Description
EHR observations reflect a complex array of information about patients. Deep Learning (DL) algorithms are increasingly applied to EHR data to untangle their complexities. Despite their often-noble predictive power, the so-called “black box” DL algorithms are hard to interpret for clinical application. Feature selection and data representation are critical to the performance and interpretability of any algorithm. We propose EHR Sequencing as a novel approach to constructing highly-predictive and yet clinically-interpretable data representations from EHR Data.
Learning Objective: At the conclusion of this activity, participants will be able to discuss how EHR observations can be temporally abstracted into chronological sequences to construct predictive and interpretable data representations.
Authors:
Hossein Estiri (Presenter)
Harvard Medical School
Thomas McCoy, Massachusetts General Hospital
Shawn Murphy, Harvard Medical School
Presentation Materials: