Google Flu Trends
In a video that I saw last year on TED talks, professor Nicholas A. Christakis (Harvard professor of Medical Sociology) discussed how studying network structures could allow us to provide better warning on epidemics (one of the students wrote a nice blog on this video here). On this video, Prof. Christakis briefly mentions Google Flu Trends, a small Google project working on making health predictions (the flu in this case) based on web searches on flu related keywords.
The main idea behind this project is that there is a direct correlation between the number of people affected by the flu and the number of people who actually have the disease. The reasoning behind this is simple: the more there are people that are sick, the more likely they are to perform a search on Google about that sickness (how to cure it, best treatments, medicine side-effects, etc.). If we can keep track of these searches, then we could try and see whether there is a direct link between the two.
Today, the main U.S. organization that provides data about epidemic outbreaks is the CDC (Centers for Disease Control and Prevention). What they do is gather data from clinics and physicians all around the United States and try to make a prediction about the current number of epidemic outbreaks. Although these predictions have been more or less accurate so far, they usually don’t show up until after the epidemic is well underway. In this case, the CDC relies on the information gathered from a knowledgeable “network” of doctors in order to make their predictions. Although this network is very valuable in terms of the content of the information, it is a “slow” network.
Google Flu Trends takes another approach by monitoring a broad network (the network consisting of all searched words on Google) that is not specific to the medical field, but that contains an incredible amount of data and is fast to acquire. Google thus trades scarce but valuable knowledge from doctors with abundant (but not very accurate) and fast data from users searching for flu-related words.
The results of this project have been extremely accurate since it started, and has proven to warn of the flu epidemic up to two weeks before the CDC can release its reports. If we can detect the flu outbreak at its earlier stages, then we can attempt to utilize the adequate measures to prevent it, or try to minimize its spread.
Another advantage of this project is that since the Internet itself is such a vast network of information and is not limited by location, Google Flu Trends can look for epidemic outbreaks in various countries and compare how the disease might spread across the network simply by keeping track of the search terms.
It is interesting in seeing how much information one can extract from something as simple as search words. But it is exactly this type of network analysis that will allow for epidemic prevention methods to become better and better.