially in those last two components, you know, enhancing care, give returns and then in the equity of care. But I really think that, you know, now in this AI era, we are going, we're just getting the glimpse of it, like just a little taste of it. You know, we're doing a lot, you know, radiology is embracing AI pretty well. There are some cool notes I'm a riser type of things, but we're not full AI in the clinical space yet. And I think once we get there, you know, of course, through, you know, change management and appropriate data and trusting of the systems, you know, we're still, I think, you know, maybe a year or two out from that. But once we get there, I think the sky's limit. And I think the docs are really going to see time spent in the EHR is decreasing. They're going home on time, you know, they're, they're starting to find joy in medicine, you know, really embracing ambient, but more than just documentation, you know, from the ambient listening, but really doing your full workflow with an ambient solution, I think is, is ultimately where we need to go. No, it's great to see technology helping practitioners that way, right? Now, it comes to, you know, no, so it's an interesting, you know, because you have this two kind of speed level experience in some ways they're connected, but what do you see, you know, like technology, spirit plate technologies, fundamentally misunderstanding and work clinical. Yeah, I, I think ultimately, on spirit heading technology is, it's not a technology issue. It is a workflow issue. It is a change management issue. It's a training issue. And so if you do not understand and embrace the workflows, if you don't understand the needs for a change management, or if you don't really have a good training program around these AI solutions, then you're just going to put them in and they're going to fail, you know, and so there are going to be a lot of times where we, you know, where we implement and it's not meeting our needs, and well, why isn't it meeting our needs? Well, chances are we put in something because, you know, maybe the operator's thought it would help solve a problem, but ultimately, you know, one of the things as being an informatic system is you learn, you know, to ask the question, what problems are you trying to solve? And maybe the problems of the physicians aren't necessarily aligning with your hospital operators. And so you need to make sure that we're in alignment from the system, what problems we truly are trying to solve. And that, you know, if you can get your, you know, your service line leaders, your physicians, auto-mean alignment, and, you know, maybe it is about access for patients, and maybe it's about, you know, physician burnout, but whatever problem you're trying to solve, we need to make sure that you're aligned from a system, and you have a singular system approach to that problem before trying to deploy the solutions. Interesting. And now we are in the biggest technology shift, not having a right to know everyone is saying that AI will transform healthcare, especially in a, and from your perspective, you know, what's actually working today? You know, is it still kind of PowerPoint deck? So are there any interesting applications that you've seen work, you know, and giving out of eye? Yeah. So, you know, I think most health systems are pretty, are advancing themselves in the back end pieces, you know, automatic coding. You know, we are pursuing fully autonomous coding, which I think would be, you know, a game changer for us. Right now it's still mostly an augmented coding aspect. We are moving towards fully autonomous verification of benefits. We're an agent, AI solution is going to do all of the verification benefits. We're also looking at agentic call return. And so agents actually participating in, you know, voice call returns and, you know, interacting with the patients. So there are some really cool solutions out there doing that. Other areas where we've actually expanded into agentic use, fully autonomous agent abuse is when we converted, we actually just did a four-hospital acquisition recently. And we put them on our tech stack in seven months, which was pretty aggressive. But you only get this system available in a backload appointments. We got it about two weeks before we went live. And two weeks to backload 100,000 appointments is not necessarily enough time. So we had to think differently. We could either hire 40 people to do this or we could use some HNTKI. And so we were able to basically do all of the ambulatory backload appointments with an agent. It took us about three weeks to build the agent. And then two weeks from that time that the system became available to backload all of the appointments. So that was game changing. And that was fully autonomous agent TKI and doing all of our scheduling for our patients. So I think that's pretty cool. We're looking at some digital FTEs and any all these repeatable processes. We have a pretty good robust robotic process automation program. We have a good intake process and a governance around the RPA. But imagine laying an agent TKI solution on top of your RPA. We're not there yet. But I think that's right around the corner as well. So those are all of your back end processes. Then when you think about the clinical aspects, we actually have a white paper that has been written about agent TKI and colon cancer screening. So really embracing the use of colo guard and as an initial screening instead of automatically going to a colonoscopy. And a lot of this has to do with access to colonoscopies. And colo guard is now being viewed as a good first solution. And so we built an agent that automatically ordered colo guard for this group of patients that qualified. So we ordered 7,000 colo guard screenings and ended up getting back. I think it was about 40% returned and ended up doing 250 colon cancer screenings additional with a colonoscopy. And we did actually have one positive colon cancer return. And that was caught early. So had a reception and full survival. So I mean, that's just-- that is game changing clinical care where an agent TKI solution can actually help prevent further mortality. So I think that's pretty cool stuff. Wow, that's awesome. So automating appointments to clinical diagnosis. So maybe you can give us oriental. So what's one use case that's often overhiked and also what's like a thing that is underappreciated, both in your experience that you've seen with AI? Yes. So one use case-- so obviously radiology-- a lot of the radiology use cases are fully embraced. I don't know that I don't know of any organizations that are fully autonomous with radiology. Yet we are using it as an augmentation to the workflow. So I think that's one that isn't overhiked, but is definitely used throughout multiple health systems. And one that I think is underappreciated is the augmented responses to the in-basket messaging. And I think part of it is just understanding the large language model and having to say, well, I have to modify the message anyway. But yeah, modifying a message that's already written up is very different than typing a brand new message. So I think that's one that we should continue to explore and push ourselves on. There are studies out there that show that-- especially our primary care docs-- are getting 85% more in-basket messages from the portal. These portal messages are since COVID. And so that's additional burden. Again, the cognitive burden for our physicians. But yet, also, it's not being reimbursed. It's not just additional time for our providers. The other thing is we need to look at the pools for these messages. Do this particular message really need to go to that provider or can it go to an MA, can it go to the front office staff? So I think routing the messages better and then when it truly is a provider message, having the AI help queue it up. I think also some referral management is not as well utilized with AI. We're diving into that refill management. So refill medications are really feel like as an opportunity with AI. Again, cognitive burden. What is a repetitive task? And so can the AI queue up the refills that are ready so the docs aren't having to click on those boxes all the time. So there are some solutions out there that are during some pretty cool stuff for refilment. Awesome. So we touched upon a few things, right? You know, you leverage in large language models to respond on behalf of doctors or even augment, you know, their answers. Now, one of the problems with LLMs is that there are a bunch of risks, you know, they can hallucinate at times, you know, they could, you know, leak PIA information and all of that of BHI in the case of healthcare. How do you, when you joined, you know, beacon, how did you approach this problem of AI governance and where did you begin, you know? Yeah. So great question. So we have begun upon my arrival. Didn't really have a good, strong IS governance process. So it was, IS was either the approvers or non-approvers. And it was kind of just tickets were submitted. People didn't hear back. And it was just not a great process. So that's the, actually the very first thing that I did when I came here was stand up in IS governance process. So we have an executive steering committee. We have eight different advisory levels. And then we have multiple work groups that report up to the advisory councils. One of the advisory councils that we did stand up is an AI council. And in that AI council, we developed two AI policies. One's basically an AI policy, meaning that you don't go out and just sign up for AI. It has come through the advisory council for approval. And that, you know, if you're, if you, we have, if you use AI, it has to be like a system-approved AI. So kind of more about like how to use AI. And then the, and then we have a permitted and prohibitive use policy, meaning I can't put a PHI into chat GPT or into open evidence, something along those lines, but that we do have permitted clinical uses, which is more like augmenting the workflow. So there's ambient listening, you know, the clinical oracle's clinical AI agent, but you might use a bridge or, you know, DAX co-pilot. So, you know, some of those solutions that, you know, are augmenting your workflow. So you're still reviewing and you're signing off on that documentation. So in the radiology space, it's an augmentation for the workflow because it cues up the severity. So, you know, if they get a, you know, hundreds of CTs a day, it will do the read first. And then it will queue up to the radiologist. These are ones with results that we think you need to review first. And so it actually helps them become more efficient because they can then, you know, read that the most urgent reads first and then go to that, you know, more of them maintenance reads. So there's just a lot of cool things that we can be doing, but we need to make sure that we are governing it. So at first, it comes through the AI Council for approval. There is a form that we created that the actual company submit and fill out themselves. So it's about, you know, what is your data modeling look like? What are you using on LLM, which model are you using? You know, what are you monitoring for DRIPT and bias, where you storing the data? You know, so basically some vendor, it's like a vendor risk assessment, but from an AI route. And then we are also going, we've done double down on our AI literacy program. So now we have, we built out an AI literacy program and it's required, prior training for our managers on up, which is essentially to help us look for draft or bias or anything that is starting to help us like manage these AI agents, you know, as you might manage an associate. You need to start managing these agents themselves too. And then it also helps us understand like what's a permitted use, what's not permitted. So that we don't have our individual team members doing something with AI that they shouldn't be doing. And then one last component of governance is, there are a few solutions out there that are doing AI monitoring. So overlaying another solution on top that is monitoring the AI for DRIPT for bias and then they would escalate their concerns to the leadership if something seems a mess. So it's kind of like you have these security for monitoring for network and it's basically another security metric, but it's monitoring for AI use. Yeah, makes a lot of sense and very relevant. And to today. So I guess fundamentally, it all comes down to trust and safety and governance. What does trust in AI look from a hospital board's perspective? Because as you think about potentially AI recommendations conflicting with the initiative management or AI maybe standardizing too much across and losing position in things like that. So, you know, how do you view of this world of like trust in AI from hospital perspective? Yeah, so I think the administrators and operators are the leadership is most most likely to embrace AI and trust it at this time because they've seen it successful in the back end processes. So for instance, our revenue cycle team, they're on full on AI. They're like the more AI can layer on, bring it on. And so, so they've actually been great to work with. They are actually kind of leading our organization with the amount of AI they've embraced, which then made it easier to tell our referral team, hey, we're going to embrace AI for your referral management. And okay, now now we've gotten that. So now let's dive more into the clinical space. The AI that's more of an augmentation tool workflow. So as I mentioned in basket messaging, the clinical AI agent, we we were really adopted the clinical AI agent, which is our ambient listening solution to queue up the documentation. So about 70 to 80% of our providers and the ambulatory clinics are using the ambient listening solution for their documentation. And about 70 to 80% of their notes are generated with the ambient solution. So, so it's being very widely used. Part of that was we train them. We spent a lot of time training them on how to use it to the to help them become more efficient. And they saw their documentation time go from seven and a half minutes per note to two and a half minutes per note. And so the fact that they saw the value in it and that they were starting to see that the note was actually producing a good quality note. So they trust that. Now if I were to say, I'm going to try to turn it on to be fully autonomous and you don't even get to read the note, I don't think we're there yet. They're not trusting it enough. And so you, you know, so you've got to move the needle a little bit, have them trust that one component and then move the needle again, have them trust that component. And so it's just a natural trust that builds on top of each other as you start to expand and expand more. Very interesting. And, you know, like I think this is a great ROI thing that you know, sort of mentioned right now healthcare industry runs on thin margins, you know, you probably have to justify your investment. Where have you seen like the path fastest be back, you know, the cost reduction, clinician productivity or new revenue. Yeah, so I think that's still kind of for us on the back end pieces. Although I do have to say the clinical AIA agent, you know, I remember when we first deployed it at another organization, it was gosh, I don't know if we can afford to play it. Do doctors pay for it? Does the hospital pay for it? And there's no ROI, you know, maybe they might see one more patient per day, maybe they won't, you know, and so, but what we have actually found is we've been live for about a year and a half and in, you know, over a 12 month period of time, we had an extra $10,000 per doc revenue and just better capturing the documentation of the visit. And so there was a definitive ROI by going live with clinical AIA agent. So then we're like, that's an over a, let's just keep rolling that one out. So, so in the other one. What does this clinical AIA agent do? Actually, it's like the average, it's within Oracle built it, but you know, it's like the code, you know, DAX 360, you know, the co-pilot, the apron, what ambient, you know, it is conversational AI that you have with your patient, the provider and the patient that then cues up your documentation. Let's review. But that one's built within Oracle by Oracle. And the other one that again, received great ROI was the colonoscopy. So why not fully, I mean, the colon guard, why not fully automate that? Like, why can't that be a fully autonomous decision? That if we have based off of your inclusion criteria and we know your exclusion criteria, why isn't, why isn't an agent just automatically ordering the colobard screening? You know, I think that's a great use case. And then, you know, for all of the decreased colonoscubes, unnecessary colonoscubes, and increased in appropriate colon cancer screening, which is tied to a lot of the four and five star metrics from your, your Medicare advantage plans. So, you know, why not do something like that? Yeah, it makes sense. And as you bring these multiple tools, like you mentioned, clinical agents developed by Oracle or average or other other systems, are you seeing like the need for a centralized control plane emerge where you are able to track all of these agents in one place and how know, how they're performing? Yeah, so that's, so AI reports to me. I've formed a department of AI and transformational technologies. And we are a centralized process at this point in time. However, you know, the recommendation and the studies are showing that you're more likely to embrace AI, it's more of a federated model, you know, more of a decentralized model. However, I still think the monitoring should be centralized. And so, what does that, you know, federated decentralized model really look like? Well, that means that, you know, you've got your department's owning, you know, just kind of day-to-day monitoring of the AI. Does this still make sense? And my, you know, what is my adoption rate? Are these metrics meeting, you know, my needs? Is this AI no longer needed? Do we turn it off? I mean, that's something we never ask ourselves in technologies. Do we turn off these systems? And so, I think the monitoring still should be a centralized process, but how well you adopt and deploy the AI agents or, you know, any RPA or anything like that. I think we can start to move towards more of this federated model. One of the things we're talking about developing, and there are some organizations that are doing this really well, which is this kind of this AI citizen program, which is you can start building out like low-code, no-code yourself. And so, we have, you know, we have Microsoft co-pilot. And if, you know, with a, you know, that higher level license, you can actually start building out some of your agents. And, or do we use that? Do we use UI path, you know, where you can maybe build out your own bots? And so, how can we actually, through an approval process that may be more centralized, then give people the tools to actually build their own bots? I think that's where we want to go, but we're not great there yet. Yeah, makes sense, makes sense. And then, like, as part of this control panel centralized observability, you know, like, is there a need for a runtime AI governance? Like, almost like the three-dacks BII before and from, you know, I think this is an audience question I'm reading out, the three-dacks that PHI before any information enters them. Yeah, I mean, obviously, I think you have to be real-time because, you know, without that, we won't be responsive. So, it is, it's critical for you, you know, you need to have kind of your standard processes, but there are going to be times and needs where you need to be a little bit more agile. So, how can you have a, you know, a governance that's, that has a standard process, but yet as agile, I think that's the sweet spot. And I don't know that a lot of organizations are doing that really well. We're not quite in a fully agile yet. But I think having our dead, having a dedicated AI team has actually made us much quicker. Hey, interesting, interesting. And then, another related question I'm just taking a couple of audience questions here of, you know, there are so many that at least keep coming them. And, so, there's one more question from John Sacriss on, uh, hi, Dr. Johnston, how does it be can ensure that he's robust in their tool evaluation while not acting as a bottleneck for innovation. I think this is kind of that trade off between innovation and governance. Yeah, no, that's, that's always hard. I think that organization is definitely clamoring for more AI, but, you know, I brought up in our large system meeting this morning that, hey, we can go as fast as you want. I just need more people to do the work, you know, so, um, so is, you know, are these tools worth their investment? I think so. But, you know, if we work on these AI solutions, what are we not working on? So I think that's kind of, but, you know, so we need, we need to understand the system priorities. Um, but also as we move into this more, um, federated model where if it's kind of this low code, no code, um, or, you know, a simple bot build program, then that, I think, helps us become more innovative and that people could actually be building some of their own. agents are building their own baths. That's I think where most organizations are ultimately going to have to go and the other thing is that do you buy or do you build and we're doing a hybrid of both. So if it's a really complex workflow or if we need it turned around really quickly we're going to buy. We're going to partner with someone and we're going to say hey we need this agent built in two weeks. Can you you know we have it mapped out you know you can you build that for us versus if you know if it's a simple workflow you know why don't we build that ourselves within our team. So you know I think I think it's a you're cannot almost have to do all. You're gonna have to be innovative and have a process. You're gonna need to be agile and also look at cost reduction. I mean you have to you have to kind of do all and so you know what I brought up this morning was I think you know yes we might be spending more now but in the long run it's gonna save us more money by hopefully you know reducing FTEs or honestly where we are is we can't even fill those FTEs who are paying contractors and so rather instead of filling it with contractors can we fill that with an AI agent that can actually help reduce the workloads you don't need as many FTEs. So for even it's not for us it's not about reducing your FTEs it's just about you know filling the positions we have open or reducing our contractors spend. That's very interesting and so I guess like for other you know AI leaders like you working healthcare what's your advice like what do they how do they think about you know bringing new AI tools to deployment what is the one thing to put in place you know before they deploy anything in terms of AI you know they're in their environment. I do think having in the AI governance council and advisory council makes sense and so we have physicians sitting on it vice presidents, hospital presidents, we have ethics, we have compliance, we have risk, we have security, I ask leaders so you know it's kind of the gamut of you know what's gonna have OHR so you know we have the kind of the gamut of who needs AI and who should be responsible for AI and that just helps to ensure that we have some responsibility for our solutions. But it also helps us understand our training and needs, our change management needs. So I think that's the first thing and then you need to understand you know the other thing we did very early on was an minimum understand what you're permitted and how habited uses are are you going to be you know what is your risk for fully autonomous in the clinical space? Are you there yet? Are you not? You know you need to decide as an organization and then I would you know so those are the first two things I think you need to do first before you really fully expand and then I would 100% what I did of standing up a department that this is all I did because if you add AI on top of your team that's already busy with code upgrades, maintaining your systems, the break fix you know the you know the request from you know I need to change this order set, this documentation template you know we have now joint commissions coming so if if your team that's doing that day-to-day maintenance and you know of your systems and then they're also responsible for going out there and and being a partner for implementing AI I think it's going to be very hard to move the needle. So I think having a team that they don't need to worry about the money their job is just to go and meet with the business partners go they're going at conferences they are you know meeting with other organizations they are out there just scanning the environment and finding out what's new their meeting with venture capitalists we're meeting with you we're partners with Notre Dame so we meet with Notre Dame frequently what are they working on rather they're seeing you know trying to we are setting up some internships with Notre Dame so they can actually do some development in our environment too so before you get there though you know it's kind of that you know crawl walk run philosophy so you've got to crawl for or so choose a couple of good select use cases that you are fairly confident in before you start to you know walk and run the other thing that we do I forgot to mention what the AI policy is we only approve at the advisory council if it has a defined ROI if there is no ROI then you need to ask yourself why are we doing it but if it's soft ROI and it's really going to give back you know back to the clinical experience then it goes to the steering committee meaning that you know do do the executives so the C suite do we buy in that we want to give this to the clinical end users even though we may not get any ROI from it and so we haven't actually had anything that we've approved without a defined ROI so right now you know finances are tight for all of us so you all should be looking for ROI first with your AI before really going to the non ROI and it's that's that makes sense so you got to measure your ROI continuously and evaluate how these things are working I guess that's important yeah so yeah there is winning the confidence of clinical staff right when it comes to introduction of new tools especially AI right what has been your journey in terms of doing that that because you know how will you success with all the new lessons learned and I'm sorry it broke up for just a second could you ask the question yes and the question was basically how are you you know one of the problems in healthcare is like how do you win the confidence of a clinical staff and who might be skeptical about AI and the tools and and you know what has been your experience doing that at B.C. Yeah so I think having key physician partners understanding their use cases so I actually spoke at a physician meeting last night kind of talked about the future of healthcare and future of healthcare technology obviously AI was probably a number one thing we talked about and people were you know somewhere reticent but somewhere like hey I have really great ideas I'm like cement them to me you know if it's a low hanging fruit we could put that in you know from the time of request to an agent being built is you know probably about you know forward and forward and eight weeks depending on the complexity right and so why not you know if that if there's gonna be ROI there I don't see why we couldn't do that if it's obviously more complicated workflow or requires an integration into our HR and that's gonna be a longer you know ask but the but you need to deliver on your promises and then they need to see the value of what they you know what we've turned on so you know one of the things I always say is don't implement AI for AI state to say you're implementing AI but it's gonna add it's gonna be meaningless to the docs or make sure you know where that data is pulling from so again if you have you know different data set definitions and you're pulling from the wrong data sets the data you know the docs aren't gonna trust the data and therefore they're not gonna use the systems and it needs to be easy to use you know just naturally intuitive it needs to be part of their workflow asking them to go out of their EHR to use another system that's gonna also not help your case and so it needs to be seamless you know integrating into your system as much as possible and and it just needs to provide some value to them either you know they can see more patients or they go home earlier they less cognitive burden you know one less click you know throughout a day adds up you know a significant time saving so you know that you're gonna gain the trust by doing solutions that are easy to use for them awesome yeah and I think once you're able to trust your AI then you can start thinking about customer-facing AI as well right that's actually a interesting you know question from an audience you know how is there how are the content why are you tracking the concerns you know how are you sort of all coming the concerns about AI agents directly work you know working with patient and and sort of a and potential patient reactions you know dealing with AI agents you know what has been you know I don't know if you're there or like how are you thinking about it we're looking at it I mean we are investigating the AI agents making phone calls and so you know studies have shown that empathy from an AI agent call is actually higher than you know a person calling and you know so so that's probably where we're gonna start with our AI true interaction with the patient now there's other things that AI could be doing and that I'd like to pursue which is really AI detecting kind of emotional or mental crises you know when they're calling to make an appointment there's AI solutions out there that can detect you know suicide or risks so is that something that you you know want to look at as well too but I think you know I think AI agents calling patients is pretty cutting edge you know the other thing that you know AI can be doing there's a lot of solutions out there that for in your inpatient stays you know, AI and falls. A lot of people are doing AI with all predictive modeling with camera used. But what about if you could take that and then that if the, you know, the patient is about to stand up, the agent itself calls into the room and says, "Hey, please sit back down." Or do you need me to call your nurse for you? You know, so having that agent interact with the patient actually in the clinical setting in the care room, that's out there. I mean, that we haven't done that yet. That's definitely on our road map. Very interesting. Very interesting. I guess, coming back to sort of co-pilates and autopilates in medicine, you know, apart from the administrative assistance and clinical diagnosis, where do you see the world going? For example, if you have to rebuild, you know, the entire healthcare system from scratch today, you know, with AI, what would you do differently? Yeah, so it's interesting. So Oracle is rebuilding their total EHR with this new AI-smantic layer. And we're going live with it in our ambulatory clinics this summer. So at least in pilot, you know, a pilot a few clinics. So it'll be interesting to see if that significantly changes. But some of the AI things that we need to embrace within the EHR, again, the augmented responses to the patient messages. Number line. Number two, summarizer. So kind of what happened between from the time I saw you, you know, six months ago to now, you know, was there admission? What, you know, what images were done? Did they follow up on any other care gaps? What care gaps still need to be done? You know, so basically kind of what happened to now and help keep you healthy? Like layering the EHR on top of a large language model to be able to queue that up combined with the ambient solutioning. So then you can review the, you know, the note summarizer and say, okay, I see that you, you know, visited that AD for Chessmean. Did you follow up with a cardiologist? No, I didn't. Well, let me place this referral for you now, you know, let's get your referred. Do you have a particular cardiologist you on? Let's go ahead and order an echo and, you know, follow up with me in a month. That conversation should be actually placing the orders as you're speaking. Instead of you then getting the document right now, it's just queuing it up in the document. You read the document, you submit it, still see the time. But wouldn't it be great if then it also actually placed those orders or at a minimum, queued them up into your in-basket so that you just had to sign off on them instead of actually having to manually type in the orders. And because it was using a large language model, it could review to make sure you have the right CPT code and the right diagnosis code. And, you know, and it also has access to your referral management system. So it knows which doctor to send it to in the right address. And then a patient gets a text, you know, 30 minutes later and say, hey, we got your referral process. You know, please, you know, we'll text you again when we have an appointment made for you. So, or text in here, you know, use your texting solution to choose which time you want. So that's where we need to go. We're not there yet, but I definitely think there are some organizations that are doing bits and pieces. Like we're doing the texting and the online scheduling. But we are not fully the notes on riser. We won't go live until the summer. Interesting. Interesting. And there was also a question on variables from the audience are, you know, from Melike, Bunar, Dr. Justin, Howardian, Visioning, utilizing lifestyle variable data from patients in their care, variable devices like smartwatches, rings, wristbands, continuously collect data. And then how can this data set me utilized by leveraging the eye to enhance patient care? I mean, great question. So, I mean, that's where I think this guy's the limit. So, you know, we've got all of this, this IoT world, you know, so we have all of this patient data. So whether it's a continuous glucose monitor, blood pressure cup, you know, we've got, you know, these telemetry devices, your smartwatch, you know, so all of this were available data, right? And so coming in. And so it can come into the EHR and adjust it pretty easily. That's not the hard part. What the hard part is from where I started at the very beginning is the workflow around that. So if you have someone's whose blood pressure is reading at, you know, 200, what are you going to do? Like who's monitoring that alert? And does that alert come to the primary care doc? Well, what if he's off that week? Does it, you know, does it go to a key management list? You know, because we, you can embed it into a registry. So, this or someone that's always monitoring this registry, or is it a nurse practitioner that's just monitoring, like managing a hypertension clinic. So there are some organizations that are doing this really well in the wearable space where we are not there yet, just being totally honest. But, you know, I do think the wearables are one component. And I think the other that I talked about last night, like the physicians is genomics. On genomics is, you know, I think the studies show that every three or four out of five drugs now have some sort of genomic marker being tested as it's going through FDA approval. And, you know, depression meds, antipsychotics, blood thinners, I mean, a lot of medications right now have some sort of genetic testing. And that's a additional cognitive burn to our docs. There is no way they are going to remember everything that needs to be done for your patients at any given time. So, embedding that genetic data to alert you and say, oh, you're about to order products. So this patient isn't a candidate for products because of their genetic profile. Please consider one of these others. And then the solution actually queues up what the other recommendations are. So, I think between both the wearables and the genetic data, you can now move from predictive analytics to truly prescriptive care, which is the right patient at the right time with the right care, by really surrounding them with the data and analytics and the AI on top of that. Wow, that's amazing. I think that gets me into the future of health care and related. I'd like, fast forward five to 10 years. What does a hospital look like with deeply embedded AI, you know, according to you? Yeah, so I think what will end up happening is, you know, there's a lot of consumer grade things that are happening right now like blood pressure cups and, you know, and then, you know, all the wearables and even telemedicine visits are, you know, I think I think I'm going to continue to expand. So I think in the primary care space, you're going to actually see sticker patients and the less sick patients are going to just kind of do their care at home. You know, even prescriptions that require an actual visit and a prescription can now be bought over the counter. So there's all, you know, like, um, Omepers all used to be a prescription only. Now it's bought over the counter. So there's an also with a large language model and chat to VT people are going to start to self care at home. I think and then and then it's going to become sticker patients going to be seeing that primary care office. And they're going to manage those at home as rest as possible. And therefore, your inpatient days were going to be even sicker. And so, you know, I think you're what we used to see in that inpatient space, some of these people that we would keep a little bit longer because they need some extra auction. And we're going to take care of those patients at home. And that that truly would then leave the beds for the the sick as of the sick. So I think we're going to really see the acuity in our hospitals going up as well too. Um, but on the flip side, I think we're going to have more automation. We're going to have, you know, these cameras in the rooms and the AI predicting the care, um, you know, AI right now without without an actual blood pressure cuff is able to start detecting blood pressure or temperature, heart rate, um, just by cameras. You know, so I think there's just more we would be able to do with the technical aspects and the technology surrounding, you know, the patient in the individual hospital room. So that, you know, that hospital room as a future. Um, so that so you're monitoring these sicker patients, um, kind of more around the clock instead of relying on people always going in and looking at them all the time. And are they going to be some new rules in healthcare because of these changes or they're like, are there's going to be job, job shifting as well happening in healthcare? Yeah, I think we're going to obviously continue to have to upskill our folks. Um, so, you know, everyone's going to need to practice to their top of the license, but don't forget we're also going, you know, already in a nursing shortage and a physician shortage. So how are we going to face both, you know, the clinical caregiver shortage, where we're going to need to upskill our CNAs, we're going to need to upskill our, you know, transport team. So, you know, the more we can upskill the folks that we may have, um, that they can maybe go into some of these other roles. Um, I know some hospital systems are looking at, um, you know, partnering with community colleges that, you know, they come directly out of community college to go into some of these, um, you know, these roles. But I think upskill and definitely upskilling on your IT team. So how to manage an AI model and, you know, building, are you, you know, a builder, a bi program, and, you know, your smaller community hospitals make you need to be a bi program. I think I'm a person in managing a large language model is probably not going to happen anytime soon. So that's why we partner, but, you know, um, I, you know, upskilling your team, but upskilling the clinical people as well too. Okay, so we're almost at the end of the webinar. Let's actually wrap this up with like, uh, you know, lightning rail. So like, you know, one or two word answers, Dr. Johnson. So I ask, quickly ask these questions, you know, one AI, I'm in health care for you. Right? No. Then I think it's going to get rid of the caregivers anytime soon. And the future, maybe five, 10 years from now, but right now, we're not seeing that. Okay. Great. And one company doing it right? We've, I, I had to say right now, I'm really impressed with the Oracle and their clinical AI agent. And, you know, so we're doubling down on our Oracle experience and going all in with their new, um, agent at EAR. Okay. And one capability that you wish existed today. Uh, I, I really think just being like totally keyboardless, like ambient everything, just being able to have a conversation, a computer, a conversation with a patient, just, you know, your thought process and, you know, queuing up your orders, everything being totally ambient, I think it's coming, but we're not there yet. Yeah. And this is probably interesting. AI has a co-pilot or auto pilot in medicine. Uh, I think it's both. I think it, for the most part, should be a co-pilot in the clinical space, but I don't see any reason why it couldn't be auto pilot in some of the backend processes. Okay. So finally, if you went back to being a physician today, how would you use AI differently? Uh, definitely, you know, I would embrace it, you know, is, you know, seeing more and more patients, um, mostly just to finish on time and go home on time. You know, I, the EHR is definitely hard work and adds more clicks to the docs. And so if I could use AI to make my life easier, I would 100% embrace it. Awesome. Well, thank you so much, Dr. Johnson. Thanks for your valuable time here with our audience. And I learned a ton of body in healthcare and here journey from being a physician to know a technologist and some great insights. Oh, thank you so much. Awesome. But that's it for this month's AI explain folks. We'll come back to you with another great guest on, um, in the month in the next few weeks. See you. All right. Sounds great. Thank you.