The correlation between Google Trends and conventional influenza surveillance data in Taiwan

Abstract

Purpose: Previous studies have showed that Google Trends of influenza-related search data were highly correlated with conventional influenza-like illness (ILI) surveillance data. A timely and effectively surveillance method for influenza monitoring and early alert is essential for Taiwan, where only conventional surveillance is currently available. The purpose of this study was to evaluate the feasibility of using Google Trends for influenza surveillance in Taiwan.

Method: This study investigated the correlation between 208 weeks of ILI and virologic influenza surveillance data and Google Trends data for influenza-related search terms (H1N1, fever, sore throat, cough, flu, influenza A, type A influenza, influenza B and cold) in Taiwan from Jan 3rd 2010 to Dec 28th 2013. Pearson correlation coefficients and lag correlations of maximal 2 weeks were calculated.

Results: In terms of the correlation between Google Trends data and ILI surveillance data in Taiwan during the study period, the search term flu showed the strongest correlation with both outpatient and emergency room ILI visit rate (r = 0.82 and r= 0.70, p< 0.001). Except for fever and sore throat, the correlation coefficients between Google Trends data and ILI surveillance data for all keywords were statistically significant (r=0.16, p= 0.02 to 0.82, p< 0.001). When compared with virologic surveillance rate, Google Trends for influenza B had the highest correlation in 2012 (r= 0.91, p<0.001). In addition, Google Trends data for all keywords had strong correlations with the outpatient and emergency room ILI visit data in 2011, and the correlation coefficients ranged from 0.25 to 0.90 (p< 0.001). The lag correlation analysis indicated that most of keywords had the maximum correlation coefficients at a lag time of 0 weeks.

Conclusion: The strong correlations between Google Trends data and CDC ILI surveillance data in Taiwan suggested that Google Trends data fitted well with the conventional ILI surveillance data. The correlation between ILI surveillance data and Google Trends data was higher than that between Google Trends and virologic surveillance data. The lag time correlation analysis did not indicate a significant preceding for most of search terms in Google Trends than CDC surveillance data, which suggested that Google Trends for the search terms used in this study did not show an advantage on the monitoring of influenza outbreak when compared with the conventional surveillance data in Taiwan. In summary, this study found that Google Trends in traditional Chinese could be applied as a complementary tool for influenza surveillance in Taiwan, especially for the ILI surveillance. Further study should develop more representative keywords and predictive model to estimate influenza activities as well as assist in detecting early alert of influenza outbreaks.