Daniel Archambault (Newcastle University) & Alma Rahat (Swansea University)

Talk Title: Generating Medium-Term Projections of Covid-19 for Wales.

Abstract:  One of the key tools used by the Welsh Government during its devolved pandemic decision-making was a stochastic, age-structured, local authority level, epidemiological model, designed to capture the key dynamics, and project the potential future spread and hospital impact under a range of formal assumptions. Some of these models are complex and often take several seconds or minutes to generate a single simulation on several indicators of interest, e.g. prevalence, deaths, hospital admissions, and intensive care burden. The typical approach to fitting the models to observed data consists of optimising model parameters to fit a weighted sum of the mean squared errors (MSE) between the model output and multiple time series data. In this work, we combined MSE and maximum absolute distance between projections and data using augmented Tchebyshev scalarisation as a measure of the error. Furthermore, to combat the computational expense of optimising up to 50 parameters of the model, we used a multi-objective Bayesian optimisation approach to simultaneously minimise the errors between six time-series projections and data. This approach yielded a good approximation of the sets of parameters that generated (near-)optimal trade-off fits between different targets, and was easily generalisable in post-processing if different weights were required for each data stream. The resultant trade-off front allowed us to interactively generate appropriate visual aids, and, most importantly, "Medium-Term Projections" (MTPs) for Wales, deemed plausible by experts. Our generated MTPs have been prepared in collaboration with the Scientific Pandemic Influenza Group on Modelling (SPI-M) and the UK Health Security Agency (UKHSA), and included in their weekly published reports. They were updated weekly, and delivered to stakeholders in Welsh Government, Public Health Wales, the NHS Wales Hospital Trusts, the Ambulance Trust, and directly impacted planning, assessment and decision-making.