Stochastic mechanistic models and Bayesian inference for direct and environmental disease transmission: applications in aquaculture

Abstract

Stochastic dynamic epidemiological models are key to data-driven understanding of infectious disease. This thesis aims to expand, develop and provide deeper insights into applications of direct (DT) and environmental transmission (ET) models as mechanistic descriptions of environmentally transmitted infections like cholera and diseases affecting aquaculture production.

We explore timescale separation between host and environmental pathogens as the factor determining whether ET may be adequately described by a DT model. When fitting DT models to data, graphical posterior-prediction demonstrates robustness to departures from DT.

Rates of environmental transmission and pathogen decay of white spot disease among penaeid shrimp are estimated from published data and used to formulate the DT, susceptible-exposed-infected-removed (SEIR), and ET susceptible-exposed-infected-removed-pathogen (SEIR-P) models. Investigation of regular partial removal of dead and diseased shrimp reveals that host-disease dynamics of both models are almost indistinguishable with 24-hourly removals. However, for 6-hourly removals the SEIR model under-estimates average reduction in outbreak size and over-estimates duration compared with the SEIR-P model, demonstrating limitations of DT models.

Novel methods of Bayesian inference are shown to reliably estimate key model parameters using data from routine immersion challenge experiments (ICE). How design of such experiments can be aided using simulation and ideas from information theory is shown. Using published data and particle-marginal Markov chain Monte Carlo (PM-MCMC), we estimate the transmission rate and mean duration of latent infectiousness for a novel strain of Piscine orthoreovirus affecting rainbow trout.

Finally, using PM-MCMC, we show that rates of environmental transmission, pathogen emission and loss of pathogen viability of an ET model that distinguishes between viable and non-viable pathogen may be estimated from routine ICE data with additional digital polymerase chain reaction measurements of waterborne pathogen load. This work opens up application of ET models to develop understanding of pathogen dynamics beyond the confining assumption of direct transmission.