A spatial EHR- and wastewater-informed modeling framework for respiratory virus forecasting under sparse and missing data conditions

Heterogeneous surveillance data, including electronic health record (EHR) and wastewater, provide complementary information for monitoring infectious disease transmission and healthcare burden. However, surveillance systems often exhibit overlapping spatial gaps, particularly in underserved communities where clinical and environmental data are sparse or unavailable, limiting the performance of existing forecasting methods. Here, we develop a spatial Bayesian renewal framework that integrates heterogeneous surveillance data through a shared latent infection process and mobility-informed spatial interactions, enabling robust disease burden forecasting under incomplete surveillance. We apply the framework to forecast county-level hospital admissions associated with three major respiratory viruses: influenza, SARS-CoV-2 (COVID-19), and respiratory syncytial virus (RSV), across South Carolina. In rolling four-week forecasting horizons, the proposed framework consistently outperforms non-spatial approaches and maintains strong predictive performance even in counties lacking direct wastewater or insufficient EHR observations. By borrowing information across spatially connected regions, the model leverages surveillance-rich communities to improve forecasts in surveillance-poor areas. These results demonstrate that integrating complementary surveillance streams within a spatial Bayesian framework provides a scalable and generalizable approach for infectious disease forecasting and supports public health decision-making where surveillance infrastructure is incomplete.