2008 — Reducing Dependency on Manual Chart Review through Automated Information Extraction Methods

D'Avolio L (VA Boston Healthcare System, MAVERIC), South B (IDEAS Center, VA Salt Lake City Health Care System, Univ. of Utah), Shen S (IDEAS Center, VA Salt Lake City Health Care System, Univ. of Utah), Garvin JH (PVAMC, Center for Health Equity Research and Promotion, NewCourtland Center for Transitions and Health, Univ. of Pennsylvania), Goldstein M (VA Palo Alto Health Care System, GRECC, Stanford University, Center for Primary Care and Outcomes Research), Samore M (IDEAS Center, VA Salt Lake City Health Care System, Univ. of Utah)

Workshop Objectives:
Manual chart review is a primary method of collecting information for performance measurement, outcome assessment, and clinical classification. It is time consuming and costly. This workshop will address the potential to replace manual chart review with automated natural language processing (NLP) techniques. Presenters will: 1) describe simple and advanced approaches to text-processing; 2) discuss the types of information that are feasibly extracted from non-structured data sources such as VistA text documents and; 3) provide examples of the use of NLP techniques to complement and expand structured data sources to support performance monitoring and health services research within VA.

Activities:
Presenters will provide information about different types of text processing systems, explain how text processing outputs are coded, and discuss methods of evaluation. Participants will be invited to share questions and interests about the use of NLP techniques in their research. New programmatic initiatives will be discussed, including the Consortium for Healthcare Informatics Research (CHIR).

Target Audience:
VA researchers interested in extracting and utilizing information found in VistA non-structured electronic free-text for research, decision support, or quality improvement efforts.

Assumed Audience Familiarity with Topic:
Although most VA researchers are familiar with the richness of structured data available from local and national data sources, few have experience extracting data available from non-structured data sources such as the text documents in VistA. Automated methods to reliably extract and codify information found in non-structured VA data sources hold great promise to gather large amounts of data for research, yet there is limited familiarity with techniques to extract information from text in this manner within the VA research community.