10.5281/zenodo.198183
https://zenodo.org/records/198183
oai:zenodo.org:198183
Adrakey, Hola
Hola
Adrakey
Heriot-Watt University
Gibson, Gavin
Gavin
Gibson
Heriot-Watt University
Spread model for PPR
Zenodo
2016
Epidemiology
Peste des Petit Ruminants
Modelling
Vector-borne disease
Bayesian approach
individual
network models
2016-12-09
10.5281/zenodo.168021
https://zenodo.org/communities/efsa-kj
Creative Commons Attribution 4.0 International
Concerning the spread of PPR within a country, there are several ways to model the susceptible population. Data from Tunisia provide information on infected sub-populations, suggesting that a meta-population approach where susceptibles are modelled as a network of interconnected nodes would be appropriate. Although not implemented in this study, this framework would offer the scope to capture heterogeneity across nodes, representing the species composition of individual farms. The strength of connections in this network would then be influenced by the perceived strength of the infectious challenge between nodes. A natural initial approach to this would be to represent the strength of interactions as being monotonically related to Euclidean distance.
Anticipating the problems regarding a lack of knowledge on the susceptible population, it is also beneficial to consider frameworks where the susceptible population is represented via a continuous spatial intensity, which then moderates the occurrence of new infections. In particular, it was noted that a framework in which spatial contact distribution models (described later) for the initiation of new infections are used to formulate a marked point process model in which marks are drawn from a distribution - representing the diversity of units forming the susceptible population - may be appropriate, and may offer a way of accommodating the lack of information on the susceptible population.
It was identified as being important that parameter uncertainty could be treated properly for any models that are fitted to data on PPR, so that these uncertainties could be propagated through to predictions. With this in mind, particular consideration was given to finding models that could be fitted using either likelihood-based methods (to extract maximum information from data) or a fully Bayesian approach, for uncertainty propagation. As previously mentioned, it was considered important that any approach to model fitting could cope with unobserved processes, or with the increased model complexity that arises from, for example, including parameters for the probability of non-reporting of infections. For methods that use data on final size, the inference of epidemic parameters is potentially relevant.
The model is implemented in R