June 5, 2017 2:00 PM – 3:30 PM
Dmitriy Babichenko University of Pittsburgh School of Information Sciences
Marek Druzdzel, University of Pittsburgh School of Information Sciences
James McGee University of Pittsburgh School of Medicine
Recent advances in machine learning, big data analytics, scalable web frameworks, and availability of large patient cohort datasets from Electronic Medical Record (EMR) systems created an opportunity to develop a new type of a virtual patient (VP) system, where cases are based on and simulate real patient clinical treatment processes and outcomes.
We are developing “Model Patient” - a novel VP system that uses probabilistic machine learning models learned from EMR data to generate a framework for VP cases. Because the underlying models are based on real patient data, each decision made by a learner affects the probability of each possible outcome in the same way as these outcomes would have been affected in real patients. Therefore, the behavior of a VP reflects how a real patient with the same medical condition would have reacted to the learners' actions.
Through examples, we will demonstrate:
How probabilistic models can be created (learned) from EMR data
How models are used in the context of virtual patient case authoring
- How decisions made by learners affect virtual patient outcomes