Ken Jung,
Research Scientist,
BMIR, Shah Lab, Stanford University

ABSTRACT:

Most published work on predictive models intended for use in clinical applications focus narrowly on technical validation rather of the models themselves, while community wisdom about the how to use these technical artifacts to move the dials we truly care about, which are usually patient outcomes and physician behavior, is often given short shrift.  In this talk I will try to convey some of the lessons the Shah Lab has learned as we have moved a predictive model intended to improve access to palliative care into actual use, with the eventual goal of a prospective study evaluating its impact on patient and physician centered outcomes.  In particular, I will focus on the use case and operational constraints that impact model development and evaluation, along with some insights into what the real stumbling blocks are for this work making a difference outside of academic circles.  There may also be amusing anecdotes about how the popular “science” media operates.