Policy Learning and Evaluation

Personalized sequential decisions and their expected outcomes.

Projects

Blood pressure (BP) is a key modifiable factor after endovascular therapy (EVT); excessive hypertension and aggressive BP lowering may both be harmful. Appropriate BP targets may differ between patients, and treatment decisions need to account for evolving BP, physiology, neurologic status and previous treatment history. Rather than relying on one universal target, we use a causal dynamic treatment regime (DTR) framework to learn personalized, history-dependent decisions that aim to improve outcomes.

Six sequential treatment decisions, backward Q-learning of personalized policies, and estimation of the expected outcome under a specified policy.