Real-time responses to epidemics: A Reinforcement-Learning approach.

Area of Science:

• Epidemiology
• Control Theory
• Computational Science

Background:

• Open-loop optimal control for epidemics assumes perfect knowledge, lacking real-time adaptability.
• Existing methods fail to integrate evolving epidemic data for policy adjustments.
• This limitation hinders effective pandemic response and preparedness.

Purpose of the Study:

• To develop a closed-loop method for real-time social distancing policy management during epidemics.
• To create a Reinforcement Learning (RL)-based decision-support framework for adaptive epidemic control.
• To balance public health outcomes with indirect socioeconomic costs of interventions.

Main Methods:

• Proposed a novel closed-loop Reinforcement Learning (RL) framework for adaptive social distancing.
• Utilized a COVID-19 model incorporating asymptomatic transmission, healthcare capacity, and quarantine.
• Evaluated the framework's performance under varying indirect cost priorities during a severe outbreak.

Main Results:

• The closed-loop RL framework demonstrated adaptive management of social distancing policies.
• Optimal real-time responses varied, leading to different policy strategies (elimination, suppression, mitigation).
• Policy outcomes were contingent on the prioritization of indirect costs by authorities.

Conclusions:

• Closed-loop adaptability enables dynamic policy adjustments, optimizing epidemic control.
• The RL framework supports informed, timely decision-making for governments and health agencies.
• This approach enhances preparedness and response strategies for current and future pandemics.