An open challenge to advance probabilistic forecasting for dengue epidemics

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Despite numerous advances in infectious disease dynamics research, practical application of forecasting tools remains limited.
• Existing tools often fail to meet public health needs, lack probabilistic outputs, or have not been rigorously validated.
• Dengue fever represents a significant global health challenge, necessitating improved forecasting capabilities.

Purpose of the Study:

• To assess the skill of probabilistic forecasts for seasonal dengue epidemics.
• To evaluate various modeling approaches and data utilization in forecasting.
• To identify key factors for improving the real-time application of infectious disease forecasts.

Main Methods:

• An open collaborative forecasting challenge was established involving sixteen teams.
• Teams generated probabilistic forecasts for three epidemiological targets (peak incidence, timing, and total incidence) over eight dengue seasons.
• Forecasts were generated for two locations: Iquitos, Peru, and San Juan, Puerto Rico.

Main Results:

• Forecast skill varied significantly across different teams, targets, and seasons.
• While midseason forecasts showed good situational awareness, early-season skill was low.
• Ensemble forecasts, combining multiple models or teams, consistently demonstrated higher skill than individual model forecasts.
• Models incorporating biological data and mechanisms did not necessarily yield higher average forecast skill.

Conclusions:

• Improving forecast skill for seasonal epidemics like dengue requires leveraging insights from collaborative challenges and ensemble approaches.
• The developed framework, emphasizing public health integration, standardized data, and open participation, can advance infectious disease forecasting.
• Enhanced forecast skill is crucial for real-time epidemic preparedness and response efforts globally.