Abstract 7157: Development of a comprehensive multigene assay to predict the prognosis of localized or locally advanced prostate cancer

Abstract Background: Prostate cancer is the most prevalent malignancy in men worldwide. However, the absence of reliable biomarkers for metastasis due to intra- and inter-tumor heterogeneity complicates the establishment of personalized treatment guidelines. To address tumor heterogeneity, a next-generation sequencing (NGS)-based multigene assay, including gene expression, tumor-related pathways and fusion variants, is proposed for developing a new assay. This study aims to develop and validate a metastasis risk prediction model and a molecular subtype classification model to enable personalized treatment. Methods: This study included surgical specimens from patients with localized or locally advanced prostate cancer at Seoul National University Hospital and Severance Hospital. Patients were stratified into three cohorts: NED (no evidence of disease; no metastasis within seven years), BCR (biochemical recurrence; PSA elevation within five years), and METS (metastasis within five years). Propensity score matching (PSM) balanced clinical risk factors, and whole transcriptome sequencing was performed on formalin-fixed paraffin-embedded primary tumor tissues to identify differentially expressed genes (DEGs) associated with metastasis. The cancer-related pathways were selected using DEGs in the following methods: Gene Ontology term enrichment analysis and curated pathways, and their scores were calculated using Gene Set Variation Analysis. A Metastasis prediction model was developed using classification and regression algorithms to predict metastatic risk based on DEGs and pathway scores. For molecular subtyping, a improved model was optimized by incorporating previous model with fusion status. Both models were validated using an independent cohort of 172 patients. Results: Following PSM matching, 354 patients were analyzed, resulting in the identification of 103 DEGs and 33 pathways. Using these features, the metastasis prediction model achieved the highest performance (AUC = 0.904) with the Random Forest algorithm and 0.82 in the validation cohort. This performance surpassed that of regression models and joint models combining classification and regression features. Kaplan-Meier (KM) analysis confirmed the clinical relevance, showing significant differences in 5-year metastasis-free survival (MFS) between high- and low-risk groups (p < 0.0001, hazard ratio [HR]: 4.958). The improved subtyping model classified samples into two subtypes: Luminal (69%) and aggressive variant prostate cancer (AVPC, 31%). The 5-year MFS by molecular subtype for all patients (n=526) showed that the AVPC tended to have a worse prognosis (p=0.084, HR: 1.389) Conclusion: This study can be used to advance risk stratification models for metastasis risk and to inform treatment decisions by comprehensively evaluating gene expression and gene fusion events in prostate cancer. Citation Format: Jiyoung Ahn, Hyunho Han, Jang Hee Han, Jongkeun Lee, Danbi Lee, Kyung Chul Moon, Young Deuk Choi, Chang Wook Jeong, Nam Hoon Cho, Woosung Chung. Development of a comprehensive multigene assay to predict the prognosis of localized or locally advanced prostate cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 7157.