This article is more than 5 years old. Information may no longer be current.
Google Flu Trends predictions inaccurate in Latin America
Click Here to Manage Email Alerts
Google Flu Trends produced substantial inaccuracies in predicting influenza activity in Latin America compared with FluNet, according to study data recently published in Clinical Infectious Diseases.
“Recent years have seen the development of several internet biosurveillance tools, which use population-level trends in Google and other internet search-engine queries about infectious diseases such as dengue, pertussis, influenza or norovirus to detect and predict epidemics,” Simon Pollett, MBBS, BMedSci, DTMH, FRACP, of the department of epidemiology and biostatistics, University of California at San Francisco, and Marie Bashir Institute for Infectious Diseases and Biosecurity, University of Sydney, NSW, Australia, and colleagues wrote. “Given rising internet access throughout Latin America, web-based surveillance tools like Google Flu Trends could provide real-time information about influenza activity, particularly in less affluent countries with limited traditional health care and laboratory-based surveillance.”
The researchers obtained weekly data on influenza-positive respiratory specimens in eight Latin American countries (Argentina, Bolivia, Brazil, Chile, Mexico, Paraguay, Peru and Uruguay) using FluNet, a biosurveillance tool operated by WHO, from January 2011 to December 2014. They then extracted Google-predicted influenza activity for the same period and analyzed the correlation between the two.
Pollett and colleagues reported “frequent prediction errors” by Google Flu Trends. In all countries during the study period, the correlation between Google Flu Trends predictions and FluNet observations was highly variable (r = –0.53 to 0.91). Google was most closely correlated with FluNet in Mexico, the researchers reported, followed by Uruguay, Argentina, Chile, Brazil, Peru, Bolivia and Paraguay. In general, Google Flu Trends showed a better performance in more temperate countries, Pollett and colleagues wrote.
“Beyond seasonal autocorrelation, Google Flu Trends’ better performance in the more temperate countries of the Americas could also be explained by higher internet access. Those nations with the poorest Google Flu Trends-FluNet correlations (Bolivia, Paraguay and Peru) also had the least population internet access compared to the other studied countries,” Pollett and colleagues wrote. “Our findings emphasize that caution should be used when interpreting the findings of Google-based digital influenza surveillance in Latin America. Our results also provide important lessons for the improvement of internet-based biosurveillance methods globally, including future versions of Google Flu Trends, for these and other low-middle income regions.”
In an accompanying editorial, Mauricio Santillana, PhD, of the computational health informatics program at Boston Children’s Hospital and Harvard Medical School, wrote that “the lessons that can be learned from [Pollett and colleagues’] evaluation are limited by well-documented improvements made to the original Google Flu Trends methodology in recent years.
“My intuition is that the implementation of these methodological improvements will change the evaluation results presented by Pollett and colleagues in an important way,” Santillana wrote. “Finally, methodologies that combine information from disparate data sources … to track and forecast flu have shown great promise in the field of digital disease and perhaps should be the next subject of research around the globe.” – by Andy Polhamus
Disclosure: The researchers report no relevant financial disclosures.
Click Here to Manage Email Alerts