Bayesian Modeling of Partially Observed Epidemic Count Data
Fintzi, Jonathan Refael
MetadataShow full item record
Epidemic count data reported by public health surveillance systems reflect the incidence or prevalence of an infectious agent as it spreads through a population. They are a primary source of information for shaping response strategies and for predicting how an outbreak will evolve. Incidence and prevalence counts are often the only source of information about historical outbreaks, or outbreaks in resource limited settings, which are of interest for researchers seeking to develop an understanding of disease transmission during ``peace time", with an eye on preparing for future outbreaks. The absence of subject--level information and the systematic underreporting of cases complicate the task of disentangling whether the data arose from a severe outbreak, observed with low fidelity, or a mild outbreak were most cases were detected. The magnitude of the missing data and the high dimensional state space of the latent epidemic process present challenges for fitting epidemic models that appropriately quantify the stochastic aspects of the transmission dynamics. In this dissertation, we develop computational algorithms for fitting stochastic epidemic models to partially observed incidence and prevalence data. Our algorithms are not specific to particular model dynamics, but rather apply to a broad class of commonly used stochastic epidemic models, including models that allow for time--inhomogeneous transmission dynamics. We use our methods to analyze data from an outbreak of influenza in a British boarding school, the 2014--2015 outbreak of Ebola in West Africa, and the 2009--2011 A(H1N1) influenza pandemic in Finland.
- Biostatistics