A Calibrated Clinicogenomic Framework for Glioblastoma Survival Prediction

Le résumé fourni par la source

Glioblastoma (GBM) remains the most lethal primary brain tumour, motivating calibrated clinicogenomic risk models that can guide treatment escalation. Existing studies often rely on single modalities, leading to optimistic discrimination, poor calibration, and limited clinical utility. We assembled a matched TCGA cohort integrating routine clinical variables with RNA-seq expression and prespecified endpoints, asking whether genomics adds incremental, decision-relevant value beyond a strong clinical baseline. To address selection bias and overfitting, we used a leakage-free pipeline with nested cross-validation, pathway-level aggregation for transcriptomic representation, probabilistic calibration, and decision-curve analysis. Relative to the clinical-only model, the clinical+genomic model achieved superior discrimination, improved calibration, and higher net benefit across treatment-escalation thresholds. Time-to-event sensitivity analyses corroborated these gains. Feature attribution and pathway aggregation located the added signal in mitochondrial-translation and microenvironmental programmes that complement age rather than replace clinical signal. These results show that disciplined genomic integration delivers statistically robust, well-calibrated, and decision-relevant improvements for GBM risk stratification.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.