The media coverage of the Swine flu and its elevation to Phase 4 status by the WHO has caused a lot of discussion across print, TV and web chatrooms. It also renews interest in predictive tools such as Google's Flu trends project. The fascinating thing about Google's flu prediction system is its apparent accuracy and ability to predict flu activity one to two weeks ahead of CDC reports. This ability is clearly critical with potential pandemics.

It seemed inevitable that services mining keywords associated with Twitter status messages would eventually emerge around the health space. Two services, FluTweet and Sickcity.org, have developed real-time tracking services to show flu activity based on Twitter messages. FluTweet updates its data every hour and tracks messages globally. Sickcity.org is attempting to track data on the local, city level. The goals of both are admirable but a quick review of Twitter messages on Swine flu and flu in general reveals how daunting the task can be. The conversational nature of the messages on Twitter means there is a lot more static to muddle through. Here are a couple of results when I search on flu:

trish_thedish: Sanjay Gupta says don't worry about swine flu, but when an obnoxious, dirty little 5 y/o projectile coughs in line at Target, I worry.

ROCKINTHAT: Swine flu, Remember what your Mother taught you. Cover your cough, Wash your hands, and if you have a fever stay at home!!!!!

How do you build that filter into an algorithm (ie, if you are tracking mentions of coughs and flu)? With 19 million users, Twitter now has thousands (perhaps millions) of conversations going on at the same time about Swine flu. The key to the success of these services is lots of historical data and the ability to continuously refine the data model.

The issue is what happens next -- Does awareness itself have a public health value? A study published in Proceedings of the National Academy of Sciences in the United States of America (PNAS) found that awareness does have a positive effect on containing a pandemic. The study was based on data on the spread of the SARS epidemic in China culled from millions of text messages sent to and from residents of Guangzhou where the epidemic originated. According to PNAS, "The authors linked a mathematical model for the spread of disease awareness to one for the spread of disease, and found that increased disease awareness resulted in a smaller outbreak within a population." A similar study is probably impossible to conduct here in the US due to privacy concerns, so the best analog we have is publicly-available Twitter and Facebook messages.

The CDC has set up several Twitter accounts, including @CDCemergency which provides accurate information. Twitter has been criticized during the last few days as a source of misinformation but it is also bringing good information to those looking for it. If the results of the PNAS study can be extended to the U.S., increased awareness (stay home, wash your hands, etc.) and most importantly, understanding when to go to an emergency room or a doctor, is a net positive.