Precision Risk Assessment in Prostate Cancer: Growing Utility in NCCN Guidelines - Rashid Sayyid & Zachary Klaassen

April 3, 2024

Rashid Sayyid and Zach Klaassen discuss the 2024 updates to the NCCN Prostate Cancer Guidelines, highlighting the integration of the Decipher Genomic Classifier tests across various stages of prostate cancer management. This detailed conversation explores the progression from the 2023 guidelines, emphasizing the evolution of risk stratification tools from prognostic to the now predictive capabilities of the ArteraAI Prostate Test. The discussion meticulously outlines the importance of accurate risk stratification in optimizing patient care and the expanded role of Decipher across low, intermediate, and high-risk categories, as well as post-radical prostatectomy settings. This analysis underpins the significance of advanced genomic classifiers in refining treatment decisions, ultimately enhancing patient outcomes and reinforcing the shift towards personalized medicine in prostate cancer care.


Rashid Sayyid, MD, MSc, Urologic Oncology Fellow, Division of Urology, University of Toronto, Toronto, ON

Zachary Klaassen, MD, MSc, Urologic Oncologist, Assistant Professor Surgery/Urology at the Medical College of Georgia at Augusta University, Well Star MCG, Georgia Cancer Center, Augusta, GA

Read the Full Video Transcript

Rashid Sayyid: Hello everyone, and thank you for joining us today in this UroToday recording. I'm Rashid Sayyid, a urologic oncology fellow at the University of Toronto, along with Zach Klaassen, associate professor and program director at Wellstar MCG Health. We'll be discussing the 2024 key updates to the NCCN prostate cancer guidelines that were published on March 8, 2024. In this recording, we'll focus on the principles of risk stratification with a focus on the Decipher Genomic Classifier tests.

If we look back at the 2023 NCCN guidelines, we see that for risk stratification we had, at that time, only one table, and we'll see in later slides how this has evolved in 2024. If we look at Table 1 that was included in the 2023 guidelines, we'll see that the tools across various categories were all included in the same table, and this included clinical tools for risk stratification of patients with clinically localized disease, but also included AI tools such as the ArteraAI prostate test, gene expression testing, such as Decipher, POLARIS, etc., and germline testing, which also includes the HRR testing. If we notice that in the third column, none of these tests were considered predictive tests in this setting, whereas all of them were considered prognostic, and we'll see in later recordings how this has changed now for the ArteraAI Prostate Test, which is now considered the first and only test that is predictive in this setting.

Now, focusing on the Decipher test, we note that in 2023 Decipher had an NCCN level I evidence for validation for risk stratification for distant metastases. So, it's not that Decipher has not been approved back in 2023, but we'll see in 2024 how really this has been further elaborated on across different risk groups in order to better inform treatment decision-making.

Taking another step back, why is risk stratification important in the first place? Risk stratification, or accurate risk stratification, is the first step in providing our patients with optimized treatment decision-making and management, and it really helps address many different questions. These are only three examples of many possible questions that could be answered in this setting. First of all, how likely is a given cancer to be confined to the prostate or have spread to regional lymph nodes? Another potential question is, well, how likely is the cancer to progress and metastasize after treatment? And a third potential question is, well, how likely is adjuvant or salvage radiation to control cancer following radical prostatectomy? So, really, the more information we have in order to inform these questions and give answers, the better or the more empowered we are to provide our patients with better care.

And so, the NCCN has recognized it, and now if we look at Table 1, we notice that this table only includes selected clinical variables or models as opposed to including the genomic classifier tests and the AI tests. And so the NCCN has done a really nice job of looking at different disease states and really proposing what different models can be used across different settings. So, opposed to just lumping them all across clinically localized stage, now the NCCN is looking at the pre-treatment setting, localized setting, the different tools, D'Amico, NCCN, CAPRA, etc., the post-radical prostatectomy setting, I highlight the CAPRA-S model, among patients who recur after radical prostatectomy, what are different tools, and patients with metastatic castrate-sensitive prostate cancer, looking at the CHAARTED volume criteria, the number of bone mets, what are the different tools across all these settings?

But there are important limitations that come with using only clinical variables or models, and really, can we do better? Can we employ advanced tools? I think the NCCN really has acknowledged this very nicely in this Table 2, which is limited to select advanced tools for localized prostate cancer. And what we'll see here is that the gene expression tools, the AI pathology, and the germline tools make an appearance, and really, not only across all settings, but the NCCN does a really nice job of highlighting how the Decipher test, for example, works in the pre-treatment setting and also in the post-radical prostatectomy setting in order to act as a prognostic model to predict different outcomes, mainly metastasis-free survival.

And so, we'll look at this Table 3 and see how the NCCN, first of all, in the pre-treatment setting, really highlights the utility of Decipher across different NCCN risk populations. If we look at this table here, we see the NCCN low risk, there's a certain score, and really, it asks what question does the Decipher test help address? And then we have the NCCN intermediate risk and the NCCN high risk.

Now, if we focus on what are the specific questions that Decipher helps with, you'll notice with the NCCN low risk as opposed to just doing active surveillance for everyone, it really helps address, what is the intensity of follow-up for these patients? Should these patients have a confirmatory biopsy within 6 months or is this patient okay for a confirmatory biopsy 18 months after their initial diagnosis? It also helps address the question of, well, are there a select subgroup of patients with lower risk NCCN prostate cancer that should be considered for radical therapy as opposed to active surveillance? In the intermediate risk patients, should we just consider radiotherapy for these patients and spare them short-term ADT? Or can we identify those patients that are going to do worse and who may benefit from the addition of short-term ADT? Now, also in the NCCN high-risk subgroup, should everybody get long-term ADT, or is there a subgroup of patients who have a lower Decipher score that can be spared the long-term ADT in favor of short-term ADT and benefit from this less lengthy treatment, as well as the quality and side effects that come with that?

Let's talk about each NCCN risk group separately and let's talk about the evidence that informs the recommendations that have been made by the NCCN guidelines. So, where does the evidence for Decipher in the NCCN low-risk cohort come from? There's been a number of studies, but probably the most prominent study was that by Randy Vince and the group at Todd Morgan that was published in 2022, where they used data from the MUSIC registry and included 855 men who underwent Decipher biopsy testing. For all these patients, they looked at outcomes which included the time to treatment after active surveillance, so active surveillance continuation, and among those with lower risk NCCN who undergo treatment, what is their time to failure? And then they further looked at this by adjusting for different variables, assessing the association between high-risk Decipher scores and discontinuation of active surveillance as well as treatment failure. So, really, we're looking at two different outcomes here, very important clinical outcomes in this setting.

I'll take a second and just explain what a cumulative incidence function curve is. It's kind of like a reverse Kaplan-Meier curve. Not exactly. But as opposed to Kaplan-Meier curves where the curve starts with 100% and goes down, this is different, whereas events occur with these cumulative incidence function scores, we see that the curves go up as events occur. What's nice is that these curves also can have competing risks. Typically, when we talk about competing risks with prostate cancer, it's an elderly cohort, it's typically other-cause mortality. So it's a more robust analysis, and really, very informative in this setting.

And so, when we look at the Decipher score, we see that patients with a higher Decipher score, and this is the top curve, they're much more likely to discontinue active surveillance compared to those with lower Decipher scores. If we look at the median time to active surveillance discontinuation, in the high Decipher-score group, it's just over a year. Whereas if we look at the low and intermediate Decipher score, it's almost 3 years. So as opposed to your average high Decipher-score patient discontinuing active surveillance within a year, your low or intermediate score patient is going to discontinue, on average, at 3 years. So this is very informative when we tell patients, "Well, doc, is this active surveillance thing a short-term thing or are we trying to do this?" This is very important in counseling our patients.

And then the next question is, while you discontinued active surveillance, well, how likely are you to fail the treatment that you received? Again, we see that the high Decipher-score patients are much more likely to have treatment failure after receiving definitive therapy. So very informative data in the NCCN low-risk patients.

Next, we can think of cumulative incidence function curves as kind of a univariable analysis. When we look at this in the multivariable models adjusting for age, PSA, BMI, the volume of disease, etc., we see that still high-risk Decipher score was considered a significant variable predicting risk of progression to definitive therapy, so active surveillance or discontinuation, and also looking at the rate of treatment failure after radical therapy after these patients discontinued active surveillance. So again, a two-and-a-half-fold increased risk of discontinued active surveillance and a three-fold increased rate of treatment failure after radical therapy.

Based on this data, the NCCN took this into consideration and issued this statement, "Whereby more intensive active surveillance frequency should be considered for patients with NCCN low-risk disease and a high genomic classifier score given the higher probability of transitioning off active surveillance and subsequent progression." So very informative data with actionable recommendations for these patients.

Next, let's look at the NCCN intermediate-risk population. Here, the treatment decision it's aiding with is, should we just give radiotherapy or should we consider radiotherapy with short-term ADT? What is the evidence to inform this recommendation? And so, this data comes from Daniel Spratt and Felix Feng and the rest of the team who took 215 men with intermediate-risk prostate cancer from the NRG Oncology 0126 trial. In this trial, patients were randomized to 70 Gy versus 79 Gy of radiotherapy without ADT. They looked at the primary endpoint here of disease progression, which was a composite endpoint of biochemical failure, local failure, distant metastasis, prostate cancer-specific mortality, and use of salvage therapy. They looked at this primary endpoint both as a composite outcome and also individually.

And so, when we look at the Decipher score, and I want you to focus here on the right side, when we look at the composite primary endpoint of disease progression, we see that for every 0.1 unit increase in the genomic classifier score, there is a 12% increase in the rate of disease progression. So that's really important. It doesn't tell us that ADT works better in these patients, but what it does tell us is that patients who have a higher Decipher score have a progressively increased risk of adverse outcomes.

So if these patients with adverse outcomes are identified, well, these are probably the patients that we should consider more intensified treatment. We don't know it's going to work or not, but logic tells us that they are more likely to benefit from that. And then when they try to look at the individual outcomes, there's really a consistent pattern here whereby patients with a higher Decipher score have worse outcomes individually. So there's no one outcome that stands out, but it's a consistent, uniform worsening of outcomes for these patients. Again, it increases our confidence in these results and increases the fidelity, internal and external validity of these results. So very, very important data published by Dr. Spratt in 2023.

Based on this evidence, the NCCN issued this second statement, which states that "Radiotherapy alone should be considered for patients with a low genomic classifier score in NCCN intermediate-risk disease. Conversely, the addition of short-term ADT should be considered for patients with a high genomic classifier score given their increased risk of distant metastases and the significant benefit of short-term ADT on distant mets, even with de-escalated radiotherapy or brachytherapy use." So again, very clinically informative and actionable recommendation for patients. Consider a Decipher score when patients are on the fence about whether to add short-term ADT or not to radiotherapy.

Next, we'll discuss the NCCN high-risk patient and how Decipher really comes into play in deciding short versus long-term ADT in these patients. I'll turn it over to Zach who will go over the evidence informing this, as well as the evidence for Decipher in the post-radical prostatectomy setting.

Zach Klaassen: Thanks so much, Rashid. We've talked about the first two populations, and as Rashid mentioned, we're going to talk about the NCCN high-risk population. Again, this is informing treatment decisions between short-term or long-term ADT for these patients. The information that informs this decision came from Paul Nguyen and colleagues in 2023, and this was 265 men with high-risk disease in three NRG RTOG trials including 9202, 9413, and 9902. This study also had a pre-specified meta-analysis, which is what we'll be discussing today, and this was NRG-GU-TS006.

The primary objective of this meta-analysis was to validate the independent prognostic ability of Decipher for distant metastases, with secondary objectives to assess prostate cancer-specific mortality as well as overall survival. Looking at the multivariable analysis, and this is adjusted for age, PSA, clinical T-stage, and Gleason score, we see that across all outcomes, distant metastases, prostate cancer-specific mortality, and overall survival, an increase of 0.1 units on the genomic classifier score was statistically significant for assessing distant metastasis hazard ratio of 1.22, prostate mortality 1.23, and overall survival hazard ratio of 1.12, and this is in the adjusted model.

In these curves, these are, on the left, distant metastases, prostate cancer-specific mortality. These are cumulative incidence curves, as Rashid eloquently outlined how these work. And then on the right is the more classic Kaplan-Meier curve for overall survival. These color schemes correlate to a high genomic classifier of greater than 0.32, this is in red, intermediate risk, genomic classifier of 0.23 to 0.32 in gold, and low, blue, genomic classifier less than 0.23. And so, looking at these curves in turn for distant metastases, we see an excellent splitting of the curves with worse outcomes for high versus intermediate versus low. We see similar for prostate cancer-specific mortality in the middle, and again, for overall survival, we see great delineation in survival based on these three risk classifications for the genomic classifier.

This leads to NCCN evidence. Statement number three, that "Radiotherapy plus long-term ADT should be recommended for most patients with NCCN high-risk disease regardless of their GC score outside of a clinical trial, even with dose escalation, radiotherapy, or brachytherapy boost. However, patients with a genomic classifier low-risk score should be counseled that the absolute benefit of long-term over short-term ADT is smaller than for patients with genomic classifier high-risk scores, and when accounting for patient age, comorbidities, and patient preference, it may be reasonable with shared decision-making to use a duration shorter than long-term ADT.

Finally, this is the second new table also highlighting Decipher, as Rashid mentioned, post-radical prostatectomy biochemical recurrence. And so, when we focus on this, this is based on population four in this comprehensive assessment of Decipher. The score, you can see triage here is less than or greater than 0.6 on the genomic classifier scale and the treatment decision here for these patients is radiotherapy versus radiotherapy plus ADT in the post-RP BCR setting.

We'll talk about two trials that informed this decision. The first one was by Felix Feng and colleagues in JAMA Oncology in 2021. This was 352 men having undergone radical prostatectomy in the RTOG 9601 trial. This trial randomized patients to salvage radiotherapy plus or minus 2 years of bicalutamide, with an excellent median follow-up of 13 years. The primary objective of this trial was to validate the independent prognostic ability of Decipher for distant metastases and secondary objectives were to assess prostate cancer-specific mortality as well as overall survival.

What we see here, these are the curves, on the left, for distant metastases, prostate cancer-specific mortality, the cumulative incidence functions. On the right is overall survival and the Kaplan-Meier curve. The genomic classifier risk group is high for the black curves, intermediate for orange, and low for blue. We see across all of these outcomes a delineation for high, intermediate, and low risk for both distant metastases, prostate mortality, and overall survival, clear separation of the curves based on these three genomic classifier risk groups.

This looks at the difference between arms by Decipher risk group. And so when we think about this, on the left side, this is panel A, this is all patients looking at 12-year distant metastases on the left, prostate cancer-specific mortality in the middle, and overall survival on the right. The bars that are above the line show a benefit of bicalutamide. And so we see here that in patients with a low genomic risk group, there's a slight benefit for bicalutamide. But in the intermediate and high genomic classifier risk groups, we see a benefit across these three outcomes.

Where things get interesting is in the early salvage patients, so same delineation of distant metastases, prostate mortality, and overall survival. They defined early salvage as PSA less than 0.7. And so for the benefit of bicalutamide, very low for genomic classifier risk group for distant metastases, and we see a benefit for intermediate and high genomic classifier risk groups with regard to distant metastases. A similar-looking curve for prostate cancer-specific mortality. But where we see an overall survival, there's actually a detriment to giving bicalutamide. This bar is below the line of zero, and a slight benefit for overall survival for genomic classifier intermediate and high-risk groups.

This is the multivariable analysis for this paper. And we see here, again, adjusting for a variety of variables. You can see on the left for both distant metastases, hazard ratio of 1.17, prostate cancer-specific mortality, hazard ratio of 1.39, and overall survival, hazard ratio of 1.17. For each 0.1 increase in the unit in the genomic classifier score, we see prediction of these three outcomes in a statistically significant manner.

The second paper was from Alan Dal Pra and colleagues in Annals of Oncology in 2022. This was also looking at Decipher in the post-radical prostatectomy biochemical recurrent setting, but this was looking at men in the SAKK 09/10 trial. This was 226 men undergoing RP and then the intervention was salvage 64 Gy versus 70 Gy without hormonal therapy. The primary endpoint for this trial was biochemical progression, secondary endpoints were clinical progression and time to hormonal therapy.

This is the multivariable model looking for prediction of biochemical progression. The top is a continuous genomic classifier score. We see in the multivariable model a statistically significant sub-distribution hazard ratio of 1.14 for increasing genomic classifier score for predicting biochemical progression. And then the bottom is the genomic classifier risk group. So this is high versus low/intermediate, and we see the hazard ratio of 2.26, statistically significant for the assessment of biochemical progression with regards to high versus low/intermediate.

These are the cumulative incidence estimates for genomic classifier low/intermediate risk. This is less than 0.60 on the genomic classifier scale versus high. High is in red, low/intermediate is in blue. We see that for biochemical progression on the left, clinical progression in the middle, and hormonal treatment on the right, that there's a clear separation and statistically significant difference between these three outcomes for genomic risk high versus low/intermediate.

Looking further at prognostic performance of other risk groups for all endpoints, we see that for biochemical progression, clinical progression, rapid biochemical failure, so less than or equal to 18 months, hormonal treatment, metastasis, or metastasis-free survival, we see that the delineation for high versus low/intermediate genomic classifier risk group was statistically significant for all of these outcomes, with hazard ratios greater than 2.21 or higher for all outcomes. So this cut point of 0.6 is really showing us the ability to counsel our patients in terms of these outcomes, which are all clinically relevant outcomes.

This led to NCCN evidence as statement number four, "For patients with node-negative disease post-radical prostatectomy planned for early secondary radiotherapy with a PSA of less than or equal to 0.5 with genomic classifier low or intermediate risk, the use of radiotherapy alone should be considered. For patients planned for early secondary radiotherapy with a genomic classifier high-risk tumor, use of secondary radiotherapy plus ADT is recommended. Thirdly, currently, it is unclear how best to use the genomic classifier for patients receiving late post-radical prostatectomy secondary RT, and this is where PSA is greater than 0.5. Finally, the optimal ADT duration, such as 6 versus 24 months, based on the genomic classifier score is unknown at this time."

To summarize, the utilization of the Decipher Genomic Classifier now has indications across several disease spaces, including NCCN low, intermediate, and high-risk patients, as well as post-radical prostatectomy biochemical recurrence. I think the take-home from this discussion today is that it should provide clear evidence for each of these specific disease spaces as to when we use Decipher and how it can inform our patients. Arguably, we should be using it for all of these treatment decisions based on these four specific patient populations. Finally, advanced tools, such as the Decipher Genomic Classifier, provide further prognostication for the treatment of prostate cancer beyond the typical clinical variables and NCCN risk groups.

We thank you for your attention. We hope you enjoyed this UroToday discussion of the 2024 NCCN prostate cancer guidelines with a focus on risk stratification and highlighting the Decipher Genomic Classifier.