Use cases

Epidemic Decision Making

picture11

Use Case Description

Emerging infectious diseases can have a significant impact on public health and the economy. We have been recently reminded of the havoc such events can inflict by the current COVID-19 pandemic caused by the SARS-CoV-2 virus. Yet, this was not an isolated event, with as other recent examples the H1N1 influenza pandemic in 2009 and the SARS-CoV-1 pandemic in 2003. Therefore, better understanding how infectious diseases are propagated and how we can mitigate epidemics is important from an economical, societal and healthcare perspective. To properly understand these dynamics, and to study emergency scenarios, epidemiological models are necessary. Such models enable us to perform scenario analyses and to study the effect of prevention strategies (e.g., vaccine allocation policies, contact reduction schemes, contact tracing procedures, school closures) in simulation. The development of prevention strategies, which need to fulfil distinct criteria (i.a., prevalence, mortality, morbidity, cost), and need to consider the viewpoints of different experts, remains a challenging process. Therefore, there is a great potential to use AI techniques to reason about epidemic mitigation strategies.

Results, Highlights & Research Outputs

Exploring the Pareto front of multi-objective COVID-19 mitigation policies using reinforcement learning

Infectious disease outbreaks disrupt public health and societal processes, making epidemic mitigation complex. Multi-objective reinforcement learning (RL) combined with epidemic models can help design prevention strategies that consider various conflicting criteria, such as mortality, economic costs, and well-being. Through an international collaboration (including FAIR partner UHasselt), Reymond et al. propose to use deep multi-objective RL with Pareto Conditioned Networks (PCN) to optimize deconfinement strategies in Belgium during COVID-19, aiming to minimize both infections and societal burdens [Reymond2024]. The approach effectively explores trade-offs between social restrictions and outcomes, showing that multi-objective RL offers valuable insights for balancing epidemic mitigation policies. 

Outputs & Resources

Publications

  • [Reymond2024] Reymond, M., Hayes, C. F., Willem, L., Radulescu, R., Abrams , S., Roijers, D. M., Howley, E., Mannion, P., Hens, N., Nowe, A., & Libin, P. Exploring the Pareto front of multi-objective COVID-19 mitigation policies using reinforcement learning. Expert Systems with Applications, 249 (Part C), 2024.

Software & Tools

  • gym-covid: A gym environment for a compartment model modelling the first COVID-19 wave in Belgium.

Uptake

  • Results will be used in FWO-SBO ANTICIPATE (VUB, UAntwerp, University Hospital Brussels; 2025-2029) on AI-guided Epidemic Decision Making in hospitals. The project has an advisory board with different stakeholders from the healthcare sector (hospitals, government, health insurance providers, scientific experts and healthcare umbrella organizations).

Team & Contact

Contact

Pieter Libin, Vrije Universiteit Brussel - Artificial Intelligence Lab

Use Case Lead(s)

Pieter Libin, Vrije Universiteit Brussel - Artificial Intelligence Lab

Research Groups

VUB AI Lab, UHasselt DSI, UA ADREM, VUB Data Lab