The Reliability of Tweets as a Supplementary Method of Seasonal Influenza Surveillance

• DOI: 10.2196/jmir.3532
• Publication Date: 2014-11-14T15:35:22Z
• Authors: Anoshé A Aslam; Ming-Hsiang Tsou; Brian H Spitzberg; Li An; J Mark Gawron; Dipak K Gupta; K Michael Peddecord; Anna C Nagel; Christopher Allen; Jiue-An Yang; Suzanne Lindsay
• Keywords: Influenza-like illness; Popularity; Disease Surveillance

Existing influenza surveillance in the United States is focused on collection of data from sentinel physicians and hospitals; however, compilation distribution reports are usually delayed by up to 2 weeks. With popularity social media growing, Internet a source for syndromic due availability large amounts data. In this study, tweets, or posts 140 characters less, website Twitter were collected analyzed their potential as seasonal influenza.There three aims: (1) improve correlation tweets sentinel-provided influenza-like illness (ILI) rates city through filtering machine-learning classifier, (2) observe correlations emergency department ILI city, (3) explore laboratory-confirmed cases San Diego.Tweets containing keyword "flu" within 17-mile radius 11 US cities selected population At end period, 159,802 used analyses with reported corresponding county health department. Two separate methods between rates: type (non-retweets, retweets, URL, without URL), use classifier that determined whether tweet was "valid", user who likely ill flu.Correlations varied but general trends observed. Non-retweets URL had higher more significant (P<.05) than retweets URL. Correlations observed most cities. The yielded highest many when using well number Diego. High values (r=.93) significance at P<.001 categories be valid classifier.Compared previous season, study demonstrated increased accuracy supplementary tool better classification 2013-2014 season those found where correlated rates. Further investigations field would require expansion regard location from,