Neeraj Agarwal: Hi, my name is Dr. Neeraj Agarwal. I'm a professor of medicine and Director of Genitourinary Oncology Program at the Huntsman Cancer Institute, University of Utah in Salt Lake City. I'm very happy to welcome two guests on our program today, Dr. Irbaz Riaz, faculty in medical oncology, assistant professor in Mayo Clinic, and a post-doc fellow in Mayo Clinic, Dr. Syed Arsalan Naqvi.

So Arsalan and Irbaz, welcome to the program, and thank you for taking the time.

Irbaz Riaz: Thank you, Dr. Agarwal. It's a pleasure to be here with you.

Syed Arsalan Ahmed Naqvi: Thank you, Dr. Agarwal. Thank you for inviting us to present our work today.

Neeraj Agarwal: So, Arsalan, your presentation at the ASCO GU was very well-attended, and it raised a lot of questions and interest. You presented on the report from a living meta-analysis, if you will, regarding choice of androgen receptor pathway inhibitors by disease, volume, and timing of metastasis in patients with metastatic hormone-sensitive prostate cancer. Can you please tell us more about your research, your presentation, and then we can talk more beyond that?

Syed Arsalan Ahmed Naqvi: Of course. Thank you so much again, Dr. Agarwal, for giving us the opportunity to present our work, which you pointed out was basically based on choice of androgen receptor pathway inhibitors by disease volume and timing of metastasis in mHSPC, which is a report from the meta-analysis.

So I'm going to quickly start by giving a brief background about the topic itself. The pace of evidence generation in oncology is rapid. And this holds true for prostate cancer, as well. Hence, we initiated and are maintaining a living systematic review for systemic therapy for patients diagnosed with metastatic hormone-sensitive prostate cancer.

A previous report of this effort showed that androgen receptor pathway inhibitors, with ADT, with or without docetaxel, is the standard of care therapy for all patients with mHSPC. This was published in JAMA Oncology in 2023. So based on the current evidence, you think about these patients in four clinical prognostic subgroups based on timing of metastatic presentation and volume of disease as defined by the CHAARTED criteria.

These patients—they are patients at one end of the spectrum, with low-volume metachronous disease. And then there are patients with high-volume disease and synchronous metastasis at the other end of the spectrum.

So our previous efforts showed—our previous network meta-analysis showed that patients who have low-volume metachronous metastasis—the preferred choice of therapy may be ARPI doublet therapy. And those who have high-volume synchronous metastases—the preferred choice of therapy could be triplet therapy. However, in patients with either high-volume metachronous metastases or those with low-volume disease and synchronous metastasis, the preferred choice of treatment could be triplet therapy or ARPI triplet therapy.

So now, in this setting, we have several ARPI agents available as potential treatment options, including abiraterone acetate, apalutamide, and enzalutamide. Recently, the ARANOTE trial demonstrated a statistically significant AFS benefit with the addition of darolutamide to ADT.

Now, this availability of multiple ARPI agents presents a clinical challenge on how do we determine the optimal choice for each patient. So understanding these nuances is critical for decision making. Therefore, we leveraged our living, interactive evidence synthesis platform to conduct an analysis assessing the comparative efficacy of different ARPI agents by volume of disease and timing of metastatic presentation.

Our analysis was conducted using the living interactive evidence synthesis framework, which allows for continuous updates as new data emerge. While we continuously update the entire evidence base, this particular analysis focused on phase III RCTs evaluating intensification with ARPIs in patients with mHSPC. And trials that reported clinical outcomes such as PFS and overall survival by volume of disease and timing of metastatic presentations were included in this analysis.

Additionally, P-scores to quantify comparative efficacy were also calculated, which measure the certainty that a given treatment is superior to its alternatives. Higher P-scores reflects, potentially, greater efficacy.

Now, in terms of results, the most recent update of this living evidence profile for mHSPC patients included a total of 11 trials with approximately 12,000 patients and 12 unique treatment options. The results were consistent with prior updates. However, this particular analysis is a subset analysis from the living meta-analysis, which supports data from only six trials with approximately 7,000 patients and four unique ARPI agents.

We will start by discussing the results for PFS by volume of disease. So on the left side of the panels, we have league tables, which outline mixed treatment comparisons. And on the right, we have forest plots. The way to interpret league tables is by reading them from top, treatment, to side, comparator. Green color indicates benefit, while red indicates lack of benefit, with darker shades of color representing a statistically significant effect.

So for high and low volume disease, we can see that there are no statistically significant differences among different agents. The results were consistent by timing of metastatic presentations in patients who had synchronous disease and those who had metachronous disease. We saw a similar pattern of results for OS, though it should be noted that the results for the ARANOTE trial is based on data with limited follow-up.

So in conclusion, based on the current evidence, there are no statistically significant differences between ARPI agents when comparing them across disease volume and metastatic timing in mixed treatment comparisons.

So how do we pick an ARPI? Choosing the right ARPI is challenging and requires a patient-centered approach. There is no single best option. So decision making should factor in cost because some agents might be more expensive or have different insurance coverage. Accessibility of certain drugs may vary based on regions, and then there are different drug interactions as well that a clinician should account for.

And finally, toxicity profile is also very important. Each ARPI has a distinct side effect profile that must be considered based on patients’ comorbidities and treatment tolerance. In the last, evidence from well-designed real-world studies may offer additional insights.

I would like to thank all the co-authors and collaborators who have worked on the living evidence profile for mHSPC, particularly my mentors, Dr. Riaz, who is here with us today, and Dr. Alan Bryce from City of Hope. Thank you very much.

Neeraj Agarwal: So, Arsalan, very nice presentation. And basically, the gist of the presentation for our audience is we have four different ARPIs out there—abiraterone, apalutamide, enzalutamide, darolutamide—and they all seem to be beneficial in our patients with metastatic hormone-sensitive prostate cancer, regardless of disease volume, and time of onset of metastasis, so de novo disease versus metachronous disease or non-de novo disease.

Especially in those patients who are presenting with metastatic hormone-sensitive prostate cancer, high-volume but metachronous, and the other category would be the low-volume but the synchronous or de novo. Is that correct?

Syed Arsalan Ahmed Naqvi: Yes, this is correct.

Neeraj Agarwal: Yeah. And you raised a very nice point about—the choice of these agents may be driven by the cost, accessibility, drug interactions, and many other social, clinical aspects which are not really related to the efficacy of the drug shown in a clinical trial.

And in my view, these findings are very pertinent. Very similar findings were presented by a STOPCAP M1 investigator, Dr. David Fisher, who also—who we interviewed on this platform last month. And again, I think the data support, from these large—huge analyses, that choice of ARPI is—doesn't matter what you choose, but do choose ARPI with ADT.

We have another problem, which is going on in the context of implementation medicine, that many of our patients with metastatic hormone-sensitive prostate cancer are only getting androgen deprivation therapy monotherapy, and they're not even being offered ARPIs. So I think that is one big lesson from this.

Let me switch gears and ask you about this living network meta-analysis. It's amazing how you update your platform on a continuous basis. It's like a continual process. You don't wait for doing any new analysis after every year. It is continuously updated. So tell me, Irbaz, how do you do that?

Irbaz Riaz: Yeah, thank you. As you can imagine, it's a really exciting project for us. And I think you asked two good questions here. The first is, what is a network meta-analysis? And second, about how we keep the evidence living in a regular meta-analysis, a pairwise or a network meta-analysis.

I'll start by answering the question about network meta-analysis. So network meta-analysis is a statistical framework that allows for indirect comparisons while doing its best to preserve the randomization.

Many important questions in oncology are now really indirect comparisons, basically. For a patient who is a new patient with castration-sensitive disease and metastases, the question, often, we face in the clinic is whether we give a triplet versus a doublet. And the trials compare a triplet against ADT and doublets against ADT. And trials show that triplet is better than ADT. Doublet is better than ADT.

So the real question is then, how do we pick between triplets and doublets? And this is where the network meta-analysis comes in. In the setting of a common comparator, like ADT, the network meta-analysis technique allows us to compare a triplet versus doublet. There are obviously statistical assumptions which must be met before this comparison is made. But in my opinion, it's a very appropriate framework and very well-defined framework that allows us to make important decisions for our clinical practice.

Neeraj Agarwal: So you raised a very good point. So before we move forward, I would—for our audience especially, I'd like to highlight that triplet therapies have not been compared with ADT plus ARPI doublet. And I don't think we are going to see that trial happening any time soon, in the near future. The field has already moved on, and we are going to see several new therapies coming up.

But we don't have that question answered, whether ADT plus ARPI plus docetaxel is better than ADT plus ARPI. And network meta-analysis gives us a fantastic avenue to address those questions, which cannot be answered by a clinical trial. So I agree with you.

Irbaz, the second part—you were going to talk about another aspect of living network analysis. Can you please elaborate on that?

Irbaz Riaz: Yeah. So AI is hot these days, especially these large language models. They are one of the biggest inventions of the century. So with the modernization and improvement in AI methods such as large language models, what it allows us to do is do assessment of evidence, summarizing it, and generating insights from it.

So we have built a framework based on these language models, plus other informatics techniques. And what we do in that context is for a given topic, such as metastatic castration-sensitive prostate cancer, we constantly scan new evidence. The language models then assess for relevance. These language models can also assess data from the relevant articles, put it together, summarize it, and they can also help with, actually, the analysis part as well.

So essentially, we have created a pipeline in which, as soon as new evidence comes, we look for it every week, assess for relevance. If a new paper is relevant to the question of interest, we add it to our analysis and update the results. And the results are pushed to a website, which is maintained, living.

We don't update the changes in real time. We stage the updates when the new relevant evidence becomes available. But in the back end, the evidence synthesis is happening all the time. Whenever there is a meaningful update, we push it to the external website so that it could be released, even without an actual publication.

Neeraj Agarwal: That's wonderful. And that seems like a very cutting-edge technology. So for our viewers, you are—tell me, how do you actually perform this AI-assisted living evidence synthesis? Seems like this is the core of what you do to maintain this living network analysis.

Irbaz Riaz: Exactly.

Neeraj Agarwal: So how do you use this tool? What kind of AI you are using?

Irbaz Riaz: So there are different pieces—there are different steps—in performing a living meta-analysis. And for each step, we have carefully thought through and picked the best model.

For instance, there are steps like analysis—you don't need an AI model to analyze it. We just program it, and we just automate the analysis. But things like looking for new evidence, things like deciding whether a new paper that was found in a PubMed search—is it relevant to metastatic castration-sensitive prostate cancer? And once we have decided that it is relevant, we have to extract the data.

Those steps we use large language models—the ChatGPT-style, the GPT-4-style models—they are very good at looking at a paper and making some decisions about it.

Obviously, this is high-stakes work, right? We cannot leave this all to AI because AI can still make mistakes. So we have created a framework in which different AI models work in tandem, checking the work of each other, working together, and then creating the final output that is delivered to a human—that is delivered to a human researcher who finally approves or disapproves the work of the AI models.

Neeraj Agarwal: That sounds so futuristic, but I'm so glad it's happening right here, done by our colleagues at the Mayo Clinic. So congratulations. So I'm really hoping that this will ultimately allow us in the clinic to truly personalize medicine.

If a patient comes, I'm really hoping to use a tool like this, maybe learn from you guys, to incorporate not only the data from the clinical trials, but patient-specific data at the time of a given presentation, and create the prognostic model for a given patient at a given point of time in the clinic, and choose the best therapy for the patients, especially with so many therapeutic options being available and more coming in the next one or two years.

So congratulations again for the groundbreaking work, paradigm-shifting work, and I'm really hoping to see you soon in one of the meetings.

Irbaz Riaz: Thank you so much. I'll just add that we have generated living evidence profiles now for metastatic-castration sensitive prostate cancer, metastatic-castration resistant prostate cancer, first-line kidney cancer treatment, first-line urothelial cancer treatment. And we are just constantly adding topics. So we are really excited to create a new evidence ecosystem so that evidence is up to date all the time. And we are adding features using AI that—you can interact with the evidence. Happy to talk more, perhaps, in a separate session about that.

Neeraj Agarwal: Absolutely. Looking forward to it. Thank you again, Irbaz, Arsalan, for taking the time today.

Irbaz Riaz: Thank you so much, Dr. Agarwal. It was a pleasure chatting with you, and thank you so much for having us.

Syed Arsalan Ahmed Naqvi: Likewise. Thank you so much, Dr. Agarwal, for having us and for giving us the opportunity to present our work. Thank you.