Speaker
Description
Influenza is a seasonal infectious disease, and real-time forecasting of outbreaks is essential for effective public health responses. In Korea, influenza surveillance relies on two types of data. Influenza-Like Illness (ILI) data collected through sentinel surveillance reflects outbreak trends, whereas confirmed case data obtained through universal surveillance better represent the total number of patients but are not updated in real time. Previous studies applied different likelihood functions to universal and sentinel surveillance periods to estimate population-level incidence. We extend this approach to a statistical time-series model.
We propose a latent INGARCH model in which the latent process represents population-level influenza incidence while observations correspond to counts recorded under different surveillance systems. Because sentinel observations represent only a subset of cases, a scale mismatch arises between the input data and the model output. To address this issue, we use the Poisson splitting property to reconstruct population-level incidence from sentinel observations and use the reconstructed counts as model inputs through an input data scaling procedure.
The proposed framework enables estimation of population-level influenza incidence using sentinel surveillance data while accounting for scale differences between surveillance systems. By addressing this mismatch, the proposed model improves prediction accuracy compared with existing approaches.
