This event is designed for U-M IOE PhD students and faculty and is also open to all U-M students, faculty and staff.

Title:
Individualized Medical Decision Learning for Chronic Diseases

Abstract:
Chronic diseases are the leading causes of death, and leading drivers of national health care costs in the U.S. Healthcare management for chronic diseases involve many medical decisions such as whether and when to conduct different types of biomarker tests and medical treatments. Optimizing such medical decisions can be a challenging engineering problem for several reasons. First, the patients’ health states can progress stochastically over time and are typically not directly observable. Second, the patient heterogeneity in the disease progression and clinical effectiveness are prevalent in many settings. Third, there are often trade-offs between alternative decisions that must be considered to prevent both adverse health outcomes and potential side effects. In this talk, I will present a data-driven optimization framework for individualized medical decision learning. I start from descriptive and predictive analytics by fitting machine learning models to the electronic medical record (EMR) data. Then, I use a model-based reinforcement learning method to find the optimal testing policy for each patient. To address the issue of parameter ambiguity caused by the patient heterogeneity and estimation error, I propose a new multi-model partially observable Markov decision processes (POMDPs) method to find the best solution when model parameters are only known with uncertainty. Although focused on prostate cancer active surveillance, this work can be easily applied to other applications in healthcare, robotics, and marketing analytics.

Bio:
Weiyu Li is a Ph.D. candidate in Industrial and Operations Engineering at the University of Michigan. He earned a Master of Science in Statistics from the University of Michigan, and a Bachelor of Science in Mathematics from Tsinghua University, China. His research focus is on the interdisciplinary study of statistical/machine learning methods and stochastic optimization. Specific applications are in data-driven sequential decision-making in healthcare operations, including individualized treatment decisions for cardiovascular disease and active surveillance strategies for prostate cancer. Weiyu has an ongoing collaboration with the Movember Foundation and the U.S. Department of Veterans Affairs.