Complex network analysis and Exponential Random Graph Models: Epidemiological implications of network structures

Hepatitis E virus (HEV) is a virus infecting pigs and humans. Its spread among pigs occurs within-farm through faecal–oral route including direct and environmental transmission [1] and live animal movements are known as the major driver of between farms spread [2]. Strongly structured in space and time, the swine production chain is made of various interconnected farms, forming a complex network.

In this study, we aim to simulate realistic between-farm movements to feed a multilevel epidemiological model. Swine movements from 2017 to 2019 provided by the National Swine Identification database (BDporc) were analysed using complex network analysis (CNA) methods. Exponential random graph models were then used to identify the key drivers of trade partner choices and the associated probabilities of contact between farms.

In the 36-month period, 2.512.174 loading and unloading records were recorded involving 16.377 premises. Because of the large size of the dataset, specific subsets of the data, using semi-annual data and type of transported animals (breeding sows, piglets (defined as 08kg animals) and growing pigs (defined as 25kg animals) were used for the analysis. For each data subset, about 2400 variable combinations were tested. In all cases, the between farms distance, the company, the size, the indegree related to the type of farming and the export and import frequencies, and only edges as structural covariable were identified as explanatory variables of the network structure. Some alternative factors such as free-ranging characteristics and the types of farms were subnetwork-specific. 

ERGM outcomes allow simulating pig trade networks having similar characteristics as the original one, to prospect on the global impact of the network structure on pathogen transmission. Simulated movements fed a multi-level model developed with SimInf package [3]. First run on French data, the model will be extended to partner countries involved in the European Joint Programme (EJP) funded project BIOPIGEE. The simulated prevalence of HEV positive pigs sent to slaughterhouses, will feed a Quantitative microbiological risk assessment model (QMRA) to evaluate the risk of human exposure to HEV through swine products [4].

[1] M. Andraud, M. Dumarest, R. Cariolet, B. Aylaj, E. Barnaud, F. Eono, N. Pavio & N. Rose. Vet. Res., 44 (2013) 102.
[2] M. Salines, M. Andraud, N. Rose & S. Widgren. PLoS ONE, 15 (2020).
[3] S. Widgren, P. Bauer, R. Eriksson & S. Engblom. Journal of Statistical Software, 91 (2019) 42.
[4] E.L. Snary, A.N. Swart, R.R. Simons, A.R. Domingues, H. Vigre, E.G. Evers, T. Hald & A.A. Hill. Risk analysis : an official publication of the Society for Risk Analysis, 36 (2016) 437-449.