Ask AI
Back Back
AI in Cancer Care: Promise, Pressure, and the Physician’s Role
Activity Information Ask a Question

Activity

Progress
1
Course Completed
Activity Information

Released: April 29, 2026

John Marshall
John Marshall, MD
 
Catch Up On All Episodes of Oncology Unscripted

Watch Now

Main Topic - AI in Cancer Care: Promise, Pressure, and the Physician’s Role

[00:08] John Marshall, MD: John Marshall for Oncology Unscripted, and I want to thank our partners at Decera Clinical Insights for the support they are giving us and the opportunity, and honestly, the platform that they are giving us to share the thoughts around our world. And what we're really trying to do is decode complex science, and one of the things we're going to talk about today is the big decoder in the sky, and that is artificial intelligence.

It's everywhere. Everyone's talking about it. The quality of AI has changed dramatically, even in the last year or so. We recently took some information that we collected on Post-it notes from a meeting 3 years ago that had been sitting in a shoebox in somebody's office, and we uploaded it, scanned in what people had written on the Post-it notes, and now put it into AI.

And it's cranking to see what was the overall theme of the feedback that we were given at a big, busy meeting where everybody was asked to give their information on Post-it notes. But that's where we are with AI at this point.

So, I want to take a little bit of time and kind of drill down on the issue of AI in health care, and I want to start with a paper. It's actually more of a news article that was put out about The Cancer AI Alliance. This thing that I'd really never heard of before is a group of academic centers that are partnering around their data, that they are sitting on top of patient data to combine it and see what they can learn from it.

So, what they've actually done is they've got millions of de-identified patient data, and the participating centers then share it, and they're promising they're maintaining patient security, privacy, and adherence to regulatory and ethical standards. And so, this Cancer AI Alliance is out there and doing their thing.

And a guy named Eliezer Van Allen, MD, who's the chief of population sciences at Dana-Farber, he's an associate professor up there, came out with an article about the pilot projects that they are doing, and it's sort of the hope of the future that they can take these millions of patient data and turn it into things like predicting treatment response, identifying novel biomarkers, things that we can understand, what makes cancers tick, and so we can target them more easily.

Take rare cancers and analyze the trends and how new therapies are working in those rare cancers. And then, look at large language models on patient data so that we can predict the future for people. Almost equally important is: can we replace placebo or control arms with AI data, right? We've got all of these data sets that we're entering into systems. Why do we have to enroll a patient prospectively, spend the money, have the patient be discouraged, they're on the control arm, when in fact we might be able to use AI to predict and replace control arms and placebo arms?

Could we predict outcomes? We're already doing this with minimal residual disease. Could we reduce this sort of trial and error of giving a patient treatment for, I don't know, 2 or 3 months and seeing if it worked or not? Couldn't we predict faster on that level?

How about manuscript production? I'm editing a book right now, and one of the things we're all really nervous about is somebody's just basically saying, AI, write me a chapter on precision medicine in this or that.

And you know, what about creating a lecture? We are already seeing people use slides and actually giving a call out that I asked AI to build me a slide that said whatever or made the point about whatever. And they're really cool slides, but they didn't make it themselves. AI made it.

And so, how long before entire lectures are maybe AI-generated? Should we be using AI to produce guidelines? We still gather in the United States around cancer, around a table, and look at the data, have base knowledge, and make decisions about what should be officially on the guidelines. How much should AI do that?

I've got a recent grant proposal that came across my desk that actually says that they can predict different tumor biomarkers depending on the life status of the patient, how much stress they're under, how much, you know, what their nutritional status is, these kinds of things, that those could be connected to actually molecular biology.

And so now I'm thinking maybe we're going to have one day a really smart watch. So, not just telling them my pulse and how many steps I got today, but really how I'm doing at that moment with regard to my health, and so this is where it's going. But wait, how do we know what's real? How do we check our results?

Doximity right now says that a whole bunch of us are using AI all the time. I think the number is somewhere around 60-plus percent, and Doximity is also establishing something that they're calling Peer Check, where a human actually gets to look at the data probably pretty quickly and say, that seems fine, or that seems shaky.

I have been thinking a lot as a scientist. When I use AI, I think what it's spitting back to me is probably pretty good quality because the data it's going for is published, peer-reviewed data. But you can see how this quickly could become a circle where the peer-reviewed data is being generated by AI, as well as the cross-checks are being created by AI.

So, we need to fundamentally have enough knowledge base in order to see the checks and balances. And we drilled down during our interview with 3 very smart people about this and how it's sort of being deployed right now and how humans still have a strong role in all of this.

When I first started this whole process, what did I do? I actually typed into AI, and I said, you know, what's health care going to look like with AI? And let me tell you what it told us. It said, artificial intelligence in health care improves diagnostic accuracy, speeds up drug discovery, enhances operational efficiency, offering personalized patient care. This is what AI says about itself in health care. However, it introduces significant risks regarding data privacy, algorithmic bias, high implementation costs, and a potential lack of empathy in patient care. So, there's a real tension there between: are we removing that patient empathy as well? And so, challenges going forward are going to include regulatory, legal, and accountability issues when errors occur.

And that's what AI says about itself. So, it even acknowledges, if you will, and because it's based on things that have been written by humans, it acknowledges that there are some risks about this.

Thinking a little bit how AI is different from other medical technologies, and there were a couple of things I found. First, AI is interactive. The more we ask it, the more we put in there, the more it learns and the more it sort of specializes where we are. But what AI is basically saying, or where a lot of people think it's going to go, is that it will be able to do informed consent. It will be able to counsel patients at end of life.

Could you imagine a computer robot counseling you at the end of life? And so, what some people feel like is that compassion is no longer really depending on doctors to deliver compassion, that a computer will be able to also mirror compassion with that interaction that you have there.

And the second positive, I guess, is that it democratizes medical knowledge. We clearly know that our patients across the way are also doing their searches as well, so they really know what's going on. They don't necessarily know what everything means. They don't have the base knowledge that we have to reflect on it, but they do know a lot more than they used to know because of AI.

I live in the DC area, as you know, and out by Dulles Airport, and there are other places around here, where they've built these data centers, and it's a real topic for small-town USA across the country, is the energy consumption that AI is causing. And so, when you Google search something by AI, as I understand it, it uses about 10 times more energy than if you did a normal search. And these data centers are major energy drains to the point where it's really stressing, as you know, the electrical grid.

But I've also been thinking about the stress and strain of AI on our medical grid, the people who work there, the ability to keep up with the speed and the knowledge that's going on out there in health care, our own personal capacity to deal with it.

So, I actually sometimes feel a lot like one of those big data centers out there by Dulles Airport in the strain and the stress on my own energy flow as we go more and more into AI. Sure, it helps, but it also adds a lot to our overall stress and strain. I want to sort of finish around this issue around the physician role, because a lot of people are talking about that doctors will be replaced by AI, that a computer will be able to do our end-of-life care, et cetera.

And there was a nice article that was written by a guy named John Lantos, and it's called The Lost Aura of the Physician in the Age of AI, and I commend it to you. And talks a bunch about this concept of aura, and he first quotes a guy named Walter Benjamin, somebody I'd never heard of. He was a German philosopher from the turn of the last century, so long gone. But he basically expressed his feelings around art, that art, live art, not the produced print that's sitting there on my wall, but live art, actually has an aura about it.

It's unique and the original creation and the display of that, it captures a moment when that artist created it and then goes on to connect that artistic aura to a physician aura. Think about it.

You know how a family reacts when you, as a physician, walk into the room. It's a different mood automatically, even different than when you're on a televisit, right? When you're physically there with them, there is an aura that you have that you share with the family and everybody around it, and it's sort of grounded, what they call practical wisdom.

And it's the base knowledge that you have that you can integrate in the context of the human in front of you, and it's a little bit like that artistic moment. You're creating a relationship. You're creating something that's not some sort of computer-generated thing over here, but it's a real-time aura that you're sharing.

And the quote that he uses in this article is that telepresence is not the same as presence. And at least for now, I think the importance of our going in and sitting at the bedside, putting a hand on a hand, looking eye to eye in the same room with a human is the large part what it is to be human.

And I'm hoping that AI will never replace that. Yep, AI's here to stay. But here on Oncology Unscripted, at least we feel like we humans still have a very important role in the overall process, and fingers crossed that the current administration or PhRMA or somebody else or the payer won't write us off the formula and that we will be able to maintain that aura and share it with others.

So, keep that base knowledge going. Keep signing into Oncology Unscripted. Use your AI to make you more efficient. Don't burn out on the power grid, and keep doing good for others. John Marshall, Oncology Unscripted.

[13:14]

This transcript was generated by AI and lightly edited for clarity.