A hierarchical model for forecasting seasonal respiratory epidemics across pathogens and regions

Since 2021, the seasonal tripledemic composed of COVID-19, influenza, and respiratory syncytial virus (RSV) has threatened healthcare capacity globally. Short-term forecasts can provide public health officials and healthcare leaders time to effectively respond to epidemics, but many forecast approaches are bespoke to specific diseases or localities. We present a Bayesian hierarchical forecast model that flexibly accounts for spatial and seasonal transmission dynamics and test its performance on hospital admissions of COVID-19, influenza, and RSV in the United States over two years. We find the model outcompetes a naive baseline model by 42%, 44%, and 41% for COVID-19, influenza, and RSV, respectively, and it was the top individual forecast model in the 2023-2024 and 2025-2026 CDC FluSight forecast challenges. We use our model’s inferences to quantify the timing and shape for influenza and RSV epidemics and the biannual seasonality of COVID-19. Additionally, we estimate regional disease burden differences across the country, finding higher burden in the South and lower burden in the West and Northeast. Given its flexible nature and robust performance, our model provides a straightforward way to expand forecasting to additional regions and for other seasonal diseases such as Dengue virus or malaria. Short-term forecasts of respiratory diseases can help hospitals and public-health agencies prepare for increases in patients. However, many forecasting models are designed for only one disease or location and therefore cannot easily be used elsewhere. We developed a model that accounts for differences in the timing and severity of seasonal outbreaks across locations. We tested it by forecasting US hospital admissions caused by COVID-19, influenza, and respiratory syncytial virus (RSV). Its forecasts were more than 40% better than simple comparison forecasts, and it was the best-performing individual model in two seasons of the US Centers for Disease Control and Prevention’s FluSight forecasting initiative. The model also revealed differences in seasonal timing among the three viruses and in disease burden across US regions. Given its adaptability, this approach provides a scalable framework for expanding outbreak forecasting to other regions and diseases, enhancing global epidemic preparedness.