Differential association of post-testosterone recovery PSA kinetics with progression and metastatic outcomes in prostate cancer.

e17122 Background: Prostate-specific antigen (PSA) kinetics following testosterone recovery from androgen deprivation therapy (ADT) may reflect heterogeneous tumor biology and distinct failure mechanisms. We evaluated the association between post-testosterone recovery PSA dynamics and progression and metastasis using Bayesian joint modeling, with sensitivity analyses accounting for competing mortality. Methods: Individual patient data were pooled from two randomized trials: (1) intermediate-risk prostate cancer treated with 6 months of neoadjuvant versus adjuvant ADT with prostate radiotherapy (RT), and (2) high-risk prostate cancer treated with 36 months of ADT with prostate and pelvic RT with 3D-CRT versus helical tomotherapy. Serial PSA measurements obtained up to 18 months after testosterone recovery to supracastrate levels (> 50 ng/dL) but before progression were analyzed using Bayesian joint models linking longitudinal mixed-effects models of log-transformed PSA with survival submodels for progression-free survival (PFS) and metastasis-free survival (MFS). Models adjusted for age, baseline PSA, Gleason score, tumor stage, and ADT duration. Associations of PSA slope and cumulative PSA exposure with event risk were estimated via shared random effects. Results: A total of 402 patients contributed 1,471 post-testosterone recovery PSA measurements. With a median follow-up of 117 months (IQR, 90-150), 167 PFS-related events and 144 MFS-related events were observed. In joint modeling, steeper PSA slope was strongly associated with higher progression risk. With approximately 1% relative increase in PSA per month - the hazard of progression or death increased by 11% (posterior median HR: 1.11; 95% credible interval [CrI]: 1.01-1.22). The association between PSA slope and MFS differed in direction (posterior median HR 0.88; 95% CrI 0.78-0.99). Cumulative PSA exposure showed no clinically meaningful association with either endpoint. In cause-specific Cox analyses censoring deaths, PSA slope remained strongly associated with progression (per approximately 1% relative increase in PSA per month, posterior median HR was 1.20 [95% CrI 1.08-1.34]), whereas its association with distant metastasis was attenuated (posterior median HR 1.00; 95% CrI 0.74-1.32). Conclusions: PSA slope after testosterone recovery is a robust marker of progression risk but has a framework-dependent relationship with metastasis. These findings indicate that post-testosterone recovery PSA kinetics should not be interpreted as a uniform surrogate of metastatic risk. Instead, PSA dynamics likely reflect failure phenotype, surveillance intensity, and clinical intervention rather than metastatic potential alone. Bayesian joint modeling provides a powerful framework for endpoint-specific interpretation of PSA behavior after RT and ADT.