A video podcast series exploring the potential of digital solutions in cardiovascular and emergency medicine

[00:00:03] Mathieu: Welcome to Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine, featuring thought leaders from leading institutions across the globe. This series of conversations unveils the power behind clinical algorithms and decision support tools, and how they are changing the way we diagnose and manage cardiac conditions from their impact on clinical practice, their benefits for patient outcomes, and the challenges they pose to health systems in terms of implementation and regulation. This podcast is intended as a precious resource for healthcare professionals in cardiology and emergency medicine, who are curious to learn more about this exciting field. I'm your host, Mathieu Chaffard, and you are listening to Cardio Insights.

[00:00:52] Mathieu: So welcome, Professor Gibson, to this episode of Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine. For our listeners who might be new to you, you are a Professor of Medicine at Harvard Medical School, Chief Executive Officer of the nonprofit Baim Institute for Clinical Research, and the Academic Research Organization PERFUSE, as well as Interventional Cardiologist at Beth Israel Deaconess Medical Center. You are known as the inventor of several coronary blood flow measures widely used today in clinical practice, including the TIMI frame count and the TIMI myocardial perfusion rate. Throughout your career, you have published more than 700 peer-reviewed papers and led numerous clinical trials, including the pioneering trial that demonstrated the superiority of dual therapy over triple therapy in patients with atrial fibrillation that require stents. You're also a member and fellow of numerous societies, including the American Heart Association Council on Atherosclerosis, Thrombosis, and Vascular Biology, and the Commission of Professional Education Committee. So in this episode, we want to explore with you the role of clinical decision support tools and solutions in cardiology, the impact that these technologies have had on the management of acute and chronic heart conditions, and also take a sort of forward-looking look at the future of cardiovascular care in the era of artificial intelligence and digital health. So from your perspective, how have clinical decision-support solutions changed the landscape of cardiology in recent years? And do you see an acceleration of that progress in terms of the capabilities and adoption of these technologies in clinical practice?

[00:02:41] Michael Gibson: Well, thanks for having me on today. It's really been breathtaking, right, to see the advances in clinical decision support science. That's the good news. AI machine learning. You're not a company unless you have AI machine learning. That's the good news. The bad news is that the implementation And science is lagging far behind. Now there's a popular expression. Culture eats strategy for breakfast. We have great strategies, but the cultural changes inside hospitals are so big and hard to surmount that it's been a challenge. While there is good science supporting clinical decision tools, we have to make sure that the tools are doing something for clinicians rather than something to clinicians. We also have to make sure all the stakeholders are aligned. You have a lot of Ps here. You have the physician. You have the patient. You have the payer. You have the person at the top of the hospital. So all those have to come together. And we don't have really great alignment of all those strategies. Each of those different groups has different goals and we have to get alignment around those goals. The patient should come first, but we have to get better alignment. So this is not the end. This is not the beginning. Beginning of the end. This is the end of the beginning where we have to reemerge now with changing culture to really, really take advantage of all the good science. I like to say that science is easy. People get mad at me when I say that, but people are hard. This. We've solved a lot of the science issues. The answers are very clear. It's the cultural barriers, the people problems to making this happen, which are the challenges that are before us.

[00:04:48] Mathieu: And so, you know, that's an element we've talked a lot about in the other interviews that we've done today with some of your fellow cardiology colleagues. And, you know, this piece of, you know, fostering adoption and it's frustrating, you know, the implementation of tools that fit the way, you know, healthcare professionals work seems to be a recurring theme. And to be a, you know, a main, major pain point that, you know, goes much beyond, let's say, the level of the technology that we have today. But in terms of what you've seen, you know, over the years in terms of new solutions coming up to support the way you treat acute and chronic heart conditions, can you tell us, you know, how your practice has evolved and how some of these tools have changed the way you deliver care on a daily basis?

[00:05:41] Michael Gibson: You know, it's fascinating. You said tools. And then at the very end, you use the words deliver care. And I think those are two separate pieces of information on the tool side. I think we've got to get to a place where MD doesn't stand for manual data entry. You know, we've been entering all this Data into the EHRs. We've got to move away from technology where we work for technology instead of technology working for us. That's got to be a transition. It's so frustrating. We've worked so hard to enter all this information into the systems. The systems have to be able to give us the information back in ways that don't require further work on our part. We don't want more clicks, okay? You use the words deliver care. I think one tool today that came up in our discussions that actually was used and is good, is a tool that says for concussions, not only assesses risk okay. There's the assessment of risk. Then there's a decision about acting on that risk. The third step is the third D is doing something about it. This tool is very helpful because it actually helped people do something about it. It's set up those outpatient visits in an automated fashion. In other words, it reduced the caregivers' workload. So there is the data. The first D, there's the decision, the second D, and then there's the doing. That's the third D. We heard from Prof. Jim Januzzi that they have a heart failure implementation team that spends an hour with the patient doing things. That's what we have to focus on. It's really the third D, the doing that really needs some work. And it's a people problem.

[00:07:41] Mathieu: And there is the doing on the clinician side and the doing on the patient side. And we see also you know the let's say remote monitoring technologies coming up or being you know, let's say common part of the way we deliver care today as well. How do you see these technologies evolve and how do you see the impact that they have on, you know, patients being empowered to take care of their condition?

[00:08:08] Michael Gibson: Well, the letter that comes after D is E. You said empowered, but you're only empowered if you're educated. I think we had those three DS, but the third D third E that has to begin the E series is the education that comes after it. We've got to make better educational tools for patients. We know patients will hear about one out of every three words you say. We have got to have our best communicators making content that's free and available and accessible to everybody to do that. Education. A doctor's visit is more than 5 to 20 minutes. It really is something that should go on for a long time after the visit, with the right kind of educational tools. A doctor or other health care professionals are not going to be able to do it all in 20 minutes. It's going to take it's going to take a lot more sophisticated, careful, studied approach with good communicators to get the job done. One of the flaws in communication is that communication actually occurred. I think oftentimes health care professionals don't communicate with the patient, not just at their educational level, but they failed to establish a connection. We usually call EBM evidence-based medicine. I think evidence is necessary, but it's not sufficient. I think you've got to also establish an emotional connection with the patient. So I call it emotional-based medicine. I think it runs the opposite of what we think in medicine. I think we often think we are neutral. We must be, you know, impartial and neutral. And, you know, no one cares how much you know until they know how much you care. And we've put up a barrier between us and our patients. I think in our communication style and how we educate people, we've got to we've also got to use our hearts to engage them and to show that we care about them and their understanding of this content.

[00:10:26] Mathieu: I think that brings an interesting piece. That's the, you know, having that empathy between, you know, and in, in the patient-doctor relationship. You touched on before that, you know, you don't want more clicks in the way you work. And something I've read, you know, in the preparation of the interview was that you had also a lot of a large part of the medical community saying that we, you know, we treat the screen. We don't treat the person that's behind the screen. And the thing I'd like to, you know talk about is, you know, we talk about we talk a lot about artificial intelligence these days and what it can do in terms of processing automatically information recording, you know, the vocal, let's say, elements of a consultation to automatically transcribe it so that clinicians don't have to do it. How do you see, you know, this type of technology, artificial intelligence playing a role in reestablishing potentially this empathy that's so important in ensuring that, you know, communication is flowing well between the patient and the clinician and that there is this level of empowerment that, you know, the patient will understand what it needs to take to have sustainable behaviors in the future to address the condition that they are concerned with. How do you see this, you know, in the future being part of care?

[00:11:45] Michael Gibson: Well, I guess I was dumbfounded when, you know, a group came in to begin to implement some of the EHR stuff in our cath lab, and they literally had the data entry tool or the computer across the room with the nurse having her back turned literally. Her back turned to the patient. It begins with, you know, looking at the patient. And I demanded that, you know, all that be rearranged so that the nurse is actually looking at the patient. So they're very simple things. You know, I just now, in the past two weeks have been using a tool that's an AI-based tool for all my meetings. So I'm on, you know, I'm on meetings a lot of the day. And this tool takes notes. It doesn't miss anything because it's transcribing everything that's said. And it's almost absolutely perfect in its transcription. And then it creates a summary of all the key points. And at the end, it has all the action items. It's just revolutionized my world because I was sitting there writing out on a yellow notepad, a legal pad, everything, and then typing it in and transcribing my notes.

[00:12:57] Michael Gibson: Now, I don't do any of that. I just sit there and I just look it over, and if it's accurate, I save it to a folder or if it's inaccurate, I edit it. Edited. I think we're there. I think we are there. Where you. I could sit there with you and look right at you in your eyes. Talk to you. And you could look at me in my eyes and we'd talk and all the notes would be taken for us. And you know, what I would do is I would look over the notes at the end. Perhaps we could look over them together. And if you disagreed with something or I disagreed with something that was said, we could do it right there on the spot. That's where I see the future's going. I do think we're there. So I think, again, we've got to have tools that facilitate engagement rather than get in the way of engagement. And I think we're there. These tools should actually improve our connection rather than be a barrier to our connection, but we have to have the cultural changes to adopt them.

[00:14:01] Mathieu: Some of these topics are, I think, absolutely valid across basically every condition. It's not just Starches specific for cardiology medicine. Is there an aspect where these technologies are potentially, in the future, even more relevant for cardiology and emergency medicine? Do you do you know certain use cases where you could say, this is really going to change the way we practice?

[00:14:32] Michael Gibson: I think what I just said about emotional-based medicine isn't going to change. But I think one thing that happens in the emergency room, for instance, is you've got to make decisions immediately. I worked a lot on decision algorithms in ways we communicate with patients surrounding life-threatening emergencies. You've got a very, very, very quickly exclude a life-threatening problem. There's no time really for establishing a relationship right then. So I do think we could have algorithms that help us very quickly identify the key things that can kill someone, for instance, with chest pain, like a dissection or a heart attack or, you know, an esophagus that's torn. You know, let's begin by excluding the things that are going to kill you immediately. Then let's move on to other things. I think an organized, hierarchical approach, with the assistance of these kinds of algorithms, could be helpful in getting excluding the life-threatening things quickly so that we can move on to other things. Now, I think one problem is in ERS, and for cardiologists if we think someone's not going to have a heart attack, that doesn't mean they don't have a problem with their GI system. Right. So that's we shouldn't say the person's fine. Yeah, they're fine from a cardiovascular perspective. But you know, we have got to take care of the whole patient, not just the cardiovascular system. And I hopefully will keep us honest in that regard.

[00:16:02] Mathieu: What do you think would be the benefits on, let's say, at the health system level, at a public health level, you know of having these solutions in place and running, let's say, on a daily basis, what would you, let's say, anticipate the benefits would be on a broader scale?

[00:16:20] Michael Gibson: We talked about the Ps. There's at the top the patient there's the physician, there's the payer. But then there's also the third, the final pea you're talking about, which is the public health all the Ps put together. I think an informed public is a healthier public. I think our biggest battle right now is not in information, but battling misinformation. I think we have got to work towards reestablishing the trust of our patients. I think Covid did a lot to really diminish the trust of providers. I think it really diminished the trust also of our government and the agencies that give us recommendations. I think we have a lot of work to do on educating people about how science works. I think we all want the truth, and people think science means truth, and they think truth is something that's permanent and never changes. The truth changes, and we've got to get the public used to the idea that as we learn more, our truth sometimes changes. And in Covid, we learned things and it seemed like the truth was changing. And that was very confusing to the public. So I think we've got to do a better job of reestablishing truth, letting people know how science works. I think that's a big challenge in front of us.

[00:17:48] Mathieu: We have a recurring question we ask every guest on the show, and that's a bit more, let's say, fantasy-driven than what we talked about. But you know, if you could have any type of digital solution in your clinical practice that addresses one of the central pain points that you face and have it tomorrow. What would that be?

[00:18:15] Michael Gibson: Well, we're working on tomorrow. Today. For a few years now, we've been working on a decision support tool that does the following. Patients want to know about them. Not the population, but what's. What about me? What about me and my risk? And what should I do? Not the population. Secondly, someone's risk changes over time. Let me give you an example. If you take a blood thinner, if you don't bleed on that blood thinner, you have passed your bleeding stress test. You're a different patient at the end of that. So we need to rather than just assess the risk right now we need to iteratively assess risk. It's a little like expecting us to predict the weather a year from now. We're going to be doing a much better job if we predict the weather a week from now. Right. So we need to iteratively assess risk. Patients want to know what's going to happen to them, not the population. We need very specific individual risk. We also want to give people a give-or-take around that assessment of risk. Your risk could be 5%, give or take 3%. And we also you know, we also want to take into account their preferences, like how do you rank death a stroke or a heart attack or bleeding? People may rate those outcomes very differently. Now there's population ratings of those. But no, we want to use the individual person's assessment. So where we have we gotten with the tool we've developed is it assesses your risk iteratively over time. It gives you a give-or-take around that risk. It provides you that risk in a way that's displayed. If you are a numeric, you know the number of bodies there. And finally, if you want if you're someone who values preventing a stroke over a bleeding event. It does a different calculation than if you're more worried about bleeding than a stroke. Yeah. So we want to take patients preferences into what they want to prevent and what their risk adversity is. And we're finally there. So I'm very excited about it.

[00:20:31] Mathieu: Amazing. Thank you so much, Professor Gibson.

[00:20:33] Michael Gibson: Thank you so much for having me today. It's been a real pleasure.

[00:20:40] Mathieu: Thank you for listening to this episode of Cardio Insights. We hope you enjoyed it. If you want to dive further into other aspects of digital solutions in cardiovascular and emergency medicine, check out the other episodes of this series. To support the show, you can share this episode with your surroundings and leave a five-star review on your preferred streaming platform. Cardio Insights is a podcast brought to you by Roche. Thank you for your time and see you in the next one.

[00:00:03] Mathieu: Welcome to Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine, featuring thought leaders from leading institutions across the globe. This series of conversations unveils the power behind clinical algorithms and decision support tools, and how they are changing the way we diagnose and manage cardiac conditions from their impact on clinical practice, their benefits for patient outcomes, and the challenges they pose to health systems in terms of implementation and regulation. This podcast is intended as a precious resource for healthcare professionals in cardiology and emergency medicine, who are curious to learn more about this exciting field. I'm your host, Mathieu Chaffard, and you are listening to Cardio Insights.

[00:00:51] Mathieu: Welcome Prof. Baugh To this episode of Cardio Insights. It's the first recording, so that's the podcast where we want to explore how digital solutions Are transforming cardiovascular and emergency medicine for our listeners who might be new to you. You are an Associate Professor in Emergency Medicine at Harvard Medical School, having recently completed a five-year term as the Vice Chair of Clinical Affairs for the Department of Emergency Medicine at Brigham and Women's Hospital, where you trained in emergency medicine. Throughout your career, you have co-authored over 120 publications on topics related to cardiology, emergency medicine, oncology, but also in healthcare, finance, policy, and clinical operations, just to name a few. So in this episode of this series, we want to discuss with you the challenges and the opportunities that digital decision-making algorithms in cardiovascular emergency medicine bring and the considerations they require when we think about implementing and driving adoption of such tools. So before we dive into all these aspects, could you share with us how the emergence of these tools changed your clinical practice on a daily basis?

[00:02:03] Christopher Baugh: That is such a great question. I feel like my career has really bridged the analog phase of providing health care. And when I say analog, I mean pencil and paper to a real digital electronic medical record. With respect to order entry, decision support, documentation. When I was training 20 years ago, my note was a written piece of paper. I was writing orders on a clipboard at the head of the bed. And it's really been a revolution that I've got to experience. And what I love seeing is the potential in the future for where we could go now. I see a lot of potential that has not yet been realized, and that's why it's kind of a passionate area of interest of mine, because I feel like there's an opportunity right now to really innovate around this space. I feel like in recent years, the kind of digital tools that we're seeing are primarily focused on safety. Usually there's a safety event, there's a review. And the outcome is that there's now going to be an alert that's going to prevent that safety. Event from happening again in the future. And as a result, they're designed to be very sensitive to not miss any cases. And as a result they're often not so specific. And so it does create quite a bit of alarm fatigue among clinicians. And that very often the alert is not relevant to the patient in front of them. And I think that creates a lot of distrust around digital tools for decision support. So I think with all the revolution we've had with AI and kind of smarter design and what we've learned about what good decision support looks like, I think we're at a point where really helpful, useful tools can be developed now. And I'm really excited to be at a point in my career where I can contribute to the development of those tools.

[00:03:56] Mathieu: Beyond the safety aspect and all these alarms that you mentioned, where do you see the biggest potential today for improvement or for the development of new solutions in that regard?

[00:04:07] Christopher Baugh: So the area where I think it would be most helpful is around decision support to help make better decisions for clinicians, especially in the emergency department who are making an astounding number of decisions in a short period of time. I feel like if you like making important decisions very quickly, work in the emergency department, because there are times when there are a line of people waiting to talk to me with a question. And some of those questions are very important. It could decide, you know, am I going to send this patient home or not? Am I going to call a consultant? Am I going to do a CAT scan to find a life-threatening diagnosis? And I'm making one decision after another. And so the idea of electronic systems helping me make better decisions in a way that I feel better off for having interacted with that tool than if I was on my own without the tool. That's where I think the future is. And so we need we need tools that are evidence-based, that are done in the population, that is, that the clinicians actually practicing in, where it might catch an error where you didn't consider a factor and you're missing a diagnosis in your differential, or you're perhaps about to send someone home with a dangerous diagnosis. And on the other end of the spectrum, I think supporting doing less for patients who don't need additional diagnostics or treatments is going to be very helpful, because right now I feel like we are misallocating our resources. We are not giving enough resources to sick patients that we're missing, and we're probably overtesting a population of low-risk patients that could just be sent home safely. But we don't have the level of decision support integrated into our electronic systems to really help clinicians make that decision with confidence and know that doing the right thing by doing less testing for those patients.

[00:05:59] Mathieu: And in terms of implementation challenges, you brought up one which was what's more related to actually the use of these devices where you have this alarm fatigue, where you know, you get too many notifications or there's too much information. What are some of the other challenges that you see within your practice, among your colleagues, in terms of implementing or adopting these solutions?

[00:06:23] Christopher Baugh: So there are challenges around making the evidence to know that the tool is properly working for your population. There are challenges around integrating those tools into electronic systems. You know, there are many different electronic medical records. So you need a tool that's going to be able to be integrated into those different systems. You need to be able to have that alert be presented to the clinician at the right time in the encounter that they're having. And then you ultimately need the clinicians to trust that recommendation and act on it. And so you need buy-in from them. And so that goes back to the evidence showing that the tool is going to give them good information. And then you need real-world evidence as well not just study evidence. And then you really need a cultural change around the way that these tools are viewed. Because again, they're kind of traditionally viewed as as not very helpful. And so the idea of saying we're in a new phase of decision support now. Smart decision support, where you're going to really miss these tools if they were taken away from you. And what I would love to have is a clinician who moves from one spot to the next, where these tools were embedded to where they're not. And one of the first pieces of feedback they give is we need that tool. Because when I was using it before, I was much better off. And so advocate for that tool. To then be integrated into that system will be very important I think. And lastly, there's a level of regulation that is happening now around these digital tools, especially ones that are more than just a clinical calculator.

[00:08:00] Christopher Baugh: There are lots of tools where it's a risk score. You put in different aspects of the patient, their age, their gender, etc. and then it gives you a number that goes back to a risk score. That's helpful. But ultimately I think we're going to go towards probably more nuanced decision support that takes into other accounts. It might take more of their risk factors into account and make their presentation the quality of their chest pain for a chest pain patient. And those more nuanced ones are probably going to need evidence, and they're going to need to be regulated to some degree. And so this is a new era in regulation for these digital tools. So I think a lot of folks are still figuring out how to navigate that regulation. And then lastly these tools are going to need to be updated because new evidence arrives all the time. And when you're talking about governance around making sure that the tool that you're using is using the most up-to-date information, You know, the kind of updates that happen often are delayed and that there's no room for that delay when you're talking about care of patients. And so, you know, when we see these big specialty society guidelines around things like chest pain, they come out every 4 or 5 years. And, you know, the update on the digital tool can't wait 4 or 5 years. It needs to be continuously updated. And I don't know if we have a good mechanism for that right now, but that's certainly going to need to be integrated in the future.

[00:09:27] Mathieu: One thing that would be interesting to understand was you mentioned the level of evidence around these tools and the importance it has on, you know, the selection from clinicians as to whether or not they would use it. How do you see, you know, how clinicians actually make the selection because there is more and more tools available. There's also this point solution fatigue that I think is quite recognized as well. What's the decision-making process as to, you know, we use that we don't use that?

[00:09:56] Christopher Baugh: So ultimately, if there are multiple tools around the same disease process that are, say, available to be integrated into the health record, the leadership at the hospital or health system record is going to need to evaluate all of those tools, just like how they might evaluate a different test or assay to be like. There are multiple different troponin assays or multiple different types of stress tests. They're going to need to figure out which one are we going to allow our clinicians to have access to. And it could be that for some tests or tools that are complementary in some way. And so we're going to have them all available. And others they're going to have some a choice to be made, right, to say, hey, we're going to we're going to go with this one tool because we feel like it's the best tool for our patient population. And it's the we believe the evidence is most robust, and there may be some discordance around the clinicians. There might be some who say, well, I like this other tool, I want to use that one, but you're going to need to have the leadership get everyone on the same page around saying, at this site, we're going to be using this particular tool. And then it's going to be important also not just to get, say emergency physicians to all be on the same page, but you need if it's a cardiac tool, you need the cardiologist in the group to also be on the same page, because very often the tool is going to make a recommendation like this is a patient who needs to be admitted to cardiology. If the emergency physician is using that tool and believes in it, and the cardiologist does not believe in that tool, you're going to have conflict. And so you need everyone in the institution to be on the same page around what the standard of care is there and have a common expectation.

[00:11:33] Mathieu: And is there also a responsibility from health systems or broader health systems beyond, let's say, the individual hospital level, where this has an influence on which types of tools are being used or implemented?

[00:11:47] Christopher Baugh: I think we're seeing a lot of consolidation, at least in the United States of hospitals, into larger and larger health systems, usually a mix of academic centers and community centers. And what I'm seeing is a lot of lot of decisions that are made at a more central level, and then tools and resources are pushed out to the whole system to have some harmony across all of those different hospital settings. I think some tools will need to account for different resources available locally at different hospitals. For example, in cardiology, you could have a community hospital that does not have a cath lab versus a big academic center that does. And so you can imagine if the recommendation is for emergent cardiac cath, it's a different kind of set of actions for a community hospital than an academic hospital a community hospital. That can mean you need to transfer that patient right away. Whereas the academic center is you need to activate your cath lab. But I think hospital systems can act to advocate for those smaller hospitals that may be on their own. If they weren't part of a larger system, wouldn't have access to these tools, because there often is a pretty big IS kind of concentration of resources in big health systems that might be able to bring these tools to disseminate them to smaller hospitals that may just have a very limited budget and expertise around IS. But being part of that larger system gets in that benefit.

[00:13:13] Mathieu: And how do you see the role of regulators in providing some degree of guidance on which tools to use, because obviously they are responsible for certifying them? They are classified according to a certain framework in the US and in Europe. But is there also a responsibility from their end on providing guidance to clinicians, to health systems as to for which conditions, which tools are the most relevant?

[00:13:37] Christopher Baugh: I think regulators are going to set some safety standards. I think that's probably where their interests are. They want to make sure that the tool is being used appropriately, and that the bar is set at a level where other tools that are being used have set, have met some minimum standard around a level of evidence, and they want to prevent the tools from being misused or applied to to the wrong clinical scenario, the wrong population at the same time. I want to make sure regulators are not putting undue burden on the development of these tools that would delay their creation and implementation or increase the, say, the cost and time needed to get a tool developed and integrated to the point where fewer tools are available because of the burden of creating them from a regulatory perspective. So and this is also a relatively new and emerging area of regulation of digital tools as we kind of move away from the calculator type of decision support to more nuanced decision support, even AI-guided decision support. You can imagine a future where there are AI agents that are kind of looking out for the patients and running all of this analysis in the background.

[00:15:01] Christopher Baugh: That's very sophisticated. That goes well beyond a calculator that are making recommendations to the clinician that they can use or not use. One would imagine that a certain amount of regulation is required for those agents to make sure that they're, I mean, right now, you see what we call some hallucinations with, with AI, right? Where it's making up references or making or, or like assuming an answer is desired and giving that answer even if it's the wrong answer. And so I think regulators are going to want to make sure that that does not happen to where there's harm caused to a patient. And I can understand that and appreciate that. At the same time, the rate of development in the AI space is happening so quickly that I think we're not very far away. And we may have tools developed by certain people that's actually today available, that has a level of sophistication to where they could really be implemented today and be right and be really helping patients. And so I'd love to be able to have those integrated and to be able to use them. I just want to make sure the regulators are allowing those to be developed.

[00:16:11] Mathieu: I see, you know, we're getting towards the end of our conversation. And we have a recurring question across all the episodes of the series for the external stakeholders and thought leaders we interact with. And that is, you know, if you had a magic wand, which digital solution would you like to have in your clinical practice that is, from your perspective, missing at the moment?

[00:16:33] Christopher Baugh: Yes. And so I would say if we're talking in the theme of cardiology, I would say around chest pain and really deciding which patients need further chest pain diagnostics is something that I feel like is missing right now. And so right now, if I have a patient with chest pain, which is very common in the emergency department, we're talking about, on a typical shift, I'll see 3 or 4 patients with chest pain. It's about 6% of all ED visits in the United States. And I feel like it can get very complicated to do a lot of the calculations in your head right now. There's the gestalt around kind of how worried you are about that patient. And then there's all the factors around their history and their EKG interpretation and their troponin interpretation. We're now using high sensitivity troponins more often, those might have multiple values that you need to interpret both the absolute delta values. And you're trying to integrate all of that into, am I done? And can I send the patient home or do you need to keep them? I feel like there's an opportunity to integrate tools there that would help give me confidence that I'm not missing unstable angina, which is the element of acute coronary syndrome. That is the hardest to diagnose in the emergency department, because by diagnosis, by definition, you're not going to have a troponin elevation in those patients.

[00:17:56] Christopher Baugh: And that's the population that really would benefit from staying for cardiac stress testing or coronary CTA. And so I feel like if there was a tool that said this patient does not have unstable angina or occlusive coronary disease, and I don't need to do further testing, that would be incredibly valuable to me. I feel like we're very close to having that tool available today. It's and then, you know, longer term, you talked about these AI agents. I'd love to have an AI agent who's kind of like a guardian angel on my shoulder, who's looking at all of my patients and thinking about other chest pain diagnoses besides ACS, looking at, you know, is this a pulmonary embolism? Is this an aortic dissection? What are the life threats that my job in the emergency department is to not miss? I'd love to have that resource available to make me better, to feel like, hey, my job is easier now, and I can rest, you know, easy at night knowing that I didn't miss something dangerous. And I'm not going to get that email in a couple of days that "you remember that patient you saw a few days ago?" I hate getting those emails or the tap on the shoulder because very rarely it's like, hey, they're back. And they brought cookies for the department because they're so happy about how well they are cared for. No. You know, usually it's something bad happened to them. And so I'd love to eliminate that from my practice and my colleagues' practice as well.

[00:19:18] Mathieu: Yeah. Understood. Thank you so much, Prof. Baugh.

[00:19:21] Christopher Baugh: Yeah. My pleasure. This was great, thank you!

[00:19:27] Mathieu: Thank you for listening to this episode of Cardio Insights. We hope you enjoyed it. If you want to dive further into other aspects of digital solutions in cardiovascular and emergency medicine, check out the other episodes of the series to support the show. You can share this episode with your surroundings and leave a five-star review on your preferred streaming platform. Cardio Insights is a podcast brought to you by Roche. Thank you for your time and see you in the next one.

[00:00:03] Mathieu: Welcome to Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine, featuring thought leaders from leading institutions across the globe. This series of conversations unveils the power behind clinical algorithms and decision support tools, and how they are changing the way we diagnose and manage cardiac conditions from their impact on clinical practice, their benefits for patient outcomes, and the challenges they pose to health systems in terms of implementation and regulation. This podcast is intended as a precious resource for healthcare professionals in cardiology and emergency medicine, who are curious to learn more about this exciting field. I'm your host, Mathieu Chaffard, and you are listening to Cardio Insights.

[00:00:52] Mathieu: So welcome Professor Papendick and Professor Daniels to this episode of Cardio Insights, the podcast where we explore how digital solutions and clinical decision algorithms are transforming cardiology and emergency medicine. So I feel lucky to have two guests on this special episode. And for our listeners who might be new to you, I'll give some background on your respective work and focus before we get started with the conversation. So, Professor Papendick, you are an Emergency Physician at the Royal Adelaide Hospital and an Associate Professor at the University of Adelaide Medical School in Australia, focusing your work and research on improving outcomes and resource utilization for patients presenting to the emergency department with chest pain, particularly those with suspected acute coronary syndrome. Professor Daniels, you are a Cardiologist, Director of the Cardiovascular Intensive Care Unit, and Professor of Cardiovascular Medicine and Epidemiology at the University of California in San Diego. You have co-authored more than 200 publications in your field, where you focus on the use of biomarkers to assess cardiovascular risk in a variety of populations. So in this episode, we want to learn from your experience in implementing and driving adoption of decision-making algorithms and digital solutions in your respective practice, understand the challenges that you have faced in that regard, and get some perspective on what these solutions or where these solutions are heading toward in the future. So the first question that we have for the two of you is: how has the emergence of decision-making algorithms and potential digital solutions changed your approach to emergency medicine and cardiovascular care?

[00:02:39] Cynthia Papendick: Well, I guess from an emergency point of view what we call chest pain pathways, so some of those decision algorithms, I think, have been a part of emergency medicine for quite some time now. I guess the most important one has been things such as the Heart score. So these allow for example in the emergency department it's quite a, it's quite a diverse group of people with undifferentiated problems. And then there's also the staff who are also a diverse group of people, not necessarily with problems, but a diverse group of people with different interests. For example, some people are more interested in trauma or, you know, ultrasound. So being able to have these pathways allows people to maintain, you know, a homogeneous and evidence-based approach to common cardinal presentations, for example, chest pain. And I think the Heart score is one of the most well-known. In Australia, we also have EDACS and things. I think for me what has changed and I'd be interested to hear what Lori thinks about this, is that now with high sensitivity troponins, the pathways now incorporate a separate, very complicated troponin algorithm. And then you need to combine all these metrics together in your head, often in the head of an orthopedic, you know, RMO or house officer who is doing, you know, ED for six months or something. So not necessarily an expert. And I think we could really use digital sort of algorithms to help calculate some of these more complicated metrics, put them into position with the other things that we get from the patient, patient-specific demographics, and then come up with an evidence-based, you know, approach that's safe for the patient.

[00:04:41] Lori Daniels: Yeah, it's a really good point. I would say, you know, we're not using digital algorithms so much, but we're using algorithms. And the most common one I would say is the chest pain pathway kind of derivations of the ESC zero-one hour approach, and we started using that when we transitioned to the gen-5 troponin to the high sensitivity troponin as a way to get all on the same page, make sure we're still practicing evidence-based medicine. We were behind the rest of the world in implementing it in the United States, but when we did, we wanted to make sure we had a pathway. People weren't familiar with the test yet, and so we wanted an algorithm to say, look, this is tested not necessarily in the US population, but in chest pain populations. It's safe. We can all be on the same page here. We made a few adaptations until we got comfortable with it, incorporating things like risk scores or high-risk EKGs and saying it's okay to leave the pathway. If you're uncomfortable, do it safe. But in general, let's follow this pathway.

[00:05:46] Mathieu: And what have been some of the challenges that, you know, regarding the, in your own practice, in adopting these tools or apprehending these tools? What were some of the issues? Or, you know, you mentioned that in the US it came later compared to the other places of the world. So what were certain things that you learned from former implementation that, you know, helped you navigate that?

[00:06:09] Lori Daniels: Yeah, I think so. One of the challenges is if we're implementing a zero-one-hour pathway, it's not that common that your testing is exactly at one hour. And that's, you know, the life intervenes and the craziness in the emergency department. And I think that's one area of opportunity where digital algorithms can be quite useful, because a digital algorithm can say, okay, well, this sample is really at 90 minutes. And so here's what you want to look for. And you don't need to rely on someone's brain to kind of interpolate for you.

[00:06:42] Cynthia Papendick: Absolutely agree with that. Yeah. So that by the time you know a sample is taken, it gets sent off by the time someone's acknowledging what the sample is and if it's in a grey zone or whatever, you have these big delays and we're calculating based on a zero and one hour, at least in my shop, that's the one we use. And often the metrics are different than that. The real-world metrics are different. So to have an, you know, like a digital platform that could do those calculations would be amazing.

[00:07:13] Lori Daniels: I also find some of the beginners you mentioned, you have a lot of people visiting in the ED. We have, you know, new trainees, also people passing through on the cardiology service who aren't necessarily going to be cardiologists and they don't necessarily understand the gray zone as well. And they think, oh, there's no delta. We're good to go after one hour, and they don't necessarily follow the pathway and see, no, if you're in that gray zone you need another sample. Yeah. And I think a digital algorithm that said stop, repeat. It would be quite useful for these more complicated situations.

[00:07:48] Cynthia Papendick: Absolutely. Absolutely.

[00:07:50] Mathieu: Do you see a lot of disparities between health systems within your respective country in terms of adopting such algorithms for clinical care, or are like the Heart score I assume it's pretty much everywhere, maybe some more, like you mentioned, next-generation troponin. Is it something that's fairly well distributed in your respective health system, or do you see a lot of disparities in terms of how these tools are used in clinical care?

[00:08:18] Cynthia Papendick: Well, it's funny because while we were walking over here, Lori and I were having a discussion about how looking at ACS in particular, you know, from an emergency physician point of view, I want something to be exquisitely sensitive, to make sure that when I send someone home, I have full confidence that they've actually ruled out. You know, whereas Doctor Daniels wants to make sure that, you know, okay, I'm taking this person to the cath lab for a reason, particularly if she doesn't have the opportunity to see a patient or to get. She's trusting me that I'm able to do the metrics to figure out, actually, this is a rule-in. So I think that's that's where sometimes the disparities are. So having a digital thing, which could really piece out those intricacies and take into account like other patient characteristics. I mean, for us, you can have a raised troponin in a patient who has a gastrointestinal bleed, let's say, and they're actually having a type two myocardial infarction. And really, the worst thing we could do, much as I appreciate and respect you, is send this patient to the cath lab. You know, you want this patient to have treatment for their GI bleed because that's what's causing their symptoms. So to be able to have something that would pull in other information like, oh, the hemoglobin is 61: have you thought about that?

[00:09:41] Lori Daniels: And renal function is a big thing, and being able to incorporate all aspects of a patient if we're thinking about heart failure. Another area I think digital algorithms would be phenomenal because you incorporate BMI, kidney function, and not all natriuretic peptide levels are created equal depending upon the patient who's providing them.

[00:10:05] Cynthia Papendick: And then ideally, maybe you could have a prior one to compare it to. So the ability to have more of these complicated metrics put in together, because I find that often the inpatient teams will overestimate the influence of things like renal function being a problem and everything's chronic, everything's a type two, you know, especially at 3:00 in the morning. That's how the best way to tell the difference between a type one and a type two is what time is it? So I think those things could be evened out by digital processes or at least assisted.

[00:10:39] Lori Daniels: Cynthia, that's a super insightful comment because having prior levels, whether it's troponin or peptide for a given patient, is so important for getting a diagnosis and understanding if this is an acute or chronic presentation. And I totally agree, that's somewhere where digital algorithms can really be useful. Yes. And helping determine a delta for instance. Yes.

[00:11:03] Mathieu: You mentioned especially Professor Daniels, that at the moment they were there were not so many digital solutions available. And you probably also see in your practice or when you go through even like just the news, new solutions coming up. Is there something where you would say you have an interest that you would like to have that type of solution at the moment that you don't in your own practice, or how do you look at these things, maybe from a use case and from an evidence level perspective?

[00:11:34] Lori Daniels: Yeah. So I think two answers there. One kind of getting also back to disparities in the health system. We do have a lot of differences back in the United States. Not everyone is using high-sensitivity troponin yet. And then there's I and T. There's point of care. So when we're a transfer center a referral center, when we get patients from outside, you know, their troponin they have that they're sending us is very different from what we're using a lot of times. So a digital algorithm potentially could help to translate that. What would I like to see? I think my biggest wish list would be in in the heart failure diagnosis realm. Like I mentioned before, taking into account age, comorbidities, body habitus, and interpreting all of that prior peptide levels and helping interpret. I think that would be quite useful.

[00:12:31] Cynthia Papendick: Absolutely. Because those other metrics are so important to really doing the right thing by the patient. Like for myself, I get what I consider to be community-acquired acute pulmonary edema, which is when patients who have an underlying heart failure diagnosis come in with their pneumonia. Their BMP is 5000. What does that mean? You know? Well, if you had something that could say, oh, well, last week or two months ago, it was, you know, 400, then we know that those things are playing in concert, but we can also pick the right team to look after the patient. Is sepsis the driver or is, you know, do we need a cardiologist to sort out their heart failure?

[00:13:16] Mathieu: And in terms of the level of evidence that you would like to see that supports the use of potentially a new tool that you know, someone would bring to you and say, okay, you need to try that out for either triaging or detecting some of the conditions you work on. How do you evaluate that based on what's available?

[00:13:38] Lori Daniels: There's different metrics. And again, it depends who you ask. If you ask Cynthia there's going to be emergency department metrics, which I'll let her answer. From a cardiologist point of view, if we're going to be using a metric outside of the ED, you know, we want to make sure that it's improving ultimately outcomes. That's a really high bar, right? You have to show that, hey, using this algorithm is going to make you miss less MIs, you know, improve mortality, reduce hospitalizations. That's kind of a high bar. I think some of the ED metrics might be more attainable, efficiency and care, efficiency and utilization of resources.

[00:14:21] Cynthia Papendick: Absolutely. And I also think that there's that zone in between of emergency departments often think, oh, the best thing for the patient is the patient should be admitted. You know, they need to go, you know, and then learning Earning which patients actually would benefit from further intervention versus which patients just need more resuscitation, or, you know, other, other bits and pieces that are important to them and not getting stuck on, oh, the troponin, is this. Really taking in holistically all the things that are going on.

[00:14:54] Lori Daniels: Yeah. And I mean, ultimately from a hospital-wide perspective, if you can make a system more efficient and even save 30 minutes or an hour, you know, that may not be a wow for physicians, but for the hospital system as a whole, it can really be important.

[00:15:10] Mathieu: Yeah. Yeah. We talked about that with Professor Than on the role of digital solutions in addressing overutilization of healthcare resources, especially like in the ED. So that's one use case. How do you see, you know, the current and future role of technologies like artificial intelligence in the development and adoption in a way of digital tools in clinical practice? Is there like a use case that you think of when you think about these technologies in your respective areas?

[00:15:45] Lori Daniels: We're seeing yeah, I think there's a lot of potential there. We have to be careful to implement it, you know, in a correct way. Yeah, in the correct way where we know we're doing good. But there's already, you know, interpretation of electrocardiograms, for instance. And if you could merge that with biomarkers, I think that could be extremely useful. But even interpretation of EKGs, like there's a lot of what we call ST-elevation myocardial infarctions, occlusions of the arteries that don't actually have ST-elevations. And if you incorporate that with AI, plus or minus biomarkers for the non-ST-elevations, and it could really improve efficiency and potentially improve outcomes as well.

[00:16:27] Cynthia Papendick: 100% I agree with that. I think that one of the things I find when I look at an ECG is you get the computer printout, which I tend to, you know, I ignore it at my peril. If it says acute MI all over the top, you know you want to have a look, but often it'll say things that don't seem to make a lot of sense. So you really spend your time. So it's based on my ability to assess that. And as you're saying, acute myocardial infarction with an OMI, you know, with an acute occlusion MI, because ST-elevation is very common and it's not always an occlusion. So being able to discern those subtler changes of posterior MIs or different changes in T waves, which are very subtle, if you're not interested in EKGs people are going to, you know, people miss it. Yeah.

[00:17:17] Lori Daniels: And I feel like that's. Oh, sorry.

[00:17:19] Cynthia Papendick: Yeah. No, just that's where AI, because I've always been disappointed with the computer interpretation of EKGs, you know.

[00:17:28] Lori Daniels: I'm kind of in shock with how bad it is. It's part of the reason I have a job. Like one day a week, I'm sitting there reading EKGs and overriding them. But that seems like low-hanging fruit that could get better pretty quickly. Yeah, and then you take the next step and you incorporate that ECG with your digital pathway for chest pain. And you could really improve things.

[00:17:48] Cynthia Papendick: Because most of the pathways that we have now don't really include ECG. Often ECG is sort of tossed aside as either ischemic or not. And it's like there's a lot more in there.

[00:17:58] Lori Daniels: I think there is a lot more.

[00:18:00] Cynthia Papendick: I think we're losing a giant amount of information about what's going on.

[00:18:06] Mathieu: It's an interesting segway for the recurring question we have across all the different episodes, which is, and you already expressed some of the wishes in terms of what solutions you would like to see. But if you had a magic wand, which would be like one digital solution you would like to have in your clinical practice that you think is very missing at the moment?

[00:18:29] Cynthia Papendick: I would say I would love to have better and more reliable ECG interpretation. I mean, I think the key thing with ECGs is that it is interpretation, because literally I could look at an ECG and then five other cardiologists look at it and you'll get 12 answers. Do you know what I mean? So I think it is still interpretation. It needs to be taken in the clinical context. But if you had some algorithm that would be able to say, you know, consider these, you know, this ST-depression and this thing, is this a high lateral problem. You know, that kind of a thing that people often miss. So just to consider that this is there, it should be able to. Yeah.

[00:19:10] Lori Daniels: Yeah. And and I mean I agree with Professor Papendick. And at the same time, we have to realize it's kind of the contrary opinion in the world of biomarkers. Right? A lot of these algorithms don't even incorporate ECGs. But as a cardiologist, I'm uncomfortable with that. And an algorithm could tell me, oh, low risk, send them home. But if I'm seeing dynamic, you know, down-home ST-changes, I don't want someone ignoring that patient. So I agree 100%. Yeah. Even though it's not a popular opinion necessarily.

[00:19:46] Mathieu: Yeah. Thank you so much.

[00:19:53] Mathieu: Thank you for listening to this episode of Cardio Insights. We hope you enjoyed it. If you want to dive further into other aspects of digital solutions in cardiovascular and emergency medicine, check out the other episodes of this series to support the show. You can share this episode with your surroundings and leave a five-star review on your preferred streaming platform. Cardio Insights is a podcast brought to you by Roche. Thank you for your time and see you in the next one.

[00:00:03] Mathieu: Welcome to Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine, featuring thought leaders from leading institutions across the globe. This series of conversations unveils the power behind clinical algorithms and decision support tools, and how they are changing the way we diagnose and manage cardiac conditions from their impact on clinical practice, their benefits for patient outcomes, and the challenges they pose to health systems in terms of implementation and regulation. This podcast is intended as a precious resource for healthcare professionals in cardiology and emergency medicine, who are curious to learn more about this exciting field. I'm your host, Mathieu Chaffard, and you are listening to Cardio Insights.

[00:00:51] Mathieu: So welcome Prof. Body To this episode of Cardio Insights, the podcast where we explore how digital solutions are transforming cardiology and emergency medicine. For our listeners who might be new to you, you are a Professor of Emergency Medicine, an Honorary Consultant in Emergency Medicine, and Group Director of Research and Innovation at Manchester University NHS Foundation Trust. Next to your teaching activities and clinical practice, you also serve as a Member of the Committee for Clinical Applications of Cardiac Biomarkers with the International Federation for Clinical Chemistry. Throughout your career, you have published over 200 peer-reviewed publications and your research focuses on acute coronary syndromes, diagnostics, clinical epidemiology, biomarkers, and the philosophy of emergency medicine. In this episode, we want to explore with you what's your perception on the use of digital solutions in your practice, and also, the methods and the techniques we use to evaluate these tools and the impact that they have on patient outcomes and the challenges that evaluating these tools poses as well. So the first question I have for you is if you could give us an overview of what type of digital solutions you use in your daily practice and how your work has also evolved over the past years with the emergence of new solutions?

[00:02:16] Richard Body: Yeah, sure. So, well, I got into this area by being interested in the development of a digital tool, actually, which ultimately became the decision aid called T-MACS, the troponin only Manchester acute coronary syndrome decision aid. And our idea was to use troponin as one component of a clinical prediction model and to pull in other information from a patient's symptoms, past history, and to take a sort of data-driven approach to this, so that we'd use the information that gives us predictive value and combine it in the optimum way to calculate the probability that a patient has an acute coronary syndrome on an individual level. Yeah. We'd handle troponin as a continuous variable. Or at least we had. We were open-minded to to using troponin as a continuous variable. I mean, I think we had to explore the data to see if that was the best way to use it. And it turned out it was, you know, no cut off that we actually troponin behaves best for predicting the probability of acute coronary syndromes when we handle it as a continuous variable. So we derived this prediction model. And then we've evaluated it in multiple validation studies. We implemented it in 2016 I led a project to implement this across our region in Greater Manchester at 12 hospitals after that. And we've been using it successfully since then. I think that it has some particular advantages, actually. I really like the fact that we get an individualized probability.

[00:03:46] Richard Body: It informs me as a clinician as to the individual risk for my patients. It allows me to share that with my patients as well, to engage in shared decision-making. And it tells me what to do. So if I just use the test in a dichotomous way, positive or negative there's, there's so much uncertainty in that, you know, the positive predictive value of a troponin above the 99th percentile is actually relatively low. And the negative predictive value below the 99th percentile is not that high. But when we bring in all of this information, it's much more rich. The first decision we've got to make in the emergency department is whether the patient needs to stay with us for more investigation or go home, and we can use troponin in a dichotomous way to do that. That's fine. But there's so much more that we've got to do. We've got to decide whether they're coming into hospital and if they are coming into hospital, where do they go to? Can they go to an ambulatory care area, or do they need to go to a coronary care unit? And when you've got a digital tool that sort of gives you a continuous variable as an output with probability, it allows you to make those decisions with a much better evidence base than if you just dichotomize results or impose any cut-off.

[00:04:55] Mathieu: Yeah. Understood. And you know, if we shift gears towards the, you know, understanding what's the evidence that's needed to support the use of these tools and how this evidence is built? Can you take us through in maybe, you know, following the example of that project that you worked on and that tool that you developed, how you came about building the evidence that would eventually create the trust towards other practitioners to use that and, you know, be, let's say, get a broader use in the cardiovascular and emergency medicine community?

[00:05:30] Richard Body: Yeah, sure. I guess I mean, it starts with refining your research question, deciding what you want this tool to do, and setting that out very clearly, because that's going to inform your whole approach to evidence generation. For us, the key thing was to decide who's coming in and who's going home. But then to risk stratify the ones who are coming into hospital. And then you set about doing your studies, observational studies initially to derive the decision rule. So you want to include the patients who are going to have who have the condition suspected. Collect all of the data that you're going to need, and then use your usual machine-learning techniques to derive the optimal algorithm. The next step is to validate that externally in different populations. And again that's usually done in observational studies, diagnostic test accuracy studies. And many decision aids actually stop there. Yeah. We get the observational data and some of the recent ones have been derived and validated in really large cohorts, tens of thousands of patients, which is fantastic because that increases external validity. But we often stop that at the observational data. And what we really need to know, I think, is to go a little step further. How does it actually change patient care? And so for that we probably need the interventional studies, randomized controlled trials to see, does it actually affect patient outcomes when we actually use it? Can we get more patients home safely from an emergency department than we would if we just used troponin by itself, for example? And can we make better decisions for those who are coming into hospital, be more selective about the use of PCI, for example.

[00:07:02] Mathieu: Yeah.

[00:07:04] Richard Body: And then we can go even further than that, you know, than an RCT. There's plenty more to talk about.

[00:07:09] Mathieu: And why do you think we stop in most cases at the observational study level? Is it a lack of means? Is it the complexity that's implied with running clinical trials? Why is that?

[00:07:20] Richard Body: Well, I think I mean, a good external validation study is the bare minimum that you need for implementation. And I'm fine with that. I think that's okay because you don't want too much inertia. It takes too long to get innovations out into practice. Yeah. You know, there's a statistic that goes around saying that on average it takes 17 years from innovation to adoption to practice. And if for every innovation we're taking 17 years, the pace of change in healthcare is not going to be very fast. So I'm all for, you know, being an early adopter after you've got the robust data that show accuracy implemented in practice. But at the same time as we've implemented in practice to get findings that people trust and, you know, on a wider scale to get more general implementation. I do think the RCT plays an important role. I reflect on my experience with T-MACS, for example. I don't think we've got that wide adoption. I think the heart score is better utilized internationally than the T-MACS score. And yet, when we've done direct comparisons, T-MACS is more accurate because you better risk stratification. So why is that? Maybe it's because we didn't do the RCT to show that actually the use of this will actually change decision-making and make us more effective in practice. To show that it's cost-effective. So I do think it's an important step, hopefully in parallel with clinical adoption for the early adopters. But I don't think you can ever replace the RCT. There are alternatives. I mean, as we as the field moves, maybe people could take the example of people like Prof. Martin Than in New Zealand leading the ICare project, where they're doing a sort of step wedge implementation for point of care testing. And this is kind of real-world data. So they put the point of care tests in practice and then evaluate how it's going outside of the context of an RCT. And I'd love to see more methods like that if we can robustly evaluate effectiveness and cost-effectiveness using such real-world approaches. I think that would be a real step forward for us.

[00:09:12] Mathieu: Yeah, it accelerates in a way as well. The access to innovation where you mentioned, you know, 17 years, it's like, you know, generation almost. And, you know, when we were preparing the conversation, you also touched on the importance of evidence that's generated, you know, towards that would speak more for patients rather, I mean, for clinicians as well, but beyond, let's say, the pure clinical evidence in terms of the efficacy what can you share with us in terms of what are some of the metrics that matter or that we should make transparent to patients with regards to the use of these tools? And yeah, it would be interesting to hear your perspective on that.

[00:09:51] Richard Body: Yeah, I mean, I think this is far from a done deal that it's decided in the field about what outcome measures we should use. But if you think about it, when we when we're evaluating a digital tool, we look at efficacy. And when we talk about efficacy with the diagnostic test, we're generally thinking about measures of diagnostic accuracy like sensitivity, specificity, the proportion of patients that might be discharged appropriately, that kind of thing. We might look at effectiveness. You know, when we actually use this in practice, how many people get home and what are the outcomes of those patients in terms of prognosis, major adverse cardiac events in the future and cost-effectiveness are the sort of quantitative outcomes. But I think there's so much more to it than that that you mentioned the patient-related experience measures that we could use. So I'd be interested to know, you know, how do patients feel? One of the key advantages of a digital tool is that we could potentially empower patients with more information about their condition, and allow them to play a more active role in their care through a shared decision-making approach. We can get richer diagnostic information through digital tools. Why don't we share that with patients and allow them to engage in the decision-making process? And if we do that, how do they feel about it? What's important to them about their care? And are we able to do that better through the use of a digital tool than we would just through using a diagnostic test with dichotomous results?

[00:11:10] Mathieu: Yeah, understood. And then you touched on the relevance of real-world data and real-world evidence mentioning the work from your colleague, Prof. Martin Than we also got on the on the podcast in another episode. How do you see real-world evidence and real-world data helping to evaluate and also provide, you know, faster access to innovation for clinicians or hospitals across other, you know, maybe potentially beyond cardiovascular and emergency medicine as well?

[00:11:42] Richard Body: Well, one of the big problems we've got with evidence generation at the moment is that, you know, we have we've talked about this. We do the bare minimum. You know, we'll do the external validation. We may even do the RCT. And it will show efficacy or effectiveness for the whole group as a whole. Yeah. And so we know how it works with a one-size-fits-all approach. But increasingly we're recognizing across health and care that this can leave people behind. You know, we get gender inequities. We get disparities by ethnicity, by socioeconomic deprivation. When we talk about digital tools as important elements of digital exclusion.

[00:12:17] Mathieu: Exactly, yeah.

[00:12:18] Richard Body: And there may be differences in performance by age or co-morbidities. It's really difficult to explore those in an RCT because you power your RCT for the whole group, not for subgroups. So your analyses in the subgroups are always going to be underpowered. And that's where I think the real-world data comes in and is so important. We don't want to slow the pace of innovation down so much that we get so much inertia that we can't adapt until we've run powered RCTs for every single important subgroup.

[00:12:44] Mathieu: Yeah, it's not sustainable.

[00:12:45] Richard Body: It's not going to be possible, is it? We wouldn't have any innovation. It would just literally not happen. But what we could do is commit ourselves to say, well, actually, once we've got enough data to get this over the line and adopted, we need to be sure that we do sort of the phase for evidence generation where we collect data in the real world, and that in a larger number of patients and with the informatics infrastructure that we have nowadays in high-income countries, it should be possible to for us to collect that data and explore how digital tools are performing in all of those different important subgroups. And that would then allow us to refine the digital tools to make them more personalized for each subgroup. And ultimately, when you have a large enough sample, you can really tailor the outputs of these digital tools to the individual based on their personal characteristics.

[00:13:40] Mathieu: And I was also asking myself, you know, in light of, let's say, the use of wearables that are, you know, they also provide, you know, very interesting information. Now in terms of ECG, in terms of heart rate, potentially blood pressure as well. I guess it also makes the acquisition, the treatment, and obviously the execution of these sort of like decentralized trials, more accessible than, let's say, a standard randomized controlled trial. So there's also a cost-benefit of running these types of studies that add to the evidence base that we have for these solutions.

[00:14:12] Richard Body: Yeah. Totally totally. And you know, there's it's such a complex field, but with digital tools, things like you mentioned the wearables. I mean, as our population ages and has increasingly more co-morbidities and comes in with more complex health problems and our hospitals fill up and emergency departments get crowded, we need to think about new ways of delivering care. So in the UK, there's a big drive for community-based care, hospital-at-home and things like that. So wearable technology that allows us to monitor patients as well as we would in hospital, but in their own environment become increasingly important. And when we think about digital tools, it'll be increasingly important to be able to assimilate that data with our digital tools so we can bring in information from all sorts of different sources, including wearables, and get sort of dynamic risk prediction that will inform the care of patients wherever they are.

[00:15:05] Mathieu: Understood. You know, we have a recurring question we ask every guest that we have the chance to have on the podcast. And, you know, you can bring some some fantasy to the discussion in a way. But if you had a magic wand, you know, which digital digital solution would you like to have in your clinical practice that's currently missing? You know, beyond, let's say, the solutions we've talked about in our conversation?

[00:15:29] Richard Body: Well, I'd like my perfect decision tool to use a sort of large language model so that we could talk to it.

[00:15:37] Mathieu: Okay.

[00:15:37] Richard Body: Yeah, I can imagine, you know, ChatGPT you realize the possibilities with the generative AI and, and and, you know, that sort of solution would be amazing. So as a clinician, I could see my patients we could have ambient documentation. So it's listening to the consultation. It's relaying it back to me. It's interpreting the consultation when I come up with my theory about what's happening with the patient. I'm going to get some validation from the AI model.

[00:16:05] Mathieu: Or be challenged.

[00:16:06] Richard Body: Or be challenged. Exactly. Make me think about things that I haven't considered. When I get my diagnostic information back, it's going to help me to apply the digital tools, and it's going to help me interpret the information from that in terms of probability, the way up the risks and benefits in a data-driven way. You know, if I decide I want to give the patient antiplatelets or go for angiography, then it's going to help me to weigh up the risks of those treatments against the benefits, given the uncertainties that are inherent in information that we've got. And I think that then it could allow us to integrate that with our communication with the patient, so that we can give them something to go away with. If we send someone home, for example, and we say, actually, it's not a heart attack, you don't need to worry, it's going to be fine for you to go home. Patients often express this dissatisfaction that we say it's not a heart attack. You can go home. We don't tell them what it is. We don't tell them how to stay healthy. We haven't addressed the fact that they're smoking 40 a day or not exercising or, you know, there are lifestyle choices that they could they could modify to stay healthy for longer. And we could integrate that with this digital tool so that there's sort of some ongoing follow-up and education for the patient that empowers them as a sort of active player in their healthcare, rather than a passive recipient.

[00:17:17] Mathieu: Yeah, it makes a lot of sense. Thank you so much, Prof. Body.

[00:17:20] Richard Body: Thank you very much.

[00:17:25] Mathieu: Thank you for listening to this episode of Cardio Insights. We hope you enjoyed it. If you want to dive further into other aspects of digital solutions in cardiovascular and emergency medicine, check out the other episodes of the series. To support the show, you can share this episode with your surroundings and leave a five-star review on your preferred streaming platform. Cardio Insights is a podcast brought to you by Roche. Thank you for your time and see you in the next one.

[00:00:03] Mathieu: Welcome to Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine, featuring thought leaders from leading institutions across the globe. This series of conversations unveils the power behind clinical algorithms and decision support tools, and how they are changing the way we diagnose and manage cardiac conditions from their impact on clinical practice, their benefits for patient outcomes, and the challenges they pose to health systems in terms of implementation and regulation. This podcast is intended as a precious resource for healthcare professionals in cardiology and emergency medicine, who are curious to learn more about this exciting field. I'm your host, Mathieu Chaffard, and you are listening to Cardio Insights.

So welcome, Prof. Than, to this episode of Cardio Insights, the podcast where we explore how digital solutions are transforming cardiology and emergency medicine. For our listeners who might be new to you, you are an Emergency Medicine Specialist and the Director of Emergency Medicine Research at Christchurch Hospital in New Zealand. As one of the most published emergency medicine physicians in Australasia, your research interest is centered on quality improvement initiatives for the emergency department and you focus on cardiac biomarkers. You were notably the principal investigator of the ASPECT study that involved nine countries in Asia-Pacific and more than 3000 patients, as well as the principal investigator of a randomized controlled trial with a two-hour diagnostic protocol for cardiac chest pain evaluation in the emergency department. So in this episode of the series, we want to discuss with you how clinical decision-making algorithms and digital tools can mitigate mortality risks in healthcare settings. And we will discuss the risks that overutilization of healthcare resources poses in this setting, and how certain digital solutions can contribute to addressing this important issue. So to kick us off, can you share with us how you see the current situation in emergency cardiovascular care and why we need better tools to address the needs of healthcare professionals and patients in that regard?

[00:02:18] Martin Than: Yeah. I think that we all know that chest pain is a very common problem for emergency departments in the developed world, and that's partly because we have public awareness campaigns encouraging the general public to seek medical help when they have any symptoms that could be related to a heart attack, to get them to the hospital. The consequence of that happening, of course, is that very many people then go to the hospital, and the number of people that have a heart attack is relatively small, and you have to try and determine from the very large number of patients which ones need urgent treatment or not which is an increasingly challenging task because there are very many patients. But it's not always that easy to work out which patients you should focus on. Cardiac troponin has obviously been a very groundbreaking test in terms of helping us do that. But as it's become a more precise test and we get a lot of information from it in terms of numbers from it, it's become harder and harder to interpret it sometimes about what the test results mean. And so now differentiating between what's a type one myocardial infarction, which it might be from coronary obstruction, from type two, from systemic supply demand issues from injury from chronic injury is actually quite a hard process in the emergency department setting.

[00:03:44] Martin Than: And also the test, we tend to use what is a very powerful and quantitative variable that runs really from naught up to thousands in a very blunt manner, which is to call it positive or negative around an upper reference limit, and it's an upper reference limit that has been determined on a general population that's probably not representative of a particular individual, particularly if that individual is female comes from a different social or economic background or ethnic group. And so the challenge is how to try and use the tools that we have now to be more specific about patient decision-making for the individual themselves and also to be able to target it so you get the right diagnosis at the end of it. The point being with that is that putting the patient in the right category then allows you to focus on which investigations or not they may need next.

[00:04:41] Mathieu: And how is your practice in that regard, or what were the evolutions you've seen in the past year in terms of, you know, getting access to these tools and how you were able to triage or redirect patients to different types of care based on the symptoms they were showing up in the emergency room.

[00:04:59] Martin Than: So I mean, this has been an evolution process over almost 15 years now that we've been doing various cycles of improvement of chest pain pathways. And we're not quite there yet with digital solutions, I think, I think there's been a lot of development since we first published a paper on machine learning in myocardial infarction called MI^3 in 2018, and it's, to my knowledge, not been formally implemented anywhere. And there have been a number of other good studies in that space by the Artemis group led out of Hamburg, and also the CoDE-ACS group led out of Edinburgh. And they're much better tools than I think have been available in the past. But actually operationalizing them is quite a challenge. And it's one thing to have a tool that is predictive. It's completely another thing to be able to make that tool easy to use. And part of the doctor's workflow in an easy, in an easy way. And there are two major challenges that I see there. So first of all, no one wants to have to reenter information that's already existing. Even little things like having to enter the patient's age or sex into the tool. That information should automatically be pulled from the patient information system into whatever digital tool you're going to use.

[00:06:20] Martin Than: Also, once the guidance is given by the tool you don't then don't want to have to transcribe it into your patient record. So ideally the tool will write your records as you go along. So to me, the ultimate aim of what we want to achieve is something that is constantly pulling information as it becomes available about the patient that you're assessing and is assimilating the way in which it forms a patient medical record as you go and gives guidance as you go, and it updates the guidance. As new patients come, new information becomes available. And when the clinician has done that, their clinical record is written so they don't have to go and spend time then typing a record up. And what is really important about that? That is that clinicians won't use tools that are a burden. But just as we've discovered with the iPhone or smartphones that we use now, people will always use things that make their life easier. That's why people like using apps, okay? Because it makes what was difficult before much more simple. So that is the ultimate game aim. You have to create tools, digital tools that actually are so easy to use that doing the right thing for the patient is the easiest thing to do.

[00:07:36] Mathieu: And I'd be curious to learn from you about what is the current level of evidence, or what is the impact that these tools have on two things, one being the, you know, the mortality risks in the emergency department. The other one that I alluded to in the introduction, which was around the overutilization of emergency departments that you also mentioned regarding patients coming up with chest pain symptoms that are actually maybe not related to that. So how do you see the impact of these solutions on these two aspects?

[00:08:06] Martin Than: Yeah. So I think it's quite hard to show mortality difference because the number of people that are actually dying is quite small. But if one looks at in a particular case example it can sometimes based on our subjective impression as clinicians of an ECG, be difficult to tell whether certain patients need immediate attention for possible coronary occlusion. And this is where we're getting away from the idea of just a STEMI to occlusive myocardial infarction or obstructive myocardial infarction. There are tools out there. They've now got a library of many thousands of ECGs so that the clinician can take a photo of these they've got in front of them and put in some simple clinical information. And it will give you a percentage likelihood that there is some sort of critical abstraction happening. So for me, that is a potential mortality saver, because there will be some EKGs which aren't immediately indicative of that scenario that the tool will allow you to identify. So that has a potential mortality benefit. I would have said. In terms of overcrowding, I think we do over-investigate patients. I don't think there's any doubt about that. And that's, that's that's human nature because we're worried about making a mistake. And for someone coming to harm, if you have a tool that allows you to refine your predictions in a probabilistic way, it gives you a percentage likelihood and gives you guidance related to that likelihood. And I think that will allow you to rationalize your decision-making and have much better resource stewardship about who you investigate and who you don't. So there is the ability there to refine decision-making and then in relation to that, perhaps be more efficient about the way you use the resources you got, which can therefore have an important impact on overcrowding.

[00:10:00] Mathieu: I see. And you know, we one one question we ask all the guests on the podcast is to imagine, you know, what would be one ideal solution that for the moment, you see as missing in your current practice, if you could have access to something new, what would that be? And for what pain point that would address?

[00:10:23] Martin Than: Yeah. I think the perfect tool would be embedded within the local IT infrastructure system. So in the US that might be an EMR electronic medical record producer like Epic or Cerner. So that as the clinician gets the information from the patient. It's being entered into the record. They could do it at the bedside or on a smart device or a tablet. Or they could do it at the desktop when they sit back down. And as they're doing that, as soon as the blood test result comes back, it populates the same tool. It weights the probabilities in the background and then updates the guidance and then sends alerts that to the clinician that the probability for the patient has changed. And actually sends an alert when a result is available that they weren't aware of as well. And at the same time writes their record. And you should then be able to have links to investigations. So for example, if the guidance because there's a certain probability of coronary disease is the patient should have a CT coronary angiogram, there should be a link there that quickly orders the test so the clinician doesn't have to go and enter a test request form, etc. So again it's about efficiency to me. I'll repeat what I said. The perfect tool makes the best decision for the patient the easiest thing to do.

[00:11:53] Mathieu: So you don't see that there would be like a new area, that's some kind of new technology, might be able to address in the near future that's that's part of your clinical practice?

[00:12:05] Martin Than: Well, actually, in our own hospital, we've developed a bedside tool that we use for assessment of patients with, for example, for chest pain. But for us, the integration isn't there. So we actually have a good tool. And actually the prediction value of it isn't bad. It is quite good. But we have to do that as a web tool. And a web tool isn't necessarily a bad way of doing things, but these tools are almost there. You know, the prediction value of the machine learning models that I was mentioning earlier, MI^3, Artemis, and CoDE-ACS, are available now, but they haven't been integrated into workflows. And that's the next step. I think that we actually have tools that we've validated across multiple populations that take into account individual characteristics of patients, and that then are integrated into the electronic system of their local hospital and environment. You can bring those things together. Then you have something that's genuinely useful.

[00:12:59] Mathieu: Yeah. So from your perspective, the biggest challenge or the biggest need is actually on integrating clinical workflows and not necessarily pushing.

[00:13:07] Martin Than: And it's quite often ignored. And I think that generally the people that are working on these things are working in silos. There'll be people that are working on the machine learning or the mathematics, but they actually don't actually understand the clinical, the IT integration, and vice versa. So what it really actually needs is a team of stakeholders that understands the needs of the end user.

[00:13:28] Mathieu: So you think there is a responsibility from maybe let's say healthcare manufacturers to engage more closely with healthcare professionals to guarantee that this is given once the new, let's say these new products are launched.

[00:13:43] Martin Than: Yeah. Well, I'll flip that question around. I'd say it's my experience that there's lots of clever, shiny tools around where there hasn't been adequate engagement with the end user and how it's going to be deployed before they've gone too far down the development process. So effectively, that answers your question. Yes, you should actually find out how something's going to be used before you really start developing it properly.

[00:14:06] Mathieu: Understood. Thank you so much, Prof. Than.

[00:14:09] Martin Than: Thank you.

[00:14:14] Mathieu: Thank you for listening to this episode of Cardio Insights. We hope you enjoyed it. If you want to dive further into other aspects of digital solutions in cardiovascular and emergency medicine, check out the other episodes of the series to support the show. You can share this episode with your surroundings and leave a five-star review on your preferred streaming platform. Cardio Insights is a podcast brought to you by Roche. Thank you for your time and see you in the next one.

[00:00:03] Mathieu: Welcome to Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine, featuring thought leaders from leading institutions across the globe. This series of conversations unveils the power behind clinical algorithms and decision support tools, and how they are changing the way we diagnose and manage cardiac conditions from their impact on clinical practice, their benefits for patient outcomes, and the challenges they pose to health systems in terms of implementation and regulation. This podcast is intended as a precious resource for healthcare professionals in cardiology and emergency medicine, who are curious to learn more about this exciting field. I'm your host, Mathieu Chaffard, and you are listening to Cardio Insights.

[00:00:51] Mathieu: So welcome, Prof. Hautz to this episode of Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine. For our listeners who might be new to you. You are an Associate Professor and Head of Emergency Medicine at Bern University Hospital, with a background in anesthesiology and medical education. Throughout your career, you have co-authored more than 100 peer-reviewed publications in areas including diagnostic decision-making in clinical medicine, digital decision-making tools in emergency care, and evidence-based medical education. In this episode, we want to explore with you how digital decision support algorithms can change the management of acute and chronic heart failure in emergency settings, notably through early identification, treatment optimization, and care transition improvements, and also the limitations that we see from these tools these days. So to kick us off, could you take us through the complexity of clinical decision-making, especially in the emergency department, and why we need solutions to address the problem of diagnostic errors in this context?

[00:02:00] Wolf Hautz: Well, thanks for having me. Yeah, that's a very good question. I think recently we saw a lot of attention going to the topic of diagnostic error in emergency medicine, which was in part fostered by the WHO's decision to make diagnostic error a focus or the theme of this year's Patient Safety Day. And I think that brought a lot of attention to the topic of misdiagnosis. And naturally, misdiagnosis is particularly frequent in fields where a lot of diagnoses are first made, which is the GP's office or the emergency department. And when I first started investigating the topic of misdiagnosis and diagnostic error, I was surprised to find how little we actually know about that and how frequent the issue actually is.

[00:03:02] Mathieu: And you know, are there like certain, let's say, to illustrate a bit the scope of the problem, can you give us some numbers on, I guess, you work in Switzerland, I don't know if you have some references like, let's say more globally or also just within the country, but just to illustrate what's the extent of the challenge?

[00:03:24] Wolf Hautz: Well, the New York Times picked up on one of the larger publications on diagnostic errors, which was published in December 23rd in the Journal of, the British Medical Journal. And in the headline The New York Times used was there's around half a million or more preventable harm or deaths annually in the United States from diagnostic error, and about half of that is attributable to emergency care. There's only very few publications from areas outside the United States, most notably Switzerland and Canada. And the numbers are pretty much the same. So one in about 10 or 1 in six patients somewhere in that range is misdiagnosed in the Ed. Now, of course, you can ask yourself, what does that mean to be misdiagnosed? It's not 1 in 10 that dies from that misdiagnosis. But it comes with inconveniences for the patients. You have to come back. Treatment isn't working. You come back three days later. Then we might figure out what's the actual problem you have. We might switch to better treatment. But as you know, for example, in infectious diseases, time really matters when we start appropriate treatment. And of course, that's the same for for many cardiac diseases as well.

[00:05:04] Mathieu: Understood. And so are there, you know, opportunities for digital decision support algorithms or some of these digital solutions to aid in the identification or the management of some of these conditions? One that we want to focus on in that episode is around heart failure. What do you think about those?

[00:05:25] Wolf Hautz: Well, that's a really tricky question. I would say yes and no. So if you look at what are the diseases that we misdiagnose it's actually the frequent conditions that are frequently misdiagnosed, in the field, we call them the big three, which is basically cardiovascular diseases such as stroke or myocardial infarction or venous thromboembolism, infectious diseases such as pneumonia or sepsis, and cancer. And those account for about 75% of the total burden of misdiagnosis globally. Now, the issue here is that, of course, the question of how much harm results from these misdiagnoses depends not only on the incidence of the diseases, but it also depends on the likelihood of the disease to be misdiagnosed and the likelihood of that misdiagnosis to actually cause harm. So in the field that you just mentioned cardiovascular diseases, take for example, myocardial infarction, which I think is a frequent cause of heart failure. In particular, if it's misdiagnosed, and maybe the thing we all fear to misdiagnose aortic dissection or aortic aneurysm. Now, aortic aneurysm has an annual incidence in the United States of about 100,000. Myocardial infarction, in turn, has an incident about 12 times as much. About 1.2 million in the United States annually. But the rate of misdiagnosis in the aortic diseases is about 1 in 3. Whereas in myocardial infarction, we have very well-developed tests, diagnostic tests, the ECG, all the biomarkers that we're familiar with. So we miss about 1.5% of myocardial infarctions on initial presentation. If you miss a myocardial infarction, the question or not harm results from that is about 50:50. Whereas if you if you miss an aortic dissection it's about 75% of those patients that actually suffer harm or death from that. And so, bottom line, although the incidences of these diseases are about a magnitude, an order of magnitude apart, the harm that results from misdiagnosing them is about the same.

[00:08:11] Wolf Hautz: Or put the other way. We are all afraid of missing an aortic dissection. But the harm we cause by missing only very few myocardial infarctions is about comparable. Just because it's a much more frequent disease. And now to get to your question of can decision-support tools help us or not? I think I would say there's there's two issues there. One is what's in the literature. And two is what is the studies we run. What is in the literature? I think there's the most recent umbrella review of reviews, which was published in 2023 in The Lancet Digital Health, and it looks at the at the effect of decision and computerized decision support in medicine in general. And their conclusion basically is there is a substantial increase in diagnostic quality and treatment adherence and all these things for tools that are disease-specific. So a tool that supports physicians and making correct diagnoses in pulmonary embolism, or as you mentioned, in heart insufficiency or heart failure there's actually good evidence that these tools support patient-relevant outcomes. That is less so with general-purpose decision-making tools. And there the second problem is most prominent: we really don't have that many randomized prospective controlled trials of these tools. But we use them in ways that we use diagnostic tests and that we use biomarkers and so forth. And, you know, the pipeline to validate these biomarkers is actually quite extensive and quite tedious. And I don't think we yet have something similar in the field of digital tools in medicine overall.

[00:10:18] Mathieu: Okay. So not only those that are specific to one condition but agnostic.

[00:10:24] Wolf Hautz: Yes. And that's a particular problem because obviously when the patient walks in your door in the ED, you don't know which of the disease-specific tools to use. And what you need is sort of a general-purpose tool first, to start with. And the evidence that those actually do good is currently very limited. And I would say it's basically non-existent. And maybe if we have the time we can go into into the reasons for that and why that is. For one, we don't do the research. We should. But second, I also think there's issues with sort of the cognitive psychology and the, the decision environment in which or the environment in which these decisions are made which are sort of different in, let's say, radiology or dermatology or looking at an ECG as opposed to the general ED population.

[00:11:28] Mathieu: Maybe you can double-click on that aspect. I think that comes to the question of what type of data and information you are leveraging for clinical decision-making in the ED versus other specialties. Is that a reason why in that setting it is more tricky to implement those solutions?

[00:11:51] Wolf Hautz: I think so, yes. So in cognitive psychology, there are basically two types of decisions. One is sort of the well-defined decision. So you want to buy a new electric car? Well, there's only so many options. All the variables that are relevant to you are known to you. Let's say price and range and these sorts of things. Space. And you can validate the values of all these variables. So if the seller the dealer tells you the range is 1000km, you can very well validate that by talking to friends, reading reports and just testing it out. And then what's the worst possible outcome? Well, your car will end up on eBay. Nobody dies, right? You just get rid of it. Whereas in decisions such as what's my patient's diagnosis? Well, there are 70,000 different diagnoses to make in the ICD 10, and the variables that are relevant to you are not prospectively known. It's different biomarkers when you are looking for this disease than when you're looking for that disease. And then in emergency medicine in particular you cannot validate frequently the value of those variables. So we look at ECGs that are kind of shaky, that are not very well fixed to the patient. The patient is shivering, you know, these sorts of things.

[00:13:28] Wolf Hautz: So the decisions become blurry, the decision environment becomes blurry. And the outcome space is vast compared to, and now if you look at those medical decisions that you refer to reading the ECG, there are only so many possible outcomes there. Right? There's only so many pathologies you could potentially identify. Same for reading a chest x-ray, and the structure of the data is also quite clear, and it's in its native format. It's accessible to a computer. The ECG is just voltages which are accessible to a computer, and the chest x-ray is really just grayscale, or it's converted to grayscale, but really it's radiation absorption that we're looking at. Also that is available to the computer. And so computers perform well in that area. Whereas most of the information we currently use to make clinical diagnoses and clinical decisions require physicians or other health professionals to make that information available to the computer, and we're not very good in doing that. We are biased, and I only provide the information to the computer that I think is relevant to the decision. And now that already biases my decision, and now the computer can't get any better than me.

[00:15:06] Mathieu: And this also contributes to the difficulty of evaluating these tools in that setting.

[00:15:11] Wolf Hautz: It absolutely does. It absolutely does. You're always evaluating the tool together with the way it's used. And to prove your point there, there's a huge difference in the effects that you see with these tools in what is called vignette studies. So I give you a piece of paper. All the information is already there. You don't need to collect it. And it's it's there at once. And now you use the tool. Well, what people do is they just transform most of that information into the tool. The tool performs very well. That's different in the clinical setting, right? Information is acquired sequentially. The question whether or not I do a troponin might depend on what's in the ECG, and might depend on what my physical exam findings are in the first place. So that information is likely not available in the first place. And it might not even be collected later on. And then we are frequently interrupted with informations that refer to other cases and so forth. So this vignette study environment is very different from the clinical environment. And in the clinical environment we're not only evaluating the the tool itself, we are evaluating the tool plus the interaction with the physician. But ultimately I think that's what matters to the patient.

[00:16:38] Mathieu: And you know, if we look at it from, let's say, a technology perspective, I think there is still some very interesting advancements happening that potentially in the future will benefit, you know, the environment as well. Are there some of those technologies or what are some of your your hopes in that regard? Or where do you think we might be in a few years from now in terms of how, you know, I guess there's the way of working and then there's the tool aspect.

[00:17:08] Wolf Hautz: Well I think I have some hope with respect to the way we change approaching tool development. I think about five years ago we developed tools, decision support tools in fields where data were easily available and the tools did what the IT people thought that frontline clinicians might need which is really making diagnoses. Now, a tool that makes a diagnosis in radiology threatens the radiologist. Now, I think we're turning more towards including what we know from cognitive psychology, from the research into human decision-making, into this tool development. And we're finding more and more that what people need in making better diagnoses is not necessarily a tool that makes the diagnosis for them. It's a tool that truly supports them in making a diagnosis. So let me give you an example. There is this patient that presents to the ED with shortage of breath. And currently I can impute the signs and symptoms I find into the tool. And the tool will come up with a list of possible differentials. I think what a future tool might look like is: the same patient comes, I tell the tool, well, there's this 50-year-old man with shortage of breath and I'm thinking of pneumonia, pulmonary embolism or maybe myocardial infarction, and then the tool could support me in figuring out which of the three it is by telling me, well, does the patient have fever? And by the way, also frequently associated with shortage of breath is maybe left heart failure. Right. And while pneumonia will be associated with fever, left heart failure will likely not be. And so forth. Right. And it could help me in selecting appropriate test strategies for the differentials that I am considering.

[00:19:30] Wolf Hautz: Because what we know is physicians are very good in considering the correct diagnoses. We're just not very good in ultimately selecting the correct diagnosis. And I think there's a lot of potential for these tools to help us better choose the correct tests that lead us to the diagnosis and sort of avoid tunnel vision on our gut feeling or preferred diagnosis. In the first few minutes, because that's what we know. Front line clinicians narrowed down their differentials and their considerations within minutes of the encounter. And once we've narrowed that down, all we do is collect information that supports our hypothesis rather than refuting it. And I think that's some that's the place where decision support tools could actually help us sort of broaden the view back a bit. Reducing confirmation bias. Reducing selection bias. There is a group in London at King's College, Olga Kostopoulou. She has plenty of studies where she shows that provided my initial hypothesis is A versus yours is B, and we both get the same additional piece of information. Let's say a neutral ECG. Neutral with respect to the likelihood of your or my diagnosis. We will both distort that new incoming information in favor of our own diagnosis. And you will end up saying, well, this ECG supports my diagnosis. I will say it supports mine. Whereas ultimately it really doesn't contribute anything. So humans are very good in maintaining cognitive coherence, right. It's very hard once we've made up our mind to convince us otherwise. And I think that's sort of a place where these tools could come in handy.

[00:21:40] Mathieu: Yeah. Makes sense. You know, we have a closing question. We ask every guest on the podcast. That's a bit of a fantasy question, but if you had a magic wand. Which digital solution would you like to have in your clinical practice that you think is currently missing, either related to what we discussed or something completely different?

[00:22:02] Wolf Hautz: I read an article with an artist recently and she said, well, I want AI to do my household chores so I can focus on the art. What I'm currently observing is AI trying to do the art, whereas I get to do the household chores. And I think that's true for medicine as well. I went into medicine to, to treat people, to diagnose people, to spend time with people. And we all currently spend a lot of time on the administrative burden on documentation and so forth. And that's where I really see potential for digital solutions, tools that listen to our conversation, that transcribe and summarize this conversation that, before my patient walks into that door, summarize for me what is in the electronic health record about this patient and so forth. Because we all accumulate those health data over, over our lifetime. And we hardly, it's hardly ever acknowledged by anybody. And if we had the tools that would summarize those data and would focus on on those information that might be relevant to your current encounter, that would be extremely helpful. I think it's less fancy, it's much less fancy, but I think it would improve medicine quite a lot if the tools would focus on getting, relieving us of administrative burdens.

[00:23:34] Mathieu: I think is very interesting because in the first episode we recorded with Prof. Michael Gibson from Harvard Medical School. That's also a point he brought up. He said his hope in, let's say, all the technological advances that we see is that we reestablish a degree of humanity in the patient-doctor relationship that kind of like, might have lost. And that's something that he sees as a very positive outcome for what's what's going to happen in the future.

[00:24:06] Wolf Hautz: I totally think so. And I think that the way we currently use large language models in medicine, I mean, we have all these fancy publications. The large ChatGPT can take the USMLE. Well, that's fantastic, right? But it doesn't help anybody. And I think ChatGPT is a language model. It's the obvious, right? And now the conversation between you and me is language, doing something to that language such as transcribing, summarizing, that's the nature of what these things are built for, and what they're good at. They're not very good at coming up with context, and relevant knowledge. And that's not what they're built for. And yes, that knowledge is ingrained within the language they are processing. But I think what these language models really do is they model language. They don't model knowledge. And so a lot of administrative burdens in healthcare refer to language. And so that's where I have a lot of hope.

[00:25:14] Mathieu: Thank you so much, Prof. Hautz.

[00:25:16] Wolf Hautz: Well, thank you.

[00:25:21] Mathieu: Thank you for listening to this episode of Cardio Insights. We hope you enjoyed it. If you want to dive further into other aspects of digital solutions in cardiovascular and emergency medicine, check out the other episodes of the series to support the show. You can share this episode with your surroundings and leave a five-star review on your preferred streaming platform. Cardio Insights is a podcast brought to you by Roche. Thank you for your time and see you in the next one.

[00:00:03] Mathieu: Welcome to Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine, featuring thought leaders from leading institutions across the globe. This series of conversations unveils the power behind clinical algorithms and decision support tools, and how they are changing the way we diagnose and manage cardiac conditions from their impact on clinical practice, their benefits for patient outcomes, and the challenges they pose to health systems in terms of implementation and regulation. This podcast is intended as a precious resource for healthcare professionals in cardiology and emergency medicine, who are curious to learn more about this exciting field. I'm your host, Mathieu Chaffard, and you are listening to Cardio Insights.

[00:00:51] Mathieu: So welcome, Prof. Brunner-La Rocca to this episode of Cardio Insights, the podcast where we explore how digital solutions are changing cardiology and emergency medicine. For our listeners who might be new to you, you are the director of the Heart Failure Clinic and Vice Chairman of the Department of Cardiology at Maastricht University in the Netherlands. Throughout your career, you have co-authored more than 400 peer-reviewed publications, and your work has been focused on using biomarkers and digital health technologies to individualize the therapy and the management of patients with chronic heart failure, notably. You notably led the TIME-CHF study, which is one of the key NT-proBNP guidance trials in heart failure and currently coordinate two large European projects. One is PASSION-HF, where you develop a physician avatar to enable patients with heart failure to treat themselves. And the other one, ICARE4CVD, where you develop precision medicine models based on AI using biomarkers in large cohorts, with the aim to prospectively validate these models. So in this episode, we want to explore with you the application of digital decision-making algorithms in the diagnosis, the triaging, and the management of coronary artery disease, and how these can enhance the accuracy with which we can detect these conditions, improve the outcomes for patients, and ultimately as well optimize the use of resources in emergency settings. So before we dive into all these aspects, can you share with us what characterizes coronary artery disease and how your clinical practice has evolved in the past years in that regard?

[00:02:33] Hans-Peter Brunner-La Rocca: Yeah. Well thank you. Well, coronary artery disease obviously is just well, a narrowing of one of the coronary arteries, which often it can be stable or unstable. But here we're now talking mainly related to well, chronic stable coronary artery disease. So not really an acute event but these patients often present themselves with with chest pain. Sometimes it's typical sometimes it's not. And the problem is that the prevalence in patients that actually see the GP, or also those that actually visit the emergency room. Really having chronic coronary artery disease is relatively low. And the diagnostic accuracy of current models is rather poor. So there is really a need to actually improve that, because we are seeing a lot of patients that may not really require specialist care. This, on one hand, is a problem related to the availability of specialized care but also costs that are created. And to be honest, in recent years, actually not that much has changed. Of course, there are now new guidelines related to the diagnosis of coronary artery disease, but it basically still remains the same. You have either the very low likelihood patients that actually can be dismissed, and you have those where it's basically almost clear, but according to the guidelines, this group is very small. So you have a large group of patients where you need additional noninvasive testing. And this is basically the same during the last ten, 20 years.

[00:04:23] Mathieu: So you think the reasons why we haven't seen so much progress is really two, due to the, you know, the pathophysiology of the condition or it's also more that certain technologies available today haven't made it yet in clinical practice for integration reasons or?

[00:04:41] Hans-Peter Brunner-La Rocca: Well, I'm mainly referring now to the diagnostic part.

[00:04:44] Hans-Peter Brunner-La Rocca: Of course, regarding treatment quite some new aspects obviously have been developed also regarding risk reduction. But I'm not referring to that. It's purely about the, well making the right diagnosis. And there I think that what we have available just purely based on symptoms risk factors is actually limited. And Then quite soon you need additional testing and this is basically the same. We also have to deal with a disease that is quite prevalent. So it's not really very seldom. And this actually makes that it's a quite burden, quite high burden to, well, healthcare. But obviously also to patients because well, if you think something is wrong with your heart.

[00:05:40] Mathieu: It's not very.

[00:05:41] Hans-Peter Brunner-La Rocca: Then it's not very nice, and patients really would like to know whether they need to be worried or whether not. So this really puts quite some burden. And in that regard, well, there is not really much progression, in recent years.

[00:05:59] Mathieu: Understood. And in terms of the treatment and the way you manage these patients. What has changed?

[00:06:05] Hans-Peter Brunner-La Rocca: Well of course, on one hand side risk reduction has become more aggressive. So really more studies really showing that with interventions, that for instance reducing cholesterol, you need to be very strict. But of course to do that you need to have a diagnosis. On the other hand side, it's also becoming clearer that stable coronary artery disease is relatively benign. So there is not really often the direct need to go to, well, angiography. But of course, both require the more accurate diagnosis with noninvasive testing. And this is one of the challenges that we face.

[00:06:52] Mathieu: And in terms of the treatment piece and the risk stratification piece, what are certain, what are some of the digital tools that you are using on a daily basis? Or that are really contributing?

[00:07:02] Hans-Peter Brunner-La Rocca: On a daily basis, is maybe not that much. I think generally speaking, and also according to the latest guidelines, it's really the assessment of a pre-test probability based on the type of symptoms, sex, and age, and then based on the pre-test probability, then either to do further non-invasive testing or exclude patients or to directly go to, well, invasive testing. But according to the new guidelines, you never get to that point that you have to do directly invasive testing. And in clinical practice, often patients are really in this intermediate group. So they undergo either CT scan for instance, stress echo, or MRI stress testing actually as further exams. And of course, this already requires some referral to to specialists.

[00:08:08] Mathieu: Yeah. And when we think about certain types of technologies in digital health or using artificial intelligence, do you think there is a space at some point for use in the diagnosis, in the management, in discharge decisions?

[00:08:23] Hans-Peter Brunner-La Rocca: Yeah, there is in fact a tool. And I'm well, just talking about that tomorrow in the presentation at ProCardio, which actually has been developed based on a cohort with some lab results, also some characteristics of patients. So in total, I think 22 variables to be able to predict the presence or absence of coronary artery disease quite accurately, without really the need for imaging. So that it can actually be done quite easily. Also in primary care or in patients that are being seen in an emergency room where acute coronary syndrome is excluded, to know quite quickly whether these patients really need further exam or whether this patient actually can be discharged. And basically to exclude the presence of coronary artery disease.

[00:09:25] Mathieu: Understood.

[00:09:26] Hans-Peter Brunner-La Rocca: So this is certainly an important step forward and has a big potential for, well, in this entire diagnostic process of coronary artery disease.

[00:09:38] Mathieu: And in terms of management, are there certain emerging technologies that you are seeing that have potential as well in the future?

[00:09:46] Hans-Peter Brunner-La Rocca: Well, I think this is certainly one it already is introduced. It's already available, of course, not very broadly, but of course, that needs to be done. Of course, similar kinds of tools are certainly possible to be developed. The problem with AI is, or with this kind of tools, that you need quite large data sets. And often when models are developed, they are not properly validated. So this makes it then very difficult to really apply to, well, in clinical practice. In addition, it may also be different depending on the cohort that you're investigating. It has been tested and validated in different cohorts that are completely independent of each other. Really reaching the same accuracy in the diagnosis of the same area under the curve. I think this really gives confidence that this is a tool that actually is also applicable in clinical practice. But this is one of what I see. One of the major challenges in using AI tools that a lot can be calculated.

[00:11:03] Mathieu: How it's calculated is the question.

[00:11:05] Hans-Peter Brunner-La Rocca: Well, not so much, because I mean, maybe a disadvantage of AI is that we do not fully understand maybe all variables that are included.

[00:11:16] Mathieu: Yeah, exactly. There's this explainability issue.

[00:11:19] Hans-Peter Brunner-La Rocca: Exactly. Because to be honest, we also don't really fully understand the entire biology.

[00:11:25] Mathieu: Yeah, yeah. For sure.

[00:11:26] Hans-Peter Brunner-La Rocca: And all the interactions between the different systems we have in our body. So this is a really an important strength of AI, but what it means is that we need sufficient validation to really know when it's applicable in which cohorts, and also to be sure that it's not just a random finding based on AI and that this remains a challenge. And often we see in publications about AI in the last paragraph: further studies are needed. And of course, we can keep writing that for the next 20 years, or we can start doing something about it. And well, there again validation is really key.

[00:12:10] Mathieu: And can you share with us - because I think in your given your research and your interest for that technology, you're very let's say open to that - how do you see, you know, potentially some of your other colleagues in terms of acceptance or perception, how do they see these tools in your area?

[00:12:27] Hans-Peter Brunner-La Rocca: Well explainability is certainly an issue. So of course, I mean, that's one approach that you really try to explain the findings or to link them to some, well, processes in biology or pathophysiological processes to make it more understandable. I mean, that's one issue or one possibility. What I see as a limitation there is that we exclude those pathways or those interactions that we don't know.

[00:13:03] Mathieu: Yeah.

[00:13:04] Hans-Peter Brunner-La Rocca: And I'm convinced that there are many that we don't really know, but they are still important. So to me, therefore I think really as mentioned, validation is key, also to do prospective testing. That's also what we do in well, just one of the projects that you mentioned, not to stop at finding a model, but really also to properly validate it, to test it prospectively, ideally also to use this as a new pathway and to compare this pathway actually with the standard of care to really show what the what is the advantage of these models. And I think there is really the potential that then also, well, colleagues start to believe in it because, I mean, if we are honest, quite a lot of things what we're doing in medicine is based on our experience on experimental findings. And in that way, I mean, if we have to prove that something works, then I think we're used to adapt it. But we are very critical on just some theories that are not properly proven. So I think this is really key to be convincing.

[00:14:23] Mathieu: Understood. And, you know, there's always a recurring question we ask every guest that, you know, we have the chance to interact with on the show. And that's the following. So, you know, if you imagine you had a magic wand, what type of digital solution would you like to have in your clinical practice that's not available yet? And that's, let's say out of the scope then that what we've already talked about?

[00:14:46] Hans-Peter Brunner-La Rocca: Yeah. Well, what I think we need, we had talked about diagnosis, but what I really also would like to know is who really needs interventions and what type of interventions, because in cardiovascular disease and medicine, we often apply a one-size-fits-all approach. So for instance, with hypertension, of course we know that we need to treat hypertension, but we need to know which drugs are really important for a patient or who really needs very low blood pressures, maybe even accepting some side effects. That's something we don't really know. And so this is really what I would like to have more information on the interaction of what we do. So treatment and the response to it. And to really be able to identify those patients that are in most need of particular treatment.

[00:15:48] Mathieu: Understood, thank you so much Prof. Brunner-La Rocca.

[00:15:50] Hans-Peter Brunner-La Rocca: My pleasure.

[00:15:55] Mathieu: Thank you for listening to this episode of Cardio Insights. We hope you enjoyed it. If you want to dive further into other aspects of digital solutions in cardiovascular and emergency medicine, check out the other episodes of the series. To support the show, you can share this episode with your surroundings and leave a five-star review on your preferred streaming platform. Cardio Insights is a podcast brought to you by Roche. Thank you for your time and see you in the next one.

[00:00:03] Mathieu: Welcome to Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine, featuring thought leaders from leading institutions across the globe. Thi series of conversations unveils the power behind clinical algorithms and decision support tools, and how they are changing the way we diagnose and manage cardiac conditions from their impact on clinical practice, their benefits for patient outcomes, and the challenges they pose to health systems in terms of implementation and regulation. This podcast is intended as a precious resource for healthcare professionals in cardiology and emergency medicine, who are curious to learn more about this exciting field. I'm your host, Mathieu Chaffard, and you are listening to Cardio Insights.

[00:00:51] Mathieu: So welcome, Prof. Giannitsis, to this episode of Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine. For our listeners who might be new to you, you are an Assistant Professor of Medicine at the University of Heidelberg in Germany, where you serve as a Senior Physician in the Cardiology, Angiology, and Neurology clinic. You are also an Attending Physician in the Chest Pain Unit and Research Group Leader for the Cardiac Biomarker Group, leading as well the Chest Pain Unit Initiative of the German Society of Cardiology and presiding as the Chair of the Certification Committee for Chest Pain Units. So throughout your career, you have co-authored more than 600 publications where you focus on cardiac biomarkers and their use in diagnosing coronary syndromes. You also specialize in cardiac magnetic resonance imaging, heading a research group that's interested in combining biomarker information with phenotypes from imaging. So in this episode of the series, we want to explore with you how digital algorithms can be used for the timely recognition and management of acute coronary syndromes in the emergency department and the impact of these algorithms on reducing treatment delays, improving risk stratification, and guiding appropriate interventions. 

So for our listeners who might not be experts on acute coronary syndromes, can you give us an overview of what these conditions represent and also of their causes?

[00:02:19] Evangelos Giannitsis: First of all, thank you very much for inviting me here. The acute coronary syndrome is worldwide a leading cause of hospital admissions. And in fact, many countries face enormous problems with the numbers of patients who are admitted to emergency departments. And only a fraction of these patients then in the end, will end up with an acute coronary syndrome. And this is our daily challenge in clinical practice to sort out who has an acute coronary syndrome, and then to find out what specific entity Is within this spectrum of acute coronary syndromes. And just to briefly highlight, for those who are not familiar with acute coronary syndromes. Acute coronary syndromes is the term that is comprehensive, including patients with St segment elevation myocardial infarction who are not the scope of biomarker testing. Here, the diagnosis is made by the EKG and the patients face a very rapid process of revascularization revascularization therapies. So our main focus is to identify those who have an unstable angina or a non-ST-segment elevation infarct. And the majority of these patients even won't have any EKG changes. So we only have to rely on the clinical presentation of the patients and the biomarkers. And for this purpose, the introduction of high-sensitivity troponin assays has helped a lot to improve the faster and more accurate diagnosis of myocardial injury and myocardial necrosis in the setting of an infarct and the introduction of fast protocols, namely the zero-one-hour protocol, or, as an alternative, the zero and two-hour protocol, or in case these protocols are not feasible or available, then the zero and three-hour protocol.

[00:04:32] Evangelos Giannitsis: These are all protocols that are based on high-sensitivity assays with validated troponin assays and pre-specified cutoffs and baseline concentrations. So if you employ these fast protocols, you can accelerate the diagnostic process. You can have faster disposition of patients. So rule out patients who do not have an infarct rule in patients with myocardial injury or an infarct. And then there's because these strategies are optimized to yield the most optimal result for both categories. So there's a category that's left over. That's the so-called observe zone. Yeah. And in this observe zone there's I think one of the greatest challenges that we have is to find out who is then lower risk or higher risk, who will at the end have an infarct, that, we missed by the serial testing if we limit the numbers of tests. So this is a very interesting but very challenging group. But this is our daily practice to get the diagnosis. And this in the environment of crowding, of overfilled, overcrowded emergency departments where we face a lot of entries and very often exit blocks.

[00:06:05] Mathieu: I see. And I was thinking in terms of you know, how these tools have changed the way or how these algorithms have changed the way or have changed your clinical practice? Or can you explain to us how it was maybe, you know, when you started your career and how things are now, and then maybe in a later stage we can look into, you know, what might be the future of the solutions in clinical practice?

[00:06:30] Evangelos Giannitsis: So far as as I mentioned, there are several accelerated protocols. Fast protocols that have been optimized to get categorization and classification of patients faster. And these protocols did not exist several years ago. So at that time, we relied on measurements based on the 99th percentile of a healthy reference population and some validated concentration changes that we deem to be relevant. Yeah. The introduction of the fast first protocols. Assumes that the troponin is released in a time-dependent way. So the longer you wait for the next blood draw, the higher will be an increasing troponin. So the concentrations that have been found to be relevant are time-dependent, and therefore, all fast protocols have different baseline and concentration change values. For example, with troponin T, the the baseline concentration for rule out is, for zero-hour rule-out is below the limit of detection. That is five nanograms per liter. If you use the zero and one-hour protocol, you should have a concentration, the initial concentration below 12, and the delta change below three nanograms per liter. And if you have a rule-in then your baseline concentration is above 52 nanograms per liter, then you are already ruled-in with the first blood draw, or you have a delta change of more than five, five equal or greater than five nanograms per liter. So that is ruled-in for the zero two-hour protocol. These values are different. And now we come to a problem: in the situation of crowding in the emergency departments, a fixed time point for the serial testing, let's say after one hour, an exact time point, exactly 60 minutes after the first blood draw is not feasible because nurses, the staff is busy, they have a break. This in general, we have a staff starvation in German EDs. The physician is busy, so you cannot achieve an exact timing for the blood draw and even the ten-minute interval plus -ten minutes that has been set and that is, is has been written in the, in the most recent ESC guidelines is not feasible. Ten minutes is very challenging. So and you cannot expect from the physician that he has all baseline concentrations and cutoffs for the delta change in his mind by heart. It's impossible to have every rule in your head. So usually hospitals decide to use one rule, one algorithm. Let's say the ESC zero-one-hour algorithm. And physicians are trained on that. But if now the rule cannot be achieved, then the interpretation is wrong. And here now comes this tool into play a clinical decision tool, as any clinical decision too, aims to help the physician with his decisions. And should I go now more into details of this tool? This tool is a so-called knowledge base tool that merely incorporates what has been provided by guidelines. And what is evidence-based on the established validated protocols. So the information that is contained in this CDS tool is for example, which protocol are you following? Are you following a US protocol with a US reference values or are you following a non-US protocol? What is the time of chest pain onset? The last episode of chest pain?

[00:11:04] Evangelos Giannitsis: Then what is the time and the concentration of the first troponin value? And then this tool already gives you an output and readout that say, according to the ESC guidelines, with the first blood draw and the presentation time of more than three hours since the chest pain onset, you all have troponin below LOD, so you have a rule-out. You can stop your triage process. The patient is ready for disposition, is either rule-out or rule-in, and you can make the decision to admit or discharge. If not, the protocol suggests, the tool suggests a second blood draw and also proposes the optimal time point for the next blood draw. If you can fulfill that, then you insert the time when is the the second blood draw and the protocol automatically reads out what is the result. The classification zone. If by chance you did not reach the goal of having 60 minutes or nearby 60 minutes, then the tool automatically reclassifies the patient according to the nearest available protocol. So for example, you are closer to this to two-hour sampling time, then the protocol will give you the result of the zero two-hour protocol. If you close it to the three hours, then it will give you adjudication based on the zero and three-hour protocol. And this ensures that you have the proper classification. You adjust for the time intervals. And this is something I think that's very important because it helps physicians to make a more accurate classification.

[00:13:06] Mathieu: Yeah. And in the era of artificial intelligence and digital health how do you see these tools in particular, and other tools evolving in the future when it comes to supporting these discharge decisions or respecting the guidelines based on the different thresholds and the evolving literature as well?

[00:13:27] Evangelos Giannitsis: So first of all, I already see a lot of chances from the CDS tools to improve our daily quality of healthcare. Why? Because the tools correct for misclassification. Thereby they allow reclassification. They also lead to a change in the prognostic assessment of patients. They change the risk group. They are transparent. The tool can show the physicians and the hospital how is their general performance regarding protocols protocol adherence and also guideline Appearance and may help to overcome some general problems. So already here we have a lot of improvement. And probably it's also more cost-effective because the tool avoids overdiagnosis. So means excessive blood draws that are costly and also avoids let's say under diagnosis because it alarms you when you need to have a second or third blood draw. So this reduces the risk of missing an infarct, by omitting unnecessary second or third blood draw. Now coming to the artificial intelligence or, let's say further development of this kind of tool. Of course, there is a very big chance for improvement. And artificial intelligence is in general a tool that helps us where we are not able to solve a problem by conventional biomarkers by conventional statistical analysis. So it's a very helpful tool with big data and with complicated scenarios, and especially for the setting of acute coronary syndrome in the ED, I see many chances. For example unmet needs for physicians is the the better risk stratification of a patient. So after you classify the patient, for example, as rule-out, you want to make the decision to admit or discharge this patient. Usually you do that by risk stratification that is based on Gestalt, on your gut feeling, or based on a clinical score, for example, the GRACE score, the TIMI score, the EDACS score, the Heart score. So there are many scores on board. But artificial intelligence could improve that risk assessment. So the tool combined with artificial intelligence could tell you yes, this is rule-out. The estimated probability for death or reinfarction within the next months is x percent. So you can give a rough estimation of the risk of the patient. So and the third is for example another unmet need is for example the estimation for the need of coronary angiography and revascularization. So such a tool would also give you information on the probability of having an obstructive coronary artery disease requiring revascularization. And then you can discharge the patient. But plan for a readmission for coronary angiography and maybe for an invasive strategy. And so far, the tools that are available to do this are very limited. So there are some tools, but their performance is let's say imperfect.

[00:17:13] Mathieu: Yeah. Understood. You know, there is always a recurring question we ask every guest on the podcast, and I think it goes in the same direction as we mentioned regarding the tool that you have developed and how you see it evolve. But the question is, if you had a magic wand or whatever, which digital solution would you like to have in your clinical practice that is not available today? Maybe something different than the tool that was just discussed?

[00:17:39] Evangelos Giannitsis: Well, first of all, I would like to have my tool because I think the implementation of such a tool is already a small challenge due to the interconnectivity due to IT problems. To have this tool is already a big achievement. And then if if one can have a very comprehensive tool that allows you, let's say also discrimination on clinical grounds, what is high probability for an ACS versus a noncardiac chest pain or solutions that can help you in the sequence on deciding on risk and management, optimal medical therapy, to check for guideline compliance regarding secondary prevention, the need for coronary angiography for revascularization. So this kind of comprehensive tool would be ideal.

[00:18:39] Mathieu: Yeah. Okay. Thank you, prof. Giannitsis.

[00:18:40] Evangelos Giannitsis: Thank you very much for having me here.

[00:18:46] Mathieu: Thank you for listening to this episode of Cardio Insights. We hope you enjoyed it. If you want to dive further into other aspects of digital solutions in cardiovascular and emergency medicine, check out the other episodes of the series to support the show. You can share this episode with your surroundings and leave a five-star review on your preferred streaming platform. Cardio Insights is a podcast brought to you by Roche. Thank you for your time and see you in the next one.

[00:00:03] Mathieu: Welcome to Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine, featuring thought leaders from leading institutions across the globe. This series of conversations unveils the power behind clinical algorithms and decision support tools, and how they are changing the way we diagnose and manage cardiac conditions from their impact on clinical practice, their benefits for patient outcomes, and the challenges they pose to health systems in terms of implementation and regulation. This podcast is intended as a precious resource for healthcare professionals in cardiology and emergency medicine, who are curious to learn more about this exciting field. I'm your host, Mathieu Chaffard, and you are listening to Cardio Insights.

[00:00:51] Mathieu: So welcome, Sara, to this new episode of Cardio Insights, the podcast where we explore how digital solutions are transforming cardiology and emergency medicine. For our listeners who might be new to you, you are a Senior User Experience Designer and Researcher at Roche with a background in design and engineering. Your work focuses on bringing the user perspective into product development to deliver solutions that are tailored to users' and customers' needs. You have a strong interest in designing products for the healthcare industry, as you've done in the past when you designed a smart system to monitor laboratory instruments in a centralized way. So through a single digital touchpoint and more recently at Roche, you've been working on the development of clinical algorithms, designing solutions that are safe and efficient for healthcare professionals. So in this episode, we want to explore and learn from you about the aspects that come into play when we think about integrating digital tools into existing health systems infrastructures and the challenges that can be that can be faced to make such solutions available and fit the current ways of working from healthcare professionals. So user experience, interoperability, data privacy, data safety, are a few of the themes that come to mind. And our aim with this episode is to give our listeners the basis to understand the implications that these digital tools pose from a technical perspective on health systems. So the first question that I would ask you is, you know, what are the different technical aspects that need to be considered - I mentioned a few in the introduction - when we think about integrating these new tools in a way into, into health systems and hospitals and existing care pathways?

[00:02:54] Sara Zimbardo: Thanks for your question and thanks for inviting me to this episode. So there are several technical aspects to consider when it comes to the integration of digital solutions in the hospital systems, and specifically in the emergency department systems. What I believe is one of the first things that is really needed to be done is really to map out all of the different processes that are within a specific clinical workflow, and to see which ones are on a digital basis or still paper-based. So it's really important to understand the level of technology, the technological maturity that the different healthcare systems have in order to provide a solution that can actually be easily implemented and easily used. We are learning through all of the research that we are doing in the scope of algorithms within Roche that the level of technology used in the different healthcare facilities can really vary from country to country, from kind of facilities to other ones. So for instance, there are the private ones that, compared to the public ones, they have a level of technology that is of course very different. So it's really important to understand the facilities that we want to work with, how available they are, how open they are to actually use some new products. And then of course, when it comes to clinical algorithms, especially in the emergency department, what is really crucial is the integration of data. And this is really a recurring topic in developing such kind of tools. And this means that all of the information, all of the data that are already in use in the hospital systems, specifically in the IT systems, can be reused, can be used also by the products that we are providing. And this is really key, this interoperability that you already mentioned in the introduction. So the ability of the system, IT system to exchange data and information. This is really important because it's the way that we have to guarantee that what we are providing can be really integrated in the current pathway, in the pathway that are already in place in hospitals, general practitioner's office, and of course, also in the emergency department. Yeah, the integration of data and the continuity in the data flow is crucial, but it's also quite something that represents a challenge to achieve because being able to speak, to open the data, the hospital data to an external products provider it's not so straightforward because, of course, there are many other topics, issues that arise in this case. So, the data privacy, data security, is a little bit challenging. And we are working on that for sure.

[00:06:02] Mathieu: Can you explain us some of the challenges that you face in that regard? So you mentioned that one part of your work is to assess how the, you know, the workflow is in the ways of working at a, you know, at an individual hospital level and see, you know, how your users are actually providing care and using or, you know, using some of the tools or let's say, prototypes that you guys are working on. The other aspect that you mentioned was that was this heterogeneity. So the differences between hospitals in terms of how they are structured from an IT perspective. So can you tell us what are the biggest challenges that you face with respect to these aspects that you've mentioned?

[00:06:46] Sara Zimbardo: Yeah, sure. So when it comes to the users, our end users. So in the emergency department, emergency nurses, and emergency physicians, I believe that one of the biggest challenges is how to drive adoption of new tools, because of course the healthcare industry in general is full of protocols and guidelines. So essentially of rules that physicians and nurses they have to, to follow. And when we provide a new solution that has to be integrated it's, it's very hard to provide a solution on a large scale that can fit most, a big portion of the market, because protocols and guidelines they change from region to region, from country to country. Like, for instance, in the US they have some guidelines that are on a national level. And then within the national level, they can have some changes on a regional level. And we are learning that they can even have some things changing based on a hospital level. So this is really a big, big, big bottleneck for us because if of course, it's easier to provide a solution, like for instance, for a pilot for a specific customer, but then when we want to provide a solution that can satisfy a big portion of the market needs when it comes to protocols and guidelines we can face some issues there.

[00:08:27] Sara Zimbardo: And, and also in regards to the users they are already used, of course, to work on a daily basis following some protocols, some pathways. They have their own clinical workflow. And we're learning through the research that we are doing that it's quite hard for them to change a little bit the way they work. And because especially in the emergency department, again, they're really overwhelmed already by a series of external inputs. And they have to interact already with so many different systems that are both software and hardware. So when we design new solutions, we have to be sure that we are providing something that is beneficial for the user. But at the same time, it's really integrated in a seamless way so that we are not going to change drastically the way they operate. And this is for sure one of the biggest challenges that we are facing.

[00:09:32] Mathieu: I understand and, you know, to double-click on the user experience piece how do you make sure you know that, that these tools are designed and used in a way that fits the workflows from HCPs? Because, to give you some context, when we, you know, we recorded a few of these episodes with healthcare professionals or professors of medicine and researchers who have a clinical practice. And that's probably the biggest topic they brought up, yeah, you know, they are all, I think, quite tech savvy and let's say more than average. And still they, you know, I think the main element that they brought up was to say, well, if we were to use such solutions, we need to, you know, they need to be super simple for us to use, and it doesn't need to require let's say a heavy learning curve, both, let's say from an implementation, but from a use case perspective. So how do you go about, you know, thinking about how these tools can be used in the best way for them?

[00:10:43] Sara Zimbardo: Yeah. Well, I think it's really key to design together with them. So as we start defining the requirements of the solution that we are delivering I think it's really important to immediately start from the very beginning to ask for feedback. The real end users and then as we come up with different different progress in the solution, always test together with our real target user. I think that's, this is for sure really crucial. And then once we have a solution, we first try such solution, maybe with a pilot site in a way that we can have some real hands-on feedback on the field. And then we can reiterate and make some changes where needed.

[00:11:43] Mathieu: I was also wondering, you know, given the diversity of countries and, and health systems overall, you know, I mean, you work with across many different geographies. So yeah, it's this complexity is part of your daily work. Are there, or how do you go about selecting some of the sites or those geographies where you can provide a certain type of solution? Because I understand that given the diversity of configurations and setups for hospitals, there might be certain that might be, let's say, more prone as well to, to test those solutions. How is this, yeah, how do you go about that?

[00:12:26] Sara Zimbardo: Yeah. Well, in this case we mainly follow the business needs and market needs. And also, as you said, exactly. There are some differences that we have to take into consideration when designing for a specific area. And so we tend to design for the areas that we already know quite well. So for instance, of course in the European market we already have quite a lot of products for which we already have many feedback and insights coming from our customers. And so this makes it also easier for us to define the requirements for some specific area. Then anyway, there are the specific guidelines and the specific standards for each region and country. Then whenever we decide to launch a product in a specific country, we will study the guidelines and the standards and we will ask for feedback, for instance, to the regulatory bodies that is responsible for each specific region. In a way, to be sure that our product is compliant With the standards that are in the different regions.

[00:13:40] Mathieu: Understood. I think this is also a topic that we discussed in, for example, the episode with Matthew Prime, who is an International Business Leader at Roche for clinical algorithms. It's this, you know, the topic of collaboration between academic and industry. I think the other aspect that we talked in other episodes with some of the key opinion leaders in cardiology in emergency medicine we got to meet through that podcast was the use of, let's say, machine learning and artificial intelligence for clinical algorithms, which in a way makes them, you know, a bit or more intelligent than what they used to be. It goes beyond just, let's say, a sequence of decisions that were programmed. You know, let's say in a more simplistic way. And I think it raises questions about how these solutions actually work, because, you know, you don't have access or you cannot understand easily the logic behind them. So you have to put more trust in the solution itself before you can use it. How do you see these advancements in the field of clinical algorithms? How do you also go about it, you know, within your work at Roche?

[00:14:59] Sara Zimbardo: Yeah, that's a very nice question. Generally speaking, I believe these clinical algorithms, they have a high potential and they can be really beneficial for physicians because generally speaking, again the aim with these tools is really to reduce the cognitive workload of physicians to save them some time on a daily basis and even if possible, to reduce some tasks that they have to do. And as you said the machine learning algorithms are also getting more and more on the market, and I'm working right now in one algorithm in the kidney stream that is based on a machine learning logic. And when we design our solution, we always have to keep in mind some risks that can occur during the interaction of the end user with our interface. And when it comes to machine learning algorithms, the biggest risk that also the regulatory bodies want us to focus on is really the so-called automation bias. So the physicians or generally speaking, people they tend to overtrust a solution or result when they know that it's based on a machine learning algorithm, because right now there is this tendency to trust more and more whatever is automated. But at the same time, the algorithms are also called clinical decision support tools. So they are just there to support the final decision of the physician. So it's very important when we design a solution to make it clear to the end user the fact that their advice is not going to be replaced by the algorithm result. Because, yeah, algorithms can really improve the clinical workflow, but they are not going to replace physicians' knowledge, experience advice. And in most of the cases, at least on the algorithms that I've been working on such tools are going to be complementary to several other sources of information. So the patient clinical history or some other test results, physician experience, of course. And so, yeah, this is another challenge that we are facing on a daily basis on the product development of the machine learning algorithms is how we can train and instruct the physicians, nurses, the end user in general, that they don't have to merely rely on the algorithms results, but just like consider the results that we are suggesting them. But of course, always keep in mind while taking the decision, additional information and sources of information.

[00:18:00] Mathieu: Understood. And it's interesting because I think it makes the link as well to what Professor Michael Gibson was saying in the episode we recorded with him, where he was already seeing the benefits it has in terms of freeing up mind space and making, in the end, the interaction between patients and doctors more human, because you can have already these days, you know, consultations where everything is recorded, there is much less work needed from the physician to make a report, write down the notes, or take the notes at the same time as the patient is speaking. And that's really something they already see now, and that they have a lot of hope for in the future. So we're coming towards the end of the episode, and we're really thankful for your for your time and your insights. The last question, which is a common question that we've asked as well the other guests on the show and in your case, it's probably a bit different because it was a question around clinical practice. But if you had a magic wand, what type of digital solution would you like to put in the hands of healthcare professionals that you work with? So you can be you can be a bit imaginative.

[00:19:14] Sara Zimbardo: Okay. Well if I had a magic wand for sure, I would give access to the different solution providers access to the data that are already in the healthcare system, because this is really exactly where we are struggling a little bit to have this communication between different systems. And so ideally, what I would like to have is only one tool to be used by physicians on a daily basis, where they can have access to all of the information about the patients, all of the test results, and also to a specific portal of algorithms that is already part of the data workflow without adding an additional touch point. And yeah. And it would be very cool if this, the software in this case could be also connected with the hardware. So automatically have all of the information coming from all of the sources. I think that would be really solving many, many problems, saving a lot of time, reducing the number of tasks, and make the interaction seamless. And that would be very cool.

[00:20:35] Mathieu: So hopefully we get there at some point with the centralized system for each hospital that also facilitates the work. Thank you so much Sarah.

[00:20:45] Sara Zimbardo: Thank you Mathieu. Thanks a lot.

[00:20:51] Speaker3: Thank you for listening to this episode of Cardio Insights. We hope you enjoyed it. If you want to dive further into other aspects of digital solutions in cardiovascular and emergency medicine, check out the other episodes of the series to support the show. You can share this episode with your surroundings and leave a five-star review on your preferred streaming platform. Cardio Insights is a podcast brought to you by Roche. Thank you for your time and see you in the next one.

[00:00:03] Mathieu: Welcome to Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine, featuring thought leaders from leading institutions across the globe. This series of conversations unveils the power behind clinical algorithms and decision support tools, and how they are changing the way we diagnose and manage cardiac conditions from their impact on clinical practice, their benefits for patient outcomes, and the challenges they pose to health systems in terms of implementation and regulation. This podcast is intended as a precious resource for healthcare professionals in cardiology and emergency medicine, who are curious to learn more about this exciting field. I'm your host, Mathieu Chaffard, and you are listening to Cardio Insights.

[00:00:51] Mathieu: So welcome, Paul, to this episode of Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine. For our listeners who might be new to new to you. You are an Access Evidence Leader here at Roche Diagnostics with a background in public health management and health policy science spanning academia and the industry. So in this role, you are leading the development of payers' relevant deliverables such as health economic models to support reimbursement requests and price negotiations for Roche Diagnostics and to ultimately advance patients' access to our solutions. So in this episode, we want to explore with you how implementing digital decision-making algorithms in cardiology and emergency medicine can lead to significant benefits for healthcare systems, including cost savings, resource optimization, and increased care quality. The goal for our listeners is to understand the role that evidence generation plays in that regard, the methods that are used to generate this evidence, and how it contributes to driving the adoption of these tools in clinical practice. So if we start with, let's say, the potential benefits that these digital decision-making algorithms can bring to hospitals and health systems, can you share with us to what extent these tools can be beneficial? And, you know, the areas where we can expect improvements if they were to be adopted and used?

[00:02:15] Paul Neveux: Yeah, sure. Thank you for the great question, Mathieu. I would say that for these algorithms two values that may be, let's say drivers. The first one is the workflow optimization meaning achieving a desired outcome for a patient. In a more efficient way, which means, for example, if you go through the editor would be achieving the accurate diagnosis with less unnecessary procedures, lab tests, or additional examination. It comes also to the point of the, let's say, healthcare professionals' time meaning optimizing the time that these healthcare professionals spend with patients and resources so that it can be dispatched to other patients or other situations apart from, let's say, the workflow optimization. There is also the the possibility offered by these algorithms to improve patient outcomes, meaning that patients who get an optimized course of action or management might be facing less adverse events or less complications based on this decision leading to cost savings. I would say today, one of the major examples we could mention is the use of triage algorithms, digital algorithms in the ED, for example. Yes. Emergency department. One of the let's say more common cases would be for patients presenting at the emergency department with chest pain. And for that, there are a few algorithms available. What they do is that they will, based on the data parameters of the patients, assess the risk for severe, let's say, disease or events with critical outcomes. And based on that, we'll derive the level of acuity of these patients and the level of medical attention that could be needed for these patients. And interestingly, is that how it plays in practice? Usually, if you compare how these algorithms work compared to standard procedures, usually these algorithms will classify more patients as being with low acuity, meaning that they might need a lower level of attention and lower resources, so that based on this algorithm, you can shift the medical attention and the medical resources also to patients with the higher needs. How does it translate from, let's say, the operational perspective going through the emergency department? It means that these patients might have a shorter length of stay through the emergency department, but also for patients needing more medical attention. The transition from the emergency department to the in-care inpatient units might be shorter, leading to a shorter intervention. Also, for these patients and ultimately potential cost savings. Because these patients are treated earlier and the turnover in the emergency department helps to manage more patients over the same period or amount of time.

[00:05:28] Mathieu: Understood. So there is, let's say at an individual hospital level, there are time savings for healthcare professionals in terms of, you know, efficiency of their daily optimization, optimizing their daily tasks. There is the optimization of what we say or what we call care utilization in a way. So like hospital bed capacity, which ultimately has an impact on costs. If we look at, let's say at a higher level, considering like a health system, like, you know, a country or multiple countries what are the benefits that, you know, based on your experience and the work you've done that you see also at a higher level. Kind of like, put differently, if we kind of sum up all these individual benefits that we see in hospitals, what does it change at the health system level?

[00:06:23] Paul Neveux: Yeah, a great question. I think what we just touched upon is exactly at the level of, let's say, a single hospital or a healthcare facility. I think the first aspect would be if you, let's say spread or develop or let's say implement these kinds of algorithms across your healthcare system, you might see this benefit multiplied by the number of, let's say, facilities, let's say widespread. But the second one is and something very important is that these algorithms, digital algorithms, they can also provide medical care standardization. And you avoid, let's say medical practice variations across your healthcare system, leading to a very consistent quality of care, which can be very important at the, let's say, at the size of a healthcare system. A third one that is a bit more, let's say forward-looking, is that implementing these digital algorithms provide some insights for these patients. Brought up together you can derive also, let's say insights at the size of a population. And you can forecast, let's say future events or derive also interventions at the size of the population based on these insights that you've collected through the algorithms that can impact, let's say, the whole health care system and the larger population.

[00:07:39] Mathieu: So yeah, the scale of the benefits is just getting bigger. What are the considerations to account for in demonstrating the benefits of these, that these digital solutions, you know, have at a hospital level or at a health system level toward the stakeholders within those systems? Meaning, you know, what are the types of methods that are used to generate this evidence? The types of studies? I think it's it's quite a fascinating field, and it's a complex field as well. So if you can take us through, let's say, the basics of that.

[00:08:12] Paul Neveux: Yeah, sure. I think first of all, what is very important is that when we work also on diagnostics, we have a sequence of events, which is, let's say, the decision making, the decision that has to be made by the healthcare professional, then the intervention and the outcomes of the intervention. And you have to be able to chain these three steps: the decision, the intervention, and the outcome, and show that for a new algorithm, for example, you have better outcomes than when the same decision is based on the standard of care or standard procedures. Usually, this evidence comes from the clinical trials. And let's say this is the prerequisite for a demonstration to your payers that you have to show a positive impact on clinical outcomes. Then when it comes to the type of evidence, it really depends on the audience you're trying to address. Because to different audiences, different requirements, be it in terms of costs that are covered, the population that is included in the intervention that is covered by your algorithm, or for example, the time horizon. What we see usually is that if you were to talk to a hospital CFO, that would be really interested into, let's say, efficiency of procedures in the hospital, like a workflow optimization study would be a very good fit to see how you have efficiency gains with your new solutions compared to the standard of care. And this can be translated, of course, in cost savings, ultimately. When you're talking, for example, if your audience is a national payer, they might be interested in more into, let's say, the impact of the intervention at a larger, let's say, at a larger magnitude.

[00:10:05] Paul Neveux: What we would be using usually in this case is, for example, a budget impact model that will help payers or the audience to understand can I afford to implement this new intervention? Meaning that you take into consideration the new intervention and the cost associated to that, and you try to demonstrate that your new intervention will lead to cost savings. Some other players might be interested into what we call cost-effectiveness models that are more long-term models, showing the long-term evolution of these patients. They usually rely on what we call a utility or let's say yeah, cost-effectiveness data and are much more targeted when the question for the payers or for the audience is, is this intervention good value for money? So you can see there are different types of evidence. There is not a one way forward. But it's more to be, let's say decided upon the needs or the requirements of the audience that you're trying to to address. One point I would like to bring is that, especially when it comes to diagnostic solutions, they always fit within a very particular healthcare organization. There are large variations in terms of medical practices and care organizations across geographies. So a step that is also very important when you work on the value demonstration, in the economic demonstrations of these solutions is actually to generate, to engage into pilot activities and generate real-world evidence to convince the local stakeholders of the value of the solution in this given environment.

[00:11:44] Mathieu: And to that point. So there is an aspect of generating local evidence to convince, let's say, stakeholders at a local level. And then there are these bigger assessments. You talked about budget impact models, cost-effectiveness simulations, or models. Obviously clinical evidence like clinical trials. So I guess the probably part of your work or your work is to figure out as well what is the correct mix that we need to have in order to support reimbursement decisions that, as we know, are kind of like the ultimate lever or leverage to unlock, let's say, access to a certain solution because I guess before there is any reimbursement for it, it's hard to drive any or a significant level of adoption, I would say.

[00:12:32] Paul Neveux: Yeah, yeah. Yeah, definitely. And what we do and what we're trying to take also in consideration working at the global level is the requirements and needs of affiliates, because they are the ones engaging into these activities. Exactly. So the way we try to develop our deliverables is to make them adaptable and try to hand them, let's say, early on to affiliates so that they have time also to tailor to their needs and populate in the appropriate way for the local value demonstration.

[00:13:02] Mathieu: I see you know, there's a there's a question we ask kind of like a fantasy question we ask at the end of each episode. We've recorded so far. If you had a magic wand, what would be one digital health solution you would like to see in health systems or hospitals based on your experience, and your interest in the field?

[00:13:27] Paul Neveux: Great question. I mean, I think what I'm really interested in at the moment is the concept of digital twin. Yeah, really excited about this. Very shortly, let's say digital queen, a digital twin is basically creating a simulation of, for example, a patient or an organ that mimics very, very precisely, let's say the patient or the organ, it's derived, it's based on information from the medical history of the patient or for example, imaging. And it's also informed in, let's say, in real-time by using sensors on your patients. What's super interesting with this, with this digital twin is that it has several applications, because once you have defined the parameters of your patient, you can compare it and match it against evidence. And let's say, for example, for a patient, it could be deciding or forecasting the, let's say the course of a disease and deciding potentially on early intervention to prevent a risk or early treatment. It could be also from, let's say, the research perspective, using this digital twin to explore the effect of potential treatment or interventions and decide on the most appropriate for your patient going into the direction of the personalized medicine. There is also, I think, an aspect that can be about the healthcare resource use and planification is that you're able to forecast maybe demands based on these insights and generate insights on workflow. And for example, how to forecast your, let's say, healthcare resources. Given a situation, for example, epidemics or let's say specific patient profiles. So there are many, many applications that are that could be possible with the digital twin. It's still I mean, there are challenges, of course, because it requires a lot of computational power to be able to create this, this digital twin. There are also, of course, considerations about like ethics and privacy. But yeah, I think it could be really a game changer to gain access into widespread digital twins.

[00:15:51] Mathieu: And I think it loops back as well. Nicely on the topic we discussed today with evidence. Because there's probably also a world in the future where we are able also to generate, you know, evidence out of synthetic data. I think we're not really there yet, but maybe that's something for the future.

[00:16:09] Paul Neveux: Yeah, yeah. You touch on a very good point. I think we could definitely accelerate time to market. Also, using this synthetic data and being able to generate hypotheses and confirm them much faster. Yeah.

[00:16:21] Mathieu: Perfect. Thank you so much, Paul. I think it's a fascinating field. I think it gives, you know, a good overview of the type of evidence, the things that you work with and, you know, the criticality of it to unlock reimbursement and drive adoption of these new tools. And yeah, thank you so much.

[00:16:41] Paul Neveux: Thanks, Mathieu for having me.

[00:16:47] Speaker3: Thank you for listening to this episode of Cardio Insights. We hope you enjoyed it. If you want to dive further into other aspects of digital solutions in cardiovascular and emergency medicine, check out the other episodes of the series to support the show. You can share this episode with your surroundings and leave a five-star review on your preferred streaming platform. Cardio Insights is a podcast brought to you by Roche. Thank you for your time and see you in the next one.

[00:00:03] Mathieu: Welcome to Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine, featuring thought leaders from leading institutions across the globe. This series of conversations unveils the power behind clinical algorithms and decision support tools, and how they are changing the way we diagnose and manage cardiac conditions from their impact on clinical practice, their benefits for patient outcomes, and the challenges they pose to health systems in terms of implementation and regulation. This podcast is intended as a precious resource for healthcare professionals in cardiology and emergency medicine, who are curious to learn more about this exciting field. I'm your host, Mathieu Chaffard, and you are listening to Cardio Insights. So welcome to this episode of Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine. For our listeners who might be new to you, you are a Director of Health Policy and External Affairs at Roche Diagnostics, with a background in digital health spanning philanthropy, academia, and the industry. So in this role, you are leading data and digital policy topics, liaising with external stakeholders to advocate for efficient data, AI and digital legislation, and policies to advance ultimately, patient access to our digital health solutions. In this episode, we want to explore with you the political and ethical implications around digital algorithms and clinical decision support tools in healthcare, but also the challenges and the potential pitfalls they bring. The goal for our listeners is to understand the role of policy and legislation in driving the adoption of these tools in clinical practice and the mechanisms at play. So as a first question for you to set, let's say the tone for the topic we're going to discuss, can you explain to us in general terms what's the role of policy, health systems, and governments in driving the digital transformation in healthcare?

[00:02:11] Afua: Thank you Mathieu. Thanks for having me. I guess this is quite a loaded question. In asking, what's the role? We're talking three separate entities here policy, health systems, and governments. I think working policies are oftentimes set by governments to ensure that there is perhaps they've identified a need or they see the need to pursue a certain strategy. And so then they go through the whole legislative process to put a policy in place to enhance that. Now, policies within health systems are often broad. They could be more at a public health level, or they could be more targeted if, for instance, there during the pandemic where a lot of resources were put to COVID-19. And so I guess maybe I could reframe the question to kind of say, what are the current health system challenges that our governments may find the need to to use policies to enhance so people, people are aging, so aging and getting older, which means that they have a higher likelihood of getting non-communicable communicable diseases, which includes heart disease. Exactly, exactly cardiovascular diseases and even the broad cardiometabolic set of diseases, which could include diabetes, among others. So this is also costing a lot of money then to take care of people. So health systems, the cost of care keeps going up. And so there are current discussions of finding new procurement measures, understanding ways to move care from hospitals to maybe a primary care setting or potentially the home to save money. And also, this is where a lot of digital tools are potentially being applied. Other challenges in the health systems, like we could go on and on, but this gives you a general sense of sort of the challenges the health systems are currently facing, and hence policies that are going along to to address these challenges. So I would say overall, to answer your question governments stipulate policies to address challenges that arise in health systems and these evolve over time. But I think currently some of the examples I just gave are the most prominent.

[00:04:35] Mathieu: Are there certain mechanisms that you know, are known at a policy level to address some of the needs that you mentioned? Can you give us some examples that you've seen from your work and that are, let's say standard in, in regulatory frameworks?

[00:04:55] Afua: Yeah, exactly. So I do access policy. It's not quite regulatory policy. There's a bit of a divide. But I do work closely with our regulatory policy colleagues. So yes, there are some processes on the normal sort of policy aspects where I work. If there is, let's say an issue, the government identifies an issue in the health system. For instance, several people are getting infectious diseases. Oftentimes they can mobilize a task force with certain public health experts. Which could be a broad range from epidemiologists to physicians. And then they could then try to analyze the issue. And then on the policy side, perhaps organize some roundtables with various experts. It could be academics as well, could be involved. And then try to understand the problem. Once they get a grasp of what that problem is, then they proceed with identifying different solutions. And so you have these big agencies like perhaps the CDC, the European CDC, CDC in the US, the European CDC, they're often tasked with those public health type challenges. And I think that is most of the ways that the government tends to operate. Of course, there are Department of Health or Ministry of Health agencies that also have people that do this. But I think it really depends on the problem that they're trying to tackle, whether a task force or a roundtable with experts, and how urgent, right. During Covid going through this process, it's not as efficient. And so you very quickly have to restrategize and say, how can we be as quick and efficient as possible? And so if it's a public health issue like non-communicable diseases, you can then be a bit slower, relatively to an epidemic where you have to be quicker with your decisions. So I guess the method that the government would use would depend on how urgent or pressing the health system need is.

[00:06:52] Mathieu: Understood. And I mentioned in the introduction that one focus, in general, for our conversation in this podcast series is digital health solutions. So I think it's, you know, it's an established term, but it's a field from a technological perspective that progresses very fast. So the latest advances we've seen in artificial intelligence, we touched on that in the episode with Doctor Matthew Prime from Roche, with Professor Michael Gibson from Harvard in another episode as well. And I guess given the pace of progress, technologically speaking, I would see that or foresee that you know, legislation is a bit like falling behind or trying to follow. And so I was wondering, you know, or be curious to hear your perspective on how, you know, governments, health systems are trying to, let's say, adapt their local frameworks to enable these technologies or drive clinical adoption?

[00:07:55] Afua: Yeah, very interesting question because I think if you follow this sort of narrative at all, you would know that the EU is perhaps the most prolific in coming up with legislations around this space. Right? So everything from the European health data space, which is more at the data level, trying to ensure that there's a single, Europe has more use for its data in terms of primary use and secondary use, all the way to the AI act, which is categorising AI according to risk. So there are some banned uses of AI in the EU. And then there's high-risk, low-risk, healthcare happens to fall into, most of the healthcare applications happen to fall into the high-risk category. So I mean, these legislations are being spurred by several I think, observations within society to recognize that AI is really here to stay. It's making decisions for us. Actually, one of my favorite books is by Cathy O'Neil, which is called "Weapons of Maths Destruction". I particularly like the title, exactly. I can highly recommend you to read it. And it goes beyond healthcare to explain how AI is being used in education, in social justice, in law, in all aspects of society. And this this is actually an old book. And so I think the EU is leading the way. But even across the world, I think there has been an observation that AI is here to stay, and it has been a big part of the legislations that are being or perhaps recommendations.

[00:09:41] Afua: So the, the US has gone more on the route of recommendations. So the national NIST, the National Institute of Standards and Technology released an AI risk management framework playbook in 20 recently. And then the Biden administration also released the blueprint for an AI Bill of Rights, which set out a set of criteria for what should be highly recommended to be available and using AI systems. And so that's like in the sense of AI, of course, data drives AI. We need data to be able to build AI tools, right, and train AI algorithms. And so that's where the other data legislations are crucial. And so the European health data space, for instance, which is trying to get everyone on the continent that has data in scope to share the data and make data sets more comprehensive and available for training algorithms. As an example, I think is crucial to ensure that, you know, we can build digital health tools, whether they incorporate AI or not. But I think the fundamental need for data is crucial. And that's also why I'm a big advocate of data governance mechanisms, because these need to be in place to ensure that you can at least have a bare minimum structure to your data to ensure that you can eventually utilize that data down the line.

[00:11:13] Mathieu: Understood. One topic we've discussed, you know, in some of the other episodes of the series, somewhere around, let's say, the utility or usability of these solutions from, let's say, a clinical perspective. So how can we drive adoption from clinicians for these tools? What are the requirements from a user experience perspective? And we also touched on the topic of reimbursement, notably with Doctor Matthew Prime that I mentioned earlier. And this is obviously a policy topic because it depends on - from what I understand - it depends on the willingness of health systems to fund certain types of these solutions and which triggers, let's say, a broader access because it aligns incentives for notably health care professionals to use these tools. And so I would be curious to understand how you see, you know, when it comes to these digital algorithms and or decision support tools what's the current, let's say, situation from, you know, health systems and sort of like attitude towards these solutions in terms of reimbursement or in terms of funding those?

[00:12:29] Afua: That is a $1 million question, I think. I mean, at Roche, we are building some of these clinical decision-support tools. We have been working with various health systems to try to get them sort of adopted. And so we have first-hand experience of the challenges it takes to adopt these tools. We know that there is no broad understanding of what the concept of AI algorithm or just even the term algorithm is. Not every physician perceives it the same way. Exactly. Which goes a long way to determine their appetite to use it or not to use it. And so that's a bit of the adoption piece now. It costs a lot of resources, whether it's monetary or human resources, to actually get a CE mark or an FDA-approved device or digital health tool, which is often a basic requirement for you to even attempt to get funding or reimbursement in several markets. Right. And so for us, that's the underlying thing. We go through the process. But unfortunately this regulatory process is completely Dissociated from all the other aspects that will come along to enable it. Eventually adoption and reimbursement or funding. And I'm really emphasizing funding here because for some of the digital tools, reimbursement perhaps may not be the right way. It would have to tap into several budgets that sit within the health system. And so that's crucial here, I think. Also, considering how long it takes to get a reimbursement code, for the most part, funding opportunities and innovative and alternative or creative funding opportunities is the best route for most of these tools now. Once you have gotten the CE mark, you need to be able to prove that you're improving patient outcomes. You need to be able to prove that you are saving the health system, you're cost effective, so you're saving the health system some amount of money. And oftentimes you think of this process as it's traditionally a health technology assessment, right? I don't like to use the term health technology assessment because it gives the traditional meaning of what that term means. But what it actually should be is that we will call it value assessment framework. And what that should be is a framework that's fit for purpose for the digital solutions. So unfortunately, most of the current frameworks do not really capture the way that the digital solution will be built, which means that they're not fit for purpose to make those claims and even most of the frameworks, if it's an app or solution that's targeting patients, then it's really easy to make that patient outcome link, which is often the benchmark that's used to say, yes you are adding value to the system, right? Unfortunately, if you're making digital solutions that are targeting physicians, then the value assessment frameworks are usually not sufficient. Right. Which means that you're not even eligible for funding or reimbursement because you can't solidify those claims. Of course, this is aside from all the other resources that's required to run a trial, and the fact that RCTs are not even fit for purpose. So, you know, there's a whole host of problems that often comes with this, but just the fact that there are no value assessment frameworks that can capture the breadth and, and sort of depth of how AI or digital solutions that could be using AI and that are meant for physicians or public health individuals do not currently exist. It makes it extremely challenging to bring those types of solutions to the market.

[00:16:23] Mathieu: I see you know, we talked about a lot of different subjects that I'm keeping an eye on the, on the time. There is a recurring question we asked all the guests on the show. Maybe for you, it's a bit different because you're working on these policy topics. And the question was initially more tailored towards healthcare professionals or let's say, people who develop these solutions. But I'll still ask it to you. So it's if you had a magic wand, what type of digital solution would you like to see in healthcare? It touches on the data topic that you mentioned at the beginning.

[00:17:02] Afua: Right? That's a really good question. You know, I wouldn't say I'm completely naive to like the development side of things because I did do my PhD where I asked users what they would expect and created a checklist for app developers. And so perhaps I've dabbled in the user research. I wouldn't quite say fully, but a bit. Gosh, I think we still have very siloed healthcare systems, which makes it difficult to integrate digital solutions. And I think the data, as you, as you can tell I'm passionate about that needs to be interoperable. The data needs to be structured. The data needs to be of good quality. I think if we can have fit-for-purpose data governance mechanisms in place and also have a high emphasis on comprehensive data sets because one of the biggest challenges we have with algorithms is it's not often validated on the data that's used to validate it. It's not often applicable to the population that it's used in, which means that it's just not as effective. I mean, it's obvious I'm a black woman, and if anyone has heard me talk, they probably heard me say this. Most data sets do not include people like me. And this is also something that I'm passionate about very much recognizing that I'm passively being subjected to algorithms that might not fully capture, you know, the breadth or diversity that is required to ensure that they can operate effectively. And so I think you're right. If I had a magic wand, I would ensure that we have comprehensive data sets. Of course, they're diverse in all senses of the word, but they're also structured. And there's a really great data governance process in place to ensure that once they're building these AI algorithms, that they're really going to work or be able to sort of fit the claims that we have. I think within the regulatory policy space, generative AI is still a big struggle. How do we regulate this? But that's all, you know, it's happening now, but it's also like a build-up of everything else that's needed. But the fundamental thing, in my opinion, is data and getting it you know, structured and having good data governance structures in place, having comprehensive data, to ensure that anything else that's built on top of it will be good for the citizens and for patients. Overall.

[00:19:52] Mathieu: Yeah, would make the life of many easier.

[00:19:55] Afua: Absolutely. Yeah, exactly.

[00:19:57] Mathieu: Thank you so much Afua.

[00:19:58] Afua: You're welcome. It was a pleasure. Thanks for having me.

[00:20:05] Mathieu: Thank you for listening to this episode of Cardio Insights. We hope you enjoyed it. If you want to dive further into other aspects of digital solutions in cardiovascular and emergency medicine, check out the other episodes of the series to support the show. You can share this episode with your surroundings and leave a five-star review on your preferred streaming platform. Cardio Insights is a podcast brought to you by Roche. Thank you for your time and see you in the next one.

[00:00:03] Mathieu: Welcome to Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine, featuring thought leaders from leading institutions across the globe. This series of conversations unveils the power behind clinical algorithms and decision support tools, and how they are changing the way we diagnose and manage cardiac conditions from their impact on clinical practice, their benefits for patient outcomes, and the challenges they pose to health systems in terms of implementation and regulation. This podcast is intended as a precious resource for healthcare professionals in cardiology and emergency medicine, who are curious to learn more about this exciting field. I'm your host, Mathieu Chaffard, and you are listening to Cardio Insights.

[00:00:51] Mathieu: So welcome, Matt, to this episode of Cardio Insights, the podcast where we explore how digital solutions are transforming cardiovascular and emergency medicine. And for our listeners who might be new to you, you are the International Business Leader for Clinical Decision Solutions and Algorithms here at Roche Diagnostics, with a rich background in clinical practice, but also in the industry as a Founder and Executive. Your work focuses on developing, validating, and commercializing clinical decision support tools and algorithms that are then deployed in hospitals and medical institutions across the globe to support healthcare professionals in their daily work. So in this episode, we want to explore and learn from you about the future innovations and potential advancements in this field, and how these tools might evolve and integrate further into health systems globally. The goal is basically to give our listeners a flavor of what's to come in the future. So to kick us off, can you give us a definition of what a clinical algorithm is and what the recent technological, let's say, advances have changed in that regard?

[00:01:54] Matthew Prime: Sure. So thanks very much for the opportunity to kind of chat about this today. I think you've started with probably the hardest question, because I'm not sure anyone really can tell you what the definition is. I think the term clinical algorithm is very broad. And we've been using algorithms in clinical care for a long period of time in general. You can think of an algorithm like any form of calculation that helps in healthcare and sort of classical things. There would be calculators treatment guides, treatment flows, nomograms. Any other thing that helps you as a clinician to kind of make a decision. I think that the recent advancement and to be fair, it's not hugely recent, is artificial intelligence, for obvious reasons. And that we're kind of coming in healthcare is coming into artificial intelligence a bit late. I think in the research space we've been doing a lot of work there for a long time, but I think now in terms of actually how it gets implemented into clinical practice and use that, that's where we are now. And perhaps if I can give you a kind of example of that. So I'm going to give you a non-cardiac example, but it's an emergency care example.

[00:03:19] Matthew Prime: You know, some of the tricky decisions you need to make in the emergency department are really around assessing patients for risk of sepsis. And in general, the way we've done that before is with kind of clinical history, physical examination, lab tests, and trying to put all of that together. And very often sepsis is not really a defined thing. Right. It's an evolving clinical picture. And so where you have a more junior clinical staff, they very often miss this. And that can have quite significant consequences for a patient and quite frankly, for the hospital system. So tools that are now emerging that can help predict sepsis can make a huge difference in clinical care. I mean, if you treat someone early with sepsis, you save their life. If you leave them, they get, you know, moved to a kind of distant part of the hospital. Then, quite frankly, that's why we have such a high mortality of sepsis. So I think the opportunity is really about artificial intelligence and machine learning models and how that gets implemented into care.

[00:04:33] Mathieu: And what do you think the impact of these new types of technologies, so AI-based algorithms or solutions that are put in the hands of practitioners, like what do you think will this change in their daily work and in terms of impact on patients and health systems?

[00:04:55] Matthew Prime: Yeah. I mean, I think hopefully the goal of all of this, right, is to improve patient care and to make health systems and hospitals more effective and efficient. If you think about just the macroeconomic picture at the moment, and you think about the pressure that health systems are under, we have aging populations with patients with more comorbidities, hospitals don't have any more money, and so we need to find, you know, the way you can kind of tackle some of these problems is we need to find and identify disease earlier. And so therefore prevent people having very significant complications. And we need to discharge patients from hospital faster. And you know, that's giving them a safety net to go home, you know, to go home with an enhanced kind of monitoring in a kind of more virtual care setting. So I think those are some of the kind of areas where there are opportunities. And maybe as that pertains to cardiology and cardiac disease. So some of the biggest things that, hospitals have to deal with, there are very much around acute chest pain.

[00:06:07] Matthew Prime: It makes up a significant proportion of patients being seen in hospitals. And unsurprisingly, most people presenting with chest pain, they don't have, they're not having a myocardial infarction. Right. They've got indigestion or they've got a musculoskeletal condition. But again, for a junior staff member who's also seeing 6 or 7 other patients, that can be quite a tricky assessment. And if it's not a, you know, there are ECGs and blood tests like troponin that help you make that diagnosis, but it also can take quite a long time. And I think that, you know, now we're in a situation where we can deploy, again, machine learning-based models that can help identify those patients who are having a heart attack, and then they can be streamlined to care and we can safely rule out those patients who aren't having a myocardial infarction. Overall, that has a clinical benefit, but it also has an operational efficiency benefit that you're not overburdening your emergency departments. So I think that's where some of the opportunities can lie basically.

[00:07:14] Mathieu: Understood. So this is a bit like looking into the future. I think there are some already some of the solutions being implemented. You mentioned algorithms to triage chest pain. And there are other use cases in cardiology and beyond. What do you think are some of the challenges currently with the solutions we have at hand in driving adoption of these tools, or what's the perception from healthcare practitioners in that regard?

[00:07:40] Matthew Prime: So we're talking here about the implementation of this, right? So there are a couple of areas that are interesting here. So the first one is, is the kind of infrastructure and resources hospitals by and large or the health sector by and large, and this varies in different geographies. But it doesn't invest enough in healthcare, IT infrastructure, and technologies. So very often it's using technology that's out of date. And that means that when you come to deploy more advanced solutions, you just simply can't. And so that's a big hurdle that needs to be overcome. The second thing is, hospitals are operating at capacity and in in many cases over capacity. And where you, if you bring a new idea into any system, you need to have space for that change to take place. And that's really difficult to drop a new idea into an emergency department where they're already busy, already managing very complex scenarios. So that kind of we need to create environments or places where we can create that change. And then the third thing, and the thing that drives everything ultimately, is we need sustainable funding and reimbursement. You're not going to take the risk to disrupt a busy clinical work environment with a new technology if you're also not getting any reimbursement for that activity. And so I would say that technical infrastructure, giving space for change for clinical teams, and reimbursement are the key things that will help drive change.

[00:09:28] Mathieu: If you can give us some perspective on where you think the most advanced health systems are in that regard. I guess through your work, you get to see what's happening in the US, in Asia, in Europe. Is there, are there some trends that you can share with us in terms of where the uptake and the incentives are most aligned with what you mentioned?

[00:09:50] Matthew Prime: Yeah. I don't think you can look too far from the US. I think it's a patchwork in the US. I think there are some health systems there that are unbelievably advanced from an IT health informatics perspective, and there are other places that are less advanced. But maybe if I just give you an example, I was in the US a couple of weeks ago. We were visiting a large academic medical center, and we were fortunate enough to, kind of have a tour of their virtual care center. In that virtual care center, and they cover about 15 hospital sites across the state that they operate in. And they, for the highest-risk patients, they are monitoring those patients in a virtual hospital. And a critical care nurse is monitoring between 12 and 15 patients. And a range of artificial intelligence algorithms are feeding information to this, to the critical care nurse, to help that person make a decision about care. It could be around initiating a sepsis bundle. It could be about preventing patient falls. There's a whole range of use cases that they have. But I mean, I practiced in the UK, in the NHS and when I saw that, I have to say I was a bit blown away because that's really going to the next level. I would say that I've also seen some incredible things happening in Asia Pacific. I think that people there feel much more comfortable with testing, trying new technologies. And there's certainly investment from the very highest levels to kind of do this, taiwan being a super example. But again, I don't think you can generalize at a country level. I think there are really good pockets in the UK, in Germany, in many countries. Finland also recently, I saw some fantastic stuff there. But as a health system where you have public health systems that are very financially stretched, they are struggling because they just simply don't have the money to invest.

[00:11:57] Mathieu: You mentioned that you visited one health system, and there are certainly many that you've seen through your work. What's your perspective on the role of collaborating between industry, academic centers, medical institutions in the development of these tools, but also in driving adoption and addressing the challenges that you mentioned earlier?

[00:12:22] Matthew Prime: Yeah, I mean, I think that perhaps unlike what we we've maybe seen with, with development of medicines, I think we need a completely different type of much more interlinked public-private, industry, health system partnership. And I think that happens across the development lifecycle. So let's talk about the early stages. Let's talk about the kind of ideation phase and the development phase. The best ideas will come from the people who are on the front line. You know, a kind of nurse running an emergency department knows what technology they need and what their teams can tolerate. And a cardiologist working every day in the clinic knows what technologies they need or what the opportunities could be. And that combination, so they know what their problems are, combining that with some of the kind of data scientists, quality regulatory people, innovation thinkers, that's a really rich area for kind of ideation and creativity. So I think that partnership needs to exist. I think when you're then developing a product, so you have a concept and you want to develop out a product, particularly when you think about algorithms, you need data. So we want to build solutions on appropriate good quality data coming from health systems. And that's an opportunity for partnership. But more than that and you'll know this here is that we try and embed a clinician right into the development team so they can sense check some of the stuff that's happening. And we support and enhance that with a network of clinicians and people that really work and do the job. So that's a little bit different to having a KOL network because actually we're not looking for the necessarily the kind of thought leader in our development space. It's way too in-depth and detailed. What we need is the person who's doing the job on the front line.

[00:14:25] Matthew Prime: So we're really trying to bring that in. And then, you know, you build if you're building a model, you build the model, you validate the model, you test the model on different data sets. But then you need to do real-world data, real-world studies. And so that's again a partnership where we want to go into the right sites that have that space for change that I was talking about before, where you can actually test this product in the actual real clinical environment. Because what makes sense on a laptop does not make sense necessarily in an emergency department. But we need places that are able to do that type of work and able to have the kind of resources to do it, but also understand the kind of needs around, you know, academic rigor. And then finally, when it comes to adoption again, it's fine for me to come along and go look at this fantastic machine learning-based algorithm. I work, you know, in the industry sector. I think where the power is around adoption is if you've really got something that's working and is a good solution, other clinicians explaining how this works in their setting that can really help build trust in this, trust in these algorithms. Right. And I think there's a really important, important part there. And then finally coming to the reimbursement space that's an engagement that needs to be made between industry and policymakers because without that funding infrastructure, a lot of this is a bit academic. And I think that that's also really important space where policymakers and even governments can start to say, okay, we want these tools. We see the opportunity, we need to make some investments here as well.

[00:16:12] Mathieu: Understood. Yeah. It's kind of like the sort of like last or the thing that comes last, but that has like one of the biggest impacts. And we have one episode with a colleague of ours, Afua van Haasteren, she's dealing with this topic. So for our listeners who want to dive more into that, I invite you to, to listen to that.

[00:16:29] Matthew Prime: I just loop back on that. What you said it comes last. And I think that's a really interesting thing. Ultimately, if it comes a bit sooner, it drives up the quality of development. Because if you if you're clear about, hey, this is a reimbursement pathway, but to get it, you need to have done all of these different things, then that helps the innovator. I'm not talking about big companies here. I mean, startups, they know, okay, I need to make sure I invest in proving the value of my product, not just in developing it and getting it approved by the regulatory authorities. But I need to show real-world data. I need to do a health economic model because that's what's required to unlock reimbursement. If that isn't there, then the temptation not to invest in those studies is huge because they don't need to. And this is complex, difficult stuff that they've not really done before. So I think it needs to come sooner. But that's my own personal biases.

[00:17:22] Mathieu: You know, we have a recurring question we ask every guest on the show, which is I mean, you're your physician by background. So if you had a magic wand, what digital solution would you like to have in clinical practice or like something out of imagination that might not be feasible today, but maybe in the future?

[00:17:44] Matthew Prime: I mean, I'll tell you one that I've seen recently that just blew my mind. I'm really excited about the opportunity around large language models. I'm not directly working, there's another team in Roche that's looking at that. I'm not directly looking at that, but I spent a considerable part of my life writing the notes from clinic appointments, writing discharge summaries in my earlier years. It's a lot of documentation we do. And the opportunity to have large language models do that documentation is just mind-blowing. Like the time that will save and to be honest, the quality that that will bring is just mind-blowing. So for me, that's the one I'm really excited about.

[00:18:33] Mathieu: I think it's it's one we've heard as well from Professor Michael Gibson from Harvard. And he was saying that the benefit of it is you change the relationship between the physician and the patient because you don't have the screen in between. And you just, you know, get humanity back into the way people are treated. And that's what excites a lot of the physicians.

[00:18:53] Matthew Prime: Yeah. I mean, classically, you're looking at your computer like what tests to order? What do I need to write here? The people must sit there and think, what is this about? They're not interested in me? And I think, yeah, to bring that humanity back, it's a lovely statement and I would agree.

[00:19:11] Mathieu: Thanks a lot, Matt.

[00:19:12] Matthew Prime: Great. Well, thanks for the opportunity. I hope that was interesting. And yeah, always great to chat.

[00:19:20] Mathieu: Thank you for listening to this episode of Cardio Insights. We hope you enjoyed it. If you want to dive further into other aspects of digital solutions in cardiovascular and emergency medicine, check out the other episodes of the series to support the show. You can share this episode with your surroundings and leave a five-star review on your preferred streaming platform. Cardio Insights is a podcast brought to you by Roche. Thank you for your time and see you in the next one.