Categorical short-term forecasts of respiratory pathogens in Europe.

Real-time near-term forecasts of seasonal respiratory infections are currently available from the European Centre for Disease Prevention and Control through the Respicast forecast hub. In this study, we explored the generation of probabilistic forecasts of short-term change in influenza-like illness (ILI) and acute respiratory illness (ARI) disease activity for over twenty countries in Europe. Building on publicly available datasets of respiratory disease activity in Europe, we describe a forecasting pipeline that includes: a) a flexible method to define categories of interest, in relation to disease activity seen in past seasons; and b) an approach based on smoothing splines to translate real-time probabilistic forecasts of continuous targets (available in Respicast) to forecasts of categorical targets. We applied the methods to forecast ILI/ARI consultation rates in Europe, and influenza-associated hospitalization in the United States during the 2024/25 season. For forecasts in the US, results suggest that the simple approximation method can match the skill of at least a third of the models with which a direct comparison was possible. Skill of the approximated forecasts of nearly half of the models in Respicast were better than the benchmark's when predicting ILI rate changes. Approximation of the hub-ensemble's continuous forecasts yielded an improvement of 24% over benchmark; further improvement through combining individual model forecasts was noted. ARI rate changes were harder to predict than ILI, and the value of model combination was less clear. Besides forecast skill, heterogeneity in models' calibration was also observed. We believe these categorical forecasts can provide a plausibly more intuitive indicator of predicted change in respiratory disease burden, and a useful addition to routine surveillance reports. The approaches are flexible, computationally inexpensive and can help realize an addon value from extant Respicast infrastructure.