Susan Halabi: Thank you so much, Andrea, for having me.
Andrea Miyahira: So tell us about the goals of the study and the clinical needs that you were aiming to address.
Susan Halabi: Yeah. The primary objective of our analysis was to develop and validate a prognostic model for progression-free survival, actually radiographic progression-free survival, which I'm going to refer to in the talk as rPFS in mCRPC patient who received first-line androgen receptor pathway inhibitors. And the reason why we wanted to do that is that rPFS is a clinically meaningful endpoint in advanced prostate cancer, and overall survival is no longer going to be possible to use as a primary endpoint. And although there are several models of overall survival that have been validated and a few of those models include tumor genomics, there are no existing models to our knowledge that have been validated in mCRP patient who received first-line RP therapies. So we felt there is an unmet need and we sought to address this gap.
Andrea Miyahira: So you used data from the Alliance trial to do this study. Can you just remind us what were the major findings from that study and how was the trial designed?
Susan Halabi: Yeah, the Alliance study is a randomized phase-three trial where it randomized 1311 patients to either enzalutamide or enzalutamide plus abiraterone acetate. Now, if you recall the combination of enzalutamide and abiraterone did not show treatment benefit in overall survival. However, rPFS was impacted by the treatment where the combination did better. There was a superior treatment benefit in the combination arm compared to enzalutamide alone.
Andrea Miyahira: So describe the methods of the study when the CTA was collected and what other features were assessed.
Susan Halabi: Yeah, we were lucky. As you know, this is funded by the Alliance. So we had sort of, I would say, a complete follow-up on the patient. And in addition, they collected plasma at baseline, at three months, and at progression. So we were able to leverage that data.
This work is funded by an R01 with three PIs, myself, Dr. Andy Armstrong, and Dr. Scott Dehm. So Dr. Dehm in his lab developed this assay that capture tumor fraction. It has copy number variants, it has pathogenic variants, and also, which is different than the other existing assays that you have like foundation medicine, what's special about this assay, you have AR androgen receptor genetic structural rearrangement. And you can develop this panel from only 1 to 3 ml of plasma. And this assay has been validated. Actually, we previously developed and validated a model of overall survival using that assay. And this has been published in European Urology in 2026.
Andrea Miyahira: Talk about the data methods also.
Susan Halabi: So again, here, the critical part is we know there are established clinical factors of overall survival and of radiographic progression-free survival, and these are hemoglobin, PSA, LDH, alkaline phosphatase, pattern of spread, pain, and ECOG performance status.
However, what's novel about this model is we wanted to combine 74 genomic variant with the clinical variables. And what we used is a machine learning approach, which is called Random Survival Forest. And the whole idea is to test whether if you use only the clinical variables versus a model with clinical and genomic variables, do we increase prediction of radiographic progression-free survival in those patients? Which is really important because as you know, prostate cancer is heterogeneous. So we want to make sure that there is a better risk stratification and for better design.
Andrea Miyahira: So what were your major findings? Were there any tumor genomic alterations that were prognostic for rPFS? And did they differ between the treatment arms?
Susan Halabi: Yeah, that's an excellent question. So we had several obviously genomic alteration that predicted rPFS. So we have AR gain, AR enhancer, which most assays out there don't even look at AR enhancer. So this is really something that's new. So AR gain, AR enhancer. Then we have MEC gain. We have FOXA1 and CCND1. In addition, we have P10 loss and TP53 loss. So these are the key genomic drivers of resistance or of progression that we found.
And in addition, the key drivers of clinical variables were hemoglobin, PSA, alkaline phosphatase. So this has come as no surprise to us. And as I said, there has been an improvement in prediction like how you can classify patient. So you classify patients as risk and they are really truly risk. So when you say high risk patients or poor risk patients are patients who die quicker or progress quicker than patients who were in the low risk.
Now it is important to note here, which I forgot to indicate to you, Andrea, that the rPFS in the trial was defined as purely radiological progression-free survival. So it did not include PSA or clinical progression. And the increase in predictive accuracy went from a clinical model of 0.66 to a clinical and genomic model that went up to 0.73. And this translate basically to 30% improvement, net improvement in correct risk classification. And that's really major. In my opinion, it's considered clinically meaningful.
And then once we did the model, the next step obviously is to translate that model into actionable categories. So what we did, we created prognostic risk groups where we're able to categorize patients into either low, intermediate or high risk group, or poor risk group. So I'm using high and poor in the same way. And patients in the low, intermediate and poor risk group have medians of 40 months versus 23, sorry, versus 25 versus 13. And there is a clear separation in the Kaplan-Meier curves.
Andrea Miyahira: Thanks. And what are the next steps in this study?
Susan Halabi: So we did not have external data set to validate this model. We used an internal tenfold cross-validation and we developed a tool that a clinician can use to input that data. So this tool could be used in clinical trial designs and in risk stratification.
We are looking at other clinical data, whether coming from clinical trials or from real world data to validate the model. And once this is validated, then I think we can deploy that tool to the scientific community.
Andrea Miyahira: And what are your major take home messages for the clinical biomarker space, including how liquid biopsies can be used?
Susan Halabi: Yeah, this is so exciting because now with ctDNA, this is a non-invasive biomarker. So we're taking it from the lab; we can take action. It could be actionable. And using models that integrate genomic and clinical variable is going to be far more superior in making sure that the treatment arms are balanced in clinical trials, it doesn't take a lot of specimen, only one to three, it's not inconvenient to the patient. So in my mind, this is really easy to implement.
And of course the big question is whether the users, the clinicians, will prescribe that to the patients. I think the patients, from the few patients I talk to, they're receptive to the idea of knowing their genomic mutations.
Andrea Miyahira: Well, thank you so much, Dr. Halabi, for joining us and presenting this today.
Susan Halabi: Thank you. It's my pleasure. Thank you for having me. As always, I'm delighted to speak to you, Andrea, and to share the knowledge gained from our research with the community of UroToday. Thank you.