Google Creates Plague Prediction System

Illustration for article titled Google Creates Plague Prediction System

Google doesn't just help you find the nearest pizza place anymore - now it helps doctors find where flu epidemics are going to strike next. Working with a group of epidemiologists, the tech megacorp has revealed a new system for tracking disease outbreaks by checking what people are searching for. By tracking the rise in searches on phrases like "cold/flu remedy," Google said yesterday in a Nature article that it can predict with almost total accuracy where flu outbreaks are occurring, far more quickly than the American Centers for Disease Control can.

Illustration for article titled Google Creates Plague Prediction System

To make its predictions accurate, researchers first looked over the data on where influenza-like illness (ILI) outbreaks took place in the U.S. over the past five years. Then they pored over search data, looking for key phrases that popped up again and again when the outbreaks occurred. You can see to the left a chart of some of the searches that correlated most strongly to a spike in ILI cases. Note that the terms listed in the chart are not exact search terms (nobody actually searches on the phrase "influenza complication"), but instead capture the idea of what people searched on. So "influenza complication" would mean people searched on things like "runny nose" or "aches and chills." After analyzing the data, researchers came up with a system where they weighted certain search topics for how much they seemed connected to ILI outbreaks. In the chart, you can see searches on flu symptoms rank high for a correlation with hospital data on lots of flu patients. Searches on specific types of antibiotics might be connected to flu, but weren't weighted as heavily. Eventually, they began using the list of weighted search topics against real-time data coming in during the early part of 2008, trying to see if a rise in certain topics in a given region meant that flu was on the rise there. And as you can see in the chart above, the Google prediction (in black) was uncannily similar to the actual rise in flu cases in the region they picked (in red). So what's the advantage of the Google system of disease surveillance? Traditional methods of disease tracking rely on data from hospitals and pharmacies, and often it takes a week or more to figure out any pattern. The Google system takes about a day. If epidemiologists see a huge spike in data that points to a massive disease outbreak, they can begin to prepare for it. And more importantly, they might be able to track where the disease is heading. What this means is that the Centers for Disease control will probably hear about the next big pandemic from Google, rather than hospital researchers. It also means that when you do a search on Google, you're revealing a lot more information about yourself than you probably realize. Detecting Influenza Outbreaks Using Search Query Data [via Nature]


Franklin Harris

Snark all you want, but this is a textbook case — like futures markets — of finding a way to access dispersed information that would otherwise go unnoticed. And I predict it will do a better job of predicting epidemics than can any number of "experts" cloistered away in the offices of the CDC.