A simulation-deep reinforcement learning (SiRL) approach for epidemic control optimization

Area of Science:

• Epidemiology
• Artificial Intelligence
• Public Health Policy

Background:

• Government decision-making significantly impacts public health outcomes during epidemics.
• Balancing disease control with economic stability presents a major challenge in epidemic planning.
• The COVID-19 pandemic highlighted the need for optimized intervention and vaccination strategies.

Purpose of the Study:

• To develop a novel Simulation-Deep Reinforcement Learning (SiRL) model for epidemic control planning.
• To compare different vaccination strategies (e.g., age-based vs. random) for prioritizing vaccine distribution.
• To evaluate the effectiveness of various intervention strategies in controlling epidemic spread and reducing fatalities.

Main Methods:

• Development of a Simulation-Deep Reinforcement Learning (SiRL) framework combining agent-based simulation with a deep reinforcement learning agent.
• The DRL agent learns to enforce interventions within the simulation environment based on epidemic status.
• The model was validated using COVID-19 data from Kansas and compared against historical government interventions.

Main Results:

• The DRL agent successfully learned effective epidemic control strategies tailored to specific situations.
• A random vaccination strategy, prioritizing super-spreaders, could reduce infections by 32% with a 4% increase in deaths.
• Combining targeted quarantines with random vaccination significantly reduced mortality compared to age-based strategies.

Conclusions:

• The SiRL model offers a flexible and effective approach for optimizing epidemic control strategies.
• Random vaccination prioritizing super-spreaders and targeted quarantines show potential for reducing epidemic impact.
• The findings suggest that current epidemic control strategies could be further optimized for better outcomes.