The Uromigos Episode 108: Peter O’Donnell Gives 5-Year Follow-Up Data on Front Line Pembrolizumab

Tom and Peter O’Donnell continue to discuss and debate long-term follow-up data of pembrolizumab monotherapy in initial treatment of cisplatin-ineligible metastatic urothelial cancer. View the ASCO abstract.

Episode Transcript

Mike:
All right. Welcome, everyone to our series of ASCO podcasts. We’re joined today by Peter O’Donnell from the University of Chicago. And we’re going to talk about his abstract regarding the Keynote-052 long term follow-up from the Phase 2 study and also work in some of the 361 Phase 3.

So, Peter, thanks for joining us. Do you want to maybe briefly introduce yourself, and then give us sort of the high-level summary of data, and then we’ll launch into questions.

Peter O’Donnell:
Sure. As you know, I’m a medical oncologist at the University of Chicago. I see mostly patients with bladder cancer, urothelial cancer, and it was my pleasure to present these long-term follow-up data for pembrolizumab in the frontline cisplatin ineligible population on behalf of my co-authors as well.

And the high-level results here that we’re showing, these are now data five years after the last patient was enrolled in the trial, the objective response rate has not changed from what’s been previously published for the overall population.

We’re seeing just under 30% objective response, but in the high PD-L1 population, it’s almost 50%. And I was actually surprised to remind myself that even in the PD-L1 negative population, it was a 20% objective response rate. And then really the value of the data is looking at the long-term follow-up now that we have four or five years of follow-up on many patients.

And you see overall that the takeaway that I had was that it’s one of the five patients that are still alive five years later. And in the high PD-L1 population, it’s one of every three still alive, which I think is pretty impressive. And we have the duration of response, which is another way to think about these data.

If you have a patient who’s a responder, they’re likely to respond for a very long time. The median duration of response in all comers was almost three years in the high PD-L1 population not yet reached. So, five years later here, we’re seeing over half of the population still having ongoing responses.

Mike:
And, Peter, the pembro stopped at two years in this study, right? Like other pembro studies?

Peter O’Donnell:
Correct.

Mike:
So, it was 24 months of therapy. And obviously then, no real drop off after 24 months. Patients responding at 24 months obviously tended to stay in response. Is that fair?

Peter O’Donnell:
Exactly.

Tom:
And, Peter, any long-term toxicity?

Peter O’Donnell:
No, no toxicity signal is long term. I think we know that, right? Most of the immune checkpoint toxicities tend to happen in the first six to 12 months.

Tom:
And, Peter, anything exciting in the biomarker from a TMB or PD-L1 perspective?

Peter O’Donnell:
We didn’t look at that in this data set, so nothing new there.

Tom:
So, I guess one of the other questions would be, how do you look at this in conjunction with the confirmatory data, which didn’t show the same? I mean, there wasn’t as much activity in the randomized Phase 3. Response rates weren’t as good. The medians weren’t quite as good. The population was a better population, because remember, it includes the cisplatin eligible as well as the cisplatin ineligible population.

And one was a large single arm Phase 2 trial, and then the second was, which was published in Lancet Oncology a few weeks ago, the other was the confirmatory 3000 patient randomized Phase 3. Do you think the two data sets are aligned?

Or do you think actually the randomized Phase 3, which wasn’t successful, although the monotherapy arm wasn’t statistically tested, do you think that actually we were unable to confirm the findings from your Phase 2?

Peter O’Donnell:
No, no, I actually don’t think that at all that we were unable to confirm the findings. So, we’ve talked about this a little bit before, but the biggest difference between these two data sets, and I’ll point out for the audience that the 52 data, the one we’re talking about here, is still the largest dataset for cisplatin ineligible patients receiving pembrolizumab monotherapy.

The biggest difference is the biomarker, right? The PD-L1 positive percentage in the 52 data was under 30%. It was 29.7% of the patients were PD-L1 high. Whereas in the Keynote-361, I think it was about half the patients. Tom, you can remind me.

Tom:
It was. It was just over half, I think, so let’s go with 50%.

Peter O’Donnell:
So, what’s going on there is the question for the field, right? Because I think anytime you’re selecting “half the patients,” you’re not really selecting at all.

Tom:
Peter, can I ask …

Peter O’Donnell:
Yeah, go ahead.

Tom:
At the beginning part of 52, there was a bio component associated with that where patients had to have a biopsy. And five years ago, when we were treating upfront immune therapy, we were probably more nervous, and we probably were selecting better. And we were at a series of academic institutions, particularly in the US and some in Europe.

I wonder if, and I’ve spoken to yourself and a number of other people, [Alogen 00:05:21] and Matt and one or two others, Noah, the list goes on. And people tell me that they think they know who we think we can give up front immune therapy to.

They can’t put their finger on them and they can’t sort of write it down, but when they see a patient maybe relapse a bit of lymph relapse after cystectomy a year out, maybe lymph node positive, maybe quite young [inaudible 00:05:42] well that I’d like to give immune therapy to, is it possible actually that because of the compulsory biopsy, the initial part of the trial, and because we were more cautious around who we were treating, because remember, we had no idea whether immune therapy was going to perform well initially.

Do you think actually there was a degree of patient which results in the discrepancies in the biomarker positivity, number one? And also, some of the discrepancies in the response rates and other bits and pieces?

Peter O’Donnell:
I actually think it’s a good point you’re making, but I don’t look at it that way. I’m not sure we were better at selecting the patients back then. I know that a number of the patients that I put on this trial were patients that had really no other option.

I was very reluctant to think about even carboplatin-based therapy. And I was giving this to them as their last chance at any type of therapy. I think we’ve gotten better over time at selecting those patients. So, I would actually argue that we probably weren’t as good back then selecting these patients …

Mike:
Tom, you’re saying it was more just the inherent selection of an academic center and requiring a biopsy. Not that you were actually selecting them based on clinical features, is that …

Tom:
My mindset in the randomized Phase 3 is everyone goes in because that’s what we’re doing. And I was enrolling everyone into the 361. I think our site, we enrolled a number of patients.

Whereas I think in the frontline trials, when we started with those single arm Phase 2 trials, I wonder if those patients with clearly aggressive disease… Remember, our response rate second line for atezo was 13% and we didn’t yet have the results of 210 or we didn’t have the results of 52.

And so, we didn’t really know. And because chemotherapy has a higher response rate maybe saying, hmm, I’m a bit nervous about 30, except that pembro and unselected patients were 21%. And we knew by then the biomarker wasn’t really working that well.

You remember we enrolled all comers, not just the biomarker positives to these trials. And actually, in the pembro trial, it looks like the biomarker hasn’t helped at all. In the atezo trial, I think there’s debate going on around that.

Mike:
Right. Right.

Tom:
And so, I just wonder if we were saying this patient goes in and this patient gets chemotherapy, and they can go on a second line immune therapy study at the time. I don’t know. Because deep down, I feel there some are differences between the data sets in terms of the response rate and how patients perform [crosstalk 00:08:04].

Mike:
There’s clear differences. I guess the question is, Peter, are there objective differences in patient characteristics or other features that you can say, well, this is why the data sets look different, or no?

Peter O’Donnell:
Why would the prevalence of the biomarker go up? It doesn’t mesh with any of the other datasets that we have, which I’ll say that about a third of patients with metastatic advanced urothelial cancer are PD-L1 high, by any metric or any of the assays. The 50% is an outlier.

Tom:
So, there is an explanation for that. And that might be that because a lot of the patients at the beginning of the trial had a compulsory biopsy, and in 361, it was always historical tissue. That actually could account for that discrepancy. And it may be that real time biopsies have a lower positivity rate. Is that a possible explanation?

But there was a difference in terms of collection of tissue. And, of course, it’s important to say that not all patients were forced to have a biopsy. There was an early amendment in the trial, and I’m not suggesting for a second that this is a selected population of patients.

But there do seem to be discrepancies in 52 and 361. And one of the broader questions is, is there an easy explanation?

Peter, I have to agree with what you said. I don’t think the biopsy issue actually covers that net. The biomarker issue maybe a little bit, but not completely. Does it explain why the response rates were less impressive in the randomized Phase 3?

Well, we’ve seen those before. We’ve seen that before. One of the things I think that was really important and I don’t think we’re talking about …

Peter O’Donnell:
Was it that different, though, Tom? Do you really think that the immune monotherapy arm in 361 is that different from 52?

Tom:
Well, I …

Peter O’Donnell:
It wasn’t formally evaluated, but the duration of response, the percentage of patients responding …

Tom:
The duration of response, I think, is the same, but you are quoting response rates above 50% in some subsets, and we didn’t get …

Peter O’Donnell:
No, no, no, I didn’t. I said high forties for the PD-L1 …

Mike:
47.

Peter O’Donnell:
Population.

Tom:
Right? Yeah. So, that’s high. We didn’t get near that, obviously, at all.

Mike:
I feel like this is your amigos debate, not an ASCO podcast, but go ahead. I guess the question is, Peter, we’ve had this debate before of which patients would you select for pembro monotherapy.

I guess putting the debate aside focusing on the biomarker, are you using the biomarker to select at all? Or is it more that clinical endoglin nodal-only disease, et cetera, whatever other features you use? What is the role of the biomarker?

Peter O’Donnell:
We’re required to use the biomarker now by the label.

Mike:
Fair.

Peter O’Donnell:
We’re still required to use the biomarker. So yes, I send it, but I’m selecting the patient before I even get a biomarker result back, if that’s your point.

Mike:
Yeah, that’s my question.

Peter O’Donnell:
So, it’s got to be a patient that’s not a rapid progressor, a patient that doesn’t have liver metastases, a patient that’s not in terrible pain that I know that I can wait weeks to get the pembrolizumab working.

Mike:
What was the median time to response in this study? Sorry, Tom. What was the… Because we’re kind of talking about, oh, delayed response in disease …

Peter O’Donnell:
I believe it was at the first scan. So just around the two-month mark.

Mike:
Okay.

Tom:
Yeah. So, the response rate in 361 floats around the sort of 30% level, and in the biomarker positive population is the same. I think the thing that was surprising for me and the thing that I wanted to talk about was the carboplatin group in 361. So those patients treated with carboplatin, their response rate was higher than we expected. Their outcomes were, obviously, impressive, and we’ve looked at that subset data at ASCO GU this year. And so …

Peter O’Donnell:
And we actually know that a number of those patients were technically cisplatin eligible but got carboplatin by the investigator choice. Correct?

Tom:
So that …

Peter O’Donnell:
So, the investigator changed them to carboplatin.

Tom:
But I think that was more marked in 130. I think in 361, it was probably more balanced, I think, because it was about a 50/50 split. But the difference, I think from my perspective, I think is that in Maria De Santis’ EORTC study and the reason why this was an area of unmet need, is her overall survival in that big trial was only about nine months for carboplatin patients.

But in the other trials, in 130 and in 361, we’ve seen those carboplatin patients come in very similar to [inaudible 00:12:40]. and one of the originals …

Peter O’Donnell:
And a number of the De Santis patients had two poor prognostic factors, if I recall.

Tom:
So, it’s possible, Peter, that we’ve been somewhat, I think with hindsight, we were probably a bit naive to think that carboplatin was so ineffective and that patients could get six cycles of carboplatin and their median survival would be between six and nine months. And I think the field’s moved on. I think we’ve got better at giving chemotherapy, and I think those patients have done better. And I think that’s why …

Peter O’Donnell:
Do you think carboplatin still has the durability? Because I don’t see that. I don’t see the CRs. I don’t see the durable response.

Tom:
Well, the CR, this is difficult. So, the CR rate in the biomarker positive population with GemCarbo, which was presented at ASCO, was 18%. So, the CR rate for GemCarbo is higher than for pembrolizumab in 361.

And we haven’t talked much about that. And the reason why I don’t talk about it intentionally is I don’t think CR correlates with cure. The data from the avelumab trial, so median duration of CR, is only four months. So, at the time of completion of chemotherapy, if you had a CR with metastatic urothelial cancer, the duration of CR is only four months. But-

Peter O’Donnell:
It is different in the 52 here, right? CR clearly correlates with cure in the 52 data. Those patients almost never recur.

Tom:
So, I think in broad terms, these radiological tools that we use have got some significant flaws with them. And I agree the quality of the responses are better. And that indeed is why the Kaplan Meier Curves ultimately cross in 361. Immune therapy catches up. We lose some patients along the way, and the biomarker doesn’t currently select the winners.

Mike:
So, I’ve got one more question, Peter. Looking at the duration of response in this table of PD-L1 negative, it was 27%. PD-L1 positive, almost 60% who are duration of response greater than four years, 48 months. Do you think those patients are cured? They’ve now been two years off pembro if they’re at 48 months, at least?

Peter O’Donnell:
Correct. I do. I think certainly the PD-L1 high patients, they appear to be cured. They appear to be never recurring.

Tom:
I guess one of the questions would be, can we cure the same group of patients with maintenance avelumab? And the answer is, I think you probably can. Can we hand that net by capturing more patients that potentially results in long term durable remission? I think the answer is probably, yes.

And so, although, obviously, I’m involved in 361 and a number of other trials and I have a number of [inaudible 00:15:30], I’m still very nervous and these 52 durable data hasn’t swayed my opinion that I think treating unselected positive patients with 361 as the preferred option is not the right answer and that we still should be …

Although I do accept Peter’s, and we debated this previously, and I don’t think it has changed, I do accept Peter’s perspective that there are some patients in some institutions who potentially can pick the winners.

Mike:
Right.

Peter O’Donnell:
I feel like I want to respond. I’m holding myself back here.

Mike:
Well, I was going to ask you for your summary. Like these new data, durability is clearly key here, a fraction of cure. We’ve debated patient selection, which we’re not necessarily going to agree upon. But what’s the take home point from the abstract? Let’s just do that for the listeners.

Peter O’Donnell:
Yeah. I’ll say in the population right now that meets the label indication, the FDA label indication, there’s one out of three of those patients that are still alive five years later with immuno-monotherapy. Right?

And so, to me, that’s a big case for retaining the current option of immunotherapy as a treatment for this cisplatin ineligible population up front, because there are clearly patients, and we could debate who those are, but clearly patients, they will never need anything else but immunotherapy in their life.

Mike:
Yeah. Tom, do you have a final question? Or do you want to want to respond since we’ve turned this into a debate?

Tom:
No, I like the fact we’ve had to. It’s the first time we’ve managed to debate this, too. I hope the listeners are okay with that. And I realize that we have talked about this before. From my perspective, I think the ODAC meetings, and I think Peter have been really helpful. I don’t think my position’s changed that much. I still think chemotherapy and maintenance avelumab is currently the safer in unselected patients.

Peter O’Donnell:
Tom, you remember that the maintenance avelumab data are totally a biased data set. They took away all the worst patients. [crosstalk 00:17:41]

Tom:
So, it’s not biased in as much as the trial was randomized in a one to one. So, the trial wasn’t biased.

Peter O’Donnell:
Patients that didn’t have stable disease aren’t even included in the randomization.

Tom:
So, the stable disease and the response patients are included. But, Peter, I think this is a really important point, and I’m pleased we could disagree on this as well. So, listen, Peter, so I had a patient came in two weeks ago.

Mike:
Oh, boy, here we go. Make that [inaudible 00:18:09].

Tom:
No, no, no. It’s important. It’s important because I’m going to start presenting this patient a lot more. Because I tend to present winners in my series, and I don’t know why I do that. And I’ve had a bit of an inward look. I suspect I’m not going to get invited to many meetings, but I think we sometimes forget those patients who we leave behind.

And in bladder cancer, about 20% of patients have progression of disease. And so, this lady who I saw, I gave two cycles of chemo to. She was admitted to the ward. She got liver metastasis and she got, not a huge number, one or two. And she got some lung metastasis. She was admitted to the ward.

I went to see her. I assumed she had urothelial sepsis, and actually she got progression of disease in the liver, and she has primary progression urothelial cancer. We reviewed, we looked at the scans and there was nothing wrong with it. Just the chemotherapy wasn’t working.

And obviously, we said to her what are we going to do? And we tried to get her out of hospital. We got her out. She wanted some immune therapy. We gave her a cycle of immune therapy. It didn’t work. She progressed. Not with us anymore.

And I’m afraid my experiences, those patients that have primary progression of disease, immune therapy doesn’t rescue those patients. Patients who progress on chemotherapy with urothelial cancer are not rescued by immune therapy. They’re not rescued by the combination of immune therapy plus atezolizumab or pembrolizumab clearly in that trial.

So, when you say it’s a biased data set, number one, I don’t think it’s biased. I think it’s a randomized trial and it’s balanced. It is a selected population and the benefit of the trials, we haven’t included those patients, but those patients don’t win with anything. And certainly …

Peter O’Donnell:
I agree, I agree. But what you and others are doing is taking the avelumab data and comparing that against immuno-monotherapy. That comparison’s never been done. I’m saying there’s limitations to both data sets. And so, you’re making that comparison even though I think it’s not fair.

Tom:
So, actually, I agree. I don’t disagree with aspects of that. What I think you can do is, one, to pick an approach and there are two approaches. Either one gives upfront immune therapy or alternatively one gives chemotherapy followed by maintenance avelumab. And unfortunately, when patients sit in my office, as they are in yours, they say, “What are the options, Doctor?” and those are the two, and therefore, one does have to make some comparisons either direct or indirect of those data sets.

You can’t say we’ll start chemotherapy and hope for the best. And then we may or may not give you maintenance avelumab. You say, we’ll give you chemotherapy. If it doesn’t work, it’s going to be really tough. If it does work, we’ll give you it. And then-

Mike:
I think just to wrap up before we get into an hour-long debate, we would agree there are select patients clearly who can benefit long term from immune therapy that we don’t really know how to select and the biomarkers inconsistent.

There are probably some clinical features, but that different docs like you guys with experience might apply that selection criteria more or less strictly. Meaning Peter’s probably a little more likely to select patients and you’re much less likely, but it’s not wrong. It’s ends of a spectrum. Is that fair?

Tom:
I …

Peter O’Donnell:
We agree on that. I bet Tom would agree more if we were talking about the IMvigor130 data set where he believes it worked.

Tom:
So, the IMvigor130, and I wasn’t involved in that trial, I have to say. And, Peter, and we’ve got some data coming out in the not-too-distant future in different areas. This biomarker issue where that has a ratio and that capital marker and the biomarker positive looks a bit more encouraging. And I realize it’s exploratory as is the case with 361.

And that’s another shortcoming of the study is that the randomized data is exploratory and not positive. So, it makes it very speculative currently for both trials. And I guess when you look at that and say it’s exploratory and not positive, and there’s a different trial which looks stronger, a stronger data set, that’s where I begin to get into problems even with the atezolizumab data. Remember, the randomized Phase 3 durvalumab trial with durva/treme in the biomarker positives, came in with a hazard ratio of 0.74.

And remember, we picked the ITT population and the biomarker positive population for the thing. Had we picked the biomarker positive population for durva/treme, we would’ve had a positive randomized Phase 3 study with upfront immune combination therapy.

Durvalumab is added or tremelimumab adds about 10% response rates to durvalumab, 10% on top, which then pushes it above all or pushes it in parallel with chemotherapy. And that’s the most robust data set here. They’re all exploratory. None of them are statistically significant. That one at least seems to be chemotherapy where the other two seem to struggle at the moment.

Mike:
We got to wrap this up. I have to get to clinic. Peter, I’m going to give you the last word.

Peter O’Donnell:
No, I appreciate the chance to share these long-term data …

Mike:
Tom could talk forever. Yeah.

Tom:
I apologize. I apologize.

Peter O’Donnell:
My conclusion is that I think it validates the FDA label indication in the United States that this is an option for some patients.

Tom:
It’s a very fair summary. Agreed. Can we move on? Peter, it’s nice to chat.

Peter O’Donnell:
Mike, likewise to both of you.

Mike:
Thanks for joining. Enjoyed the debate.