Forecasting Influenza Epidemics in Hong Kong

Area of Science:

• Epidemiology
• Mathematical Modeling
• Infectious Disease Dynamics

Background:

• Seasonal influenza forecasting is established for temperate regions.
• Subtropical and tropical regions face challenges due to year-round, irregular influenza epidemics.

Purpose of the Study:

• Develop and validate influenza forecast systems for non-seasonal epidemics in subtropical/tropical regions.
• Improve real-time prediction of epidemic peak timing and magnitude.

Main Methods:

• Utilized ensemble adjustment Kalman filter and modified particle filter with a susceptible-infected-recovered (SIR) model.
• Retrospectively forecasted influenza epidemics in Hong Kong (1998-2013).
• Analyzed forecasts for individual strains (H1N1, H3N2, B) and aggregate epidemics.

Main Results:

• Accurate forecasts achieved for peak timing and magnitude across 44 individual and 19 aggregate epidemics.
• Average forecast accuracies: 37% at 1-3 week leads, 50-51% at 0 lead.
• Forecast accuracy improved with decreased ensemble spread, reaching up to 93% (timing) and 89% (magnitude) for specific strains.

Conclusions:

• Accurate influenza forecasts are feasible at least three weeks in advance for subtropical and tropical regions.
• Developed model-filter systems offer a valuable tool for public health surveillance and intervention planning.
• Findings highlight the potential for real-time epidemic prediction in challenging epidemiological settings.