INFORMS conference talk focused on dynamic machine learning for medicine. Based on Joint work with Jon Seymour, MD (Peers Health) and Brian Denton PhD (University of Michigan).

Time is a crucial factor of clinical practice. Our work explores the intersection of time and machine learning (ML) in the context of medicine. This presentation will examine the creation, validation, and deployment of dynamic ML models. We discuss dynamic prediction of future work status for patients who have experienced occupational injuries. Methodologically we cover a framework for dynamic prediction health-state prediction that combines a novel data transformation with an appropriate automatically generated deep learning architecture. These projects expand our understanding of how to effectively train and utilize dynamic machine learning models in the service of advancing health.