Study applies AI tool to STAMPEDE data in prostate cancer

In this video, Charles Parker, MD, discusses the background for the study “External validation of a digital pathology-based multimodal artificial intelligence (MMAI)-derived model in high-risk localized (M0)/metastatic (M1) prostate cancer (PCa) starting androgen deprivation therapy (ADT) in the docetaxel (Doc) or abiraterone (AAP) phase III STAMPEDE trials,” which was presented at the 2023 European Society for Medical Oncology Annual Congress in Madrid, Spain. Parker is a clinical research fellow and PhD student at UCL Medical College in London, the United Kingdom.

Transcription:

Please describe the background for this study.

The background to this is really the STAMPEDE clinical trial, which is a multi-arm, multi-stage trial that's been recruiting since 2005, testing a number of treatments in hormone-sensitive prostate cancer in both the locally advanced and the metastatic setting. So within prostate cancer in general, the big problem that we've got is accurate prognostication, so being able to tell patients how long they're going to live on treatment and using that to help guide decisions on treatments for who can be de-escalated from treatment, and who needs treatment escalation. So as part of this, the work I'm connected to through my PhD with Professor Attard is part of the STRATOSPHere Biomarker Development Consortium, trying to find these prognostic markers from the data associated with the STAMPEDE trial. That's really where the background of this project came from.

What was the methodology for the research?

With this specific study, what we wanted to do is find out whether there was more information that we could leverage from preexisting material within the STRATOSPHere Biomarker Development Study; namely, the diagnostic H&E sample. We already know that there are various variables that get reported as part of routine histology reporting, which have a degree of prognostic impact, particularly in the early stage of the disease. But these aren't as powerful in the latest stages of disease, which is affecting the STAMPEDE patients. And so, as part of the study, we partnered with ArteraAI, which, in the localized disease setting has developed a multimodal artificial intelligence model. It's multimodal because it combines both clinical data and digital images of histology slides, and artificial intelligence because we use a deep learning pathway to extract features from both the clinical data and the whole slide imaging data and combine these into a prognostic score that is trained against prostate cancer-specific mortality, which, coincidentally, is the main outcome reported in the STAMPEDE trials. As part of this, I reviewed over 7000 slides from over 3000 patients to check for completeness of the tissue, tumor content, and a quality check to look for any slide artifacts. At the end of the day, we ended up with 3167 patients who had both complete clinical data and acceptable-quality imaging slides. These patients had the data then transferred to ArteraAI, which ran the algorithm and generated the prognostic score based on prostate cancer-specific mortality.

This transcription was edited for clarity.