A variant-informed decision support system for tackling COVID-19: a transfer learning and multi-attribute...

Area of Science:

• Epidemiology and Public Health
• Data Science and Artificial Intelligence
• Health Policy and Management

Background:

• The COVID-19 pandemic's evolution, marked by variants of concern (VOCs), necessitated agile decision support for governments.
• Existing systems struggled to adapt to the dynamic nature of emerging viral strains.

Purpose of the Study:

• To introduce the Variant-Informed Decision Support System (VIDSS) for dynamic adaptation to specific VOC characteristics.
• To enhance forecasting accuracy for future variants by leveraging historical data and transfer learning.

Main Methods:

• Utilized multi-attribute decision-making (MADM) techniques to assess country performance based on past improvements and peer comparisons.
• Incorporated transfer learning from previous VOC forecast models to improve predictions for new variants.
• Employed K-fold cross-validation and SHAP plots to evaluate model accuracy and feature importance.

Main Results:

• The VIDSS framework demonstrated robust predictive accuracy, with neural networks significantly enhanced by transfer learning.
• The hybrid MADM approach provided insightful country-specific scores, identifying key factors influencing COVID-19 spread.
• Vaccination rates, ICU admissions, and hospitalizations were consistently critical features across different variants.

Conclusions:

• Leveraging historical variant data substantially improves the prediction of future variant impacts.
• VIDSS offers dynamic, data-driven decision support for policymakers to optimize pandemic strategies and resource allocation.
• The system provides crucial insights for navigating the complexities of the evolving COVID-19 pandemic.