event-icon
Description

Analyzing information on whole-person functional activities is important for improving quality of life. However, extracting this information is challenging and suffers from poor coverage in controlled terminologies. We use automatically-extracted text features to classify the types of mobility-related activity described in a recent dataset of annotated physical therapy notes, using a variety of popular methods. We find that neural concept embeddings outperform other text features for the task, and identify robustness to label skew as an area of future work.

Learning Objective: Understand use of automatically-extracted text features to detect types of mobility activity described in free text clinical records.

Authors:

Denis Newman-Griffis (Presenter)
National Institutes of Health Clinical Center

Ayah Zirikly, National Institutes of Health Clinical Center
Pei-Shu Ho, National Institutes of Health Clinical Center
Jonathan Camacho Maldonado, National Institutes of Health Clinical Center
Maryanne Sacco, National Institutes of Health Clinical Center
Alex Marr, National Institutes of Health Clinical Center
Albert Lai, Washington University in St. Louis
Eric Fosler-Lussier, The Ohio State University

Presentation Materials:

Tags