Date of Award

2015

Document Type

Dissertation - MUSC Only

Degree Name

Doctor of Philosophy (PhD)

College

College of Graduate Studies

First Advisor

Andrew B. Lawson

Second Advisor

Elizabeth G. Hill

Third Advisor

Jeffrey E. Korte

Fourth Advisor

Jane E. Joseph

Fifth Advisor

Bo Cai

Abstract

Relative risk estimation or disease mapping concern the global smoothing of risk and estimation of true underlying risk level. However, it is also appropriate to investigate association with the local properties of relative risk surface. These local properties include peaks of risk and local heterogeneity in risk, and cluster detection is often the main focus on local features of the risk surface where elevations or depression of risks happen. Cluster analysis of disease incidence has a long history, and a variety of approaches can be adopted for this analysis ranging from testing-based methods to fully parameterized cluster. Although a range of models available with a variety of goals in disease mapping applications focuses on retrospective analysis, prospective analyses are essential in many public health situations when timeliness is a key component. The importance of the early detection of unusual public health events is the ability to detect rapidly any substantial changes in disease, thus facilitating timely public health interventions. There are two methods of detection: retrospective and prospective. A retrospective analysis is carried out for the whole dataset to decide on the presence of a change based on the information from the past. To detect changes prospectively, observations are added to the process and a decision is made whether to collect more data or declare as an outbreak. The later detection of changes is our focus of surveillance. The Centers for Disease Control and Prevention (CDC) defines an outbreak based on the number of cases occurring after an investigation of the disease. This definition is not adapted to the prospective analysis because an alarm should be triggered before the investigation and thus before the determination of a potential epidemiological link between cases. To assist public health practitioners to make the decision, statistical methods are adopted to assess unusual events on the fly. In this research plan a range of novel Bayesian spatial models and measures for disease cluster assessment and public health surveillance are proposed and evaluated. The general aims of the proposal are structured as follows: Aim 1: Evaluation of Cluster recovery for small area relative risk models. The analysis of disease risk is often considered via relative risk. The comparison of relative risk estimation methods with ‘true risk’ scenarios has been considered on various occasions. However, there has been little examination of how well competing methods perform when the focus is clustering of risk. In this paper, a simulated evaluation of a range of potential spatial risk models and a range of measures that can be used for a) cluster goodness-of-fit, b) cluster diagnostics, are considered. Results suggest that exceedence probability is a poor measure of hot spot clustering because of model dependence, whereas residual–based methods are less model dependent and perform better. Local deviance information criteria (Local DIC) measures perform well, but conditional predictive ordinate (CPO) measures yield a high false positive rate. Aim 2: Bayesian detection of small area health anomalies using Kullback – Leibler divergence. The importance of early detection of unusual health events depends on the ability to rapidly detect any substantial changes in disease, thus facilitate timely public health interventions. To assist public health practitioners to make decisions, statistical methods are adopted to assess unusual events in real time. We introduce a surveillance Kullback-Leibler (SKL) measure for timely detection of disease outbreaks for small area data. We investigate the performance of the proposed surveillance technique and compare with the surveillance conditional predictive ordinate (SCPO) within the framework of Bayesian hierarchical Poisson modeling using a simulation study. Finally, the detection methods are applied to a case study of a group of respiratory system diseases observed weekly in South Carolina counties. Aim 3: Prospective Bayesian surveillance for spatial case event data. There has been little development of surveillance procedures for epidemiological data with fine spatial resolution such as case events at residential address locations. This is often due to difficulties of access when confidentiality of medical records is an issue. However, when such data are available, it is important to be able to affect an appropriate analysis strategy. We propose a model for point events in the context of prospective surveillance based on conditional logistic modeling. A weighted conditional autoregressive model is developed for irregular lattices to account for distance effects, and a Dirichlet tessellation is adopted to define the neighborhood structure. Localized clustering diagnostics are compared including the proposed local Kullback-Leibler information criterion. A simulation study is conducted to examine the surveillance and detection methods, and a data example is provided of non-Hodgkin Lymphoma data in South Carolina.

Rights

All rights reserved. Copyright is held by the author.

Download

Share

COinS