Accurate and timely early warning of seasonal influenza epidemics continues to pose a critical public health challenge. Innovative methodologies leveraging network data sources, such as internet search queries, serve as a valuable complement to traditional laboratory surveillance in terms of timeliness. This study aimed to develop and validate a deep-learning model for influenza prediction based on multi-source data. Using etiological surveillance data of influenza viruses among influenza-like illness (ILI) cases in four megacities (Beijing, Tianjin, Shanghai, and Shenzhen) during 2013 and 2018, we developed prediction models based on weekly Baidu index data and meteorological indicators. A long short-term memory (LSTM) model was compared against three machine learning algorithms. The optimal model was used for weekly forecasting of influenza activity. The LSTM model exhibited superior performance (maximum R2 : 0.80‒0.94 across cities) when compared to the three machine-learning models (maximum R2: 0.73‒0.82), and effectively predicted the weekly positive detection rate of influenza viruses with a lead time of 1‒3 weeks for the four megacities. Its accuracy was robustly maintained in medium-term rolling forecasts (14‒35 weeks) across all four megacities. Although the significance of predictors varied geographically, the Baidu index for “Tamiflu” was consistently a dominant predictor in three megacities. This study validates the significant potential of integrating network big data and deep-learning algorithm for influenza surveillance. The developed LSTM model provides a streamlined and effective tool for the early detection and warning of seasonal influenza epidemics.
Shan et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: