Recent machine learning applications show great potential in patient care. However, clinical data like patient medications usually contain highly dimensional attributes that may negatively affect machine learning algorithm performance or model fit. We hypothesized that by leveraging medication classifications and mapping to higher levels of abstraction, we could translate historical information entered in the legacy EMR to the ATC terminology and reduce the dimensionality for clinical prediction in a scalable way.

Learning Objective: Formulate an approach to map patient medications to higher levels of abstraction for machine learning models

Authors:

Sina Madani (Presenter)
Vanderbilt University Medical Center

Colin Walsh, Vanderbilt University Medical Center
Joseph Huenecke, Vanderbilt University Medical Center
Jeff Byrd, Vanderbilt University Medical Center
Asli Weitkamp, Vanderbilt University Medical Center

Presentation Materials:

• Abstract/Manuscript - 1