Uncertainty Quantification for Expensive COVID-19 Simulation Models (UQ4Covid)

Accurate mathematical models of transmission are crucial for targeting successful interventions to combat the spread of SARS-Cov2. In the UK, established models are used to provide real time policy support to Government through the Scientific Pandemic Influenza group - Modelling (SPI-M). Modellers in SPI-M have a proven track record, and models are continually adapted to respond to the evolving pandemic.

When using models to inform decision making, it is crucial that all sources of uncertainty are properly accounted for when calibrating and predicting. For 30-years the UK has been a world-leader in developing Uncertainty Quantification (UQ); delivering methods for formal treatments of uncertainty when using models to understand the world, allowing efficient and robust calibration and prediction.

Despite this, these techniques are not currently in place for COVID-19 simulation models, leading to slower-than-necessary adaptive model development-UQ allows for fast re-calibration-and an under-representation of uncertainty in predictions delivered to policymakers.

This project will adapt and deliver UQ techniques, code and tutorials for models of COVID-19 in the UK, providing SPI-M modellers with tools to facilitate rapid re-calibration of their models when changes are made in response to the evolving pandemic, and to more accurately represent uncertainty in their predictions. We will work closely with MetaWards, a spatial meta-population transmission framework (Danon et al. 2009, 2020) that contributes to SPI-M, to develop and apply these tools as we move into the winter; enabling fast evaluation of interventions responding to localised outbreaks, efficacy of vaccine rollout strategies, duration of immunity and more.