Integrating Genetic Variants and Clinical Factors for Improved Prostate Cancer Prognostication - Susan Halabi

June 11, 2024

Susan Halabi discusses her presentation on prognostic factors in metastatic castration-resistant prostate cancer (mCRPC) based on the Alliance trial. The phase three trial, which enrolled 1,311 patients, compared the effects of enzalutamide alone to enzalutamide combined with abiraterone and prednisone. Despite no difference in overall survival between the treatment arms, Dr. Halabi's team developed a new prognostic model that integrates genetic and clinical variables. This model includes significant genetic markers like AR enhancer gain and RSPO2 gain, alongside clinical factors such as alkaline phosphatase and hemoglobin, improving the accuracy of predicting patient outcomes. Dr. Halabi emphasizes the future potential of this integrated approach for more personalized patient monitoring and treatment. She highlights the importance of data sharing and collaboration, recognizing the critical role of patients and the contributions from her team and junior researchers.


Susan Halabi, PhD, Professor of Biostatistics & Bioinformatics Chief, Division of Biostatistics, Duke Cancer Institute, Durham, NC

Alicia Morgans, MD, MPH, Genitourinary Medical Oncologist, Medical Director of Survivorship Program at Dana-Farber Cancer Institute, Boston, MA

Read the Full Video Transcript

Alicia Morgans: Hi, I'm so excited to be here today with Professor Susan Halabi, who is joining me from Duke, also joining me from ASCO 2024, where she presented a wonderful oral abstract looking at prognostication and risk factors for patients with metastatic castration-resistant prostate cancer within an Alliance trial.

Thank you so much for being here with me today.

Susan Halabi: I'm delighted to be here. Thank you, Dr. Morgans, for having me.

Alicia Morgans: Well, thank you so much for being here and for taking the time to talk to us about the work that you did. Before we get into the presentation that you gave at ASCO, if you could, would you mind reminding us what the Alliance trial that really provided the fuel for this analysis was?

Susan Halabi: Absolutely. So the Alliance trial is a phase three trial that randomized 1,311 patients, mCRPC patients, to either enzalutamide or enzalutamide plus abiraterone and prednisone, and the randomization used two stratification variables: prior chemotherapy and prognostic risk group as determined by our prior clinical nomogram. The primary endpoint for this study was overall survival, and even though the study was successful in recruiting patients, we found no differences in overall survival in the combination arm versus the enzalutamide arm alone. That really formed the basis for this study for our new prognostic model that integrated genetic variants and clinical factors.

Alicia Morgans: Well, and this is so important, because as you said, you previously had a prognostic model, the holobiome model as many of us refer to it, or the holobiome nomogram, and this model helped you as a stratification factor because it really did distinguish patients who are going to have better or worse prognoses, and is so important, especially as we're thinking about prospective studies. But in the new analysis that you did, as you said, you were able to integrate genetic factors to really, hopefully, at least the hypothesis was, I'm sure, enhance the ability of the model to predict outcomes for patients. So tell me a little bit about how you did that genetic analysis and what you ended up finding in terms of factors that were prognostic.

Susan Halabi: Yeah, that's really an excellent question, Dr. Morgans. So just as background, as you know, there are a lot of prognostic models of overall survival in first-line mCRPC patients, and most of these models are based on clinical variables only, and these variables are related to tumor or host characteristics. In general, what we found out is that if you use clinical variables, you're able to assign patients into different prognostic groups.

However, with those models, what we found is that the prior validated models' predictive accuracy is not doing really well, especially now with a shift in stage migration. So we used the A03-1201 trial where we had banked plasma, and we had a hypothesis that if you integrate some of the genetic variance information, because we know there are several studies that show that survival is impacted by these factors, such as circulating tumor DNA fraction, such as AR and non-AR ctDNA alterations.

We hypothesized that if we incorporate clinical and genetic variances in the model, we are going to be able to prognosticate patients better, so we will know who are in a high risk, who are in an intermediate risk, and who are in a low risk, and obviously, this information is very important because one of the key principles in using a stratification variable is that you want the two arms in a phase 3 trial to be balanced. You don't want to have heterogeneity between the arms, and if you use the prognostic risk group as a stratification variable, you are ensuring that there is comparability at baseline between those two arms, so you're not introducing any bias, and that is going to be really important, I would say, for the next phase 3 trials in mCRPC.

Alicia Morgans: Absolutely. Now, this genetic analysis was absolutely new, as you said, using ctDNA and looking at various aspects there. What was the analysis, or what did you pull out of that as a potential contributor to your new risk model?

Susan Halabi: Yeah, this is an excellent question. Again, we are not the first to show that circulating tumor DNA fraction or other genetic variants are important, because there have been really a lot of studies out there in the literature. However, the studies have been limited by the number of patients.

We had data banked on 776 patients, and DNA sequencing was performed on those patients at the University of Minnesota Genomics Center. When we looked at incorporating those genetic variants, we ended up including 16 genetic factors in our model, of which the androgen receptor enhancer gain was the most influential variable, followed by MEK gain and RSPO2 gain.

Now, the model did also include top influential variables that were clinical, such as alkaline phosphatase and hemoglobin, and what we've noticed is if you use a model with just the clinical variables, our time-dependent area under the ROC curve was about 0.72. However, if we include the clinical and the genetic variants together in one model, our tAUC, the mean, was about 0.77, and this was statistically significant, so clearly the genetic features are contributing. They're really contributing a lot to the model, to the prognosis of these patients.

Alicia Morgans: And I think there were some genetic factors that you found that really haven't been described previously. I wonder if you could highlight one or two of those, just for us to have a sense of some of the innovation of the team.

Susan Halabi: Yeah, absolutely. So AR gain has been documented before as an important prognostic factor. I think we are maybe the first to look at AR enhancer gain, and of course, there is a high correlation between AR gain and AR enhancer gain, but biologically they may be different. I think really this is an area that we ourselves don't know, what's the implication of that, but definitely this is something worth exploring in the future to understand the dynamics and to understand the underpinnings of the molecular biology and how each of those might work on prognosis, specifically on overall survival.

So that was one, and I believe the RSPO2 gain is something that no one has identified as an important prognostic factor of overall survival. But clearly, we used only one data set. Our next step is to use data, an external data set, to validate this model. It is really critical to validate this model, and we would like to validate the model in more contemporary patients who have used other therapies because in this trial we're using enzalutamide versus enzalutamide, abiraterone, and prednisone, so clearly now there's a shift in treatment, and we would like to validate this model in contemporary patients.

Alicia Morgans: Wonderful. So if you had to answer to a clinician who said, "Can I look into this tomorrow? Can I order this? Is it a CLIA-certified test?" it sounds like the answer is, "Not yet. Not a certified test. Not something that's been validated yet," but is this something that you think in the near future, relatively near future, might be potentially validated, integrated into prognostic scenarios and clinical trial development?

Susan Halabi: Yeah, this is a very good question. I think the model, just to clarify, the model has been validated. Without going too much into the statistics or how we did the model and the validation, in terms of the validation, what we did, we split the data a hundred times randomly and did the fitting, and then we evaluated the model on each of the hundred models, and then we computed a measure, which is the tAUC of the predictive accuracy, and we found the median.

But I think I will feel more comfortable if we have an external data set and could validate the model. I think it will be more solid. I think people can start using it. So, for instance, if one is interested in doing an enriched design and they want to select high-risk patients, I'm sure the model will do great. It will tell you who are the high-risk patients, regardless of whether you use a three-level scheme, prognostic scheme, or four.

Just to give you an idea, the patients who were in the poor-risk group had a median of 16 months, or almost 17 months, so that's a very, very short median. Compared to patients who were in the low-risk group, they had the median overall survival that hasn't been reached. So clearly, you can see within even the same trial, you can see the impact of those variables, those prognostic factors, and the heterogeneity of patients, and if someone wants to do a quick study, you can just select the intermediate or the poor-risk patients and have a quick readout, so I think that will work well.

I've already mentioned another possibility of how you can use the model as a stratification variable, and this will ensure that your model is balancing the randomization balance between the two arms. Rather than have 10, let's say, stratification factors, you can have one factor, which is the model-based prognostic risk grouping, and that will balance all patients between the two arms, let's say an experimental arm and standard of care, so that's another way of using it.

And I would say when it comes to prediction on an individual patient level, that's the part where I feel we are not ready for prime time because the model really needs to be validated externally, and then once it's validated, then a clinician may be able to use it to predict survival on an individual patient level.

Alicia Morgans: Wonderful. Well, if you had to sum all of that up into one final message to the listener, what would that be?

Susan Halabi: I would say this is the area we are looking at. This is now, this is the future. Looking at only clinical factors alone is the past because what we have seen is a stage migration. The prognostic factors are changing over time. The beauty of this assay is the presence of ctDNA allows for non-invasive testing, and you can use that... This is very powerful. You can use it for monitoring patients for progression, and even more importantly, identify patients who are becoming resistant to therapy.

Our next step is we're not only going to validate the model externally, we're already working on looking at on-treatment changes. We have about 450 patients where we have plasma collected on them, and that's our next step, is how can we incorporate dynamic changes in monitoring the patients and counseling the patients, and also identifying who are having resistance and what are the other options for these patients.

Alicia Morgans: Fantastic. And also fantastic that you did this work with the team through the Alliance, one of our NCTN Cooperative group systems. These are not easy lifts, and I really commend you and the team and thank all of the patients for participating.

Susan Halabi: Yeah, thank you. And I would like to also thank the patients and highlight the importance of data sharing. The patients have been incredible to allow us to share their data and to have access to the plasma.

And I also want to take a minute and greatly highlight the importance of my other PIs. This work has been funded by the NIH, really, NCI R01 that's been co-led by Dr. Andy Armstrong, Dr. Scott Dehm, and myself, and it really provided an opportunity, not only for us and working with an Alliance investigator, Dr. Mike Morris, who was also the PI of the Alliance trial, but others like Dr. Small, Dr. Kevin Kelly, where we have a history of developing the prognostic model, Dr. Chuck Ryan, Dr. Emmanuel Antonarakis, but also this funding provided a great opportunity for our junior trainees. We have five post-docs between the University of Minnesota and Duke University who learned so much, and I'm really thrilled that we're passing this information to the next generation of scientists.

Alicia Morgans: So we're learning for our patients, we're educating our clinical teams, our research teams, and we're coming together across the country in many ways to do this work so effectively. So thank you for sharing that. Thank you for highlighting your team, and I appreciate your time.

Susan Halabi: Thank you very much, Dr. Morgans. It's really a pleasure talking to you, as always.