Read the Full Video Transcript

Alicia Morgans: Hi, I'm so excited to be here with Dr. Ashley Ross, an Associate Professor of Urology at Northwestern University. Thank you so much for being here with me.

Ashley Ross: Thanks for having me.

Alicia Morgans: Wonderful. I wanted to talk with you a little bit, Ashley, about the current clinical landscape and how we define risk in localized prostate cancer. What do you rely on?

Ashley Ross: Obviously, the risk of the patient's tumor and somewhat the patient's health dictates everything we do in localized prostate cancer. Can they be watched or do they need treatment? If they need treatment, how intense should that treatment be and with what type of regimen?

A long time ago, radiation oncologist, Dr. D'Amico had made the framework of what had eventually become our NCCN risk tools or AUA risk tools that were based on pathology, what the pathologist saw under the microscope and a couple other clinical features, what we felt with our finger on clinical exam, and with the PSA was. And when we look at the current NCCN risk stratification, those are based largely on those things. What's the Gleason grade group? What's your PSA? What's the digital rectal exam show?

Recently, in the updated guidelines for the beginning of this year, there's a risk stratification modeling that they did to say what's most supported, and they rated that based on category one evidence because we have lots of evidence looking at that clinical pathologic risk stratification. They also looked at two other features, genomics that we'll talk about in a little bit, and then also actually a new to the market multimodal digital pathology based AI, called Artera AI.

Alicia Morgans: Wonderful. Those old categories that we've used, they have been good, but are they good enough?

Ashley Ross: Yeah, they're definitely better than nothing, better than a coin flip, but they're not as individualized as there can be. Even among every NCCN risk categories, and as the year progressed D'Amico had three risk categories, and now you go from very low risk, to low risk, to favorable intermediate, to unfavorable intermediate, to high, and very high. Even with these six categories, there's still a spectrum in every category. If you want to personalize a treatment for that patient, even just based on prognostic risk, it can be sometimes challenging.

For example, surveillance now can be appropriate in some men with favorable intermediate risk prostate cancer. I think that's why that category was called out, but there's certainly men that are better and worse candidates for starting out their management with surveillance among that risk group. There's some men with high risk disease that you may want to treat more intensely, and there's some that probably behave more like unfavorable intermediate risks. How do we get more discrimination?

There was tools that have been developed in are continuing to develop, that can take us that one step forward and make it more personalized for that individual.

Alicia Morgans: What are some of these more developing tools that we can use?

Ashley Ross: I think over the last decade, probably the biggest thing in localized prostate cancer for risk stratification, was genomics, specifically in that call out table in the NCCN guidelines, whereas what's been linked to category one evidence, there's an NCCN risk stratification, and then the next thing that was linked, they said was the decipher genomic classifier. That is, much of our audience and you certainly know, is based on gene expression. They're taking the routinely stored tissue, they're extracting the RNA, looking at the gene expression levels, and then giving you this additional prognostic risk for how the patient's doing. That's been extremely well validated, tied to these phase three trials, and hence the designation by the NCCN that has support from level one evidence.

That's actually widely used at this point. And actually, I would say the trajectory is maybe up, but there are limitations to it. There's limitations in, you need to use some of the tissue that you have for it. That means there's certainly people that have limited amounts of cancer found on the biopsy, that may actually not pass quality assurance because there's just not enough RNA to extract. There's always new biomarkers and molecular markers being developed and there's sometimes a question like, if I use this tissue for one thing, am I going to have it available for the next thing? There's a limitation for limited tissue and now in limited tissue that you might want later. And then turnaround time, it's obviously some manual labor to extract, run the arrays, et cetera. Those all also bundle into the idea of cost and availability.

Now in the United States, I think that these things are on almost every payer plan. They're readily available. You can get them. Turnaround time is pretty decent, but it certainly was affected when COVID was flaring and stuff like this. But in other areas as well, the accessibility and dissemination of these things might be more challenging. Certainly, genomics I think has been transformative, in terms of risk prediction.
The other kind of really new on the market, but called out by the NCCN, also supported by category one evidence, is this Artera AI multimodal AI test that's run on clinical features and digital pathology. That's a very new thing. The evidence has just been presented over this last year and it's just come to clinical market where I can order it and I have ordered it. For me, in the last month or so, I've just started to get going with that.

Alicia Morgans: Tell me, how does that help get rid of some of the complications or some of the more challenging points related to even things like genomics?

Ashley Ross: Yeah, and just maybe to back up for a second, what the Artera AI is, it was a partnership between the NRG or radiation oncology groups and an AI machine learning company. And what they decided to do, is they said, "Look, we have all these patients with known outcomes from clinical trials, adjudicated outcomes, so we have more confidence in how they were followed, how those were reported. We have enough of these patients to allow for the computer to learn in an unsupervised way what's important." Their numbers were in the thousands for just these NRG collaborations.

Essentially, they looked at still things that some of those nomograms would look at the or the NCCN would look at: Clinical stage by T stage, PSA, Gleason score. The computer looked at that data and made the best model for that on the AI, and then it also looked at the digital pathology and made the best data from that, put it all together, and developed this AI score. Consistently across outcomes that are important to us, distant metastasis, prostate cancer specific survival, all these things, it outperformed those NCCN risk categories.
The advantage for this test is, obviously you don't sacrifice any tissue. You send the slide and you get that with the digital image. They just read the digital image and give you your results. You do not need to lose tissue. Obviously, that takes care of patients that have limited tissue for analysis, the guy who has 5% of four plus three equal seven, and you wonder is that risk different or not? You also have the advantage of as it gets more and more operationalized, a very rapid turnaround time. You can imagine a future where turnaround time becomes instantaneous.

As I mentioned, it's just launching, it really just came to market in my area maybe a month ago or so, and so it has to ramp up, but in its full form you're going to see that go away. And then finally, quite frankly, almost around the world, everybody is moving towards digitization of their pathology slides. In the same way that I never go down to the reading room and pull up a plain film, or an x-ray, or a CT scan that was on an old development film, I'm always pulling up my computer. Same thing is happening with pathology. You can imagine sharing the ability to share things, even in underserved areas, et cetera, just astronomically can be surmounted. The dissemination of this technology, I think is much easier than in other things.

It's not going to be affected by supply chain either, as I was mentioning. There's been new developments, at least from the literature done on it, of a powerful tool that's widely disseminatable, that is not subject to sacrificing any tissue or any of the supply chain problems that we have.

Alicia Morgans: Well, that's all really, really encouraging. And I know that, of course there's ongoing research and investigation because really what's available now in your clinic, I think is a starting point, with the expectation that there will be maybe multiple other ways that we can use these models, both prognostic and predictive, hopefully in the future, to help identify which treatments may be appropriate for which patients. Now, I know you presented data at SUO, that was a continuation of this research, and I wonder if you can share that?

Ashley Ross: Yeah, it's important. With all new technologies for us as researchers, obviously we get a large amount of excitement. I think seeing digital medicine and AI brought into the mainstream is phenomenal. As we go to use it in our patients, we want validation, validation, validation. At the SUO, I was happy to present a study done with collaboration between the NRG and Artera AI, where they wanted to look at two things: One, a completely independent phase three clinical trials. It had never been used for developments of models or anything like this. They just take a lock signature and run it on these folks and just see, does it still risk stratify? And secondarily, that study looked primarily at men that had high risk or very high risk prostate cancer. The question was there, or the secondary question was, could you still see discrimination among these men that you would say, well, we know their outcomes are going to be bad, so how much discrimination can you really have?

The Artera AI model worked and worked well. It both showed that it would validate in that series, had much better independent, accuracy for prognosis, and additionally, it still risk stratified in a fairly clean way who's going to be the really, really bad actors, even among guys that NCCN would classify as very high risk or high risk? It just builds on the sets of data, and it shows this nice continuation of using that trial data that we have that we all believe, that was carefully documented to elevate these technologies.

Alicia Morgans: That's great. Now, can you tell me a little bit about predictive versus prognostic biomarkers and when these might be most useful, in what situations and how do they come into play in the clinic?

Ashley Ross: Yeah, so obviously the patient wants to know not only what's the prognosis of their disease or how well are they going to do in general, but they also want to know can they pick a specific therapy that they're going to do better on? And obviously, when we think about this term therapeutic index, how much bang buck are you getting? How much can you avoid toxicities of treatment?

Now, people think about things that are it's nice in your mind to think specifically about something that's purely predictive. Meaning, if you have this thing, you know won't do worse than the guy who doesn't have that thing, but you will respond to this therapy much better than not. In the clinical realm, I don't know how much it matters if there's a little bit of overlap or not between predictive and prognostic. But what matters, is that there's groups like Artera AI and the NRG that are asking specific questions regarding therapy.

In the Artera AI test, for example, they asked can they develop an AI multimodal AI model to tell you for men with say, intermediate risk prostate cancer, will they benefit from antigen deprivation therapy or not? So specifically, not looking at who's more risky or not, and then maybe you'll add more treatment, but specifically if you look at that regimen versus radiation alone, who's going to benefit from the ADT? I think that they should be plotted for their efforts there in the design of those studies, and they do have a classifier that can say, you're not going to benefit because of the biomarker selection or you're going to benefit. And then obviously, will both save men morbidity of overtreatment, and then also will save men down the road who would've potentially been undertreated if we had decided on our own.

I think that design, where you're framing the question like, Hey, now we have the therapeutic choice, which choice is right? And the development of those markers that are predictive, that's a huge thing. That's what it's all about for the field. And as more data develops, that can be run on their AI platform for Artera AI or on other platforms that look at, again, genes and whatnot, we've seen great things using NexGen sequencing in advanced disease for DNA damage repair, et cetera. As we further on, that's the question we need to ask. Okay, we know that you need treatment, treatment with what? What's the ideal thing? That what personalized and individualized medicine really is.

Alicia Morgans: Well, I could not agree more. And as you think about using these tools in your clinical practice and you think about the utility in general, I wonder what your message would be?

Ashley Ross: We've come a long way. We've come a long way, in terms of developing new technologies and tools that can help us better risk stratify patients and individualize their care. Those are both on the prognostic area and in the predictive area. We've seen a lot done in genomics over the last decade, and that has helped a lot for both prognostic signatures and even some predictive data. That has some limitations however. And now we're seeing in a parallel way, development of things that use digitized pathology with all the clinical and pathological information we had to also develop these prognostic and predictive tools. It might be a way that can be better disseminated. It might be a way that can have lower cost. It might be a way that that can have faster turnaround time. And in my mind, I think it's a way that's going to perhaps also synergize complement or maybe be superior in some areas or not. And it's great to have these tools.

I would say nowadays, the use of these types of tools in my practice nears ubiquitous. Nearly all my patients are getting individualized, predictive, and prognostic tools leveraged for them, so that we can make the right decision for that patient. It's a great time to practice because of it.

Alicia Morgans: Well, at the end of the day, that's most important, getting the right treatment to the patient that needs it and avoiding overtreatment whenever we possibly can. Thank you so much for talking through Artera AI and the prostate test, as well as the landscape in general. I think we all feel a little bit better off for it. Thank you for your time.

Ashley Ross: Thank you.