Modelling the spatial and temporal distribution of Bacillus anthracis suitability across Uganda and Kenya: A Bayesian Approach

Valentina Awino Ndolo

Anthrax disease, caused by Bacillus anthracis, poses a significant global disease burden, leading to sporadic outbreaks across a variety of climatic regions. Through the action of two virulence plasmids, pXO1 and pXO2, the bacteria can evade host immunity and produce two toxins that can rapidly kill susceptible hosts, which then shed the bacteria back into the environment, resulting in recurrent outbreaks and catastrophic loss of animal life. The location, timing, size, and duration of anthrax outbreaks usually vary from year to year, making the characterization of the spatial-temporal patterns of disease occurrence challenging. It is necessary to consider environmental and host-specific drivers of the spatial distribution of B. anthracis suitability across different contexts. Ecological datasets have become richer, and there are statistical challenges associated with their analysis. Accurate insight and prediction can only be developed using a novel hierarchical approach that deals with structural dependencies in data, and so produces more robust predictions of B. anthracis suitability to guide efficient prioritization of scarce resources for disease prevention.

In this thesis, I examine the environmental and socio-economic drivers of B. anthracis suitability, applying both conventional and novel hierarchical modelling approaches. In Chapter 1, I describe the biology, ecology, clinical signs and diagnosis, treatment, prevention, control, and epidemiological burden of anthrax disease. In Chapter 2, I review published literature to understand the type of input data, suitability modelling algorithms, and evaluation methods previously used. I discuss the application of the Integrated Nested Laplace Approximation (INLA), a spatial method that can be applied to structurally complex data, to model the suitability of B. anthracis. In Chapter 3, I apply a common conventional algorithm to analyse a 15- year dataset comprising confirmed and probable wildlife, livestock, and human anthrax cases collected across Uganda to determine the spatial distribution of B. anthracis suitability across the country. In Chapter 4, I apply INLA to develop a model for B. anthracis suitability in Uganda using the same dataset as in Chapter 3.

In Chapter 5, I use a spatiotemporal model to investigate the socio-economic and climatic drivers of anthrax occurrence and incidence in Kenya at the national and sub-county levels. In Chapter 6, I assess the knowledge, attitudes, and practices of livestock farmers in Northern Uganda that might influence anthrax prevention and control. In Chapter 7, I summarise my findings and discuss how the new insights into the drivers and the spatial distribution of B. anthracis suitability can avail new opportunities for more robust anthrax prevention and control and future research.