For studying cancer biology, multiomics data integration has become an invaluable tool. With the amount of -omics data, its availability and the pace of its acquisition increasing rapidly, multiomics data integration is even more pivotal. This work employs a novel adaptation of the expectation maximization routine for joint latent variable modeling of multiomics patient profiles. Along with traditional methods of biological feature selection, the data-centric approach toward latent distribution optimization can adequately cluster patients from well-studied cancer types and does so with lower computational expense. Crucially, this work modifies the optimization subroutines in the relatively standard joint latent variable -omics workflow for improved survival analysis and run-time performance. This work also provides a framework that identifies distinctions between cancer subtypes and proposes potential biomarkers for breast cancer.