Download files
Abstract
Epidemiological models are used to inform health policy on issues such as target vaccination levels, comparing quarantine options and estimating the eventual size of an epidemic. Models that incorporate some elements of the social network structure are used for diseases where close contact is required for transmission.
The motivation of this research is to extend epidemic models to include the relationship with a broader set of relevant real world network properties. The impact of degree distribution by itself is reasonably well understood, but studies with assortativity or clustering are limited and none examine their interaction.
To evaluate the impact of these properties, I simulate epidemics on networks with a range of property values. However, a suitable algorithm to generate the networks is not available in the literature. There are thus two research aspects: generating networks with relevant properties, and estimating the impact of social network structure on epidemic behaviour.
Firstly, I introduce a flexible network generation algorithm that can independently control degree distribution, clustering coefficient and degree assortativity. Results show that the algorithm is able to generate networks with properties that are close to those targeted.
Secondly, I fit models that account for the relationship between network properties and epidemic behaviour. Using results from a large number of epidemic simulations over networks with a range of properties, regression models are fitted to estimate the separate and joint effect of the identified social network properties on the probability of an epidemic occurring and the basic reproduction ratio. The latter is a key epidemic parameter that represents the number of people infected by a typical initial infected person in a population.
Results show that social network properties have a significant influence on epidemic behaviour within the property space investigated. Ignoring the differences between social networks can lead to substantial errors when estimating the basic reproduction ratio from an epidemic and then applying the estimate to a different social network. In turn, these errors could lead to failure in public health programs that rely on such estimates.