AI-derived Digital Pathology-Based Biomarkers in Localized Prostate Cancer - Felix Feng

March 1, 2022

Alicia Morgans is joined by Felix Feng to discuss an AI-derived digital pathology-based biomarker to predict the benefit of androgen deprivation therapy (ADT) in localized prostate cancer with validation in NRG/RTOG 9408. The current standard of care for men with intermediate- and high-risk localized prostate cancer treated with radiotherapy is the addition of ADT. This AI-derived predictive biomarker demonstrates that a majority of patients treated with radiotherapy on NRG/RTOG 9408 did not require ADT and could have avoided the associated costs and side effects of this treatment

Biographies:

Felix Y. Feng, MD, Vice Chair of Faculty Development, Director of Translational Research, Department of Radiation Oncology, Director of the Benioff Institute for Prostate Cancer Research at the University of California of San Francisco

Alicia Morgans, MD, MPH, Genitourinary Medical Oncologist, Medical Director of Survivorship Program at Dana-Farber Cancer Institute, Boston, Massachusetts


Read the Full Video Transcript

Alicia Morgans: Hi, my name is Alicia Morgans, and I'm a GU medical oncologist at Dana-Farber Cancer Institute. I am so excited to have here with me today a good friend and colleague, Dr. Felix Feng, who is a professor of radiation oncology medicine and urology and the associate director for translational research at the UCSF Cancer Center. Thank you so much for being here with me today,

Felix Feng: Alicia, thanks for having me.

Alicia Morgans: Wonderful. Well, I am really excited to talk with you about a presentation that Dan Spratt made at GU ASCO 2022. And I know that you have been involved with this company for a long time. So, of course, may be biased in some comments, but I would love your thoughts, as a radiation oncologist and someone who's enthusiastic, of course, about treating men with prostate cancer, I'd love to hear your comments on this presentation and how we might be able to predict whether or not ADT could help a person who's receiving radiation for his cancer.

Felix Feng: Yeah, so, great points, Alicia, and I got to tell you, I'm very high on AI. I'm a believer. And I've always been an earlier adopter of technology, so that's consistent with who I am. And as you mentioned, my disclosure here is I'm an advisor for the company. And so, therefore, I'm likely slightly biased, but it's because I love artificial intelligence, and I think it's really going to improve outcomes for patients with prostate cancer. And so the genesis of this idea occurred a while back, and honestly, as you know, there's a lot of efforts to try to personalize therapy for prostate cancer patients. It's clear that not all prostate cancers are the same. It's clear that some men need more therapy. Some men need less therapy. Some men need to figure out which of various options of therapy will best treat their disease.

And, in that context, genomics has seen a revolution, and we've had a lot of genomic tools now that are geared at trying to identify patients with more aggressive disease or less aggressive disease. But as you know, one of the issues with genomics is that it's clunky. And so, as a physician, if I have a patient shows up in my clinic, I have to order... Oh, I order a genomic test. We have to figure out where that tissue is, send it off to the company. That tissue gets consumed. DNA, RNA gets extracted out and two to three weeks later, we get results back. And the issue is the two to three weeks. So if you have a patient who's really concerned about prostate cancer, and you say, "Listen, give me three weeks and I'll get some information for you," it's a lot of anxiety. And we just don't like that anxiety.

And the other thing is that genomics isn't readily available outside the United States in the sense that there are a lot of countries that we run clinical trials in within NRG Oncology, which is a clinical trials group that I'm involved in, where it's hard to get at the genomic assays to initially places in Canada or abroad to the United Kingdom, Australia, even places like that. And so the question we had was could we come up with an approach to personalize therapy for prostate cancer patients that was performed extra extraordinarily well, gave patients and physicians the information they needed, and was scalable geographically across the board, accessible to patients, accessible to people who might not be able to get genomic tests and so forth? And that was the genesis of this AI approach.

And what Dan showed was that using data from clinical trials, we could develop a AI algorithm that takes a picture of a pathology slide, a standard pathology slide, and identifies, let's say in the context of patients with intermediate risk prostate cancer, who needs radiation alone or who needs radiation and hormone therapy. And as you know, hormone therapy has a lot of devastating effects to patients. A lot of patients don't want hormone therapy. And if you look at the trials showing the benefit of hormone therapy, let's say, an intermediate at risk prostate cancer, the majority of patients actually don't need hormone therapy, but we over treat a lot of patients to benefit some.

And so what Dan did was he collaborated with this company, Artera, and they developed and then validated this AI classifier, and it showed that two-thirds of patients with intermediate risk prostate cancer did not need hormone therapy. And when they validated in the trial, those two-thirds of patients literally had no benefit. And then the other one-third of patients had big benefit with their hazard ratio 0.33, which was quite substantial, as you and I both know. And so I think this is exciting. This is clearly a disruptive technology. This is different than what else is out there. There's a lot of advantages to it. It can be obtained quickly. It can be accessible broadly. It's, from a cost perspective, going to be a lot better than genomics, but more importantly, it's specifically predictive of response to hormone therapy. And that's, as you know, where I think the field needs to go.

Alicia Morgans: Absolutely. And I think that... I just want to recap, and you said this, but just to be very clear, there was an initial training set of multiple studies, where the algorithm was developed. And then there was a validation set that actually used the same trained algorithm applied to that set and demonstrated that hazard ratio that was strongly beneficial to patients who had the signature of interest, of course.

Felix Feng: Exactly.

Alicia Morgans: So it was designed appropriately, was validated, and is, I think, one of the first predictive biomarkers that we really have for men with prostate cancer. And I think especially in the intermediate prostate cancer set, we're only talking about four to six months of ADT. I would imagine that there's probably comfort among physicians in saying, "If you have this strong evidence, I actually feel comfortable for going a DT on this subset of patients," but I'd love to hear your thoughts on that.

Felix Feng: Exactly. And I'm sure your clinic is similar to mine. My clinic is full of prostate cancer patients who don't want hormone therapy, and they don't want hormone therapy because of the side effects of hormone therapy. And I, for the last decade, have told patients "Listen, there are randomized clinical trials demonstrating a small benefit to hormone therapy in this setting, and that is why I'm recommending the hormone therapy." And each patient then has to be faced with the question of what constitutes a small benefit versus a large benefit? Is a 10% benefit big or small to them, because that's roughly what it is? And we go through this agonizing process for the patient, where they're trying to choose between how much benefit they want to see in terms of outcomes in the side of effects and so forth. And here now this tool splits patients into two classes, and one class really has no benefit from hormone therapy. And the other one has a big benefit from hormone therapy. And as a physician, and I think you're the exact same way, I want the best care for my patients. And I don't want to deal with too many maybes, and there's always going to be maybes, but I want less maybes than more maybes.

Alicia Morgans: I agree with that. I completely agree. I think you also are in a unique position being a leader in NRG for the GU group, and I wonder how this technology may impact future trials or even potentially ongoing investigations in the NRG looking at this set of patients.

Felix Feng: Yeah. And so one of the great things about the NRG is my predecessors, who were leaders in the NRG, collected samples from all our phase three studies and banked them into one central repository. And so we're actually led by Dan and Osama Mohammed and Fu Tran and many others. We're actually applying this AI-based approach to 16 phase three trials in prostate cancer. And so it's not just going to be which patients with intermediate risk prostate cancer benefit from short term versus no hormone therapy. It's going to be which patients with high-risk prostate cancer benefit from years of hormone therapy versus months. It's going to be which patients, if they get surgery, need additional therapies versus no additional therapies after surgery. And so it's going to help radiation oncologists because we're going to try to develop classifiers of who should get bigger radiation field sizes versus less. And I think that's part of the excitement that this technology can be applied across many different settings in collaboration with NRG and even outside the field of prostate cancer.

Alicia Morgans: Absolutely. And I think one other benefit is that as trials are being designed and as they're moving forward to actually be enacted, we're going to be able to maybe shrink the sample size, because you can identify that highest-risk patient population where you would expect the rates of events to be so much higher, and you won't have to include those patients who aren't going to benefit, of course, number one, but who also will not be high enough risk to actually have events. So smaller sample size, fewer patient burdens, I think.

Felix Feng: Absolutely. I mean, I think this kind of approach will be integrated into probably the majority of clinical trials going forward, if we project long enough into the future, just because it's so accessible.

Alicia Morgans: Yeah. Well, wonderful. Well, we'd love to hear a closing statement to the viewers about this, as you said, disruptive, really exciting technology.

Felix Feng: So I spent a lot of time in Michigan at the University of Michigan and that's close to Canada, so I speak in hockey phrases of sorts. That's what I've been taught to do now. And so there's a phrase that you don't want... And probably I'm butchering this phrase, but you want to skate to where the puck is going, not where the puck is now. And that's actually always true, right? It's true for new therapies. You want to project in the future. It's also true for better ways to personalize care. And for us, I think that the future is about better personalization of therapy by identifying patients with higher risk or less risk of aggressive disease and also patients that specifically benefit from one therapy and not another. And that's where I think AI comes in. It's going to be... I think it's going to be very transformative.

Alicia Morgans: Well, I really appreciate you helping us, as a field, skate to where the puck is going to be. And I love the analogy, and I certainly appreciate your time and your expertise today.

Felix Feng: Thanks again for having me here.

email news signup