Influenza: Improving flu predictions using “big data”

Predictions of influenza spread can be improved by integrating “big data”, in the form of Google Flu Trends, into traditional surveillance methods, a study in Scientific Reports proposes. Combining the two systems can predict US influenza infections one week into the future, the study demonstrates. The findings may have implications for the prevention and control of influenza outbreaks at a local and national level.

Seasonal influenza infects approximately 5-20% of the US population every year, resulting in over 200,000 hospitalizations. Accurate assessments of infection levels and predicting which regions have higher infection risk can assist targeted prevention and treatment efforts. Google Flu Trends uses search query data to estimate influenza activity in real-time - two weeks earlier than traditional surveillance, which gathers data on the number of potential and confirmed influenza cases and classifies the influenza viruses collected from patients.

Although Google Flu Trends has not consistently predicted real-time trends, using it in conjunction with established monitoring systems can produce better estimates of actual cases of influenza, Michael Davidson and colleagues report. They attribute these improvements to the use of methods borrowed from social network analysis, which they use to construct networks of connected geographic regions affected by influenza that allow for better predictions of future spread. The new model benefits from both the accuracy of traditional data collection and the real-time predictions generated by Google Flu Trends.