Title:   Hypertension management: a value of information perspective

Abstract: The traditional technique for blood pressure measurements involves noise. Moreover, patients exhibit short-term and long-term variabilities in their blood pressure. Estimating the patient’s true underlying blood pressure, based on which the prescription decisions are made, will involve some degrees of the physician’s subjectivity. Such a noisy, stochastic, and subjective environment, however, constitute the basis for hypertension management. We present an analytical framework, which is a combination of Machine Learning and Dynamic Optimization, to characterize the optimal treatment policies in the above environment. We also evaluate the value of information (obtained from more accurate technology) as well as the value of optimal learning. We present several interesting results regarding the optimal prescription policies and the value of information/learning. Joint work with Mehmet Gumus (McGill), Vedat Verter (Michigan State), and Stella Daskalopoulou (McGill). Keywords: Clinical decision making, hypertension management, Markov Decision Processes, Machine Learning, value of information.