Predictive value of a pathomics signature in <i>de novo</i> metastatic prostate cancer: An ancillary study of the PEACE-1 phase 3 trial.

Abstract

221 Background: The addition of abiraterone acetate plus prednisone (AAP) to docetaxel and androgen deprivation therapy (ADT) has become a standard of care for metastatic castration-sensitive prostate cancer (mCSPC) following the PEACE-1 trial. This ancillary study aimed to identify pathomics-based predictive biomarkers associated with radiographic progression-free Survival (rPFS) and overall survival (OS) to guide therapeutic decision-making. Methods: Hematoxylin-Eeosin-Safran (HES) slides were digitized using an Olympus VS120 scanner at a 20x magnification. A multiple-instance learning framework was applied integrating patch-level treatment attention with tissue composition inferred from HoVer-Net segmentation to generate a composite pathomic treatment-benefit score. Patients were stratified into high- and low-score groups using the median value. Cox proportional hazards models adjusted for age, ECOG performance status, disease burden, Gleason score, type of castration, and other treatment received (radiotherapy, docetaxel) assessed prognostic and predictive associations. Interaction tests evaluated whether the pathomic score predicted AAP benefit. Results: Among the 1172 patients (pts) randomized in PEACE-1 (NCT01957436), 595 had available FFPE tumor samples and 526 HES slides were analyzable by pathomics after central review. Baseline characteristics were comparable between the full and pathomics cohorts. In patients with low AAP-benefit scores (n=263), the addition of AAP to standard of care (SOC) did not significantly improve rPFS or OS (HR = 0.80; 95% CI 0.59–1.07; p = 0.14 and HR = 1.00; 95% CI 0.72–1.40; p = 0.98, respectively). Conversely, patients with high AAP-benefit scores (n=263) derived substantial benefit from AAP (rPFS: HR = 0.38; 95% CI 0.28–0.51; p &lt; 0.001; OS: HR = 0.48; 95% CI 0.34–0.67; p &lt; 0.001). Interaction tests confirmed the predictive effect of the pathomic signature for both rPFS (p &lt; 0.001) and OS (p = 0.001). In term of model performance, the area under the curve (AUC) for interaction between AAP and the two score groups to predict rPFS and OS were AUC 0.712 and AUC 0.722 respectively. Pts with low AAP-benefit scores displayed lower ki67 log score in IHC (p=0.013), higher mean AR z-score in transcriptomics (p=0.033), more non-neoplastic cells (p&lt;0.001), fewer neoplastic cells (p=0.004) and less necrosis (p&lt;0.001). No additional predictive biomarker was identified across (IHC), genomic or transcriptomic analyses. Conclusions: This study identified a pathomics-derived histological signature predictive of benefit from AAP in patients with de novo mCSPC treated with ADT ± docetaxel. External validation is warranted to confirm these findings and evaluate its potential for clinical implementation in precision treatment selection.