Influenza Epidemic Trend Surveillance and Prediction Based on Search Engine Data: Deep Learning Model Study

Area of Science:

• Epidemiology
• Public Health
• Digital Health

Background:

• Influenza outbreaks present a significant global public health challenge.
• Traditional surveillance methods often lack accuracy and timeliness in predicting outbreaks.
• Emerging infectious diseases like influenza and COVID-19 require advanced surveillance tools, especially when official data lags.

Purpose of the Study:

• To develop an influenza outbreak predictive model using Baidu search data and virological surveillance.
• To enhance early detection and preparedness for influenza in China.
• To provide evidence for supplementing modern epidemic surveillance with digital data.

Main Methods:

• Collected virological data (National Influenza Surveillance Network) and Baidu search query data (Jan 2011-Jul 2018).
• Analyzed correlations between influenza-related search terms and influenza-positive rates.
• Developed a predictive model using gated recurrent units and attention mechanisms for forecasting influenza trends.

Main Results:

• Specific Baidu search terms strongly correlated with influenza-positive rates in China.
• Search terms related to influenza prevention and symptoms showed significant lag correlations (1.4-8.0 days).
• Baidu data predicted influenza rates 14-22 days in advance in southern China but caused interference in northern China.

Conclusions:

• Integrating web-based data sources with traditional surveillance can improve early detection of influenza outbreaks.
• Caution is advised when supplementing modern surveillance with search engine data.
• Further research is needed to optimize search terms for regional and linguistic variations in digital epidemiology.