Yes, perfect. Good morning. Thanks for everyone for joining us for our next session here, and we're pleased to have Dr. Yin Ho, CEO of Veradigm, along with Will Manidis?
Yep.
CEO and co-founder of ScienceIO. You know, today we're going to be talking about sort of the business outlook for Veradigm. Obviously, I know there's a lot of questions that people have regarding sort of the audit. Obviously, I think at this point, we know that at this point, we're still in a waiting situation, and
I'm afraid we can't talk about it.
Right.
Thanks for asking.
Okay, so instead, let's, you know, but there's still a lot going on at Veradigm, and I think that's probably important to kind of share with investors that, you know, having been through this process with other companies before, you know, I do think they, in time, get resolved. And there's still a business being operating underneath there, and there still could be opportunities for investors. So, maybe with that, maybe we'll start to help with an overview, just to remind folks, you know, sort of where the business is at, sort of to talk about sort of the business that you're running, and sort of, you know, sort of what you're seeing in terms of the growth outlooks for those segments.
Yeah, happy to. So first of all, I'd like to say that what's really. Thank you for including us today. And this is actually a really exciting moment in time for Veradigm, even though I know there's a lot of things going on around it. But we are actually the only company out there who actually cover the three end markets of the largest ecosystem in healthcare, which is the physicians and providers, the payers, and life sciences. And what I think is really exciting about that is that we have made a lot of moves to get to this point, and then we are a really strong company, having serviced all of these customers.
But more importantly, we understand where the market is going, and we understand where the solutions are going, and more importantly, we understand where it is possible to actually do a competitive leap. And so from our perspective, we're in a really great place. But again, we actually sell electronic health records to our providers and our physician practices, ambulatory, not in the hospital, but in the ambulatory setting where care is being given, probably predominantly in the United States. We do help a lot of payers, both national and regional, but mainly more regional, which actually have a lot to do with, like, helping with care gap closures.
For life sciences, we're just scratching the surface because we actually have been supporting them with not only real-world evidence analysis, but also data, and also additionally, safety type of studies, but we're about to go into a really exciting place with them. So, I think that we're sitting on a really wonderful business that has just poised to, like, take a big leap.
And if you think about those three segments, maybe if you could help characterize what you think the growth opportunities look like. Obviously, EHR, probably more mature, payer, more, a little bit more established, life science. But if you were to give sort of any type of ranges on what you think growth could look like in those segments, not that that would be right now, but what do you think people could, should, could expect, I guess?
Okay, I'm not going to answer your questions with numbers, but I will give you sort of a general heuristic perspective about this. EHR market is definitely a more mature market, but that does not mean that there is no room for innovation. In fact, I would say that the providers have been asking for quite some time for additional types of features and functions that are a little bit more driven around their preferences.
So if you think about what Veradigm has, is we not only hold the ability to do the electronic health record with them, we own the workflow, we also own the data, but more importantly, we actually have an opportunity to use technology to support their ability to be more, more engaged with their patients, and also be able to provide care in a more meaningful way. I think that that's something that we haven't talked about in this space in a long time. I think we've looked at EHRs for such a long time as a mature market, as just sort of share shifting, as opposed to recognizing there's a moment in time where you can actually support physicians' ability to handle their workflow under their control.
I think that that's one of the areas we can think of in terms of innovation in electronic health records. For payers, I think it really comes down to the fact that we're talking about value-based health. We're talking about contracts that are due to basically trying to depend heavily on accurate information around the care that is given, appropriate care that is given, and the outcomes that are expected. If you look at it that way, the providers and payers are actually starting to merge. The needs are starting to merge a little bit, and I think by recognizing that is also one of the important sort of growth opportunities in that direction. Then finally, with life sciences.
Life sciences is really, it's only been real-world evidence and real-world data has only been on the horizon for the better part of the last 8 years, and now we're sort of accelerating in a direction which I think is going to be more useful. If nothing else, there is a desire for more diverse information. The FDA has already been moving in that direction and wanting more diverse populations that are going to be represented in the data that is used to run analysis that support research, whether that's in life science products or in life science therapeutics, or in anything that has to do with just general care.
This is the highest place where everyone is going for, and if you think about it from our perspective, we're actually in a really good place to be able to do that, to support that, because our ambulatory settings, for our, physician practices are really quite diverse.
Yeah. Maybe before we kind of move on to, 'Cause it does seem like when you talk about innovation in the EHR, you know, I think ScienceIO can be a real enabler of that, and we'll definitely want to jump into that. But, you know, before getting there, you know, I think we've seen from other companies in the real-world evidence space, you know, any kind of companies providing commercial end services to the life science space, you know, has seen a bit of a slowdown. If people call that a softness, I think there's a lot of optimism that we may see that turn around.
Maybe, you know, obviously, also being part of that market, maybe give your perspective on what we saw last year in general, in the market, and maybe what you're seeing in terms of your conversations with life science, in terms of how they're looking at commercial spend.
I think really, I and I understand where that question is coming from. You're right, the last couple of years have been a little bit more of a slowdown in pharmaceutical spending, in particular, in areas of research. But that hasn't changed the fact that we are still producing, that the pharmaceutical companies are still developing drugs and are still developing the need to be able to understand where which populations will do better with certain therapeutics. So the need has not changed, the demand has not changed. Perhaps a slowdown has happened on the spending side, but I don't actually think that is a continual situation.
Actually, I would argue that right now, with the ability to actually know precisely what data you want, is actually the demand that hasn't necessarily been met, and that, I think, is where this gets exciting. Veradigm sits on the largest, and the richest clinical data set, that exists, around, patient care and patient, outcomes, that we are looking forward to using, this acquisition to help monetize, but more importantly, to actually, liberate this information in a way that supports the research enterprise. I think that there is never going to, be a slowdown in the demand for good data, and there's never gonna be a slowdown for the demand for data that's more precise and that will actually support pinpointing what, therapeutics work for whom and when.
Yeah, I would also just add, I think it's, it's a cost factor, right? If you think about a business like Flatiron Health, they're hiring 2,000 oncologists to sit in a room with a medical record on one screen and a spreadsheet on the other, and abstract the data by hand. There's just not that many disease modalities that are worth that drive that much cost to be worth that degree of abstraction. So you see pharma rotating out because of the huge amount of cost to prepare these data sets. If we can prepare much larger data sets at a much lower cost, there's much more demand on the pharma RWE and HEOR side because those data sets are a thing you can actually afford.
Yeah, and that's a great segue. Maybe, you know, Will, the announcement of acquiring ScienceIO, I think, caught a lot of people by surprise. Maybe just to help people understand, you know, what is ScienceIO? Like, what are you offering here? And then maybe, you know, you can talk about sort of the how this kind of came about and the need to bring it in-house.
Absolutely.
Sure. So ScienceIO is a large language model provider for healthcare. We started an LLM company in 2019, coming from Foundation Medicine, where we focused on data abstraction there. We were really focused on building the smallest possible language models, trained exclusively on healthcare data, to drive higher cost-value ratio for healthcare. So as you see, OpenAI and everyone else go big, we're going really efficient, really narrow models, trained on high-quality bio-debiased, audited data, focused on driving really narrow solutions across the value chain. We came to Veradigm because everyone has GPUs, no one has data on the scale that Veradigm has, right? Thinking about 200 million lives represented in a format that's ready for training is a really unique asset.
And maybe talk about what is the importance of being very focused on healthcare data versus training it on, you know, the big large language models? What is the difference it-
Yeah, I mean, if you, if you take every big model, look at OpenAI, look at Claude, look at all the massive models, they're trained on the same data, right? It's largely The Pile . It's a bunch of research papers. It's a bunch of social media data. It's data that you would not trust a physician to treat you with if they had only read that information, right? The vast majority of easily accessible information online is not healthcare info, right? I don't know why we would trust these big generalist models to do highly specialized tasks when they've never seen data in that distribution, right? Models are directly a reflection of the data they're trained on. Large language models, in particular, learn incredible bias from the data they're trained on.
Being able to have access to this many tokens that are proprietary, non-public, to train a model of this scale is really unique, and it also fixes the cost problem, right? OpenAI is a tremendously expensive model to run at scale. Our models are tiny and can be run on edge in the experience where the workflows already touch. So you're not looking at massive compute bills, you're looking at incredibly cheap and efficient models trained on high-quality data.
And maybe talk a little bit more about, you know, your decision to go with Veradigm. Was it really the data that key?
Yeah, it was purely the data. I mean, there is an abundance of capital and options available in the healthcare AI space right now. We could have gone to plenty of places that had hundreds of millions of GPUs, but there was only one place that had hundreds of millions of free text patient records that were ready to work with and ready to go. Like, that data and that diversity of data, not just kind of a certain cut of patients that are seen at academic medical centers, but a data set that is truly representative of American health and all its diversity, is an asset that only exists within Veradigm.
I think it's important to also recognize that by basically acquiring ScienceIO, we actually create our own proprietary large language model. And that's really important because a lot of our competitors are probably renting LLMs, and when you do that, you have no control over the data that goes in there, and you have no control over how it trains, and then over time, the output degrades. And what we have the ability to do, and this was one of the things that Will and I talked about very early on, was that we had a shared mission of figuring out how to do this in the appropriate way. More importantly, do responsible AI development in a manner that actually supports our healthcare customers, helps, helps supports patients, helps supports physicians, helps supports payers, and more importantly, supports research, because research is the thing that actually affects all of us.
And so maybe, again, like, how did we get to ScienceIO? Like, you know, what led you to. Obviously, you had a view sitting on the board. Maybe talk about sort of the strategic thinking and getting that we needed to have this added capability?
What's interesting about this is that this is actually was a very natural step for us. This wasn't something out of the blue. I have been on the board for a little over a year at this point, and when I joined the board, I was recruited to join the board because of a background in life sciences, but more importantly, in digital health. And the perspective was, we knew that there was something there with the data that existed inside of Veradigm, but we weren't 100% sure how we could actually access it.
And as we started to work through a strategy through the better part of this past year, in 2023, it became very clear that with the leap in generative AI and with the leap in large language models, that if should we bring that in-house, we would not only be supporting a life sciences monetization of care, oh, sorry, of data that we had, but also be able to support all our product development too. It would actually be a competitive leap, and it is from that direction that we recognized that we had something much larger.
We had a really great opportunity to actually affect all parts of care, from the patient care and the physician workflow to the payers and their ability to deal with value-based care and accurate charts and at the same time, be able to support research. It just became very obvious that this was the way to go, and that we needed to do this in-house, and that was the reason why we went forward with the acquisition.
Can you talk about some of the, we talked a little bit earlier, right? How it can be applied across various businesses. Maybe go a little bit deeper. Maybe give some examples of use cases of how you see ScienceIO capabilities being integrated and, you know, they adding more value to your offerings.
Yeah, let's, well, start.
I mean, if you think about the most obvious low-hanging fruit, it's our life sciences-based segment. If you think about what you are to a doctor, you are a very small amount of structured information, and then a massive clinical note and a stack of PDFs that's inside the report folder of the EMR, right? All of our data sales to date have largely been focused on the structured data that exists inside the record. That's less than 2% of the data that's relevant about a patient. Demographic factors, genomic reports, biomarkers, lab results, all of that data is things that are only accessible via high-quality extraction from the unstructured data.
So if you think about taking the vast majority of the data that Veradigm is sitting on, making it automatically usable with, like, literally no humans in the loop or extremely limited humans in the loop, to make that data accurate, accessible, and structured. And then selling that to life sciences to be able to access, patient factors or demographic data that would otherwise be inaccessible, that massively increases the premium on our data, increases our ability to scale that at a high margin, and actually build high-value data products that sell across the board.
So, even extracting from the PDFs as well?
PDFs, records, patient conversations, any piece of unstructured text that touches the patient can flow through the ScienceIO platform and become structured information.
Okay, so that makes, that's more. That seems very obvious from a life science standpoint. What about maybe the payer side? You know, is it similar there, or?
When you think about payers, you have to also think of providers at the same time. You actually. The two worlds sort of merge a little bit closer, or they're blurring a little bit. On the provider side, the physicians have been asking for AI tooling for quite some time at this point, and largely because there's a compression of time efficiencies in terms of gathering information and having it structured right away. And in that respect, that's something that only enhances, like, the data capture on the EHR. And if you look at it from a payer's point of view, the data capture being structured is also important, particularly if you're dealing with any kind of payments around value-based care or value-based contracting.
When you are looking at whether some outcomes are worth paying for or certain therapeutics are worth paying for, you are starting to need more and more accurate and structured information coming from the EHRs in order to be able to close those gaps. Because really, at the end of the day, what a payer is simply doing is they are basically figuring out if they are spending their dollars in an efficient way. And at the same time, if you look at a physician who's using an EHR, they are trying to figure out how to make documentation the most efficient thing that they can do, so that they can spend more time with their patients. So in this respect, the two actually converge.
No, that makes sense. Acquisition's done, you're starting to work together. How should we think of timelines in terms of how quickly can we get, you know, these, these capabilities integrated into the existing products?
I mean, we are actually starting work right away, but more importantly, this year we should be able to see some pretty exciting, high-margin data products.
I guess, maybe talk about sort of, particularly in life science, you know, your existing products today, when you sell, like, a real-world evidence study or you're selling data itself, like, how are those contracts typically structured? Is it project based? You know, is it a subscription-based model. Having the ScienceIO capabilities, could you start moving more to a recurring subscription-based model?
Absolutely. I think right now we have a little bit of a mix.
But I think that as we move forward, and especially in novel data products, you definitely go down a much more subscription orientation. What's more interesting about this is that not only are you looking for recurring revenue type of streams, but you are also, because of our cost structure, basically, structuring data, that's also now with a little bit of a lower human labor to it. So now you're talking about a really great efficiency on how you can put this together.
If you think about competing on cost, right? Not needing teams of thousands of physicians to label that data by hand. If you think about competing on speed, right? A customer being able to come to you, asking for a custom data cut or a custom element, being able to do that in basically real time. And then being able to compete on customer service because you're actually able to mold the datasets to customer preference, rather than pushing them onto an external platform that may or may not be able to perform the linkages they need, right? Today, largely, customers are consuming our data by cross-linking it with other common datasets. We think about building that value in-house by taking our unstructured data and massively increasing the scope of use cases that it's applicable for.
How long do you think it will take for particular life science to get used to that idea? 'Cause my understanding is right now they are doing a lot of that in-house, right? They're collecting all the different sources of data. They're seeing what all these people are offering in terms of analytics, and then they're kinda looking at all together at themselves and say, "All right, let's figure it out." You know, I was talking to one large big biopharma company, and right, I think they've hired 100 data scientists alone last year.
I mean, if you think about it, it's about raising the ambition of what's possible, right?
If you talk to an RWE head at a big pharma today, they're saying: "Well, look, we can buy a dataset, but we have 100 data science resources we're gonna put on that dataset over a year to make it useful. Have to figure out the data dictionary, have to figure out the schema, have to make sense of it." It is a massive ramp and a huge amount of services work to get to the top of that ramp. I used to be at Foundation Medicine. When we sold a dataset, 50% of that was literally just explaining to them what exactly they were buying. With large language models, it's fundamentally different, right? The ScienceIO platform can conform the data into the format that is ready for their analysis out of the box.
So instead of having this massive operational lift of, I buy the dataset, I have a huge amount of FTEs to make that dataset usable, and then I can consume it, it's I've bought the dataset, it's already loaded into my Snowflake instance. I can instantly visualize and make sense of that data purely with an end-to-end visualization platform that is, you know, commodity cloud tooling, rather than needing these massive OpEx stacks.
Okay, then that sounds pretty interesting. W hat do you think would be pushback from pharma clients? Like, where in your early conversations, what have they been asking for? 'Cause otherwise, I mean, it just seems so intuitive, like, why wouldn't they all just be signing up right now?
Well, we haven't been out selling yet, so that's why they're not signing up right away. But I can say that having been in this space for such a long time, having both been on the side of pharma buying data and also at the same side of selling data or products to pharma, I can say that pharmaceutical companies, biopharmaceutical companies, they want to know anything and everything about their drugs or about their therapeutics, and that's just basically how research kind of goes forward. And so, as a result, any information, any new information, that is possible to have a better understanding of how a drug's efficacy works over time, how its real-world effectiveness plays out in non-controlled circumstances, has become of greater and greater interest, and therefore, they're always looking for more and more information.
Plus, at the same time, we have regulatory authorities like the FDA, who are pushing harder and harder for there to be a greater and greater diversity representation inside of whatever analysis is going on. So I think in that respect, bringing information, bringing data, and bringing data that's consumable right away, to help with analysis that actually represents that level of diversity across a wider population, is going to be incredibly useful to the research efforts of the RWE teams, but also of the clinical research teams as well.
Yeah. I wanna touch on AI in general a little bit, right? Big topic, everyone's been talking about for the last, you know, couple of years, particularly with large language models. You know, one of the things that gets brought up a little bit, though, in healthcare, it's a little bit tricky, right? Because you're, you're talking about patient outcomes, patient care, and my understanding is large language models, and I'm not, you know, expert in this, it's not deterministic, right? You don't get a definitive answer, right? It's gonna be more a probable answer. Can you talk about maybe how you apply it such that, you know, life science or anyone who's using it can, can get to more of a definitive answer?
Yeah, absolutely. I would say a model is purely a reflection of the data that it's trained on, in addition to the reinforcement learning that is provided by its users, right? If you wanna go from generic model to great model, the two steps are first, get great data, and then get great humans in the loop that are an expert in their domain, giving feedback to that model quickly in rapid iteration cycles, to be able to prune out behavior that is unwanted. If you take what's happening in the rest of the industry with OpenAI, et cetera, these are monolithic models that are being tugged in a thousand different directions. These different expert users in different domains tug them in different directions, and the data is largely irrelevant to the task at hand.
What Veradigm has that's interesting, it's a huge amount of data in-house, you can train the model to begin with, but also, and perhaps more importantly, access to the workflow. So when providers and payers and life sciences companies are using our models, they're actually providing incredibly, incredibly valuable real-time feedback to push the models in the direction where they're more aligned. We think safety is the number one priority. There is very limited window for any kind of error here, and we, what we see with the rest of the industry is they're taking these massive models and encode huge amounts of racial and gender bias across dozens of factors, and just using them with very little safeguards, genuinely terrifying the degree to which these are being used. We're taking a fundamentally different approach.
Small models that we control tightly, monitor deeply, benchmark throughout the entire process, and tie the human feedback into the improvement process. It's our view, it's the only responsible way to deploy these models in healthcare, and I think five years from now, we'll be looking back and saying: "Wow, we wish we didn't allow all these big open models to be used at the scale that they've been used at.
I mean, it's healthcare at the end of the day, right? If you are a patient and a fake piece of data gets attached to you, how do you ever erase it on the way down? And if it propagates through research, how do you deal with that? The one problem with large language models that are a little less discerning about the inputs that go into it, is you have a higher likelihood of actually developing hallucinogenic data, and those are just basically fake data, which actually now we don't have. Our laws and our policies have not caught up with that. In fact, we're far ahead in that respect, and what happens is how do you sort of take care of an actual value, of a clinical value that's attached to a person?
How do you trace it back when it actually propagates its way through research? You augment the ability to move much quicker with generative AI models. But more importantly, if you're not careful, and you're not thinking through the ramifications at the individual level and at the research level, you will end up developing very irresponsible AI. And we would argue that what we're doing is actually, we're on the vanguard of responsible AI development in healthcare.
You know, a little off topic, but well, like, you're able to safeguard 'cause you're starting with small, a very narrow focus. If you were to advise a client that comes in, "Hey, we used this large language model and have all these issues," are you able to backtrace and help someone get back to the original source and eliminate those errors, or does someone have to start all over again?
It depends on the task you're performing. I would say with large language models, they are certainly explainable insofar as you can ask the models to explain themselves, and they will generate reasonably convincing explanations. But those explanations will not be the reason those outputs were produced, right? These models essentially are incredibly capable liars at explaining what happened. With smaller models, it's much easier to do essentially the full stack trace of finding where data causes error because you're just dealing with a much more contained stack, right? We have spent a huge amount of time on benchmarking, observability, and output monitoring just because we believe it's so important, and that starts by starting small. But also, as we grow our models over time, the same infrastructure we build to make small sub-billion parameter models observable scale with us to much, much larger models.
Where do you think healthcare is in terms of, even though healthcare talks about AI a lot, where is healthcare, do you think, in general, and maybe across the different segments, providers payers, and, pharma, in terms of real adoption, effective use?
I mean, this is the only healthcare break glass moment I've seen, right? What's interesting is that every other technology shift in healthcare required huge amounts of investment before it actually got to the provider, right? The cloud migration stuff pretty much universally made provider lives worse, right across industry. Providers are using ChatGPT, right? It is a tool they can install on their phone and use it in point of care. If you go on TikTok today and search provider ChatGPT, you will see physicians using generative AI in their workflow today. It is crazy. I have never seen provider demand at this scale for solutions.
I think while in general you should be skeptical of adoption curves of technology in healthcare, the one time you should believe that that's gonna beat expectation is when providers are grasping for the solution and using it in contexts where even their employer hasn't provided it to them. That's a huge change from what we've seen historically, and I think will lead to a dramatically faster adoption curve.
Do you envision that this will allow for premium pricing in terms of your products? I know providers, in particular have, r ight? Unless the government was giving a big subsidy, weren't willing to swap out new systems. But, you know, how do you envision providing these new capabilities? Is it to retain existing providers, or is it to say these enhancements?
I think what we'll end up doing is improving the provider's experience.
I think that's what's gonna matter the most, and then over time, you know. And if you think about it, it's a premium approach, maybe [that's] a way to think about it, in the sense that we're basically providing a user experience that more mirrors how their workflow and their preferences. And with that, I think what you will also generate is almost even better information, better data, and I think the feedback loop will make this incredibly useful, not only for the physician but also for their patients. And I think that's really where we're playing.
Wanted to see if there's any questions from the audience? All right.
Think about bringing the best [audio distortion].
Oh, okay. Well, I think, well, from our perspective is we're a company that continues to invest in our organization, and I think that if you think about these sets of capabilities and talent that we've brought into our house, we have the ability to use our own proprietary large language model in a way that will immediately create a competitive advantage for us and a differentiated set of products. So I think that that's a large part of where our focus is.
Any others? Maybe I'll just finish off. So, you know, if we think about the opportunity set in front of Veradigm now, looking aside, you know, obviously, the listing issues. But, you know, what do you think-- what do you see as the sort of biggest, you know, one or two challenges that you need to tackle right now to unlock sort of this opportunity in front of you?
I mean, it's like everything: How fast can we run, right? So I think that for us, from a challenge perspective, I think as an organization, we have a lot of things that we're investing and reinvesting in our company. I think that some of that takes time. I think also, at the same time, we are also trying to rapidly build out some differentiated data products, too. So I do think it's a combination of speed and both the slow of what and methodical, in some respects, on just reinvestment and the speed at which you can deploy a new investment. So I guess I didn't fully answer your question. I'm sorry.
No.
I would say the velocity is really the thing that is always the biggest challenge and also the biggest opportunity.
Okay, and I think with that, we're pretty much on time, and we'll stop there. Thank you very much, and we'll see you on the next panel.