Using heterogeneity in the population structure of U.S. swine farms to compare transmission models for porcine epidemic diarrhoea

Eamon B. O’Dea; Harry Snelson; Shweta Bansal

Both the objectives, identifying variables relevant to the risk of infection, and challenges of our data analysis, uncertain reporting rates and many correlated candidate predictors, are common in epidemiological studies. Two reasonable steps toward such an objective are to assemble as many relevant explanatory variables as possible about reporting rates and measures of exposure based on prior scientific knowledge and then to determine if the available data support the conclusion that these variables are relevant. Our general contribution has been to provide a worked-out example of how variation in the structure of the population across a large scale may allow for the identification of variables with relevance to mechanisms of spread. We have also demonstrated the use of stability selection and regularised regression for the task of filtering out noise variables from a set of candidates. These examples may serve to provide analysts with new ideas about how to make the most efficient use of often limited epidemiological data, hopefully leading to more rapid understanding of transmission and how to stop it. Table 3 . Summary of models. The models chiefly differ by how contact is assumed to depend on flows. In the null model, denoted by none, contact was independent of flows. In the internal model, contact was a function of within-state flows. In the directed model, contact was a function of flows moving into a state and within-state flows. In the undirected model, contact was a function of within-state flows and both flows into and out of a state. The column "Fit η?" indicates whether we estimated the value of η, which corresponds to risk that is independent of the number of infective farms. The symbol θ denotes the dispersion parameter of the negative binomial response. The symbol σ denotes the standard deviation of the random effect of (geographic) state on transmission rates. To display their density, they have been made transparent and jittered along the y axis. The y axis was transformed using y = log(Positive accessions + 1).