Google Flu Trends: wrong prediction or wrong expectations?

Google Flu Trends (GFT) is a web service launched by Google in 2008, with the aim of estimating influenza activity by aggregating search terms to draw a real-time picture of both the spread of the flu and its seriousness. Google even claimed that their predictions were 97% accurate comparing with those from the Centers for Disease Control and Prevention (CDC), which are based on surveillance reports from laboratories across the United States.

Unfortunately, it turns out that such a promising tool to predict influenza outbreak could have been not very accurate, especially between 2011 and 2013.

In February 2013, Nature reported that the Google’s predictive model overestimated peak flu levels, particularly at Christmas 2012, when the predicted peak almost doubled that from CDC. It was not the first slither for GFT, since in 2009 it underestimated the swine flu – a mistake attributed to differences in people’s way of searching on the web, due to the exceptional event of the pandemic. More recently, a paper published on Science showed that GTF’s predictions have been too high for 100 out of the last 108 weeks. Which means since August 2011.

However, it is also important to remind that GFT was not meant to replace traditional surveillance networks. Also, their data can be extremely useful when combined with other near–real time health data, like those from CDC, as pointed out by the authors of the Science’s paper, a team of scientists from Northeastern University (Boston), Harvard University (Cambridge) and the Institute for Scientific Interchange Foundation (Turin).