Okay, I'm Rohan Khera, I'm a general cardiologist and data scientist at the Yale School of Medicine. And I have the honor of moderating this interesting conversation around data-driven management of heart failure and the role of AI-driven innovations. I have to acknowledge from the get-go that we had a change in our speaker group so that now it's currently five men on stage and that's unintentional, but I still apologize on behalf to whoever's attending that that's not, as a group we acknowledge that that's not ideal, but we will go on with the conversation with that acknowledgement in place. And then I wanna introduce our guests real quick before we get on to this conversation. So on my left is Dr. Jay Edelberg, who's a co-founder and head of research and development at Proleo. On the very right is Dr. Murad Farim, who's an associate professor and a heart failure cardiologist at Duke University. Then I have Dr. Adrian Hernandez, who's the executive director and the vice dean at DCRI, the Duke Clinical Research Institute at Duke University. And then on the end I have Alex Hajduk, who's an advanced heart failure and transplant cardiology fellow at UCSD. So the topic at hand today is our discussion on how would we consider data-driven management of heart failure in the era of artificial intelligence. Because it's an open-ended conversation, the hope was that if you have questions in the middle of the conversation, please feel free to ask them. We can address them to the panel today. The spectrum of folks here represent those in practice, in training, those leading the institutions, and those leading commercial ventures. So our hope is that we, through these conversations, get across the point of what it takes to first innovate and then translate those innovations to practice. And some of the work that we've done in our group at Yale has focused on one end of the spectrum, is to look at advancing the diagnosis of heart failure using AI applications to accessible modalities like electrocardiograms and echocardiography. But as we've had a conversation before the meeting, we could see that the translational aspects of AI into heart failure go way beyond that. It goes from discovery to the point of measurement of inpatient care and how inpatient care is delivered, and how we develop and care for patients as they leave the hospitals, and then ensuring and building systems of care that can affect every aspect and facet of heart failure care. So without going into too many details, I wanna give a moment to all of our guests here, speakers here, maybe a minute or two on how you see might be a role for AI in the spectrum of heart failure care and where we feel data-driven innovation is truly possible. Jay, why don't you start? Well, thanks. So when we think about where AI plays a role in the outpatient setting, there's both the diagnosis, where our patient comes to us in the clinic, maybe they're in primary care, and we're identifying patients who are at risk, and we know that AI ECG has been really quite revolutionary in identifying patients who are at risk for a cardiomyopathy or a low ejection fraction. And we also see the opportunity for AI in our management of our patients who already have been diagnosed with heart failure, and whether it's the progression of a cardiomyopathy, or it's being able to manage their risk as an outpatient when they've left after a heart failure discharge. Being able to use this, we now see that we actually have great tools, that it can identify patients who are progressing in their risk, that complements the great advances we've had in the diagnosis. And then the great challenge for us all is to really be able to now translate these into better outcomes for our patients. Perfect. So I think this really leads on to Adrian, you. How would we think about these innovations? And when we're scaling up, you know, as somebody who's led large innovations in the device space, the drug space, how do you see the next frontier evolving in AI for heart failure care? Did you say 20 years, or 20 seconds? No, no, no, no. Oh, yeah. No, well, you know, I think you kind of highlighted, you know, what I think are largely three buckets here, which is, you know, how can we diagnose heart failure better so with greater sensitivity and specificity, especially now that we have new therapies that can be targeted for different types of heart failure? And I think that's gonna continue to improve. Secondly, how can we optimize the therapy for heart failure? Again, getting to evidence-based therapies as quickly as possible, especially for different phenotypes of heart failure. And then thirdly, how can we better manage patients, especially in areas where they may have difficulty coming in? And so each of these may need to have different frameworks of evidence. And so how to diagnose different types of heart failure. You know, really, you know, we're looking for what's the sensitivity and specificity of these new algorithms? How do we optimize it? How can we do so quickly and safely and effectively? And focusing on quality of care. And then thirdly, managing heart failure is gonna be really important to understand are we actually improving outcomes, or are we potentially causing harm? And so each of these will have different types of evidence. And the issue is, how do we have a platform or platforms so that we can actually keep up with the evolving technology? And that's excellent. And I think, you know, we'll come back to the point because the evidentiary standard needed for different regulatory burden will be different. But I wanna, you know, put a pin in that for a later conversation in this meeting. So Alex, what's your take on, from your perspective, where you think most of the innovation in AI-related heart failure care will come, especially as you're in the trenches delivering most of the care? Yeah, exactly. I think starting from the trenches, you know, starting from the CCU, early identification of shock. There's recent data just from last weekend at ESC, really highlighting the fact that those first 24 hours are critical. And, you know, we've all been on service and these patients are always coming in overnight and they're not really kind of getting the best immediate care due to lack of resources that are present at the time. So utilizing, you know, these data-driven approaches and embracing AI can really serve as an adjunct there to steer the course right in that first, you know, when minutes and seconds counts for these patients. And then kind of taking it across, you know, when you're moving to the floor and moving to the outpatient setting. I think that we're taking what we learned from the large trials over the past 15 years, device, other monitoring systems, to really implement the care that we've seen that are, you know, are able to give life-saving therapies with GDMT for heart failure patients. And I think that utilizing the data and specifically AI to help with that, I think is a big topic of interest moving forward for us. So Murad, so I think, you know, you've certainly focused on a very unique aspect of data and looking at small populations, but looking at really ingrained physiologic monitoring and physiologic subtypes. You know, first off, would you describe your work and then maybe in your perspective, how that work could be even further innovated upon by access to technology that can learn more from data? Yeah, so we do, you know, deep phenotyping is sort of a terminology we use. You know, physiologists have been doing a lot of things to human beings to generate very, you know, rich data sets. And I think where we use more and more data-driven approaches is to have a staff and softwares and machines look at that data from different angles. And you have done a lot of clustering work yourself where we can use these type of approaches to help us come to conclusions about physiology. So a very specific field is a heart failure preserved ejection fraction. It's a big conundrum. It's the majority of heart failure patients have this specific phenotype, but it's so heterogeneous and we had a lot of neutral approaches to try to, and you publish actually on that, you know, if you to apply common drugs to all phenotypes, you will likely be unsuccessful. But if you now start do phenotype-directed interventions, you're actually much more likely to have success. Whether that will then translate itself to clinical trials and we can run it at DCRI is a whole different story to do upfront phenotyping. I love the plug-in for DCRI there. To, you know, actually deliver care. So I think the future is bright, certainly, but it's coming, I think that's probably where AI is advancing a lot faster. Maybe coming to the point in research, it's easy to develop and deploy all these tools. So Adrian is my boss. He's a visionary. You know, he has big plans, big ideas. I'm the little man who has to deliver on that at Duke. You know, I have to grease the wheel to make sure that these initiatives- What's taking you so long, Raj? Yeah, exactly. So those things they had to implement. Little man. So, you know, on the research side, it's actually easy to implement this tomorrow. I'm telling you, in clinical practice, you mentioned early identification. We talked to Jay earlier. It's actually surprisingly hard to implement anything and to keep up with the pace of you all on the engineering side. You know, I think that's important. I think I'm gonna come back to Jay. Jay, you had this, before this conversation, you had a framework about how to think about innovation and heart failure in AI. So do you wanna share with the group? Sure. When we think about it, we spend a lot of time thinking about the science of AI, and it's pretty incredible. Whether it's the AI ECG or the AI Echo, this is fabulous. And what that's allowing us to do then is to really get to, really, that better phenotyping of patients. Because I'll be honest, actually, it's kind of embarrassing that we're talking about heart failure like it's a disease. It's actually multiple different cardiomyopathies. We just don't know which ones they are and which ones respond. So you can think of me like an oncology wannabe. But the science is there, and we can take advantage of it. But that's only the start. The next thing is we have to have, actually, a structure, a system that actually, we can actually gather the data, process the data, and deliver the data. And that needs to be, for our healthcare, that needs to be cleared by the agency. And so that infrastructure, the second is. And then the last part, which I think is really important for us, is actually the supervision. We're not going to hand over the care of our patients to a black box that we can't actually understand and explain to our patients or their family. We actually have to supervise it. It could be, basically, a quality system for supervising it, which is, of course, important for any system. But at the end of the day, we, as the clinicians, are responsible for being able to supervise the science that's delivered on a cleared structure, a system, and then we supervise it. So basically, it came down to three S's. No, that's excellent. So, Adrian, can I ask you one specific question? So we started this conversation on regulation and thinking about different pathways. And there's this huge divide right now, one in therapy and one in diagnosis. Like, you know, it's pretty separating. How do we think about regulating the diagnosis of cardiomyopathy at scale versus new therapies that come in the, because the current FDA strategy is that it doesn't require randomized clinical trials or any large-scale studies at all to get these diagnostic tools on the market. And we've seen a few AIECG tools already approved. So do you think that's sufficient for us to go after? What would be the ideal strategy thinking on scale? Yeah, I mean, I also think of it, often you're working through different pathways and regulations through CDRH in terms of device technology. And so there is this kind of inverse to say like going through a drug approval or a biological approval. And I think it comes down to some of the principles here. It's like, hey, is there a public health benefit for having better diagnosis of a disease? And so I think the answer is largely yes. Do we see the tip of the iceberg when there's actually a very large proportion of patients below that tip of the iceberg that we could have some benefit from that events already exist in terms of managing them? It's a different story where there may be situations where we're actually employing algorithms to help with decision-making. At least there is some type of potential judgment that's also incorporating things that we don't know. Over time, I imagine we will see more and more going from so-called assisted and semi-automatically decision-making to more automated decision. And as you get further out towards that, then I think you actually probably need to have better evidence around that because then it becomes more like a pill in a software box, so to speak. And that's the concept. That's excellent. I think that's an excellent framework. So I'm gonna take on Murad next for one thing. So you've discussed phenotyping specifically, and say there were extra phenotypes that were generated through some data-driven assessments of physiologic monitoring. How do you see that playing out in therapeutic strategy decisions? So we find spironolactone works well in certain phenotypic subsets. How do you see that playing out in clinical practice? Do you expect, how do you, I mean, envision a clinical scenario? Because I think quite a few people wanna think about that clinical scenario specifically. So that's a good point because in clinical practice, to date, we have, let's speak for heart failure with reduced ejection fraction, for example, where we have many, many drug therapies available. Let's say six, seven drugs are today available you could consider for your patients. There's an underutilization of those drugs for many, many reasons. And today, the big push by the societies and our guidelines is to get everybody on everything because that's how it was tested. There was never a phenotypic approach to testing and approving these drugs. I think in the future, we will see maybe a slightly more nuanced approach when we have even more drugs, that we will say risk-based prescription versus benefit-based prescription. And we know these patients are at greater risk, just don't prescribe that drug because it will cause hyperkalemia or whatever. Or this drug maybe has been more likely to be benefiting this patient. I can tell you, though, that till we have very high prescription rates across the board, I don't see ourselves to be taking a phenotype-specific approach. That's just not happening. It's also easier to practice one Bill Fitz-Everybody approach. That's what we're practicing in medicine right now. And to that point, I think we can all think about this. If we design a strategy-based trial around phenotypes, a strategy-based trial on phenotypes, that trial will be hard to conduct because you have multiple sites deploying that strategy slightly differently. We already have challenges with like, yeah, there are great examples of something like SPRINT, easy, measurable target. But then other strategy-based trials like remote patient monitoring trials have had challenges where every trial comes up with a different conclusion. So I think there is definitely some of that in play. So Alex, as you think about, and I know you worked on fitness trackers and other elements, too. Have you seen any studies related to some of the fitness tracking devices in heart failure care? So, yeah, I think that... We can envision them, too, if they're not. Right, yeah, exactly. I think from a fitness tracker, kind of the consumer wearable space, the patients, for example, might be more familiar with that and may be more likely to maybe uptake that. But the science doesn't really back the use. And I think that there's a disconnect there. I think one element where you can kind of circumnavigate some of that, some of those issues, is from a patient empowerment standpoint. And I think that there's a lot of support from the clinician side and the consumer side is saying, how can I use whatever tools at hand to help me help myself? Personalized feedback, you have these kind of coaching models in some of these apps, or even just very basic, like Fitbit, and aiming for 10,000 steps. You have your patient in front of you and you're giving basic recommendations for exercise. It's a little bit easier when you're giving them an objective target to reach. And then, basically, you take that moving forward to say, okay, that's the basics. What's the next step, what's the next step? And then how can you have that cross, almost that two-way street of saying, okay, now my patient is walking this much or their activities levels are here. Can you turn that around and alert that to help you up titrate GDMT, or get an assessment of their functional status that way, where you don't have to ask them. And really, just trying to focus on these passive monitoring systems versus active, which takes a lot more effort. And I kind of got away from the wearables question that you posed initially, but I'm sure we'll get back to that. But I think it's on point. And again, this is where, whenever you're in a consumer health space, actual consumer health-based regulation is really complex because you can't have patients be regulating and working on medical devices on them without the knowledge base. But then this subterranean, sub-regulated space where you talk to the patient and figure out what they're doing is where things are going. But then, Jay, you've worked in this space, specifically in using device-based post-remote patient monitoring. And so, how do you see that space in terms of regulate? I mean, you've gone through FDA and other processes. How does that differ from the vision from consumer-based wearable devices? So, for this, obviously, things need to be cleared by the agency through either DeNovo or 510K. But basically, CDRH has to actually clear them so that we can actually use the information and put this in the medical record. We're all familiar with watch-based approaches and someone shows up to your office and it shows that they didn't have a heartbeat for 30 minutes yesterday. And, you know, perhaps that's not as trustworthy as we would like it to be. We know that, basically, our FDA-cleared devices, that's trustworthy and you can act on that data. Can I ask, what was the tool that you sent through the FDA? What was the indication, AIFU? Right, so what we've done is to actually take the individual, we can either take the entire system, soup to nuts, or we take each individual piece. What we've chosen to do is take our individual pieces. We can take these physiologic measures, we can calculate this risk, and we've gotten that one approved. And so we get each one of our components cleared so that, basically- Can you give an example of what might be a- So, one of them is our multivariable change index here that basically takes multiple tools, multiple physiologic inputs, heart rate, respiratory rate, activity, establishes the baseline, and then allows us to be able to then see whether that move from a, basically, digital twin identifies patients at risk. That was generated, the agency has provided a clearance, and so we can actually use this as a tool in clinical medicine for you to be able to employ, and that's, of course, the backbone of what we're doing at Proleo. So we actually have got that cleared and gone through that pathway there. And we've used this, and others have used that as a precedent for trying to get their own tools cleared. That's awesome. Murad, so you also have this other footprint in doing remote patient monitoring on a clinical site among patients. How have you seen that translated? Again, one area is this somewhat data-rich approach to having tools, but what has been the real-world RPM utilization like, and what does it actually look like for our patients right now? Right, so that's a good question. So I think the framework, we actually just published on that in JCF last week. I think we have a lot of tools or gadgets out there to give us data. I think that we are, as a clinical community, moving away from pure RPM technologies, because if you just give us data to monitor, we don't like that because we can't keep up with the data. Now it has evolved to where each individual company, in most cases, has some form of algorithm for some form of data cleaning and filtering to give you only the patients in, let's say, alert status. They are at high risk, so that is a must. Don't even start with raw data in first place in a clinical space, because then you have too many patients you cannot scale it. But the evolution has also been that more and more companies now, and I see some in the audience here, is that provide also an additional layer of clinical support on top of the data, because what happens is that 95% of health institutions in the United States will not be able to act on the data, even if it's cleaned and filtered, and act on that data in a meaningful way. So I think we have to go away from RPM, remote patient monitoring, to remote patient care. We actually have to act on the data we've delivered on the case. So that was really what that piece was all about. But now in clinical practice, well, we have ICDs. That's one. We have the CardioMEMS and the Cordellas of this world. We have patches, a number of clinical patches. And the limitation right now is human power. Our institution does not want to outsource most of this work. It want to keep it internal. I think in the majority of institutions in the United States, the smaller they are, the more likely they are willing to outsource, because they either don't have the pride or the lawyers to keep the data internalized. But I think the future will be that more and more data will have to go external, and patient management will have to be a shared approach with industry. There's no way around it. I think just one thing that comes up, actually two things that come up is clinicians are also worried about having another app for that. And so if it's not integrated into their workflow, and so I'll just use our example, we're an epic institution, we have Haiku on our phones. If you have to open even another app, even if it's on your phone, like that's just like one extra step as opposed to something that's fully integrated. And so any extra click causes problems. And so that's an institutional concern that comes back constantly in feedback regarding burnout and so forth. The second thing which Marat touches on, and I think there's still attention on this, is that there's interest to optimize the system to see all the people that are out there that are new. And so how do you bring them in? Well, you have to figure out some other way to manage the patients that you're already seeing. And so that offers opportunity to do that. But getting that, what's the tipping point where you say, okay, let's do this on a big scale because then it becomes, then you start really getting to economies of scale. I think we're still at that tipping point realizing where that is. Yeah, I think we're gonna switch into the funding stuff because I think we are at that point now, we have to decide how we fund these. But if there are any- Money matters, is that what you're saying? Yeah, money might matter at some point. But if there are any questions among the audience, this is a great window to come up while we jump into this treacherous domain of money. So I think I will start with you, Dijay, because I think being at an organization that has to value, again, that has to understand the financial, I would say, bottom line. All of us work in institutions that care about it, but it's transparent in an entity. It's more transparent. So how would you say has the funding model been for developing these technologies, scaling these technologies, and eventually getting reimbursed for these technologies? Yeah, so when you look at it, we have to start with the end user in mind. We have to start with our patients there. We know that our patients, what can't we charge for? We can't charge, we can't have our patients paying for these tools. That's not where we're gonna go. We have to actually know that we have to be providing value to the actual system, whether it's reducing hospital readmissions, reducing resource utilization, et cetera. And so therefore, those can be risk-based approaches, risk-based models that we can do, or basically a subscription service. But you, the clinicians, you, the system, you, the patients, have to see the value from the insurers. Based on those models, then we can actually then be able to actually raise the money to be able to actually generate the science to be able to do it. But it is a very different model system than when I'm developing a drug, because it's just a very, very different reimbursement strategy. But if we don't actually put that one in mind, money matters, and you gotta put that together. As somebody in healthcare leadership, and you think about, there are these spare reimbursed models for our more patient monitoring that are gonna be in flux in 2025 for sure. The COVID burst, it's gonna change. But are healthcare institutions invested sufficiently that even if the payer model changes, the RPM ecosystem will survive? What's the thinking on institutional setting? Are the institutions seeing enough value? So there's a lot of heterogeneity in terms of the financial models for healthcare systems. But as Jay is kind of noting, as systems getting more towards population-based management, then that really does matter. How can we manage a population that's high-risk that can be very costly to the system by better managing them outside the so-called traditional walls of the hospital? Those places that are more fee-for-service oriented, we're immune to it. And so they come in, they get their care, and it doesn't drive the incentive there, except in certain settings where that changes. The thing that we have found, and Murat alluded to this, is how do you get to a critical mass of patients where you start getting to efficiencies and it's a step function that actually people start realizing that. And that can be modeled and demonstrated that yes, this is a better approach, so it so-called lets more people into the system. That makes a lot of sense. So Alex, you've been in multiple institutions over the last several years, and you've seen this evolution at play. And as I said, as a practicing clinician, have you seen some examples that stand out or any of the other things that might suggest that this is a viable strategy? Yeah, so over the last couple of years, I started with John Boehmer at Penn State. That's how I got into heart failure and remote monitoring. And he'd done a lot of work with Boston and HeartLogic, and we did some work in non-invasive remote monitoring strategies. Then I moved to Jefferson in Philadelphia, which was a completely different system. No real champion in the remote monitoring space, although phenomenal heart failure clinicians. But the system really wasn't built for expansion and taking care of a large patient population by utilizing technology just because it was the opportunity cost. If you were gonna allocate any more nursing staff to monitoring this, you were taking away from something else. Now at UCSD, where I'm fairly new, just started in July, we're using a lot of different data-driven and AI tools kind of across the board. If you were listening to the prior talk on this stage on generative AI in medicine, which was phenomenal, Karen Deep is now at UCSD and look forward to everything that will come from that. Specifically in the advanced heart failure world, Eric Adler and colleagues worked on MarkerHF, which is basically a tool, risk-gratification tool, to look at advanced heart failure patient and outcomes there using less biomarkers than other traditional models and with pretty good sensitivity and specificity for the outcomes. Point being, just like we've all been at different institutions, everybody does things a little bit differently and some of the models I think are set up for more success. I think that the ones that are really trying to push innovation and are open to conversations, open to partnerships with industry, I think that that's really where I've seen the most success and fortunate to be a part of those. Perfect, and Murad, this is actually a perfect then direct question to you. As you think about scalable innovation in data-driven heart failure management, you've seen in your own local practice and we've discussed that the business model has to be aligned, but there are institutional variation. How do we get to a future where all 5,000 acute care hospitals in the US are using some standard strategy because this is now standard of care? So how do you see from day now to that day where we have, if this is care, then this has to be standard? Yeah, so it's interesting. I am thinking about it quite a bit. The power play here that should be done is Epic. The majority of the country is now moving towards Epic. I don't understand why Epic wouldn't buy VisAI, Merge, et cetera, et cetera, because that, once that is on one platform, well, or just develop themselves. Now we're talking because then it's a lot easier for us. I'm going to violently disagree with Murad here in a minute, so. So I mean, I think, to be honest. It's always fun to have your boss disagree with you on stage, yes. I love disagreeing. You know, the reality is that, it's that comment by Adrian about the apps. Do not make me wanna or have to click on any more things that I already have to provide care to the same patient across multipliers. It's just too complicated. So I think streamlining, and I think we're moving to the world. I think your comment, by the way, earlier regarding reimbursement. I think if CMS is right now to cut reimbursement for RPM, which they tried to do, actually. We petitioned in the Congress this early this year. It would be catastrophic. Catastrophic. So that luckily didn't happen. But I think if, money needs to be infused to sustain this, and right now we're very, very thin stretch to sustain the system by streamlining workflows, by having it all on the same platform, all integrated for providing care, and billing is very, very important. Because if you don't do those two things congruently, we cannot survive as a system. Now I have to go to Adrian. There is no other way. You have to yield your time. Yeah, well, I have teenagers, and so they're a big fan of Fortnite. And so, and one of their biggest worries is that Fortnite goes away, and they won't have great games. And so one of the concerns I have, and look, it's great that we have electronic health records. It's great that we're getting, hopefully, to greater uniformity around the use of that. However, I worry about a monopoly situation where it could stifle innovation. So what I would hope for is actually that there's actually more openness at Epic that allows apps to come in. And so it is more of an operating system that allows those innovators, those innovative applications to come in, and be actually a service integrated there. And I fully respect that each party should derive fees and revenue streams for that, but how do we get it so it's a little more of a shared process? And so the Fortnite example is like, how do they get into the app store? How can they get back into it? And then who sets the pricing for those fees? And so I think there is some risk here for healthcare if we were all just under one tent, so-called the Epic tent, that's only control there. So I think there's gotta be some healthy innovation and tension around that. Absolutely, and I think there is an app ecosystem, Epic developing, we all have to see how well it actually pans out, or Epic has that as we have seen in other examples where apps are developed, and then the parent company develops a similar app, takes it on, and that's their app. So I don't know how innovators will feel about it. The problem is with that, for example, we just signed on Pacemate. How long did it take to integrate Pacemate at Duke to do EP and hard for remote monitoring? Over a year, and at every institution you actually have to reinvent it, even though there's a plug-and-play with the Orchard and wherever to get it into Epic, but it's still, on an individual level, Epic is still like the United States where every state has a little bit of a different interface, A, and you have local regulatory bodies that want to have their say upon integration. So it sounds, that's exactly what it needs to be, an open play field. I think through regulation we actually need to standardize certain things. Right now it's a wild west. It appears to me, because I'm not in the weeds of it. And Adrian, maybe because I grew up in a communism, maybe that's why I want to have a single system. You know? I see that. I'm just saying, you know? I was trying to see that and not say that, but you said it yourself, so it's okay. So in our free market economy, in our free market economy, how do we drive innovation? So one is, we can drive innovation by value. We can drive integration by some sort of regulation. So where do we see, I think we have last, like maybe seven minutes left for our conversation. Where do we think the next frontier is? And Jay, I'll let you take on that. Where do you think, what are the next step for us to innovate on? We've discussed that we need to do science, we need to innovate on the regulation and others, but are there frontiers that are not being touched upon in heart failure care? I think what we need to do is to actually pull in what we're all talking about here. And we start out basically by establishing that heart failure is not a monolithic disease. It's got diverse phenotypes. And as Murat pointed out, some patients benefit from one therapy and they don't from others. And the fact is, we need, let's make it really simple. Let's learn from our colleagues in oncology. They actually figured out that one drug benefits one patient and it doesn't benefit somebody else. And then they basically developed the biomarkers, often genetic biomarkers, but for us, we can actually have digital biomarkers that identify those as unique phenotypes. And we can do it both in the inpatient and the outpatient setting here. And we're very, very lucky that we've got ECHO, we have ECG, which are incredible data sources for being able to phenotype these patients. And we will be able to then stop talking about HEF-PEF and HEF-REF and we'll be able to start talking about very specific digitally-based or phenotypes that are basically, basically on responses and progressions. Again, much like our colleagues in oncology. And I will just state, I went to medical school a few years ago, we had heart failure. Now we've progressed, when I was in medical school, we had lung cancer with large cells and small cells. Now there are over 250 types of lung cancer based on the responses to phenotype. But we in cardiology has also progressed. Now we have heart failure with preserved EF and with reduced EF. So I think it's outstanding. So we basically are only 30 years behind oncology. I think that AI and ECG and ECHO give us that opportunity to catch up. Perfect. I'm gonna actually pose the questions a little differently. So say we have found out there are 35 different kinds of heart failure subtypes. How are we going to go back on our heart failure therapy landscape? Our guidelines. I'm just like, I shudder at the thought of looking at that guideline document and revising the class indications for every drug we do. How do we go to that next frontier where we test all our ACE inhibitors, ARBs, financial models are completely off? Oh, wow, okay. That's an easy question. Yeah, yeah. No, well, again, I think you point out there's gonna be certain situations where we're gonna have to employ data science to actually have so-called best guidance in terms of what we do. Sequential therapies in terms of what to do. How do we get to personalized therapies based on someone's comorbidities and benefit and risk profile potentially? So chronic kidney disease, the risk for hyperkalemia, other issues that we may consider. And so that's gonna hopefully be really data science driven. You won't do a randomized clinical trial in every setting. My hope though is actually we have a better system for doing so that's across geographies, across different clinical management spectrums. Because right now, we're either the Wild West or we're like at the colony stage for this. We're not States and we're not United States. And so we gotta have a better system to do so. Perfect. So Alex, what do you think? You're practicing now and how would you like to practice five years from now? One of the ideas, maybe it's just crazy because we were talking about how more data is sometimes not good. You need to clean it and make it digestible. But you see someone in clinic, you make your assessment, you have all these diagnostic tools, you send them home, you have them get labs, you have titrating meds, et cetera, et cetera. I think in the future, we're probably not far off in some of these biosensors for real-time monitoring of electrolytes and kidney function and other biomarkers. I think that that would be an extremely useful tool to integrate on top of the system that we already have. If you knew everybody's drug levels at all the time, in your transplant patients, you're monitoring drug levels and you know what the creatinine is, you know what the K is, you know what their blood pressure is, you know what their symptoms based off of an app, you could take that conglomerative information and you could actually rapidly uptitrate GDMT and hopefully incorporate the 35 different pathways that you've basically delineated, okay, this is gonna be the sequential order here based off of your personal biomarker fingerprint. I think that that would be the pie in the sky idea that I envision. I think we're pretty, I don't know how far off we are, but hopefully we'll get there someday. I mean, that's an important idea to have at least. Murad, I think I'm gonna give you a slightly more challenging question, because the boss is here, I have to make you look special. So I think, so now you've been given the charge to run every company that's generating new evidence on new devices or new innovation in AI is now coming to you at DCRI to run these large mega studies because DCRI has done that forever. Now you are there at the helm of running digital health innovation at scale. How do you offer, because they don't have the same financial model as a drug company, they still need the regulatory bar for getting therapies approved, but this is not the same financial reimbursement model. How do you envision trials to run in that setting, in the digital health setting, that's different and more cost effective than say in the pharmaceutical space? Barry, sorry to put you there. Yeah, that's a tough one. So I think the key part is unlike, and I think the same actually applies to pharma companies as well more and more, is A, you want to run trials more pragmatic, sure. I mean, you need to be more innovative about how you implement it. But I would argue that particularly with software and with device-based strategies, you need to be ready to implement when the trial is done. And to some degree, running the trial, and the term of learning health system comes in, to some degree you need to run the trial in a healthcare system that might actually implement it. It would be the death of something when it works on paper, you derive the score, you validated it, and then you try to bring it to clinic, but the clinic cannot implement it because it just doesn't fit the workflow. Unlike with drugs, which find their way through established prescription pathways. We have prescribed things for many times. It's just an easy plug and play. With device-based approaches, software-based approaches, it's never a plug and play. And it's reinventing the wheel at every institution from scratch, at every physician from scratch. So I'm actually curious what Adrian has maybe, contribute to the thought of learning health system, how you bring trials quickly into practice. That's an important point. I think if people would take away, it's the idea that trial is now the pathway to pilot testing the final implementation. We've started an experiment on this in a way, it's called Cardio Health Alliance. And so the idea was how do we get to real-world data to real-world action? And so testing different strategies of care, and if it works, then the assumption is how can we get it to automatically implement it into practice on day one? And so the group basically has to evaluate and say, okay, if this works, we're gonna push it forward. Perfect. I think so we're at the close of the session. Any last closing remarks that we didn't cover? Anything people have any burning thoughts or ideas? Yes. Can we ask you a question? Sure. Well, what about you? What do you see in the next five years? You know, I think- Three years, it's going so fast. Three years, a year. So next year, I work a lot in the space of diagnostics and prognostication, and what I've felt has really been progressing really rapidly is this idea that you can diagnose patients anywhere and diagnose a series of conditions anywhere without having access to our current diagnostic care pathways, which I think is the one area that's ripe for evolution because we under-diagnose a series of cardiomyopathies. We've diagnosed diseases like amyloid cardiomyopathy at scale, hypertrophic cardiomyopathy at scale, which was previously required advanced systems of care to get there. And so I think that is where we'll see the most innovation, plus the regulatory pathways are simpler, making it easier to get into practice as long as we find a business case for them to happen. The second place I think I see most value is that we can finally start getting to the point where AI creates a level playing field across systems because currently, if you don't have an amyloid center of excellence, you don't have even the thought of process going where you would get your patients into PYP scans or other testing strategies. Now you have this tool that applies on electrocardiograms. So now you've leveled the playing field for amyloid diagnosis at a smaller center that can now create a hub-and-spoke model to that center, but has something to start the process. It doesn't start at heart failure. It starts at amyloid suspected on ECG. So I think those are the key areas that I see evolving, and plus, again, all the things you have mentioned are all still at play, but these are things closest to my heart. So with that, I think I want to give a big round of applause to our speakers today for an excellent session. Thanks for folks who stuck around to the very end. It's a big accomplishment. And I want to add one more thing, which is something that we haven't really talked about, is we need to use AI or need some help with getting more people into the heart failure sphere because you're having more and more patients, more and more data that we're talking about accumulating, and less and less providers that are able to deliver that care. So really thinking outside the box there, I think is a major topic that we need to address moving forward. Absolutely, perfect. Thanks so much.

heart failure

AI-driven innovations

data-driven management

clinical practice

digital health

tailored interventions

real-world data

cardiomyopathy