Development and external validation of a psychological stress-related gene signature for overall survival prediction in bladder cancer

Background: Psychological stress may influence tumor progression through neuroendocrine, inflammatory, immune, and microenvironmental mechanisms, but its prognostic relevance in bladder cancer (BLCA) remains incompletely defined. We aimed to develop and externally validate a psychological stress-related gene (PRG) signature for overall survival (OS) and characterize its biological context.

Methods: This retrospective study used The Cancer Genome Atlas bladder cancer cohort (TCGA-BLCA; n=404, 176 deaths) for development and Gene Expression Omnibus (GEO) GSE48075/GSE48277 (n=73, 45 deaths) for external validation. Candidate PRGs underwent differential-expression analysis, univariate Cox screening, and least absolute shrinkage and selection operator (LASSO)-Cox modeling. The fixed 23-gene model was evaluated by survival and prognostic-performance analyses. Post hoc analyses assessed functional annotation, clinicopathological subgroups, stress-related pathway scores, immune functional-state and Tumor Immune Dysfunction and Exclusion (TIDE) features, drug-pathway associations, and network-defined hub genes.

Results: High-risk patients had worse OS in TCGA (P<0.001) and GEO (P=0.006). Apparent 1-, 3-, and 5-year TCGA areas under the receiver operating characteristic curve were 0.741 [95% confidence interval (CI), 0.679–0.798], 0.742 (95% CI, 0.684–0.800), and 0.774 (95% CI, 0.716–0.831); GEO discrimination was modest. Prognostic associations were directionally consistent across major subgroups without significant interaction. Risk score correlated positively with all six stress-related programs, most strongly extracellular matrix (ECM) organization (rho =0.509) and angiogenesis (rho =0.497; both Benjamini-Hochberg-adjusted P<0.001). High-risk tumors showed higher checkpoint expression, cytolytic activity, T-cell-inflamed scores, and TIDE/exclusion signals. Forty of 54 drug-pathway correlations remained significant after false-discovery-rate correction. MYC, JUN, EGR1, SREBF1, and FOSL1 were the top network hubs; all were prognostic in TCGA after correction, whereas GEO estimates were directionally concordant but not significant after correction.

Conclusions: The fixed PRG signature provided research-level prognostic stratification and was associated with stress-related inflammatory, immune, and tumor-microenvironment programs in BLCA. Because external discrimination was modest and the post hoc biological analyses were exploratory, independent prospective and experimental validation is required before clinical use.