Abstract: Machine learning approaches to understanding patient risk are becoming increasingly popular in health care. Unfortunately, these models often present patient risk as a single aggregate measure of "high" or other risk, rather than illustrating the specific factors driving a given patient's overall risk. We developed a random forest prediction algorithm for congestive heart failure and a visualization tool for clinicians to understand population risk distribution, overall patient risk, and patient-specific factors driving those risk scores.

Describe the new knowledge and additional skills the participant will gain after attending your presentation.: Session attendees will be able to understand why risk prediction models illustrating aggregate risk scores for any given outcome are unlikely to be clinically actionable. We illustrate effective interactive visualizations of patient-level factors driving aggregate risk scores derived from prediction algorithms. By extending these examples to their own organizations, attendees will be better equipped to develop visualizations of actionable, patient-specific drivers of patient risk for use by clinicians and care managers.

Authors:

Nate Apathy (Presenter)
Indiana University

Anna Roberts, Regenstrief Institute
Ross Hayden, Regenstrief Institute
Christopher Harle, Indiana University
Joshua Vest, Indiana University