Good day everybody. I want to welcome all of you to this session here, a roundtable, where we will be talking about, you know, AI considerations and how they enter into the business world. We really understand the potential that this technology has for everybody, but one of the bigger concerns is bringing the theoretical actually into the practical with all those components that sometimes create barriers to entry or create blocks that make it somewhat difficult. We have a great faculty with a diverse background, so I'm going to let everybody first of all introduce themselves, and we'll start off with you, Shelby, and we'll go around the table. Shelby Doblich here. I'm founder and CEO of DocPace. We have a healthcare company with a suite of AI tools that help across practice, coordination, communications, and performance analytics. So we have patented technology that's helping in the clinical workflow, patient flow, leveraging historical data from the EMR system to optimize and improve today's scheduling and patient flow, and then expanding that into communications and helping really improve the patient-provider relationship. So I'm Renee Arnold. I'm a PharmD by training, and I'm here representing the NIH, actually. I'm an entrepreneur-in-residence, a very fancy term. I'm part of an innovation and commercialization group that works with entrepreneurs that have gotten small business grants to send them to investor conferences to get dilutive funding, as opposed to all the non-dilutive funding that the NIH gives out. I think people are surprised in some cases to know that the NIH helps small businesses commercialize their technologies, so it's really a very interesting program, and I also have my own small business. My area of expertise is health economics, outcomes research, and evidence development, leading to reimbursement, and I'm also adjunct faculty at Mount Sinai School of Medicine. I teach a course on pharmacoeconomics. Thanks. Wow, that's cool. I'm Gareth Hall. I'm on Microsoft's global health care team. I am in Seattle, despite my accent, and I lead the solutions approach for Microsoft, so we have a pretty big investment in health care, and it's summed up as what do we do internally to make our products more useful for the health care industry, and the reason I'm here I think today is actually for AI today, the biggest take-up we're seeing is in administration, is in kind of, yes, there's lots of clinical stuff, but there's a huge take-up outside of that, so hopefully we can learn together. John Crane, I'm the senior vice president of Emory's Heart and Vascular Service line. I am here today to help understand what it looks like at a local level. We have a bunch of great ideas that we're going to be talking around around this table, but inevitably folks interested in this will probably end up speaking to someone like me at a hospital as we go into contracting, as we go into proforming, and here to talk about what that looks like. And I'm Tom Dearing. I'm a cardiac electrophysiologist here in Atlanta at Piedmont Heart Institute, and I've worked with John closely for many years, so we'll share some dirt afterwards. My name is Kirsten Toulier. I'm the senior vice president of payment health care delivery policy at the Advanced Medical Technology Association, or ADVMED. We represent about 500 medical device companies across the United States, and I'm here representing the regulatory side of the world. All the things that the government tells you you can't do, or could do, or should squint at a little harder to see where you can find some space to get paid for these technologies. Thanks for having me, Tom. My name is Uday Kumar. I'm a cardiac EP by training. I'm currently the founder and CEO of a company called Element Science. We're building a wearable defibrillator, machine learning algorithms. Previously, I founded a company, iRhythm, which makes heart monitoring patch and has a deep neural network AI algorithm powering that. Started a couple companies in natural language processing, and I teach at Stanford Biodesign as a faculty for the last 17 years, thinking about how you think about unmet needs. And so I probably represent kind of hardware, software, and the intersection from the innovation side and the entrepreneurial side, and obviously reimbursement, and how AI gets regulated, and how people then translate that to implementation and trust and use by physicians is everything that I've been involved in for the last 10 plus years. Well, thank you all. This is going to be a great session. Uday, I'd like to actually start it off with you. You have a, you know, great background in clinical electrophysiology, and you've brought products to the, you know, use in the marketplace. How do you initially look at it from the point of view of getting a validated algorithm that is effective and will work for patients to create clinical value? Because we can talk about reimbursement, but if there isn't good statistical data to support its use and good background validation, then it becomes an unknown. So how do you think that should be done? And that will set the stage for how we talk about reimbursement. Great question. So I think this has been, especially when we think about machine learning and AI algorithms, it's kind of the focus of this particular panel, it's been a learning curve over the last 10 to 15 years, along with regulatory bodies, as well as less so payers, but more regulatory bodies. So I think ultimately starts with an unmet need and kind of really trying to figure out who are you going after, and more importantly, in that population you're going after to solve a particular problem, what data set do you need or should have versus what's available? Because oftentimes what's available may not be as representative, may be biased because of historic norms. So in terms of our pathway, both at iRhythm, which was kind of a cloud-based algorithm, and at Element, which is a device-based algorithm, which is a machine learning algorithm, a lot of it had to do with really understanding sample and populations and figuring out where can we get the data along with what the regulators think is appropriate. And that's actually sometimes been, there's been tension with that in terms of understanding, you know, a learning curve they're on to kind of get to the same point where what is the up-to-date science. So there are no rules. I think it boils down to a sample size where you feel, and they feel, and probably clinicians will ultimately feel that there's enough data to have this product as something they would use and trust. That being said, even when it is regulated and approved, there's still a lot of physician uptake that has to be garnered, as well as payment through clinical trials. So there's a whole another reimbursement paradigm, and iRhythm is a great example of having to do many, many, many, many studies to get payers to believe why taking an EKG for longer is okay. Great point, Uday. And maybe that segues into your role, Rene, you know, working at the NHLBI and coming up with research studies to, you know, also validate these things and to create more credence that this is something that goes forward. So when people like Uday bring a product to the marketplace, what do you think then needs to be done? What additional data is necessary for validation and for improvement? Okay, so great point. But we were talking at the, our previous panel about reimbursement also. It scares me when people who have created algorithms and AI, you know, based on those algorithms, and they say, but we didn't expect it to do that. Okay, it's not a sentient being. What do you mean you didn't expect, you programmed it. What do you mean you didn't think it was going to do that? But to that point, there needs to be this continuous evaluation. So I think this is something that even in health outcomes, my real area of expertise, you don't just do a study and then you're finished and you put it in a journal article and you're done. As you collect more data, you need to now be updating your outcomes data and your patient evaluation. And I think that certainly at NIH and looking at the small business SBIR grants and STR grant applications that come in, we want to see that there's this continuous evaluation, continuous learning. And you need to really make sure that there's an external reliability here, external, you know, generalizability. So the previous panel also we were talking about tech equity, so equity. Make sure that you're including all populations that you possibly can. Maybe you're not going to do that in a phase one grant when you're developing a prototype, but certainly in phase two when you start getting into clinical trials. Can you get this more into the community? Can you, you know, validate these algorithms to as great extent as possible? I mean that's a very good point because once something is released, there are utilizations that were not looked at in the clinical trials. And tech equity is an important consideration for us to look at going forward so that we can give everybody the opportunity to get the same level of care. You know, looking at it from the Microsoft perspective, you know, I know it's a very small startup company, but maybe you can give us your perspective from that, you know, side of the fence and where this might sit. Yeah, a tiny little software company in Seattle. It's funny, I said this yesterday, there's a partnership thing here. Big Tech, us, and competitors need to do certain things. Organization partners that go and build actual solutions that deliver the end point on top of this. We need to work together with them and then we need to work together both with kind of the organizational customer and then the practitioners, the users, the administrators. So I think we're all getting to the point where we're realizing this kind of mutual wins for everybody here, but it's gonna take some time. I think there's almost, maybe I propose something, you might not like this, there's almost two paradigms. There's all the predictive machine learning that we've had for 10-15 years that works really well. Clues in the name, it's predictive, it's predictable, great. And then there's the whole new world of generative AI, which is doing, which everyone thinks is a silver bullet. It's not. It can be for some things, but you've got to be careful, you've got to be thoughtful, particularly in health care. And to your point about the constant learning, generative AI, not only does your algorithm learn, the platform, the model underneath it is learning and changing in unpredictable ways. We're gonna have to collectively work out how we manage this. And we can't do it, I suspect you can't do it, I think we have to all do it together. So as we get more and more thoughtful about the industry, I think there's more and more partnership opportunities for us to learn, because everybody wins at that point. Now I think a key thing you just said is the partnership. We have to look at it as that we all come to the table from different perspectives, but when we can put all of us together, we're gonna get hopefully to a larger and better place sooner, because we all then understand what the others are thinking. And clearly, you know, as we're talking about in this session, payment is important, Kirsten. So I'll, you know, ask you to weigh in there, because that's one of the issues. We come up with these great ideas, and then how do we get it to a point where it actually can be compensated for those organizations that utilize it? Absolutely. And I was actually thinking, as you asked your first question, for reimbursement we can kind of turn it on its head. That no matter how good your data is, sometimes you will not get reimbursed for it, because there are a set of rules that the government operates under when they're going toward reimbursement for technologies. So I already used this joke with you guys earlier, but the regulations that govern Medicare payment were written at a time in the United States where iron lungs were in active use, and they did not contemplate that we would be walking around with supercomputers in our pockets. So when you're working with the Medicare regulators, you're working with a group of folks who are having to really squint at these regulations to say, can I make this fit? Is there a way to argue that an AI is actually going to be used incident to a physician service? And so they are doing that, but it is also a very protracted process. And then went probably to something Uday has experienced very personally. There's also even once you get in the door or in the window in the case of AI, where you're kind of getting anywhere you can to get into the house of Medicare, you have the question of what your rates look like. Because all of the Medicare systems work on what's called a balanced or a balanced budget. And so every time you want to pay for a new service, you have to take that money out of somewhere else within the same system. So you are constantly robbing Peter to pay Paul, and nobody likes that. Because in this case, you now get physicians that are kind of fighting against one another. Heart is a really great example of where AI has been revolutionary. We've seen a lot going on in remote physiologic monitoring. We've seen a lot of changes in the way that CMS is looking at this space and trying to get patients beyond the four walls of a physician visit. But that does have the side effect of taking it out of other specialties. And so, you know, they get thousands of comment letters, hundreds and hundreds of meetings every year. And they're getting both sides of that argument. It puts them in a very difficult spot. So I am, while I am sympathetic, there is definitely space for a lot of improvement in that conversation. And to your point about collaboration, I think a lot of it is bringing groups of people together to talk to the agency. So that it's not, you know, MedTech says you should pay more for technology. That it's, you know, it's patient groups, it's physicians, it's the producers of the technologies. And that they can prove that they're validated, and that they can also demonstrate that patients will use these technologies. Because uptake is a huge component, obviously, as well as physicians using them. So there's a lot there. There certainly is a lot there. And I think you're absolutely right. As a zero-sum game, we've got to figure out, out of all those things that we can compensate, what truly adds value and will improve patient outcomes, patient cares. And I loved your commentary about putting all the stakeholders together to go and make an argument for that. You're right. If you have MedTech going, or you have a physician group going, people are going to not listen as much as if you have the entire spectrum of individuals involved. So John, as an administrator who has to, you know, pay money to keep the lights on, and, you know, pay physicians and nurses and everybody else at a fair market value, when you're seeing all this potential new technology that can really create value, but oftentimes hasn't quite done it yet, how do you look at it? And how do you adopt these considerations? And what components do you use to analyze where you're going, and how you're going to hold people accountable? It's great. So, and it's fun to hear all the innovative and the thought pieces that you guys have come up with at the end of the day, until something comes out of Medicare, or some payer that says, this is what we can reimburse you. It's generally a pretty short conversation when someone comes to me and says, hey, we need to look at X. And if the first answer to the question, well, what's reimbursement for that? Well, I don't know. Or, well, this isn't approved yet. Or, well, we're still working on our pass-through payment. It's unfortunately sometimes a short conversation. But once we do get to that point, right, once there is reimbursement, then we get into, okay, fine. What's our value proposition? And that value proposition, despite what some might say, can take the form of what's the clinical output improvement, or what's the improvement in XYZ metric? It's not just all about dollars. Now, the fun part is, then we can turn around and say, fine, we're going to decrease our length of stay by two days. I can put a dollar sign on that, right? So then we start putting together the all-fearing pro forma. Here's what the top line is, here's what the bottom line is, and then here's what the number at the bottom of the page is. We have really thin margins in health care. You guys know this. And they're exceptionally thin relative to what some of the innovators around this table or others in the room might do. But we have to hit those every year, every quarter, in order to hit payroll. So I need a rock-solid pro forma that shows up across my desk, and I'm happy to partner with it to put it in, but that at the end of the day is what sells it. And if that is not at X amount, and each facility has a different level on it, we're not going to be able to move forward very well. So, not to show a lack of creativity but very creative I think actually John and one of the things I've had the opportunity in working with John over the years to learn is you know when an idea is put forward and it has potential benefit is to do all that upfront work like you said the pro forma the analysis the decrease in length of stay or decreasing cost elsewhere in the system but also just don't go in you know blinded and hope that it will work but to actually do a trial a pilot project you learn two things I think with those pilot projects number one where could we make it better because rarely does something get rolled out perfectly and then we can improve thereupon and also is it actually achieving the goals that it was designed to do and I guess you could say thirdly is it making at least neutrality fiscally or positively otherwise John will fire me so you know you have a lot of information shall be at your company that you set up on workflow issues and I think one of the concerns that we always have is these inefficiencies you presumably work a lot in the outpatient setting and there's as we know cancellations there's no shows there's workflow issues that could be optimized how can that be done with AI so that we can make that more effective and diminish that you know loss that occurs there and that inefficiency that has a fiscal consideration yeah I want to jump back real quick to your comment on I think it's so important to have buy-in from everyone in the office and really what's going to get us there is transparency to build the trust that's needed to buy into these solutions and that's one area that we've been really focused on and and found that worked is really sharing how our algorithms and our outputs work and where the information that we're sharing is coming from and so then it becomes very clear how everything is working why this is working and to your point of you know sharing before and after results that's a key thing that we do in the diagnostic side of first running reports analysis on okay where are we today and how can we improve going forward and then tracking that moving forward sharing updates on those metrics a key area that we're helping on the efficiency side is actually looking at appointment duration specifically and how can we help optimize schedules from a patient throughput perspective and where our appointments getting backlogged what's causing that and then what are some other solutions AI solutions that can help improve moving patients through their visit and new data kind of drilled on that particular comment you know looking at your existing company or your prior company poor perhaps more accurately so patients now are using a lot more wearables and different types of rhythm devices that they can get in the retail market in addition to that there are those that are medical grade you know devices as well but what I find is problematic a lot is there's oftentimes siloing in ER doc orders at a hospitalist orders at a primary care doc orders their cardiology and there's all these inefficiencies and how we basically deal with abnormalities and how we deal with normal results because there's a lot of energy and effort I think that is expended that is really not value-adding its redundancy or its inefficiency so how do you think just not taking a product to the marketplace but actually putting together workflows that allow us to say like you were saying earlier we reduce costs here so we need better reimbursement there so probably 10 or 15 years ago at the Stanford biodesign program we realized that a big barrier for a lot of companies after they came up with a good need came up with a solution found the right regulatory pathway they couldn't get paid and they went on business and so we spent a lot of time with our fellows when we have an unmet need and just the idea of a concept of doing deep value analysis we teach them how to do Markov modeling to really understand all of the different options to really think about the full length from like here's when the here's when the intervention or the diagnostic whatever it might be all the way to an outcome and diagnostics is even harder but then you have to think about what do people do with that information does it really change the outcome in a clinically and economically meaningful way because if you don't do that you're going to waste a lot of time and effort to get to something that where you don't have a good sense of whether it's going to be useful so the inefficiency years are because of this to your point and I gave this example earlier when the Apple watch came out I rhythm stock went down because people thought oh you can get an EKG and then people realize like that's not really a regulated product at that time and it's kind of screening so then people realize it's just going to lead to more CEOs so stock went back up so total inefficiency and it gets back down to very simple things of the pre-test probability in the patient population wearing Apple watches is so low that means false positives are higher and then what are you going to do with that I'm sure every EP here or every cardiologist is like someone will come in like look what I found and you're like why were you looking are you looking and now you're kind of obligated in the society we live in in which liability and other things are high to now get on the ride and start spending money and since we're so far away from a value-driven system except in certain ID ends even though that the promise was during Obama and it's we're still far away that's when you'll start to get efficiencies when there really is it's it is a cost of care for a given patient and so ordering this extra test unless it's really going to valuably do something is not just a thing you do one other comment I'll make is in this process people also learn about reimbursement of really understanding like if it's a DRG you care if it's a prescription for an outpatient product you don't care in this well you care but in a different way from an economic perspective so really APC's DRG's CPT's new codes existing codes working within the laws from 1960s to sign a figure out is it possible all of these impact how long it'll take a technology how much it'll cost which then means how much you have to recoup at the end when you charge patients so it goes back to the beginning do the economic modeling up front it'll kill most ideas but the ideas were worth taking forward we'll we'll get through there's a different number of zeros between the CPT and a DRG reimbursement right you know I wanted to really quickly first of all it does my heart good to say here Markov modeling I love that but anyway something that people don't consider is they'll consider the short term you know it reduce hospital length of stay by you know one day or two hours or whatever but what are the downstream costs and especially if it's an integrated delivery network or you're in a capitated system which a lot are these days they care about what are the downstream what's the cardiac rehab that's involved how often you're going to come for unscheduled clinic visits you know coming back into the hospital within that 30 day that dreaded 30 days that you know Medicare as soon as they see that oh we're not going to reimburse because you you know discharge them too quickly or whatever so I think you need to think also longer term and not just in the short term is there a cost-saving yeah I think that's a really good point and it's something that we have been trying to kind of fight against in the advent space as well is that this concept of bending the cost curve which when it's applied to health care has always been the concept of reduced cost it's not focused on the outcome of the patient and it's specifically not focused on the long-term outcome of the patient which is something that we see the role of AI and ML really helping with is to be able to say this patient exists beyond the 30 days they are still a person on day 31 and they deserve to be treated like a person on day 32 and so you know trying to look at appropriate investment on the front end which might look like cost but saves you money over time because you're identifying for those patients who aren't just actively monitoring their Apple watches you're finding arrhythmias you're finding patients who are candidates for AMI or infarction so much earlier in the process that their prognosis is better that they don't require crashing into an ED or they don't crash into ESRD for example so you don't have patients who come in in renal failure because nobody saw them before that point in their care because we were only thinking about them for individual instances it's really a transformative piece of value it's something Medicare is really struggling with they are still very focused on when is this patient being seen and how much does that cost as opposed to who is this person and what do they need long term I think that goes to some of your value points of can we start to transition the system CMS has a concept for doing that they had a goal of having I think 96% of Medicare patients in an accountable care relationship within the next some and five years which they're getting close to with the Medicare Advantage program there are other problems there but especially when we try to transition the fee-for-service system I think that's going to be a real challenge because there are upfront costs associated with that transition that the system can't currently bear so you know call your congressman I guess I'm not a lobbyist so I'll say that on the front end but you know it's it's a very daunting prospect but it's a necessary shift in the system I would just add that there's some disincentives based on the commercial in terms of insurance churn like most people most insurance companies commercial will think about their beneficiaries in two-year three-year increments because and I never understood it because even if you're losing a patient you're getting somebody else's patient with the same problem but just that's the way it is so the focus on short-term outcomes is because on a given beneficiary an insurance company might be looking at it in a short term so until that changes we just have disincentives to looking at the cost of care across a lifetime of a disease and I think you both make a very good point we don't look at the long-term situation we look at the short term perhaps those administrators in the insurance company get quarterly or yearly bonuses based on that so they're willing to look at the short term only rather than looking at the long term you know and you raised two very interesting points I'd like to get John you to weigh in on it first so look looking at this perspective if you know sure hospitalization within 31 days is important because all the negative fiscal and other clinical implications that can occur but how do you as an administrator take a long-term perspective and how do you engage your clinical staff to look at that effectively to make it work because I think you make a very good point a patient isn't a 31-day episode a patient is a long-term care delivery process I'm not gonna suggest that we're great at pro forma modeling a per click basis you start asking me to look over two years this gets even harder and we get worse at it you start looking over a lifetime whoo scary right up this sounds like a epic solution a electronic medical record solution to look at the patient longitudinally but how do I value that that's going to be a challenge and we are not sophisticated organizations from an analytics perspective we're great at giving care right we start trying to model some of that to understand the value proposition even more challenging so maybe that brings us back to the point of getting all the stakeholders together not just when you go to get for reimbursement but in designing systems with new products and with reimbursement and with processes to create efficiency Gareth what are your perspectives on all this I'm gonna give maybe a slightly different one so I'll Microsoft perspective minor Microsoft perspective is hey you're all building loads of data we've got to find a way of helping you connect it drive the value so you can get that long-term visibility most importantly we all have the partnership connection thing has to be important and we have to do it with open standards so the fact that we're all not inventing our own standards anymore we're all talking cool the other perspective I'll just give you but maybe I'll try and cheer you up so I came from the National Health Service in the UK which has a very different incentive model it does incent that it's not a competitive it's designed to incent long-term care it can't do it either it's really difficult to do this the payment models make it harder here but it's gonna have to be the payment piece it's gonna the reimbursement piece it's gonna have to be the data piece it's gonna have to be the trust piece there's no one simple fix fix for this and I think we all I'm gonna keep saying it we haven't we all have a role to try to keep driving this and I think you're right some of this is phoning your congressman and again getting back to that comment of the trust I mean I think we need to have more situations like this where we're all at the table discussing and we can argue we can disagree but we ultimately hopefully trust each other enough to come forward and to make things like that benefit just to go back to you again Kristen we are moving and you noted that you know from a fee-for-service world to a basically fully capitated world you know as we look at Medicare shared savings programs and other similar third-party you know payer programs how do you see that working in the payment world both for the clinicians who provide you know the services for the administrators who have to at least balance those dollar amounts and for the outside payers who will be providing it I know it's a hard question but how do you see that evolving over the next five ten years I think you just summed up 12 years of my career and one question so it is a it is a very hard question to answer because there are so many different if just focusing on federal because individual insurers do so many different things in different markets but in the federal space there are so many different little pieces of this puzzle everywhere so we've had a long-standing quality reporting program system in the inpatient hospital setting for example it used to be called rock-a-poo I still don't remember what it stands for now it's hospital IQ our program but that program has been collecting data on patients for almost 20 years the thing is are we using that data in a meaningful way you know we we publicly report some of it patients can in theory look at that when they're selecting a physician we know they don't because they don't understand it and they might not even know it's there and then you get into value-based purchasing which is kind of the next step of the puzzle the thing is that's also key to quality metrics that the government has endorsed and adopted for these programs and they're very limited so that it is things 30-day readmission 30-day mortality you know other complications when you get to things like joint replacements and so you are putting a piece of the hospital's puzzle or reimbursement at risk in these programs but it's a I say this from my side of the table it's a small percentage I know it's not when you're looking at close margins and so and then you go to things like the Center for Medicare and Medicaid innovation which was stood up under Obamacare and it was intended to test large-scale model changes in healthcare so you would look at things like total comprehensive comprehensive joint replacements which was a long-standing model in the system it was looking at lower extremity joints and that was the closest we have seen a model get to meeting the requirements for national implementation and we still didn't hit it so 50 some models over the last 10 years nothing has met the thresholds for national implementation through that program and so we are constantly looking for ways to evaluate value to find ways to pay for value it's just a really complex puzzle and there's also this piece of how do you establish accountability when it comes to quality is it the physician is at the hospital is it the patient adherence is it something else in the environment that prohibits them from being able to perform the care that they need and so there's I don't want to give CMS and out for it they have been really trying it's just that these are all little bits and pieces and it's one of those things where you know not to draw the ire of anyone at the table I think you kind of have to clear the table and start over because you can't keep trying to adapt a fee-for-service system to become value based it's just never going to translate in that way because that's not the way that the incentives are structured. It's looking at things totally different I think I like your idea of clearing the table and we have about 13 minutes before we have to clear this table so Uday you were gonna say something yeah I would say that maybe we can learn because obviously I mean I won't say single-payer because that's not in the cards anymore which is fine but there are closed systems, the kind of the Kaisers of the world and others, who maybe have a different view on kind of value, positive and negative, in terms of their beneficiaries and how they get cared for. So there may be some learnings there because they're looking longer term because they're the payer of record and the provider of record, or even maybe the VA system. So I think there are models, not great, but there might be some learnings there because I know, you know, for instance, when we started the Zio, the VA was one of our first hospitals, they saw the value of it and they could make that decision, as opposed to the fragmented system of commercial payers, Medicare as well, you know, as kind of for a given physician. So I do think there's something, it's not all still, there are places we can learn from, but still there's still a paucity of data about a long-term evaluation, how do you model that? No, I think you're right. Most of us are starting at the Model T level, but there are places like the VA and the Kaiser group that are maybe driving 1932 Plymouth. So maybe we can get together and we can get to a new Tesla at some point in the very, very near future. You know, but you mentioned earlier, Uday, and I think this is important. One of my big things is that we need programs and processes in place. Like I get frustrated, I'll have to say, when an ER doc or a hospitalist or a cardiologist or an EP doc orders a monitor and they don't know exactly why, and they don't know exactly who's gonna follow up on it, and we create all these inefficiencies. I think putting together programs that then get institutionally based so that you're doing the right thing at the right time for the right reason, et cetera. And we have so many new modalities, all these things that can predict perhaps ischemia or can predict, you know, risk for a-fib development or other components. I think it's very important for the institutions to work with partners, to work with colleagues, and put together programs that lead to how we deliver care. What do you think, John? And I know it's really easy. It's easier to get cardiac doctors to agree than it is to herd rabid cats. Electronic medical records have pop-ups for just what you're describing, right? They're really popular. Clinicians- Oh, very, yeah, oh yeah. What you need is four more pop-ups and then we can direct it the right way, right? That's what you're advocating? Right, right. Boy, going back to clearing the table idea, the singular payment and, again, monitoring long-term, it's a great idea, but when we get to the quality payments, I mean, Tom, we spent two years trying to come up with appropriate quality criteria in our prior lives and we got to some good metrics. Still tough to measure, still tough to communicate, still tough to get buy-in across the clinical community. I mean, that was a labor of love and we had some success, right? But there's no great way to get to pay for quality. There is not. It's prioritize, focus, start, have metrics, show benefit, and then hopefully broaden it. I want to bring up one last topic about AI that I think is relevant. How often in the future do you think we're going to need the clinician? I mean, we all think that we are the raison d'être, but in reality, there are many, many studies out there now, and Uday, you're familiar with it as are the rest of you, that show EKGs can be read just as well by a machine. A monitor can be read just as well by a machine. An echo can be read by a machine. An MR, you can go down the list. So are we reaching a point where we have to start looking at AI technologies not as just a tool? It clearly is a tool, but actually as a partner, as a colleague. So I'd like to get everybody's thought on that as we can go around the table, and we only have nine minutes left, so probably be a little focused in your commentary. I was going to comment on the space of data and AI and helping on the administrative side. That's a key area where we can really leverage data and AI and help on the efficiency side. And then within the clinicians. So you still need the touch with the patients, and I absolutely believe that AI used appropriately is useful and helpful, and it's good in training as well, as you can imagine. Training young staff, training different kinds of staff, nurse practitioners, et cetera, et cetera. It doesn't always have to be the physician, for example. So there may be efficiencies there that we can take advantage of with AI, but AI still scares me a little bit. And you look at chat GPT and the hallucinations and whatever, and I use perplexity more so because you've got references, but every once in a while, it's still not a real reference. Wait, how did you come up with that? I was just talking to somebody here and she said, oh yeah, she loves perplexity because of the references, but then the reference was at the end, but it didn't refer to something that she was interested in. So she asked perplexity, well, where did you specifically get that? Oh, sorry, I overreached. That's really not a conclusion I can make from these references. Whoa, okay. So I think, yeah, from the soft touch, from the patient perspective of trust, and I think the clinicians have to understand more, or we have to do a better job, especially the people that are developing these types of algorithms and machine learning, of saying to the clinician, this is the training data set. This is how we generalize it. This is how we validate it. So that if the patient says, well, you put this thing on me, I don't know what it's doing. You don't have to be an engineer as the clinician, but I think you have to understand it well enough to say, this has been evaluated in 10,000 patients. We really feel that this is going to give us a good idea of how often you have AFib episodes, something like that. That's very nicely said. I think it, just like you enumerated, trust is very important, communicating about where the technology is and how we are using it. And I think if we can use it effectively and free up clinician time, there's more opportunity, like you started off with your comments right now, to have a conversation with the patient about what they need, what their goals are, where they want to be, what they're hoping to achieve. Because medicine is both an art and a science and has always been, and I believe it always will weigh in in that manner. Couldn't agree more. Don't get excited about Microsoft branding, but we call all of our AI stuff co-pilot. It's not called pilot, it's called co-pilot. It's job is to help. How do you operate at the top of your license? How does technology remove some of the stuff that just gets in the way of you doing the job that you're actually incredibly uniquely trained for? AI's got a massive role there. Did you all see the study, was it a few months ago, that said a kind of a chat GPT generated response seems more human than a physician response? I did see that, yes. That was really interesting. How do we combine that? Because obviously, what it actually means is chat GPT's got more time to write nice things at the start and in the end of the letter. Yeah, it does. How do we combine that? That would be a really powerful use case. Clinical validation, physician in the loop, making sure that everything's right, but AI's helping make it feel more relevant based on personal knowledge of that patient. Cool, that's really valuable for all of us. And that's key. They had more personal knowledge. They also had more time, and they had more ability to put it together. But freeing up physicians and actually learning from that, like you said, because many of us aren't trained to communicate to patients and to communicate about sometimes very difficult issues. Mr. Administrator. Does AI carry malpractice? So, you know, let's say that we have 1,000 studies and 999 of them are read perfectly. One misses. Our friends in the legal profession will want to know about that, and they'll want to come after someone with sharp knives. So how do we make sure that some of the companies out there are willing to be liable for that if we're going to put that level of responsibility in their hands? Because I haven't seen that yet in a contract. A new teammate, the lawyer. Speaking of, I actually trained as a lawyer. Never practiced to everyone's benefit. So, to my own benefit as well. But I think that's, obviously malpractice is a huge piece of it. That's, I mean, malpractice is such a major part of the Medicare system that it is part of the calculation for how they pay physicians. It is also the piece that is most likely to go up over time because they are seeing increases in the rates of malpractice cases. I think the piece that I look at from the reimbursement perspective is, obviously, the AI as a partner is great. It is instrumental. How do we get the AI paid for? And then how do we ensure that as we increase efficiency in the system, we don't inadvertently say that now physicians are going to be paid less as a result? Because, again, going back to this balanced budget system, you become more efficient. Medicare says, great, now I don't have to pay you as much for that service because you can do more of them. And physician burnout is a very real problem for folks who are treating Medicare patients. The volume is insane. The need of the system is massive. And because the payment rates are so inadequate, we get into systems where folks just don't want to play the game anymore. So, you know, if we are able to find ways to get these reimbursed, and we've got bits and pieces, we've got this AI stroke predictor, we've got the FFRCT heart flows, computed tomography technology, but these are just little bits and pieces. It's not standard policy. And if we could get to that, if we could increase the certainty for investors and for developers, that they will be able to commercialize, I think we could get to a point where they are really being able to use these technologies in a standing fashion that produce some of the data that we need to support long-term adoption. Yeah, maybe just to build on this, I might say that if I think about physicians 20 years from now, the things that I don't think AI will be there will be procedures, robots doing procedures still a pretty long way off. So I think things, empathy, thinking about like, because we're still humans, but things with data, EKGs, radials, that's already here. I live in San Francisco and there are driverless Waymos everywhere. And I've actually noticed now in the last year, they've become more aggressive. No, seriously, like they go, I mean, it's amazing. And I just use that as an example, tongue in cheek, and that it's already here, it's already happening. And now, at least in my city, everyone, it's fine. It's totally fine. I'd put my kids in that bed more than an Uber, because there's different issues in terms of liability risk in terms of a driver versus a driverless, in terms of safety, et cetera. So I would just say that in terms of where AI is going, it's how can humans do more what humans can do better and just seed, it's going to happen. And it's already, if I think about, if I gave three EPs here in EKG, I am 100% sure there will be inter-observer variability. Whereas if I give our algorithm a deep neural network, which has seen millions and millions, it's probably better. And so from the liability perspective, proving that difference and how badly we accept variability among humans compared to computers is something we haven't done a good job about, but it exists. Because we just trust it more today. I think that's a great summary on where we are with the digital reliance and the empathy and interaction component and procedures is where we're going to be in the short term. We only have two or three more minutes left, so what I'd like to do is go around the table and give each one of you an opportunity to say, in a word or two words or a very short sentence, what you think about, you would like everyone to have as a message on bringing together AI with business-related considerations. We'll start with you, Shelby. I think there's a huge opportunity for it to help across the board on the efficiency side within the clinic. And that's really where within the administration offices that we can streamline, improve processes, workflows, and really help our teams operate and work better and do their jobs and deliver the care and be face-to-face with patients and spend the time doing what is improving healthcare the best. Thank you. So from an NIH standpoint, I encourage people to de-risk their technology using non-dilutive funding. So we had talked about the risk and whatever, but if you can show in NIH study, oh, you know, it's validated, da-da-da-da-da-da, and more and more of what we're seeing has an AI component. Everybody says AI, even if it has nothing to do with AI. It says AI in the title because it's become sexy now. But I think as long as we understand where we are with AI, what it can do, what it can't do, what we have to be careful about so you don't over-promise, but the more we can validate what's being done with AI, the better. Find the most obvious, simplest, lowest risk use cases. Probably administrative, probably less clinical. Get going on those. And then I loved your comment about pilots earlier. We say this in the NHS. There's more pilots in the NHS than there are in BA. I guess the equivalent here is Delta. Have an eye on pilot plus one. Make sure when you're doing your pilot, you're doing it with the administrator or the nurse who hates technology. Not just the shiny, always get involved using the coolest stuff. Go and find the most difficult person because we've got to get this stuff out of pilot and actually making a difference. So yeah, find the use case and then turn it real. And so I guess from the hospital side, despite kind of what I've said being very conservative, we remain bullish on AI and healthcare. In the exam room, in the procedure room, we think it can help. We just need to make sure that it has a rock solid business case. It does have proven outcomes and something that we can get on board with and get our most reluctant nurses and physicians on board with as well. I think to go back to a point Uday made earlier, and this is my battle cry of my career. Start early when you're thinking about your reimbursement. Understand the environment you are coming into and if the environment needs to change, start today. Because the conversation you have after you get through FDA approval is four years too late for us to be able to get you to market before your tech dries up, your startup money dries up, your people get bored and leave. And so start today. If you need big change, start the day before today because that's what it's going to take to get into the systems. Yeah, just I would say maybe for a forward looking comment, if you're telling your, and many of you I'm sure have children or people or young people in your life, I would say figure out what it means to be human and do those human things and everything else, it's already here. So let it go. It's because it's going to help you actually enjoy your life as a human more. So I think it's this letting go and we as physicians are very bad at that. Not to say that we don't have to have the right guardrails and the right, but it is this fear of the unknown. But like I said, in my own city, I just, it happened before I knew it. Cars are driving around without people and it's just normal in one year. I think becoming human is really the key. I guess my thing is what I'd say is build an effective, it's T-tubed. Build an effective team. Talk, number two. Talk among yourselves. Disagree, but come to an agreement on certain things that you can prioritize and make go forward. And then walk out and trust. But as was said by a leader one time, trust but verify as we go forward. I want to thank all of the panelists for coming and joining us here today. I love the fact that you come from different backgrounds and perspectives and we're able to provide that to our audience. Look forward to continuing to have these discussions going forward. Enjoy the rest of the meeting and thank you all for your participation.

AI integration

healthcare

clinical workflows

patient communication

tech equity

business cases

reimbursement

collaboration