April 15, 2024

I, Science

The science magazine of Imperial College

Complaining about your symptoms on social media might be more useful than you think.

“This flu is horrendous. Can’t breathe, can’t sleep or eat. Muscles ache, hot fever. Should have gotten the shot. Time for a movie marathon.” Tweet… sent. Your 140 characters of misery are released into the world. Your friends, on their smartphones, scroll past with an eye roll, but somewhere in Boston, Massachusetts, a computer has found your tweet, singled it out from countless others, and is analysing the time and location data. The computer scours Twitter for messages like yours, searching for keywords that betray illness; it’s trying to predict where the flu will strike next.

Disease forecasts are crucial for managing epidemics, like the Ebola outbreak of 2014. Computer programs called ‘models’ calculate the trajectory of an illness based on what we know about it, like where and when it started, and how infectious it is. “Paradoxically, although the seasonal flu is the disease that we experience most, it’s one of the most difficult to model,” says Professor Alessandro Vespignani, from Northeastern University, in Boston. Because the flu is so widespread, scientists often can’t tell when or where each epidemic starts.

Tracking seasonal flu is so problematic that the Centre for Disease Control and Prevention (CDC), in the US, runs an annual “Predict the Influenza Season Challenge”. Their prize encourages researchers to design accurate models that can aid in planning vaccination campaigns and targeting resources. Government flu surveillance relies on doctors’ reports, but the information they provide is often too little, too late.

“If you get the flu, you may stay in bed for two days, but you probably won’t go to the doctor”, says Vespignani, whose team has taken part in the CDC’s challenge since it started in 2013. “However, you might write a message on Twitter to tell your friends that you feel miserable.” Researchers like Vespignani are turning to social media because the elusive initial conditions of each flu epidemic can often be traced through our online moaning. “Tweets won’t tell you how many people are sick, but they can tell you the relative impact of flu in different places. This is all we need.”

Competitors in the challenge are getting very creative with their data gathering: other teams have analysed Google and Wikipedia searches, work absences and even restaurant cancellations as measures of the flu’s impact. Vespignani’s Twitter-based model can forecast the flu’s trajectory up to six weeks in advance, and it knows which week it will hit hardest.

So, next time you catch the flu, don’t be shy, you can whine—just make sure you do it online.

Bruno Martin is studying for an MSc in Science Communication

Banner image: couple with flu