Multi-Modal Artificial Intelligence (MMAI) Prognostic Biomarker in High-Risk Prostate Cancer - Daniel Spratt

March 10, 2023

Alicia Morgans and Daniel Spratt discuss an AI-derived multi-modal digital pathology-based biomarker test designed towards improving risk stratification in high risk prostate cancer and then supporting personalized treatment decisions in these men. 

Daniel Spratt, MD, Chair, Department of Radiation Oncology, UH Cleveland Medical Center Professor, CWRU School of Medicine, Cleveland, OH

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 with Professor Dan Spratt, who's joining me today to talk about the Artera Prostate Test and the really exciting artificial intelligence model that's been developed to help us understand how to care for patients with high risk prostate cancer, among other things. So Dan, thank you so much for being here with me today.

Daniel Spratt: Thank you so much. It's a real pleasure.

Alicia Morgans: It's always a pleasure to talk to you. And in this case, I'd love to hear a little bit about the work that you and the team at Artera have done to develop this model and to help us understand how to best care for our patients.

Daniel Spratt: Absolutely. So I think as we've talked before, and I think clinicians are starting to realize, our standard clinical and pathologic tools, such as NCCN risk groups that tells us how to treat patients every day, unfortunately, we keep repeatedly showing it's just not very accurate. It has about moderate ability to tell us which men will have aggressive or lethal disease.

So there's this unmet need for us to have improved prognostic and predictive biomarkers. So there's been many different attempts for us to improve risk stratification, but one of the most exciting things with this collaboration with Artera is using artificial intelligence. And this is a big buzzword right now. But really, what this is it's a very novel method of taking patients' clinical information that we use every day, their pathology information, as well as immense digital imagery. So whenever a patient has a biopsy, that tissue is imaged. It can be digitally imaged and run through a deep learning model to extract features, things that maybe a pathologist can't even interpret.

And so when you put all this together, it's called the multimodal, because it's multiple forms of data, artificial intelligence, or the MMAI model. And there's been multiple of these models trained for different important clinical endpoints, distant metastasis, death from prostate cancer, and more to come. And they're all packaged in to this Artera Prostate Test.

So this test was trained and validated, and it's now published on five phase three randomized trials, which is unprecedented for a biomarker to be trained on such high quality data. And it was validated in these trials and showed that it was superior in prognostication and we call it discrimination of which patients will have these adverse clinical outcomes, and so superior to our NCCN risk groups, which, as you know, tells us how to treat patients. So our current work is trying to build upon that within a specific risk group such as high risk prostate cancer.

Alicia Morgans: Well, and to that end, really, I think you've updated this model over time with so many patients, and I'd love to hear if you can give us an estimate of how many patients, and really then are able to apply that model to additional data sets to see if we can learn even more and have it function in that high risk group.

Daniel Spratt: Absolutely. So what's really amazing is the first five phase three trials was over 5,000 patients. It was about 16 terabytes of imagery data. And we've built upon that. There's now been seven phase three randomized trials that have been reported to validate this biomarker.

And so the current work adds two more phase three trials in high risk prostate cancer of over a thousand patients when you combined all the high risk patients together. And very importantly, the follow-up in this cohort, these are very well done, multi-centered trials, and it's over 10 year median follow up in this cohort. So it's really something that's unprecedented to have this diversity of centers, hundreds and hundreds of centers across both Canada and the United States in an extremely diverse population.

As you may know, the NRG Oncology Group probably has the highest rate of African-American enrollment. It was about 20% in men with high-risk prostate cancer, versus many trial groups, it's only about 3%. So one of the strengths, especially when you're developing and training a model, is that you want to make sure that it's representative of the patients we see and we treat. And what this test is by having so many different centers, as you know, the way patients' biopsies are different at centers, the way their pathology slides are processed are different.

And that heterogeneity and diversity is really, I think, where artificial intelligence shines. When you and I go on the internet, we have very different probably patterns of clicking things, but that heterogeneity enables it to probably select things you want to see and select things I want to see. That's really what this model is able to do, is see the features that matter most to predict outcomes.

Alicia Morgans: And that's what's so important, I think, in high risk prostate cancer. When we think about radiation treatment at least, we're talking about long-term androgen deprivation therapy and, of course, combined with radiation to definitively treat the disease, hopefully cure a majority of our patients. But we do know that we're exposing many, many patients to a long period of androgen deprivation therapy, which has multiple complications and certainly is not the preference of any individual to really endure that.

So how can this model, how can the Artera Prostate Test help us in clinic when we're trying to have a treatment discussion with patients and think about, "How long should I be on ADT? Should I be on ADT at all?" Well, yes, if you have high-risk prostate cancer, but maybe not for everybody.

Daniel Spratt: Absolutely. So what we showed in this new independent validation in high-risk patients was it's pretty incredible that in men with high-risk disease, treated fairly similarly in these trials with long-term follow up, if you take, let's say the bottom quartile of scores, so let's say a more favorable score on the test versus a higher quartile score, a more unfavorable score, there was about a 30% difference, an absolute difference in patients' 15-year death from prostate cancer. But we're treating all those men the same today in clinic. That's crazy.

And so as you said, there's a lot of treatment decisions. Do we add nodal radiotherapy? Do we give higher doses of radiation? Do you add abiraterone now? Do you give longer term or shorter term hormones? For patients with a lower MMAI score, what you are going to see is that if, let's say you've got five to 10% risk of death from prostate cancer, how much can you improve that from further treatment intensification? You may be able to, but that might only be 1%, maybe 2%.

And we talk often to patients about, we call it the number needed to treat. You may have to treat 50 to a hundred men to prevent one from dying of prostate cancer. So you're exposing all those other men to the side effects you've mentioned, cardiovascular potential side effects, especially higher radiation doses, pelvic complications. But if you had that higher, more aggressive score on that Artera test, you may have a number needed to treat of five or 10. That's an incredible... That's a no-brainer that you would absolutely discuss intensifying treatment.

But probably even more importantly is instead of just making a one-size-fits-all approach for the patient sitting in front of you, when you know their comorbidities, you know their preferences, that you're going to be able to have that more accurate discussion, "How much is this likely to benefit you and the potential side effects we know that are well established?"

Alicia Morgans: Absolutely. So something that you can do in clinic right now to really help patients make a personalized choice about what's going to happen to them in their treatment to get the outcome that they hope for with as little treatment or with as appropriate treatment as is needed. And is this something that individuals can use in clinic right now if they want to?

Daniel Spratt: Yeah, absolutely. So it's a test. I encourage anyone, you can go to their Artera AI website. There's a portal, and patients can or providers can order the test. And I think what's really fascinating about this is that it's going to keep improving over time as more and more data, more and more prognostic and potentially predictive biomarkers are going to keep expanding into the space from Artera. And so the compendium really of what patients are going to be able to have, as I always say, go on to that website, understand the science, the evidence that's on there. And I think it's going to be something that I think more and more patients are going to be asking providers about. And I think it's going to be something that providers really need to understand not only how to order the test, but how to then communicate the benefits to patients.

Alicia Morgans: And I think it is on the NCCN guidelines at this point too, right?

Daniel Spratt: Yeah. So it was added in 2022 to NCCN guidelines. There's a criteria of level of evidence we use for biomarkers, a little different than for therapeutics or drugs from, it's called the Simon criteria. And so it got level one evidence from that initial validation work, because, again, it was in five randomized trials, and so it's only built further. And so, yeah. So it's in NCCN guidelines, and really excited to see the future.

Alicia Morgans: What are your key takeaways with this really exciting technology that is something that I think can ultimately be deployed for patients around the world to help inform their treatment decisions? We're talking today about high risk prostate cancer, but this is going to be something, I think, that can be applied to many clinical situations. So what is your key takeaway from this data?

Daniel Spratt: Yeah. So I think there's a couple points, and I think you just touched on them, is number one, as you know, we're expanding the number of therapeutic options and a lot of ongoing trials in this space. We really need tools to be able to personalize treatment for these men across the spectrum, but definitely in high risk prostate cancer. I think, two, we really need a scalable technology. And three, there's real advantages of a test that does not consume tissue, right? Because it enables that tissue to be retained and returned and may need to be used in the future. So I think I'm just very excited for the opportunity, A, to be part of this work, but, B, I know it's going to have immediate impact for patients and providers.

Alicia Morgans: Absolutely. And helping patients and providers have treatment decision making conversations that can be so personalized and can really help the patients get the best treatment for them is an advance that we are very, very happy to see. So thank you so much for your time and your expertise.

Daniel Spratt: Thank you so much.