Adaptive multi-model ensembles for improved epidemic projections and decision support.

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Multi-model ensemble projections are standard in infectious disease modeling but are resource-intensive.
• Current methods limit refining projections or updating scenarios with new data.
• Coordinated efforts require significant computational power and research team input.

Purpose of the Study:

• To introduce an adaptive ensemble approach for infectious disease modeling.
• To dynamically select individual model trajectories based on observed data.
• To improve the efficiency and accuracy of long-term ensemble projections.

Main Methods:

• Developed an adaptive ensemble method analogous to multi-model particle filtering.
• Dynamically selected individual model trajectories based on observed data.
• Validated the approach using U.S. Flu Scenario Modeling Hub (SMH) projections for influenza hospitalizations.

Main Results:

• The adaptive ensemble demonstrated improved predictive accuracy compared to the original SMH ensemble.
• The approach successfully identified the most plausible epidemic scenarios for U.S. influenza seasons.
• Retrospective analysis showed superior short-term forecasting performance against a baseline model.

Conclusions:

• The adaptive ensemble approach offers an efficient strategy to enhance multi-model epidemic projections.
• It provides real-time support for modeling teams, public health authorities, and decision-makers.
• The method shows potential for real-time collaborative forecasting challenges like CDC's FluSight.