Some Recent Developments in Machine Learning for Addressing Multiple Outcomes in Data Driven Decision Support
Michael Kosorok, (UNC-Chapel Hill)
In this presentation, we will discuss some recent developments in machine learning based precision health for estimating dynamic treatment regimes which seek to balance two or more outcomes, such as, for example, balancing treatment efficacy against side effect burden. The first approach involves obtaining individual patient input on the relative priority of outcomes through preference instruments based on, for example, item response theory or discrete choice modeling. The second approach adapts inverse reinforcement learning to infer physician outcome priorities, allowing for the possibility of errors in decision making. Both approaches are illustrated with practical examples from mental health.
Michael R. Kosorok, PhD, is W. R. Kenan, Jr. Distinguished Professor of Biostatistics and Professor of Statistics and Operations Research at UNC-Chapel Hill. His research expertise is in biostatistics, data science, machine learning and precision medicine, and he has written a major text on the theoretical foundations of these and related areas in biostatistics (Kosorok, 2008, Springer) as well as co-edited (with Erica E. M. Moodie, 2016, ASA-SIAM) a research monograph on dynamic treatment regimes and precision medicine. He also has expertise in the application of biostatistics and data science to human health research, including cancer and cystic fibrosis. In particular, he is the contact principal investigator on an NCI program project grant (P01 CA142538), which focuses on statistical methods for novel cancer clinical trials in precision medicine, including biomarker discovery and dynamic treatment regimes. He has pioneered machine learning and data mining tools for these and related areas.