Read the Full Video Transcript

Evan Yu: Hello, my name is Evan Yu. I'm at the University of Washington Fred Hutchinson Cancer Center. I'm a genitourinary medical oncologist and I have the pleasure to be here today chatting with my good friend and colleague, Oliver Sartor, who recently moved to the Mayo Clinic and just doing wonderful things in sunny Minnesota. Thanks for being here today, Oliver.

Oliver Sartor: Thanks, Evan. Pleasure to be here.

Evan Yu: You were an author on a pretty interesting abstract that was from the VISION trial, looking at Lutetium-PSMA-617 and trying to pick out predictive and prognostic markers at baseline. And so first off, it's great to have you here today. How are you enjoying Minnesota so far?

Oliver Sartor: Well, it depends on the day, of course. There are days that it's minus 11 degrees and they're, shall we say, challenging. Other days are fairly pleasant, but glad to be here.

Evan Yu: Yeah, well, I guess it's good. And on those minus 11 degree days, you can get a lot more work done. Right?

Oliver Sartor: Of course. You don't have to wander around outside and get sunshine or anything. You could skip all that sunshine.

Evan Yu: Excellent, so let's talk a little bit about this abstract. I thought that it was really interesting, but the one thing I want to talk about before we dive into the details is the term predictive. I hear this all the time, people talk about something being predictive when really it's prognostic. For our viewers, can you just tell how do you define predictive versus prognostic?

Oliver Sartor: Yeah, great question. So prognostic is really easy. It means it's a potential biomarker. And by the way, a biomarker, it could be a scan, it can be a lab, it can be a bone scan, it can be opioid usage. There are all sorts of things. And if that happens, you're going to do worse. It really doesn't depend on the therapy you're on. You're just going to do worse. And that's prognostic because it doesn't really predict much, except you're going to do worse.

Now, predictive is a little different. Predictive means that you might do better with one therapy over another. Let's take a classic predictive biomarker, let's say BRCA2. If you've got a BRCA2, we can predict that you're going to respond to a PARP inhibitor and that's a great predictive biomarker. Now, it's also prognostic because it's a poor prognostic finding with a variety of other therapies, but it's predictive for the use of PARP inhibitors. So prognostic means you're going to do bad, predictive means you're going to do better with one therapy over another.

Evan Yu: Great. Thanks for clarifying that because I see this all the time and people will sometimes use the term predictive and it might be a single-arm study. And you can only establish predictive if you have a randomized trial and it's tied to that therapy.

Oliver Sartor: Exactly.

Evan Yu: Exactly. Right.

Oliver Sartor: In VISION, by the way.

Evan Yu: Exactly. And so let's talk about this abstract. So my understanding is you took the patients enrolled on the VISION trial and you took some of the baseline markers. I think it was 29 markers, is that right?

Oliver Sartor: Yeah, a bunch of markers. Bunch of markers.

Evan Yu: And you really drilled it down based on looking at its association both in univariable and multivariable fashion, association with overall survival, radiographic progression-free survival, and PSA response or PSA decline 50% or greater. Is that correct?

Oliver Sartor: That's correct.

Evan Yu: All right, so what came out?

Oliver Sartor: Well, first thing is the PSA decline, the rPFS, and the OS are not the same. And that's an important sort of first pass at the data. Now, if you want to go about PSA decline, you want to know particularly about PSMA SUV mean or max. Now, let's talk about that for a second. PSMA PETs we all know, but how do you quantitate them? And how do you say, "There's a lot of uptake here or there's less uptake there. And what about heterogeneity and what about max? What about mean?" And I'll simply say that SUV mean, which is the one that we really analyzed, it turned out to be the best single image-based biomarker, is complex because you in essence have to go look at all the lesions and then take an SUV mean. Now interestingly, we did SUV mean and SUV max and they were not that distinct, so pretty damn close.

But for things like PSA, then the PSMA PET uptake at baseline is a really good predictor and a predictive biomarker as well as potentially prognostic biomarker. If you are in the control group, these patients do pretty badly, but here's what's interesting. When you start looking at overall survival, there's a whole lot of other important things other than just a scan. You can't just look at the scan and predict what's going to happen. And guess what? A lot of the things that we saw are the things we're very familiar with. Let's go back to the old Halabi nomograms. Okay, hemoglobin, guess what? If you're very anemic, you're going to do poorly. You're going to die sooner. High LDH, high alk phos, opioid usage, these are the type of parameters that we're accustomed to in a prognostic nomogram like the Halabi nomogram. And we found them all over again in a careful prospective analysis, 29 different variables, overall survival endpoint. It's actually a very solid analysis.

Evan Yu: Great.

Oliver Sartor: And I'll just simply say it was really reassuring in a way to see those familiar biomarkers come up, but now we have the PSMA PET as an alternative, too. So that's a big broad overview just in a couple sentences.

Evan Yu: Right. Great. So those are well-known prognostic markers.

Oliver Sartor: Yes.

Evan Yu: Were they predictive as well?

Oliver Sartor: No, no. The things like the LDH and the alk phos and the opioid usage, hemoglobin, those were not predictive. The ones predictive is the PSMA PET uptake, which you kind of expect. You see a lot of target, you've got a ligand that's going to bind to the target. You can radiate more to the target when you got it and it predicts you're going to get a response. Is that surprising? Not to me. That's why we laid out the trial, but excluded individuals with very low PSMA PET uptake in the first place. Now, I might sort of leap to another abstract to help provide some enlightenment. Would that be okay?

Evan Yu: Yeah, sure.

Oliver Sartor: Okay, so Phil Koo presented at the EANM another analysis looking at the PET scans as a predictive biomarker. And one of the things that they did there was really sort of shave down and look in great detail at the various hazard ratios, rPFS and OS, as a function of PSMA SUV mean. And what's really interesting is a lot of linearity. I'm going to call it shades of gray. There's not an individual cut point. You can't say, "Your SUV is below five. You're going to do terrible." And you can't say, "Your SUV mean is seven and higher. You're going to do great," but there's a beautiful curve in there where you see the gradation. And the higher the SUV, the better you're going to do, but there's not a cut point. So it's got shades of gray.

Evan Yu: So a continuous variable?

Oliver Sartor: It's like a continuous variable.

Evan Yu: Okay.

Oliver Sartor: And when we came to the nomogram paper, if you look at the nomogram, it's basically a continuous variable. And that really is true.

Evan Yu: Not surprised.

Oliver Sartor: Yeah, it's not surprising, but it was kind of reassuring to see and it was interesting to me how well the imaging would be able to predict certain important parameters. Now, we also go a little bit further and I'm probably rambling on now, but if you delete the imaging altogether, guess what? Those old familiar prognostic biomarkers are still critically important. So guess what? If it was a short interval between the time of diagnosis in prostate cancer and when you get treated with Pluvicto or it's a high LDH, high alk phos opioid using patient, those patients are going to do poorly. And that's independent of the actual biomarker on the imaging.

Evan Yu: Got it.

Oliver Sartor: Anyway, lots of interesting stuff there.

Evan Yu: That's great. Now, just remind me and the audience. When you say mean SUV, are you talking about the total body mean SUV that had been done in those analysis where 10 was kind of ... They use it for simplicity. 10 was a cutoff, but I think we all know it's more of a continuous variable. Is that right?

Oliver Sartor: Yes. So the SUV mean is literally an arithmetic average of the lesions that you see on the scan. And it's not easy to do. It takes a little software and a little bit of time to do, which is one of the reasons I didn't like it. The SUV max can substitute for the SUV mean.

Evan Yu: And that's what you used in this, right?

Oliver Sartor: We used mean, but max also works pretty well.

Evan Yu: Right.

Oliver Sartor: And max is more practical, even I ...

Evan Yu: Way more practical.

Oliver Sartor: I can go and look at the image and I can circle the spot and say, "That's ...

Evan Yu: Because I've been told by my nuke med colleagues that total body SUV mean is something that is not necessarily easy to reproduce from, nuke med to nuke med. And it takes a lot of time. Is that right?

Oliver Sartor: Correct.

Evan Yu: Yeah.

Oliver Sartor: Now, we're moving into the AI world and that AI world is going to fix the problem. And I've seen stuff mentioned like Exini. They're working hard to be able to do this in an automated fashion. They do segmentation, they do automatic, then they get visual like cognitive verification. And then after you have the cognitive verification, they put everything together and, boom, just spit it out. So there's technology coming that will reduce that variability, but right now if you go to Topeka, Kansas, Peoria, Illinois, and Washington DC you may get three different calculations for SUV mean because it's not fully automated yet.

Evan Yu: Okay, got it. All right, so what's our takeaway? What should we learn from this? Can we apply this to our clinic? How can we take this nomogram and start using it?

Oliver Sartor: Yeah, so number one is I feel like I'm better informed about a patient's prognosis. I can actually potentially manage that patient a little bit better. If they're falling into a particularly good prognostic category, I can be a little more reassuring to the patient. If I can look at a patient and sort of say, "Look, this is not looking so good," I think it's incumbent upon us to inform the patient of the parameters that we know can influence their future. So that's one. Now for the predictive stuff, well, guess what? I can say this is a really good therapy, let's give it a shot, but you know as well as I do, Evan, that at times the therapy that's right for that individual is also predicated on the other options available. If I've got a great clinical trial over here and then I have a so-so outcome I might predict on the Pluvicto, then maybe I want to steer that patient through the great clinical trial.

Evan Yu: Makes sense.

Oliver Sartor: And cabazitaxel. Don't forget about cabazitaxel. It's a real drug. We saw that in TheraP, the overall survival outcomes identical between the PSMA lutetium and the cabazitaxel in the TheraP trial. Phase two, not phase three, but what did we take home? There are important parameters at baseline that we can utilize to look at prognosis for a patient. And there's predictive biomarkers in terms of the PSMA PET uptake that we can utilize in our practice. I think those are pretty practical and good take-home messages.

Evan Yu: Excellent. Well, thanks so much, Oliver. This has been really great. It's a wonderful abstract. Thanks for making it simple for us to understand and giving us the practical take-home messages.

Oliver Sartor: Right. Thank you so much, Evan. Glad to be here.

Evan Yu: All right, take care.