Google said its system detected increases in flu-related queries in Mexico City and several other Mexican states during the April 19-25 time period.
At the same time, Google cautions that Experimental Flu Trends for Mexico is, as the name suggests, an experiment and that the data should be viewed as such.
"While we would prefer to validate this data and improve its accuracy, we decided to release an early version today so that it might help public health officials and concerned individuals get an up-to-date picture of the ongoing swine flu outbreak," explained Google engineers Jeremy Ginsberg and Matt Mohebbi in a blog post. "As with our existing Flu Trends system, estimates are provided across many of Mexico's states and updated every day."
Ginsberg and Mohebbi said flu activity in the United States remains low, as might be expected given the relatively few cases reported in the country. But Google plans to continue its monitoring to look for rising infection.
Google introduced Google Flu Trends in November as a way to visualize the correlation between flu infections and flu-related search activity. Google maintains that searches provide an early warning about the spread of the flu because search data can be gathered and analyzed almost instantly, unlike traditional epidemiological reporting methods.
Ginsberg and Mohebbi explain that Google's "model tries to filter out search queries that are more likely associated with topical searches rather than searches by those who may be experiencing symptoms." Thus the spike shown in the graph on the Experimental Flu Trends for Mexico Web page should correlate with actual flu activity rather than searches prompted by worry.
As of 11 a.m. EDT Wednesday, data from the Centers for Disease Control and Prevention indicated that 91 confirmed swine flu cases have been detected in the United States and that one person -- a child in Texas -- has died as a result of the outbreak.
Google planned to hold a media call at 11 a.m. PDT Wednesday to field questions about its flu-tracking project.
InformationWeek has published an in-depth report on predictive analysis. Download the report here (registration required).