Find this paper

Abstract

We present results from the second season of Influcast, a multi-model collaborative forecasting hub focused on influenza in Italy. During the 2024/25 winter season, Influcast collected one- to four-week-ahead probabilistic forecasts of influenza-like illness (ILI) incidence alongside influenza A and B ILI+ incidence signals. New ILI+ targets were constructed integrating syndromic surveillance data with virological detections collected weekly by the Italian National Institute of Health. Forecasts were submitted by six independent models (including compartmental, metapopulation, and statistical approaches) and combined into an ensemble. Ensemble forecasts for ILI+ consistently outperformed both the baseline (a naive persistence model) and most individual models in terms of Weighted Interval Score (WIS), Absolute Error (AE), and prediction coverage. Importantly, ensemble ILI+ forecasts achieved significantly lower WIS and AE ratios (i.e., ratio between the ensemble and the baseline models) and improved calibration compared to ILI forecasts. Our findings support the integration of virological surveillance data in forecasting target definition to improve the reliability of epidemic forecasts and strengthen their utility for situational awareness, communication, and targeted intervention.

Posts about this paper

No posts about this paper yet. Mention it with @ in a post to start the conversation.