Towards forecasting flu dynamics using a regionalized state space model

The emergence of H1N1 in 2009 and a subsequent pandemic onset illustrated the importance of developing effective models with useful predictive capabilities for infectious diseases. The early identification of epidemic peaks can help the authorities to strategize effective anti-epidemic plans. In this regard, we propose a particle filter approach using the Susceptible-Exposed-Infected and Recovered (SEIR) epidemic model. The epidemic model was integrated with an observation model characterizing real-time influenza-like illnesses (ILI) data originating from general practice/family doctors (GPFDs) located in Singapore. The systematic resampling algorithm used in our approach resulted in better parameter estimates and the approach was able to predict the ILI peak registered around Day 40. However, the current approach suffers a serious limitation in its sensitivity towards the initial prior distribution assumed. The shortcomings of the current approach can be overcome by using a regionalized approach. Future efforts will be dedicated towards initializing the prior based on region-specific socio-demographic variables. Such regionalized models can provide good insight concerning the execution of efficient region-specific anti-epidemic plans for preventing future pandemics.