Bayesian spatio-temporal hierarchical models for disease mapping and detection

Abstract

In recent years, emerging computational algorithms have revolusionised the application of sophisticated Bayesian hierarchical models in numerous fields, including public health. Such models have become widespread in disease mapping and detection studies, where data of complex spatio-temporal structures arise, due to their flexibility to accommodate associated statistical challenges. This thesis contributes to the scientific literature by building upon existing disease mapping and detection spatio-temporal models within the Bayesian hierarchical framework in order to answer important epidemiological questions. Motivated by chronic respiratory diseases and road traffic accidents, two major public health concerns characterised by inherent geographical structures, three main objectives are established. First, a Bayesian model is proposed for the joint analysis of road traffic accidents by taking into account dependences across space, time and also between severity levels, in order to perform disease mapping and highlight increased risk of accident. Second, the chronic respiratory disease is investigated by using a recently proposed Bayesian detection model. To achieve this, three data sources are analysed: mortality, hospital admissions, and drug prescriptions, each capturing a different level of disease severity. Along with providing spatio-temporal patterns of the chronic respiratory disease, the detection of unusual areas in terms of temporal behaviour is enabled, thus providing insights for hypothesis generation. Following on from the previous work, a more flexible detection model is finally developed and evaluated through an extensive simulation study. Comparisons against the benchmark model are performed, suggesting great improvements of the proposed model. To illustrate the model, two case studies are carried out, on road traffic accidents and chronic obstructive pulmonary disease.