Multivariate Bayesian Modelling of tick life-stage count data incorporating spatial and time variations.

Abstract

Increasing tick abundance and tick-borne pathogens constitute a growing threat to public health. Five major ticks species were reported to be active and responsible for transmitting a variety of pathogens of both human and animals during the period of the original research, namely: Ixodes scapularis, Amblyomma americanum, Dermacentor variability, Amblyomma maculatum, and Haeaphysalis longicornis. This study uses tick life stages monthly data collected from 2009 to 2018 from 12 random sample sites within the edges, grass and woods habitat types in eight southeastern counties in Virginia, United States. The availability of time series and geo-referenced datasets for modelling has necessitated the development and application of dynamic time series and spatio-temporal statistical methods. In this study, two Bayesian estimation techniques were investigated. The study compares the model overall performance by the Markov Chain Monte Carlo (MCMC) method with the Integrated Nested Laplace Approximation (INLA)technique. Multilevel Bayesian models were introduced using the INLA technique and its flexibility and computation time was demonstrated and compared with the MCMC technique. This dissertation contributed by developing models that enabled incorporation of the association among the components of response vector over time, while modelled as a function of time and space-time covariates. The models developed were successful in revealing environmental and time variations effects on the distribution of tick life stages. Alternative to the frequentist approach, the use of prior distributions in Bayesian modelling was useful for improving the model accuracy. Bayesian models offers flexible specification of complex models through the inclusion of random effects, hyperparameters and time-varying coefficients. The study investigates the temporal behaviour of tick life cycle stages in each month, taking into consideration the effects of time trends, seasonal, and environmental variations. The association between tick life stages is modelled using a multivariate Poisson, zero inflated Poisson and/or negative binomial distributions. Similar monthly time effects results were found in chapters 4, 5 and 6. Secondly, variation within the random sample sites was investigated using different prior distributions. This type of model was able to reveal that some areas in York, Portsmouth, Cheasapeake and Northampton counties had a significant higher tick variations while some areas in Portsmouth and Norfolk had significant lower variations. Lastly, the study employed spatio-temporal models to unpack the effects of space over a period of time. The results showed that tick abundances were influenced by environmental factors and seasonal changes and it was concluded that tick abundances depend on the type of habitat where they are closer to their hosts and the time when their hosts are more likely to be targeted.