Predicting progression events in multiple myeloma from routine blood work

Abstract

Abstract This study introduces a system for predicting disease progression events in multiple myeloma patients from the CoMMpass study (N = 1186). Utilizing a hybrid neural network architecture, our model predicts future blood work from historical lab results with high accuracy, significantly outperforming baseline estimators for key disease parameters. Disease progression events are annotated in the forecasted data, predicting these events with significant reliability. We externally validated our model using the GMMG-MM5 study dataset (N = 504), and could reproduce the main results of our study. Our approach enables early detection and personalized monitoring of patients at risk of impeding progression. Designed modularly, our system enhances interpretability, facilitates integration of additional modules, and uses routine blood work measurements to ensure accessibility in clinical settings. With this, we contribute to the development of a scalable, cost-effective virtual human twin system for optimized healthcare resource utilization and improved outcomes in multiple myeloma patient care.