Google's Flu Trends

A facebook friend made some comments about the flu epidemic (H1N1 or otherwise) starting to wind down, and referenced Google Flu Trends, which I was unfamiliar with. Taking a quick look, one is struck by the high correlation between the Google estimate of flu activity, as measured by search activity on flu related terms, and the Center for Disease Control (CDC) flu activity reports.

However, on reading the paper (also published in Nature February, 2009), the correlation is not surprising, since the researchers selected search terms out of tens of millions that exhibited the best correlation with the CDC data. So I'll grant their statement that Flu Trends can predict the CDC measurement rapidly (<24 hours), where the CDC process lags by a couple weeks collecting the data.  Anyway, the correlation is inevitable from Google's search term selection process to fit the data.

What is Flu Trends actually measuring? Certainly not disease, or even symptoms of a disease. It is measuring public interest in a disease, or its symptoms, or whatever other search terms that Google has selected (and won't divulge). So at best, Flu Trends is measuring the spread or intensity of a meme, not a biological virus. This leads one (e.g., me, the skeptic) to wonder what the CDC is actually measuring in their index. I haven't researched this, and couldn't immediately find details about the CDC process, but the CDC surveillance seems to include multiple factors, such as or perhaps even primarily influenza-like illness (ILI) doctor visits, independent of clinical test or diagnosis.

One wonders, then, if both measurements are more driven by the public's level of exposure to INFORMATION (TV, radio, newspaper, web pages) about the disease than to the actual disease itself.

Just wondering....