Every year, millions of people become infected with one of the many seasonal influenza viruses. These infections may have dire consequences as local hospital Emergency Rooms (ERs) experience sudden surges of influenza patients, causing ambulance diversions and shortages of medical supplies. Current influenza surveillance techniques lack the necessary spatial and temporal fidelity to benefit local hospital systems. This dissertation helps correct that issue through three chapters. Chapter one identifies an approach to augment social media data using the Digital Interaction Program (DIP). DIP uses application program interfaces to digitally converse with and seek social media users' participation in an online questionnaire. This questionnaire is designed to collect spatial and temporal data and augment social media data, such as demographic information. Chapter two uses DIP to identify where and when influenza tweets posted across New York City and London at fine spatial and temporal scales. It was found that on average influenza tweets tend to occur closer to a user's home ZIP Code, in comparison to those users' non-influenza tweets. Therefore, this information suggests that influenza tweets can predict influenza cases at a finer geographic scale than current research suggests. Influenza tweets are most often posted when a user is experiencing peak symptoms, not symptom onset. Finally, Chapter three of this research tests if, when, and to what degree influenza tweets can predict local hospital ER admissions. It was found that most hospitals can use influenza tweets to predict influenza ER admissions on average of about eight days advanced. Chapter three speculates that influenza tweets have the potential to identify influenza propagation between the different age groups in New York City. Therefore, Twitter has the spatial and temporal potential to provide a more timely and spatially accurate influenza surveillance system that is focused on local hospital systems.
Read
