Multi-scale simulation of COVID-19 epidemics

TL;DR

This paper compares compartmental and agent-based models for COVID-19, combining them to improve epidemic predictions and policy impact assessments.

Contribution

It introduces a hybrid approach that leverages both models' strengths to enhance epidemic forecasting and policy analysis.

Findings

Agent-based models refine compartmental parameters.

Combined models better predict policy impacts.

Methodology bridges microscopic and macroscopic epidemic views.

Abstract

Over a year after the start of the COVID-19 epidemics, we are still facing the virus and it is hard to correctly predict its future spread over weeks to come, as well as the impacts of potential political interventions. Current epidemic models mainly fall in two approaches: compartmental models, divide the population in epidemiological classes and rely on the mathematical resolution of differential equations to give a macroscopic view of the epidemical dynamics, allowing to evaluate its spread a posteriori; agent-based models are computer models that give a microscopic view of the situation, since each human is modelled as one autonomous agent, allowing to study the epidemical dynamics in relation to (heterogeneous) individual behaviours. In this work, we compared both methodologies and combined them to try and take advantage of the benefits of each, and to overcome their limits. In particular, agent-based simulation can be used to refine the values of the parameters of a compartmental model, or to predict how these values evolve depending on sanitary policies applied. In this report we discuss the conditions of such a combination of approaches, and future improvements.