Validation of a genomic classifier to predict adverse pathology in men diagnosed with low risk prostate cancer.

Abstract

71 Background: The Decipher 22-feature genomic classifier (GC) has been validated to predict metastasis and prostate cancer specific mortality in needle biopsy (Bx) tissue of men with intermediate and high-risk prostate cancer. We validate GC in diagnostic Bx specimens for the prediction of high-grade/stage (HGS) disease at radical prostatectomy (RP) in men with low and favorable-intermediate (fav-int) NCCN risk group disease. Methods: We identified 176 men diagnosed with low or fav-int NCCN risk group disease who had available Bx GC scores and pathological information after RP from Cedars-Sinai, University of Calgary, Cleveland Clinic, MD Anderson, and Johns Hopkins. The GC score was calculated based on a locked random forest model. Scores range from 0-1 with cut points for GC low, intermediate and high risk groups as <0.45, 0.45-0.6, and >0.6, respectively. The primary endpoint was HGS (Gleason group 3-5 or pT3b or lymph node invasion (LNI)). Univariable (UVA) and multivariable (MVA) logistic regression models were used to evaluate GC and CAPRA. Results: Median age of the cohort was 62 years, 87% and 13% had Bx grade group (GG) 1 or 2 disease. 76% and 24% were NCCN low and fav-int risk, respectively. CAPRA classified 70% as low (0-2) and 30% as average risk (3-5). GC classified 80% low, 16% intermediate and 4% high genomic risk. After RP, 41% had RP GG 1, 46% GG 2 and 13% had GG 3-5 disease. pT3b or positive lymph nodes were observed in 7 men (4%), overall 27 (15%) of men had HGS at RP. In the UVA and MVA, GC was the only significant predictor of HGS with odds ratio (OR) of 1.38 and 1.34 per 10% unit increase, before and after adjusting for CAPRA (p=0.011, 0.027). A low risk score (GC<0.45) had a negative predictive value (NPV) of 92% to identify men who do not have HGS at RP. In exploratory analysis, a very low risk cut-point (GC<0.2) was found which had a sensitivity of 96% and an NPV of 99%. 26% of men had GC<0.2. Conclusions: We validated GC in a multi-institutional study to predict HGS at RP among men with NCCN low and fav-int risk disease with high sensitivity and NPV. Future studies will aim to validate the very low risk genomic cut-point to guide decision-making and follow-up biopsy protocols for men considering or in active surveillance.