Yair Lotan: Oh, thanks so much.
Tian Zhang: So we're going to talk a little bit about the CHAI biomarker that you're presenting here. So tell us a little bit about the biomarker and the study itself.
Yair Lotan: We all recognize that we use pathology all the time to get stage grade of tumor, but there are a lot of features that are not necessarily readily utilized. And some of them will recognize lymphovascular invasion, for example. But beyond that, there are features that probably have prognostic significance, but are not something that our pathologists readily identify. The CHAI biomarker is essentially an artificial intelligence tool that is taught a lot of factors ahead of time. So for example, it's taught what tumor cells look like, what fibroblasts, what immune cells look like, can look at vascularity, et cetera. Once it was taught this information, then it is given a set of patients for training to identify patients that have worse outcomes. In other words, a worse prognosis, and then, subsequently is validated. And this has been done for patients with non-muscle-invasive bladder cancer.
So there's an assay commercially available right now for patients to assess BCG response. It's also been studied in BCG-unresponsive disease and BCG-exposed. In this particular aspect, we applied it towards muscle-invasive bladder cancer. So the way the assay was developed with using the TCGA patients who had TURBT with clinical T2 disease, and we identified patients who subsequently had cystectomy in the TCGA and who had a higher risk for recurrence, and that's how it was trained. And then it was validated in two cohorts from UT Southwestern, University of Kentucky, to help to see if it would identify patients who had a higher risk of recurrence and cancer-specific survival and overall survival.
Tian Zhang: Yeah. Did you use their TURBT specimens or their cystectomy specimens for this particular actual study?
Yair Lotan: Yeah. So we use the TURBT specimens. And some of that is because if patients get neoadjuvant chemotherapy, they may have no residual disease in best-case scenario, or it may alter the features in the tumor. And so, when you're trying to evaluate maybe a prognosis upfront, if you're trying to decide should I give systemic therapy or not, you have the TURBT specimen as your baseline information.
Tian Zhang: Yeah. And so, is it helping you find a poor prognostic portion? We're hearing a lot about circulating tumor DNA at this meeting. And so, what do you think that sweet spot is of using a histology marker like this versus a circulating tumor DNA marker?
Yair Lotan: Right. So what we found in our cohort was that the biomarker gave independent prognostic information, identifying a group that was significantly higher risk of recurrence and of dying of bladder cancer. We looked at multiple factors, including receiving neoadjuvant chemotherapy, having CIS, age, sex of the patient, and this was independent of that. So as a prognostic marker, I think it was quite valuable. One of the nice aspects of it is that it doesn't use up tissue, so it's an H&E slide, and so it's a fairly rapid test. How it'll be incorporated clinically, something we'll have to see. Because if you think about what a biomarker can do, it can either identify patients who should intensify treatment or de-intensify treatment. Well, some of the patients here receive neoadjuvant chemotherapy, we both know that the paradigm is shifting and patients are going to start getting EV pembrolizumab likely as the neoadjuvant treatment of choice, and possibly adjuvant treatment. That still doesn't change the fact that we would like to know if a patient is more likely to need those treatments. And so, this might be a tool that could be useful. It'll still need to be validated though.
Tian Zhang: Yeah. So with a changing neoadjuvant paradigm, do you think that it will need to revalidate similar characteristics in a new cohort?
Yair Lotan: I think so. I think that there will always be patients who don't want systemic therapy or can't get systemic therapy, or where you're on the fence. Should I give treatment a systemic therapy or not based on comorbidities, age, patient preference? And so, a biomarker that will lean you one way or another might help you make that decision. And this biomarker obviously was evaluated in patients who didn't get neoadjuvant therapies, and so it may not need to be validated for that. On the other hand, when you are trying to look at a patient who's going to get EV pembrolizumab anyway, then you'll have to have an evaluation of a new cohort. And we do have multiple trials now that we could theoretically evaluate the biomarker in that dataset.
Tian Zhang: And the good news is this is a slide-based assay. Everybody has a TURBT H&E slide ready to go. So it seems easy to collect it on folks.
Yair Lotan: There are aspects of it that are very easy. And then there are some obstacles from the way the FDA and Medicare work. So right now, it's still somewhat archaic a process. You still have to send a slide and for it to get uploaded into a server, it hasn't been approved yet that a pathologist could just upload the slide into the cloud and then get a response within five seconds of the prognosis, which theoretically, if you just license the software to a hospital, they could immediately get the results back. Right now, it's still cumbersome, but I think there will be many such tools, AI tools where a pathologist at an institution could just immediately get information back, and that would just part of the standard report to a clinician. Yes, this is a T2 with unfavorable characteristics according to this assay, or this is a favorable T2, for example.
Tian Zhang: Sure. That sounds like a great future state to shoot for. And with the keynote yesterday, really thinking through what AI is going to bring to medicine. So that would give real-time feedback on prognosis.
Yair Lotan: Yeah, I think so. And because it's an easy test to do, I think that people will find a way to incorporate it. But I think, as you mentioned, validation is important because dataset that's been developed in part with cisplatin-based chemotherapy may not be fully relevant to a patient who's going to get EV pembrolizumab.
Tian Zhang: Any takeaways that you want the UroToday audience to remember?
Yair Lotan: I think that there's a growing number of these AI tools. The one we studied from VALOR is just one of them. There's other companies developing these tools. So like any biomarker, I think it's important to assess, first of all, what the clinical utility will be. And then also, make sure that the assays are validated in good-size cohorts because we don't necessarily want to obtain information that's not, A, very accurate, and then try to apply it clinically. So I think each marker that comes out needs to be really assessed both in terms of its accuracy and predictive ability.
Tian Zhang: Awesome. I think that's a great summary and hopefully we'll get to a future where we have real-time markers to help us in muscle-invasive disease. Thanks Yair.
Yair Lotan: Thank you.