"It could identify about 2/3 of men that normally we'd be recommending hormone therapy would appear to have no benefit from hormone therapy," says Daniel E. Spratt, MD.
In this video, Daniel E. Spratt, MD, highlights the background and findings from the study, “Artificial intelligence predictive model for hormone therapy use in prostate cancer,” for which he served as the lead author. Spratt is the chair of radiation oncology at University Hospitals Seidman Cancer Center and a professor at Case Western Reserve University School of Medicine in Cleveland, Ohio.
Could you describe the background of this study?
The ArteraAI biomarker that we developed and validated is really transformational. What we did was, using a novel pipeline and novel methods, we took 5 phase 3 randomized trials. These are big clinical trials done across North America with long-term follow-up of patients treated with radiation therapy with or without hormone therapy for localized prostate cancer, the most common cancer in men. One of the big things that men face when they receive radiation therapy is do they need to have androgen deprivation therapy, or some people call it hormone therapy. The current national guidelines say some men, definitely we should be giving it, some men maybe don't need it. But it's very crude. We know that we're overtreating, and we know we're undertreating some men. Trust me, no patient wants to receive hormone therapy if it's not going to help cure them, because it has lots of different [adverse] effects that can bother men. But obviously, if it can help cure them or prevent them from developing metastatic disease, that's of great value to patients.
So, using these trials for the trials, we used their clinical information such as their age, their PSA, the grade of their cancer, but we combined it with taking the pathology slides from their prostate biopsies and they digitize those slides. So, you have a digital image, and they ran artificial intelligence, machine learning to extract features from that pathology slide. Some of them probably are human interpretable that maybe a pathologist can see, and some of them are non-human interpretable. By combining them together with the clinical features, we trained a model that can predict which patients benefited or did not benefit from hormone therapy.
What were some of the notable findings?
A very important part of this groundbreaking study was not only did we train this model, we independently validated it. Meaning that we took the largest phase 3 trial that has ever been done with the longest follow-up. It's called RTOG 9408, and it's a randomized trial of men who got radiation with or without 4 months of hormone therapy. Very importantly, this study validated the biomarker once it was developed and locked in this trial.
It could identify about 2/3 of men that normally we'd be recommending hormone therapy would appear to have no benefit from hormone therapy. That's pretty meaningful. About 1/3 of men would be recommended to have hormone therapy. So that's one big take home point is this is something that ordering the test or using this biomarker, it's not something that would necessarily help 1 in 100 men, it's about 2/3 of men that order this likely could omit or not need to have the hormone therapy.
I think another really important finding from this was that this was a very diverse cohort. It was from hundreds and hundreds of centers across North America, and about 20% of patients that enrolled in these trials were African American. Which, that's been a chronic challenge in clinical trials, to have good, adequate diversity and representation. So, this gives a lot more confidence when using this test that it's such a large sample size from diverse centers, diverse backgrounds, [and] diverse people to the power of this finding.
This transcription has been edited for clarity.