Optimising pandemic response through vaccination strategies using neural networks

Area of Science:

• Epidemiology
• Public Health
• Computational Science

Background:

• Traditional epidemic risk assessments struggle to balance health outcomes with economic constraints.
• Developing effective vaccination strategies requires considering both disease dynamics and economic impacts.

Purpose of the Study:

• To present a data-driven decision support tool for optimizing vaccination strategies.
• To minimize disease spread and economic losses during epidemics.
• To inform policymakers for better pandemic preparedness.

Main Methods:

• Developed a stochastic SVEI3RD compartmental model to capture epidemic dynamics and stratify infections by severity.
• Formulated an optimal control problem to minimize pandemic-related expenditures.
• Employed Physics-Informed Neural Networks (PINNs) to overcome computational challenges in high-dimensional stochastic control problems.

Main Results:

• The optimal vaccination strategy reduced cumulative hospital-days by 85.3% and deaths by 84.4% compared to no vaccination.
• Total pandemic costs were reduced by 22.3% with the optimal strategy.
• The optimized strategy outperformed the actual government rollout by 4-6 percentage points.

Conclusions:

• The proposed economic-epidemiological framework effectively balances health and economic considerations in epidemic control.
• AI-driven tools can significantly enhance pandemic preparedness and response strategies.
• Policymakers can utilize this framework for continuous strategy updates to minimize costs.