On Benefits and Limitations of Social Media for Early Outbreak Detection

• Activity Number: 580
• Type: Invited
• Date/Time: Thursday, August 7, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Elena Naumova*+ and Marwah Soliman
• Companies: Tufts University and University of Texas at Dallas
• Keywords: forecasting ; social media ; epidemiology ; surveillance ; computational statistics ; outbreak detection

Abstract

While recent years have seen substantial progress in improving surveillance of infectious diseases, a few major challenges hindering operational, real-time forecast of infectious diseases still exist. Indeed, many currently employed forecast models: a) lack flexibility, due to overly restrictive assumptions; b) suffer from misspecifications or inclusion of data elements not universally available in real time; and c) have technically demanding computational requirements. Most importantly, these models do not offer observations for near real-time forecasting, which substantially reduces practical utility of these prediction models. In this talk we discuss how social media input can be incorporated to a time series model for operational tracking and out-of-sample real-time prediction of influenza in a city of approximately 1.5 million population and verify our results against rates of hospitalizations and laboratory-confirmed cases of flu. We demonstrate benefits and limitations of data from social media to operationalize early outbreak detection.