ESTRO 2025 - Abstract Book

S2021

Clinical - Urology

ESTRO 2025

2920

Digital Poster A hypoxia-based prognostic Bayesian network for high-risk prostate cancer patients treated with radiotherapy Casper Reijnen 1 , Vanesa Biolatti 2 , Conrado Guerro Quiles 2 , Ann Henry 3 , Jim Zhong 3 , Alan McWilliam 2 , Jane Shortall 2 , Marcel van Herk 2 , Robert Bristow 4 , Catharine M.L. West 2 , Johan Bussink 1 , Mark Reardon 2 , Peter Hoskin 5,6 , Ananya Choudhury 2,5 1 Radiotherapy, Radboud university medical center, Nijmegen, Netherlands. 2 Division of Cancer Sciences, University of Manchester, Manchester, United Kingdom. 3 Leeds Institute of Medical Research, University of Leeds, Leeds, United Kingdom. 4 Faculty of Biology, Medicine & Health, University of Manchester, Manchester, United Kingdom. 5 Department of Clinical Oncology, The Christie Hospitals NHS Foundation Trust, Manchester, United Kingdom. 6 Department of Clinical Oncology, Mount Vernon Cancer Centre, London, United Kingdom Purpose/Objective: Bayesian networks (BN) are valuable in medical decision making because of their ability to handle uncertainty, integrate prior knowledge, and provide probabilistic inferences for complex relationships between variables. The primary aim was to develop and externally validate a hypoxia-based prognostic BN to predict biochemical recurrence (BCR) in high- risk prostate cancer patients (based on d’Amico risk classification) treated with external beam radiotherapy (EBRT) or high-dose rate (HDR) brachytherapy combined with androgen deprivation. Additionally, we explored if the BN could identify subgroups in which the hypoxia signature was prognostic. Material/Methods: A retrospective multicentre study was performed to construct the BN including 400 patients treated with EBRT or HDR brachytherapy. The network was developed using score-based machine learning in addition to expert knowledge. Clinicopathological biomarkers, including a 28-gene expression hypoxia signature, were included. Model performance was tested with Akaike information coefficient (AIC). Next, external validation for the prediction of BCR was performed using an independent cohort of 127 patients. Model performance was quantified with area under the curve (AUC) analysis and calibration plots. Risk group assessments were performed. Sensitivity analyses were used to explore the impact of the hypoxia signature on the prediction of BCR. All statistical analysis was performed using R (4.3.1).