“Decoupling Tumor Detection and Survival Signals Reveals a Context-Dependent Role of NDRG2 Across Cancers”

Cancer transcriptomic studies frequently report strong classification performance; however, whether genes associated with tumor detection retain independent relevance for patient survival remains unclear. In this study, we evaluated the relationship between tumor-discrimination and survival-associated gene-expression signals using a revised and reproducible cross-cancer analytical framework. Publicly available RNA sequencing and clinical datasets were obtained from The Cancer Genome Atlas (TCGA) through the UCSC Xena platform. Breast invasive carcinoma (BRCA) was used as the discovery cohort, while lung adenocarcinoma (LUAD), colon adenocarcinoma (COAD), and liver hepatocellular carcinoma (LIHC) served as independent validation cohorts. To reduce information leakage, feature selection was performed exclusively within the training subset using repeated Random Forest–based evaluation, followed by independent validation. Classification performance was assessed using receiver operating characteristic analysis and complementary performance metrics. Survival analyses included Kaplan–Meier and multivariable Cox regression models. The revised workflow identified a nine-gene panel consisting of GPR125, SCHIP1, NDRG2, FBXO7, FGF2, BCAM, SLC2A12, GYS2, and COL10A1. Internal validation in BRCA showed strong classification performance (AUC = 0.999), while external validation retained performance in LUAD (AUC = 0.967), COAD (AUC = 0.999), and LIHC (AUC = 0.852). Exploratory enrichment analysis suggested associations with signaling, endothelial regulation, and metabolic processes. However, survival associations weakened after clinical adjustment, and NDRG2 did not retain independent significance in multivariable analysis. Overall, the findings suggest that transcriptomic features associated with tumor discrimination do not necessarily translate into independent prognostic relevance and highlight the importance of reproducible validation strategies in biomarker research.