Ethnicity-specific molecular subtypes and a machine-learning risk model in Asian patients with non-muscle-invasive bladder cancer.

Non-muscle-invasive bladder cancer (NMIBC) has high rates of recurrence and progression, yet its molecular landscape across different ethnicities remains poorly defined. We assembled a meta-cohort of 202 Asian NMIBC transcriptomes and identified three novel molecular subtypes (AC0, AC1, AC2) using non-negative matrix factorization clustering. These subtypes, each associated with distinct biological pathways and clinical outcomes, were externally validated across 22 datasets. Subtype-specific biomarkers (RABL6, MYBL2, RAD54L, and FAM64A) were confirmed by immunohistochemistry in an independent Asian cohort (n = 210), enabling the development of a machine-learning-based risk stratification model. AC1, characterized by aggressive features and enrichment of cell cycle and oncogenic signaling pathways, exhibited the poorest prognosis. Notably, after adjustment for T-stage and grade, the combined risk model was independently prognostic in Asian cohorts but not in European cohorts, underscoring ethnic differences in NMIBC biology. Our findings delineate distinct molecular subtypes and introduce a clinically applicable, ethnicity-specific prognostic model for Asian patients with NMIBC. This study emphasizes the importance of incorporating ethnic diversity into precision oncology frameworks and provides a foundation for personalized therapeutic and surveillance strategies in underrepresented populations.