Multimodal AI Informing Treatment Decisions in Advanced Prostate Cancer - Gerhardt Attard

November 20, 2023

Alicia Morgans interviews Gerhardt Attard about groundbreaking research presented at ESMO 2023, utilizing the STAMPEDE data set. They discuss a new prognostic tool developed with Artera, the ArteraAI Prostate Score, which predicts prostate cancer-specific mortality using digitized images of H&E slides from over 3,300 patients. This tool has proven that primary tumor analysis can provide significant prognostic information, even in metastatic cases. It offers more nuanced risk stratification than current methods, which could lead to better-tailored treatments. Dr. Attard envisions integrating this with other tests like PSMA PET for even more precise predictions. The conversation also touches on the potential of predictive markers for treatments like docetaxel, aiming to personalize therapy further and balance treatment intensity with patient quality of life.


Gerhardt Attard, MD, PhD, FRCP, John Black Charitable Foundation Endowed Chair in Urological Cancer Research, University College London, London, UK

Alicia Morgans, MD, MPH, Genitourinary Medical Oncologist, Medical Director of Survivorship Program at Dana-Farber Cancer Institute, Boston, Massachusetts

Read the Full Video Transcript

Alicia Morgans: Hi. I'm so excited to be here with Professor Gerhardt Attard of University College London to speak about a recent presentation at ESMO 2023. Thank you so much for being here with me today, Professor Attard.

Gerhardt Attard: Thanks for inviting me.

Alicia Morgans: Wonderful. I am always so excited to hear the work that you and your team are doing in the STAMPEDE data set to try to understand the basic biology and really help us make decisions related to prognosis and prediction even eventually, I hope, in terms of what therapy is going to be helpful for a patient. Now, you presented some really exciting data at ESMO 2023. Can you share a little bit about this please?

Gerhardt Attard: Yeah, so this was data presented by my research fellow, Charles Parker, who's a pathologist who's been in my lab for a couple of years, presented in the mini oral session at ESMO. We focused on patients who were treated in four phase III trials in STAMPEDE. I'm sure your listeners will know that STAMPEDE is a platform protocol where we have conducted a number of phase III trials, many of which were positive. And here, we focused on two trials that tested docetaxel plus ADT versus ADT alone, and two trials that tested abiraterone combinations, plus ADT versus ADT alone.

And the trial recruits both patients that have very high-risk non-metastatic disease, so locally advanced or biochemically recurrent, and metastatic patients. In total, we had slides and clinical information from just over 3,300 patients, so a large number of patients, with prospective follow-up collected as part of the trials. And all these trials have closed in the past two years. And the median follow-up was 6.7 years, so nearly seven years of follow-up.

We designed our trial to ask the question about whether we could predict death from prostate cancer, so prostate cancer-specific mortality. And we worked with a company called Artera. And using digitized images of the H&E slides taken at diagnosis, which we retrieved from just over a hundred centers in the UK, we with Artera generated a risk score, which is the ArteraAI Prostate score, and then tested whether that predicted prostate cancer-specific mortality. And it does. And I think that's really exciting for a number of reasons. The first is, it's proof of concept that the primary tumor contains prognostic information even in patients where most of the disease has spread elsewhere.

When we started this, we debated at length. Is this going to work? Is a [inaudible 00:02:56] biopsy from a prostate really going to have that information? And here, we prove that. We prove that that is clinically informative and strongly prognostic. We then asked the question, how can this be used in general practice? And what we are using, I think most physicians in the world are using, to decide the risk of a patient dying from prostate cancer is metastatic volumes. Metastatic volume in a metastatic patient and nodal volume in a high-risk localized patient.

So, we then took the patients by that disease burden and we split them by ArteraAI quartiles, and we really found that the quartiles divided into the fourth quartile, which had the worst prognosis, and then the first three, which really bunched up together. And when we look at the patients divided by disease burden and then ArteraAI score, we start to see some better granularity than we're currently working with.

For example, the node positive patients that are in the fourth ArteraAI quartile have the same risk of prostate cancer death as the low volume patients that fit in Q1 to Q3. And we debated several times at ESMO this year how really, the disease is a continuum. Patients don't fit into these divided categories. They sit somewhere along that continuum. And with these types of tests, we're now starting to have better information on where that patient sits along that risk stratification.

And then similarly in the metastatic patients, we divide our patients into groups with very significant prostate cancer mortality. So, low volume splits into these two groups. One group, same risk as the node positive high ArteraAI score, and then the bad low volume group have the same outcomes as our high volume metastatic patients.

Alicia Morgans: Phenomenal, of course, but let's just back up to a really practical aspect of this that you said quickly. And I want to make sure everyone hears because I think it's so important as we try to deploy these types of strategies ultimately for the world, for patients with prostate cancer. You were able to go back within this study that is not a pharmaceutical-run study, it's a study that is being administered across many centers in the UK, and you were able to pull slides that were not created specifically for the purpose of running this particular Artera MMAI assessment on them. And these were slides made just for clinical practice and you were able to calculate this MMAI within those slides, again, made for standard of care purposes. Is that right?

Gerhardt Attard: Yeah. So yeah, over a hundred centers in the UK, and we have both slides that were generated for clinical practice. We also retrieved the tumor samples, and we have previously presented on a lot of the molecular analysis we've done on those. And as part of that process, we section new H&E slides. So, those are representative of what's going to be used in everyday clinical practice. A slide that's been cut in the past month or so, that got scanned. And that was the input to this test.

And you're right. That's what's exciting about it. This is information that was just lying around there. It's free. Yeah, we're not paying extra for that to get those images. They're available.

Alicia Morgans: That is fantastic and really pretty incredible, I think, again, if we're trying to deploy this sort of strategy in our practices. Now, one thing that I think is so interesting, we are trying to approach prognostic testing from multiple different angles. We're using things like MMAI. We're also using things like PSMA PETs and other imaging strategies. But PETs are certainly not deployed everywhere, and we did not have PET data from our STAMPEDE population. So, it might be interesting over time to even think about how PSMA PET data could integrate with these models to, again, even get down to even more granular and granular predictions ultimately, I hope, as well as prognostic data for these types of patients.

Gerhardt Attard: Yes, I agree. PET will give another element of information. We didn't have PET PSMA when we started these studies a decade ago, but we're now thinking of ways to implement tests like ArteraAI into practice, and then we'll build those cohorts up and really understand what's the metastatic burden based by ArteraAI score. I don't know the answer, but I'm confident we will get different levels of information from every different test. And then as you say, we'll pull it together to get the most accurate prediction for our patients.

Alicia Morgans: Wonderful. As I think about this, let's think really about the abiraterone cohort. And these patients with locally advanced disease are treated in many different ways in many different areas, and it's really important that we understand from a prognostic perspective which patients are going to have the highest-risk disease. How do you think about using this prognostic information when you see a patient with locally advanced disease in the clinic now?

Gerhardt Attard: I think all patients should get ADT plus an ARSI, so hormone intensification. I think the question is, who should get additional treatment? We already have a number of, or at least one option, docetaxel, to consider, but not all patients, in my mind, should receive docetaxel. We have shown in the final follow-up data from STAMPEDE that nearly half of the low volume metastatic patients have not progressed at eight years. Clearly, those patients are unlikely to benefit from additional treatment.

And of course, similarly in the localized high-risk patients, there's a group who die from prostate cancer, eight to 10%, and we need to identify those. But we cannot treat all patients with additional intensified treatments because we'll start to cause more harm. So, I think we need prognostic tests to work out who should get additional treatment.

The second question then is, what treatment should that be? And for that, we need predictive tests. And the work going forward with ArteraAI over the next year will be to start to answer that question. And probably initially with docetaxel where we already have the data sets and working with Chris Sweeney and the charted group to pool our data sets together and hopefully have an answer in the next year or so about how can we predict who is going to benefit from docetaxel and who is not.

Alicia Morgans: Well certainly, if you can do that, the field will be very excited to hear it. So, it sounds like predictive markers may be on the way.

Gerhardt Attard: I hope so.

Alicia Morgans: I hope so too. If you had to sum up this really impressive body of work, presented at ESMO 2023, what would it be?

Gerhardt Attard: Using clinical information and digitized images assembled into the ArteraAI score, we can extract new and very informative prognostic data that can be used to predict the chance of our patients dying from prostate cancer. And then doing so, then tailor who should receive additional treatments, who we should consider for deescalation.

Alicia Morgans: Fantastic. It's one step closer to really personalizing our care. So, thank you so much for your time and your expertise.

Gerhardt Attard: Thank you.