Beyond the NCCN Localized Prostate Cancer Risk Category: The MMAI Prognostic Risk Stratification Model - Jonathan Tward

March 13, 2023

Jonathan Tward joins Alicia Morgans to discuss advancing prognostic capabilities in prostate cancer risk stratification using multimodal deep learning with digital histopathology within the context of a series of NRG Oncology phase III trials. The Artera AI prognostic biomarker model is based on five prospectively randomized trials of over 5,500 patients with low-risk prostate cancer on radiation therapy trials. Four of the trials were used to train this AI  biomarker, which is powerful due to its lack of bias and meticulous outcomes collection.  The Artera AI MMAI prognostic risk stratification model can further risk-stratify patients within a given NCCN risk category in localized prostate cancer.  This improved prognostic risk stratification can optimize treatment intensification decisions in localized disease. 


Jonathan Tward, MD, Ph.D., Professor, Department of Radiation Oncology, University of Utah, Salt Lake City, UT

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 Dr. Jonathan Tward, who is a radiation oncologist who's worked over some time with the Artera Prostate Test now integrated into the NCCN Guidelines that really helps us understand the prognostic expectations we should have for a given patient and how we can really help counsel patients more accurately. Thank you so much for being here.

Jonathan Tward: Oh, my pleasure. I appreciate the opportunity to chat with you about this.

Alicia Morgans: Well, it's wonderful to chat with you too, and I'd love for you to sort of set the stage for us. Tell me why do we need new ways to sort of have these prognostic principles around our patients? Isn't the NCCN enough?

Jonathan Tward: That's a great question. The NCCN risk groups were actually designed and built off of the risk groups first proposed by Anthony D'Amico in 1998. And actually, those risk groups were sort of designed around a biochemical failure endpoint. And that's a complicated endpoint in oncology because not all men who have biochemical failure actually require additional therapies. And so there's been a recognition that additional endpoints like metastasis or prostate cancer-specific mortality are probably more relevant endpoints when risk stratifying patients. And NCCN risk groups do a reasonable job of doing that, however, it's become clear in the more recent era that we can probably do better with our prognostic capability, which is why people have developed molecular tests, people are developing AI tests to try to get really a better prognosticator than something that was really designed to look at biochemical failure.

Alicia Morgans: And tell me a little bit more about the purpose of prognostication. Is this something that this treatment is going to work better than that treatment, or is it something a little more basic to help our patients understand what to expect?

Jonathan Tward: Yeah, prognostication is actually really important because it can inform decisions really on whether or not doing more or less things would be futile. Another way to say it is whether or not one should consider treatment intensification or one should consider treatment de-intensification. The way we sort of in oncology use clinical trials to determine if an intervention has an effect is we tend to use the hazard ratio, right? Drug A versus drug B hazard ratio might be 0.6 for the intervention, we say, isn't this great? It reduces the event by 40%. And that might be great if the event you're talking about happens in 60% of patients. Dropping it by 40% may be very important, especially on something like mortality. But what if the risk of something bad happening to you is only 1% or 2%? You might question whether or not you want to drop that 1% or 2% number down to relatively half to that by taking on a potentially toxic treatment or additional treatments that could really impact your quality of life.

So I think that prognostication is still really important in the doctor-patient discussion. And I know we talk a lot about shared decision-making, and we should, because patients can best understand their own goals. We try to interpret their goals, but when we can give them some real hard numbers about what is the actual probability this thing happens to you and what is the probability that that thing would be reduced by doing more, then we can really have informed choices rather than just broadly following guideline recommendations that are one-size-fits-all.

Alicia Morgans: That definitely makes a lot of sense and really I think is so patient-centric as we're trying to ensure that we match patients with the right intensity of treatment and de-escalate or escalate according to the risk that something bad is going to happen. So as I understand the Artera Prostate Test is an AI model that helps us understand a patient's prognosis. Can you tell me a little bit about how that sort of compares to the NCCN, which we've already acknowledged is something that's been around for a long time, has been helpful clinically, but maybe something on which we can improve?

Jonathan Tward: Yeah. One of the things that's somewhat unique about the Artera AI model is that it was actually based on prospectively randomized trials of many thousands of patients who actually enrolled with low-risk prostate cancer on radiation therapy trials, where the trials were asking questions such as should we or should we not add ADT might be the randomization question in other trials that were included, or maybe should we use short term versus low term. But the fact that you have prospectively randomized data that is used as the source for generating a prognosticator by itself is very powerful because it doesn't suffer from a lot of bias and it suffers from meticulous outcomes collection on the patient population. And even though the Artera AI model was trained on radiation patients, because it is a prognosticator, I think it still has relevance for any newly diagnosed patients because it is based on the biopsy tissue up front.

And there's still a lot of decisions that need to be made, not just by radiation oncologists, although that's currently where this test I would say is validated, but urologists as well. And when we face a patient for the first time with localized prostate cancer, we're considering should they be actively surveilled or should we treat them. And then if we decide that with the patient that treatment is warranted, a surgeon might be interested in thinking about what is the chance that they have lymph nodes. Maybe I should do an extended nodal dissection, which as of yet, this test doesn't specifically prognosticate that. But on the radiation side, we have to consider, do we use ADT. What duration of ADT? Do we add pelvic nodes?

So this AI does a much better job of prognosticating the risk of metastasis, prostate cancer-specific mortality, even biochemical recurrence than the NCCN risk groups does. And when you look at AUC, which is some measure of the quality of the prognosticator, for these endpoints, it tends to be up around the 0.75 to 0.8 area, which is actually extremely high for oncology. Most of our prognosticators don't really achieve that kind of level of prognostic ability.

Alicia Morgans: Well, that's impressive. And as we think about how it specifically kind of compares to the NCCN risk groups, which they are what they are, you put the data in, you get the risk group. Are there any differences? Might a patient be an NCCN intermediate risk and then by MMAI maybe be a lower risk patient? Does this ever happen?

Jonathan Tward: Yeah. I think my interpretation of this data is that we have a tendency in the radiation oncology community to probably overtreat a lot of patients with combined modality androgen deprivation therapy added to radiation therapy. Now, of course, it's with great intent, right? We're trying to help our patients. But if you look at patients with, for instance, NCCN low-risk disease, we know that those patients have such a low risk of developing metastasis or death after interventions that we often just surveil them. Or certainly, if we're going to provide a therapy, we're not going to be talking to men with low-risk or even unfavorable intermediate-risk disease, combined modality therapies.

Men on these trials, when you look at what was the risk of NCCN low-risk men of having METs, we sort of validated indeed they only have a 1% or 2% risk of METs. But we also found that 60% of the men on these RTOG trials, 6 out of 10 on trials designed to look at adding ADT or not to radiation therapy, which became the care standard because of these trials probably don't have a very significant reduction in their risk to warrant adding it. So in other words, even though the men on these trials, 1 in 10 of these men were considered NCCN low risk, and nobody would argue we shouldn't do anything but a monotherapy or watch, 6 out of 10 men on these randomized trials had the same risk of metastasis with the MMAI risk score implying that it would very much be likely overtreatment if we were to use combined modalities in them. So we can probably spare a huge percentage of men, I'll call it the indignity of adding ADT when the clinical benefit might be so low, too low for them to want to do it.

Alicia Morgans: Absolutely. And intermediate-risk disease, so a patient comes in, just as a clinical scenario, they are considering radiation in six months or short term of ADT. There is a likelihood that if they had this test that they may actually end up with MMAI low-risk disease that would really have such a low risk of recurrence, progression, metastasis that may not be worth it for those individuals.

Jonathan Tward: That's correct. When these trials were originally conceptualized and started, there wasn't really the concept yet of favorable versus unfavorable intermediate risk. So from the trial point of view, men were lumped into a category of intermediate risk, and that's why the NCCN Guideline for a long time until just a few years ago basically said plus or minus ADT for intermediate-risk patients. When the NCCN Guideline further stratified based on conventional clinical criteria into unfavorable and favorable intermediate risk, there was kind of a nod that this was a heterogeneous group. And the implication from NCCN is if you're unfavorable intermediate, you can probably withhold multimodality therapies. But even now, it says, "For unfavorable intermediate-risk patients that you should add combined modality."

Looking at it from the MMAI Artera risk score point of view, if you look at all the men who have intermediate risk, which is a conglomerate of favorable, unfavorable, but on these trials, a very healthy proportion presumably were unfavorable intermediate risk. It turns out that 83% of men in the NCCN intermediate risk category would be classified as an MMAI marker low, implying that they have no greater risk than an NCCN low-risk man of developing metastasis. So that's pretty substantial when something like 8 out of 10 men could consider not doing ADT when a guideline for most of those men would suggest that they should have it.

Alicia Morgans: Wow, that is quite a lot. And from your perspective, does this actually help us better as clinicians match patients with the right intensity of treatment?

Jonathan Tward: I think so. One of the challenges just thinking about this globally is when you consider risk groups or stages, they're really cohorts of persons, right? They're population metrics. And so it's a little bit challenging to take these risk scores that are now so personalized that actually give a personalized risk. And how do you deal with an individual person's information when our whole oncologic thinking has been stratified into groups? So I think the real power of what we're doing here is that yes, you can possibly help guide a clinician into conglomerating these individually precise risk scores into groups that might cue them to have the discussions with their patients that maybe we should or shouldn't intensify.

But I think what's really mostly interesting about the MMAI test is that it'll spit out a number for an individual that says, "We think your risk of metastasis following therapy is 6%." And then an individual patient, once they hear about what treatment intensification might look like, which might drive that number lower or de-intensification a little bit higher, can say, for me, a number needed to treat of 30 patients for one a benefit might not be sufficient for me to want to do the ADT. But for a different man, that number might be a hundred. And so I think there's sort of tremendous power in being able to not look at risk groups on hazard ratio reductions, but really reframe the conversation into what are the absolute risks of things happening to you, what are the absolute risk reductions, because then the light bulb will go off on both the physician's head and the patient's head to truly understand how much therapy they're willing to tolerate.

And what's kind of interesting when you poll men with localized prostate cancer about what are thresholds for absolute risk reductions at decreasing metastasis at which point the vast majority men would be willing to do more, about 80% of people, both experts and laypeople alike, will say, "For a 5% absolute difference, I'm willing to accept more therapy." But it's all over the map, those other 20% would say, "I won't do it unless there's a 20%," and some for 1%. So I'm sorry for the very convoluted long answer, but I think there's power in grouping to sort of cue physicians on when to have certain discussions. And at the same time, there's power in the precision because now the individual patient understands it at the individual level, what it means for them.

Alicia Morgans: But I think that's a great thing to talk about. And shared decision-making is all about taking the knowledge of medicine from the doctor, communicating that to the patient so the patient can contribute what am I willing to tolerate, what am I willing to pay essentially, and those kinds of discreet choices and experiments to get this benefit, is it worth it for me personally. And these are decisions that we have been informing with less information and maybe less accurate information for years. So I think that speaking about that and really understanding the context of this test in that decision-making framework is really, really important and I think can really benefit patient decision-making and clinician and patient satisfaction too. So if you had to sum it all up, what would your summary be of the Artera Prostate Test for prognostication?

Jonathan Tward: My summary is that this is an outstanding new tool based on level one evidence from numerous prospective randomized trials that in the most evidence-based way we know, with massive follow up can give a very high level of precision to an individual's risk of developing metastasis after certain kinds of therapies and that that would help that patient and their physician craft a right size therapy for them once they understand all the side effects and toxicities of their treatment choices. So a patient can go away, hopefully with a lot less decisional regret about what's going on, and in a lot of respects takes away some of the paternalism from the practice of medicine that we've been practicing for so long frankly, by just putting so much emphasis on guideline-directed care.

Guidelines are critically important but I do think that if you are a physician who never follows the guidelines, you're not a very good physician. But if you're a physician who always follows the guidelines, you may also not be a very good physician. So the more information to inform the discussion is a very powerful tool for patients and their families.

Alicia Morgans: Absolutely. And this tool is actually recommended on the NCCN Guidelines available to patients and to clinicians to help inform that treatment choice.

Jonathan Tward: That's correct. The Artera AI test was recently added to the NCCN Guideline as a test that's acceptable to use in the localized prostate cancer space. And it's my understanding that the company that can run this test has the ability to accept test orders, and I look forward to seeing how these results are integrated in the clinical practice.

Alicia Morgans: Wonderful. Well, thank you so much for your time and your expertise today.

Jonathan Tward: Thank you so much. I appreciate being here.