Predicting Coronavirus Outbreaks Using Google Searches for Gastrointestinal Problems

Google search results could help predict coronavirus outbreaks.

Google search results could help predict coronavirus outbreaks. Shutterstock

 

Connecting state and local government leaders

New research found that Google searches for symptoms associated with the coronavirus might be able to predict outbreaks before positive test results confirm them.

When people don’t feel well, many turn to search engines for answers. This is particularly true during an outbreak, when the quest is to find out whether an ailment is a symptom of the dreaded disease. It happened during the Ebola outbreak, when the H1N1 flu hit, and as measles tore through certain communities. 

During the coronavirus pandemic, searches for certain symptoms might become useful for predicting the next places to experience an increase of cases. New research from a team of gastroenterologists and epidemiologists at Massachusetts General Hospital found that searches for common gastrointestinal symptoms associated with Covid-19 spiked four weeks prior to a rise in cases for most states they examined.

One of the researchers, Dr. Kyle Staller, an epidemiologist and the director of the Gastrointestinal Motility Laboratory at Massachusetts General Hospital, said he had the idea to examine this data after he was “redeployed” as a Covid doctor at the beginning of the pandemic. “I saw a lot of patients with gastrointestinal issues that were a herald before traditional respiratory symptoms,” he said. “We used this analysis method because we had done similar work looking at eating disorders and irritable bowel syndrome in search results.”

The research uses Google Trends data for search terms like abdominal pain, loss of appetite, diarrhea, vomiting, and ageusia, the medical term for the loss of taste, which is commonly associated with the onset of Covid-19. The researchers examined search results from 15 states with high, medium, and low Covid-19 burdens for a 13-week period from January to April.

In the states with a high incidence of Covid—New York, New Jersey, California, Massachusetts, and Illinois—searches for ageusia and loss of appetite correlated most strongly with spikes in cases four weeks later. That lag time is likely due to one of two explanations, Staller said. The first is that GI symptoms could pre-date respiratory symptoms, leading people to think they might have food poisoning or another illness. The second is that in states where testing capacity was stretched thin during the early months of the pandemic, people weren’t getting tested until they were seriously ill.

“Covid testing may not have been as available as it is now,” Staller said. “You may have had to be classically ‘Covid sick’ before you could access testing.”

One thing complicating research on gastrointestinal symptoms is the lack of a medical consensus on how commonly they occur. In March, researchers estimated that between one half and two-thirds of patients with a confirmed case of coronavirus had at least one gastrointestinal symptom. Other research found that one-third of patients had GI symptoms. Later research suggested that GI symptoms might appear in only 10% of people with the virus. 

The changing news around coronavirus could impact search data if similar studies are conducted again because “search results provide a very behavior oriented dataset” and people now have a “stronger understanding of GI symptoms associated with the coronavirus,” Staller said. 

In general, search trends on Google associated with the coronavirus haven’t been static throughout the pandemic. Early on, common terms were “N95 masks” and “social distancing definition.” Then searches for “anxiety” and “panic attack” rose, speaking to the mental toll of quarantine and the uncertainty of the virus’ implications for daily life. Amid the early lockdown measures, another term that spiked was, well, “amid.”

If public health officials want to use search data to potentially predict outbreaks of the virus moving forward, another confounding variable this fall and winter they’ll have to deal with is how to differentiate searches for flu symptoms from searches for Covid symptoms. They’ll also have to be careful not to exclusively rely on search data, which can sometimes be inaccurate. Google Flu Trends, a now-defunct algorithm that was supposed to predict flu outbreaks before they happened, notoriously failed to predict the H1N1 pandemic because most people searching for “the flu” didn’t know which symptoms were associated with influenza and were probably sick with something else. That might not be the case with coronavirus, however, because of widespread media coverage of specific symptoms like shortness of breath and loss of taste.

If public health officials can identify a symptom that is closely correlated with only the coronavirus, search data could be an invaluable tool, Staller said, especially as data gets more specific to the county level. 

“As an epidemiologist, I still believe old fashioned epidemiology beats everything: testing, interviewing people, what I call ‘boots on the ground,’” he said. “But in a fast moving situation like Covid, this could provide useful adjunctive information for people at the state and local government level who have to make public health decisions. It’s another source of information that can potentially confirm suspicions that they’re hearing from their public health officials on the ground.”

Emma Coleman is the assistant editor for Route Fifty.

NEXT STORY: Tips for Living Online—Lessons from Six Months of the Covid-19 Pandemic