Ep27: Amy Abernethy The Long Run with Luke Timmerman ยท 18 Sep 2018 Source: https://soundcloud.com/longrunpodcast/ep27-amy-abernethy Transcript: https://podcasts.opennash.com/shows/the-long-run/ep27-amy-abernethy-bef5ff/ (via OpenNash Podcasts, machine-generated) Welcome to the Long Run. This is a podcast for biotech adventurers. I'm your host, Luke Timmerman. Today's guest is Amy Abernethy. Amy is the Chief Medical and Chief Scientific Officer of Flatiron Health in New York City. She's an MD, PhD, and previously was a tenured professor at Duke University. She made a name for herself there with more than 400 peer-reviewed publications, and she honed her ideas there around using technology platforms to improve cancer clinical trials, outcomes research, and health policy decisions. She's definitely smart and a forward thinker. In the last few years, Amy has been working to translate some of her long-standing academic ideas into practical reality at Flatiron. The company makes electronic medical records for cancer physicians. It aggregates data from their practices, cleans it up so that it's clear and consistent enough to use in FDA new drug applications, and sells that quality data on things like how patients perform on certain medications, etc. to pharmaceutical companies. Flatiron was acquired by Roche in February for $2.1 billion. In a world where lots of healthcare software companies are all hat and no cattle, Flatiron is unusual. It doesn't just hoover up data and claim it's changing the world. It works hard to make sure that data has meaning for drug development. Now, I often find tech press releases to be so full of buzzword gobbledygook that I can barely intuit what a company does when I'm done reading. This conversation with Amy was a breath of fresh air because she was able to explain things in a way that I could understand. I think you'll agree. Before we get started, if you enjoy listening to these in-depth interviews, you'll love reading Timmerman Report, my subscription publication. You can subscribe to Timmerman Report for $149 a year per person and expect a couple in-depth articles per week. Discounted subscriptions are available for academic institutions and for corporate groups that obtain sharing licenses. For details, ask me at luke.timmerman at protonmail.com. Lastly, are you planning a conference, a team-building event, or a leadership retreat? I've developed a presentation based on my successful Mount Everest summit expedition. I'm sharing the experience over the coming year through a select number of corporate talks. These Everest talks feature gorgeous photos from the world's highest mountain, along with lessons on leadership, teamwork, and what it takes to overcome adversity to achieve the big goals. Interested? Ask me about an Everest talk at Luke.Timmerman at ProtonMail.com. Now, join me and Amy Abernethy for the long run. So with me today is Amy Abernethy. She's the chief medical officer and chief scientific officer at Flatiron Health in New York, a company that was recently acquired by Roche for $2.1 billion. A pretty big deal, so I'm eager to hear more about it. Thanks for joining me today on The Long Run, Amy. Thank you, Luke. I'm delighted to be here with you. So if you've heard this show, Amy, I think, you know, I like to start out with a bit on who the person is that we're talking to. And so I like to go way back to the origins. So can you tell me where were you born and raised? Well, so I was born in Houston, Texas, but I was really raised in Orlando. And it's kind of funny because most people don't imagine that there were people from Orlando. But indeed, I grew up just near Disney World, worked at NASA over in the Cape, and really have been from Orlando almost all my life. Okay, okay. So you did make a few trips there to the Magic Kingdom as a kid? Yes, I definitely did. Lots of jobs at the Magic Kingdom, starting off with scooping ice cream when I was 16. Well, good. That's a character building sort of thing. So how did you end up getting interested in science? Does this go back to like when you're really small? Well, you know, I got interested in science in two different ways. So I was particularly interested in math originally because my mom had written a textbook for nurses called Dosage Calculations. And when she was writing the textbook, I was helping her to edit it. And I was about eight years old. And from the time I was eight to maybe 12 or 13, I would help her edit it. And I was responsible for not only helping to edit, but to do all the math problems. And she would use me as a gauge to whether or not the math problems for the textbook were appropriate. it. And subsequent to that, I went to Duke for a summer program called TIP. I was a part of the second cohort ever to go to TIP as a part of their residential program. And while I was there, I took biology, I took computer science, and it was really the biology class that got me interested in science specifically. And I'll never forget that we did this project. You know, you got to imagine I'm like 12 years old, we do this project where you need to map the meaning of time by distance. And we were mapping it all across Durham and the Duke campus. And to this day, it was sort of a foundational experience that brought together the idea of some aspects of science, such as physics and distance and other aspects of science, such as biology and implications of how biology changes over time. And it was incredibly formative. That was really how I got interested. Huh, huh. And your mom was a nurse, you say. So obviously there's health care in the family. Did you decide early on you wanted to be a doctor or a scientist? That's often a fork in the road people address. Well, you know, I don't know if you'll remember this, but there was a show on TV around this period of time called Quincy, and I was convinced I was going to be Quincy the pathologist. You know, I was sort of fascinated by all the cases. I loved the discovery Quincy style. You know, I sort of loved the health and science in it. And so anyway, when I was a kid growing up, there was always this joke that I said I wanted to be Quincy. I ultimately went to university. I was at the University of Pennsylvania. And I was pretty convinced by, let's say, my junior year that I wanted to be a molecular biologist and had decided that I wanted to go and do a Ph.D. in a hardcore science, was planning to do a Ph.D. program in the Northeast subsequent to Penn when things didn't go the way they were. I was expecting them to go in my social life. And I said, oh, I need a new plan. And believe it or not, my next plan was medical school. So I hurriedly, this is now my senior year of college, decided to take the MCAT. And I ultimately decided to go to Duke as my medical school, really because there was a third year of research. And a requirement of being a med student at Duke was that you do a third year of research that at that time was mostly in basic science labs. And so I thought this would be a great way to pressure test whether or not I wanted to be a basic scientist. I realized I loved medicine much more than I expected while I was at Duke. And so ultimately became not only a physician, but a practicing oncologist. But really, for the longest period of time, I really thought I was going to be a bench scientist. And this was a way to kind of hedge your bets in a way. Exactly. And it was interesting because I then thought maybe I'll do an MD-PhD. This was in 1993 and decided I really liked healthcare enough that I wanted to go straight into it. And I didn't do my PhD until about 15 years later. Wow. So you get the MD, you're treating patients. What else? Well, so, you know, and I think this is going to be foundational as I think about what's happening here at Flatiron. So I do my MD as chief resident at Duke. I do my training in oncology. And just before I'm about ready to go on to faculty, my husband gets a job in Australia. And so now suddenly we have a opportunity to move to Australia. But what in the world am I going to do? And I end up doing a couple of things. So just around this period of time, I was becoming more interested in cancer pain. And it turned out there was a whole bunch of research and work going on in Australia and cancer pain. So I wrote to the National Cancer Institute and sort of wrote a rather radical proposal that I was going to be in Australia anyway. And if they would fund me to do training in Australia, I could learn all these things about cancer pain and bring that back to the U.S. if they would fund me to be a clinician and a researcher in Australia. I also was a part of Duke and very closely aligned with the Duke Clinical Research Institute. And at that time, the DCRI had large-scale clinical trials, including sites in Australia. These were the old Gusto trials. And so I thought, well, maybe I can learn clinical trials and clinical trials development while I'm in Australia. And finally, while I was there, I was very interested in this issue of how do you use technology to aggregate data sets and now use them in service for research and clinical care. And so I thought, well, you know, why don't I work on that while I'm in Australia? And so during this period of time, which was 1999 to 2003, I was there as a rogue U.S. physician in Australia thinking about these problems. And I ended up being responsible for one of the coordinated care trials in Australia, which was a large randomized trial of a health service delivery intervention. And wait a second. Now, the NCI agreed to fund your work while you were carrying it out in Australia with the understanding, of course, that science is international by its nature. You've got collaborators there at Duke. It's going to benefit people in the U.S. and everywhere. And that if we, you know, there were things that they were doing in Australia that I could learn and ultimately potentially bring back to the U.S. Exactly. Uh-huh. Uh-huh. So how long were you in Australia? You came back to Duke, and how did you set out to apply some of those learnings? Well, so let me kind of back up and give you one more detail. So while I'm there, as I mentioned, I was nestled up next to the Gusto trials. I was thinking about these issues of data and technology, and so somehow or another I ended up responsible for this one large clinical trial in Australia, which was one of the trials being funded by the Commonwealth. And so what we did was we sourced the data set for the trial from the data sets available from Medicare in Australia, from the hospital systems in Australia, from the insurers in Australia. And so what we did was we sourced the underlying data sets that then fed the clinical trial data set to understand whether or not the intervention made a difference. through the underlying technical infrastructure that already existed. And in fact, what I did my PhD in was in informatics and this idea that you could pull together currently available data and then add on a question about an intervention that gets delivered within the context of the healthcare delivery system. So when I come back in 2003, I really had that as now my core interest. Now, is there something special about Australia? they do have single-payer health care, don't they? And is there some kind of consistency in how they collect data and organize it? So it's interesting. So this is all night, say, 2000, 2001. So people were talking about electronic health records, but that was pretty nascent. Like the U.S., their Medicare system did have a large single repository of all the claims data. It is a single-payer system, but it's a single-payer system supplemented with insurance. And so all the insurers also had claim systems. The other things that were interesting was that the pharmacies were fairly connected to each other. And so there was a consistent system for pharmacy data. Home care was using the same kinds of reimbursement claim systems and basic medical data collection systems across each of the different home care organizations. So while they were all disparate, the thing that was unique about Australia was that each of those available data systems could be pulled together into essentially what in the U.S. we would call a single healthcare exchange or healthcare data exchange. And so we were able to take advantage of pulling them together into a single data exchange and then use that information for the clinical trial. Being a physician, physician scientist, I mean, you want to extract, not just pull together a bunch of data, but to learn something from it, to achieve some knowledge, presumably, advance the field or advance individual patient care. So were you able to do that or even just improve the way the clinical trials were conducted? So we were actually able to do both, which was what was so interesting. So we ended up conducting a factorial design, pragmatic randomized control trial. So a fairly complicated design where we studied three different interventions. One was a new way of having doctors and patients talk to each other. Another was a new way of educating doctors about how to take care of patients with cancer pain. And the third was a new way of educating families about how to take care of the patient. And we deployed those different interventions at scale across the state of South Australia. And then we used the data sets sourced from that health data exchange to now evaluate whether or not those interventions made a meaningful difference and identify what combination of interventions did the best to improve patients' cancer pain over time. And what's interesting about that story is that we were able to randomize. So we literally randomized patients to each of those different interventions in a two-by-two-by-two factorial way. We were able to source the data so that it was complete enough to evaluate the meaningful outcomes. and for data points that weren't naturally available in the systems, such as measurement of pain, we were able to supplement the data that was already in the health data exchange with information sourced directly from patients to measure outcomes. So then indeed the answer was yes. Now this is done retrospectively, right? The intervention was done prospectively. For the cancer pain, because usually cancer pain is not really the thing that is the main objective of the study. It might be a secondary question. Exactly. And so if I go back to the point I made before about asking the NCI if I could study cancer pain in Australia, they in Australia had prioritized cancer pain to the point where from the standpoint of a clinical trial, it was the primary question. Oh, okay. Okay. Well, and of course, you know, to patients, this is super important aspect of their quality of life, even if it's not always the primary endpoint that we're asking here in our trials in the U.S. Exactly. And, you know, and the other part of your question a few minutes ago was, did we, so you asked me two things. Did we learn something meaningful and did we learn something about how to do clinical trials better? So we learned something meaningful about better managing cancer pain. But I think probably even more impactful is that we learned something very meaningful about how to use currently existing data coupled with intentionally collected data to now basically make clinical trials run faster and more efficiently at scale. So indeed, we really learned something important. So there's a process learning here. Wow. So you come back to the U.S. What year was this? I come back to the U.S. in 2003. And I went back to Duke. Okay. And then what happens? Are you all excited about trying to do something kind of like this in either the Duke health system or, you know, elsewhere in the U.S.? Yeah. So I come back in 2003. And, you know, you can kind of imagine that everybody looks at me like I have three heads. What are you talking about? As you pointed out, there's not that many clinical trials in the United States around an issue such as cancer pain. And it was hard to look past this being about cancer pain, really. In my mind, it was about, you know, how did we do this whole new way of conducting clinical trials at scale? But for most of the people who were looking at what I was working on, it seemed like the predominant question was about better symptom control. So I come back to the U.S. And at that time, I was a faculty member on the leukemia lymphoma service. I subsequently take over the melanoma program at Duke, but that's not for a couple years longer. And while I'm doing leukemia and lymphoma as my day job, I type up a sticker. And the sticker that I type up says the Duke Cancer Care Research Program. And believe it or not, I stick the sticker on my door. I then go to the local Kinko's and make myself some business cards. And starting in 2004, I became the Duke Cancer Care Research Program, which was nothing more than a sticker on my door and a few business cards. You sound like a budding entrepreneur right then, Amy. So I now have a name. And, you know, we had a pretty remarkable success with this trial in Australia. I was just finishing my Ph.D. now in informatics. And a couple of things happened. One was that in the U.S., importantly, the way to build credibility, especially as a scientist, is to do work that's compelling enough and of an adequate research standard to be accepted by the NIH. So one part of my quest as the Duke Cancer Care Research Program, which at the time called DCCRP, was to now set out to basically build a clinical trials program where the trials were NIH funded. And indeed, that's what we did was we got a series of RO1s. Ultimately, I had a U, a U01, which is a big type of coordinating grant to conduct clinical trials. The second thing that I did as a part of this period of time was start to think about how do I train faculty members in thinking about not only answering the questions sitting in front of us in both cancer care, symptom research, and how do we take better care of patients predominantly in the clinic. I also started thinking about how do we train new faculty in informatics and data science. So I started building a faculty base. And then the third thing I did during this period of time was started to realize, and I think this is pretty important to the Flatiron story, that this is all well and good, but the infrastructure from a technical standpoint, when I say technical now, I mean software and data, the infrastructure wasn't there to do it at Duke, let alone at scale. If you enjoy listening to these interviews with biotech newsmakers, you'll love reading Timmerman Report, my subscription publication. You can subscribe to Timmerman Report for $149 a year per person and expect a couple in-depth articles per week. Discounts are available for academic institutions. For details, ask me at luke.timberman at protonmail.com. Is your company interested in raising its profile? Consider sponsoring the Long Run Podcast. I'm only allowing room for one or maybe two sponsors of this show over a year's time. If you work in a top-notch organization with something of value to offer the biotech industry, ask me about sponsorship opportunities at luke.timberman at protonmail.com. These years that you're talking, I mean, there's still a lot of old-fashioned pen and paper records, right? It's 100 % pen and paper records. You're 100 % right. I mean, we had laboratory data that sat in the lab system, but most of everything we did was on pen and paper. And so I did two things. One is I experimented with building new technical systems. In the beginning, I predominantly focused on patient reported data collection because that was what I knew best. And also because it didn't piss anybody off. Like people were perfectly happy for me to ask patients questions and collect data in the clinic. And I could use that as a learning laboratory. And then the other thing that I did was I went to Washington. And in Washington, I, you know, basically spent a lot of time on the road talking about the need for electronic health records, the need for digital infrastructure. that story ultimately translates to what becomes the High Tech Act. And the idea that there is a stimulus, you know, something that gooses the uptake of electronic health records in this country. And you can, you know, we can step back and say whether or not the EHR makes clinical life better or worse, whether or not the EHR is yet what we expect it to be. But there's no question that if you want to be able to do research at scale and improve the leading edge of taking care of patients, you have to have an efficient technical infrastructure to do that. And the electronic health record is by far the best way to do that. We ended up with all kinds of other problems like interoperability problems, proprietary formats, just confusing interfaces, I think. I mean, yeah, I've talked with doctors and they tend not to be too happy with a lot of their EHRs. Yeah, and I think, you know, ultimately that's a set of problems that we're still working through. And we'll probably continue to be working through for another five to ten years. but because of the availability of electronic health records that sets us up in a very different place to use data and technology to accelerate research and improve patient care and that really was the foundation of what we were doing in let's say 2005 to 2012. the last piece I'll mention is that circa 2011-2012, I, at that point, was really focused predominantly in cancer. And the question came to me, why is this just cancer-focused? Why shouldn't this be something that we're doing for the whole university or health system and we're thinking beyond cancer to cardiology and aged care and other areas? And so together with Rob Califf, who subsequently becomes the commissioner of the FDA, and Victor Zhao, who's currently the president of the National Academy of Medicine, we started something called the Center for Learning Healthcare at Duke. And the Center for Learning Healthcare was intended to be almost like an incubator or a discovery zone where we figured out how do we take the core principles that I've been working on for the prior decade, this idea of using data and technology to accelerate research, and make that happen at scale. And that was really the next major step before leaving Duke. Right, right. Right. So you got to first gather the data in electronic form, get it in a curated, some kind of reasonably coherent form before you can begin to learn from it. Exactly. Exactly. And, you know, what's interesting is that I was doing all of this within an academic institution. And a lot of people ask me, you know, why did I stay in academia for so long? And in fact, I would argue that the private or commercial side, even the venture side, was not ready for all of this until circa 2011, 2012, where we had enough distribution of electronic health records and technical infrastructure in the U.S. to be able to push it forward. I first saw that as a part of Athena Health. So I sit on the board of directors of Athena Health. And, you know, when I got to Athena, what was really interesting was that I now could see a national infrastructure where there was one cloud-based EHR that acted almost as a blanket across the U.S. and then each individual physician and healthcare practice was operating within this one common framework and the ability now to start to see and think about what does healthcare look like and how can you improve it in many different ways was one of the first times now I could see, all right, I think the private side or the commercial side is ready to start thinking about this. Okay. So you sort of dipped your toe in the water with industry there. you saw some interesting possibilities. I guess in retrospect, your Duke colleagues could have said, oh, the writing's on the wall there for Amy. But how did you end up getting introduced to these Flatiron guys? So the other couple of things that were happening about this period of time is that I had written a treatise in 2009 about this idea of what we call a rapid learning system for cancer. So using data and technology to continuously learn and improve cancer care delivery and research. That white paper ultimately became something with an oncology that's known as CancerLink. It got funded by the American Society of Clinical Oncology, and I was the founding chair of the advisory board for CancerLink. So it sort of acted as almost like executive chairman for CancerLink. and that then meant that I was really thinking about this space and pushing it forward and about the same time Rob Califf who as I mentioned I had been working with in the Center for Learning Healthcare and was one of my mentors he calls me and says you know I've got this phone call this afternoon with these two guys from Google who are thinking about building something around cancer data very similar to what you've been talking about inside a cancer link you know do you want to come chat with them with me. So I get on the phone with them. And, you know, frankly, I wasn't paying that much attention. And then they kept talking and I got more intrigued. And so during the end of the phone call, I said, well, you know, that's not going to work for the following reasons. And what you need to do is X, Y, and Z. And they were like, okay. And so about a month and a half later, they call me and they said, all right, so we did X, Y, and Z. Now what should we do? Well, that's actually a really good sign that they were listening. And there was a certain degree of humility there that I find often lacking when I sit down and talk with tech entrepreneurs who confront the healthcare system. So you've probably had this experience that I've had where a couple years ago I went to South by Southwest and watched some of the people in the audience come up and say things like, well, what healthcare needs is a lot more data. It just needs more data. And I sit there as a life sciences person or a pharmaceuticals person and think, no, no, no. It needs more quality data. and this all needs to be adequately consented in the first place. It's a different kind of data than maybe you're used to working with. There's all this unstructured stuff too. How do you make sense of that? I mean, so these are the kind of conversations I guess you had with these guys. Well, and as I mentioned, you know, that X, Y, and Z thing I mentioned before, it was all about unstructured data. And my point to them was, you know, basically aggregating healthcare data is cute, but at the end of the day, for cancer research, at least half of the critical variables you need live in unstructured documents, essentially digital paper. Now, for those unfamiliar, what do you mean by unstructured data? Yeah, so these are documents such as the pathology or radiology report, maybe it's the biomarker report, or even the medical case notes where the doctor has written a note to summarize the clinical visit. Many times these are now faxed to the clinic. Yes, in 2018 they're still faxed and then scanned in the electronic health record. They get scanned in and now they exist within the EHR literally as a PDF, which is an image file, really. And one of the things that always makes me laugh is that you can imagine that they're faxed and scanned and they're often, you know, tilted to the side. So they're not even a straight up piece of paper. There might be the crease in it if it came through in the envelope. So it's still. Oh, yeah. Yeah. And somebody's penmanship, the pen starts running out of ink halfway through the sentence or it trails off the end of the page and you can't quite tell what that abbreviation was. I mean, it sounds like a mess. Yes, exactly. And yet that is the document that describes the critical biomarker for whether or not, for example, a patient is predicted to respond to a drug. And that's half of the data that's in the EHR systems for cancer patients. or even other patients too. Right. You know, I would argue it's even more than half of the data. But if I think about the data points you need to do credible research, it's at least half of the data. And obviously that doesn't fit in neat columns and rows that can be, you know, sliced and diced and munched together with data from, say, other clinical sites or other parts of the world. Yep. And if you, you know, go back to why I think the Australia experience was so formative, is that it was really clear to me when I was in Australia that I only still had part of the story by all those systems we were pulling from. And that most of the critical stuff that I needed if I was going to do complex research existed in these pieces of digital paper. In 2009, we had a grant from the National Cancer Institute to try and figure out how to solve for this. And we were looking at natural language processing and other techniques, which were still a little rudimentary in 2009. But nonetheless, we had the participation of some phenomenal computer scientists to try and solve for it. And the thing that I kept coming back to over and over again was that in order to pull out the critical data points in a consistent and high-quality way, you still needed human experts who could read those pieces of digital paper, identify the right data point, and curate it in a consistent fashion. And that was what I had told Nat and Zach. Now this is the point where I would imagine the conversation would stall with a lot of tech entrepreneurs because, you know, this sounds really labor intensive. Like you're going to have to hire a lot of people, and they're going to have to be knowledgeable, and they're going to have to scrub through these documents. That doesn't sound like get big fast and change the world in 18 months. This is, you need data janitors. I think I saw, I think Rob Califf might use that term. You're going to hire a bunch of data janitors. I call them experts. They really are experts because we train them really, really carefully. The interesting thing, and so I'll break this part into two kind of key points. One is that, you know, your point about the tech side, the hubris that, you know, I'm not going to do something that hard is I think actually what means that a lot of groups don't win, right? They assume that there's no way that doing the hard stuff is going to scale. And in fact, if you build a solution that allows you to solve for conducting the hard work, that is both a competitive moat and what will allow you to scale. And so the number of times that I've heard, even in 2018, a tech entrepreneur say to me, I can't do that because it's not going to scale. And I almost kind of giggle now when I hear it. It's remarkable. The part of it is to ask what does scale and figure out how to build that and then ask the things that don't scale, how do I make it as efficient as possible? Well, one thing that when you put in the hard work, it does create a high barrier to entry, which is a nice thing too. Tell me about the Flatiron business model. How does this thing work? So Flatiron itself is a two-sided business model. And you need to remember, because I think it's core to our story, that our core customer is still the oncologist. So the first thing that we do is we make software used by oncologists. Our number one software solution is an electronic health record. So we make the largest electronic health record used in the community oncology setting in the U.S. And so through that electronic health record, we're able to pull in a complete copy of the medical record of the patients who are receiving care in those oncology practices. The second kind of software we build is software that sits on top of somebody else's electronic health record, such as the EHR used at an academic medical center. And they're usually using our software in those settings to do things like improve quality monitoring or other things that rely on then pulling, again, a complete copy of the medical record in and cleaning up the data and then getting it back to the health system to try and help them take care of patients at the point of care. So those two software solutions are the first side of our business and half of our company, really literally half of the FTEs, are focused on building that software and serving that customer base. Now if I'm a community oncology physician, do I have to buy this? Why would I want this? So, you know, not only do you need to buy it, some community oncologists would say our software is kind of pricey. I don't think it's necessarily pricey. I think it's, you know, competitive in the market. But you buy our software because it's what you need to do your day job. So it's your electronic health record, it's your quality monitoring tools, and you buy it from Flatiron, expecting Flatiron to be the best in the business. You know, we are constantly thinking about how do we invest to improve those software solutions so that our oncologists on our systems are consistently believing that we will be basically in service to them building the software that they need and constantly updating it. So they buy it from us. So is it easier to use or just, you know, more thorough or helps you with billing as opposed to somebody like Epic? Well, so it's much cheaper than Epic. So, you know, Epic is often a hundred plus million dollar proposition, sometimes up to multiple billion. So it's usually a software solution that's built for large health systems and academic medical centers. So if you're a five oncologist practice in Gettysburg, Pennsylvania, or even 160 oncologists in the state of Florida working together, the EPIC systems of the world, they're just too big for your needs and way too costly. What we build is the kind of software for the community oncologist. Okay. So you've got a price advantage, and it's clearly made with their needs in mind as oncologists. pharmacologists. There are not a lot of extra bells and whistles that they don't need, presumably in there, which is good. But then they buy it for a year or a couple, some kind of a subscription license? Yeah, so it's a license. It is a software product that they're now going to use in their practice. It's fairly sticky. So even though they're licensing it for one, three years. It sort of depends on the license and the practice. If you're the electronic health record of record, if you're the platform that that practice is using, the switching costs to another EHR are pretty substantial. Oh, yeah. Once you've gotten used to doing this, you don't really want to switch. I get that. As long as it works for you in your practice. So how many people or how many practices are you in? So about 2 ,800 oncologists in the United States right now use our EHR, which represents about 2 million cancer patients that are currently flowing through our systems. 2 million cancer patients. That's what's the market share? It's about 15, 17 % of all the cancer patients who are actively getting treated in the U.S. today. Now, I also see on your website that you work with the pharmaceutical industry, of course, the top 15 cancer drug makers. So this, I guess, is the other half of your business model? Exactly. So you can imagine that then there's a bridge to the other half of what we do. So in the middle, so right on top of the bridge, is that we're pulling all the electronic health records into a central repository. And then a key part of what Flattern does is clean up the data. So the data that is already structured, we put all that into a common format so that it can be used quickly for consistent research. And then the data that is unstructured, we have this cohort of over 1 ,100 human abstractors. These are the human experts that curate the data. Therefore, what we end up with in the central zone of our business are these huge repositories of exceptionally clean, high-quality data. The data then can be used by the oncology practices to understand what care are they providing, how to do that better. But then also the data can be used in a de-identified way for all kinds of research. And that is the second part of our business, which is predominantly focused on the life sciences clients. Okay, so where's the value for those pharmaceutical companies? What kinds of questions are they able to ask from this aggregated data set that they can't ask otherwise? So it's a really interesting and emerging space. So it is this landscape of what's now becoming more and more known as real-world evidence. And I think it starts with some of the basics. It is the questions about, you know, so there's like questions about where are my drugs? Who's using them? So what kind of oncologists are using them? What kinds of patients are using them? What's the proportional market share? What clinical settings are they being used? So are they being used when patients are first diagnosed with metastatic cancer or when the cancers come back after somebody else's drug? So really very descriptive sets of questions. Then the second set of questions are comparative. So how is my drug performing compared to another drug in the same class or perhaps in a different class but now being used for a similar clinical question or clinical situation? Then there are what I would call sort of the progressively more high-value and important questions. So when my drug is being used off-label, how do I understand what the uses are and whether or not there's compelling evidence of that drug having action in that off-label setting? And I can give you a really kind of interesting example of that in a second. And whether or not then information about my drug being used in the off-label setting can now be used in service of, for example, presentation back to the FDA for label expansion and approval of an updated label. Then there's also, can you use these data to understand drugs after they're marketed to answer questions that the FDA requires of us, such as post-marketing commitments and requirements. And most recently, our clients are asking us to think about, can you simulate the control arm of a clinical trial, either providing information that demonstrates what current contemporary care looks like and provides a reference for a single arm study, or maybe even starts to provide the control arm for a randomized study so that you don't need to randomize as many patients so you can back off to like a 2-to-1 or 3-to-1 randomization or even potentially simulate the entire control arm in the future and not need to randomize at all. This is pretty wild stuff, I would think, to the medical evidence crowd out there. I mean, it leads you down the road to eventually sort of like everybody being in a real-time clinical trial, even though you're just getting treated at your community oncology center and you're not really being randomized in a traditional prospective sort of way. Yeah, so it's really this kind of core question of how do we learn from the care of every patient and acknowledge that many times what the oncologist is doing in caring for a patient is tailoring what he or she thinks is the best medicine or combination of medicines or dose for this particular person in their clinical setting. And why aren't we aggregating that information and making sense of it and asking the question, well, if it works for this patient, should we be applying those same solutions then to subsequent patients who've got similar features in the future. And so, yes, it really starts to move us to this idea of continuous learning and the idea of building data sets that allow us to continuously reevaluate the questions sitting in front of us. And hopefully you can learn really fine-grained things in close to real time about, well, maybe this adverse event appears more often for patients who have this certain biomarker. Exactly. And the interesting thing is that it also provides us a framework to keep learning. So what I mean by that is when I think about doing a clinical trial, right? So I was trained as a, you know, classical clinical trialist. I think about doing a clinical trial. It occurs within a fixed construct. So this is the beginning. This is the end. And when I get to the end, I get to an answer, right? But we know that in fact, healthcare is dynamic and the aspects of what are the other drugs that are available, supportive care, features of our understanding of the disease, features of our understanding of biomarkers, that's all changing, right? So it allows us now to keep answering even one specific question over and over again as all of those other things are updating so that we can continuously have the most updated answer to that question rather than the fixed construct of what is a clinical trial. Earlier this year, I think it was February, Roche decides to buy the company. It had already invested, so in typical Roche way, they kind of got their nose under the tent, looked around, liked what they saw, presumably, and bought it. what's changed since then? Actually, you know, I'm sitting here in the office and not much has changed at all. You know, right now, we lovingly call it the anti-integration. So in other words, we're, you know, becoming a part of the Roche group and the Roche family. But for the most part, we don't feel it on the ground every day. I think there are little ways that it's changed. First of all, our ability to focus on the longer term and have access to resources, i.e. money, to invest in getting to longer term is really, truly different. Where I'm seeing that the most right now is on our ability to focus on making sure that we continuously invest in building our electronic health record. If you recall, the community oncologists who are using our EHR still are our core client. That's who we have to worry about first, because if they don't use our electronic health record, then none of the rest of it actually happens. And so I see us being able to invest more. I think that there are other little things such as, you know, some of the Roche systems that we need to interface with, such as some of their finance systems and those kind of things. But otherwise, for the most part, I think Roche has acknowledged that we're a tech company and they're not used to being managers of tech companies. And so it makes more sense to leave us alone and allow us to function as a tech company with tech ethos within their overall circle of the Roche Group family. But now back to your business model, you're still very much actively selling the software solutions to the community oncology practices. but also the other 14 top cancer drug makers. I mean, this is not just proprietary for in-house Roche questions. Yeah, and if I keep going back to our core customer, our core business model, I think it becomes progressively more obvious why it has to be that way. If we only use the data in service of Roche, then if you're an oncologist, it feels like a pretty strong conflict of interest, right? So, you know, as an oncologist, how am I going to trust Flatiron is, you know, now working with my best interest and my patient's best interest in mind? Are they just really trying to get my data and sell me stuff. And so if we're working with all of the life sciences companies, as well as with academia and government, then as the oncologist working on our systems, those oncologists can have more confidence that we're magnanimous in the way that we're thinking about things. Because in order for us to work with all of those different groups, we have to develop the trust with all of those different groups that we do not give preferential treatment to any one group. And I think the best evidence of that trust is BMS's announcement a couple of months ago of a very substantial three-year commitment to working with us. Now, another thing that I did see in the, I think this was in the Roche press release when they announced the deal, that they said something to the effect of the integrity of segregated patient protected health information will be preserved, as will dedicated sales and marketing, provider facing and life science business activities. Now, when I read that, I thought, okay, so this, what you're saying is there will be an arm's length kind of thing, we're going to look at aggregated data and answer some of these questions that maybe we can fold into our new drug applications where, you know, time really does matter. If you can get a drug approved more quickly based on greater real world insight or, you know, a secondary indication a few months faster, that's great. But what they can't do is you can't have a Roche salesperson looking at individualized data and saying, well, wait a minute, this doctor prescribed the Pfizer-Alk inhibitor and that patient failed, you know, a couple of weeks ago. Why isn't he prescribing, our ALK inhibitor. Call them up. I'm sure there's a natural business temptation to do such a thing. Yeah, so that's a perfect example. I think that the kinds of examples that you're highlighting, the fact that we have to have a firewall between Roche and Genentech and the oncologist and what doing is very, very clear. Further, we have to have a firewall between what we're doing and the decisions we're making on the ground with our life sciences clients and Roche. So, you know, you just mentioned two different companies that make ALK inhibitors. Both companies have key questions that are part of their developmental strategies, right? We cannot disclose any of that information. In fact, have created internal structures to wall off the teams so that even inside of Flatiron now we have very carefully circumscribed management of confidential information. You don't want to let those different clinical development pathways outside of Flatiron or even a lot of times outside of the teams. And we also have to respect the fact that Roche often is making competitor drugs. And so we are very careful about how we manage that information. What do you worry about a lot? I mean, what, you know, the proverbial, what keeps you up at night when you're into this job? So, you know, I'm going to answer this, honestly, Luke is Amy. So it's, you know, in this job is the CMO and CSO, but it's also in this job as I watch the pointy tip of the spear in where the science is going. So within the landscape of Flaturn, the two things that keep me up at night, one of them is always going to be a data breach. I've just lived in this overall space of data and tech long enough to know that data breaches happen and being able to do everything we can to secure that from not happening and also figure out if it were to happen, how we're going to manage it quickly is something that keeps me up at night and I think about often. But knock on wood, that hasn't happened? No, it's not happened. And hopefully being preemptive is the way to solve for that. I'm the one who was, you know, really, really pushing us early on before most tech companies in our space would have done this to make sure that we've got, you know, as many of the high tech and SOX compliance boxes checked as possible, et cetera. I think honestly, though, as the chief scientific officer, the thing that makes me the most anxious and what I worry about most is that the way to do this science, how we get from key question to answer, especially in the space of continuously aggregating data and the ability to look over and over and over again, is science that's not been clearly worked out. Thinking about the stats, thinking about the application of models. And so my other kind of key point of anxiety is how do we maintain a very rigorous view of how do we do good science? How do we think carefully about what we're finding and not get out of our skis of promising things that just aren't possible yet? And so I think that that's our responsibility. It sounds like, I mean, hearing some of this makes me think, you know, five or 10 years from now, we could wake up in a world where a lot of the pain with randomized controlled trials could be alleviated. So the time and expense and all the trouble people go into with enrollment, you can imagine a much more quick and seamless and adaptive sort of learning system. But there would also be maybe a new set of headaches could emerge, like jumping to conclusions, say, making snap judgments before we really should. Is that kind of what you're getting at? Yep. And, you know, as a person who has been working and thinking about this now for what amounts to literally 20 years, it's kind of crazy to think about. I feel like the momentum and excitement around it, which is just awesome, needs to be maintained in check with making sure we are honest with ourselves about what we can and cannot do every day. And so part of my job is to simultaneously talk about the what's possible with having people hear me say as clearly as possible, but here's what we need to plan for. And that's, I think, where we are right now. And so the irony is like sometimes now when I'm speaking publicly, I sound like the voice of caution where I used to be the maverick. and that's because now that I'm seeing these things escalate I want to make sure we're also doing the right thing and I do see some amazing things coming in front of us like literally the ability to accelerate how we think about clinical trials is our next big frontier Very interesting Amy I think that's a great spot to wrap up. Thank you very much for joining me today on The Long Run I really really appreciate it. Thank you for your time Thanks for listening to The Long Run, a production of Timmerman Report. Pedro Rosado of Headstepper Media was the producer and editor. Music comes from D.A. Wallach. And thanks for listening. If you're interested in sponsoring the show and in raising awareness of your work among industry thought leaders, send me an email at luke.timmerman at protonmail.com. See you next episode. you