Age-specific representations of social contact networks using egocentric survey data

点击翻译按钮获取中文摘要

When modelling the spread of a disease outbreak epidemiologists often use the assumption that the population of interest is 'well-mixed'. Under this assumption, all individuals interact at the same rate, providing a powerful mechanism to predict disease prevalence at a societal level. However, in many cases there exists groups of individuals ('superspreaders'), who infect disproportionately many more people than the mean. These heterogeneities in the secondary case distribution are pervasive in different pathogens and settings, with widespread use of the 80/20 rule throughout epidemiological literature (80% of cases are attributable to 20% of infected individuals). The factors governing the identification of superspreaders are not well understood, and could be behavioural, biological or a mixture of both. A widely used approach to incorporate heterogeneity of secondary case distribution in the modelling process is to induce disease transmission on a contact network. This project will implement an individual-based network approach, to better understand behavioural heterogeneities. In particular, contact data will be inferred from the CoMix study, and possibly other pre-COVID surveys such as POLYMOD, the BBC pandemic survey [https://doi.org/10.1016/j.epidem.2018.03.003] and the Warwick contact survey. These data sets provide a rich body of information on egocentric contact structures in the UK (e.g. the number and type of contacts from a respondent), however, a standardised mechanism for constructing the complete population level contact network from this type of data does not exist. There are many competing approaches to create a representative contact network from this egocentric data, with many failing to provide the clustering or assortative mixing known to be present in human social networks. One further difficulty arises from the intractability of true link weights, which emerge from some interactions being much more likely to spread disease, due to duration of interaction, the closeness of the interaction or other factors. This project will address the importance of duration, physicality and setting in which a contact is made on the network link structure and ultimately epidemicdynamics. It will bridge the gap between the power law degree distributions of networks commonly reported in society and the negative binomial secondary case distributions described for a wide range of pathogens.This can be achieved by incorporating link weights based on each contacts characteristics and fitting these using an epidemic model to known outbreak secondary case distributions. Danon et al. proposed methods to translate connection properties to link weights, which we can explore in the case of the COVID-19 outbreak and the novel mixing structures these lockdowns caused. The project will further consider the implications of network clustering (often observed as cliques of highly interconnected individuals). Throughout, I will use a mixture of simple analytical approaches and simulation models to gain a greater understanding; both will be informed by or fitted to the wealth of observations that are being collected.