SEMINAR

We all behave: patients, clinicians, everyone. In most clinical practices, the states and conditions of clinical situations change sequentially. We can define a trajectory to characterize a patient’s health status. Clinicians strive to improve a patient’s health through timely interactions, which include ordering laboratory tests, prescribing drugs, and conducting procedures. These interactions are repeated over time.

In this context, much of healthcare and clinical information exhibits the characteristics of time-series data. Both patients and clinicians experience feedback loops resulting from decisions, management, and interventions in clinical practice. In this seminar, we will first investigate the glucose control framework from the perspectives of reinforcement learning (RL) and simple control theory. We will then expand this framework to address general clinical problem settings, which can be resolved not only by machine learning (ML) and artificial intelligence (AI) but also by leveraging prior domain knowledge. What is considered optimal in healthcare?