Roger Li: Thanks for having me.
Ashish Kamat: You and I have talked about many things over the years and today we're going to talk about something you're discussing here at the AUA, which is the computational AI-based assessment of patients with intermediate-risk bladder cancer. So before you tell us about that, I want to congratulate you on a couple of things.
Number one is obviously the intermediate-risk bladder cancer grouping system that you led that's now making a new NCCN and of course, all the other work that you're doing in the field. So congratulations on that.
Roger Li: That we're doing. So congratulations to you as well.
Ashish Kamat: So tell us a little bit about where you think this AI-based marker for histology is going to fit into the whole classification for patients with low-grade intermediate-risk bladder cancer.
Roger Li: Yeah. So it's a great question and it's a question that we face in the clinical practice every day. Ashish, as you know, I don't have to explain to you, but for the audience, intermediate-risk non-muscle-invasive bladder cancer is a really heterogeneous disease that some patients will not need to have any treatment after their initial TURBT and the postoperative intravesical chemotherapy.
Whereas others, you're really faced with the risk of progressing to high-grade, sometimes even muscle-invasive disease. And so starting in 2014, you yourself and IBCG started this sustained effort to better stratify patients based off of clinical risk factors. I was privileged to partake in that effort by identifying the risk factors, the five that we've identified as the IBCG risk factors, and then categorize these patients into three different subcategories of IR and MIBC.
And we're able to show, with others too, that these risk factors do stratify patients in terms of their progression-free survival and also for patients that are on active surveillance for their bladder cancer, if you were not to have any of these risk factors that you're more likely to not need to have any surgical intervention.
Ashish Kamat: And just to emphasize for our audience again, some of this progression actually is to invasive disease, right?
Roger Li: That's right.
Ashish Kamat: It's not just some people think that if you have a low-grade tumor in the bladder, you're never going to have high-grade, let alone invasive disease. But if you have a lot of risk factors, you can actually progress to invasive disease, which is the important thing and that's where the work that you're doing is going to come in.
Roger Li: That's exactly right. Yeah. So we kind of hypothesized that using the CHAI platform, which by the way, we already did a study a couple of years ago looking at BCG-treated patients and just looking at the histomorphology of the tumor itself, so both tumor cells as well as the immune cells, we're able to classify those patients who would respond to BCG.
And then based on a similar idea, we also asked the question whether or not we can predict those patients with low-grade IR and MIBC that would eventually progress on to have grade progression.
Ashish Kamat: And tell us a little bit for the audience again, the development of the CHAI platform, because it was developed for different reasons for us, BCG unresponsiveness, then gemcitabine/docetaxel. How was the algorithm for this prediction developed?
Roger Li: Yeah. So initially when the Valar Lab Group approached me and wanted to study the BCG story, I thought that they were out of their minds. In my mind, I think you have to do this in a stepwise fashion where the first predictor has to be a prognosticator, looking at patients who weren't treated with any intravesical therapy at all and seeing whether or not they had recurrences or progressions.
And the low-grade patients were the natural group for these because as you know, some of these patients will have a recurrence, others will not. So if we can come up with an AI predictor to tell us which of these patients will have a recurrence, maybe we can then cut down on the number of cystoscopies for those that will not.
But as the project evolved, they were able to show a really compelling story in the BCG setting. And as an extension to that, this CHAI predictor that we used for the low-grade setting was completely different than the one that they had used for the BCG story, specifically tailored for the low-grade patients, IR and MIBC patients by the IBCG definition to predict for high-grade recurrences.
Ashish Kamat: Now we have clinical parameters. The IBCG thing is, of course, in the NCCN guidelines. We have people developing nomograms, simplifying things. And now, of course, we have machine-learning AI-based tools. For the busy urologist that's out there listening to you right now, how would you recommend they factor in something like the CHAI platform in their counseling of patients?
First of all, would you rely on it? Would you select it for a subgroup of patients? How would you recommend they actually look at this?
Roger Li: Yeah, great question. And ultimately, we would love to be able to bring this to the clinic. But I think as it currently stands, we'll still need to probably do some validation work. So the way that the biomarker was developed, it was done, as you know, with a multi-institutional cohort of just under 800 patients splitting it up into a discovery cohort of 270 patients or so followed by a validation of 500 patients.
And again, specifically looking at whether we can use the discovery cohort to predict for high-grade recurrences, but
ultimately, I think this is still exploratory. We need to do this in a prospective fashion to understand whether this can be brought into the clinic.
But with that being said, I think the early access program that Valar Labs has put out for BCG treatment has been very well received in the setting that because of the BCG shortage, there is a clinical unmet need there for us to have that predictor to allocate BCG versus gem/doce for those patients. In the low-grade IR and MIBC setting, there's not as much of a need for that.
So I think there is room for us to do that. But getting back to your question about how this is going to eventually become incorporated into our clinical practice, I think as great as the IBCG risk factor is, it still takes a little bit of work to gather all of the information, especially in a busy clinic. A lot of urologists may not necessarily know whether the patient had frequent recurrences within a year or when their last recurrence was.
And so the nice thing about the AI platform is it is very complex, but at the end of the day, all of the complexity is reduced down to machine learning, and that can be brought to the clinician in one simple predictor, a go/no-go. So I think it would be really nice for us to incorporate that and figure out all the workflow and everything, but ultimately be able to easily incorporate into our clinical workflow.
Ashish Kamat: Yeah. It'd be great if we could have a simple score, right? Whether it's a CHAI-based tissue marker or, for example, the work out of Toronto that's happening with the progression AI, which is again, clinical AI. I think AI is certainly here to stay in our daily life and we have to essentially utilize it to our best benefit when it comes to helping patients as well. So congratulations, Roger, once again, and thanks for taking the time.
Roger Li: Thank you.