Soumyajit Roy: Thank you. Thank you, Dr. Agarwal. It means a lot.
Neeraj Agarwal: Yeah. I had the honor of being a co-author on this paper, but I see that Dr. Dan Spratt, Dr. Karim Fizazi, Dr. Kim Chi, Dr. Maha Hussain, who actually originally established the concept of PSA response as a surrogate for overall survival in metastatic hormone-sensitive prostate cancer and many other esteemed faculty like Dr. Chris Sweeney, Dr. Amar Kishan. All of them are here. We would like to hear from you about this publication, what it means for our practicing clinicians who are out there, practicing urologists and medical oncologists. Congratulations again, and please tell us about your paper.
Soumyajit Roy: Sounds good. Thank you so much, Dr. Agarwal, for this opportunity to present our work today on this UroToday platform. The work, as you described, was recently published in Nature Communications where we trained and validated a tool to predict early favorable PSA response in metastatic hormone-naive prostate cancer patients treated with ADT plus an androgen receptor pathway inhibitor or called ARPI. So why do we need to predict early PSA response in metastatic hormone-sensitive prostate cancer population. Well, as you mentioned, Dr. Maha Hussain has shown and subsequently many others, including Dr. Simon Chowdhury, you, all of you have shown that early PSA response less than 0.2 nanogram per mL by six to seven months of treatment initiation is one of the strongest predictors for overall survival in metastatic hormone-sensitive prostate cancer patients.
However, waiting for six months to assess treatment response could delay critical intervention decision and disease progression during this window may result in a more treatment-resistant state. There are certain risk-stratification tools at present, for example, disease volume by CHAARTED definition or risk stratification by LATITUDE criteria, but they have limited accuracy. For example, in patients with favorable PSA response, more than 50% could still have high-volume disease. And unfortunately, there is no good pretreatment prediction tool that exists to identify patients who will achieve early PSA response after treatment with ADT plus an ARPI. So our work, this NADIR model, actually provides a solution to a number of these problems. So what we did is we developed a multivariable simple logistic regression model to predict probability of early favorable PSA response, which enables risk stratification and provides baseline prognostic insights. And it supports earlier treatment intensification or deintensification decision before any change in the disease status occurs. However, I want to emphasize one thing, it complements formal PSA monitoring. It doesn't replace it. So what this means for our patients is that if we could identify low risk or the early favorable responders, we could spare them some unnecessary treatment, while high risk non-responders could receive upfront therapy intensification when the therapeutic window is still open.
This supports trends towards concomitant over sequential treatment personalization. So how we designed this study, how we trained the model or validated the model, well, we included individual-patient data from the ADT plus ARPI treated arm of LATITUDE, TITAN and ARASENS. The three ARPI agents in these three trials were abiraterone, apalutamide, and darolutamide respectively. We combined them and split them randomly into a 60:40 ratio to a training and an internal-validation cohort. The model was trained in the training cohort, then the model was locked and tested in the internal-validation cohort. And subsequently, it underwent further independent validation in the ADT plus enzalutamide treated cohort of the ENZAMET trial, which was sponsored by the ANZUP Group. So what were the variables that we included in the model? Well, very simple variables like baseline PSA, hemoglobin, body mass index, age, disease volume by CHAARTED definition, whether the patients had visceral metastasis at presentation or not, and then what was their metastatic stage at presentation, whether they were de novo metastatic disease or recurrent metastatic disease. And as I mentioned, the endpoint was a binary endpoint, whether the patients achieved a PSA of less than equal to 0.2 nanogram per mL by six months of ARPI initiation. So you can see some of the key model performances here.
Our model actually achieved robust discrimination and calibration across the two validation cohorts. As you can see here on the left-hand side, the model showed good ability to distinguish between individuals with and without the favorable PSA response, and the AUC or the area under the curve score was 0.82. And the model was also well calibrated with a Brier score of 0.16. And as you can see in the calibration plot that the model predicts a certain level of risk or probability, and the real outcomes are very closely aligned to that level. So you see on the X-axis, you have the prediction and on the Y-axis is the observation, and you see they're very well aligned. Now, we stratified patients based on the tertile of predicted probability of early PSA response based on our model. And as you can see, the patients were well-spread by the predicted probability of PSA response. As you can see, the early PSA response rate in the upper tertile are those who had the highest probability of PSA response based on the model was 92% in the external-validation cohort compared to 39% among those who had the lowest probability of PSA response as predicted by our model. And as I mentioned, because PSA response early, like at six to seven months, after treatment initiation is a strong surrogate for overall survival, we looked at the four-year overall survival across these three tertiles.
And as you can see, there was a 24% difference in four-year overall survival between those who were in the upper tertile versus those who were in the lower tertile for whom the four-year overall survival was 56%. So what this means, how to use this model in clinical practice? Well, number one, as I mentioned, it can provide prognostic insight and health risk stratification at baseline, and it can guide some early treatment decisions rather than waiting for six months to assess treatment response. Further, it opens up opportunity to help design further biomarker-driven trials, and we all know that there are new innovative biomarkers coming up, for example, digital pathology, liquid biopsy, imaging biomarkers, so on and so forth. So this model basically provides a foundation for measuring any added value of these advanced novel biomarkers on top of this simple prognostic tool. And the next steps for clinical translation from our side would be to prospectively validate this model and then integration, as I mentioned, with additional novel biomarkers, try and establish this model as a basis for enriching patient population for biomarker-driven trials. We need to focus on implementation science, and we already have an online tool to use, which is free. And then further, and most importantly, we need to have a shared decision-making with patients, and this tool should help in patient counseling.
With this, I would like to thank Dr. Dan Spratt, who is the chair of radiation oncology at the University Hospitals Cleveland Medical Center, and many other authors, including you, Dr. Agarwal, Dr. Maha Hussain, Dr. Karim Fizazi, Dr. Kim Chi, Dr. Simon Chowdhury, Dr. Neal Shore, Dr. Fred Saad, and of course, my two statistical mentors, Dr. Susana Halabi and Dr. Yilun Sun, for their tremendous support. Without them, this study could not have been completed. Thank you so much.
Neeraj Agarwal: That's a great presentation of your data. So just for clarity, all patients with newly diagnosed metastatic hormone-sensitive prostate cancer should be offered treatment with ADT plus ARPI. That's standard of care. And we do not know until six or seven months what the PSA response is, and then we start prognosticating based on the PSA at six months. So your model actually predicts who is going to develop or achieve a PSA-undetectable rate or undetectable response or who is not going to achieve a PSA-undetectable response. And that provides an opportunity to us for our patients to modulate treatment upfront. So if I know my patient is not going to have a PSA response, which is we call it less than 0.2 or 0.2 or less, I may push for more aggressive therapy upfront. There are multiple biomarker-based targeted therapies coming up and docetaxel is always there as a triplet therapy. So I may talk to patient about intensifying beyond ADT plus ARPI upfront without losing the window of opportunity of first six months.
However, if I know from your model, a patient is going to achieve a PSA response, which is 0.2 or less, then maybe I can reassure patients. It will be tremendously helpful for prognostication and counseling and how I'm going to follow the patients, those patients, and maybe offer them deescalation trials down the line based on the PSA response, which is going to be great. It is another step towards personalization of medicine in patients with metastatic hormone-sensitive prostate cancer. Am I correct?
Soumyajit Roy: That is very true, Dr. Agarwal. Based on the absolute-risk estimates, first of all, there is a difference between prognostic tool and predictive tool. Our tool is a prognostic tool which tells us whether the patients... What is the risk or what is the probability that the patient is going to achieve PSA response or the other way? Now, it doesn't tell whether the patient is going to benefit from therapy intensification or deintensification or not, but in a common physician mind, if they see that the absolute risk of not achieving early PSA response is high, that guides them to consider therapy intensification versus if they see that the risk or the probability of achieving early PSA response is very high, they might already start thinking, "Okay, at six months, if I see a very good PSA response, the patient achieves nadir by six months, I might gradually come out of the ADT-ARPI combination, maybe considered just intermittent ADT or maybe even ARPI monotherapy," so on and so forth. So even though it's a prognostic tool, the absolute probability estimates could be used as an actionable tool.
Neeraj Agarwal: And I think first of all, it's huge for our patients from prognostication, counseling perspective, and how do I monitor those patients? For example, if my patient is expected to have a poor survival and a rapid disease progression, I would probably change the way I scan them. I obtain CT and bone scans. And we saw from recent trials that PSA may not be the best marker for those patients who have poor-prognostic disease. So I think this research has implication beyond the academic discussion and obviously has the potential to impact how we follow our patients in our clinic beyond prognostication and counseling and how we offer them different types of treatment. You mentioned the online tool which is available out there. It would be awesome if you could share the online tool to accompany this video so that our colleagues out there in the community and academia can use that tool.
Soumyajit Roy: Absolutely. Absolutely. I am happy to share the tool, like the link to the tool through the online video, and that would help. As you mentioned, a lot of patients even can put their information on the tool and they can see what is their probability of getting PSA response or not.
Neeraj Agarwal: And by the way, Soum, the tool requires very basic information, which is readily available, such as the baseline hemoglobin and BMI and so on. So I think you're right, patients can use it by themselves.
Soumyajit Roy: That is correct, Dr. Agarwal. And that's the beauty that there is, as you mentioned, there is importance or practical abilities beyond academic publication or just having an impactful publication. As long as we can implement what we are doing research-wise, that's what is more important. It should eventually transform the lives of our patients.
Neeraj Agarwal: Well, congratulations, first of all, for publishing this seminal article, and congratulations to you and your mentors, including Dr. Dan Spratt. And thank you for sharing your wisdom with us today.
Soumyajit Roy: Thank you. Thank you for the opportunity once again. I'm deeply indebted. Thank you so much.