Timely and accurate epidemic surveillance is critical for informing public health responses and minimizing disease spread. Traditional surveillance systems often suffer from reporting delays, incomplete data, and lack of adaptability, hindering real-time situational awareness. This study develops a robust framework for real-time statistical modelling of infectious disease outbreaks, integrating time-series forecasting, spatio-temporal modelling, and Bayesian updating techniques. We propose a dynamic hierarchical model that fuses multiple data streams—such as syndromic reports, laboratory confirmations, mobility data, and digital surveillance signals—while accounting for delays and underreporting through latent variable correction. The model is designed for sequential updating, enabling near real-time estimation of key epidemiological parameters, including the effective reproduction number (Rₜ), case incidence, and hotspot detection. To ensure scalability and responsiveness, we implement online inference using particle filters and variational Bayes, combined with adaptive shrinkage for high-dimensional covariates. Simulation studies and empirical applications to COVID-19 and influenza datasets demonstrate the method’s capacity to produce accurate nowcasts and short-term forecasts, with uncertainty quantification suitable for decision support. This framework offers a statistically rigorous and operationally feasible tool for epidemic early warning and response optimization under uncertainty.

Keywords: Epidemic Surveillance, Real-Time Modeling, Hierarchical Models, Spatio-Temporal Models, Particle Filters