Reinventing Discovery - Michael Nielsen [47]
Hoping to speed up the CDC’s system, the Google and CDC scientists wondered if search queries entered by users into Google’s search engine could be used to instantaneously track where the flu is occurring. The idea is that if there’s a surge of people in the city of Atlanta searching for (say) “cough medicine,” chances are there’s been an increase of flu in Atlanta. To get good results, the Google and CDC scientists took the CDC’s historical flu data from 2003 to early 2007, and looked for correlations with common Google search queries. They found 45 search queries that were especially well correlated with the historical flu data. Using those queries they built a model that they hoped could be used to instantly figure out where the flu is occurring, just by monitoring Google searches. They then tested that model by comparing it with a new set of data, the CDC data from the 2007–08 flu season. Their model gave nearly perfect (97%) agreement! In other words, Google’s search queries can be used to determine where flu outbreaks are happening, and how large they are, but without the time lag suffered by the CDC. What’s more, Google search queries can be used to track influenza not only in the United States, but anywhere large numbers of people are using Google, including places where there is no CDC-like organization tracking disease. Google has built a website called Google Flu Trends that uses search queries to track influenza in 29 countries.
The Google Flu Trends results require a couple of caveats. First, many doctors in the United States now use electronic medical record-keeping systems, and the CDC has recently partnered with the makers of one of those systems, General Electric, to develop a new tracking system that should give it a near real-time ability to track reports of influenza from 14 million patients. It’s possible and perhaps likely that the CDC’s new system will obsolete Google Flu Trends, at least in the United States. Second, the CDC data used to build the Google-CDC system did not, strictly speaking, track influenza. Rather, it tracked “influenza-like” illnesses from reports of symptoms such as cough and sore throat that are often associated with the flu. Other conditions such as colds can produce similar symptoms. A follow-up study done in 2010 confirmed that, not surprisingly, Google Flu Trends is significantly better at tracking influenza-like illnesses than it is at tracking actual laboratory-confirmed cases of influenza. It’s a helpful diagnostic tool, not a perfect way of tracking the flu.
Using Google to predict the flu is interesting, but even more interesting are the other possibilities it suggests, possibilities that go beyond medicine and into every aspect of life. Follow-up research has already shown that search queries can be used to predict trends in unemployment and in housing prices, and even to predict how well songs will do on the music charts. What else might be possible? Could Google figure out which search queries predict changes in the stock price of some company, say, Microsoft? What about the behavior of the Dow Jones Industrial Average? Or which technology startup is the best target for acquisition? Or the outcome of the next US presidential election? Or a coup d’état in an unstable country? Suppose Google was tracking the searches of law clerks working at the US Supreme Court—might it be possible to predict court decisions? Or perhaps to figure out what concerns an individual justice has while a case is being heard? Suppose a Google user is making searches that