Krishnan R. Patel, MD, on High-Risk Prostate Cancer: Clinico-Transcriptomic Risk Stratification

This study emerged from a clinical question and a clinical need that we saw in our clinics, which was how can we integrate standard NCCN risk stratification with transcriptomic risk stratification with the 22-gene Decipher score into our clinical practice to guide patient selection for abiraterone treatment intensification. What we noticed was that some patients were coming in with a high clinical risk and low genomic biomarker risk or high genomic classifier risk and low clinical-transcriptomic risk, and we weren't really sure how we should be treating those patients. So what we did was we pooled 4 NRG/RTOG studies, which were RTOG 9202, 9413, 9902, and 0521, and we basically broke the study down into 3 parts. The first part, we wanted to assure ourselves that the 22-gene genomic classifier score could improve prognostic estimates over clinical variables alone. So to do that, we constructed multivariable Cox regression models. And in summary, what we found was that we verified what was already known in the literature, that the 22-gene genomic classifier score is additive and can improve upon prognostic estimates above clinical factors alone. The second thing we wanted to do is we wanted to understand if risk discordance was sufficiently prevalent in the population to merit study. For example, if it only happened about 1% of the time where people had discordant clinical and biomarker risk, then it probably wouldn't merit much study because it's pretty rare. And what we found was that it happened about 25% of the time, and we felt that that was a reasonable number of people to merit further study. The last part of our analysis was really to develop an algorithm that would be very simple and that clinicians could use on a daily basis until prospective trials like NRG-GU009 have resulted to help guide treatment intensification for men with high-risk prostate cancer. So we were able to construct such an algorithm. It's a little bit complicated, but suffice it to say that you give people points for both their NCCN risk stratification and their genomic classifier risk stratification based on commercial cut points. Patients get 0 points for less-than-high-risk features, 1 point for high-risk features, and 2 points for very-high-risk features. You then sum the points from the NCCN risk stratification and the genomic classifier risk stratification. And if the sum is less than or equal to 2, they fall into the clinical-transcriptomic high-risk subgroup. And if the sum is greater than or equal to 3, they fall within the clinical-transcriptomic very-high-risk subgroup. And in summary, we found that the very-high-risk subgroup had a very similar prognosis to the STAMPEDE control arm, who received radiation and ADT, I mean radiation and long-term ADT alone. We took that as evidence that they would have a very similar model-estimated benefit of abiraterone as in STAMPEDE and thus may merit treatment with abiraterone, but further prospective studies would be required to validate this. The takeaway is that in the absence of prospective data, this may be a reasonable way to integrate clinical and biomarker risk in clinical practice in 2026.