Evaluating COVID-19 vaccine allocation policies using Bayesian m-top exploration.

Area of Science:

• Epidemiology
• Computational Biology
• Public Health Policy

Background:

• Individual-based epidemiological models are crucial for simulating fine-grained public health interventions like vaccine allocation.
• These models are computationally intensive, necessitating efficient strategies for identifying optimal policies within budget constraints.
• Communicating decision uncertainty to policymakers is vital for high-impact public health strategies, aligning with Bayesian approaches.

Purpose of the Study:

• To develop and evaluate a novel technique for optimizing vaccine allocation strategies using a multi-armed bandit framework.
• To integrate a Bayesian anytime m-top exploration algorithm for efficient identification and uncertainty quantification of top-performing policies.
• To provide policy advisors with flexible computational options and confidence levels for decision-making.

Main Methods:

• Utilized a multi-armed bandit framework where each unique vaccine allocation policy was defined as an 'arm'.
• Implemented a Bayesian anytime m-top exploration algorithm to identify the 'm' policies with the highest expected utility.
• Applied the method to the Belgian COVID-19 epidemic using the STRIDE individual-based model to minimize infections and hospitalizations.

Main Results:

• The proposed method efficiently identified the top 'm' vaccine allocation policies.
• Analysis revealed clear trends in prioritized age groups and vaccine types within the optimal policies.
• Vaccination policies were explored under various social contact reduction scenarios and vaccine uptake proportions.

Conclusions:

• The developed technique effectively identifies optimal vaccine allocation strategies and quantifies associated uncertainties.
• Findings offer valuable insights for designing future vaccination campaigns, highlighting age and vaccine type prioritization.
• Vaccine uptake proportion demonstrated a limited impact on the overall optimality of the identified strategies.