"The MMAI model had significantly improved accuracy and prognostication to discriminate which patients would develop distant metastasis or die from prostate cancer," says Daniel E. Spratt, MD.
In this video, Daniel E. Spratt, MD, discusses findings that showed successful validation of an MMAI prognostic biomarker, called ArteraAI Prostate, in identifying patients that would develop distant metastasis or mortality from prostate cancer. The findings were presented at the 2023 ASCO Genitourinary Cancers Symposium in San Francisco, California. Spratt is a professor and chair of radiation oncology at University Hospitals Siedman Cancer Center in Case Western Reserve University in Cleveland, Ohio.
In localized prostate cancer, as well as recurrent prostate cancer, the way we guide treatment decisions is using NCCN guidelines. And what NCCN guidelines use are prognostic biomarkers. When you hear that word, people start thinking of fancy AI tests, maybe genomic tests, but really even things like PSA, Gleason Score, NCCN risk group stage, those are prognostic tests. The problem is, all of our treatment decisions rely on NCCN risk groups. This however, is a fairly moderate to modestly performing prognostic system. It doesn't accurately tell men do you actually have aggressive or indolent disease.
What that results in is we overtreat and undertreat a significant majority of our patients. There's been a lot of work trying to identify more accurate prognostic biomarkers. An effort we started a little over a year ago and presented last year here at GU ASCO was the training and validation of a multimodal artificial intelligence biomarker, where we took 5 phase 3 randomized trials and split the cohorts for training and validation. We showed that this multimodal artificial intelligence biomarker had improved risk stratification compared to NCCN risk groups.
Now, what is this model? It takes your standard clinical variables – Gleason score, T-stage, PSA – as well as the digital imagery of the histopathology slides. All of these patients’ slides were imaged and then fed through a deep learning algorithm. Then it was combined in a novel way with the clinical information to result in a model score. That model was trained and validated, but that work was done on men with low-risk, intermediate-, and high-risk prostate cancer per NCCN guidelines.
We wanted to understand well, within a given risk group, and I see a man in clinic, he may have high-risk disease, do these models and these biomarkers still help us risk stratify these patients? Because right now, we treat them with a one size fits all approach. With radiation, it might be radiation with long-term hormones. With surgery, it might be surgery with early salvage radiation and hormone therapy. We know there's a lot of heterogeneity within even NCCN high-risk.
What we did in this study is we took the validation cohort from that last work I spoke of, which was 4 of the randomized trials that enrolled men with high-risk disease, as well as 2 newly profiled trials, RTOG 0521 and RTOG 9902. These are randomized phase 3 trials. These were all combined, so 6 phase 3 randomized trials, with 1088 men with long term follow-up. The median follow up is over 10 years, most of the men had either PSAs over 20, clinical T3 or T4 disease, Gleason 8 to 10. We compared, within this NCCN high risk group, the performance of this MMAI model versus any of our standard clinical variables again like Gleason, PSA, etc.
Unfortunately, most of our clinical variables are barely better than a coin flip to risk stratify once you're already in an NCCN high-risk group, and that's why we treat them with a one size fits all approach. However, the MMAI model had significantly improved accuracy and prognostication to discriminate which patients would develop distant metastasis or die from prostate cancer.
What I would say was really impressive was that in high risk disease, if you look at men had the lowest quartile score of that MMAI model versus the highest quartile, there was 20% to 30% difference in absolute risk of metastasis or death from prostate cancer. Again, these are patients that we all would lump together as the same type of patient, when they have vastly different prognosis where some of these men would have a low chance of dying from prostate cancer, whereas some men have very high rates of dying from prostate cancer – almost 6-fold higher.
This transcription was edited for clarity.