Talk to the Veterans Crisis Line now
U.S. flag
An official website of the United States government

VA Health Systems Research

Go to the VA ORD website
Go to the QUERI website

HSR Citation Abstract

Search | Search by Center | Search by Source | Keywords in Title

Relatedness Analysis of LVEF Qualitative Assessments and Quantitative Values

Kim Y, Garvin JH, Heavirland J, Meystre S. Relatedness Analysis of LVEF Qualitative Assessments and Quantitative Values. Poster session presented at: American Medical Informatics Association Spring Congress; 2013 Mar 20; San Francisco, CA.


Relatedness Analysis of LVEF Qualitative Assessments and Quantitative Values Youngjun Kim1,3, Jennifer Garvin2,3, Julia Heavirland3, St phane M. Meystre2,3 1School of Computing, 2Department of Biomedical Informatics, University of Utah, 3VA Health Care System, Salt Lake City, Utah Abstract: The goal of this research is to analyze how left ventricular ejection fraction (LVEF) qualitative assessments and quantitative values are related, to improve the usability of qualitative assessments when no quantitative value is available. From pairs of those indicators extracted from clinical notes, we analyzed the most frequent qualitative assessment terms for each quantitative value range. We also trained a classifier to classify qualitative assessments as normal (LVEF 40%) or abnormal (LVEF < 40%), and reached an accuracy of over 98%. Introduction: LVEF qualitative assessments (e.g., "markedly diminished EF") and quantitative values (e.g., "LVEF 45-50%") are important indicators to monitor the progression and treatment of congestive heart failure (HF) and our efforts aim at automatically extracting these indicators from clinical notes1, 2. In this study, we analyzed how qualitative LVEF assessment terms are associated with quantitative LVEF values, to improve the usability of qualitative assessments when no quantitative value is available, and automatically classify these qualitative assessments as normal (LVEF 40%) or abnormal (LVEF < 40%) Methods: We extracted 615 possible pairs of qualitative assessments and quantitative values found within two sentence boundaries of each other, from a corpus of 769 VHA echocardiogram reports containing 1,973 LVEF mentions (e.g., "EF"), 1,176 quantitative LVEF values (e.g., "60~70%, 0.45), and 1,198 qualitative LVEF assessments (e.g., "NORMAL", "severely reduced"). After manual review, 606 pairs were annotated as valid relations. We trained our Support Vector Machines (SVM) classifier with those pairs to classify qualitative LVEF assessments as abnormal (less than 40%) or normal. We only used the qualitative assessment terms as features. Results: We normalized and grouped quantitative LVEF values in tens (e.g., 14% to 10%, 0.45 to 40%), and associated them with the corresponding qualitative assessments, as displayed in the table below. For example, "severe" was the most frequent qualitative assessment with 10-19% LVEF values. Our classifier reached 98.02% accuracy in 5-fold cross validation. Most errors happened with the 40-49% range instances. When excluding those instances, accuracy reached 99.46%. EF (%) Five most common EF qualitative assessments # 10-19 severe (3), diminished (1), markedly diminished (1) 5 20-29 severe (14), moderate to severely reduced (8), severely reduced (8), severely decreased (6), moderate to severely decreased (5) 47 30-39 moderately decreased (13), mild to moderately reduced (5), moderately reduced (5), diminished (3), moderate (3) 33 40-49 mildly reduced (11), mildly decreased (8), mild to moderately decreased (6), diminished (4), mild (4) 50 50-59 normal (196), low normal (24), preserved (19), lower limits of normal (15), borderline reduced (3) 265 60-69 normal (98), preserved (54), remains normal (1) 153 70-79 normal (20), hyperdynamic (16), preserved (14) 50 80-89 hyperdynamic (1), normal (1), preserved (1) 3 Conclusion: We analyzed the relatedness of LVEF quantitative values and qualitative assessments and focused on the most frequent qualitative assessments for each range of quantitative values. Our classifier can be utilized to predict the LVEF quantitative value range when such quantitative values are absent or unavailable in clinical notes. Acknowledgments: Research supported by VA HSRandD IBE 09-069 and by HSRandD HIR 08-374 (Consortium for Healthcare Informatics Research) and HIR 09-007 (Translational Use Case - Ejection Fraction). References 1. Garvin JH, Duvall SL, South BR, et al. Automated extraction of ejection fraction for quality measurement using regular expressions in Unstructured Information Management Architecture (UIMA) for heart failure. JAMIA. 2012 Mar 21. 2. Meystre SM, Kim Y, Garvin JH. Comparing Methods for left Ventricular Ejection Fraction Clinical Information Extraction. AMIA Summits Transl Sci Proc, CRI. 2012:138.

Questions about the HSR website? Email the Web Team

Any health information on this website is strictly for informational purposes and is not intended as medical advice. It should not be used to diagnose or treat any condition.