Artera Multi-Modal AI Platform - Guiding Treatment Decisions in Men with Prostate Cancer - Andre Esteva & Felix Feng

March 22, 2023

Andre Esteva and Felix Feng join Alicia Morgans to discuss the Artera MMAI platform in prostate cancer intended to identify patients who will benefit from therapy intensification and help guide treatment decisions for men with high-risk localized prostate cancer. Artera has developed a unique multi-modal artificial intelligence (MMAI) algorithm composed of two pieces. One piece learns from imagery, the patient's digital pathology, and the other piece learns from a patient's clinical data. It fuses those two streams of information to be able to predict two things about the patient; first, their likeliest prognosis, and second, their response to a particular therapy. No tissue is consumed in carrying out this test, and it is intended to complement the way physician care is performed today.

Biographies:

Andre Esteva, PhD, Co-Founder & CEO, Artera

Felix Feng, MD, Vice Chair for Translational Research, Department of Radiation Oncology; Professor of Radiation Oncology, Urology, and Medicine, UCSF

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, I'm so excited to be here discussing the Artera platform in prostate cancer with two friends and colleagues.

Andre Esteva: My name's Andre Esteva, co-founder and CEO of Artera.

Felix Feng: I'm Felix Fang. I'm a professor of radiation oncology, urology and medicine at the University of California in San Francisco, and also an advisor to Artera.

Alicia Morgans: Wonderful. Well, thank you both for joining me today. I'd really love to hear from your perspective, Andre, how your clinical AI platform has been developed in prostate cancer. Can you tell us a little bit about what went into the building of this process?

Andre Esteva: The crux of AI is really about having the right data to power the learning behind the AI. Unlike traditional computer programs that are told line by line what to do, AI figures it out on its own by learning from data and learning to predict some task at hand, such as a patient's prognosis or predicting their response to therapy. So the way that we built Artera's multimodal AI, which learns from both pathology images and clinical data as well, was by leveraging Phase 3 clinical trials that were run over 15 to 20 years with long-term outcomes as well. And so data would be fed into the AI such as a patient's pathology and clinical data. The outcomes of that patient would also be fed into the AI and it would learn by looking at this data millions and millions of times, how to identify the features in images and in clinical data that would allow it to predict response of therapy and patient prognoses.

Alicia Morgans: So it sounds like there were many patients and their clinical data and their pathologic data that went into the building of this. Can you estimate how many? Hundred patients, 200 patients, some order of magnitude higher than that?

Andre Esteva: Our AI was developed with tens of thousands of patients, hundreds of thousands of pathology samples and millions of images. It is very robust and generalizable.

Alicia Morgans: Wonderful. Well, I'd love to hear from your perspective, Felix, as a clinician, we have risk stratification criteria. We have clinical features that we use in our practices all the time. We've been using them for years and years. What are the shortcomings of those factors and those processes that we do in clinic that might maybe benefit from some of the AI work that Artera's developed?

Felix Feng: It's a great question, Alicia. And so when we think about why is risk stratification valuable, it's because it gives us information that might be useful to patients and to physicians. And so thinking about what kind of information is useful, what I would say is what is a patient's chance of recurrence? What is a patient's chance of metastasis? What is a patient's chance, unfortunately of let's say dying from prostate cancer? Those are things that I want to know because knowing that allows me to potentially decide whether to intensify therapy or not intensify therapy.

And it turns out that the current risk stratification approaches based on clinical and pathologic features that are in national guidelines actually were developed quite a long time ago. And they're good, but when you look at their performance, they're actually not great. They're good. And so what these AI approaches allow us to do is potentially to become great at better identifying what are the chances that each individual patient will have a particular outcome. And I think knowledge is power, information is key in being able to provide this information to patients and to physicians is probably very, very, very important here.

Alicia Morgans: Well, I'm sure that that is very, very true. And I wonder, Andre, if you could share a little bit about the specifics of the prostate cancer NCCN recommended platform. What is this platform actually allowing clinicians to do?

Andre Esteva: Artera's multimodal AI has now been recommended by NCCN as you mentioned, Alicia. And what it does is there's a suite of tests that will tell you the prognosis, the likeliest outcomes of your patient. And depending on their NCCN risk group, it will give you a sense of their response to certain therapies. For instance, suppose you have an intermediate risk patient in front of you, they've been diagnosed and you are considering providing radiation as part of their primary localized treatment. We have a test that will tell you if they would benefit from hormone therapy in addition as an adjuvant to that radiation.

Alicia Morgans: So it sounds like that's really a bit of a predictive test that can predict whether someone's going to benefit from a given treatment as opposed to a prognostic test, which can help us understand whether a patient's going to have a good or a poor outcome regardless of treatment type. Is that true?

Andre Esteva: That's absolutely right. Our test includes both predictive and prognostic elements.

Alicia Morgans: Wow. Great. So Felix, from your perspective as a radiation oncologist, you are seeing patients in your clinic every week on a regular basis. Is this something that could be useful as you're trying to make decisions with patients whether or not they're going to use hormonal therapy, for example, with radiation?

Felix Feng: Oh, I think it's tremendously important. My clinic is full of patients who don't want hormone therapy, but oftentimes get their recommendation for hormone therapy because of potential aggressive features of their cancers. And what we know about hormone therapy is that unfortunately we over-treat a lot of patients. Let's say for every 10 patients we treat probably less than half, actually even less than that, probably only a third of those patients actually ever really need that hormone therapy. And so what the Artera AI test does is it actually better identifies who does, but also importantly, who does not need that hormone therapy.

And so when we think about why do we treat patients with hormone therapy, there have been randomized clinical trials in 2000 patients that have shown that if you add hormone therapy to radiation for intermediate risk prostate cancer patients, they live longer. That's very powerful. But when you look at the data, it's only five to 10% of those patients that actually really benefit. And so the Artera test, actually the main job with the Artera test is to identify the majority of patients who don't need that hormone therapy. And then these patients are spared from the side effects involving sexual function, cardiovascular risk, depression, anxiety, hot flashes and things that patients don't want to go through.

Alicia Morgans: Well, Andre, from your perspective, if you put yourself in the role of a patient or a loved one of a patient with prostate cancer, would you imagine that this might be a tool that would interest you if you were thinking about how to best take care of that cancer?

Andre Esteva: Absolutely, Alicia. There's this notion of finding the optimal therapy for a patient that maximizes the effectiveness of the therapy while minimizing its toxicity. You don't want to under-treat the patient and as a patient, you don't want to be undertreated. Your cancer will progress, but you also don't want to be over-treated and suffer the damaging side effects of say, hormone therapy and other therapies. And what we have found, which is relatively known in the community, is that the vast majority of patients are over-treated because you want to be conservative in your treatment of patients. And what we can offer through our tests is the ability to know definitively if a patient would benefit from a particular therapy or not.

Alicia Morgans: Well, thank you for that and that certainly makes sense from a patient perspective. Now, I know you're not a urologist, Dr. Feng, but I do know that you are a professor of urology in addition to being a radiation oncologist. So I'm just curious from your perspective, is this something that might be useful to urologists as well?

Felix Feng: I think it's going to be very valuable to urologists, and I think it's a reflection that medicine always advances that technology always advances, and that our job as physicians is to continually improve the care of our patients. And the reason why I think these AI tools are important is because number one, it tells us, let's say, which patients may or may not need more aggressive therapy. Number two, and this is quite important, these AI tests theoretically can evolve over time.

As you get more and more and more data, the AI learns from that data and continues to improve. And so the AI test that we have today will not be the exact same test theoretically in a few years from now and down the road as well. And that actually differs to some of the existing risk stratification tools in that breast cancer, the main test that's used to identify who does or does not need chemotherapy has been unchanged for three decades. And the optimist in me believes that a test that is three decades old is not as good as something that's based on the data that's been collected since those three decades.

Alicia Morgans: Absolutely. Continuously learning algorithm certainly makes a lot more sense, is quite promising. How does this approach to testing differ from currently available tests?

Andre Esteva: In the space of prostate cancer to be able to develop effective algorithms and classifiers to prognosticate and predict response to therapy, you need to run long-term trials that run over let's say 15 to 20 years. And it turns out that if you go back two decades and try to collect DNA and RNA out of old tissue samples, the vast majority will not pass quality control. But when you talk about developing an AI classifier that just has to look at a pathology slide that is formaldehyde fixed tissue, it turns out that preserves exceedingly well. And so as a result, AI has access to orders of magnitude more data than conventional genomic techniques. Its performance is superior. Additionally, it runs much quicker than genomics. Seconds of cloud compute will turn a sample into a test report and it doesn't consume the tissue like a genomics test would.

Felix Feng: Yeah, and just to add to what Andre has said, I think data is power. Information and more information is key. And I've spent my entire career trying to develop what we call predictive classifiers for cancer, meaning specifically identifying which patient benefits from a particular treatment. And to be honest with you, using other approaches, I haven't been very successful. And part of the reason is that we just can't get enough information. It turns out that you need large amounts of information to create predictive tests. And because AI can access that large amounts of information, I actually think that's why it's been successful in this particular regard. And that's important.

The other point I want to bring up is that part of my passion is making sure that as many patients have access to technology that can personalize medicine as possible. And I mean not only patients in the United States, but I mean patients internationally. And right now it's really hard to get a tissue sample sent from another country into the United States. There's just rules and regulations about that. But with a digital technology in a digital world, you don't have to do that. You can basically use a cloud in another country to run an AI test. And that's the power. We want to make sure that all patients get the best of care, not just patients in the United States, but patients beyond.

Alicia Morgans: So I think this is really exciting. And as a clinician, I'm really curious, is this something that I can order now?

Felix Feng: Absolutely, yes. I think that's the exciting part. Clinicians can now order this for their patients.

Alicia Morgans: So as we wrap up, I'd love to give each of you the opportunity to give us some final thoughts. Andre.

Andre Esteva: What we really hope to do at Artera is be able to positively impact patient lives. And our long term mission is to be able to globally personalize medical decision-making. The suite of tools that we've developed and begun to commercialize now in prostate are just the starting point. And in the long term, we plan to support the vast majority of cancers and expand into other disease sites.

Alicia Morgans: Well that is definitely exciting. And I wonder from your perspective, Felix?

Felix Feng: So as physicians, there's always earlier adopters of technology and later adopters with technology. What I think is AI is the ability to use things that we think didn't think would be possible until the future, now. We can identify who does or doesn't need hormone therapy. We might be able to identify in the future, let's say, who needs longer hormone therapy versus shorter hormone therapy, who needs chemotherapy versus no chemotherapy, who needs any treatment versus no treatment. And that's the power in the sense that the AI learns quickly, can be adapted to different situations and is versatile. And again, I think the key is the future may be now.

Alicia Morgans: Absolutely. Well certainly we have some excitement available to us in the clinic right now, but so much more to look forward to in addition to this. And I am excited to see where things go. So thank you both so much for your time today.

Andre Esteva: Thank you, Alicia.

Felix Feng: Thank you.