Seasonal Dynamics and Forecasting of Toscana Virus Infections in Italy, 2016–2027: A Modelling Study

Abstract

Background: Toscana virus is a sandfly-borne phlebovirus and a relevant cause of central nervous system infections in Mediterranean countries, including Italy. Its incidence exhibits strong seasonal patterns, suggesting suitability for time series modelling approaches. This study aimed to compare statistical and time series models for describing and forecasting Toscana virus incidence in Italy. Methods: Monthly confirmed autochthonous Toscana virus cases in Italy (2016–2025) were obtained from national surveillance data. The series was split into a training period (2016–2023) and a validation period (2024–2025). Three models were evaluated: SARIMA(1,0,1)(1,0,0)12, Poisson regression with harmonic seasonal terms, and Negative Binomial regression accounting for overdispersion. Predictive performance was assessed using MAE and RMSE. The best model was re-estimated on the full dataset to generate forecasts for 2026–2027. Results: All models captured strong seasonality and temporal dependence. The SARIMA model showed the best predictive performance (MAE = 4.75; RMSE = 7.63), outperforming Poisson (MAE = 4.99; RMSE = 11.59) and Negative Binomial models (MAE = 5.39; RMSE = 13.57). The Negative Binomial model confirmed significant overdispersion (α = 0.143, p < 0.001) but did not improve forecasting accuracy. Forecasts indicated persistence of seasonal oscillations through 2027. Conclusions: Toscana virus incidence in Italy follows stable seasonal dynamics. SARIMA provided the most accurate forecasts and is the preferred model for short-term prediction.