Real-time modelling and inference of Covid-19 transmission and control

The COVID-19 pandemic, caused by the novel virus SARS-CoV2, has grown to more than 215 million cases and 4.5 million deaths worldwide since the first case was reported in China in December 2019. It has had wide-ranging direct and indirect effects on population health and created a significant burden on national health and social care services. As a major public health emergency, it is of paramount importance to understand its transmission dynamics and identify effective control strategies to stem transmission.

Methods for control include vaccination and non-pharmaceutical interventions (NPIs) such as social distancing, mask wearing, travel restrictions, school closures, and full national lockdowns. NPIs often have significant social and economic costs, and vaccination comes with considerations of costs and supply. In the face of these costs and the uncertain situation posed by the pandemic, it is important to only enact NPIs when strictly necessary and to deploy vaccinations strategically for maximum impact. However, the current state of the pandemic is only reflected in case numbers and hospital admissions with a significant time lag and can change quickly, and the impact of new measures is often not immediately clear. This has been a barrier to timely and effective decision making.

Real-time modelling aims to alleviate these issues by providing information on the current state of the pandemic ('now-casting'), its trajectory (forecasting) and on the effect of control strategies (scenario modelling). This has been key in aiding decision-making around COVID-19 interventions and much progress has been made to improve methods for real-time modelling. A key challenge with real-time modelling of COVID-19 lies in its novelty: there is little historical data and a relatively limited body of evidence about the pathogen and the disease. There are many uncertainties around transmission, case ascertainment, effective control strategies, the impact of human behaviour, and the impact of new variants. This makes it hard to know a priori which modelling approaches and which data streams are most suitable for modelling. A promising approach to this challenge is to ensemble multiple base models into a single forecast.

In this project, we will compare and expand methods for real-time modelling and forecasting of Covid-19 and assess their utility for evaluating interventions and real-time decision making. This will be done using a combination of mathematical modelling and statistical analysis applied to subnational time series of Covid-19 cases, hospitalisations, vaccinations, and deaths from official sources such as the UK Government COVID-19 data dashboard and the Office for National Statistics. We will also explore wider novel data streams which may be incorporated to improve models, such as Google mobility data and CoMix study data to estimate changes in social behaviour, and data from prominent studies such as REACT and SIREN which have produced useful findings and data on seroprevalence and risk of reinfection respectively.

This project will develop the student's quantitative skills, including statistics, mathematical modelling, predictive evaluation, and computational skills such as advanced programming in R. The student will learn to apply these skills to a variety of data sources as described above. The student will also develop interdisciplinary skills at multiple interfaces, for example the application of computational skills to biomedical and epidemiological contexts; the interface between science and political decision-making; and the role of the social sciences such as behavioural science and economics in pandemic response.