Online behavioral patterns in a health crisis setting: the 2009 pandemic.

• Author(s): Vieira, Cláudio
• Date: 2017
• Persistent ID: http://hdl.handle.net/10400.7/841
• Origin: ARCA - Access to Research and Communication Annals
• Subject(s): data mining; online behavior; public health; pandemic; influenza
• Document Type: Doctoral thesis
• Language: English
• Contributor(s): ARCA

Description

Seasonal flu places a heavy burden on both human populations and health care ser- vices every year, warranting permanent surveillance. Online-based surveillance mod- els harness the collective online search activity of flu-infected individuals to provide real-time monitoring of flu activity. These models assume that most flu-related online behavior is motivated by a flu infection. However, when the flu pandemic emerged in 2009 it resulted in abnormal search behaviors that confounded these models, as several reasons, beyond infection, can motivate individuals to seek flu information. In practice, and despite their potential, current models cannot distinguish whether such activity is related with actual flu infection or not, rendering them useless, at least in pandemic settings. If the different motives that prompt flu-related searches can be pinpointed, then this information can be used to train the models to recognize what is infection-motivated and what is not. Moreover, if online behaviors reflect real-life behaviors, then valuable public health insights can be extracted by analyzing the public’s online response to a pandemic. To test these assumptions, we collected flu-related online search trends regarding the pandemic period. We estimated real-life behaviors, anxiety and risk perception, through data obtained from surveys conducted during the pandemic. As possible explanatory variables of online search trends, we collected flu-related media coverage as well as laboratory-confirmed flu cases. We found that a specific set of search trends was more associated with media activity, whereas another set of search trends was more associated with flu infections. The media-related search trends proxied the public’s anxiety levels and the infection- related search trends proxied the public’s risk perception. Having determined which factors correlated with specific search trends, and what real-life behaviors might have corresponded to these search trends, our findings place online sources as suitable tools for monitoring the public’s response to a flu pandemic. Our findings additionally support the possibility of separating search trends that are more sensitive to media activity and search trends that are more sensitive to flu activity. Thus, we provide proof-of-principle that it should be possible to infer human behaviour from online behaviour and, in practical terms, our system is flexible and general enough to be applied both to pandemic and seasonal flu, as well as to other infectious settings.