Good morning, everyone to another episode of Signals and Symptoms Podcast. And we are so, so happy to bring our friend, Talia Baker. We are excited about today's episode because this is really what we think about moving towards frontier of this. She's essentially a transplant surgeon and wants to improve communication and downstream. And one of the things that we're to talk about in this particular episode is how to generate real world evidence for treatment. This is going to be something truly amazing, which she is trying to build. And if she has inspired me as well, I mean, we are retuning some of the stuff that we have already built for stroke, but also make it available for liver cancer.

One of the important thing to understand is called OpenTSLM, which is time series language models. Now the problem with the language models is that they are really great in generating data. But when you have to actually go through the fourth dimension, which is time, the large language models used to suck. But this one, there's a new open source model, which is essentially from Stanford, that they went to, first of all, they helped us create this. And then more importantly, they were able to understand that how this is gonna move forward in terms of methods. So those who are interested in looking at the technological part of things, you're gonna see that patient reports this, then the time series in quotes, heart rate and oxygen at the same point. Again, remember these are two different events, then pre-tains and then text and quotes, and then actually gives you the prediction what is called possible sleep apnea. This is a significant important way of improving large-language models in terms of time series and in terms of prediction of events in the future. The second thing I want to do is, Artifact Read is to bridge transitional gap. And this is what Talia is doing, which is what we are trying to help as well, is to make sure that we support not just clinical reasoning and decision-making, but in a process where the transitional gap lies and make sure that we enhance that process. It is very important when we are doing experiments and when we were doing our internal experiments to understand the chain of influence. We always keep talking about chain of thought. Chain of thought is again, you know, describe basically multiple steps and then the LLM gets multiple, you know, inputs and then they go to the chain of thought. But it is also more important to understand chain of influence. And then lastly, of course, this is another one that I have shared before, but I just want to reiterate why we are doing this is large language volatilization health trajectories. So I'm going to stop here. I apologize it took a little longer, but I just wanted to give you the whole scape of things. And then I'm going to let first, Talia go ahead and introduce herself, the Lory project. And then of course, and Harvey is going to take over.

Perfect. Thank you so much. What an amazing introduction. So this is a really exciting time. I feel very grateful. am, my name is Talia Baker. I'm a clinical liver transplant and complex appetitory surgeon. I've been in the space for a long time, for about 25 years. Having the privilege of really building a career around liver transplant, living donor liver transplant, and ultimately kind of I'm developing into someone who really was deeply into the liver transplant oncology space, which is consideration of liver transplantation as a solution for patients with liver cancers beyond the traditional criteria that we've always thought of. In that process, I really came to a point where I realized that as a clinician and having deep personal privileged relationships with patients that the liver cancer journey is really broken. And in today's world, I think we can do better. And that's why I stepped away from clinical practice and really have committed myself to developing a platform where we could take all of the problems that I really lived and saw in my patients journey and use the new technology that we have. embedded deeply in the human component of what I believe is the heart of what we do as physicians and as patients going through these journeys and make a process which is better for patients, saved lives, but importantly also addresses all the institutional challenges which cost institutions and the health system in general a tremendous amount of money. So I feel really lucky because I've kind of. My passion is liver cancer, but I really see this generalizable to all cancers because as much as I understand that the liver cancer journey is broken and I have lived and have deep knowledge and am privileged to have a great network around the country of people who are committed to making liver cancer better. If we can figure out this process, I think we can make cancer care in general a process which will benefit everybody.

Well, this is amazing. And I'm so thankful whenever, you know, I meet new people that are doing amazing things to help improve the life of other humans. I've always feel so privileged myself and so lucky to be just having this conversation. But it reminds me of something someone said, I think on one of our podcasts that we're practicing on today's patients using yesterday's technology, hoping for a future better outcome. Something like that. I wrote that down. But people like Dr. Baker are definitely attacking that. Yeah, I'd be super curious to talk about at a high level, because I know we have a ton of questions, like what is today's, what's the old process, today's process for most organizations, and then leveraging your system, what could it be? Like how does your system solve this? So what is the old way for that a liver patient might go through with cancer, and what is the new way with your technology? I think that would really help bring it to the forefront.

Yeah, and to be honest, our technology is very much in the visionary stage right now. But the problems that we're addressing most significantly are exactly what you're describing. As patients enter the health system with a cancer diagnosis, the system is super fragmented and siloed. As you described, we know a lot and we're not applying it. Multidisciplinary care of cancer patients has been proven time and time again to improve patient outcomes, to decrease system costs, and to ensure that patients have timely access with limited care gaps to their sequence treatment. That being said, if you look around the country right now in liver cancer care, less than 25 % of institutions around the country have access to true multidisciplinary care. So the very high end academic institutions, which are about the top 1 % in the healthcare system in North America have really beautifully designed multidisciplinary teams. But even those are flawed because they are limited by that institution's biases, local expertise and incentive structures.

No, I love that. And Dr. Baker, you're right. You said it earlier that what you're developing is necessary, not just with liver transplantation, the whole system. Because when you're going through that story, my gosh, I know that story all too well.

I just wanted to clarify these points for my audience. If you have a blank page and a wish list, how would you convince? And that's the problem. And I'm going to tell you, look, technology part I'll solve for you or you'll solve for yourself, whatever. But how do you change behavior and institutional incentives in a way that, at the point of entry, use the system? How would you tackle the hardest problem?

So that's the hardest problem. And the hardest problem is actually as a physician, I can say, and I think you have lived this as well, that the real hardest problem is our hubris as physicians. We go into this because we, all of us truly believe that our mission is to help patients to save lives. That's why we went through medical school. We went through residencies. I loved every moment of that and felt very privileged to do that. So when I came out, Like my lens is, I wanna figure out if transplant's the right thing to do. The smartest surgeon who I ever learned from what told me on the first day of my internship, and I will never forget this. I was at Columbia Presbyterian. He was an endocrine surgeon and he said, He looked at all of us, were 10 of us who were starting as interns, including the neurosurgeon interns, orthos, et cetera. He said, you know, I can teach all of you how to operate. I can teach you how to do any operation, any complicated operation. I could teach a monkey how to do an operation. What I can't teach you and what you're gonna struggle with for the rest of your life is when not to operate. And that statement has stuck with me forever. And I think I bring it up only because I think it really addresses what you're saying, which is the biggest challenge that we have getting something like this implemented is that if a patient ends up in a surgeon's office, if it's technically respectable, they're gonna wanna respect it, even if there are other multimodal treatments which could get that patient to a better outcome not because they're bad people and they wanna do the wrong things, but they wanna bring the skills and the privilege that they've had throughout their career to help that patient in that moment. I say all of that because we've thought a lot about this. Again, our biggest problem is not gonna be trust with patients, not trust with health systems. Our biggest problem is to be able to implement trust with the physicians we're going to have to get these patients to for point of care treatment. And that's why we have deliberately designed this as an independent digital entity outside of the specific institution. But I have the privilege of having relationships with really forward thinking, multidisciplinary people around the country and globally who are invested in making liver cancer care better.

Yeah, a couple of things, Dr. Baker. Thank you for being here. Big picture, I just gave a talk at Proceed for pancreatic cancer. Really interesting. And the current theme is the data is siloed, the institutions are not helping each other. We can do so much more. And so I love your platform. And I definitely want to talk to you. Just look at your framework to see if maybe that's something Proceed could do, because I'm helping them with the AI building

Harvey, before you go into how do you teach LLM? I don't know, which is a bigger challenge. So as we have to teach humans not to cut from a technological perspective. this is and I need your opinion on that. Like LLM wasn't their hallucinations. How do you teach LLMs? don't know.

No, no, but on a serious side, that is such a good question. And I have seen some programs that they are trying to incorporate that into LLM. And obviously that's why we're doing rag models and we're increasing the token size to get around those issues. But with that said, you're right. My favorite phrase is we don't know what we don't know and the LLM doesn't know that it doesn't know. So that's why it's making it up because it doesn't know. And going into the geeky science, the vectorizing of the information, the way you're asking the prompting and how it gets that information could help alleviate some of those issues.

Yeah, part of, if I can say something provocative, that we focused too much on quality that we lost sight of the bigger picture. And so we have to move from products in healthcare, like in a hospital setting, to services. so we got really good at, hey, we're really good at lab and look, our percentage of how we do lab, we're so good in our, you know, imaging, radiology, we're so good and we focus in silos. so to Talia's point, your patient comes in and if they, enter wherever they enter first, whatever specialist they might encounter first. They have a great process within that specialty and they're gonna take that patient through and for them they're gonna have great statistic. But in the bigger picture, it's not gonna be a great experience, it's gonna be costly, won't be the optimum clinical care. And so we have to take a step back, I think as hospitals, as leadership and think about services that cut across all the different specialties cut across all the different modalities like lab and radiology. So I tell you, once the vision is complete and the product is there, so I'm at a hospital, how would I interact with your organization?

So that's a great question. That's something that we're just developing and I would love your opinion of how this would be best. So we've had a lot of kind of exploratory ideas about how this would be. Would this come from, for example, a primary care physician who would have a patient or a GI person who as much as they love this problem, it's really kind of, it's burdensome because the administrative drag of taking care of someone with cancer is huge, right? So they'd like to hand this off. So how do we get it from that diagnosis to a platform before they get to that gastroenterologist appointment that's scheduled three months out? And to your point, one of the things that we really need to solve with this, and this is one of our struggle points is the care gap between the diagnosis and the point of care implementation a plan and then even worse when that plan will be implemented. As you say, can mean the difference between presenting at stage one completely curable cancer versus stage three or four, we're looking now for survivability.

Dr. Talia, anything you want to share? More important that we have missed, anything you would like to say to the audience, which is mainly CIOs, CMOs of the hospitals and clinician-led innovators of this particular podcast that you would like to message, whatever that may be.