Individualized dynamic risk assessment and treatment selection for multiple myeloma.

Abstract

Background: Individualized treatment decisions for multiple myeloma (MM) patients require accurate risk stratification that accounts for patient-specific consequences of cytogenetic abnormalities on disease progression.

Methods: Previously, SYstems Genetic Network AnaLysis (SYGNAL) of multi-omics tumor profiles from 881 MM patients generated a mmSYGNAL network of transcriptional programs underlying disease progression across MM subtypes. Here, through machine learning on activity profiles of mmSYGNAL programs we have generated a unified framework of cytogenetic subtype-specific models for individualized risk classifications and prediction of treatment response.

Results: Testing on 1,367 patients across five independent cohorts demonstrated that the framework of mmSYGNAL risk models significantly outperformed cytogenetics, International Staging System, and multi-gene biomarker panels in predicting PFS at primary diagnosis, pre- and post-transplant and even after multiple relapses, making it useful for individualized risk assessment throughout the disease trajectory. Further, treatment response predictions were significantly concordant with efficacy of 67 drugs in killing myeloma cells from eight relapsed refractory patients. The model also provided new insights into matching MM patients to drugs used in standard of care, at relapse, and in clinical trials.

Conclusion: Activities of transcriptional programs offer significantly better prognostic and predictive assessments of treatments across different stages of MM in an individual patient.