20909 Detection of Widened Mediastinum in Radiology Text Reports: BioSense

Sunday, August 30, 2009
Grand Hall/Exhibit Hall
Stephen R. Benoit, MD, MPH , NCPHI, CDC, Atlanta, GA
Roseanne English, BS , NCPHI, CDC, Atlanta, GA
Jerome I. Tokars, MD, MPH , NCPHI, CDC, Atlanta, GA

Background:  Bacillus anthracis is one of the most feared biological weapons.  Although not specific for inhalational anthrax, a finding of widened mediastinum on chest radiography is a sensitive indicator of pulmonary anthrax that may be available before laboratory testing.  The purpose of this study was to determine the overall and facility-specific proportion of chest radiographs with a finding of widened mediastinum in BioSense, an automated surveillance system receiving radiology data from 46 hospitals. 

Methods:  We identified all free-text chest radiograph reports sent to BioSense from January 1, 2008 through February 28, 2009.  We searched for the terms “mediastin” and “wide” within the same sentence and eliminated records with negations (e.g., “absent”, “free of”, “negative”, “without”, “no”).  Finally, for facilities with ≥1,000 radiographs, we calculated the overall and facility-specific proportion of chest radiographs with widened mediastinum.

Results:  There were 1,451,665 chest radiographs from 46 hospitals in 14 states during the 14-month study period.  Of those, 11,121 mentioned “mediastin” and “wide”.  After applying negation rules, 5,546 radiographs (3,908 unique patients) showed a widened mediastinum.  Of 43 facilities with ≥1,000 chest radiographs, the overall and facility-specific proportions were 0.38% and 0.05-1.44%, respectively.  Of those with widened mediastinum, the mean/median age was 61.8 and 63.7, respectively.

Discussion:  A widened mediastinum on chest radiography is a sensitive indicator for inhalational anthrax and by knowing the background proportion of chest radiographs with this finding, anomaly algorithms can be applied to detect clusters that may indicate bioweapon attacks.  We plan to extend this study using formal natural language processing methods.

See more of: Posters
See more of: Submissions