Read the Full Video Transcript

Alicia Morgans: Hi, I'm so excited to be here with Professor Felix Feng, who's a Professor of Radiation Oncology Medicine and Urology, as well as, being the Associate Director of the Cancer Center at UCSF. Thank you so much for being here with me.

Felix Feng: Alicia, thanks for having me here.

Alicia Morgans: Wonderful. Well, I wanted to talk with you, Felix, about the use of AI or Artificial Intelligence in prostate cancer care. I think this is really a technology that is disruptive in the best of ways and I'm just wondering if you can talk us through that a little bit.

Felix Feng: Absolutely. So first I'm going to tell you my bias in the sense that I'm a huge believer in Artificial Intelligence approaches. And the reason why is because I think this is a very transformative approach, not only for cancer management, but for all of medicine. And when we look at large advances in patient care, ultimately the largest advances, being able to take the most amount of information and personalize treatment for every individual patient. And that's where I think AI really has potential.

Alicia Morgans: Well, and along that line, AI is really these algorithms that are developed on large amounts of patient data that can inform themselves over time even more, as more data is added into that algorithm. Can you tell us a little bit about how it works? Because that's just a big concept, I think, for us to think through.

Felix Feng: Absolutely. And so I'm going to actually put it in the context of a specific example and one of my interests, which is, looking at pathology samples from patients who have prostate cancer. And so when a patient has prostate cancer, as you know, they undergo a biopsy. That biopsy gives a sample of the cancer that's then put on a glass slide. At that point in time, you can take a picture of that glass slide and then you can use Artificial Intelligence approaches to identify how aggressive a patient's cancer is or in certain cases, whether they need a certain treatment or they don't need a certain treatment and what the course of their medical care ideally should look like.

And the way it works is, AI basically infers information from whatever data you present at it. And again, I'm not an AI scientist, I'm a physician because I'm going to say it in kind of the terms that I understand it. But let's say when you look at a picture of a patient's cancer sample, every little speck on that cancer sample is a different data point that very sophisticated algorithms can then take a look at and try to identify what is really driving differences between, let's say, Mr. Smith, a random patient who has prostate cancer that may never need to be treated versus Mr. Jones a patient who really needs upfront treatment with heavily intensified treatment now.

And so the beauty of AI is that it tells us things about a patient's cancer that we as physicians can't identify ourselves, in the sense that we can't tell what all the specs mean. That the AI can look at millions and millions of specs on a single patient's sample and identify which of those specs are more important versus less important. And kind of how to theoretically personalize treatment decision based on those patterns.

Alicia Morgans: And how would those patterns necessarily be better than the clinical or pathologic features that we already use. I mean, we have these NCCN guideline risk categories that we use, D'Amico categories. Is this any better?

Felix Feng: I think it's going to be a lot better. And already we have data that suggests that it is markedly better. So I shouldn't say it's the future tense, I should actually talk about it in the present tense. And the reason why it's better is several fold. So in the most simplistic setting, if you take a single patient sample for prostate cancer and you show it to four pathologists, unfortunately there's oftentimes going to be slight differences in how that pathologist interprets that.

And the pathologists are looking at something called Gleason Score, which is a measure of aggressiveness of prostate cancer that's been established for many decades at this point in time. So number one, Gleason Scores vary a little bit depending on who reads the scan. And that's why, for example, a lot of patients will take their cancer samples and get second opinions from an outside pathologist who specialize only in prostate cancer.

But even when the pathologist looks at the slide, they actually are just looking at that particular area of cancer and they don't look at, let's say, the non-cancer cells that are outside of the cancer cells. And it turns out that there's a lot of studies now that suggest that this microenvironment, the tumor microenvironment, which is the areas that are adjacent to the cancer, actually play a role in terms of who has more aggressive disease versus less aggressive disease.

And it makes lot of sense because there are immune cells in that space, as well. The other thing is that if there's a hundred thousand cancer cells on a slide, are we confident that in the span of a minute when a pathologist looks at it, and again, these numbers aren't exactly accurate but the premise is true, are they going to be able to capture all those details?

And so if you take an AI that basically snapshots everything all at once and basically runs complex algorithms up in the cloud and can identify characteristics of every cell, characteristics broadly across the entire slide, characteristics of the non-tumor areas versus the cancer areas, then at some point in time you start understanding, "Well, why is the AI actually contributing additional value?"

And now we've done studies where we've taken data from a number of randomized clinical trials of patients with prostate cancer. We've applied AI algorithms to them and we've shown that the AI algorithm significantly outperforms standard clinical and pathologic staging. And so, I think the proof is in the pudding, to be honest with you.

Alicia Morgans: Yeah. Well, and you're a practicing radiation oncologist in addition to all the other things that you do. And when you're in clinic, are these tools that you might already consider using for your prostate cancer patient decision making?

Felix Feng: Yes, these tests are available in the clinic. And just for full transparency, I'm an advisor to the company Artera, which makes the first digital pathology AI tool that's available clinically for a prostate cancer prognosis. And when you think about how amazing these tools are, number one, they've been shown to add to beyond how we currently take a look at patients in terms of their disease aggressiveness and so forth. Number two, what we do know is that medicine is becoming more digital and that records are being digitized.

And in the digital world, theoretically what you can do is, if I have a patient in front of me in clinic, I might be able to order a AI test and literally while that patient's still in the room with me, the results may come back, right? And so that's the difference between, let's say, ordering a tissue based genomic test where we have to wait two to three weeks. And as you know, one of the attractive aspects for a instantaneous tool is our patients get really anxious if I say, "Listen, there's a genomic test I want to run on you, why don't you come back in two to three weeks and I'll have the results?" That's a little bit anxiety provoking.

And so if we can reduce patient anxiety with a quicker tool, that's important. I think that from an accessibility perspective, it's easier to get an AI tool on a picture than a genomic tool order. Let's say if there's somebody outside the United States in an International country, you can imagine it's hard to get a genomic tool sometimes because you have to ship the sample across International border. Whereas in a digital role, these technologies really might be much more accessible.

And I think there's work to be done in the sense that there's workflow issues that have to be addressed and we still kind of need to work those out. But if you're asking me, can you order it in clinic right now? Absolutely. Is it useful? Yes, because it's better than what we use in clinical and pathologic staging. And actually for that reason, this Artera AI test is now mentioned in NCCN guidelines with level one evidence in the most recent guidelines. And I think that's a testament to how promising this technology is looking in the sense that it's a relatively new technology, physicians are, in general, slow to embrace newer technologies, yet this one is gaining a lot of traction.

Alicia Morgans: So are you saying that we won't need pathologists at some point. That we'll replace them all with AI?

Felix Feng: No. I think the purpose of AI is to help physicians and pathologists along, not replace anybody. We're always going to need physicians and we're going to need people, real life people in the care of patients. And so what I would think about is this, okay, so you have a pathologist who looks at a slide who identifies the presence or absence of cancer.

And then on top of that you add AI that now tells you exactly how aggressive that cancer is, what is that patient's prognosis, the chance of having a recurrence or metastasis five or 10 years later, and also what treatment that patient would most benefit from. And so that's where the AI comes in. The pathologist right now doesn't tell you exactly what treatment the patient needs. The pathologist tells you that the cancer is there and the pathologist is needed to do that.

Alicia Morgans: Let's talk through a patient case. A patient that might come to see you, maybe an intermediate risk localized prostate cancer patient. And, of course, NCCN guidelines can sort of separate these patients into favorable and unfavorable intermediate risk. And generally, we recommend four to six months of ADT for this patient population when they're having radiation to treat their prostate. So can this test help us better make that treatment decision? Because sometimes, I have to be honest, it feels like overtreatment for some of these patients and nobody wants to have the side effects of ADT if they don't need to have them.

Felix Feng: So you bring up a great case to discuss in the sense that, as you know, my clinic is full of these patients. And to be honest with you, in the field of prostate cancer, we have a lot of evidence that hormone therapy benefits subsets of patients with more aggressive prostate cancer. And you're a leading expert in this space. You know that hormone therapy has a lot of side effects that patients don't want.

And so my clinic is filled with patients who want to try to figure out ways to avoid a treatment that wipes out sexual function, at least temporarily that may or may not cause increased cardiovascular risk in some patients in the future, that causes hot flashes and symptoms of menopause in men that can precipitate depression, anxiety, causes fatigue. And in a lot of patients it reduces their get up and go, can cause memory issues as well. And so hormone therapy is not a treatment without side effects.

And so my clinic is spent with a lot of patients where I say, "Listen, hormone therapy has been shown to benefit treatment in patients getting radiation for intermediate risk prostate cancer, I recommend it." And they say, "Doc, do I really have to take it?" And when we think about it, not only for intermediate risk prostate cancer but for almost all settings, when we have a trial that tells us, "Hey, we should use hormone therapy for, let's say, intermediate risk prostate cancer, the benefit in each of those trials is only five to 10%.

So what that means is that if you have a hundred patients who enroll on a clinical trial, usually only five to 10 of those hundred patients benefit from the intensification of therapy. And so like you said, what it means is that we are fundamentally over-treating a lot of patients. And so we've developed AI tools now that can actually accurately identify which intermediate risk prostate cancer patients do or do not need hormone therapy. And that's a very powerful tool.

If I'm a patient and I have to decide on whether I want to get treated with six months of hormone therapy where the effects might be 12 to 18-months long, I very much want to know whether I need to have those side effects. And listen, if I need to have hormone therapy to get cured, I want hormone therapy. But if I don't need hormone therapy and I have a equally good chance of being cured without hormone therapy, I really don't want those side effects.

And to some degree, as a physician, it's tough to be in clinic and to have to treat patients with a one size fits all approach. And that's never been my approach. And so I embrace the use of tools that can personalize therapy. And, for example, this AI tool would be one of those in this setting.

Alicia Morgans: Great. And I think, in general, we try to use the NCCN risk and other strategies to be more personalized but we can never be as personalized as this, actually, this tool allows us to be, which I think is pretty incredible.

Felix Feng: I agree with you and I think we are just scraping the tip of the iceberg at this point in time. And so the vision is that for any decision we need to make regarding a prostate cancer patient's treatment, that there is a AI tool there to help support whether that patient should get a therapy or shouldn't get a therapy. And so, right now, AI tools have been created as prognostic tools, meaning patients to identify, let's say the rate of recurrence or the rate of metastasis five or 10 years later.

And that's valuable information because that's information beyond what current clinical risk stratification approaches provide. But you've brought up the example of intermediate risk patients. Do they need hormone therapy or not? And, yes, there's an AI tool now validated that answers that question. For high risk prostate cancer patients, do they need two years versus six months of hormone therapy? There are tools that will be created to be able to address that.

For patients with metastatic prostate cancer, how much treatment intensification do they need? Can we identify the ones that need chemotherapy versus not? And there will be tools eventually created for that, as well. On the far opposite end of the spectrum, do men need any treatment for prostate cancer or not? And wouldn't it be wonderful if tools were created to identify who can just leave their prostate cancer alone for decades and not need any treatment? And I think that's the power of AI, in the sense that if we just wait a little bit, we're going to see an explosion in terms of the number of tools available to help personalize that treatment journey for our patients. And that is what gets me up in the morning.

Alicia Morgans: Absolutely. Well, that is the ultimate survivorship, isn't it? Really preventing exposure to treatments and therapies that are not going to benefit a patient, preventing those complications by just not using those treatments. I think that's a fantastic way for the field to move. And as you think about this, as a radiation oncologist who sees patients in his clinic every week, what would your bottom line be?

Felix Feng: My bottom line is that the way we approach medicine now is going to be very different than the way we approach medicine in two years from now. And the real difference is going to be the massive improvements in technology that allow us to personalize treatment decisions. And that personalization of treatment decisions will be who gets treated, any treatment versus no treatment, who gets hormone therapy versus no hormone therapy, in the advanced stages who gets next generation hormone therapy in chemotherapy versus not, in surgical patients who gets more treatment after surgery versus less treatment. And so I am so fortunate to be part of this field and to have the privilege of being able to do research with amazing collaborators, which I think will very much change the field to improve outcomes for our patients.

Alicia Morgans: So the time is now for the use of these kinds of strategies and certainly there is more to come in the future. I sincerely appreciate your time and your expertise.

Felix Feng: Thank you so much and, thanks again, for having me here.