Bayesian models and methods to estimate age-specific infectious disease transmission dynamics: integrating disease surveillance time series, mobility, and vaccination data

Abstract

As of 2022, the coronavirus disease 2019 (COVID-19) pandemic is an ongoing global public health issue. To mount efficient public health control programs, it is essential to understand infection patterns involving, for example, the demographics of vulnerable populations or populations driving transmissions and how these vary over time. In this thesis, we make contributions to the use of Bayesian models and methods to capture fine-scale transmission dynamics of infectious diseases, with applications to Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). From a statistical point of view, transmission dynamics may be informed by various data, and we make additional contributions by extracting and using novel data types. We aim to push the complexity of the model as much as the stratification of the data and computational power allow us to, and we perform inference on these complex models using appropriate Bayesian methods. The thesis is structured around two parts. In the first part, we build on age-specific renewal models and develop a Bayesian framework for estimating age-specific transmission dynamics in the context of time-varying social contacts using daily, age-specific human mobility, and mortality data. The model is applied to reconstruct and evaluate the resurgence of COVID-19 between March to October 2020 in the United States (US). We identified the population age groups driving SARS-CoV-2 spread across the US through October 29, 2020 and projected the changes in transmission dynamics after school reopening. This work was published in the Science journal. In the second part, we develop and characterise a computationally efficient, non-parametric smoothing prior, a low-rank Gaussian process projected by regularised B-splines, as an approximation to a standard Gaussian process. We embed this prior into a Bayesian hierarchical model to estimate the evolution of age-specific COVID-19 deaths by 1-year age bands across 50 US states from May 2020 to January 2022. We uncover longitudinal trends in the age profile of COVID-19 deaths, identified states with unusually high mortality in older age groups, explored associations with deaths in carehomes, and showed that strong resurgences in deaths during the summer of 2021 are associated with low vaccination coverage. This work was published in the Bayesian Analysis journal.