Forecasting influenza hospital admissions within English sub-regions using hierarchical generalised additive models

• DOI: 10.48550/arxiv.2302.11904
• Publication Date: 2023-01-01
• Authors: Jonathon Mellor; Rachel Christie; Christopher E. Ov...; Robert S. Paton; Rhianna Leslie; Maria Tang; Sarah R Deeny; Thomas R. Ward
• Keywords: Generalized additive model; Smoothing; Exponential Smoothing

Background: Seasonal influenza causes a substantial burden on healthcare services over the winter period when these systems are already under pressure. Policies during COVID-19 pandemic supressed transmission of season influenza, making timing and magnitude potential resurgence difficult to predict. Methods: We developed hierarchical generalised additive model (GAM) for short-term forecasting hospital admissions with positive test virus sub-regionally across England. The incorporates multi-level structure spatio-temporal splines, weekly seasonality, spatial correlation. Using multiple performance metrics including interval score, coverage, bias, median absolute error, predictive is evaluated 2022/23 seasonal wave. Performance measured against an autoregressive integrated moving average (ARIMA) time series model. Results: GAM method outperformed ARIMA scoring rules at both high low-level geographies, different phases epidemic wave turning point. 14-day forecast horizon was comparable in error 7 days. found be most sensitive flexibility smoothing function that measures national trend. Interpretation: This study introduces novel approach using hierarchical, spatial, temporal components. data-driven practical deploy information realistically available prediction, addressing key limitations approaches. used operational planning by UK Health Security Agency National Service