Weeks before Europe knew it had a COVID-19 problem, warning signs were already appearing on Twitter. Moreover, the locations of tweets referring to “pneumonia” and “dry cough” match the places where SARS-CoV-2 would be found to be most common a month or so later. The discovery represents a massive missed opportunity to have controlled the spread of the virus before it took hold, but also offers hope for preventing future pandemics.
Tracking search engine use of terms associated with colds and flu has been shown to be an effective way for health authorities to detect upsurges in respiratory diseases before conventional testing procedures catch up. There is even evidence reviews of scented candles have turned more negative because people with COVID-19 blame their inability to smell on the candle. Professor Massimo Riccaboni of IMT School for Advanced Studies thought social media might offer something similar. Riccaboni tested the idea by looking for evidence of posts before official diagnoses began, testing the question of whether SARS-CoV-2 was circulating in Europe well before news from Wuhan alerted health authorities to the problem.
Riccaboni searched Twitter for references to “pneumonia” (and its equivalent in six other languages) in December 2019 and January 2020, comparing it with data going back to December 2014 to serve as a control. In Scientific Reports, Riccaboni and co-authors reveal a dramatic surge in pneumonia-related tweets during that fateful winter. Although this is flu season in the Northern Hemisphere, the influenza outbreak that year was mild compared to the previous one. Nevertheless, pneumonia tweets were up, even after removing those discussing news reports.
The same pattern was seen for “dry cough”, the significance of which would have escaped people at the time, but turned out to be a warning of one of COVID-19's more distinctive symptoms.
If Riccaboni's observations were limited to timing, there would be many alternative explanations. For example, as soon as news seeped out of China about a new respiratory disease, people with flu might have become more inclined to talk about their symptoms online.
However, when the team looked at the regions where more than 13,000 tweets including the term pneumonia originated, they found the surge was concentrated in places where COVID-19 was first found to be common a month or two later. The fact there was a surge in Lombardy and Madrid, but not in parts of Europe that didn't get hit by the pandemic until substantially later, is very unlikely to be a coincidence in the authors' view.
"Our study adds on to the existing evidence that social media can be a useful tool of epidemiological surveillance. They can help intercept the first signs of a new disease, before it proliferates undetected, and also track its spread," Riccaboni said in a statement.
The paper proposes the establishment of an “integrated digital surveillance system” to identify signs of infectious diseases through social media and web searches and pinpoint cities and regions in danger.
The benefits would be particularly large for fast-moving diseases like COVID-19, where interventions to reduce transmission made just a few days earlier would have a dramatic effect on the size of the eventual peak.
