Candidate plasma proteins associated with all-cause mortality among patients with prostate cancer: a UK biobank analysis

Evidence remains limited regarding circulating protein signals associated with long-term mortality among patients with prostate cancer (PCa). This study aimed to identify circulating proteins associated with long-term mortality in PCa patients and assess their potential incremental prognostic information and biological plausibility. We enrolled 353 male PCa patients with available baseline plasma proteomic data from the UK Biobank. Candidate proteins were prioritized using multivariable Cox regression with multi-model adjustment, Bonferroni correction, and bootstrap stability assessment (coefficient of variation (CV) < 20%). For further robustness testing, 200 bootstrap resampling iterations were performed, with full proteome-wide screening and Bonferroni correction repeated per resample; candidate proteins were prioritized according to final reselection frequency. Restricted cubic spline, landmark, time-window and cross-sectional analysis of post-baseline residual survival assessed the robustness and cross-sectional baseline protein patterns of key protein–all-cause mortality associations. Exploratory risk-estimation models were built in an internally validated framework with bootstrap optimism correction; missing-value imputation, covariate selection, variable standardization, and model fitting were all conducted within each bootstrap training sample. Fine–Gray competing-risk models were applied to explore proteins linked to PCa-specific mortality. Mechanistic analyses covered protein–protein interaction networks, pathway enrichment, module construction, mediation analysis, module interaction, and module-combination survival analysis. During follow-up, 113 all-cause deaths (36 PCa-specific) were recorded. growth differentiation factor 15 (GDF15), tenascin C (TNC), tumor necrosis factor receptor superfamily member 10B (TNFRSF10B), and ribonucleotide reductase regulatory subunit M2 (RRM2) were the four proteins most strongly associated with all-cause mortality, with the highest reselection frequencies. The composite model integrating these four proteins with available clinical and selected covariates showed the highest internally validated discrimination, with a 15-year all-cause mortality AUC of 0.7835 (95% CI, 0.7303–0.8265), corresponding to a modest 6.77 percentage-point increase over the baseline covariate model (ΔAUC 95% CI, 1.84–10.63 percentage points; P = 0.01; Supplementary Table S10). Within this cohort, this model showed acceptable calibration and a higher estimated net benefit in decision curve analysis, but its clinical utility requires external validation. Death receptor–mediated immune apoptosis, extracellular matrix remodeling, and replication-stress-related mechanisms may be involved. galanin (GAL), prokineticin 1 (PROK1), and guanylate-binding protein 2 (GBP2) were identified as PCa-specific mortality–associated proteins with varying nonlinear associations and cross-sectional baseline patterns by future time to PCa-specific death; these findings are hypothesis-generating given limited PCa-specific death events. GDF15, TNC, TNFRSF10B, and RRM2 may represent candidate circulating protein signals associated with post-baseline all-cause mortality risk among prevalent PCa survivors.