Disease Models: Determining the Network
Network theory provides a number of different ways to model disease spread in a network. There is the SIRS model, the SIR model, the SI model, and branching tree model, and many more complex models. However, all of these models assume you know the network structure of the population. In a big city, it is infeasible to simply ask people who they see regularly as you might be able to do in a small disease transmission experiment. Thus, given a big city and an outbreak, how do you create the network in order to model the spread?
In New York City, the New York City Department of Health and Mental Hygiene uses a computer model to determine the network structure. It plots outbreaks on a map of the city, assuming that people who get sick and live near each other are likely to have been in contact. In general, the system does very well, but one problem that the article notes is that a group of people may all get a strange disease not because of a local outbreak, but because they traveled together, got sick, and then came home. Furthermore, just based on this model, a disease wouldn’t be able to jump from one side of the city to another. However, a person may live on one side of a city, and work on the other side. Thus, if he gets infected, he might transmit the disease to his colleagues, bringing the infection to a different area of the city.
Given the limited amount of public information, how could an automated system do better? This simple geographical model lacks bridges in the network. However, there is a ton of information online: a computer program could scan Facebook to determine where a person’s closest friends live, and add some edges into the disease network to reflect them. A program could also scan LinkedIn, grouping people together by workplace, and add edges into the disease network to reflect that. To get even more reliable data, the program could use data from past outbreaks to determine where in the networks bridges are likely to exist. For example, if during the outbreak of a disease the West and South portions of a city get infected, but not the North and East portions, the program could add extra edges in the graph between the West and South areas because there must be some link which allowed the disease to spread that quickly. With these simple additions, the model would be able to predict spread of the disease from one area to another, and also better able to predict the impact of releasing a disease in a specific area as compared to a simple geographic model.
Source: http://fivethirtyeight.com/features/how-new-york-hunts-for-early-signs-of-disease-outbreaks/