The metrics expert Francesc Pujol, a professor at the Faculty of Economics at the University of Navarra, has investigated this correlation, which, with regard to infectious diseases, is not new. As explained in its blog SM Reputation Metrics, before this pandemic a close relationship was confirmed between the results of Google searches for the word “flu” and the official count of flu positives.
In fact, in January of last year, before the outbreak of the coronavirus, Pujol predicted the flu “spike” two weeks before it happened. The economics professor, who also works as a COVID-19 statistics analyst on Twitter (his account is @NewsReputation) is not an epidemiologist nor does he have a crystal ball. It was based on previous scientific studies – such as Eysenbach, 2006; Carneiro and Mylonakis, 2009; Valdivia et al, 2010, and an article in “Nature” from 2013 – to make their prediction.
“The similarity between Google searches and the evolution of flu cases in various European countries is so striking that in some cases the two series merge”
The Covid-19 pandemic, in addition to wiping the flu off the map, changed the dynamics: neither the term “coronavirus” nor “Covid” served as indicators in the first wave, from March to June. “This is because in March 2020 the coronavirus was not just a contagious virus or a pandemic like the flu was: it was a systemic crisis, a collapse in all areas and it was something totally new,” explains the expert.
The end of the first wave in June also marked the end of the exceptionality in the pandemic, and from there “Covid symptoms” (sic) began to be an indicator for monitoring the incidence of cases through the Google Trends portal ( Google trends). Note that “symptoms” is written without an accent mark, as most Internet users write it that way.
The coupling between both curves (searches and diagnosed cases) did not occur until October. According to Pujol, it was like that because in summer “the attention in Google soared much more than the speed of the cases “. The correlation coefficient between both variables was 0.6 between June 2020 and April 2021, it rose to 0.85 from the beginning of October and 0.92 (the curves overlap) from mid-November.
Pujol adds that Google Trends not only reflects, but anticipates the start of the new waves, since “Google searches are ahead of the notification of positives”. In economic terms this is called nowcasting, a technique based on observation, remote sensing and nowcasting.
A study published in February in “Nature Digital Medicine” by Vasileios Lampos, from University College London, and other authors, proposed using online searches to track Covid-19. The work, carried out in the United Kingdom, used models that, based on searches for specific symptoms – for example, loss of smell -, predicted coronavirus cases and deaths 16 and 22 days in advance, respectively.
In the case of the flu, the prediction worked so well years ago that the world’s most popular search engine created the Google Flu Trends service in 2008. However, he abandoned it in 2015 after he had a 140% error in 2013 predicting the peak of the flu in the United States. A later study in “Science” attributed it to a flaw in the algorithm. And it is that the interpretation of big data is not always correct.