Dissertation

Bayesian Spatiotemporal, Sample Survey, and Forecasting Methods for Analyzing COVID-19 Infections and Mortality

Slater, Justin James Ian

2023

Overview

For decades, mathematicians and statisticians have been modelling infectious diseases to forecast case/death counts, estimate important epidemiological quantities, and understand the dynamics of disease spread. This dissertation offers methodological insights into each of these three challenges using novel spatial, spatio-temporal, and Bayesian modelling methods, with applications to COVID-19 data. Alongside methodological contributions, this thesis also presents estimates of important epidemiological quantities which, subject to peer review, could be utilized by public health professionals and policy makers. There are four primary contributions of this work: 1) a subnational, single-wave COVID-19 mortality forecasting model that accounts for day-of-the-week effects, which was shown to outperform the most highly-cited model during the first viral wave; 2) a mobility-augmented spatial model for COVID-19 case counts, where cellphone-derived mobility data is shown to capture dependence between areal units better than physical proximity; 3) a novel, interpretable spatio-temporal infectious disease model where infectiousness is a function of mobility between areal units, resulting in estimates of the risk associated with travelling in two Spanish Communities; 4) a modular Bayesian framework based on mixture modelling of serological data and disaggregated deaths data to estimate COVID-19 incidence and infection fatality rates, resulting in estimates of these quantities across Canada for various strata. Although the applications in this thesis are to COVID-19 data, the proposed methodology can be applied to a wide spectrum of problems across infectious disease epidemiology.

• Publisher: ProQuest Dissertations & Theses
• Subject: Epidemiology / Public health / Statistics
• ISBN: 9798379764012