Researching with Google Flu Trends
May 18th, 2010 | Published in Google.org
When creating Google Flu Trends, we decided early on to make all of our estimates of flu activity available to the public for download. We’re always excited to hear about researchers using this data and recently learned about a study which compares Google Flu Trends estimates of flu activity with estimates from the Centers for Disease Control and Prevention (CDC).
Almost any public health expert will tell you that there isn’t a single perfect measure of flu activity. Instead, there are several good measures, each of which tells a slightly different story. A good picture of flu activity emerges when these stories are examined together. One such measure is the CDC’s influenza-like illness surveillance network (ILINet) which shows the proportion of people who visit a physician with flu-like symptoms such as fever, cough and sore throat. This network can highlight areas where activity is unusually high given the time of year but cannot provide insight into the specific pathogens which are causing the illness. Another is the CDC’s Virologic Surveillance which shows the proportion of people who visit a physician, get tested, and have lab-confirmed influenza. This network shows the specific strains of influenza in circulation but can be affected by changes in testing practices, making it difficult to interpret the data over long periods of time.
We collaborated with the CDC when designing Google Flu Trends and decided to build a system that would make estimates based on the CDC’s ILINet data. Historically, we’ve shown a strong correlation with CDC’s ILINet data, typically with a correlation of greater than 0.9 (or more than 90% accurate as compared with CDC’s ILINet data). This recent study shows that Google Flu Trends is 72% accurate as compared with CDC's Virologic Surveillance data. This doesn't come as much of a surprise since the virologic data is telling a different story than the ILInet data. However, it’s good to see research showing the differences in these measures, as it highlights the need for multiple networks to exist.
We provide Google Flu Trends as an additional source of information for helping to understand the flu season. Google Flu Trends helps paint the picture of flu activity by providing estimates daily while some traditional flu surveillance systems can take days or weeks to collect and release data.
Almost any public health expert will tell you that there isn’t a single perfect measure of flu activity. Instead, there are several good measures, each of which tells a slightly different story. A good picture of flu activity emerges when these stories are examined together. One such measure is the CDC’s influenza-like illness surveillance network (ILINet) which shows the proportion of people who visit a physician with flu-like symptoms such as fever, cough and sore throat. This network can highlight areas where activity is unusually high given the time of year but cannot provide insight into the specific pathogens which are causing the illness. Another is the CDC’s Virologic Surveillance which shows the proportion of people who visit a physician, get tested, and have lab-confirmed influenza. This network shows the specific strains of influenza in circulation but can be affected by changes in testing practices, making it difficult to interpret the data over long periods of time.
We collaborated with the CDC when designing Google Flu Trends and decided to build a system that would make estimates based on the CDC’s ILINet data. Historically, we’ve shown a strong correlation with CDC’s ILINet data, typically with a correlation of greater than 0.9 (or more than 90% accurate as compared with CDC’s ILINet data). This recent study shows that Google Flu Trends is 72% accurate as compared with CDC's Virologic Surveillance data. This doesn't come as much of a surprise since the virologic data is telling a different story than the ILInet data. However, it’s good to see research showing the differences in these measures, as it highlights the need for multiple networks to exist.
We provide Google Flu Trends as an additional source of information for helping to understand the flu season. Google Flu Trends helps paint the picture of flu activity by providing estimates daily while some traditional flu surveillance systems can take days or weeks to collect and release data.