Diving for Pearls: Indexing Mobility Information in Social Security Administration Clinical Records with a Neural Relevance Tagger
Loading...
Date
2020-02
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Locating sparse information in medical text that is relevant to self-reported functional limitations is a key challenge in the US Social Security Administration's process of determining disability. We investigate the effectiveness of a recent relevance scoring model for retrieving information related to mobility limitations, one of the most frequent allegation types in disability applications. Descriptions of mobility status are complex and difficult to extract with existing methods. We demonstrate that tagging for relevance at the token level achieves high recall on retrieving true mobility descriptions, and ranking documents by the amount of predicted mobility-relevant information achieves very strong correlation with ranking by the true number of mobility descriptions in each document. Additionally, experiments on a dataset of long, highly heterogenous documents show that our approach performs nearly perfectly at ranking documents with mobility-related information higher than those without, indicating that relevance estimation has high potential utility as a document triage tool for managing high-volume disability applications.
Description
Engineering: 2nd Place (The Ohio State University Edward F. Hayes Graduate Research Forum)
Keywords
Natural Language Processing, Artificial Intelligence, Information Extraction, Disability, Functional Status Information