Integrated molecular classification has led to the discovery of new classification systems in individual cancer types, as well as pan-cancer patterns driven by cell-of-origin or shared molecular alterations across cancer types. In order to consider clinical outcomes (such as treatment response and survival) simultaneously in tumor classification, we developed a new algorithm called SurvClust for supervised clustering. The algorithm learns a weighted distance matrix from each molecular data type with effect sizes as weights. To facilitate integration, quantile normalization is utilized to standardize the distance vectors. Multidimensional scaling (MDS) is then used to map the subjects into an n-dimensional space that preserves between-subject distances for clustering. Application to the TCGA pan-cancer datasets revealed survival associations driven by mutation burden and specific mutation and co- mutation patterns of cancer driving genes. We will further demonstrate SurvClust’s utility by integrating treatment responses and toxicity profiles in immunotherapy-treated lung, melanoma, and bladder cancer cohorts.