The Impact of Generative AI on Healthcare with Dr. Stefano Bini, an endowed professor of Orthopedics at UCSF and Chief Technology Officer

[00:00:01] This podcast is produced by Outcomes Rocket, your healthcare exclusive digital marketing agency. Outcomes Rocket exists to help healthcare organizations like yours to maximize their impact and accelerate growth. Visit outcomesrocket.com or text us at 312-224-9945.

[00:00:29] Hey everybody, Saul Marquez with the Outcomes Rocket. I've got Dr. Stefano Bini with us again. Dr. Bini, why don't you say hello? Hello, Saul. Welcome. I mean, thank you. You're so used to being a host. I'm so used to being a host. Thank you for having me on your show again. It's a pleasure to have you back on. Folks, Dr. Bini needs no introduction, but he is an endowed professor of Orthopedics at UCSF, also functioning as the Chief Technology Officer.

[00:00:59] And he is also the host of the DocSF Conference, as well as the podcast. And I'm excited to have him here today because he's been doing something that caught my eye that I wanted to share with everyone. And that is the Generative AI in Orthopedics webinar series and discussions. And we were going to cover a little bit of that, but also a couple other things that Stefano's got going on. So Stefano, thanks for being with us today. Awesome, awesome.

[00:01:27] And to clarify, DocSF stands for the Digital Orthopedics Conference San Francisco. It's a digital health conference that's kind of focused on the musculoskeletal vertical, which is a pretty big vertical. It's around 20% of Medicare spend goes all the way from trauma to wellness.

[00:01:43] And so we used it as a way to bring the heady discussions about the opportunities of digital health down to the ground, literally to the front door of health care at the use case, which we just happen to think of as orthopedics. That's awesome. Yeah, it's a great meeting, folks. If you haven't had a chance to attend it, definitely one that you have to check out. It happens every October. So still plenty of time to plan for 2024.

[00:02:09] So, Dr. Beeney, talk to us about this webinar series you're doing it on Gen.AI and Ortho. Why are you doing it? And what are some of the early learnings? So Gen.AI obviously caught us all by surprise almost, right? Unless you're paying attention to ChatGPT 2 and 3 and the early promise of Open.AI's vision. And most people weren't. It sort of came out of, seemed to come out of nowhere. In fact, it was, it's 50 years in the making.

[00:02:34] What happened though, because it caught our attention and because the user interface became the chat interface, which is really what opened it to us all, plebs, is that it was very important for us to bring it to the audience. Okay, let's clarify what this can and cannot do. Let's talk about the things that people are worried about in health care, particularly hallucinations.

[00:02:53] And let's understand how it sits in the world of digital health, which hasn't had this powerful form of AI at this point yet really come in. And so in order to do that, when something crazy like that shows up in our doorstep, DocSF generally does these webinars. We tend to be first out of the block with them often. And what I do for these, rather than get a bunch of doctors involved, I usually try to get a broad spectrum to give us perspective, spectrum of speakers.

[00:03:21] So in this case, we have Prashant Nathrajhan, who is vice president of strategy and products at H2.ai, which is an open access platform that's used in general AI. And we can talk about that. Peter Schilling is a colleague of mine who's an orthopedic surgeon in the East Coast at Dartmouth. And he does a lot of AI research. I got the research scientist, I got the person that I used to build the technology. And then what we had in each of the two webinars, a different venture capital person to come in and tell us a little bit about where they see the opportunity.

[00:03:48] Because it's important to think through not just the technology, but its financial viability. This is what these guys really do. They come in and ask the tough questions. Like, how are you going to pay for this? Sounds great. Saves the world. How does it get paid for? And then I came along for the ride to ask some tough questions. So it was a really exciting series of two webinars. We got tremendous positive feedback. Our goal in them was to define what generative AI was, put in the context of AI in general.

[00:04:14] We actually on the first one really went through the use case of large language models. And then the second one, we actually went into more multimodal AI, sorry, multimodal generative AI models. And then transition more to understanding what foundation models as a whole are about. And focusing on some early applications in healthcare. And also we spent a great deal of time talking a little bit, talking about hallucinations and how we're going to get around them. So let's talk about some of the specifics there, Stefano, like AI capabilities.

[00:04:44] Maybe talk to us about one of those integration into orthopedics and then the potential to revolutionize the field as we know it. Is it or is it not? Integration of the orthopedics. Here's the thing about that. Basically speaking, we have healthcare that falls into two massive buckets. Actually, it's probably more than two, but we've got the surgical side and clinical side. Then you've got pediatrics and there's some other iterations because peds is its own thing. But look at those two, they have different sets of needs. And I think it's important to understand that.

[00:05:13] But within like surgical specialties, the way we're going to be using some of these technologies in orthopedics, basically can be the same for general surgery, OBGYN, etc. It's a series of talks. So I do like to use orthopedics as a use case because it makes it very palpable and very obvious. But it's not like these technologies only live in orthopedics. So let's talk about large language models, right? So my favorite part of a large language model is the ability to answer questions. It goes into a database, it sorts it, it takes that information,

[00:05:40] and then it reformats it in a way that I can understand, right? And it can give it to me in any context and it can give it to me in any language, which is also phenomenal, the translation side. So now we're talking about accessing information in ways that it's easily digestible by patient. Now that information everybody's been worried about is the one that is relative to diagnostics. Everyone's worried that it's going to get the wrong diagnosis, the patient does the wrong thing. But there's so much more in healthcare where patients are needing information.

[00:06:07] It could be about their clinic access to a renewal of their medication. It could be access to a clinic visit. It could be an explanation of what the doctor told them in their clinic visit, but didn't have time to really explain at their level. Maybe you need to have the information simplified. So some of these large language models can totally help in healthcare. So chatbots are super cool because it creates this user interface. Now, if you look at some of the data where we're using Boolean algorithms, which is the old school way, if this, then that,

[00:06:37] we're still getting about 60% of patient questions coming into a clinic answered. Can you imagine how much further we can take that, how much better we can take that using a chat interface that is powered by AI? Now, when we talk about it being powered by a generative AI model, the interesting thing about where we're going with this is not that it's going to the web to get an answer. It's going to a database that is structured within the healthcare system.

[00:07:03] It's the same information that we already offer our patients in print format or on our phone calls, because now, generally, I can actually listen to our phone calls and bring out what we told patients for the last six months. And so we can control the information that goes out, or we can limit the extent to which the model is generating information and limit it just simply to making it contextually relevant to the patient. So that's clearly the unimodal version of generative AI.

[00:07:31] Large language models are going to be obvious for a selection, but there's so much more. Thank you. Thank you, Dr. Beeney. Super interesting. And so is that where the hallucinations thing comes into play and gets resolved, possibly, when you limit the amount of information that you input to the large language model? Correct. So there's this RAG format, which is becoming quite popular conceptually in terms of how these models are being applied to industry, because a lot of this work has been done outside of healthcare,

[00:07:59] where a company that could be a shoe manufacturer needs to understand and pull data from their diverse data systems. And they don't want to necessarily have an engineer spend hours and hours building a query to pull out the questions and answers. Simply tie the software into their databases, and then it becomes a chat interface. Tell me how many shoes I sold in Japan. Now, last quarter of this color to this age shirt. So that is super easy for this. It's easy for this software. It's actually not that complicated.

[00:08:30] Transition that into the healthcare systems. How many patients did I see who were 45 years to 50 who came in for this problem? So you can start asking those kinds of questions, which may not necessarily be healthcare related. We're not talking about what drug should I give this patient. It's how do I adjust my resources to meet the need of my community? Are there patterns in this demand over time that allows me to model my OR capacity,

[00:08:57] my staffing models, my X resource allocation? So where I believe strongly that where we're going to see the first ROI isn't necessarily in getting these models to be good enough to act as physicians, clinicians, advice givers, because that's fraught with all kinds of complications. Though we know it's coming. Palm 2 is quite strong already, but it doesn't have to be the starting point. Love that. Folks, tip of the iceberg.

[00:09:24] What I wanted to do today is let you know that these conversations are being had. And Dr. Beeney is the spearhead of these with the groups that he puts together. In particular, two webinars that have already happened. One of them that's going to happen soon. Today, I'm just giving you a sneak peek. The conversation is being had. So if you're in orthopedics and you want to know where Gen AI is going, you have to tune in. We're going to leave the link to the two webinars that have been hosted. Both of them live on the website at DocSF as well as on YouTube.

[00:09:54] So just go in the show notes. You'll get a link there. And then Stefano, you're about to go on a sabbatical. Tell us about that. That's amazing. I am. It is amazing. It's an opportunity to start thinking about a little bit broader. I mean, we're in the United States. We're doing great things with technology. But it's remarkable to me how the future is here, but it's not evenly distributed. And my goal is to travel around the world meeting with really interesting people

[00:10:20] actually solving real problems, especially talking about them in New Zealand, Australia, Singapore, India, all of Europe, Turkey, Italy, Spain, Germany. I've set up meetings everywhere. That's all on the agenda? That's all on the agenda. And I'm looking forward to hearing from any of your listeners who want to talk a little bit about what they're doing with digital technologies to solve problems in healthcare. Amazing. Hey, you're a renaissance man. You live the life, you work hard, and you bring insights forward. So I really appreciate you jumping on again today, Stefano.

[00:10:48] And everybody, make sure you check out this webinar series and you reach out to Stefano if you have any questions on how you could get engaged and participate. Can't thank you enough for being on, Stefano. Awesome. Saul, such a pleasure.

[00:11:13] This podcast is produced by Outcomes Rocket, your healthcare exclusive digital marketing agency. Outcomes Rocket exists to help healthcare organizations like yours to maximize their impact and accelerate growth. Visit outcomesrocket.com or text us at 312-224-9945. Outcomes Rocket exists to help healthcare organizations like yours to maximize their impact and accelerate