Predicting the dynamics of norovirus infection using time series models

Abstract

Introduction. Norovirus infection (NI) is the most prevalent cause of acute gastroenteritis and outbreaks in semi-closed settings. Forecasting of NI may improve situational awareness and control measures.The aim of the study is to evaluate accuracy of time-series models for forecasting of norovirus incidence (on Sverdlovsk region dataset).Materials and methods. Simple ARIMA time-series models was chosen to forecast NI incidence via regression on its own lagged values. Dataset including passive surveillance monthly reports for Sverdlovsk region was used. All models were trained on data for 2015−2018 and tested on data for 2019. Models were benchmarked using Akaike information criterion (AIC) and mean absolute percentage error (MAPE).Results and discussion. NI incidence in Sverdlovsk raised in 2015-2018 with strong winter-spring seasonality. The time-series incidence data was stationary. Nine significant models were found and the most accurate model was SARIMA (1,0,0)(0,0,1). Despite its accuracy on 2019 test sample, forecast on COVID-19 pandemic period was failed. It was supposed that including additional regressors (climate and herd immunity data) and choosing of more robust time-series models may improve forecasting accuracy.Conclusion. ARIMA time-series models (especially SARIMA) suitable to forecast future incidence of NI in Sverdlovsk region. Additional investigations in terms of possible regressors and improved model robustness are needed.