Statistical methods for characterising the severity of an emerging pathogen: case studies of the COVID-19 pandemic
File(s)
Author(s)
Hawryluk, Iwona
Type
Thesis
Abstract
Epidemiological modellers face significant challenges during outbreaks of novel pathogens, particularly the need for rapid and precise modelling with limited and delayed data. The COVID-19 pandemic revealed the insufficiency of current methods in addressing these challenges and highlighted the need for more innovative approaches to infectious disease modelling. Motivated by these issues, this thesis aims to aid epidemiologists in building more robust models by developing novel Bayesian methods that enhance the accuracy and reliability of epidemiological models, especially for emerging pathogens such as SARS-CoV-2.
Specific contributions of this thesis are as follows: Chapter 2 introduces a new method for estimating marginal likelihoods using thermodynamic integration, facilitating model selection through Bayes factors. Applied to COVID-19 data, this approach reveals the pitfalls of model selection and emphasises the importance of rigorous methods. In Chapter 3, a range of probability density functions is fitted to hospitalisation distributions for COVID-19 patients in Brazil, providing crucial inputs for epidemic models. Spatial heterogeneity in hospitalisation times is explored, offering insights into regional variations in disease dynamics. Chapter 4 shifts focus to data quality, investigating reporting delays in COVID-19 mortality data for Brazil. A novel method using Gaussian Processes is proposed to correct reporting delays, enabling real-time monitoring of epidemiological trends with greater accuracy. Building on these methodological advancements, Chapter 5 explores the impact of regularisation in a renewal-equation-based R_t model, demonstrating the importance of informative priors in accurately estimating importations during emerging epidemics.
The methods proposed here aim to assist infectious disease modellers in rapidly responding to emerging threats using Bayesian statistical tools. Overall, this work combines traditional epidemiological approaches with modern statistical and machine learning methods to address the challenges faced during the COVID-19 pandemic, providing a framework for improving infectious disease modelling in future outbreaks.
Specific contributions of this thesis are as follows: Chapter 2 introduces a new method for estimating marginal likelihoods using thermodynamic integration, facilitating model selection through Bayes factors. Applied to COVID-19 data, this approach reveals the pitfalls of model selection and emphasises the importance of rigorous methods. In Chapter 3, a range of probability density functions is fitted to hospitalisation distributions for COVID-19 patients in Brazil, providing crucial inputs for epidemic models. Spatial heterogeneity in hospitalisation times is explored, offering insights into regional variations in disease dynamics. Chapter 4 shifts focus to data quality, investigating reporting delays in COVID-19 mortality data for Brazil. A novel method using Gaussian Processes is proposed to correct reporting delays, enabling real-time monitoring of epidemiological trends with greater accuracy. Building on these methodological advancements, Chapter 5 explores the impact of regularisation in a renewal-equation-based R_t model, demonstrating the importance of informative priors in accurately estimating importations during emerging epidemics.
The methods proposed here aim to assist infectious disease modellers in rapidly responding to emerging threats using Bayesian statistical tools. Overall, this work combines traditional epidemiological approaches with modern statistical and machine learning methods to address the challenges faced during the COVID-19 pandemic, providing a framework for improving infectious disease modelling in future outbreaks.
Version
Open Access
Date Issued
2024-04
Date Awarded
2024-08
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Bhatt, Samir
Mellan, Thomas
Sponsor
Medical Research Council (Great Britain)
Publisher Department
School of Public Health
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)