Extending Space-Time Putative Hazard Models to Detect Latent Risk Features

In recent years small area risk assessment modeling and data analysis around putative hazard sources has become a fundamental part of public health and environmental sciences. This dissertation work examines a novel development of three different space-time Bayesian hierarchical modeling methods. In Part I, we have addressed a fundamental problem in the analysis of small area health outcomes data, when intermittent operation of facilities could lead to evidence for latent periods of risk. This work examines the development of Bayesian models for the latent switching operating period of putative hazard sources, such as nuclear processing plants, waste disposal incinerators and cement factories. The developed methodology is applied in a simulation study as well as to a real data example. In Part II of this project, we have considered the modeling of sparse small area health data. Often when sufficiently rare diseases are considered for analysis, a large number of areas will have a zero count of disease incidence. This existence of zero counts of diseases (or any health outcomes, such as death due to rare disease) in small administrative region units has been a problem within small area studies (Lawson and Clark, 2010). Thus, we have considered a variety of space-time Bayesian hierarchical models for sparse small area count data with a focus on the time-varying distance-decline effect and possible latent period switching detection. The application of the proposed method is applied to a real small area health outcome data example. In Part III, the analysis of multiple health outcomes around a putative hazard source is considered. There are situations in which interest lies in studying the association of two or more possible health outcomes that are thought to be linked to the same risk factors such as the same putative hazard source within the study region. Hence, the joint disease analysis is a powerful tool that may help detect similar geographical variation patterns in the risk of two or more related diseases. We have extended the current space-time Bayesian hierarchical shared component models by allowing for possible distance effect. The proposed method is illustrated by analyzing jointly Non-Hodgkin\u27s lymphoma (NHL) and soft tissue sarcomas (STS) cancers