#Outbreak: An Exploration of Twitter metadata as a means to supplement influenza surveillance in Canada during the 2013-2014 influenza season

Abstract

This study explored the utility of Twitter metadata as it relates to influenza surveillance in Canada. Twitter metadata posted between July 2013 and August 2014 containing influenza-related keywords (e.g. influenza, flu, cough) was analyzed using a variety of methodologies. Predictive regression models demonstrated differential utility of specific keywords; Tweets containing several keywords were strongly associated with influenza activity (flu, influenza, grippe), whereas a weaker association was observed with Tweets containing other keywords (e.g. cough, fever). Correlation analysis demonstrated that non-retweets and Tweets that did not contain a URL link were better correlated with influenza cases than retweets and Tweets containing a URL link, respectively. Geospatial cluster analysis showed that Twitter metadata could be used to identify local clusters of influenza-related Twitter chatter; clusters matched traditional surveillance reports in both space and time. Geospatial cluster analysis also identified clusters in areas not reported by the national Fluwatch program.