Feed

Survival analysis plays a critical role in predicting patient outcomes and guiding personalized cancer therapies. Although multi-omics data provide rich biological insights, their high dimensionality poses significant challenges for robust analysis and clinical implementation. While many studies rely on the traditional Cox proportional hazards model, few have explored alternative survival algorithms combined with rigorous feature selection to identify low-dimensional, clinically feasible prognostic signatures that retain strong predictive power comparable to models using the full feature set. To address these gaps, we developed PRISM (PRognostic marker Identification and Survival Modelling through Multi-omics Integration), a comprehensive framework aimed at improving survival prediction and discovering minimal yet robust biomarker panels across multiple omics modalities. PRISM systematically evaluates various feature selection methods and survival models through a robust pipeline that selects features within single-omics datasets before integrating them via feature-level fusion and multi-stage refinement. Applied to TCGA cohorts of Breast Invasive Carcinoma (BRCA), Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma (CESC), Ovarian Serous Cystadenocarcinoma (OV), and Uterine Corpus Endometrial Carcinoma (UCEC), PRISM revealed that cancer types benefit from unique combinations of omics modalities reflecting their molecular heterogeneity. Notably, miRNA expression consistently provided complementary prognostic information across all cancers, enhancing integrated model performance (C-index: BRCA 0.698, CESC 0.754, UCEC 0.754, OV 0.618). PRISM advances cancer prognosis by delivering scalable, interpretable multi-omics integration and identifying concise biomarker signatures with performance comparable to full-feature models, promoting clinical feasibility and precision oncology.