Understanding the epidemiology of Bluetongue virus in South India using statistical models

Bluetongue is an economically important midge-borne disease affecting domestic and wild ruminants worldwide. The disease, caused by the bluetongue virus (BTV), is highly endemic in South India, occurring with varying severity every year since 1963, causing high morbidity and mortality, resulting in huge economic losses to subsistence farmers, impacting the GDP of the country and affecting food security. The bluetongue epidemiological system in South India is characterized by an unusually wide diversity of susceptible ruminant hosts, many potential Culicoides vector species and numerous pathogen serotypes and strains. These factors (intrinsic and extrinsic) contribute to disease impacts that vary widely over geographical space. Chapter 2 deals with identification of remote sensed variables in discriminating between presence and absence of bluetongue outbreaks and development of a risk map using non-linear discriminant analysis (NLDA) approach. Chapter 3 deals with understanding the role of extrinsic factors such as monsoon conditions in driving seasonality in BTV outbreaks over two decades in Andhra Pradesh, India using a Bayesian Poisson regression model framework, accounting for temporal autocorrelation. In chapter 4, the mean annual numbers of outbreaks in each district in South Indian states were examined in relation to land-cover, host availability and climate predictors using a Bayesian generalised linear mixed model with Poisson errors and a conditional autoregressive error structure to account for spatial autocorrelation. In chapter 5, the annual number of outbreaks in each district in South India was examined in relation to climate predictors (temperature and precipitation) at different lags using a Bayesian generalized linear mixed model with Poisson errors. In chapter 6, a range of suitable predictors was considered for identifying their relationships with bluetongue outbreaks using Bayesian Network Modelling (BNM), and the important variables were used to develop a Bayesian geostatistical model accounting for spatial autocorrelation. The analysis resulted in the development of spatial risk maps at district and village levels, district level yearly predictions and monthly state level predictions, which can contribute to the development of an early warning system for the disease in South India.