Analysis of genomic and clinical predictors of overall survival and PSA response: Results from the PRECISION registry.

Abstract

5051 Background: Outcomes in advanced prostate cancer vary substantially according to clinical and genomic characteristics among patients with metastatic castration-resistant prostate cancer (mCRPC). We sought to identify predictors of overall survival (OS) and prostate-specific antigen (PSA) response using data from the PRECISION real-world registry of mCRPC patients treated with poly(ADP-ribose) polymerase inhibitors (PARPi). Methods: Patient-level data from five centers (Dana Farber Cancer Institute, Thomas Jefferson University, University of British Colombia, University of Washington, Yale University) were aggregated, including demographics, clinical characteristics, genomic alterations, treatments, and clinical outcomes. OS was defined as the time from the first date of PARPi treatment to death or the last date the patient was known to be alive. PSA response was defined as a confirmed ≥50% decline from the PSA level measured closest to the start of PARPi treatment. Ten candidate variables per endpoint were selected via a random forest-based feature selection. Cox models stratified by study were used for OS, and logistic regression models with study included as an additional covariate were used for PSA response. Similarly, non- BRCA mutation status was defined as the presence of any observed mutation in genes other than BRCA , including ATM , CHEK2 , MSH2 , PALB2 , BRIP1 , and BARD1 . Results: The combined dataset included n = 327 patients, with heterogeneity in baseline characteristics and treatments across studies. The prevalence of BRCA ( BRCA1 or BRCA2 ) and non BRCA mutations was 39% and 37% respectively. Median OS for patients with BRCA mutations was 21 months (95% CI:16–28), compared with 15 months (95% CI: 14–18) for those without BRCA mutations. No difference in median OS was observed between patients with and without non- BRCA mutations: 16 months (95%CI: 14-22) and 17 months (15-21), respectively. Cox models identified age, N stage, PSMA–lutetium therapy, PARPi use within a clinical trial, prior taxane treatment (docetaxel or cabazitaxel), and BRCA as important predictors of OS. The corresponding hazard ratios with 95% confidence intervals are shown in the table below. Conclusions: These results from real-world data highlight the importance of incorporating genomic information and treatment-specific variables when predicting outcomes in advanced prostate cancer and demonstrate the utility of multi-study integrative analyses. Predictor Hazard Ratio (95% CI) Age 1.02 (1.01–1.04) N stage 1.91 (1.18–3.12) PSMA–lutetium therapy 0.35 (0.20–0.62) PARPi use (within clinical trial) 0.68 (0.48–0.98) Prior taxane treatment (docetaxel or cabazitaxel) 1.44 (1.02–2.02) BRCA mutation 0.50 (0.36–0.69)