Predicting Epidemics Using Social Networks
In 2010, Nicholas Christakis gave a fascinating TED talk about how we might use social networks predict epidemics before they happen. He compares this hypothetical advanced warning to the current system of epidemic information gathering used by the CDC. By compiling data from doctors and reporters in various regions about how many patients are sick, the CDC accurately determine what epidemic happened two weeks in the past. Christakis postulates that by monitoring certain people within social networks we can know about outbreaks with enough time to get ahead of them.
He claims that the reporters should be nodes close to the “center” of the network. That is to say, nodes with a higher than average number of connections. Using the friendship paradox it is actually easy to get these “central” nodes fairly reliably. It works like this. If you select a number of random nodes in a network and then tell them to nominate one of their connections. The group of nominated nodes will actually be more central in the network than the original group of randomly selected ones. He tested this approach for the first time at Harvard during the outbreak of H1N1 in 2009. What he found was exactly as he expected. By monitoring the friends, he was able to observe the peak of the epidemic of H1N1 16 days before the randomly selected group. When he looked at when the two groups first diverged as opposed to the differences in peaks, that prediction extents to 46 days before the peak of the epidemic.
These results are compelling and Christakis provides other examples of how this technique may be used. One of the most exciting examples has to do with herd immunity. Usually, 96% of the population needs to be inoculated before the entire population would be immune. However, if you immunize 30% of the nodes while insuring that all nodes were immunized (using the friendship paradox) then the population is as immune as if 96% had been immunized randomly. This means that herd immunity can be achieved at a much higher efficiency.
In the era of what Christakis calls “massive, passive, data collection” this technique can be applied to situations even outside of pathogens. By monitoring phone speeds on a highway you can predict traffic jams. By monitoring prescriptions you can measure how fast a new drug is adopted. By collecting only from central individuals, you can monitor all types of “epidemics” at low cost to the observer and low impact on the observed.