Expert-Driven Survival Machines: Improving Stratification and Interpretability in Multiple Clinical Cohorts

The article introduces the AdaCSM framework, a mixture-of-experts enhanced adaptive deep clustering model for survival prediction in clinical settings. This model employs a routing-based expert mechanism to dynamically allocate patients to specialized risk predictors, improving individual representation and interpretability while preserving survival and subtype clustering objectives. The AdaCSM shows enhanced predictive performance compared to state-of-the-art models across multiple clinical cohorts, highlighting its potential for more accurate risk stratification in healthcare.