Forecasting of influenza activities using multi-stream surveillance data in Hong Kong

Abstract

Background: Influenza is a severe public health problem that seriously affects the human health system and even life, attracting more and more researchers to study its transmission dynamics. Despite many efforts that have been paid to investigate influenza transmission in temperate regions, the transmission in the (sub-) tropical region is seldom studied since it often presents a complex irregular multi-wave phenomenon and is easily affected by seasonality factors, ambient climates, and human behaviours, etc. Methods: We forecasted the influenza activity in Hong Kong, a subtropic city, by considering multiple factors based on multi-stream surveillance data, i.e., absolute humidity, temperature, and school holidays. We first built five mechanism models to interpret influenza transmission by considering different combinations of multi-stream surveillance data. Then we used a temporal cross-validation approach to perform retrospective forecasting to identify the forecasting performance of different models. By introducing a factor to denote the proportion between the counterfactual and the COVID-19 NPIs-based transmission rates, we compared the peak timing and peak magnitude and the effective reproduction number to evaluate the impact of COVID-19 NPIs on influenza transmission. Findings: We forecasted two peaks of influenza activities in the 2020-21 influenza season based on counterfactual forecasting, which were estimated to occur in February for the winter season and July for the summer season. We evaluated that the model considering absolute humidity and temperature had a better forecast performance than other models. Following the strict COVID-19 NPIs, we estimated the influenza incidence with a 14.01% (95%CrI 2.36%-23.38%) reduction for the first week, with a 66.91% (95% CrI, 62.30%–70.49%) reduction for the first 5 weeks. The COVID-19 NPIs were estimated to reduce influence incidences by 65.90% (95%CrI 56.14%-71.20%) for the whole winter period and 99.99% for the summer period. Moreover, we estimated the transmission rate to reduce by 48.23% (95%CrI: 41.26%-56.60%), and the effective reproduction number was even reduced to less than 1 during the strict NPIs period. Conclusions: The factors of multi-stream surveillance data were considered to model and forecast influenza activities for a subtropical city Hong Kong. Our forecasts suggest that the COVID-19 NPIs would significantly suppress influenza transmission.