Obesity Risk Estimation Accounting Spatial Dependency, Error in Covariate Measurement, and Factors Operating at Multiple Levels
Date
2023-12-19
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ORCID
0000-0003-4703-2184
Type
Thesis
Degree Level
Doctoral
Abstract
Disease mapping has long been a part of public health, epidemiology, and the study of disease in human populations. Hierarchical spatial models for areal data address the competing goals of accurate small area estimation and fine-scale geographic resolution in disease mapping simultaneously, and it has become a fertile area of research over the last two decades. More recently, there has been increased uptake of the methods in applied research. Nonetheless, there is still scope for the methodological developments. This thesis contributes to the uptake of disease mapping in applied health research through key areas: methodological development, implementation, and application. Chapters 1 and 2 of this thesis provide a brief review of literatures on spatial model, measurement error model, and obesity research. These chapters also summarize methods and data to be utilized in this thesis. Chapter 3 presents an applied research work that demonstrates the importance of incorporating spatial autocorrelation from the observed data into a statistical model, via real and simulated data. The analysis of real data across 117 health regions of Canada is of practical interest, as it identified several obesity clusters with discernible spatial patterns throughout Canada. Chapters 4 and 5 of the thesis present two research works on methodological development. First, covariate measurement error provides biased estimates in standard regression model, violating underlying assumption. In Chapter 4, the classical and Berkson measurement error models were integrated with the well-known Besag-York-Mollie (BYM2) model to incorporate covariate measured with error. The simulation results revealed that the use of a measurement error model for an error-prone covariate in BYM2 model has the advantage of producing a superior fit. The results also demonstrate that a BYM2 model without taking into account covariate measurement error may lead to highly biased estimates for certain parameters. The proposed method was applied for estimating socio-economic and environmental factor’s effect on the obesity counts using aggregated data for 117 health regions of Canada. Second, optimal prediction of risk for an adverse health condition risk at population level requires integrating covariates from multiple levels into a single modeling framework. However, it is a common practice to estimate effects of individual and group-level covariates using multiple models independently. To overcome this methodological gap, this thesis formulated the joint BYM2 model in Chapter 5, that integrates individual- and group-level models through association parameter. The simulation results revealed that the joint BYM2 model performed the same or better than the independent estimation for recovering parameter values. The capability of the proposed model was demonstrated through estimating the risk of developing unhealthy health condition among Canadian secondary school students, integrating individual-, school-, and neighbourhood-level covariates. The neighbourhood-level model incorporated spatially correlated count data and covariates measured with error. Finally, in Chapter 6, the overall findings from this thesis and potential directions for future work are discussed.
Description
Keywords
Berkson Measurement Error Model, Classical Measurement
Error Model, Spatial Statistics, Bayesian Hierarchical Model, Joint Model, Disease Mapping
Citation
Degree
Doctor of Philosophy (Ph.D.)
Department
School of Public Health
Program
Biostatistics