VacSIM: Learning effective strategies for COVID-19 vaccine distribution using reinforcement learning

Area of Science:

• Computational epidemiology
• Artificial Intelligence in Public Health

Background:

• Limited COVID-19 vaccine availability necessitates optimal distribution strategies.
• Inequitable access and geographically varied disease outbreaks complicate vaccine allocation.
• Existing allocation methods may not effectively address real-world complexities.

Purpose of the Study:

• To develop and evaluate VacSIM, a novel AI-driven pipeline for optimizing COVID-19 vaccine distribution.
• To compare VacSIM's performance against naive allocation strategies.
• To demonstrate the potential of AI in mitigating pandemic spread through efficient resource allocation.

Main Methods:

• Integration of Deep Reinforcement Learning (DRL) with a Contextual Bandits approach in the VacSIM pipeline.
• Simulation of vaccine allocation across five Indian states (Assam, Delhi, Jharkhand, Maharashtra, Nagaland).
• Evaluation against a baseline of distributing vaccines proportionally to COVID-19 incidence.

Main Results:

• VacSIM demonstrated the potential to prevent up to 9039 infections over 45 days.
• Significant increase in the efficacy of limiting COVID-19 spread compared to naive allocation.
• The VacSIM framework showed adaptability and potential for real-world implementation.

Conclusions:

• AI-powered optimal vaccine allocation, as demonstrated by VacSIM, can significantly enhance pandemic control.
• The VacSIM model offers a scalable and extensible solution for global vaccine distribution challenges.
• Open-sourcing VacSIM and its associated reinforcement learning environment promotes further research and application.