Following Accuracy Problems, Google Revamps Flu Trends

 image

In a recent blog post, Google announced major changes to its Flu Tracking program that will take effect as the 2014/2015 flu season approaches. Originally billed as a shining example of big data technology, Google Flu Trends earned its initial accolades during the 2009 H1N1 outbreak, when it forecasted a spike in flu a full two weeks before the CDC was able to confirm the outbreak. At the time, big data analytics was being touted as the inevitable replacement to a myriad of manual reporting processes in place in the public health space, though Google themselves repeatedly clarified that this was not the intention of its Flu Trends project.

The glory days of Google Flu Trend began to unwind after researchers noticed that, when compared to CDC data, Google seems to be consistently overestimating total annual flu counts. The CDC totals come from physicians who report confirmed flu cases within their patient populations. Google uses data from its search engines, measuring increases in a secret list of search terms, like fever and nauseous, that it says strongly correlates with the flu. Both the CDC and Google are able to use location data to then follow flu as it spreads geographically. Because the CDC has to wait for a patient to go to a doctors office and receive a confirmed diagnosis, its data lags behind the real-time spread of flu by a few weeks. Google is able to improve on this because its program rely on real-time search data. However, because Google uses search terms as a barometer, rather than confirmed diagnosis, its forecasted outbreak totals have missed the mark by wide margins in the past. Many speculated that Google’s data was vulnerable to spikes in flu-related searches triggered by other events, like an increase in news coverage on flu. Researchers investigated and this summer published a report which concluded that Google’s data was in fact less accurate than the CDC’s. In that study, researchers suggested that an ideal tracking system would be one that combines Google’s faster search term-based analysis with the CDC’s more accurate physician reporting data.

image

Google responded to the research Friday, confirming that for the 2014/2015 flu season, it would be including the CDC’s flu data in its program in an effort to boost accuracy. The result, it hopes, will be a flu tracking tool that is able to track the geographic spread of flu faster than the CDC, while also forecasting the prevalence of flu with a higher degree of accuracy that it has been able to do thus far.

The changes will only go into effect for the US version of Google’s Flu Tracker, but the company noted in its blog post that if the new platform performs measurably better than its older version, the company would look to incorporate public health data from other countries to improve their accuracy. Currently, Google Flu Trends tracks influenza in 29 countries, and tracks Dengue in 10.


Enjoy HIStalk Connect? Sign up for update alerts, or follow us at @HIStalkConnect.

↑ Back to top

Founding Sponsors

Platinum Sponsors