JCO Precision Oncology Conversations – Détails, épisodes et analyse

In this JCO Precision Oncology Article Insights episode, Natalie DelRocco summarizes "Genomic Risk Classifiers in Localized Prostate Cancer: Precise but Not Standardized" by Góes et al. published on September 10, 2025.

TRANSCRIPT

Natalie DelRocco: Hello and welcome to JCO Precision Oncology Article Insights. I'm your host, Natalie DelRocco, and today we will be discussing the editorial "Genomic Risk Classifiers in Localized Prostate Cancer: Precise but Not Standardized."

This editorial by Góes, Li, and Chehrazi-Raffle, and Janopaul-Naylor et al. describes genomic risk classifiers, or GRCs, for patients with localized prostate cancer. Like any risk prediction model, GRCs are intended to help identify groups of patients that may benefit from less intense or more intense anticancer therapy. Risk prediction tools can be difficult to bring into clinical practice; they require a lot of validation. And as the authors describe, GRCs in localized prostate cancer are no exception.

The authors of this editorial contextualize an article by Janopaul-Naylor et al., which attempts to retrospectively explore the clinical use of three available GRCs for localized prostate cancer: Decipher, Oncotype DX, and Prolaris. Each of these three GRCs is being used in clinical practice currently.

In the original article, all three GRCs were associated with less intense therapy being prescribed in practice. However, the editorial authors note that this is likely selection bias due to the observational nature of the study design. It is conceivable that GRCs were more likely ordered to make decisions for patients who were already thought to be good candidates for less intensive therapy.

Another weakness of the retrospective study design is that patient level covariates known to be associated with clinical prognosis in localized prostate cancer, such as staging, Gleason score, prostate specific antigen, were unavailable. The authors note that sampling bias may also be an issue. Uninsured patients are not included in the original article, and therefore may impede the ability to make conclusions about the association of GRC use with income level.

The editorial authors highlight important study findings as well as these limitations, such as the heterogeneity of interventions following GRC result return. The Prolaris GRC was found to be associated with more surgical interventions, while the Decipher GRC was associated with more androgen deprivation therapy plus radiation. Additionally, patients with active surveillance were more likely to have a GRC in general ordered.

While these conclusions are very interesting, the editorial authors note that further exploration and validation, given the retrospective study design and limitations outlined, are needed to fully understand the impact of GRCs in the practice of treating localized prostate cancer.

Thank you for listening to JCO Precision Oncology Article Insights. Don't forget to give us a rating or a review and be sure to subscribe so that you never miss an episode. You can find all ASCO shows atasco.org/podcasts.

The purpose of this podcast is to educate and to inform. This is not a substitute for professional medical care and is not intended for use in the diagnosis or treatment of individual conditions. 

Guests on this podcast express their own opinions, experience, and conclusions. Guest statements on the podcast do not express the opinions of ASCO. The mention of any product, service, organization, activity, or therapy should not be construed as an ASCO endorsement.

Authors Drs. Jessica Ross and Alissa Cooper share insights into their JCO PO article, "Clinical and Pathologic Landscapes of Delta-Like Ligand 3 and Seizure-Related Homolog Protein 6 Expression in Neuroendocrine Carcinomas"  Host Dr. Rafeh Naqash and Drs. Ross and Cooper discuss the landscape of Delta-like ligand 3 (DLL3) and seizure-related homolog protein 6 (SEZ6) across NECs from eight different primary sites.

TRANSCRIPT

Dr. Rafeh Naqash: Hello and welcome to JCO Precision Oncology Conversations, where we bring you engaging conversations with authors of clinically relevant and highly significant JCO PO articles. I'm your host, Dr. Rafeh Naqash, podcast editor for JCO PO and an Associate Professor at the OU Health Stephenson Cancer Center.

Today, I'm excited to be joined by Dr. Jessica Ross, third-year medical oncology fellow at the Memorial Sloan Kettering Cancer Center, as well as Dr. Alissa Cooper, thoracic medical oncologist at the Dana-Farber Cancer Institute and instructor in medicine at Harvard Medical School. Both are first and last authors of the JCO Precision Oncology article entitled "Clinical and Pathologic Landscapes of Delta-like Ligand 3 and Seizure-Related Homolog Protein 6 or SEZ6 Protein Expression in Neuroendocrine Carcinomas."

At the time of this recording, our guest disclosures will be linked in the transcript.

Jessica and Alissa, welcome to our podcast, and thank you for joining us today.

Dr. Jessica Ross: Thanks very much for having us.

Dr. Alissa Cooper: Thank you. Excited to be here.

Dr. Rafeh Naqash: It's interesting, a couple of days before I decided to choose this article, one of my GI oncology colleagues actually asked me two questions. He said, "Rafeh, do you know how you define DLL3 positivity? And what is the status of DLL3 positivity in GI cancers, GI neuroendocrine carcinomas?" The first thing I looked up was this JCO article from Martin Wermke. You might have seen it as well, on obrixtamig, a phase 1 study, a DLL3 bi-specific T-cell engager. And they had some definitions there, and then this article came along, and I was really excited that it kind of fell right in place of trying to understand the IHC landscape of two very interesting targets.

Since we have a very broad and diverse audience, especially community oncologists, trainees, and of course academic clinicians and some people who are very interested in genomics, we'll try to make things easy to understand. So my first question for you, Jessica, is: what is DLL3 and SEZ6 and why are they important in neuroendocrine carcinomas?

Dr. Jessica Ross: Yeah, good question. So, DLL3, or delta-like ligand 3, is a protein that is expressed preferentially on the tumor cell surface of neuroendocrine carcinomas as opposed to normal tissue. It is a downstream target of ASCL1, and it's involved in neuroendocrine differentiation, and it's an appealing drug target because it is preferentially expressed on tumor cell surfaces. And so, it's a protein, and there are several drugs in development targeting this protein, and then Tarlatamab is an approved bi-specific T-cell engager for the treatment of extensive-stage small cell lung cancer in the second line. SEZ6, or seizure-like homolog protein 6, is a protein also expressed on neuroendocrine carcinoma cell surface. Interestingly, so it's expressed on neuronal cells, but its exact role in neuroendocrine carcinomas and oncogenesis is actually pretty poorly understood, but it was identified as an appealing drug target because, similarly to DLL3, it's preferentially expressed on the tumor cell surface. And so this has also emerged as an appealing drug target, and there are drugs in development, including antibody-drug conjugates, targeting this protein for that reason.

Dr. Alissa Cooper: Over the last 10 to 15 years or so, there's been an increasing focus on precision oncology, finding specific targets that actually drive the cancer to grow, not just within lung cancer but in multiple other primary cancers. But specifically, at least speaking from a thoracic oncology perspective, the field of non-small cell lung cancer has completely exploded over the past 15 years with the discovery of driver oncogenes and then matched targeted therapies.

Within the field of neuroendocrine carcinomas, including small cell lung cancer but also other high-grade neuroendocrine carcinomas, there has not been the same sort of progress in terms of identifying targets with matched therapies. And up until recently, we've sort of been treating these neuroendocrine malignancies kind of as a monolithic disease process. And so recently, there's been sort of an explosion of research across the country and multiple laboratories, multiple people converging on the same open questions about why might patients with specific tumor biologies have different kind of responses to different therapies.

And so first this came from, you know, why some patients might have a good response to chemo and immunotherapy, which is the first-line approved therapy for small cell lung cancer, and we also sort of extrapolate that to other high-grade neuroendocrine carcinomas. What's the characteristic of that tumor biology? And at the same time, what are other targets that might be identifiable? Just as Jesse was saying, they're expressed on the cell surface, they're not necessarily expressed in normal tissue. Might this be a strategy to sort of move forward and create smarter therapies for our patients and therefore move really into a personalized era for treatment for each patient? And that's really driving, I think, a lot of the synthesis of this work of not only the development of multiple new therapies, but really understanding which tumor might be the best fit for which therapy.

Dr. Rafeh Naqash: Thank you for that explanation, Alissa. And as you mentioned, these are emerging targets, some more further along in the process with approved drugs, especially Tarlatamab. And obviously, DLL3 was something identified several years back, but drug development does take time, and readout for clinical trials takes time.

Could you, for the sake of our audience, try to talk briefly about the excitement around Tarlatamab in small cell lung cancer, especially data that has led to the FDA approval in the last year, year and a half?

Dr. Alissa Cooper: Sure. Yeah, it's really been an explosion of excitement over, as you're saying, the last couple of years, and work really led by our mentor, Charlie Rudin, had identified DLL3 as an exciting target for small cell lung cancer specifically but also potentially other high-grade neuroendocrine malignancies.

Tarlatamab is a DLL3-targeting bi-specific T-cell engager, which targets DLL3 on the small cell lung cancer cells as well as CD3 on T cells. And the idea is to sort of introduce the cancer to the immune system, circumventing the need for MHC class antigen presentation, which that machinery is typically not functional in small cell lung cancer, and so really allowing for an immunomodulatory response, which had not really been possible for most patients with small cell lung cancer prior to this.

Tarlatamab was tested in a phase 2 registrational trial of about 100 patients and demonstrated a response rate of 40%, which was very exciting, especially compared with other standard therapies which were available for small cell lung cancer, which are typically cytotoxic therapies. But most excitingly, more than even the response rate, I think, in our minds was the durability of response. So patients whose disease did have a response to Tarlatamab could potentially have a durable response lasting a number of months or even over a year, which had previously not ever been seen in this in the relapsed/refractory setting for these patients.

I think the challenge with small cell lung cancer and other high-grade neuroendocrine malignancies is that a response to therapy might be a bit easier to achieve, but it's that durability. The patient's tumors really come roaring back quite aggressively pretty quickly. And so this was sort of the most exciting prospect is that durability of response, that long potential overall survival tail of the curve really being lifted up.

And then most recently at ASCO this year, Dr. Rudin presented the phase 3 randomized controlled trial which compared Tarlatamab to physician's choice of chemotherapy in a global study. And the choice of chemotherapy did vary depending on the part of the world that the patients were enrolled in, but in general, it was a really markedly positive study for response rate, for progression-free survival, and for overall survival. Really exciting results which really cemented Tarlatamab's place as the standard second-line therapy for patients with small cell lung cancer whose disease has progressed on first-line chemo-immunotherapy. So that has been very exciting. This drug was FDA approved in May of 2024, and so has been used extensively since then. I think the adoption has been pretty widespread, at least in the US, but now in this global trial that was just presented, and there was a corresponding New England Journal paper, I think really confirms that this is something we really hopefully can offer to most of our patients.

And I think, as we all know, that this therapy or other therapies like it are also being tested potentially in the first-line setting. So there was data presented with Tarlatamab incorporated into the maintenance setting, which also showed exciting results, albeit in a phase 1 trial, but longer overall survival than we're used to seeing in this patient population. And we await results of the study that is incorporating Tarlatamab into the induction phase with chemotherapy as well. So all of this is extraordinarily exciting for our patients to sort of move the needle of how many patients we can keep alive, feeling functional, feeling well, for as long as possible.

Dr. Rafeh Naqash: Very exciting session at ASCO. I was luckily one of the co-chairs for the session that Dr. Rudin presented it, and I remember somebody mentioning there was more progress seen in that session for small cell lung cancer than the last 30, 35 years for small cell, very exciting space and time to be in as far as small cell lung cancer.

Now going to this project, Jessica, since you're the first author and Alissa's the last, I'm assuming there was a background conversation that you had with Alissa before you embarked on this project as an idea. So could you, again, for other trainees who are interested in doing research, and it's never easy to do research as a resident and a fellow when you have certain added responsibilities. Could you give us a little bit of a background on how this started and why you wanted to look at this question?

Dr. Jessica Ross: Yeah, sure. So, as with many exciting research concepts, I think a lot of them are derived from the clinic. And so I think Alissa and I both see a good number of patients with small cell, large cell lung cancer, and then high-grade neuroendocrine carcinomas. And so I think this was really born out of a basic conversation of we have these drugs in development targeting these two proteins, DLL3 and SEZ6, but really what is the landscape of cancers that express these proteins and who are the patients that really might benefit from these exciting new therapies. And of course, there was some data out there, but sort of less than one would imagine in terms of, you know, neuroendocrine carcinomas can really come from anywhere in the body. And so when you're seeing a patient with small cell of the cervix, for example, like what are the chances that their cancer expresses DLL3 or expresses SEZ6? So it was really derived from this pragmatic, clinically oriented question that we had both found ourselves thinking about, and we were lucky enough at MSK, we had started systematically staining patients' tumors for DLL3, tumors that are high-grade neuroendocrine carcinomas, and then we had also more recently started staining for SEZ6 as well. And so we had this nice prospectively collected dataset with which to answer this question.

Dr. Rafeh Naqash: Excellent. And Alissa, could you try to go into some of the details around which patients you chose, how many patients, what was the approach that you selected to collect the data for this project?

Dr. Alissa Cooper: This is perhaps a strength but also maybe a limitation of this dataset is, as Jesse alluded to, our pathology colleagues are really the stars of this paper here because we were lucky enough at MSK that they were really forethinking. They are absolute experts in the field and really forward-thinking people in terms of what information might be needed in the future to drive treatment decision-making.

And so, as Jesse had said, small cell lung cancer tumor samples reflexively are stained for DLL3 and SEZ6 at MSK if there's enough tumor tissue. The other high-grade neuroendocrine carcinomas, those stains are performed upon physician request. And so that is a bit of a mixed bag in terms of the tumor samples we were able to include in this dataset because, you know, upon physician request depends on a number of factors, but actually at MSK, a number of physicians were requesting these stains to be done on their patients with high-grade neuroendocrine cancers of of other histologies. So we looked at all tumor samples with a diagnosis of high-grade neuroendocrine carcinoma of any histology that were stained for these two stains of interest. You know, I can let Jesse talk a bit more about the methodology. She was really the driver of this project.

Dr. Jessica Ross: Yeah, sure. So we had 124 tumor samples total. All of those were stained for DLL3, and then a little less than half, 53, were stained for SEZ6. As Alissa said, they were from any primary site. So about half of them were of lung origin, that was the most common primary site, but we included GI tract, head and neck, GU, GYN, even a few tumors of unknown origin. And again, that's because I think a lot of these trials are basket trials that are including different high-grade neuroendocrine carcinomas no matter the primary site. And so we really felt like it was important to be more comprehensive and inclusive in this study.

And then, methodologically, we also defined positivity in terms of staining of these two proteins as anything greater than or equal to 1% staining. There's really not a defined consensus of positivity when it comes to these two novel targets and staining for these two proteins. But in the Tarlatamab trials, for some of the correlative work that's been done, they use that 1% cutoff, and we just felt like being consistent with that and also using a sort of more pragmatic yes/no cutoff would be more helpful for this analysis.

Dr. Alissa Cooper: And that was a point of discussion, actually. We had contemplated multiple different schemas, actually, for how to define thresholds of positivity. And I know you brought up that question before, what does it mean to be DLL3 positive or DLL3 high? I think you were alluding to prior that there was a presentation of obrixtamig looking at extra-pulmonary neuroendocrine carcinomas, and they actually divvied up the results between DLL3 50% or greater versus DLL3 low under 50%. And they actually did demonstrate differential efficacy certainly, but also some differential safety as well, which is very provocative and that kind of analysis has not been presented for other novel therapies as far as I'm aware. I could be wrong, but as far as I'm aware, that was sort of the first time that we saw a systematic presentation of considering patients to be, quote unquote, "high" or "low" in these sort of novel targets.

I think it is important because the label for Tarlatamab does not require any DLL3 expression at all, actually. So it's not hinging upon DLL3 expression. They depend on the fact that the vast majority of small cell lung cancer tumors do express DLL3, 85% to 90% is what's been demonstrated in a few studies. And so, there's not prerequisite testing needed in that regard, but maybe for these extra-pulmonary, other histology neuroendocrine carcinomas, maybe it does matter to some degree.

Dr. Rafeh Naqash: Definitely agree that this evolving landscape of trying to understand whether an expression for something actually really does correlate with, whether it's an immune cell engager or an antibody-drug conjugate is a very evolving and dynamically moving space. And one of the questions that I was discussing with one of my friends was whether IHC positivity and the level of IHC positivity, as you've shown in one of those plots where you have double positive here on the right upper corner, you have the double negative towards the left lower, whether that somehow determines mRNA expression for DLL3. Obviously, that was not the question here that you were looking at, but it does kind of bring into question certain other aspects of correlations, expression versus IHC. Now going to the figures in this manuscript, very nicely done figures, very easy to understand because I've done the podcast for quite a bit now, and usually what I try to do first is go through the figures before I read the text, and and a lot of times it's hard to understand the figures without reading the text, but in your case, specifically the figures were very, very well done. Could you give us an overview, a quick overview of some of the important results, Jessica, as far as what you've highlighted in the manuscript?

Dr. Jessica Ross: Sure. So I think the key takeaway is that, of the tumors in our cohort, the majority were positive for DLL3 and positive for SEZ6. So about 80% of them were positive for DLL3 and 80% were positive for SEZ6. About half of the tumors were stained for both proteins, and about 65% of those were positive as well. So I think if there's sort of one major takeaway, it's that when you're seeing a patient with a high-grade neuroendocrine carcinoma, the odds are that their tumor will express both of these proteins. And so that can sort of get your head thinking about what therapies they might be eligible for.

And then we also did an analysis of some populations of interest. So for example, we know that non-neuroendocrine pathologies can transform into neuroendocrine tumors. And so we specifically looked at that subset of patients with transformed tumors, and those were also- the majority of them were positive, about three-quarters of them were positive for both of these two proteins. We looked at patients with brain met samples, again, about 70% were positive.

And then I'd say the last sort of population of interest was we had a subset of 10 patients who had serial biopsies stained for either DLL3 or SEZ6 or both. In between the two samples, these patients were treated with chemotherapy. They were not treated with targeted therapy, but interestingly, in the majority of cases, the testing results were concordant, meaning if it was DLL3 positive to begin with, it tended to remain DLL3 positive after treatment. And so I think that's important as well as we think about, you know, a patient who maybe had DLL3 testing done before they received their induction chemo-IO, we can somewhat confidently say that they're probably still DLL3 positive after that treatment.

And then finally, we did do a survival analysis among specifically the patients with lung neuroendocrine carcinomas. We looked at whether DLL3 expression affected progression-free survival on first-line platinum-etoposide, and then we looked at did it affect overall survival. And we found that it did not have an impact or the median progression-free survival was similar whether you were DLL3 positive or negative. But interestingly, with overall survival, we found that DLL3 positivity actually correlated with slightly improved overall survival. These were small numbers, and so, you know, I think we have to interpret this with caution, for sure, but it is interesting. I think there may be something to the fact that five of the patients who were DLL3 positive were treated with DLL3-targeting treatments. And so this made me think of, like in the breast cancer world, for example, if you have a patient with HER2-positive disease, it initially portended worse prognosis, more aggressive disease biology, but on the other hand, it opens the door for targeted treatments that actually now, at least with HER2-positive breast cancer, are associated with improved outcomes. And so I think that's one finding of interest as well.

Dr. Rafeh Naqash: Definitely proof-of-concept findings here that you guys have in the manuscript.

Alissa, if I may ask you, what is the next important step for a project like this in your mind?

Dr. Alissa Cooper: Jesse has highlighted a couple of key findings that we hope to move forward with future investigative studies, not necessarily in a real-world setting, but maybe even in clinical trial settings or in collaboration with sponsors. Are these biomarkers predictive? Are they prognostic? You know, those are still- we have some nascent data, data has been brewing, but I think that we we still don't have the answers to those open questions, which I think are critically important for determining not only clinical treatment decision-making, but also our ability to understand sequencing of therapies, prioritization of therapies.

I think a prospective, forward-looking project, piggybacking on that paired biopsy, you know, we had a very small subset of patients with paired biopsies, but a larger subset or cohort looking at paired biopsies where we can see is there evolution of these IHC expression, even mRNA expression, as you're saying, is there differential there? Are there selection pressures to targeted therapies? Is there upregulation or downregulation of targets in response not just to chemotherapy, but for example, for other sort of ADCs or bi-specific T-cell engagers? I think those are going to be critically important future studies which are going to be a bit challenging to do, but really important to figure out this key clinical question of sequencing, which we're all contemplating in our clinics day in and day out. If you have a patient, and these patients often can be sick quite quickly, they might have one shot of what's the next treatment that you're going to pick. We can't guarantee that every patient is going to get to see every therapy. How can you help to sort of answer the question of like what should you offer? So I think that's the key question sort of underlying any future work is how predictive or prognostic are these biomarkers? What translational or correlative studies can we do on the tissue to understand clinical treatment decision-making? I think those are the key things that will unfold in the next couple of years.

Dr. Rafeh Naqash: The last question for you, Alissa, that I have is, you are fairly early in your career, and you've accomplished quite a lot. One of the most important things that comes out from this manuscript is your mentorship for somebody who is a fellow and who led this project. For other junior investigators, early-career investigators, how did you do this? How did you manage to do this, and how did you mentor Jessica on this project with some of the lessons that you learned along the way, the good and other things that would perhaps help other listeners as they try to mentor residents, trainees, which is one of the important things of what we do in our daily routine?

Dr. Alissa Cooper: I appreciate you calling me accomplished. Um, I'm not sure how true that is, but I appreciate that. I didn't have to do a whole lot with this project because Jesse is an extraordinarily smart, driven, talented fellow who came up with a lot of the clinical questions and a lot of the research questions as well. And so this project was definitely a collaborative project on both of our ends.

But I think what was helpful from both of our perspectives is from my perspective, I could kind of see that this was a gap in the literature that really, I think, from my work leading clinical trials and from treating patients with these kinds of cancers that I really hoped to answer. And so when I came to Jessica with this idea as sort of a project to complete, she was very eager to take it and run with it and also make it her own.

You know, in terms of early mentorship, I have to admit this was the first project that I mentored, so it was a great learning experience for me as well because as an early-career clinician and researcher, you're used to having someone else looking over your shoulder to tell you, "Yes, this is a good journal target, here's what we can anticipate reviewers are going to say, here are other key collaborators we should include." Those kind of things about a project that don't always occur to you as you're sort of first starting out. And so all of that experience for me to be identifying those more upper-level management sort of questions was a really good learning experience for me. And of course, I was fantastically lucky to have a partner in Jesse, who is just a rising star.

Dr. Jessica Ross: Thank you.

Dr. Rafeh Naqash: Well, excellent. It sounds like the first of many other mentorship opportunities to come for you, Alissa. And Jessica, congratulations on your next step of joining and being faculty, hopefully, where you're training.

Thank you again, both of you. This was very insightful. I definitely learned a lot after I reviewed the manuscript and read the manuscript. Hopefully, our listeners will feel the same. Perhaps we'll have more of your work being published in JCO PO subsequently.

Dr. Alissa Cooper: Hope so. Thank you very much for the opportunity to chat today.

Dr. Jessica Ross: Yes, thank you. This was great.

Dr. Rafeh Naqash: Thank you for listening to JCO Precision Oncology Conversations. Don't forget to give us a rating or review and be sure to subscribe so as you never miss an episode. You can find all ASCO shows at asco.org/podcasts.

The purpose of this podcast is to educate and to inform. This is not a substitute for professional medical care and is not intended for use in the diagnosis or treatment of individual conditions.

Guests on this podcast express their own opinions, experience, and conclusions. Guest statements on the podcast do not express the opinions of ASCO. The mention of any product, service, organization, activity, or therapy should not be construed as an ASCO endorsement.

Disclosures:

Dr. Alissa Jamie Cooper

Honoraria
Company: MJH Life Scienes, Ideology Health, Intellisphere LLC, MedStar Health, Physician's Education Resource, LLC,  Gilead Sciences, Regeneron, Daiichi Sankyo/Astra Zeneca, Novartis, 

Research Funding: Merck, Roche, Monte Rosa Therapeutics, Abbvie, Amgen, Daiichi Sankyo/Astra Zeneca

Travel, Accommodations, Expenses: Gilead Sciences

In this JCO Precision Oncology Article Insights episode, Natalie DelRocco summarizes "Prognostic Value of the G2 Expression Signature and MYC Overexpression in Childhood High-Grade Osteosarcoma" by Roelof van Ewijk et al. published on May 29, 2025.

TRANSCRIPT

Natalie Del Rocco: Hello, and welcome to JCO Precision Oncology Article Insights. I'm your host, Natalie DelRocco, and today we will be discussing the original report, "Prognostic Value of the G2 Expression Signature and MYC Overexpression in Childhood High-Grade Osteosarcoma." This original report by van Ewijk et al. describes a study of the association between 2 biomarkers and survival outcomes among patients with high-grade osteosarcoma. Osteosarcoma is a disease where not much progress has been made in risk stratification factors that could potentially help patients target lower-risk therapies, less toxic therapies, or therapies that might be more toxic but could help their high-risk osteosarcoma.

So, it's important to identify risk factors that can help target therapies. The G1/G2 gene expression signature is a prognostic risk score developed by a French osteosarcoma group in 2022. They showed in a cohort of 79 osteosarcoma patients that risk score was associated with poorer event-free survival and overall survival. This considers expression of 15 individual genes. MYC amplification was shown in 2023 by a North American osteosarcoma group to be associated with poor overall survival in a cohort of 92 osteosarcoma patients, and this group validated that finding in a localized cohort in the same publication. 

The goal of this particular original report was to assess the prognostic significance of each of these biomarkers in a population independent to those prior publications and, hence, to serve as an external validation of prior findings and to assess these 2 biomarkers in the same study. The investigators considered MYC amplification, defined as having greater than 7 copies; MYC expression as a continuous rather than the previously categorized variable; and G2 expression defined as a continuous variable; and then G2 expression defined as a dichotomous variable with the cut point at the median, as done in the original paper. 

What the investigators found in their primary multivariable Cox proportional hazards regression model, which controlled for additional clinical risk factors such as age, tumor site, tumor size, is that G2 expression and MYC expression as continuous variables were associated with increased hazard of EFS and OS event. MYC amplification was not found to be prognostic. This is not surprising. When we have continuous variables, we have greater statistical power, we decrease the likelihood that an identified cut point in a previous study does not generalize well to either our genetic assay or our patient population. So, we don't have to worry about finding the optimal cut point in our particular patient sample.

Thank you for listening to our JCO Precision Oncology Article Insights. Don't forget to give us a rating or review, and be sure to like and subscribe so that you never miss an episode. You can find all ASCO shows at asco.org\podcasts.

The purpose of this podcast is to educate and to inform. This is not a substitute for professional medical care and is not intended for use in the diagnosis or treatment of individual conditions. 

Guests on this podcast express their own opinions, experience, and conclusions. Guest statements on the podcast do not express the opinions of ASCO. The mention of any product, service, organization, activity, or therapy should not be construed as an ASCO endorsement.

In this JCO Precision Oncology Article Insights episode, Natalie DelRocco summarizes "Real-Time Monitoring in Renal Cell Carcinoma With Circulating Tumor DNA: A Step Forward, but How Far?" by Zeynep B. Zengin et al. published on February 28, 2025.

TRANSCRIPT

The guest on this podcast episode has no disclosures to declare.

Natalie DelRocco: Hello, and welcome to JCO Precision Oncology Article Insights. I'm your host, Natalie DelRocco, and today we will be discussing the editorial, "Real-Time Monitoring in Renal Cell Carcinoma With Circulating Tumor DNA: A Step Forward, but How Far?"

This editorial by Zengin and Kotecha discusses the impact of circulating tumor DNA (ctDNA) and its potential applications in renal cell carcinoma - we'll call this RCC for the remainder of the podcast.

This article was published in February of 2025, and I think this is really timely because ctDNA is currently an emerging biomarker of interest in many different cancers. Having shown promise in certain cancers, other types of cancers are really targeting ctDNA to see if it can be used as a prognostic or a predictive biomarker in their specific field of oncology. Sometimes it is found that ctDNA is a prognostic marker that's associated with outcome, but it's not always clear whether it is a predictive biomarker that can help modify treatment and to what extent it could be helpful modifying treatment.

This is what the authors of this editorial really focus on. They focus on the applications of ctDNA in RCC by interpreting the accompanying article, "Longitudinal Testing of Circulating Tumor DNA in Patients With Metastatic Renal Cell Carcinoma" by Basu et al.

So, the editorial authors begin by giving examples of cancers where ctDNA has been shown to be useful in cancer monitoring - for example, locally advanced urothelial carcinoma - and they give examples of when it has not been shown to be useful in monitoring colorectal cancer. And this just highlights the variability of ctDNA as a biomarker. It's not always a useful biomarker, but sometimes it is.

The authors note that RCC may fall into the latter category - that is, the "not useful" category - due to the low ctDNA shedding which characterizes RCC. However, metastatic RCC - we'll call this 'mRCC' for the remainder of the podcast - may be a target for use of ctDNA clinically due to advanced assay development, according to the authors. Basu et al, in the original work that the editorial accompanies, showed in a retrospective study of 92 patients with mRCC that ctDNA detectability was associated with poorer PFS, regardless of receipt of active treatment versus no receipt of active treatment. That's important because ctDNA can be directly affected by therapy.

The authors of the editorial believe that this is a particularly promising result for a few reasons. Firstly, the estimated hazard ratios were quite large. A hazard ratio of 3.2 was seen in the active treatment group versus a hazard ratio of 18 was observed in the no-active-treatment group. I will note that a hazard ratio of 18 with an extremely wide confidence interval is an unusual observation. So, when interpreting this result, I would consider the direction and magnitude of the effect to be suggestive of promise but needing to be validated in the future to improve precision. And the authors of the editorial do agree with this; they note the same.

The authors also note that a single-patient example was used to show how that ctDNA positivity can be used in mRCC to monitor and prompt imaging if disease progression is suspected. And then that way, disease progression can be caught earlier. That to say, there is a real target for clinical use, which isn't always the case. Sometimes we know that ctDNA is associated with outcome, but we don't quite know how we can modify when we know that ctDNA is positive. In this case, the editorial authors show that we can use ctDNA positivity to monitor patients for disease progression.

Despite the promise of the study, the editorial does highlight that the study inherits typical retrospective study limitations. For example, there is a heterogeneous cohort. There is variability in data collection, particularly nailing down specific time points, which can always be a challenge when collecting biological samples as part of a study. And small sample size - although 92 patients is great for renal cell carcinoma, it is a challenging sample size with respect to precision of those hazard ratio estimates, which we've already talked about.

The authors additionally note that ctDNA could be used to direct therapy, not just to monitor for disease progression. So, both monitoring and changing therapy would certainly require further study and validation, which is discussed by the authors of this editorial. We would want larger, prospective studies showing the same association before we would be comfortable modifying treatment for patients based on their ctDNA positivity level.

Thank you for listening to JCO Precision Oncology Article Insights. Don't forget to give us a rating or a review, and be sure to subscribe so that you never miss an episode. You can find all ASCO shows at asco.org/podcasts.

The purpose of this podcast is to educate and to inform. This is not a substitute for professional medical care and is not intended for use in the diagnosis or treatment of individual conditions.

Guests on this podcast express their own opinions, experience, and conclusions. Guest statements on the podcast do not express the opinions of ASCO. The mention of any product, service, organization, activity, or therapy should not be construed as an ASCO endorsement.

JCO PO author Dr. Philip Philip at Henry Ford Cancer Institute and Wayne State University shares insights into his JCO PO article, "Incorporating Circulating Tumor DNA Testing Into Clinical Trials: A Position Paper by the National Cancer Institute GI Oncology Circulating Tumor DNA Working Group." Host Dr. Rafeh Naqash and Dr. Philip discuss how prospective trials are required to clarify the role of ctDNA as a valid surrogate end point for progression-free or overall survival in GI cancers.

Dr. Rafeh Naqash: Hello and welcome to JCO Precision Oncology Conversations, where we bring you engaging conversations with authors of clinically relevant and highly significant JCO PO articles. I'm your host, Dr. Rafeh Naqash, Podcast Editor for JCO Precision Oncology and Assistant Professor at the OU Health Stephenson Cancer Center at the University of Oklahoma.

Today, we are excited to be joined by Dr. Philip Philip, Chair of Hematology and Oncology, as well as leader of GI and Neuroendocrine Oncology. He's also the Professor of Oncology and Pharmacology, as well as Co-Leader of the Pancreatic Cancer Program and Medical Director of the Cancer Clinical Trial and Translational Research Office at the Henry Ford Cancer Institute at Wayne State University. Dr. Philip is also the Senior Corresponding Author of the JCO Precision Oncology article entitled, "Incorporating Circulating Tumor DNA Testing into Clinical Trials: A Position Paper by the National Cancer Institute GI Oncology Circulating Tumor DNA Working Group."

At the time of this recording, our guest's disclosures will be linked in the transcript.

Dr. Philip, welcome to our podcast, and thank you so much for joining us today.

Dr. Philip Philip: Thank you so much, Dr. Naqash, for providing me this opportunity to be discussing this with you.

Dr. Rafeh Naqash: This is a very timely and interesting topic. We've done a couple of podcasts on ctDNA before, but none that is an opinion piece or a guidance piece based on what you guys have done. Could you tell us what led to this perspective piece or guidance manuscript being published? There is some background to this. Could you tell us, for the sake of our listeners, what was the initial thought process of why you all wanted to do this?

Dr. Philip Philip: The major reason for this was the fact that investigators were considering using ctDNA as a primary endpoint in clinical trials. Obviously, you hear my focus will be on gastrointestinal cancers. So, the idea was, can we use ctDNA instead of using the traditional endpoints such as disease-free survival, progression-free survival, or overall survival? And the question was, do we have enough data to support that in patients with gastrointestinal cancers? Now, the article obviously goes over some review of the data available, but the core of the article was not to do a comprehensive review of ctDNA use and the evidence so far, although we used that in really putting our recommendations. So, we really had to evaluate available data. But the focus was, what are the gaps? What do we need to do? And are we ready to use ctDNA as a primary endpoint in clinical trials?

Dr. Rafeh Naqash: Thank you for giving us that background. Obviously, a very broad, complicated topic with a bunch of emerging data that you've highlighted. But most importantly, for the sake of, again, trainees and listeners, could you help us understand the difference between tumor-informed and non-tumor-based ctDNA assessments?

Dr. Philip Philip: Sure. So, the tumor-informed is simply meaning that you're taking the genomic makeup or the DNA fingerprint of the cancer in a given patient, and you create a profile, and then use that profile to see whether that DNA is present in the blood. So, it's very simple. It's like barcoding DNA and then going and looking for it in the blood, which means that you have to have the primary tumor. When I say primary tumor, you need to have the tumor to start off with. It doesn't really apply, maybe easily, if you just have a fine-needle aspirate and things like that. So, you really have to have a good amount of the tumor for you to be able to do that. So, that's a tumor-informed, and from the name, you can easily understand how it's done, compared to the other one, which is uninformed, whereby off-the-shelf probes are used to look for tumor DNA. And again, they're based on prior experience and prior identification of the key DNA changes that will be seen in tumors. So, that's the difference between the two in terms of the principle of the test.

The uninformed will not require you to send the original tumor that you're trying to test. However, the informed, you do. The turnaround time is, again, a bit different because, as you would expect, it's shorter in the uninformed. And the reason for that, again, is the initial preparation of the profile that is going to be used in the future when you do serial testing. The sensitivity has been a bit of a discussion. Initially, people have thought that tumor-informed assays are more sensitive, more specific, more sensitive, et cetera. But in our review, we come to the conclusion saying that we don't think that's going to be a major difference. And there are obviously improvements happening in both types of assays. The sensitivities have been improving. So, at this point in time, we do feel that you have two types of assays, and we didn't feel strongly about recommending one over the other.

Dr. Rafeh Naqash: Thank you for that description. You mentioned something about sensitivity, specificity. Obviously, many of us who have ordered both tumor-informed and tumor-uninformed, we understand the differences with respect to the timing. The tumor-informed one can take more time. The uninformed one, being a sort of a liquid biopsy, may not necessarily have as much of a turnaround time. Could you briefly speak to those limitations or advantages in the context of the two versions?

Dr. Philip Philip: I just really want to also highlight that when we say turnaround time, so for the tumor-informed assays, the first assay that we do will be requiring a turnaround time. But once the pattern has been set and the profile has been documented, the subsequent testing doesn't require much in the way of waiting. However, when you're using this for the minimal residual disease, then you have a window of opportunity to work at. That's number one. So, it means that in patients who have resected cancer, you may end up having to wait longer than the tumor-uninformed assay, especially if you don't have easy access to your material for the baseline material to send.

And also, what we'd like to do is not do the test immediately after the operation or soon after the operation. Give it some time. There's a window where you can work at, and starting minimally two weeks after the surgery. But in my experience, I'd like to wait at least four weeks just to make sure that we got an accurate reading. Sometimes when you do it very early after surgery, because of the effect of the surgery and the release of the normal DNA is also, it may dilute the tumor DNA, and then you may get a false negative. So, basically, it depends on the clinical situation.

And your question is, is one better to be used than the other? I think ultimately, it ends up with the turnaround time not being as much of an issue. It might be in certain situations, depending on when you see the patients after the operation or any definitive treatment you've done and you want to look for minimal residual disease. But in general, I don't think that's going to be a real major issue.

Dr. Rafeh Naqash: I remember discussing this with one of the tumor-informed platforms with regards to this barcode you mentioned. They generate a fingerprint of sorts for the tumor on the tissue, then they map it out in the blood and try to assess it longitudinally. And one of the questions and discussions we had was around the fact that most of the time, these barcoded genes are not the driver genes. If you have a KRAS mutant tumor, it's not going to be the KRAS gene that they map out. It's something that is specific.

So, is there a possibility that when you are mapping out, let's say, a metastatic tumor where there is truncal and subclonal mutations at different sites, that you capture something that is not necessarily truncal, and that does not necessarily reflect some other metastatic site having a recurrence? So basically, over time, you don't see a specific mutational pattern or the signature on the tumor-informed, and then you see something on the scan which makes you think, "Well, it was not the right test," but actually it could be a different subclone or a clone mutation at a different site. Is there a concept that could help us understand that better?

Dr. Philip Philip: I think you raise a very important point. Although, I have to say from my practical experience, that is not a common thing to see. In fact, for some reason, we don't see it that often in any frequency that should, at this point in time, make us concerned about the serial testing. But what you were mentioning is a real challenge which can happen. Now, the question is, how often does the clonal evolution or the divergence happen to the point that it's going to be like a false negative, is what you're saying. At this point in time, we don't really have good information on that, or any good information, practical information. And when we went through the literature and we were looking for the evidence, that wasn't something which was there clearly telling us. Although, this is something that has to be studied further prospectively. And I don't know of a study, but I might be missing it, I don't know of a study which is systematically looking at this. Although it's a very valid hypothesis and theoretical basis for it, but in real life, we still have to see how much does it really interfere with the validity of this kind of testing.

Dr. Rafeh Naqash: Which brings us to the more important discussion around your manuscript. And I think that the overarching theme here is the consensus panel that you guys had recommended that ctDNA-based metrics be used as a co-primary endpoint.

Could you tell us, for early-phase trials, maybe phase two studies for that matter, could you tell us what were some of the aspects that led to this consensus being formed from your working group?

Dr. Philip Philip: Well, there were a number of reasons, in any order of priority, but one of them is we don't have a good sense of dynamics of the ctDNA. And again, remember this article was about gastrointestinal cancers. Maybe we know more about colon cancer, but, or colorectal cancer, but we don't know that well about the upper GI, like gastroesophageal, pancreatic, et cetera. So, we don't know what is the false negative percentages. And in fact, we know that there are certain sites of the disease, metastases, that do not lead to enough shedding of the DNA into the circulation. So, that was something else. I mean, false negativity, not knowing exactly what the dynamics are, especially in different disease types. So, that was another reason, which we felt that it may not be at this time primetime to really have those ctDNA tests as a primary endpoint.

We wanted to make sure that, on the other hand, we wanted to make sure that people consider including ctDNA more like a secondary endpoint so that we can gain the information that we're lacking, at least the ones I mentioned to you. So, that was an important point of our discussions and deliberations when we were writing the article.

Dr. Rafeh Naqash: And I myself have been on both sides of the aisle where - I treat people with lung cancer, you mentioned appropriately that most of the data that we have for ctDNA is generated from GI cancers, especially colorectal - on the lung cancer side, I myself had a patient with an early-stage cancer, had treatment, surgery, immunotherapy, and then had ctDNA that was tumor-informed, was positive four to five months before the imaging actually showed up. And on the other side, I've also had an individual where early-stage lung cancer, surgery, immunotherapy, and then had PET scans that showed a positive finding, but the ctDNA, tumor-informed ctDNA, was negative multiple times. So, I've seen both aspects of it, and your paper tries to address some of these questions on how to approach a negative, radiologically negative imaging but positive ctDNA potentially, and vice versa. Could you elaborate upon that a little bit?

Dr. Philip Philip: Well, obviously, we do see this in practice. Again, I do GI oncology. I have patients who, you do ctDNA. I mean, my advice to anyone, when you order a test, you have to make sure that you know what you're going to do with the test, because that's the most important thing. You get a positive test, you do something. You get a negative test, you do something. But most importantly, our patients who you're following up, they are very anxious for a diagnosis they have that is not- I mean, it's cancer. If you're doing these tests, if we get continuous, repeatedly negative testing, then you really have to also tell the patient that there's a false negativity. And I mentioned to you earlier, there are certain sites of disease, like peritoneal, they may not be producing enough, or there are some tumors, their biology is such that they don't release as much to be detected in the blood. Now, one day we will get maybe a more sensitive test, but I'm talking about the tests we have now.

On the other hand, if you get a positive testing, you have to make a distinction for ctDNA in the minimal residual disease situation. If you get a positive test, there is enough evidence that the patient has a worse prognosis. There's evidence for that. No one can dispute that. Again, I'm talking about colorectal cancer where there are a lot of data for that. So, in that situation, there are studies that are looking, if you get a positive test in someone who you're not intending to give any adjuvant treatment, there are studies looking into that, both in terms of intensifying, like chemotherapy, in certain patients. And also, there's work being done, if you have a negative test in someone who has stage III disease, for example, or definitely stage II disease, they may not need to give them chemo. Those things are happening. But in metastatic disease, it's a different situation. Or even in someone who has received surgery, adjuvant chemotherapy, in those patients where they, whether they're now under, in the surveillance mode, those patients, if you have a positive, it may be positive. I had a recent patient like those, eight months before we saw anything on the scans. So, the question is, if you have a positive test, is there any advantage in giving them treatment, systemic treatment? Of course, we're assuming that the PET scan is negative. So, is there really any advantage in giving someone treatment ahead of time, before you see the imaging changes? That kind of data, in my opinion, is not really available or strong. You can always think of it in different ways, explain it in different ways. It's minimal disease, maybe you get a better response. But I don't know if we really can justify at this time. Therefore, in my practice, my own practice, I do not treat just a positive ctDNA. Again, that's different than after surgery when you're thinking of whether to give adjuvant treatment, no adjuvant treatment. But someone who's finished treatments and then you're just serially monitoring the disease, those patients, I do not treat them with chemotherapy. And that was something which, based on the literature we reviewed, there was nothing out there to definitely- I mean, if you see something positive, you will do a scan earlier, you will talk to the patient, examine the patient, whatever. But if there's nothing there, starting a treatment, that's not justified at this point in time.

Now, you need to do a study like that. Definitely, you need to do a study. But I can tell you that from my experience, having been involved with study design and all that, it's not an easy trial to do. It's going to be a trial- at a minimum, it will take many patients, it will take longer time to complete, and there are a number of variables there. If someone is willing to put a lot of money into it, it can be done. But I can tell you that that kind of intention to do a study like that has been very much a challenge at this time.

Dr. Rafeh Naqash: Of course, as you mentioned, the follow-up time that you need for a study like that is going to be very long to get to meaningful outcomes.

Dr. Philip Philip: You need to be very patient to do such a study. But the problem with a very long study is that things change, standard of care changes with time, and the assays will change. So, that's why we don't have that kind of data. I'm not sure if there are people in the community or in the academic centers who do treat based on only positive ctDNA.

The other thing is that you really have to always consider the psychological impact of these tests on patients and caregivers. Sometimes it can be really very stressful, burdensome to people to sit there just waiting for the disease to show up on a scan. And therefore, in my opinion, I'm not saying definitely don't use it in that situation, I'm just saying that you have to personalize it also, to see the patient who you would like to do it and then other patients who may not do it, or you think that it's not good for them to do it. And the patient also has to understand the outcome of the test and how you're going to be interpreting it.

Dr. Rafeh Naqash: That's a lot of great insights, Dr. Philip, and I know you've been involved in trial designs. I'm sure NCT and cooperative groups are actively thinking and incorporating ctDNA-based metrics as one of the endpoints in their trial. I know of a GU study that's, I think it's an Alliance study, trying to de-escalate treatment based on ctDNA. I have one of my colleagues who's also a GU investigator at OU, he's doing a ctDNA-based, tumor-informed-based de-escalation. So, obviously, more and more data, hopefully, that'll be generated in the next couple of years.

Dr. Philip Philip: But remember, these studies are not using it as an endpoint. They're using it as a means of optimizing treatment, which is a bit different. So, as an endpoint, can you do a phase III trial of, let's say, a thousand patients, and your primary endpoint is not survival, but you're saying, "Can I reduce the ctDNA, clear it earlier, or whatever?" That's the sort of thing this article was about. We can't do that at this time.

Dr. Rafeh Naqash: I totally understand. Thank you for explaining the difference, and hopefully more to come in this space in the next couple of years.

I briefly wanted to touch upon your personal career and journey based on all that you've done and accomplished. Could you tell us about how you started, what your journey has been like, and how that connects with what you're doing right now, including mentoring other trainees and junior faculty?

Dr. Philip Philip: Well, when I was in high school, I wanted to be an engineer, but I grew up in Baghdad, and all my friends wanted to do medicine, so I went with the tide, so I did medicine. I don't regret that. I would do it again if I had the opportunity. The reason why I did oncology was, I left the country and did a PhD in clinical pharmacology at the University of London. And that really got me, it was a topic which included, which was on cancer. So, I really got interested in a disease that is really a lot of science, and things are new, or were new at the time. And if I want to look back what I was doing, the beginning of my training in the 80s, second half of the 80s, and now, it's unbelievable how things have changed.

But one of the things which I really have to say is that almost all my life I've been in what we call academic institutions. But I firmly believe that for people, whether academic or not, you have to be a very good, astute clinician, because many of the things we do, really, we're trying to put the patients in the center. It's not only doing fancy science, it's to do things that help the patients. And you can bring in bits and pieces of fancy science or less fancy science, but that's something which is really extremely important for us to think about, being a very good clinician, very good doctor, because medicine is a science, whether you're practicing as a solo practitioner or you're part of a large academic center. It's the way you think, the way you interrogate things that you're not sure of, the way you collaborate, the way you learn every day. I mean, at my age, I still don't like to miss any tumor board, because in each tumor board, there's something you learn, even if you think that you know everything. So, that's really the whole thing of it, is that be a very good clinician, be open-minded. Always, you have to think of things that, they look interesting, they look somehow unexplained. Always try to help find the solutions and do that.

One of the major things that I feel that people should do is being also very focused on things. I mean, you have to also know what you want to do in the next 5, 10, 15 years. Because although everyone is in it in the same way when we start, but there are different things that drive people, people who want to do more of the formal research, like being an academic-like institution. But there are also a lot of people who are very successful outside of a- what we call an academic setting. In the United States, most people are not working in an academic kind of setting. Although, for me, the distinction between academic and community is getting less and less, because if you think that you do phase I trials in academia only, that's not true, because there are, in fact, in the state of Michigan, the most active phase I doctor is not even in academia, he's in private practice. So, you can do all these things. It's a matter of what you like to do, and you really have to make sure you know what you want to do. Because sometimes people are, especially early on, they get a bit confused, "What I want to do." There's an issue of doing general oncology versus subspecialist. If you're a subspecialist doing only GI, you have to make sure that you really also have some kind of recognition that you're only a GI oncologist, recognition regional, national, international, but some degree of recognition that you feel that people are coming to you for advice as a second opinion or whatever it is. But again, you have to decide what you think you want to be, how you want to be, because there's a lot of options here between community practice, academic practice, industry, and of course, there's always the administrative thing. Some people tend to be more like going into the line of being an administrator. So, there's a lot of options for you.

Dr. Rafeh Naqash: Well, thank you again, Dr. Philip, for those pearls of wisdom. I think that was very insightful. I'm sure all the trainees and early-career investigators will find all that advice very helpful. Thank you again for joining us today.

Thank you for listening to JCO Precision Oncology Conversations. Don't forget to give us a rating or review, and be sure to subscribe so you never miss an episode. You can find all ASCO shows at asco.org/podcast.

The purpose of this podcast is to educate and to inform. This is not a substitute for professional medical care and is not intended for use in the diagnosis or treatment of individual conditions.

Guests on this podcast express their own opinions, experience, and conclusions. Guest statements on the podcast do not express the opinions of ASCO. The mention of any product, service, organization, activity, or therapy should not be construed as an ASCO endorsement.

Honoraria: Bayer, Ipsen, incyte, Taiho Pharmaceutical, Astellas Pharma, BioNTech SE, Novocure, TriSalus Life Sciences, SERVIER, Seagen

Consulting or Advisory Role: Celgene, Ipsen, Merck, TriSalus Life Sciences, Daiichi Sankyo, SynCoreBio, Taiho Pharmaceutical

Speakers' Bureau: Incyte

Research Funding: Bayer (Inst), incyte (Inst), Merck (Inst), Taiho Pharmaceutical (Inst), novartis (Inst), Regeneron (Inst), Genentech (Inst), halozyme (Inst), Lilly (Inst), Taiho Pharmaceutical (Inst), merus (Inst), BioNTech SE (Inst)

Uncompensated Relationships: Rafael Pharmaceuticals, Caris MPI

In this JCO Precision Oncology Article Insights episode, Jiasen He summarizes "Predictive Impact of Tumor Mutational Burden on Real-World Outcomes of First-Line Immune Checkpoint Inhibition in Metastatic Melanoma" by Dr. Miles C. Andrews, et al. published on June 07, 2024.

The guest on this podcast episode has no disclosures to declare.

Jiasen He:

Hello and welcome to the JCO Precision Oncology Article Insights. I'm your host, Jiasen, and today we'll be discussing the JCO Precision Oncology article, "Predictive Impact of Tumor Mutational Burden on Real-World Outcomes of First-Line Immune Checkpoint Inhibition in Metastatic Melanoma," by Dr. Miles C. Andrews and colleagues. This study was supported by Foundation Medicine, a for-profit company that conducts FDA-regulated molecular diagnostics, including assays used to measure tumor mutational burdens, or TMB, as described in this article.

Immune checkpoint inhibitor (ICI) therapy has become a cornerstone in the treatment of metastatic melanoma. They work by activating the patient's own immune system, representing a fundamentally different approach from traditional chemotherapy. Several biomarkers have emerged as promising tools to predict ICI therapy response, and TMB is one of the most extensively studied. TMB is defined as the number of somatic mutations per megabase of an interrogated genome sequence. In the KEYNOTE-158 study, patients with high TMB showed better response rates and longer progression-free survival compared to those with low TMB, which led to the FDA tumor-agnostic approval of TMB as a biomarker to guide ICI therapy.

In this manuscript, Dr. Andrews and colleagues set out to answer an important question: does TMB predict outcomes of ICI therapy in real-world patients with advanced melanoma? To explore this, they analyzed de-identified data from the nationwide Flatiron Health-Foundation Medicine Clinico-Genomic Database (CGDB). To be included, patients needed to have had at least two visits to a Flatiron Health clinic and a Foundation Medicine Comprehensive Genomic Profiling report. Eligible patients had received first-line treatment with either monotherapy (nivolumab or pembrolizumab) or dual therapy with the combination of ipilimumab and nivolumab for metastatic melanoma. They also needed a tissue-based TMB score from either the FoundationOne or FoundationOne CDx genomic test. For this study, TMB less than 10 mutations per megabase was considered low TMB; TMB equal to or more than 10 mutations per megabase was considered high TMB; and TMB equal to or more than 20 mutations per megabase was considered very high TMB. Of the 497 patients in the final cohort, 29% had low TMB, while 71% had high TMB, and 50% had very high TMB.

The authors observed that patients with very high TMB were more often male, had BRAF wild-type tumors, and were more likely to receive anti-PD-1 monotherapy. This group also had tumors more commonly sampled from brain and lung metastases. Patients with high TMB but not very high TMB were more likely to carry the BRAF V600K mutation and were least likely to have lung metastases. Meanwhile, those with low TMB tended to be younger and had disease limited to non-visceral sites. As expected, the presence of ultraviolet mutation signatures, a known driver of melanoma, was strongly associated with TMB. UV signatures were found in just 18% of the low TMB group, but in 89% of the high TMB and 93% of the very high TMB group. High TMB was found to be prognostic of improved real-world progression-free survival (PFS) and overall survival (OS) in patients receiving both monotherapy and dual immune checkpoint inhibitors, even after adjusting for other established prognostic factors. Interestingly, in the low TMB group, overall survival was likely confounded by the availability of effective second-line targeted therapy, particularly for BRAF-mutant patients. These patients had better outcomes compared to their BRAF wild-type counterparts, likely reflecting a greater reliance on salvage therapy in low TMB patients who derived less benefit from first-line immunotherapy.

The authors then further examined the ICI outcomes using stepwise TMB thresholds, with TMB less than 10 as low, 10 to 19 as high, and equal to or more than 20 as very high. For those receiving ICI monotherapy, both PFS and OS were highest in the very high TMB group, followed by the high TMB group, and lowest in the low TMB group. However, in patients treated with dual ICI therapy, the results diverged. While low TMB patients still had the poorest outcomes, those with high TMB (mutations 10 to 19 per megabase) had better PFS and overall survival than those with very high TMB (mutations equal to or more than 20 per megabase).

The authors then conducted exploratory multivariable modeling, showing that among very high TMB patients with BRAF mutations, dual ICI therapy was associated with a significantly higher hazard ratio compared to monotherapy. They concluded that dual ICI may not benefit, and could even harm, patients with very high TMB, whereas those with TMB between 10 and 20 mutations per megabase may get more from the intensified regimen. Importantly, as the authors stated in the manuscript, we need to note that in this cohort, very high TMB patients were more likely to have brain metastases at treatment initiation, be male, and lack BRAF V600E/K mutations—all factors associated with poorer prognosis. This might partially explain inferior outcomes to dual ICI in very high TMB patients, as patients were not randomly assigned to therapy in this retrospective, real-world study. As such, these findings should be interpreted with caution and validated in future studies.

In summary, this study showed that in a real-world setting, high tumor mutational burden predicts better outcomes with immune checkpoint inhibitor therapy in patients with advanced melanoma. Interestingly, the authors found that dual ICI therapy may offer no added benefit for patients with very high TMB compared to ICI monotherapy. However, this was a retrospective, non-randomized study, and the cohorts were imbalanced for some known risk factors, which could confound outcomes. As a result, these findings should be interpreted with caution and will need to be validated in future prospective studies.

Thank you for tuning into JCO Precision Oncology Article Insights. Don't forget to subscribe and join us next time as we explore more groundbreaking research shaping the future of oncology. Until then, stay informed and stay inspired.

The purpose of this podcast is to educate and to inform. This is not a substitute for professional medical care and is not intended for use in the diagnosis or treatment of individual conditions.

Guests on this podcast express their own opinions, experience, and conclusions. Guest statements on the podcast do not express the opinions of ASCO. The mention of any product, service, organization, activity, or therapy should not be construed as an ASCO endorsement.

JCO PO author Dr. Dean A. Regier at the Academy of Translational Medicine, University of British Columbia (UBC), and the School of Population and Public Health, BC Cancer Research Institute shares insights into his JCO PO article, "Clinical Effectiveness and Cost-Effectiveness of Multigene Panel Sequencing in Advanced Melanoma: A Population-Level Real-World Target Trial Emulation."

Host Dr. Rafeh Naqash and Dr. Regier discuss the real-world clinical effectiveness and cost-effectiveness of multigene panels compared with single-gene BRAF testing to guide therapeutic decisions in advanced melanoma.

Dr. Rafeh Naqash:
Hello and welcome to JCO Precision Oncology Conversations, where we bring you engaging conversations with authors of clinically relevant and highly significant JCO PO articles. I'm your host, Dr. Rafeh Naqash, Podcast Editor for JCO Precision Oncology and Assistant Professor at the OU Health Stephenson Cancer Center in the University of Oklahoma.

Today, we are excited to be joined by Dr. Dean A. Regier, Director at the Academy of Translational Medicine, Associate Professor at the School of Population and Public Health, UBC Senior Scientist at the British Columbia Cancer Research Institute, and also the senior author of the JCO Precision Oncology article entitled "Clinical Effectiveness and Cost-Effectiveness of Multigene Panel Sequencing in Advanced Melanoma: A Population-Level Real-World Target Trial Emulation."

At the time of this recording, our guest's disclosures will be linked in the transcript.
Dean, welcome to our podcast and thank you for joining us today.

Dr. Dean Regier:
Thank you. I'm delighted to be here.

Dr. Rafeh Naqash:
So, obviously, you are from Canada, and medicine, or approvals of drugs to some extent, and in fact approvals of gene testing to some extent is slightly different, which we'll come to learn about more today, compared to what we do in the US—and in fact, similarly, Europe versus North America to a large extent as well.

Most of the time, we end up talking about gene testing in lung cancer. There is a lot of data, a lot of papers around single-gene panel testing in non-small cell lung cancer versus multigene testing. In fact, a couple of those papers have been published in JCO PO, and it has shown significant cost-effectiveness and benefit and outcomes benefit in terms of multigene testing. So this is slightly, you know, on a similar approach, but in a different tumor type. So, could you tell us first why you wanted to investigate this question? What was the background to investigating this question? And given your expertise in health economics and policy, what are some of the aspects that one tends or should tend to understand in terms of cost-effectiveness before we go into the results for this very interesting manuscript?

Dr. Dean Regier:
Yeah, of course, delighted to. So, one of the reasons why we're deeply interested in looking at comparative outcomes with respect to single- versus multigene testing— whether that's in a public payer system like Canada or an insurer system, a private system in the United States— is that the question around does multigene versus single-gene testing work, has not typically tested in randomized controlled trials. You don't have people randomized to multigene versus single-gene testing.

And what that does, it makes the resulting evidence base, whether it's efficacy, safety, or comparative cost-effectiveness, highly uncertain. So, the consequence of that has been uneven uptake around the world of next-generation sequencing panels. And so if we believe that next-gen sequencing panels are indeed effective for our patients, we really need to generate that comparative evidence around effectiveness and cost-effectiveness. So we can go to payers, whether it be single payer or a private insurer, to say, "Here are the comparative outcomes." And when I say that uptake has been uneven, uptake there's been actually plenty, as you know, publications around that uneven uptake, whether it be in Europe, in the United States, in Canada. And so we're really interested in trying to produce that evidence to create the type of deliberations that are needed to have these types of technologies accessible to patients. And part of those deliberations, of course, is the clinical, but also in some contexts, cost-effectiveness.

And so, we really start from the perspective of, can we use our healthcare system data, our learning healthcare system, to generate that evidence in a way that emulates a randomized controlled trial? We won't be able to do these randomized controlled trials for various, like really important and and reasons that make sense, quite frankly. So how can we mimic or emulate randomized controlled trials in a way that allows us to make inference around those outcomes? And for my research lab, we usually think through how do we do causal inference to address some of those biases that are inherent in observational data. So in terms of advanced melanoma, we were really interested in this question because first of all, there have been no randomized controlled trials around next-gen sequencing versus single-gene testing. And secondly, these products, these ICIs, immune checkpoint inhibitors, and BRAF and MEK inhibitors, they are quite expensive. And so the question really becomes: are they effective? And if so, to what extent are they cost-effective? Do they provide a good reason to have information around value for money?

Dr. Rafeh Naqash:
So now going to the biology of melanoma, so we know that BRAF is one of the tumor-agnostic therapies, it has approvals for melanoma as well as several other tumor types. And in fact, I do trials with different RAF-RAS kinase inhibitors. Now, one of the things that I do know is, and I'm sure some of the listeners know, is the DREAMseq trial, which was a melanoma study that was an NCI Cooperative Group trial that was led by Dr. Mike Atkins from Georgetown a couple of years back, that did show survival benefit of first-line immunotherapy sequencing. It was a sequencing study of whether to do first-line BRAF in BRAF-mutant melanoma followed by checkpoint inhibitors, or vice versa. And the immune checkpoint inhibitors followed by BRAF was actually the one that showed benefit, and the trial had to stop early, was stopped early because of the significant benefit seen.

So in that context, before we approach the question of single-gene versus multigene testing in melanoma, one would imagine that it's already established that upfront nivolumab plus ipilimumab, for that matter, doublet checkpoint inhibitor therapy is better for BRAF-mutant melanoma. And then there's no significant other approvals for melanoma for NRAS or KIT, you know, mucosal melanomas tend to have KIT mutations, for example, or uveal melanomas, for that matter, have GNAQ, and there's no targeted therapies. So, what is the actual need of doing a broader testing versus just testing for BRAF? So just trying to understand when you started looking into this question, I'm sure you kind of thought about some of these concepts before you delved into that.

Dr. Dean Regier:
I think that is an excellent question, and it is a question that we asked ourselves: did we really expect any differences in outcomes between the testing strategies? And what did the real-world implementation, physician-guided, physician-led implementation look like? And so, that was kind of one of the other reasons that we really were interested is, why would we go to expanded multigene panel sequencing at all? We didn't really expect or I didn't expect an overall survival a priori. But what we saw in our healthcare system, what happened in our healthcare system was the implementation in 2016 of this multigene panel. And this panel covered advanced melanoma, and this panel cost quite a bit more than what they were doing in terms of the single-gene BRAF testing. And so when you're a healthcare system, you have to ask yourself those questions of what is the additional value associated with that?

And indeed, I think in a healthcare system, we have to be really aware that we do not actually follow to the ideal extent randomized controlled trials or trial settings. And so that's the other thing that we have to keep in mind is when these, whether it's an ICI or a BRAF MEK inhibitor, when these are implemented, they do not look like randomized controlled trials. And so, we really wanted to emulate not just a randomized controlled trial, but a pragmatic randomized controlled trial to really answer those real-world questions around implementation that are so important to decision making.

Dr. Rafeh Naqash:
Sure. And just to understand this a little better: for us in the United States, when we talk about multigene testing, we generally refer to, these days, whole-exome sequencing with whole-transcriptome sequencing, which is like the nuclear option of of the testings, which is not necessarily cheap. So, when you talk about multigene testing in your healthcare system, what does that look like? Is it a 16-gene panel? Is it a 52-gene panel? What is the actual makeup of that platform?

Dr. Dean Regier:
Excellent question. Yeah, so at the time that this study is looking at, it was 2016, when we, as BC Cancer—so British Columbia is a population right now of 5.7 million people, and we have data on all those individuals. We are one healthcare system providing health care to 5.7 million people. In 2016, we had what I call our "home-brew" multigene panel, which was a 53-gene panel that was reimbursed as standard of care across advanced cancers, one of them being advanced melanoma. We have evolved since then. I believe in 2022, we are using one of the Illumina panels, the Focus panel. And so things have changed; it's an evolving landscape. But we're specifically focused on the 53-gene panel. It was called OncoPanel. And that was produced in British Columbia through the Genome Sciences Centre, and it was validated in a single-arm trial mostly around validity, etc.

Dr. Rafeh Naqash:
Thank you for explaining that. So now, onto the actual meat and the science of this project. So, what are some of the metrics from a health economy standpoint that you did look at? And then, methodology-wise, I understand, in the United States, we have a fragmented healthcare system. I have data only from my institution, for that matter. So we have to reach out to outside collaborators and email them to get the data. And that is different for you where you have access to all the data under one umbrella. So could you speak to that a little bit and how that's an advantage for this kind of research especially?

Dr. Dean Regier:
Yeah. In health economics, we look at the comparative incremental costs against the incremental effectiveness. And when we think about incremental costs, we think not just about systemic therapy or whether you see a physician, but also about hospitalizations, about all the healthcare interactions related to oncology or not that a patient might experience during their time or interactions with the healthcare system. You can imagine with oncology, there are multiple interactions over a prolonged time period depending on survival. And so what we try to do is we try to—and the benefit of the single-payer healthcare system is what we do is we link all those resource utilization patterns that each patient encounters, and we know the price of that encounter. And we compare those incremental costs of, in this case, it's the multigene panel versus the single-gene panel. So it's not just the cost of the panel, not just the cost of systemic therapy, but hospitalizations, physician encounters, etc.

And then similarly, we look at, in this case, we looked at overall survival - we can also look at progression-free survival - and ask the simple question, you know, what is the incremental cost per life-year gained? And in that way, we get a metric or an understanding of value for money. And how we evaluate that within a deliberative priority setting context is we look at safety and efficacy first. So a regulatory package that you might get from, in our case, Health Canada or the FDA, so we look at that package, and we deliberate on, okay, is it safe and is it effective? How many patients are affected, etc. And then separately, what is the cost-effectiveness? And at what price, if it's not cost-effective, at what price would it be cost-effective? Okay, so for example, we have this metric called the incremental cost-effectiveness ratio, which is incremental cost in the numerator, and in this case, life-years gained in the denominator. And if it is around $50,000 or $100,000 per life-year gained—so if it's in that range, this ratio—then we might say it's cost-effective. If it's above this range, which is common in oncology, especially when we talk about ICIs, etc., then you might want to negotiate a price. And indeed, when we negotiate that price, we use the economic evaluation, that incremental cost-effectiveness ratio, as a way to understand at what price should we negotiate to in order to get value for money for the healthcare system.

Dr. Rafeh Naqash:
Thank you for explaining those very interesting terminologies. Now, one question I have in the context of what you just mentioned is, you know, like the drug development space, you talked about efficacy and safety, but then on the safety side, we talk about all-grade adverse events or treatment-related adverse events—two different terminologies. From a healthcare utilization perspective, how do you untangle if a patient on a BRAF therapy got admitted for a hypoxic respiratory failure due to COPD, resulting in a hospitalization from the cost, overall cost utilization, or does it not matter?

Dr. Dean Regier:
We try to do as much digging into those questions as possible. And so, this is real-world data, right? Real-world data is not exactly as clean as you'd get from a well-conducted clinical trial. And so what we do is we look at potential adverse event, whether it's hospitalization, and the types of therapies around that hospitalization to try- and then engage with clinicians to try to understand or tease out the different grades of the adverse event. Whether it's successful or not, I think that is a real question that we grapple with in terms of are we accurate in delineating different levels of adverse events? But we try to take the data around the event to try to understand the context in which it happens.

Dr. Rafeh Naqash:
Thank you for explaining that, Dean.

So, again to the results of this manuscript, could you go into the methodology briefly? Believe you had 147 patients, 147 patients in one arm, 147 in the other. How did you split that cohort, and what were some of the characteristics of this cohort?

Dr. Dean Regier:
So, the idea, of course, is that we have selection criteria, study inclusion criteria, which included in our case 364 patients. And these were patients who had advanced melanoma within our study time period. So that was 2016 to 2018. And we had one additional year follow. So we had three total years. And what we did is that we linked our data, our healthcare system data. During this time, because the policy change was in 2016, we had patients both go on the multigene panel and on the single-gene BRAF testing. So, the idea was to emulate a pragmatic randomized controlled trial where we looked at contemporaneous patients who had multigene panel testing versus single-gene BRAF testing.

And then we did a matching procedure—we call it genetic matching. And that is a type of matching that allows us to balance covariates across the patient groups, across the multigene versus BRAF testing cohorts. The idea again is, as you get in a randomized controlled trial, you have these baseline characteristics that look the same. And then the hope is that you address any source selection or confounding biases that prohibit you to have a clean answer to the question: Is it effective or cost-effective? So you address all those biases that may prohibit you to find a signal if indeed a signal is there.

And so, what we did is we created—we did this genetic matching to balance covariates across the two cohorts, and we matched them one-to-one. And so what we were able to do is we were able to find, of those 364 patients in our pool, 147 in the multigene versus 147 in the single-gene BRAF testing that were very, very similar. In fact, we created what's called a directed acyclic graph or a DAG, together with clinicians to say, "Hey, what biases would you expect to have in these two cohorts that might limit our ability to find a signal of effectiveness?" And so we worked with clinicians, with health economists, with epidemiologists to really understand those different biases at play. And the genetic matching was able to match the cohorts on the covariates of interest.

Dr. Rafeh Naqash:
And then could you speak on some of the highlights from the results? I know you did survival analysis, cost-effectiveness, could you explain that in terms of what you found?

Dr. Dean Regier:
We did two analyses. The intention-to-treat analysis is meant to emulate the pragmatic randomized controlled trial. And what that does is it answers the question, for all those eligible for multigene or single-gene testing: What is the cost-effectiveness in terms of incremental life-years gained and incremental cost per life-years gained? And the second one was around a protocol analysis, which really answered the question of: For those patients who were actually treated, what was the incremental effectiveness and cost-effectiveness? Now, they're different in two very important ways. For the intention-to-treat, it's around population questions. If we gave single-gene or multigene to the entire population of advanced melanoma patients, what is the cost-effectiveness? The per-protocol is really around that clinical question of those who actually received treatment, what was the incremental cost and effectiveness? So very different questions in terms of population versus clinical cost and effectiveness.

So, for the intention-to-treat, what we found is that in terms of life-years gained is around 0.22, which is around 2.5 months of additional life that is afforded to patients who went through the multigene panel testing versus the single-gene testing. That was non-statistically significant from zero at the 5% level. But on average, you would expect this additional 2.5 months of life. The incremental costs were again non-statistically significant, but they're around $20,000. And so when we look at incremental cost-effectiveness, we can also look at the uncertainty around that question, meaning what percentage of incremental cost-effectiveness estimates are likely to be cost-effective at different willingness-to-pay thresholds? Okay? So if you are willing to pay $100,000 to get one gain of life-years, around 52.8% of our estimates, in terms of when we looked at the entire uncertainty, would be cost-effective. So actually that meets the threshold of implementation in our healthcare system. So it's quite uncertain, just over 50%. But what we see is that decision-makers actually have a high tolerance for uncertainty around cost-effectiveness. And so, while it is uncertain, we would say that, well, the cost-effectiveness is finely balanced.

Now, when we looked at the population, the per-protocol population, those folks who just got treatment, we actually have a different story. We have all of a sudden around 4.5 or just under 5 months of life gained that is statistically significantly different from zero, meaning that this is a strong signal of benefit in terms of life-years gained. In terms of the changes in costs or the incremental costs, they are larger again, but statistically insignificant. So the question now is, to what extent is it cost-effective? What is the probability of it being cost-effective? And at the $100,000 per life-year gained willingness-to-pay, there was a 73% chance that multigene panel testing versus single-gene testing is cost-effective.

Dr. Rafeh Naqash:
So one of the questions I have here, this is a clarification both for myself and maybe the listeners also. So protocol treatment is basically if you had gene testing and you have a BRAF in the multigene panel, then the patient went on a BRAF treatment. Is that correct?

Dr. Dean Regier:
It's still physician choice. And I think that's important to say that. So typically what we saw in both in our pre- and post-matching data is that we saw around 50% of patients, irrespective of BRAF status, get an ICI, which is appropriate, right? And so the idea here is that you get physician-guided care, but if the patient no longer performs on the ICI, then it gives them a little bit more information on what to do next. Even during that time when we thought it wasn't going to be common to do an ICI, but it was actually quite common.

Dr. Rafeh Naqash:
Now, did you have any patients in this study who had the multigene testing done and had an NRAS or a KIT mutation and then went on to those therapies, which were not captured obviously in the single-gene testing, which would have just tried to look at BRAF?

Dr. Dean Regier:
So I did look at the data this morning because I thought that might come up in terms of my own questions that I had. I couldn't find it, but what we did see is that some patients went on to clinical trials. So, meaning that this multigene panel testing allowed, as you would hope in a learning healthcare system, patients to move on to clinical trials to have a better chance at more appropriate care if a target therapy was available.

Dr. Rafeh Naqash:
And the other question in that context, which is not necessarily related to the gene platform, but more on the variant allele frequency, so if you had a multigene panel that captured something that was present at a high VAF, with suspicion that this could be germline, did you have any of those patients? I'm guessing if you did, probably very low number, but I'm just thinking from a cost-effective standpoint, if you identify somebody with germline, their, you know, first-degree relative gets tested, that ends up, you know, prevention, etc. rather than somebody actually developing cancer subsequently. That's a lot of financial gains to the system if you capture something early. So did you look at that or maybe you're planning to look at that?

Dr. Dean Regier:
We did not look at that, but that is a really important question that typically goes unanswered in economic evaluations. And so, the short answer is yes, that result, if there was a germline finding, would be returned to the patient, and then the family would be able to be eligible for screening in the appropriate context. What we have found in economic evaluations, and we've recently published this research, is that that scope of analysis is rarely incorporated into the economic evaluation. So those downstream costs and those downstream benefits are ignored. And when you- especially also when you think about things like secondary or incidental findings, right? So it could be a germline finding for cancer, but what about all those other findings that we might have if you go with an exome or if you go with a genome, which by the way, we do have in British Columbia—we do whole-genome and transcriptome sequencing through something called the Personalized OncoGenomics program. That scope of evaluation, because it's very hard to get the right types of data, because it requires a decision model over the lifetime of both the patients and potentially their family, it becomes very complicated or complex to model over patients' and families' lifetime. That doesn't mean that we should not do it, however.

Dr. Rafeh Naqash:
So, in summary Dean, could you summarize some of the known and unknowns of what you learned and what you're planning in subsequent steps to this project?

Dr. Dean Regier:
Our North Star, if you will, is to really understand the entire system effect of next-generation sequencing panels, exome sequencing, whole genomes, or whole genomes and transcriptome analysis, which we think should be the future of precision oncology. The next steps in our research is to provide a nice base around multigene panels in terms of multigene versus single-gene testing, whether that be colorectal cancer, lung cancer, melanoma, etc., and to map out the entire system implications of implementing next-generation sequencing panels.

And then we want to answer the questions around, "Well, what if we do exomes for all patients? What if we do whole genomes and transcriptomes for all patients? What are the comparative outcomes for a true tumor-agnostic precision oncology approach, accounting for, as you say, things like return of results with respect to hereditary cancers?" I think the challenge that's going to be encountered is really around the persistent high costs of something like a whole-genome and transcriptome sequencing approach. Although we do see the technology prices going down—the "$1,000 genome" or "$6,000 genome" on whatever Illumina machine you might have—that bioinformatics is continuing to be expensive.

And so, there are pipelines that are automated, of course, and you can create a targeted gene report really rapidly within a reasonable turnaround time. But of course, for secondary or what I call level two analysis, that bioinformatics is going to continue to be expensive. And so, we're just continually asking that question is: In our healthcare system and in other healthcare systems, if you want to take a precision oncology approach, how do you create the pipelines? And what types of technologies really lend themselves to benefits over and above next-generation sequencing or multigene panels, allowing for access to off-label therapies? What does that look like? Does that actually improve patients? I think some of the challenges, of course, is because of heterogeneity, small benefiting populations, finding a signal if a signal is indeed there is really challenging. And so, what we are thinking through is, with respect to real-world evidence methods and emulating randomized controlled trials, what types of evidence methods actually allow us to find those signals if indeed those signals are there in the context of small benefiting populations?

Dr. Rafeh Naqash:
Thank you so much, Dean. Sounds like a very exciting field, especially in the current day and age where cost-effectiveness, financial toxicity is an important aspect of how we improve upon what is existing in oncology. And then lots more to be explored, as you mentioned.

The last minute and a half I want to ask about you as an individual, as a researcher. There's very few people who have expertise in oncology, biomarkers, and health economics. So could you tell us for the sake of our trainees and early career physicians who might be listening, what was your trajectory briefly? How did you end up doing what you're doing? And maybe some advice for people who are interested in the cost of care, the cost of oncology drugs - what would your advice be for them very briefly?

Dr. Dean Regier:
Sure. So I'm an economist by training, and indeed I knew very little about the healthcare system and how it works. But I was recruited at one point to BC Cancer, to British Columbia, to really try to understand some of those questions around costs, and then I learned also around cost-effectiveness. And so, I did training in Scotland to understand patient preferences and patient values around quality of care, not just quantity of life, but also their quality of life and how that care was provided to them. And then after that, I was at Oxford University at the Nuffield Department of Population Health to understand how that can be incorporated into randomized control trials in children. And so, I did a little bit of learning about RCTs. Of course, during the way I picked up some epidemiology with deep understanding of what I call econometrics, what others might call biostatistics or just statistics.

And from there, it was about working with clinicians, working with epidemiologists, working with clinical trialists, working with economists to understand the different approaches or ways of thinking of how to estimate efficacy, effectiveness, safety, and cost-effectiveness. I think this is really important to think through is that we have clinical trialists, we have people with deep understanding of biostatistics, we have genome scientists, we have clinicians, and then you add economists into the mix. What I've really benefited from is that interdisciplinary experience, meaning that when I talk to some of the world's leading genome scientists, I understand where they're coming from, what their hope and vision is. And they start to understand where I'm coming from and some of the tools that I use to understand comparative effectiveness and cost-effectiveness. And then we work together to actually change our methods in order to answer those questions that we're passionate about and curious about better for the benefit of patients.

So, the short answer is it's been actually quite a trajectory between Canada, the UK. I spent some time at the University of Washington looking at the Fred Hutch Cancer Research Center, looking at precision oncology. And along the way, it's been an experience about interdisciplinary research approaches to evaluating comparative outcomes. And also really thinking through not just at one point in time on-off decisions—is this effective? Is it safe? Is it cost-effective?—not those on-off decisions, but those decisions across the lifecycle of a health product. What do those look like at each point in time? Because we gain new evidence, new information at each point in time as patients have more and more experience around it. And so what really is kind of driving our research is really thinking about interdisciplinary approaches to lifecycle evaluation of promising new drugs with the goal of having these promising technologies to patients sooner in a way that is sustainable for the healthcare system.

Dr. Rafeh Naqash:
Awesome. Thank you so much for those insights and also giving us a sneak peek of your very successful career.

Thank you for listening to JCO Precision Oncology Conversations. Don't forget to give us a rating or review, and be sure to subscribe so you never miss an episode. You can find all ASCO shows at asco.org/podcast. Thank you.

The purpose of this podcast is to educate and to inform. This is not a substitute for professional medical care and is not intended for use in the diagnosis or treatment of individual conditions.

Guests on this podcast express their own opinions, experience, and conclusions. Guest statements on the podcast do not express the opinions of ASCO. The mention of any product, service, organization, activity, or therapy should not be construed as an ASCO endorsement.

In this JCO PO Article Insights episode, host Harold Tan summarizes Low Kynurenine Levels Among Exceptional Responders on Phase Ib Trial of the HDAC Inhibitor Abexinostat with Pazopanib by Tsang et al, published November 07, 2024.

Harold Nathan Tan:

Welcome to JCO Precision Oncology Article Insights, where we explore cutting-edge discoveries in the world of cancer treatment and research. I'm Harold Nathan Tan, your host, and today we're taking a focused look at a compelling phase Ib trial led by Dr. Tsang, which investigates a combination of abexinostat, a histone deacetylase inhibitor, with pazopanib, a VEGF-targeting tyrosine kinase inhibitor, in patients with advanced solid tumors.

VEGF inhibition has long been an established therapeutic strategy across a wide range of tumor types, including colorectal, ovarian, sarcoma, and renal cell carcinoma. These agents function by disrupting tumor angiogenesis, effectively limiting oxygen and nutrient delivery to malignant cells and contributing to improved survival outcomes. However, over time, acquired resistance remains a significant challenge.

A key mechanism implicated in this resistance involves the upregulation of hypoxia-inducible factor 1-alpha, or HIF-1-alpha for short, a master regulator of angiogenesis that restores VEGF signaling under hypoxic conditions. Interestingly, HIF-1-alpha overexpression is mediated by histone deacetylases, especially HDAC2. Preclinical studies suggest that HDAC2 inhibition can suppress tumor cell migration and downregulate HIF-1-alpha activity, effectively disabling a critical escape pathway used by tumors under VEGF pressure. Moreover, combining HDAC inhibition with VEGF blockade has demonstrated synergy in pazopanib-resistant tumor models, forming a compelling rationale for this dual approach.

The phase Ib trial by Tsang et al. was designed to evaluate the safety, tolerability, and preliminary efficacy of this dual-targeted approach in patients with heavily pretreated advanced solid tumors. A dose-expansion cohort focused on individuals with renal cell carcinoma, allowing for more detailed evaluation in this population.

A central component of this study was the incorporation of biomarker analysis, particularly regarding HDAC2 expression levels. The results were noteworthy. Patients with high HDAC2 expression achieved a progression-free survival of 7.7 months compared to only 3.5 months in those with low expression. Even more compelling, overall survival reached 32.3 months for those with a high HDAC2 expression versus just 9.2 months for those with low expression. This suggests the potential role for HDAC2 as a predictive biomarker for response to combination HDAC and VEGF-targeted therapy.

The authors also explored the metabolic landscape of these patients, conducting metabolomic analysis focused on kynurenine, a key tryptophan catabolite known to contribute to the immune suppression in the tumor microenvironment. Its reduction is driven by HIF-1-alpha and inflammatory cytokines, including interleukin-6 and tumor necrosis factor-alpha. What they found was striking. Exceptional responders, defined as patients with treatment responses lasting more than 3 years, had consistently lower levels of kynurenine both before and after treatment. This finding introduces kynurenine as a potential metabolic biomarker. It suggests that patients with lower kynurenine levels may have a less immunosuppressive microenvironment, making them more responsive to the combined effects of HDAC inhibition and VEGF blockade. Of note, VEGF levels themselves did not significantly differ between responders and nonresponders, highlighting that the treatment benefit is not purely VEGF-mediated but likely driven by epigenetic and metabolic modulation.

On the safety front, the combination of abexinostat and pazopanib was generally well tolerated. However, this study did report a correlation between higher plasma concentrations of abexinostat and an increased incidence of thrombocytopenia, a class effect associated with HDAC inhibitors.

This trial introduces several key considerations for future research. First, it calls for validation of HDAC2 as a predictive biomarker. If confirmed in larger cohorts, HDAC2 expression could be used to select patients most likely to benefit from HDAC inhibitor-based regimens, transforming how we approach trial enrollment and treatment planning. Second, the link between low kynurenine and exceptional response supports further investigation into how metabolic pathways can influence treatment response to combined HDAC and VEGF inhibition.

Overall, HDAC inhibitors hold significant promise in precision oncology. Realizing their full therapeutic potential requires a deeper understanding of HDAC biology, refined combination strategies, and thorough preclinical and clinical evaluations tailored to individual patient profiles. This study exemplifies the potential of epigenetic-metabolic crosstalk as a therapeutic vulnerability and underscores the importance of precision stratification in clinical trial design. As research in this space progresses, the integration of molecular, epigenetic, and metabolic profiling will be essential in optimizing the use of HDAC inhibitors and expanding their role within precision oncology.

Thank you for tuning into JCO Precision Oncology Article Insights. Don't forget to subscribe and join us next time as we explore more groundbreaking research shaping the future of oncology. Until then, stay informed and stay inspired.

The purpose of this podcast is to educate and to inform. This is not a substitute for professional medical care and is not intended for use in the diagnosis or treatment of individual conditions.

Guests on this podcast express their own opinions, experience, and conclusions. Guest statements on the podcast do not express the opinions of ASCO. The mention of any product, service, organization, activity, or therapy should not be construed as an ASCO endorsement.

JCO PO author Dr. Timothy Showalter at Artera and University of Virginia shares insights into his JCO PO article, "Digital Pathology–Based Multimodal Artificial Intelligence Scores and Outcomes in a Randomized Phase III Trial in Men With Nonmetastatic Castration-Resistant Prostate Cancer" . Host Dr. Rafeh Naqash and Dr. Showalter discuss how multimodal AI as a prognostic marker in nonmetastatic castration-resistant prostate cancer may serve as a predictive biomarker with high-risk patients deriving the greatest benefit from treatment with apalutamide.

TRANSCRIPT 

Dr. Rafeh Naqash: Hello and welcome to JCO Precision Oncology Conversations where we'll bring you engaging conversations with authors of clinically relevant and highly significant JCO PO articles. I'm your host, Dr. Rafeh Naqash, podcast Editor for JCO Precision Oncology and assistant professor at the OU Health Stephenson Cancer Center at the University of Oklahoma.

Today, we are excited to be joined by Dr. Timothy Showalter, Chief Medical Officer at Artera and professor of Radiation Oncology at the University of Virginia and author of the JCO Precision Oncology article entitled, "Digital Pathology Based Multimodal Artificial Intelligence Scores and Outcomes in a Randomized Phase 3 Trial in Men with Non-Metastatic Castration Resistant Prostate Cancer."

At the time of this recording, our guest's disclosures will be linked in the transcript.

Dr. Showalter, it's a pleasure to have you here today.

Dr. Timothy Showalter: It's a pleasure to be here. Thanks for having me.

Dr. Rafeh Naqash: I think this is going to be a very interesting discussion, not just from a biomarker perspective, but also in terms of how technologies have evolved and how we are trying to stratify patients, trying to escalate or deescalate treatments based on biomarkers. And this article is a good example of that.

One of the things I do want to highlight as part of this article is that Dr. Felix Feng is the first author for this article. Unfortunately, Dr. Felix Feng passed away in December of 2024. He was a luminary in this field of prostate cancer research. He was also the Chair of the NRG GU Committee as well as Board of Directors for RTOG Foundation and has mentored a lot of individuals from what I have heard. I didn't know Dr. Feng but heard a lot about him from my GU colleagues. It's a huge loss for the community, but it was an interesting surprise for me when I saw his name on this article as I was reviewing it. Could you briefly talk about Dr. Feng for a minute and how you knew him and how he's been an asset to the field?

Dr. Timothy Showalter: Yeah. I'm always happy to talk about Felix whenever there's an opportunity. You know, I was fortunate to know Felix Feng for about 20 years as we met during our residency programs through a career development workshop that we both attended and stayed close ever since. And you know, he's someone who made an impact on hundreds of lives of cancer researchers and other radiation oncologists and physicians in addition to the cancer patients he helped, either through direct clinical care or through his innovation. For this project in particular, I first became involved soon after Felix had co-founded Artera, which is, you know the company that developed this. And because Felix was such a prolific researcher, he was actually involved in this and this research project from all different angles, both from the multimodal digital pathology tool to the trial itself and being part of moving the field forward in that way. It's really great to be able to sort of celebrate a great example of Felix's legacy, which is team science, and really moving the field forward in terms of translational projects based on clinical trials. So, it's a great opportunity to highlight some of his work and I'm really happy to talk about it with you.

Dr. Rafeh Naqash: Thanks, Tim. Definitely a huge loss for the scientific community. And I did see a while back that there was an international symposium organized, showcasing his work for him to talk about his journey last year where more than 200, 250 people from around the globe actually attended that. That speaks volumes to the kind of impact he's had as an individual and impact he's had on the scientific side of things as well.

Dr. Timothy Showalter: Yes. And we just had the second annual Feng Symposium the day before ASCO GU this year with, again, a great turnout and some great science highlighted, as well as a real focus on mentorship and team science and collaboration.

Dr. Rafeh Naqash: Thank you so much for telling us all about that. Now going to what you guys published in JCO Precision Oncology, which is this article on using a biomarker approach to stratify non-metastatic prostate cancer using this artificial intelligence based H&E score. Could you tell us the background for what started off this project? And I see there is a clinical trial data set that you guys have used, but there's probably some background to how this score or how this technology came into being. So, could you superficially give us an idea of how that started?

Dr. Timothy Showalter: Sure. So, the multimodal AI score was first published in a peer reviewed journal back in 2022 and the test was originally developed through a collaboration with the Radiation Therapy Oncology Group or Energy Oncology Prostate Cancer Research Team. The original publication describes development and validation of a risk stratification tool designed to predict distant metastasis and prostate cancer specific mortality for men with localized prostate cancer. And the first validation was in men who were treated with definitive radiation therapy. There have been subsequent publications in that context and there's a set of algorithms that have been validated in localized prostate cancer and there's a test that's listed on NCCN guidelines based on that technology.

The genesis for this paper was really looking at extending that risk stratification tool that was developed in localized prostate cancer to see if it could one, validate in a non-metastatic castrate refractory prostate cancer population for patients enrolled on the SPARTAN trial. And two, whether there was a potential role for the test output in terms of predicting benefit from apalutamide for patients with non-metastatic prostate cancer. For patients who are enrolled on the SPARTAN study, almost 40% of them had H&E stain biopsy slide material available and were eligible to be included in this study.

Dr. Rafeh Naqash: Going a step back to how prostate cancer, perhaps on the diagnostic side using the pathology images is different as you guys have Gleason scoring, which to the best of my knowledge is not necessarily something that most other tumor types use. Maybe Ki-67 is somewhat of a comparison in some of the neuroendocrine cancers where high Ki-67 correlates with aggressive biology for prognosis. And similarly high Gleason scores, as we know for some of the trainees, correlates with poor prognosis. So, was the idea behind this based on trying to stratify or sub-stratify Gleason scoring further, where you may not necessarily know what to do with the intermediate high Gleason score individual tumor tissues?

Dr. Timothy Showalter: Well, yeah. I mean, Gleason score is a really powerful risk stratification tool. As you know, our clinical risk groupings are really anchored to Gleason scores as an important driver for that. And while that's a powerful tool, I think, you know, some of the original recognition for applying computer vision AI into this context is that there are likely many other features located in the morphology that can be used to build a prognostic model.

Going back to the genesis of the discovery project for the multimodal AI model, I think Felix Feng would have described it as doing with digital pathology and computer vision AI what can otherwise be done with gene expression testing. You know, he would have approached it from a genomic perspective. That's what the idea was. So, it's along the line of what you're saying, which is to think about assigning a stronger Gleason score. But I think really more broadly, the motivation was to come up with an advanced complementary risk stratification tool that can be used in conjunction with clinical risk factors to help make better therapy recommendations potentially. So that was the motivation behind it.

Dr. Rafeh Naqash: Sure. And one of the, I think, other important teaching points we try to think about, trainees of course, who are listening to this podcast, is trying to differentiate between prognostic and predictive scores. So, highlighting the results that you guys show in relation to the MMAI score, the digital pathology score, and outcomes as far as survival as well as outcomes in general, could you try to help the listeners understand the difference between the prognostic aspect of this test and the predictive aspect of this test?

Dr. Timothy Showalter: So let me recap for the listeners what we found in the study and how it kind of fits into the prognostic and the predictive insights. So, one, you know, as I mentioned before, this is ultimately a model that was developed and validated for localized prostate cancer for risk stratification. So, first, the team looked at whether that same tool developed in localized prostate cancer serves as a prognostic tool in non-metastatic castrate-refractory prostate cancer. So, we applied the tool as it was previously developed and identified that about 2/3 of patients on the SPARTAN trial that had specimens available for analysis qualified as high risk and 1/3 of patients as either intermediate or low risk, which we called in the paper 'non-high risk'. And we're able to show that the multimodal AI score, which ranges from 0 to 1, and risk group, was associated with metastasis free survival time to second progression or PFS 2 and overall survival. And so that shows that it performs as a prognostic tool in this setting. And this paper was the first validation of this tool in non-metastatic castrate-refractory prostate cancer. So, what that means to trainees is basically it helps you understand how aggressive that cancer is or better stratify the risk of progression over time. So that's the prognostic performance.

Dr. Rafeh Naqash: Thank you for trying to explain that. It's always useful to get an example and understand the difference between prognostic and predictive. Now again, going back to the technology, which obviously is way more complicated than the four letter word MMAI, I per se haven't necessarily done research in this space, but I've collaborated with some individuals who've done digital pathology assessments, and one of the projects we worked on was TIL estimation and immune checkpoint related adverse events using some correlation and something that one of my collaborators had sent to me when we were working on this project as part of this H&E slide digitalization, you need color deconvolution, you need segmentation cell profiling. Superficially, is that something that was done as part of development of this MMAI score as well?

Dr. Timothy Showalter

You need a ground truth, right? So, you need to train your model to predict whatever the outcome is. You know, if you're designing an AI algorithm for Ki-67 or something I think you mentioned before, you would need to have a set of Ki-67 scores and train your models to create those scores. In this case, the clinical annotation for how we develop the multimodal AI algorithm is the clinical endpoints. So going back to how this tool was developed, the computer vision AI model is interpreting a set of features on the scan and what it's trying to do is identify high risk features and make a map that would ultimately predict clinical outcomes. So, it's a little bit different than the many digital pathology algorithms where the AI is being trained to predict a particular morphological finding. In this case, the ground truth that the model is trained to predict is the clinical outcome.

Dr. Rafeh Naqash: Sure. And from what you explained earlier, obviously, tumors that had a high MMAI score were the ones that were benefiting the most from the ADT plus the applausive. Is this specific for this androgen receptor inhibitor or is it interchangeable with other inhibitors that are currently approved?

Dr. Timothy Showalter: That's a great question and we don't know yet. So, as you're alluding to, we did find that the MMAI risk score was predictive for benefit from apalutamide and so it met the statistical definition of having a significant interaction p value so we can call it a predictive performance. And so far, we've only looked in this population for apalutamide. I think you're raising a really interesting point, which is the next question is, is this generalizable to other androgen receptor inhibitors? There will be future research looking at that, but I think it's too early to say.

Just for summary, I think I mentioned before, there are about 40% of patients enrolled on the SPARTAN study had specimens available for inclusion in this analysis. So, the SPARTAN study did show in the entire clinical trial set that patients with non-metastatic castrate-refractory prostate cancer benefited from apalutamide. The current study did show that there seems to be a larger magnitude of benefit for those patients who are multimodal AI high risk scores. And I think that's very interesting research and suggests that there's some interaction there. But I certainly would want to emphasize that we have not shown that patients with intermediate or low risk don't benefit from apalutamide. I think we can say that the original study showed that that trial showed a benefit and that we've got this interesting story with multimodal AI as well.

Dr. Rafeh Naqash: Sure. And I think from a similar comparison, ctDNA where ctDNA shows prognostic aspects, I treat people with lung cancer especially, and if you're ctDNA positive at a 3 to 4-month period, likely chances of you having a shorter disease-free interval is higher. Same thing I think for colorectal cancers. And now there are studies that are using ctDNA as an integral biomarker to stratify patients positive/negative and then decide on escalation/de-escalation of treatment. So, using a similar approach, is there something that is being done in the context of the H&E based stratification to de-intensify or intensify treatments based on this approach?

Dr. Timothy Showalter: You're hitting right on the point in the most promising direction. You know, as we pointed out in the manuscript, one of the most exciting areas as a next step for this is to use a tool like this for stratification for prospective trials. The multimodal AI test is not being used currently in clinical trials of non-metastatic castrate-refractory prostate cancer, which is a disease setting for this paper. There are other trials that are in development or currently accruing where multimodal AI stratification approach is being taken, where you see among the high-risk scores, at least in the postoperative setting for a clinical trial that's open right now, high risk score patients are being randomized to basically a treatment intensification question. And then the multimodal AI low risk patients are being randomized to a de-intensification experimental arm where less androgen deprivation therapy is being given. So, I think it's a really promising area to see, and I think what has been shown is that this tool has been validated really across the disease continuum. And so, I think there are opportunities to do that in multiple clinical scenarios.

Dr. Rafeh Naqash: Then moving on to the technological advancements, very fascinating how we've kind of evolved over the last 10 years perhaps, from DNA based biomarkers to RNA expression and now H&E. And when you look at cost savings, if you were to think of H&E as a simpler, easier methodology, perhaps, with the limitations that centers need to digitalize their slides, probably will have more cost savings. But in your experience, as you've tried to navigate this H&E aspect of trying to either develop the model or validate the model, what are some of the logistics that you've experienced can be a challenge? As we evolve in this biomarker space, how can centers try to tackle those challenges early on in terms of digitalizing data, whether it's simple data or slides for that matter?

Dr. Timothy Showalter: I think there's two main areas to cover. One, I think that the push towards digitalization is going to be, I think, really driven by increasing availability and access to augmentative technologies like this multimodal AI technology where it's really adding some sort of a clinical insight beyond what is going to be generated through routine human diagnostic pathology. I think that when you can get these sorts of algorithms for patient care and have them so readily accessible with a fast turnaround time, I think that's really going to drive the field forward. Right now, in the United States, the latest data I've seen is that less than 10% of pathology labs have gone digital. So, we're still at an early stage in that. I hope that this test and similar ones are part of that push to go more digital.

The other, I think, more interesting challenge that's a technical challenge but isn't about necessarily how you collect the data, but it certainly creates data volume challenges, is how do you deal with image robustness and sort of translating these tools into routine real-world settings. And as you can imagine, there's a lot of variation for staining protocols, intensity scanner variations, all these things that can affect the reliability of your test. And at least for this research group that I'm a part of that has developed this multimodal AI tool can tell you that the development is sophisticated, but very data and energy intensive in terms of how to deal with making a tool that can be consistent across a whole range of image parameters. And so that presents its own challenges for dealing with a large amount of compute time and AI cycles to make robust algorithms like that. And practically speaking, I think moving into other diseases and making this widely available, the size of data required and the amount of cloud compute time will be a real challenge.

Dr. Rafeh Naqash: Thank you for summarizing. I can say that definitely, you know, this is maybe a small step in prostate cancer biomarker research, but perhaps a big step in the overall landscape of biomarker research in general. So definitely very interesting.

Now, moving on to the next part of the discussion is more about you as a researcher, as an individual, your career path, if you can summarize that for us. And more interestingly, this intersection between being part of industry as well as academia for perhaps some of the listeners, trainees who might be thinking about what path they want to choose.

Dr. Timothy Showalter: Sure. So, as you may know, I'm a professor at the University of Virginia and I climbed the academic ladder and had a full research grant program and thought I'd be in academia forever. And my story is that along the way, I kind of by accident ended up founding a medical device company that was called Advaray and that was related to NCI SBIR funding. And I found myself as a company founder and ultimately in that process, I started to learn about the opportunity to make an impact by being an innovator within the industry space. And that was really the starting point for me. About four years ago, soon after Felix Feng co-founded Artera, he called me and told me that he needed me to join the company. For those who were lucky to know Felix well, at that very moment, it was inevitable that I was going to join Artera and be a part of this. He was just so persuasive. So, I will say, you know, from my experience of being sort of in between the academic and industry area, it's been a really great opportunity for me to enter a space where there's another way of making an impact within cancer care. I've gotten to work with top notch collaborators, work on great science, and be part of a team that's growing a company that can make technology like this available.

Dr. Rafeh Naqash: Thank you so much, Tim, for sharing some of those thoughts and insights. We really appreciate you discussing this very interesting work with us and also appreciate you submitting this to JCO Precision Oncology and hopefully we'll see more of this as this space evolves and maybe perhaps bigger more better validation studies in the context of this test.

Thank you for listening to JCO Precision Oncology Conversations. Don't forget to give us a rating or review and be sure to subscribe so you never miss an episode. You can find all ASCO shows at asco.org/podcast.

The purpose of this podcast is to educate and to inform. This is not a substitute for professional medical care and is not intended for use in the diagnosis or treatment of individual conditions.

Guests on this podcast express their own opinions, experience and conclusions. Guest statements on the podcast do not express the opinions of ASCO. The mention of any product, service, organization, activity, or therapy should not be construed as an ASCO endorsement.

In this JCO Precision Oncology Article Insights episode, Natalie DelRocco summarizes "Digital Pathology–Based Multimodal Artificial Intelligence Scores and Outcomes in a Randomized Phase III Trial in Men With Nonmetastatic Castration-Resistant Prostate Cancer" by Felix Y. Feng, et al published January 31, 2025.

Come back for the next episode where JCO Precision Oncology Conversations host, Dr. Rafeh Naqash interviews the author of the JCO PO article discussed, Dr. Tim Showalter.

TRANSCRIPT

Natalie DelRocco: Hello and welcome to JCO Precision Oncology Article Insights. I'm your host Natalie Del Rocco. Today, we'll be discussing the article, "Digital Pathology-Based Multimodal Artificial Intelligence Scores and Outcomes in a Randomized Phase III Trial in Men With Nonmetastatic Castration-Resistant Prostate Cancer." We will also be discussing the accompanying editorial, "Leveraging Artificial Intelligence to Improve Risk Stratification in Nonmetastatic Castration-Resistant Prostate Cancer." So, we're going to start by summarizing the original report and then we'll jump into a few of the high-level interpretations that were supplied by the editorial.  

The original report by Feng et. al. describes the application of multimodal artificial intelligence to data collected on a nonmetastatic castration-resistant prostate cancer. We will call this disease moving forward NMCRPC, a Clinical Trial. So, we're looking at data from an NMCRPC clinical trial. The SPARTAN trial was a randomized phase three trial and this study compared metastasis-free survival as the primary endpoint for those treated with traditional androgen deprivation therapy or ADT to those treated with androgen deprivation therapy plus apalutamide. Other secondary endpoints included progression-free survival and overall survival, but the primary endpoint there was metastasis-free survival or MFS. This study found that the addition of apalutamide resulted in a significantly longer median metastasis-free survival compared to androgen deprivation therapy alone. And we should note that this is a double-blind placebo-controlled trial. In the overall study, 1,207 patients participated and over the course of this study histopathology slides were collected and they were digitized for future use. And that future use is what we are going to be discussing today. 

The authors do note that there are currently no good biomarkers for use in NMCRPC. The authors seem to be inspired by the ArteraAI prostate test, which was a recent application of multimodal artificial intelligence models. But in localized prostate cancer as opposed to NMCRPC, the authors constructed a multimodal artificial intelligence model or an MMAI model. They applied this to the SPARTAN trial with the intention of developing a risk score that could be used for risk stratification in NMCRPC. And we should note here that multimodal artificial intelligence or MMAI is a broad class of artificial intelligence models, and they can analyze different types of data at one time, hence the term multimodal. So in this example, the author's primary data source of interest were those digitized histopathology images because histopathology tells you a lot about NMCRPC. The authors though also wanted their model to consider traditional clinical factors that are known to be prognostic such as Gleason score, tumor stage, PSA level, and age. So those two different types of data, those histopathology images and that traditional clinical data are the two different types of data that make this model multimodal. So we should note here importantly, after dropping missing data, 420 patients contribute to this model, the MMAI model. 

The authors generate a risk score from this MMAI model and they categorize that risk score into low, intermediate, and high risk groups using clinical knowledge. The authors found in their results that an increase in this MMAI risk score was associated with an increased hazard of metastasis-free survival event with a hazard ratio from a Cox proportional hazards model of 1.72. To summarize how the authors arrived here, they derived a risk score from this MMAI model which incorporates both imaging and regular data. They plugged this risk score into a Cox proportional hazards mode,l modeling metastasis-free survival and they found that an increase in that MMAI based risk score is associated with increased hazard of metastasis-free event with a hazard ratio of 1.72, which is quite large. Additionally, the risk score seemed to be associated with PFS2 and OS, which were two of the secondary endpoints from the SPARTAN clinical trial, though the effect sizes were more modest. Those are the highlights from the original report, the methods and the results. 

The accompanying editorial notes that both histopathology and Gleason score specifically are very critical to understanding prostate cancer, and Gleason score alone is not sufficient to summarize the complexity of the disease, although it is a well validated prognostic factor for prostate cancer. So this makes MMAI an excellent tool in the setting described by the authors. We have an existing prognostic factor that doesn't describe the entire picture of the disease by itself and so we can use those digitized histopathology slides to help bolster our understanding and provide the model more information. MMAI allows you to do this because it can take in different types of data. So that was the main conclusion of the editorial. 

They also summarize a number of recent validations of MMAI models in prostate cancer research, noting that it will be an important tool for risk stratification and has already been shown to outperform classical techniques. The editorial though does highlight a number of weaknesses of this paper, limitations and I think the most important one to highlight, and we touched on this earlier, is that 420 patients from the SPARTAN clinical trial contributed to the development of this MMAI score. That is a small proportion of the roughly 1200 patients that did participate in the SPARTAN clinical trial. So we have a small subgroup analysis that can be limiting and this model will need to be validated in a broader population in the future.

Thank you for listening to JCO Precision Oncology Article Insights. Don't forget to give us a rating or a review and be sure to subscribe so that you never miss an episode. You can find all ASCO shows at asco.org/podcasts.

The purpose of this podcast is to educate and to inform. This is not a substitute for professional medical care and is not intended for use in the diagnosis or treatment of individual conditions.

Guests on this podcast express their own opinions, experience and conclusions. Guest statements on the podcast do not express the opinions of ASCO. The mention of any product, service, organization, activity or therapy should not be construed as an ASCO endorsement.