The RAPIDD ebola forecasting challenge: Synthesis and lessons learnt.

Area of Science:

• Epidemiology and Biostatistics
• Computational Biology and Bioinformatics
• Public Health and Infectious Disease Modeling

Background:

• Infectious disease forecasting is increasingly important for public health preparedness.
• Systematic comparisons of different forecasting model performances are limited.
• A synthetic forecasting challenge was designed, inspired by the 2014-2015 West African Ebola crisis.

Purpose of the Study:

• To systematically compare the predictive performance of diverse infectious disease modeling approaches.
• To evaluate model accuracy under varying data availability and intervention scenarios.
• To identify characteristics of high-performing forecasting models.

Main Methods:

• 16 international academic and US government teams participated in a synthetic forecasting challenge.
• 8 independent modeling approaches were compared using 140 epidemiological targets across 4 synthetic Ebola outbreaks.
• Predictions included weekly case incidences, outbreak size, and peak timing under simulated 'fog of war' conditions.

Main Results:

• Ensemble predictions, using a Bayesian average of 8 models, outperformed individual models for weekly case incidence.
• Model complexity did not correlate with prediction accuracy; simple, reactive models performed best for short-term forecasts.
• Prediction accuracy improved with data accuracy and availability; final size estimates were within 20% of the target by the epidemic peak.

Conclusions:

• Synthetic forecasting challenges provide valuable insights into model performance under controlled conditions.
• Ensemble modeling and simple, reactive models show promise for infectious disease forecasting.
• Regular 'peace time' forecasting challenges are recommended to improve collaboration and preparedness for future pandemics.