A Visual Analytics and Multi-Objective Optimisation Approach for Balancing Economic and Public Health Objectives through Compartmental Models

Lead Research Organisation: Swansea University
Department Name: College of Science

Abstract

Modelling of disease spread continues to play a crucial role in the response to the COVID-19 pandemic. There is light at the end of the tunnel with effective vaccines, but it will take until the summer of 2021 for distribution to be widespread, and re-vaccination may be an ongoing requirement. In the meantime, hybrid solutions are required to manage non-pharmaceutical interventions (NPI), that minimise the restrictions to our daily lives while suppressing transmission and maintaining the integrity of our healthcare systems.

Realistic models are publicly available to predict the spread of the virus. Varying the parameters of these models can be used to represent tentative policy actions, and the consequences are deduced in simulations of the model. Typically, the main objective for identifying effective policy actions has been to reduce the infection rate. However, often there are multiple, potentially conflicting, objectives that require optimisation in parallel. For instance, we may want policies that reduce the hospital occupancy, while simultaneously
reducing economic impacts.

Our goal is to provide a generic visual analytics framework to explore the parameter space of complex models as well as the trade-offs between objectives to inform policy makers. Specifically:
1. A scalable visual analytics framework for parameter space exploration of feasible regions of the parameter space for complex compartmental models in order to identify effective policy actions.
2. Extend this framework to handle multiple objectives: reduction of transmission to high risk groups, overall cases and deaths, hospital costs, thresholds for circuit breakers, and economic factors.
 
Description Our work has resulted in a practical optimisation framework for generating medium-term projections of key parameters of Covid-19, such as death, hospitalisations, and hospital occupancies, using a sophisticated compartmental model (referred to as the Swansea Covid model) and the data collected from the health boards in Wales. These projections are generated now on a weekly basis, and are then communicated to the First minister of Wales through the director of health economics. Furthermore, we send these projections to the UK health security agency (UKHSA). They use our projections as part of their ensemble projection and inform the UK government. The First minister acknowledged our contributions towards policymaking in Wales on national television.

We have also developed a visualisation framework for evaluating the efficacies of the Swansea Covid model. We showed that the model performs well when no new variants are considered. Furthermore, we also demonstrated that, even when new variants emerge, we have the capability to adjust our projections with only about a week's worth of data on the change in direction.

Our current focus is on reflectively analysing policy decisions during crucial past events, e.g. firebreak at the end of 2020, and evaluating opportunity costs of alternative policies, from a multi-objective perspective, i.e. trading-off between policy stringency (reflecting economic circumstances) and health objective (e.g. average total occupancy). The framework for this has been developed, and some initial analyses have been performed. However, we are continuing to improve the framework before presenting analyses before the Welsh government.
Exploitation Route The projections we produce are discussed by the Welsh Technical Advisory Group (TAG), and then communicated to the First minister. They are directly used in policymaking in Wales.
Sectors Healthcare,Government, Democracy and Justice
 
Description Yes. Our tool helps us produce regular projections of essential parameters of Covid-19, such as deaths, hospital admissions and occupancies. These projections in turn are communicated to the Welsh Technical Advisory Group (TAG) and the Director of Health Economics, who would discuss it between them, and present the findings to the First minister. In this way, our work has directly influenced the policies taken in Wales since late 2021.
First Year Of Impact 2021
Sector Healthcare,Government, Democracy and Justice
Impact Types Societal,Economic,Policy & public services
 
Description Providing Medium Term Predictions for Wales to Scientific Pandemic Influenza Group on Modelling (SPI-M)
Geographic Reach National 
Policy Influence Type Participation in a guidance/advisory committee
Impact The projection we produced were used in informed policymaking. This directly helped in saving lives in Wales. Also, since the process of informed policymaking was communicated to the public, this helped change their attitude toward adhering to restrictions.
 
Title Fitness landscape analysis via extrema graphs 
Description Fitness Landscape Analysis (FLA) often aims to visually represent a perspective, or map, on the search space, allowing for reasoning before optimisation. Currently, the dominant approach for visual FLA is the Local Optima Network (LON) where the local structure around a potential global optimum is visualised using a network with the nodes as local minima and the edges as transitions between those minima through an optimiser. In this work, we developed an approach based on extrema graphs, where transitions are captured between both maxima and minima embedded in two dimensions through dimensionality reduction techniques (multidimensional scaling in our prototype). These diagrams enable EC practitioners to understand the entire search space by incorporating global information describing the spatial relationships between optima. In a paper under review, we demonstrated the approach on a number of continuous benchmark problems from the literature and highlight that the resulting visualisations enable the observation of known problem features, leading to the conclusion that extrema graphs are a suitable tool for extracting global information about problem landscapes. This tool is expected to serve as an avenue to explore the landscape generated by compartmental models when fitting them to the observed data. 
Type Of Material Computer model/algorithm 
Year Produced 2023 
Provided To Others? No  
Impact This technique will enable optimisation specialists to analyse new problems and their landscapes, and help develop effective problem-specific optimisation algorithms. 
 
Title Optimiser for compartmental models 
Description The software framework consists of a Python wrapper for an R compartmental model and a multi-objective Bayesian optimiser that can simultaneously fit the model response to a range of observed parameter values. Once the model is fitted, the framework allows us to generate a range of medium and long term projections. Also, this framework can be used to optimise over policy parameters to identify effective policies that may yield target outcomes as proejcted by the compartmental model. 
Type Of Technology Webtool/Application 
Year Produced 2022 
Impact This software is being used by the team on a weekly basis to generate medium term projections of Covid-19 parameters such as death, hospitalisations and occupancies, which in turn is informing the Welsh government.