Abstract
BACKGROUND: Acute respiratory infections (ARIs) in young children pose a significant global health challenge, leading to high rates of illness and death. They are estimated to be the fourth leading cause of mortality worldwide, particularly impacting children under five. This study aimed to identify the most effective time series model(s) for forecasting the epidemiological season burden of ARIs for the current 2023/2024 period in Italy. METHODS: Data on the burden of ARIs' in children aged 0-14 years were retrieved from Pedianet, an Italian paediatric primary care database which includes over 200 family paediatricians. We analysed monthly incidence rates of ARIs from September 2010 to September 2023, following the typical seasonal pattern of these infections. Several forecasting models were compared to predict the future burden of ARI: Error, Trend, Seasonality (ETS); Seasonal Auto-Regressive Integrated Moving Average (SARIMA); Unobserved Component Model (UCM); and Trigonometric, Box Cox, ARMA errors, Trend, Seasonal (TBATS). We evaluated each model's accuracy by examining the residuals and the Mean Absolute Percentage Error (MAPE). The period between March 2020 and February 2022 was forecasted to represent the normal trend without COVID-19. Model parameters were estimated using the in-sample and out-of-sample approach. RESULTS: The analysis included data from over 1.4 million cases of ARIs retrieved in children aged 0-14 years. The ETS model was implemented to predict the pandemic period. Overall, our findings suggest that exponential smoothing models as ETS (MAPE = 6.85) and TBATS (MAPE = 6.87) were most effective in predicting future trends in monthly ARIs' burden compared to other methods (i.e., UCM MAPE = 11.08, and SARIMA MAPE = 25.33). CONCLUSIONS: These findings suggest that exponential smoothing models are preferable for forecasting pediatric ARIs' burden trends in Italy. However, epidemiological data from the ongoing season are crucial for understanding whether residual pandemic effects continue affecting respiratory infection patterns.