Flu in the College Social Network
Link: http://www.sciencemag.org/news/2010/09/social-network-predicts-flu-spread
Author Carrie Arnold discusses the findings of a 2010 study conducted by social scientists Nicholas Christakis of Harvard University and James Fowler of the University of California, San Diego, on social network sensors during the 2009 H1N1 swine flu epidemic. Based on their understanding of the “friendship paradox,” the idea that an individual’s friends are more popular, have more friends, and are more connected to the central network than the individual, Christakis and Fowler were able to predict the flu outbreak in the general Harvard College population as soon as two weeks in advance. The researchers randomly selected 319 Harvard undergraduates (the randomly selected control group) who named 425 of their friends (the friend group). From September 1, 2009 to December 31, 2009, Christakis and Fowler collected data on the health of these 744 students via a biweekly email survey that asked if the students had flulike symptoms and the students’ records at the campus health clinic. They found that the “better connected” friend group showed signs of the flu between 14-69 days before the outbreak of the flu in the control group of the 319 students. This ability to predict the onset of an infectious disease 14 days in advance exhibits the potential of using social network detectors to prepare for or prevent epidemics.
Despite the small scale of the research, given that its scope included students of the same college, the results of the study serve as a good example of the branching (Galton-Watson) process discussed in class. The friends of the 319 undergrads in the randomized control group were “more connected” to the central network, meaning that they were nodes with a large k and connected to many other nodes with a large k, where k = number of contacts. Given that the basic reproductive number Ro = pk, where p = the probability of transmission, if there is a large k, then Ro increases, and so the disease persists longer in the network, which allows more time for the “more connected” individual (i.e. anyone in the friend group of 425 students) to get sick. This idea aligns with the results of the study. Although p was deemed an independent probability in lecture, we could assume that the p would be inherently high for college students because they live in close quarters and may use the same silverware in the dining halls. A higher probability for transmission then could also drive up the Ro.