Real-time forecasting of infectious disease epidemics
JT Wu, BJ Cowling
School of Public Health, The University of Hong Kong
 
 
1. The validity and predictability of three epidemic models were evaluated: a hybrid-type homogeneous stochastic model, an age-structured variant of the previous model, and a power-law logistic model.
2. Reporting rates affect the interpretation of model parameters only but not the performance of parameter estimation or real-time epidemic forecasting.
3. Reliable and precise real-time epidemic forecasting is improbable during the early phase of an epidemic and unlikely to be robust until the epidemic has peaked, when using only epidemic curve data and any of the three models.
4. Robust real-time epidemic forecasting, if possible at all, requires other sources of epidemic data, such as seroprevalence, household transmission data, and phylogenetic data.
5. Epidemiologists and public health policymakers should be aware of these results when using models for real-time epidemic forecasting.