Challenges for Network Epidemic Models
https://www.sciencedirect.com/science/article/pii/S1755436514000334
Modeling epidemics using networks can be a useful tool in understanding and predicting the spread of a disease, as well as using this information to come up with effective containment and prevention solutions. However, due to the vast number of variables that can be manipulated in modeling such a complicated situation, current models still lack the depth and comprehension to accurately model the spread of infection to a high degree of accuracy. This paper highlights eight challenges for network epidemic models looking into the future of epidemiology. Below are some highlights amongst the different challenges. The networks they consider are undirected, consist of individuals as nodes and edges as acquaintances between them, and follow general SIR models.
- SIR epidemics do not inherently consider sensitivity in changes in probability with regard to degree distribution, as well as varying susceptibility/infectivity of individuals. Values such as R0 and p need to be related to ‘duration of epidemic and peak incidence’ as this heterogeneity impacts the outcome of the epidemic.
- Analytical methods are needed to study networks that are not uniform. For example, nodes vary based of attributes such as age, which impacts the level of immunity/susceptibility amongst their neighbors.
- Incorporation of waning immunity in network epidemic models: Analysis becomes more difficult as the level of immunity begins to change in individuals (the SIR models we have been considering result in permanent immunity rather than impermanent), making time of occurrence much more important, thus impacting percolation theory results.
- Development of a robust approximation method that can accurately describe an epidemic is needed. Currently, stochastic epidemics modeled by differential equations, and other approximation models cannot be compared in such a way as to know which method would be better given the constraints of the network being modeled.
- Linking network models with updated data with regards to the spread of disease: This can be done by determining where ‘weights’, or positively correlated interactions, need to be assigned within a network structure. For example, two friends at a school could have a weight between them for multiple reasons: age, school, clubs, etc. Which relationship or combination of relationships best dictates proper weight assignment?
In conclusion, epidemic network modeling is a quite complex field of current research that despite the many obstacles and challenges in modeling structure, holds potential into providing insightful information into our understanding of the spread of disease and improving future predictive models.