The vector representation of electronic health record contents has been increasingly applied to clinical text classification problems. Our study investigates what information contained in clinical notes contributes to classifying patient cancer types. We jointly used words and standard terminology concept terms found in the notes. The model using word embeddings with refined synonymous terms identified by our automated method allowed for significantly improved cancer type classification.

Learning Objective: jointly use words and standard terminology concepts to improve document classification for identification of patients with specific cancer types

Authors:

Youngjun Kim (Presenter)
Medical University of South Carolina

Stephane Meystre, Medical University of South Carolina

Presentation Materials:

• Abstract/Manuscript - 1