Dr. Yair Saperstein, CEO of Avo, discusses how winning the American Heart Association's Health Tech Competition catapulted their AI-powered clinical decision support tool into the spotlight. Learn how Avo is revolutionizing cardiac care and beyond, leveraging their AHA recognition to drive innovation in healthcare technology.
Dr. Yair Saperstein, CEO, Avo
Megan Antonelli, CEO, HealthIMPACT
Intro 0:00: Welcome to Digital Health talks. Each week, we meet with the healthcare leaders making an immeasurable difference in equity, access and quality. Hear about what tech is worth investing in and what isn't. As we focus on the innovations that deliver. Join Megan Antonelli, Janae sharp and Shahid Shah for a weekly no BS deep dive on what's really making an impact in healthcare.
Megan Antonelli 0:31 : Hi, everybody. Welcome. This is Megan Antonelli with digital health talks, and I'm here today with Dr Yair Saperstein, CEO of Avo, and he is going to talk about how we met and the American Heart Association Health Tech competition, and tell us a lot about his founding story with avo and D and you know, we're excited to be here. Hi, here. To be here.
Yair Saperstein 0:56: Good. Thanks for having me on Megan.
Megan Antonelli 0:59: It's great to see you. It's so funny to think that that was, I guess it was, it almost, almost three years ago now, right
Yair Saperstein 1:07: sometime back, the excitement is still there. We're quite excited to be a part of it, quite excited to win, and then it's still excited to be able to work together. Yeah,
Megan Antonelli 1:19: it was funny. It was the first competition, and the first sessions back after the pandemic. And I have to say, I think it was, it must have been November, and it was really one of the first meetings that I felt like had a had a real energy to it. But let's start, let's start a little bit about your, you know, your founder story, how you started Avo, and and kind of the the trajectory,
Yair Saperstein 1:44: yeah. So when I was a resident in internal medicine in 2018 one of the biggest things that I wanted to do is to take clinical guidelines, whether those produced by the H, A, C, C and K, D, go and gold respectively, right or up to date, because I often use that as my reference tool and put it into the electronic health record like allow for the electronic health record with all the patient information to guide me on what I needed to know from the evidence for my patient in the right time, in the right place, In the right way, and the way that systems currently do things like this is with alerts. It's like, Hey, you're a bad doctor. Think about doing it this way instead. But the alerts are not really part of my workflow. It's like, Hey, you forgot to put on the flu vaccine. I was just trying to order some blood products for my patient that's bleeding out. Like, that's not the right timing for you to tell me about these things, right? So it's like, how can you put it in a way that will allow me to practice with higher quality, but at the same time allow me to really have a better workflow, reduce my burnout, increase my wellness, allow me to practice as a true physician, as a true clinician as I am. That was the dream back in 2018 the way that I was starting to dream about it didn't really become an actual product until I met my co founder, PJ. PJ was building this for about five years before I met him, and we met at the Columbia informatics program. And I met him, and he was showing me what he built. It was like, Dude, you're building my dream. We got to work together. Long story short, we started working together, reached out to the Columbia Business School to find an advisor who really can guide us on more of the business side right. The two of us are docs with an informatics background with an internal medicine background, so I ended up meeting our third co founder, Lawrence, who was previously a private equity investor, who then started to guide us more in this direction of how to actually turn it into a business. Three of us came together as co founders. When we came together, the original intent was to be able to take data from the electronic health record, take the evidence and be able to populate that into a way that can guide me at the point of care. The thing is that with the advent of chatgpt, and a lot of docs using chatgpt, a lot of clinicians, right? So it's, I always think in terms of docs, as a doc myself, but really we are used by RNs, MDS, dos, AP, it's pas dietitians and everyone in between. So, you know, clinicians often turn to chatgpt. I think I saw a recent study, 76% use chatgpt or llms as part of their decision making process. It's like whoa. What happened to safety checks? Right? And so if you're able to put registered guidance that is known, that is trusted into these flows to get instant answers that are informed by the electronic health record. That's even better. So we have a chat bot function. We have the ability to turn it into automated workflows, meaning, every time that I go into the electronic health record in the morning, as a hospitalist, I am always looking at a 24 hour look back period for my patient, and I always want. To compare all the consultants recommendations, the vitals, the labs, the meds, to the guidelines. To tell me, Hey, here's what you might want to address today. Like, let me have that automated workflow or discharge. Like, let me have an automated discharge workflow for my patient, right? These different different aspects. So that's the second the third is there's things that I do I have conversations with the patients. So that's a data element, like an AI scribe. But I don't want to use that alone to create a note. I want that to combine with my EHR data, combine combined with guideline data. I want to be able to produce orders and pen the orders. I want to be able to produce documentation. I want to be able to have next best step, sometimes within a hey, I'm asking for it, chatbot workflow, sometimes within an automated workflow, sometimes even within a prescriptive pathway. Allow me to go down an MD, calc style pathway to where I need to go as part of this prescriptive here's your next best step of what you actually need to do with automatically generated notes and orders to make me more efficient. And each different clinical type sometimes has these different aspects of the workflow that are more necessary, right from each of these different elements that we put together to the platform. And so this vision of, okay, how can I practice with more quality, and how can I practice in a way that will reduce my burnout? You know, allow me to really practice. Allow me to do medicine in a way that works for me. What's that North Star that really fueled our creation and still drives us to today?
Megan Antonelli 6:32 : Yeah, that's amazing. And first of all, it's amazing that in 2018 after people, I mean, to me, when, when we thought about the electronic health record, you know, and we're talking about it in 2000 678, you know that in my brain, now, not a physician, it was meant to have clinical information in it. You know that there would be clinical tools to drive it. And what's crazy is that here you are, 20 years later, and it still doesn't have that you know, that you could bring that innovation to it is kind of remarkable.
Yair Saperstein 7:10: The worst kept secret is that clinicians are still practicing with flow sheets and diagrams that are pasted like, literally, like taped next to their computers, that there is no aspect on whether the data is and at the same time, using chat, GPT without HIPAA security to be able to find information, right? And that discrepancy between what exists in the medical world and what exists in the general world to be able to come together is really what we're doing, right? It's like, let's bring medicine into the 21st century in a way that is safe, so that we can do the best that we can in giving clinical care.
Megan Antonelli 7:48: Right? The idea that a 20 year old electronic health record still doesn't have clinical decision support tools sort of built into it appropriately, is kind of crazy, because why is it there? Oh, but, because it was a billing system, I forgot. But, yes, but, but amazing. And, you know, I think, you know, one of the things, obviously, we met at the American Heart Association, and that on that health from that health tech competition. But you guys do a lot more than just cardiology. But tell us a little bit about, tell me a little bit about kind of that experience. And also, you know, you know how the product has evolved, and what, what you take with that?
Yair Saperstein 8:30: Yeah, so when we first presented, I think we were just at the point that we were pathways system right. So we've proven out now that our pathway system right, being able to go down step by step. Let's say a patient has hypertension. Did you try these medications? What was the impact? Here's how you can escalate. You know, here's the choices that you might want to make to get to here's the combo pelvic recommended. Here's your next steps that recommended in a prescriptive pathway, applicable for cardiology and applicable across broadly. You know, we partner with about 25 different content producing societies, and so we have this ability to really work across COPD with gold, to work across chronic kidney disease with K DECO, to work in behavioral health disorders, with the APA and so on and so forth, right? So that's where we were, the ability to take guidelines and turn it into these pathways, where the pathways can auto fill with data that's in the electronic health record and guide towards the right next step. It resonates a lot more as a product, right? The Pathways product with APCs, you know, pas dieticians, RNs, it did not really resonate as much with MDS, dos, because the thinking is, hey, I don't really want a prescriptive pathway when I'm doing medicine like allow me to do it differently in the you. A artist way that I like to practice, right? So what we found is that when there are RNs pas that are taking control of the system, health systems are able to reduce their staffing costs by over 70% and that's because they're able to put the decisions in a prescriptive way, into a trusted way into the hands of not as expensive providers. So that's the pathways part. And we really wanted to be able to serve all clinicians, right? That's why we were known as Avo. MD, we are now known as Avo, to be able to serve all clinicians, which we're doing. And so it's not just that we want to serve pas and RNs, which we're serving well, but have you served MDs and DOs also? So what we built is these other workflow tools, and we have a partnership with OpenAI. We have a BAA with them, and we have a zero retention agreement with them when that means that we're able to use their software. We don't have to build our own equivalent foundational models, we can use GPT four, oh, in a way, that tip is secure. Soc two, secure doesn't use any of the data to train their model. It's used only for our instance. And then we're able to run EHR data, guideline data together into workflows. What does that mean? I can say, hey, here's my patient. Suddenly, my patient now has chest pain. What's my next best step? What's the workup I need to do? Okay, cool. I confirmed that the you know, their patient needs an urgent calf like, put in the referral for me to cardiology, like, give me my dosing that I'm going to need for, Plavix, like anything you want, you can ask and be able to get in those answers. And that's the chat bot functionality. And you might want to do it that way, or you might want to have more of these automated workflows like, you know, I was starting to describe it for but in this cardiology flow, right? It's like a patient is coming into cardiology clinic with heart failure, and it's the heart failure clinic I know every time I'm going to be assessing against their medications and their labs and their symptoms, these guidelines. And here's the next steps. Pen my orders for me. Have it all done for me so that it's set. It makes me much more efficient. In this way, you're able, really, to have the suite of products across these different instances, so that it can serve all these different clinician types across these different disease states. And we have over 1200 guidelines that are in the system that are dynamically updated so that we make sure that we're in accord with the evidence as we're basically putting in these workflows into the hands of clinicians. So
Megan Antonelli 12:43: I have two questions. One is, you know, with that, with all the guidelines, did you go to the societies and, you know, how did you work with them? How did you get to that, that point where you were, you know, integrating that into the into the workflows, and using that data, and then also on the other end, where you're saying you have a no retention um agreement with open AI. How did that? How did those come to pass, both the the arrangements with the societies and then also with open AI. So
Yair Saperstein 13:15: we started with the arrangements with societies, and we did content licensing agreements, and that we first built out these pathways right within the first product. As we started playing with the early versions of GPT, when they became commercially available, we were like, This is the future. And so going back now years, you know, we started to integrate that into our software, first working with Azure, because Microsoft Azure was the first to allow us to do a BAA with them. OpenAI did not have that. And so we were using Azure in order to make sure that it would be HIPAA secure and zero retention and so on. And then at some point, open AI allowed us to go directly with them to be able to work in this way that was safe. We because of that agreement, we are able to get now open source documents from online. So anything that's been posted which is an open source document, we're able to read, and then we can read that even without a content licensing agreement, which allows us to expand much more rapidly the information that we can pull in, so as an example, right? If you want to be able to get information about and I don't even have a perfect example, but like some disease that we somehow don't have a licensing agreement already with the society, but they're open source articles that can be read, like those can be read, and they're, of course, sourced appropriately, so you can understand where it's coming from and go back to the original source to be able to read it as you wish. So it's really a combo between the two of them. And the way that the the way that the large language model works, is a concept that's called mixture of agents, which basically means that you have different instances of the large language. Model that are speaking to each other so they could figure out the right paths to go down. To pull. Should we pull from this guideline set? Should we pull from here? How should we construct our answers, constructing the answers and passing it on to additional large language model agents that are effectively working together, which allows for much higher accuracy, much lower hallucination rate, a much more precise answers, right? Because that's
Megan Antonelli 15:24: the other that was my next question, which is, when there's so many sources of information, how does the tool, you know, then you know, sort of know the the right one, if you will. Or do you know, how are you even testing that, or, you know, coming up with validity.
Yair Saperstein 15:42: So the first part is like, what's the structure of how it works? Then there's this multi step, right? So you make sure that you understand the intent of the question, or the automated workflows, and then you pass on the intent to, well, where's the appropriate place to look which are the appropriate guidelines to pull from? What does it say within those guidelines? How can we match the characteristics of the patient against the characteristics of the guidelines to come up with recommendations, whether it's diagnosis, treatment, dosing, etc, right? So multi step allows it to work better, and then we have, like an AI safety review effectively across our large set of questions that we have that we're able to run against to be able to check the answers and how well it's doing against those answers. Of course, it's impossible to check everything. That's why we always have a human in the loop, right? It's not like we're replacing doctors, replacing clinicians. We're allowing for more efficiency and better treatment. Actually, we found that we can improve adherence to guidelines on the range of 15 to 60% better than without using these tools, right? So it's definitely faster and more accurate, and you save cost, right? That's definitely the way that you want to go. Will we eventually be able to automate some of the processes? Maybe, right, but that's that mixture of well, there's rule based logic that you know, that you can trust, assuming you set it up correctly, Gen AI based logic that sets you up amazing drafts and amazing inferences that you could look at, but you probably need to source trace or use your own judgment as you continue to progress. Putting them together in the right way allows for efficiency with a human in the loop. Might be ways of setting up with more of the rule based logic and not as much of the Gen AI as you set up these processes, that more of it can be automated.
Megan Antonelli 17:32: And where are you seeing it have the most you know efficacy, or you know where it's working the best, whether that's in a specialty based or in a particular situation?
Yair Saperstein 17:46: Yeah, we have found a lot of traction among generalists across er internal medicine, pediatrics, inpatient and outpatient. We found it to resonate among nephrology, cardiology, respiratory and endocrine as our top four specialties. And that's just for now, right? It's like, when you find that it starts to resonate within a market, it spreads more within that market, like, I think the tool just as applicable beyond here, kind of like the up to date model, as it started in Nephrology and then expanded beyond there. And then we're finding that, again, the different products will resonate with different types. So the pathways are resonating much more with APCs, and then with systems that have APCs, and they're trying to reduce staffing costs with higher efficiency and better guideline adherence on prescriptions. So we're finding that to resonate a lot within that market, we're finding ask avo and the ability to use the chat bot and ask any questions to really resonate among trainees, because it's an education tool at the same time. And then it allows for systems to not have to deal with chat GPT, instead, they have a trusted system. And then we're finding some of these automated workflows to resonate more both in the non epic market, so Meditech Athena, which really allows for them to bring their EHR experience up to a higher level. And even within the epic market, you know, epic has some of these functionalities, but they charge an arm and a leg and then some for them. And here you're able to really achieve this functionality without paying millions of dollars for it by using avo that's integrated. So even within there, we're finding some traction within the epic market. Really, what we're finding is that there is an interest in the utilization management aspects. So like, can you be able to tell me, should I discharge this patient or not? Are they ready? Can you compare to the checklist? Can you tell me what they need, or an admission workflow? Can you tell me if they should be admitted in order to increase reimbursements, reduce 30 day readmission, reduce length of stay, right? Or it ties into these financial aspects that cannot be achieved in other ways. So this is where we're finding more resonance, within the epic market, when you apply the tools to these specific use cases, like Art Fair. Layer, right? It's like, that's a very high risk one. So how do you ensure that that's done correctly? You just plug in this workflow in those areas. But then as things are being plugged in, it's like, oh, wait a second, we would love to use your AI scribe also, because it's pretty similar to the AI scribe solutions that are out there, just six times cheaper. It's like, oh, we'd love to be right. It's like each of these different aspects that come in are like, Oh, they function very similarly, but now it's a singular platform that has crosstalk within it, across specialties, easy to install, and we could just flip a switch and have it on in the next product component, that would be great. So that's really where we're finding, like a land and expand,
Megan Antonelli 20:39: yeah. Well, it's, it's interesting. I mean, thinking back to, you know, the competition a couple years ago, and where you guys were, and it was also new. And, you know, chatgpt wasn't really, at least in the vernacular with everyone, you know, I think we all talked about AI then, but in terms of where things were going. And now it's, it's almost, you know, I mean, I wouldn't say ubiquitous, but there's certainly most health systems are making decisions about that now, and what you know, where they're going to pilot it, and what they're going to use. But let's go back a little. Let's take a walk back and and say, you know, in terms of that decision, that that you guys, you know, to apply to the American Heart Association, I think one of the things that did set you apart was, at the time, I believe you were using the AJ guidelines, which is one of the criteria of which, it's not a requirement, but it's certainly something that is is looked upon fondly by the judges at the competition. But tell us about that experience a little bit, and you know what, what that was like, and you know how it might have impacted, you know some of these changes and the direction that the organization took.
Yair Saperstein 21:52: So the judges wanted to see, how do you make sure that your health tech solution is actually evidence based? For us, it's easy, right? We're using the guidelines as our evidence source of what we're doing, and working in conjunction and a content licensing agreement with the HA for the AHA ACC guidelines like that is there's no, there's no better answer than that, right? For How To ensure that it's evidence based. For us, really the biggest thing that we got out of the out of winning, was more marketing, right, the ability to reach people. We were in the newsletter The next day, and suddenly everybody at the conference is talking about Avo, and everyone beforehand was not now we're at a level that people are talking about avo anyway, but at the time, it really helped launch. It Right? It helped launch where Eva was within the context of everybody that was coming to the conference, and that's a big deal for a startup. So we're working really, within what I like to call the club of HA, within the, you know, health tech umbrella, and able to have these conversations, able to work together on by putting the guidelines into a health tech format that can be within the actual workflow. And it's really aligning objectives, right? It's like, from the society standpoint, you want to be able to have the guidelines in the workflow. From the health tech standpoint, we want to be the vendor and the vector to make that happen. And from the health system and clinician perspective, it's like, Please help us. Like, allow us to do this too, right? So it's allowing everyone to do this together. And then the competition was really cementing what we had known, which is that this is something which is valuable and something which is helpful, not just in cardiology, but even beyond there. And that was one of the points which we made at the competition, which was, you know, this is something like one of the question was, how do you know that health systems are going to actually implement this? And it's like, well, health systems are looking for these solutions across not just cardiology, but other specialties and other disease states. So if you have a platform that can solve this for them across all of these different areas with a single install, so it's not going to clog up their IT side, suddenly you're serving their needs across all of these things with evidence infusion. And that's something that resonated. Well,
Megan Antonelli 24:15: yeah. I mean, I think that's so important. I mean, well, you know, certain technologies and certain companies have that have won the competition have been cardiac specific. It doesn't have to be righted. And if it has an application in cardiology, being there is, is certainly valuable, and it's, it's an interesting for me, and I think Did, did you guys go on to win the digital health, not tech awards too, like later that year, also at health, right? Yeah, yep, you know. So when you think about all the competitions that are out there, and of course, all the conferences that are out there, the being at a conference where it's predominantly clinicians or physicians is a very different. Experience from being at a health or a HIMSS, and, of course, now you guys go to all of those, but tell us a little bit about that, that distinction and kind of the value of, you know, being with the physicians.
Yair Saperstein 25:15: Yeah, I mean, it's, I put, like the AHA and ACC conferences together, where, right? It's like, if you're focused on cardiology and physicians and being able to reach the clinicians themselves, like that's different than trying to liaise with other health tech companies and others within a broader scope. And so that's that's the big difference between these specialty focused conferences and the like digital health focused conferences. I think they're both important to go to for us, right? But depending on the company, it really depends on what your goals are and what your value is. If you're trying to reach individual clinicians, you're trying to reach the physicians, like, of course, you need to go to the specialty conferences. If you're trying to hit cardiology, like, Aha, should be at the top of your list, right?
Megan Antonelli 26:02: But, and then, in terms of, like, even now or or then, has the decision maker that you try to get in front of changed at all? Or is it? I mean, I would imagine, at least coming from my perspective, and kind of seeing all those different events and the different audiences and even the different types of discussions that with physicians and clinicians, there's almost, you know, there's an enthusiasm. But if they're and and if they are bought in, you know, that's going to be a slam dunk when you go to the leadership, you know, oh well, you know, you have this physician who's already, you know, bought and sold on this. They're, they're excited about it, as opposed to the opposite. Is that true? Am I just imagining more value than
Yair Saperstein 26:47: I wish it were that easy. I think back in the day like it was that easy, where you have a ground swell of clinicians, and then they bring it up, and it's like, okay, let's do this now. It's not quite as easy, right? It's like, there's AI review boards, and there's, like, a lot of bureaucracy to go through, which is important, but it definitely slows down the process. And what I found is that sometimes, if you go to the top level and you go to the bottom level at the same time, that's the fastest way to do it. So you go to the CMO, the CIO, the CEO, and you go to the clinicians and then share it with each other, and then all of a sudden it comes in the middle. It's like, okay, let's put this in. You know, it's harder to just go to the practicing clinicians and say, Okay, you guys like this. Now push it, and then expect for that to happen by itself. So that's there's kind of a combo, but there's definitely importance in that, right? If you don't have a clinical champion, or clinical champions, it's not going to happen, right? It can come from the top down of, like, Okay, this needs to happen, and then no one's using it, and then it's gone, like, it has to be something that's useful, that people like,
Megan Antonelli 27:52: yeah. And what about, like, feedback, you know, in terms of, you know, in those early stages, as I, you know, sort of the founder, and figuring out what, you know, what is really needed. You know, I think sometimes we're always trying to find, you know, we've got a technology, and we're looking for the problem it's going to solve. You know, obviously, you went at it from a totally different way. You know, as a physician, you saw the problem and you were bringing that to it. But in terms of those perspectives, tell me a little bit about
Yair Saperstein 28:22: that. It's been like, it's been almost like a wave a couple different times as we've released these new products, right? So at first it was like, Okay, we have this product this. It was built as a product, not as a company like, this serves my need, like it's working. This pathway system is perfect. Okay, great. But then it's like, Wait a second. Chat GPT is out. I rely on chat GPT. Like, if I could have this in my EHR with guidelines, that's like, that's that's my answer. So it's like, product number two, well, how did this fit in, like, almost a new round of customer discovery and feedback, right? It's like, at each step as we release the products, feedback becomes, again, important as we release it. And so for the new products, we follow suit to be able to have this, you know, early adopters get initial feedback and then release it through that time to the initial institutions, and then get feedback there and then go, you know, at the larger scale marketing for the new products,
Megan Antonelli 29:26: yeah, yeah. I think it is, you know, I mean, it's, it's hard to get in into a healthcare system, right? It's just difficult. And so figuring out the the right fit, both from who you know, winning everyone over, but also that you know which problems you choose to solve and how broad you want to go, I think, you know, we talk a lot about, you know, fix the problem a problem for one small group and make it very valuable, and then you can pass that through, you know, scale it out, right? And so that that's. Um, you know what you guys have done, and have done a great job of it. So tell me what's next for Avo, what's you know? What's on the what's on the horizon? We've talked a little bit about it, but where, you know, where do you see this going,
Yair Saperstein 30:13: continuing to scale up and continuing to add in these use cases that tie directly to financial opportunities? So there's right the products themselves, and the resonance that they have within users and within the market is palpable. But as we continue to think about, well, how do we scale to institutions that have other products or other things in place that and have more than enough money that they're not worried about a price differential that it's like, how do we capture them, too? Right? It's like, well, if we have something that ties directly back to financial revenue, that is a different use case that will resonate, and what we found from our customer discoveries that will resonate within that market. So we're building out some of these finer tuned use cases using the same product as these automated workflows that will tie directly back to financial revenue among additional use cases. So that's that's next for the product standpoint. You know, as we continue to expand and scale, we're really moving across primary care hospitalists on the inpatient side, emergency medicine, peds as our main targets, and then expanding out to specialties, especially cardiology and nephrology, endocrine and respiratory as our top or not restricted there. So as we continue to capture additional specialties using the same product suite and the same tools, no additional expansion needed there. It's just marketing towards different areas will continue to capture additional clinicians and provide that usage for them. And then last but not least is we have some publications that were recently out, but additional studies that are ongoing with good data that's coming back from initial like initial waves of it, and so we're excited for some of those publications to be out some presentations at upcoming conferences, and the additional opportunities that come with that, of course, as we've now proven that our solution can actually help.
Megan Antonelli 32:16: Yeah, well, it's amazing when you're born of evidence, then the evidence just comes right? The value is there so well, yeah, thank you so much for joining us, and I'm excited. I assume you will be at American Heart Association scientific sessions.
Yair Saperstein 32:31: I'll see you there. Yeah, I
Megan Antonelli 32:32: will see you there. And it's always great to connect. And thank you everyone for listening. If you would like to get in touch with Dr Saperstein, please go to what's the website for for Avo.
Yair Saperstein 32:48: Avo, md.com, is the easiest way to get there.
Megan Antonelli 32:53: And also, I assume, are you on LinkedIn as well for folks to follow you? Absolutely, great. Well, thanks again. And thank you listeners. This is Megan Antonelli with digital health talk.
Thank you 33:06: Thank you for joining us for this week's health impacts digital health talk. Don't miss another podcast. Subscribe@digitalhealthtalks.com and to join us at our next face to face event, visit healthimpactlive.com you