Eliezer Van Allen: Appreciate it. Thanks for having me.
Zachary Klaassen: So I just want to start at a very high level. I mean, we're talking to people that are starting to incorporate artificial intelligence, maybe in their practice, in their personal life. What are some of the sort of just key things in medicine with artificial intelligence today that most people are probably aware of?
Eliezer Van Allen: So yeah, there's already been some adoption across the board. So it may not be as relevant to this particular clinical audience, but so for example, there are mammography AI tools that are now effectively built into the machines that help radiologists find lesions that might be concerning. Similarly, there's a variety of AI tools that I think, we as clinicians, are already using to help us on decision support or ambient scribes, things that are listening in in our conversations, really to help reduce the burden of documentation. Those kinds of tools built off of AI that's very easily portable from AI developed in other domains to our world have already started to permeate all of our clinical practice.
Zachary Klaassen: Yeah, absolutely. And I think even in my practice, look at ArteraAI prostate test. I look at some of the predictors now for bladder cancer and non-muscle-invasive disease, what we should be using. What we learned today from your talk is that this is just scratching the surface. So you had three sort of pillars, how you broke your talk down. Maybe just give our listeners some highlights from each of those.
Eliezer Van Allen: Right. So the way I structured the talk was I used a patient example as a premise. I said, "There's a patient of mine that had this extraordinary response to cancer immunotherapy." And that kind of a situation prompts three questions. Why does this happen? What's the biology that explains this phenomenon? For whom does this happen? What are the biomarkers that might be able to help us stratify which patients will likely respond or not respond? And then, how do we actually deliver those insights such that everyone can benefit and we're not restricting ourselves to very small or tailored patient populations? And so for each of those questions, we're now in this space where AI has the potential to really accelerate our ability to answer all of them in spades across the board.
Zachary Klaassen: Yeah. And I think what's interesting, your group is doing some really interesting work. And just maybe delve into, particularly, you gave some examples of how you may put in information in other cancer types and sort of working through those algorithms really can help maybe predict what we can treat these patients with.
Eliezer Van Allen: Yeah. So one example that I talked about in this lecture was, if we start with some of the AI models that exist for very tailored, specific questions. You mentioned the Artera digital pathology tests. There's also molecular tests, there's clinical tests, there's radiology tests. But they're all very focused on one or two or maybe three questions in totality. Where we think the field could go is moving away from these static and very focused, specific kinds of biomarkers into generalists and actively-learning biomarkers. Such that you deploy these AI, they're called agents, into electronic health records. And they're basically constantly learning from larger and larger sets of patient experiences within a hospital. And so as an example, I showed what happens when you build one of these things in our hospital and you train it on 30,000 patients and then you propose, "Here's a new patient, who's similar to my patient." And in this case, the patient had prostate cancer, the molecular version of the model would suggest that actually some of the most similar patients were not prostate cancer patients. And then they got different therapeutics. In some cases, those did or didn't work. And it's those kinds of insights that we can imagine delivering at the point of care really across the board, which is quite exciting.
Zachary Klaassen: Yeah. I think you guys are obviously at the cutting edge of this. How do we implement some of these, even at a, not a basic level, but an intermediate level in our practices across the country?
Eliezer Van Allen: So you hit on a huge point. So there's one thing to sort of develop innovation and try some new tools. Implementation is a very hard and difficult challenge. We need to make sure these tools are vetted and they're safe and that is a very hard problem still for a lot of AI tools. Many of them are biased or they don't generalize to populations well or they have other quirks or weird things that need to be addressed. And they need to be done in a way that is sort of ethical and meaningful for patients and providers alike. We need to understand what the ethics are of how to actually do this properly because this is such a transformative and disruptive technology at some level. All of those things are not fully worked out for some of these more generalist kinds of approaches. And we're about to stress test what's going to happen when they get introduced. And we know this because some other similar kinds of tools are already in everyone's hands.
Many people are already using language models and sort of agentic language models that you can probably imagine what I'm referring to. Things like ChatGPT or Claude or whatever. And asking medical questions of it. And then coming to their doctors maybe even to see what they think or in some cases it's possible that providers are using those things too. So we already know this is happening outside of a proper framework, and we need to make sure that there's actually a vetted and safe framework to make sure this is deployable safely for patients.
Zachary Klaassen: Yeah, great points. And I think you showed one picture I thought was interesting, your discussions with the US Senate. So what's the government discussions like? If you can just take us behind the curtain a little bit.
Eliezer Van Allen: Yes. So actually one of the, I'd say, honors of my professional career was getting to participate in a couple of cancer AI panels in the US Senate. This was all moderated by the former NCI director, Ned Sharpless. I participated along with members of, largely, folks from the technology and biomedical sectors. I was the only physician-scientist and the only oncologist on the panel. I think I may have been the only MD on the panel. And we had a chance to articulate the challenges we're each facing in this space. And I do think interestingly, even though we all came from different spaces, what we were really asking these senators and their staffers... So the senators were all sitting up front, the staffers were behind, but they're the ones taking notes.
Zachary Klaassen: They're the ones taking notes.
Eliezer Van Allen: Right. Is, help us all work together to make the rules. If we don't have rules, we don't know how to build this. But we can't make a lot of those rules, but we can help guide them. And so it should be, and hope it will continue to be, a collaborative effort between the stakeholders in the room, the physicians, the patients, and everyone in between, and the lawmakers who are the ones that are supposed to oversee all of this coming together. I think that's still... I think it's happening. I actually think there's a blue sky version where it'll work out really nicely.
Zachary Klaassen: Excellent.
Eliezer Van Allen: But it depends on constant engagement.
Zachary Klaassen: Yeah. Thanks for sharing that with the listeners. I thought that was great. Last question. As we fast-forward, or 10 years seems like a long time in this space, let's go five to 10 years. How do you see this sort of coming together and what's the goal for the patient?
Eliezer Van Allen: Yeah. So before I even get to that, I think the pace thing is key. Things are moving so fast. There are things that I showed in a keynote that three months ago didn't even think was possible, and now it's almost trivial. And three months from now, it's hard to say what's going to happen.
Zachary Klaassen: Sure.
Eliezer Van Allen: I think a blue sky version of what the fields will look like or could look like, is this notion of the best of human intelligence paired with the best of artificial intelligence working together. There are things that AI is very, very good at, and better than us at already. And that certainly will continue to get better. Coding is a good example of what we're seeing in real time. Why don't we just take advantage of that and use that to make all of our work go faster? And that could be scientific discovery work, that could be clinical care. And then especially on the latter, the more we adopt in a safe and meaningful way, I actually think it may help bring the humanism back to medicine. I mentioned it in, I think it may have been the Q&A or maybe it was in the talk. We all are taught first day of medical school, you make eye contact with the patient, you've got to make them feel comfortable.
There's this sort of human element to doctoring. But in real life, what we do is oftentimes we turn the swivel over the chair, we're typing away in our electronic health record, and we're all having an epically miserable time doing this. But if you have even just now, like ambient scribes, if they worked well. And that's the kind of thing that could be deployable to everyone, all of a sudden I can swivel back and just do what I was supposed to do. The patient's happier, the doctor is happier, there's less burnout, and everyone wins. And that's not 10 years from now, five years from now, that's today. The five years from now version is that blown out to the nth degree, where every element of care delivery has these kinds of enhancements built in. And that they're grounded on the sort of advances in biology and drug discovery and drug development that help facilitate the best drugs for our patients. And that is the true marriage of sort of the best version of us and our minds freed up to do the things that we're really good at, paired with the best versions of these AI that are helping us and working to make the hard parts for us go easier.
Zachary Klaassen: It's a fantastic answer. Congratulations. Great keynote lecture. Really enjoyed the discussion. Anything we haven't hit on you want to share the listeners? Your wrap-up points.
Eliezer Van Allen: In terms of wrap-up points, I think perhaps the most important thing people should all do in real time is get used to these kinds of tools. Try them out. Everyone can access these things. Doctors are nervous. Will AI replace us? I don't think so, but doctors who use AI will definitely replace doctors who don't use AI. Similarly, you want to be an empowered patient, use these tools. They have challenges and problems, but they're increasingly sophisticated on a medical level. And then ask your provider what they think if you have questions. The more people get comfortable, that also will help them advocate for and speak out when they see things that are wrong, that need to be fixed. And that's, I think, the best way we can all participate in real time.
Zachary Klaassen: That's a great concluding statement. Eli, thanks for joining us on UroToday. Appreciate your time.
Eliezer Van Allen: Thanks for having me.