Last month, Levitt blogged about how Google Flu Trends, which measures flu queries in real-time, can serve as an early-warning system for flu outbreaks. The catch: it only works if Google’s data analysts are paying attention to the data — which they weren’t last week, when Flu Trends showed a bump in flu-related search terms right before the swine flu outbreak. One biosurveillance company claims we should have listened to Twitter and bloggers instead. [%comments]
When You Don’t Watch Your Data
TAGS:
and 
. You can also watch it on 
The problem with trying to expand Flu Trends beyond the US is that there is a lack of accurate real world data to make a post check on the validity of the graph.
I am not aware about Mexico, but unless there is a consistent, accurate, indicator of flu the Flu Trends data cannot be checked. Even if the real world data is a few weeks or months delayed, it is still required.
Isn’t this sort of like Nostradamus where something happens and they find a verse that “predicted” it?
How many spikes occur that do not correlate with an outbreak?
If one looks at a set of Google searches and Twitter tweets for the same term, it seems that the tweets have more signal (because there’s more context) and more noise (because of the retweeting and advertising).
rishi –
I think the point of checking the Flu Trends is to then go looking for real data. Even a day or two earlier than standard medical reporting would be helpful.
But they do have to watch for the Spanish “gripe” instead of “flu”, especially coming from .mx domains — vocabulary matters.
Google has an experimental Flu Trends for Mexico, and the uptick in April is surprisingly small.
The problem with Flu Trends, is it is an indirect measure of the number of people with the flu. The swine flu hasn’t infected that many people compared to the regular flu. It’s just unusually deadly. If the next flu that comes along spreads quickly, I expect Flu Trends will detect it very quickly.
It does seem like something is going around right now, based on what I see around me – but I’m presuming it’s just the regular flu.
I played around with Google Search Insights on Monday 27 April. As of then, there wasn’t any evidence of an uptick prior to the official announcements; however, the massive spike in searches over the weekend of 25-26 knocked out any variance in search terms prior to that weekend. The system ought to allow date ranges that cull the most recent dates.
http://offsettingbehaviour.blogspot.com/2009/04/nowcasting-flu.html
I’m with communications at Google and I wanted to clarify that Experimental Flu Trends for Mexico didn’t exist at the time of the initial outbreak of H1N1 in Mexico. Google Flu Trends for the U.S. was published in November and Google had not yet expanded it to other countries. After health officials asked the Flu Trends team if they were detecting increased flu activity in Mexico, our engineers quickly developed an experimental system for Mexico and shared preliminary data with health officials. It is a work in progress but given the level of concern and the importance of additional sources of information, we published the site on 4/29.