ESTRO 2025 - Abstract Book

S2022

Clinical - Urology

ESTRO 2025

Results:

The constructed BN included Gleason grade, extraprostatic extension, seminal vesicle invasion, age, prostate specific antigen, hypoxia signature, number of positive biopsies and perineural invasion as well as androgen deprivation and the type of radiotherapy (Figure 1). In external validation, the AUC was 0.74 (95%-confidence interval 0.64 – 0.83) for the prediction of BCR. The network was well-calibrated. Risk group assessment revealed that the quartile with the lowest predicted risks for BCR (< 1%) showed a false-negative rate of only 3.3%. Sensitivity analysis showed a worse model performance based on AIC, when the hypoxia signature was omitted from the BN. In external validation, AUC decreased to 0.66 (95%-confidence interval 0.55 – 0.78) when excluding the hypoxia signature. The hypoxia signature was prognostic within the ultra high-risk subgroup (T3-4, Gleason grade 8 10). Conclusion: Our BN can predict BCR in high-risk prostate cancer patients treated with radiotherapy and is improved with the addition of a hypoxia signature. If validated for metastasis-free survival, our BN signature could triage selected patients to hypoxia-modifying therapies to improve outcome, particularly in the ultra high-risk subgroup.

Keywords: prostate cancer, Bayesian network, hypoxia