AI tool predicts survival in patients undergoing RP in the PLCO trial

Author(s):

"Specifically when we stratified by tertiles, we saw that at 15 years there was a significant difference in cancer-specific survival for those who had a high MMAI score vs lower MMAI scores," says Eric Li, MD.

In this video, Eric Li, MD, discusses the background and notable findings of the study “A.I. Derived Digital Pathology Biomarker Predicts Prostate Cancer Specific Survival for Patients Undergoing Radical Prostatectomy from PLCO Trial with Over 15 Years of Follow-Up,” presented at the 2023 Society of Urologic Oncology Annual Meeting. Li is a urology resident at Northwestern University Feinberg School of Medicine in Chicago, Illinois.

Transcription:

Please discuss the background for this study.

We collaborated with ArteraAI, who previously developed this multimodal AI algorithm in patients mainly undergoing radiation. And so they collaborated with NRG Oncology and developed and validated this model previously in patients undergoing radiation. However, it hasn't really been looked at in patients undergoing surgery. So that was really the inspiration for our project here; to basically identify people from the PLCO cohort who underwent surgery and had digitized radical prostatectomy slides to see where this algorithm could be applied to them as well.

What was the study design like?

PLCO was originally a PSA screening trial started in the US in the 1990s. We identified a cohort of just over 1000 patients who had available digitized radical prostatectomy slides, as well as had relevant follow-up data. Our median follow-up was 17 years. Basically, we did a Cox model to look at factors associated with both prostate cancer-specific mortality as well as overall survival. And we also did a Kaplan Meier survival curve analysis.

What were some of the notable findings? Were any of them surprising to you and your coauthors?

First of all, we always think about patient selection. And so because it was a PSA screening trial, many of our patients had low-risk disease; actually, 90% of patients had NCCN low- or intermediate-risk prostate cancer at diagnosis. And we found that the MMAI models were predictive of both prostate cancer-specific mortality, as well as overall survival. And specifically when we stratified by tertiles, we saw that at 15 years there was a significant difference in cancer-specific survival for those who had a high MMAI score vs lower MMAI scores.

This transcription was edited for clarity.