Trending on Google

Every day about 400 millions searches are done using the Google search engine. Many of these searches are looking for health information and for every one the search terms, location of the computer and the time of search are stored. This data can be explored freely using Google Trends but can these trends tell us anything useful about disease patterns and do they have any potential to improve healthcare?

---

Search volumes for "flower delivery" in the UK have two sharp peaks - Valentine's Day and Mothers' Day, no surprises! Flu-related searches are more common in the winter and more allergy-related searches are made in the summer. There is a trend in the number of searches for "swollen ankles" showing an association between this symptom and climate. In the UK an annual peak occurs in July/August whereas in Australia it peaks in December/January (see Figure below). It is also interesting that 2006 was the hottest July on record and this was associated with the peak search volume of all recorded.  Google Trends also lets you analyse trends over time to look for temporal variation. The number of searches for "weekend breaks" have declined year on year since 2004 whereas the number of searches for "payday loans" is rising - practical results of the economic downturn and recession. These might be interesting observations but can we use trends in searches for to draw conclusions about health trends and predict disease epidemiology in realtime.

Recent studies have shown a close relationship between the number searches for flu-related topics and how many people actually have flu symptoms. Although not everyone searching for flu has the condition, when all the millions of search queries are added together trends begin to emerge especially when the volume of data is so enormous.  Analysis of this "Big Data" compared with traditional flu surveillance systems shows an excellent correlation and the Google Flu Trends data is collected continuously and accurately estimates the level of weekly influenza activity in each region of the United States with a reporting lag of about one day. This gives the potential for an online worldwide realtime disease epidemiology with no complex data collection mechanisms. At present only a few diseases have been studied using this methodology. These include influenza, Dengue, Lyme disease and Tuberculosis. It is likely however that other trends could emerge in the future. The world of "Big Data" has only just started and the possiblities for prediction in other domains have already emerged. In financial markets there is evidence that weekly transaction volumes of Standard & Poor 500 companies are correlated with weekly search volume of corresponding company names.  This doesn't allow you to pick winning stocks but it might help understand factors leading to financial crashes.