2020 ESA Annual Meeting (August 3 - 6)

OOS 69 Abstract - Forecasting for public health: Forecasting challenges for vector-borne diseases

Tuesday, August 4, 2020: 3:30 PM
Michael Johansson, Centers for Disease Control and Prevention
Background/Question/Methods

Arboviral diseases like chikungunya, dengue, yellow fever, West Nile, and Zika are important public health problems globally and in the U.S. Timely interventions can prevent or control the adverse impacts of arbovirus epidemics on human health, but planning interventions is challenging because, despite many years of research, the timing and intensity of epidemics are difficult to predict. Since 2014, the Centers for Disease Control and Prevention (CDC) Epidemic Prediction Initiative (EPI), in collaboration with an expanding group of researchers and public health professionals, has been developing and evaluating real-world forecasting tools to support public health in the United States. To help improve arboviral epidemic forecasting, EPI has developed collaborative forecasting challenges for dengue, Aedes mosquitoes, and West Nile.

Results/Conclusions

These forecasting challenges have connected researchers and forecast users, identified forecasting targets that are important to public health, facilitated data sharing, improved forecast evaluation, driven further forecasting research, and led to advances in forecast communication. In 2019, for example, 20 teams submitted monthly forecasts to the real-time EPI forecasting platform (https://predict.cdc.gov) for the probability of collecting Ae. aegypti and Ae. albopictus mosquitos in 95 counties across 8 states. Forecast evaluation for the various diseases and vectors have highlighted key public health needs (e.g., better early season dengue prediction) and components that can help improve forecasting research (e.g., ensuring model calibration, avoiding overfitting). Critically, every challenge has shown that using multiple models, either by a single team or by combining forecasts from all participating teams, provides more robust predictions than individual forecasts. Open forecasting challenges thus not only increase the ability of teams to assess methods in a real-time public health setting, but directly enable the development of more reliable, ensemble forecasts that specifically leverage broad participation.