You wake up feeling like someone has taken a jackhammer to your head. You’re feverish, aching all over and your stomach is doing somersaults. There’s no doubt about it: You have the flu.
You also have reservations for dinner tonight. So after a mug of tea and an ibuprofen, you grope for your phone and cancel the reservations you’d made through OpenTable.
That cancellation might be a signal to public health officials of a flu outbreak. Because, according to a study by HealthMap’s John Brownstein, PhD, and Elaine Nsoesie, PhD, reservation data from OpenTable could offer another view into the seasonal spread of the flu.
“We’re always on the search for sources that can give us real-time data at high geographic resolution,” says Brownstein, HealthMap’s co-founder and a researcher in Boston Children’s Hospital’s Informatics Program. “There is a lot of work in the digital epidemiology field to find early signs of disease outbreaks in population data, especially where clinical data is lacking.”
While there are several ways of and sources for tracking the flu—such as Google Flu Trends, influenza reports from the CDC and HealthMap’s own Flu Near You—there’s one form of data that Brownstein says has always been hard to capture: data on behaviors like absenteeism.
“Absenteeism has been shown to be a really good indicator for flu and other diseases,” Brownstein says. “The challenge is that the data are hard to collect. They aren’t recorded in standardized ways, and you’re lucky if you can get city-level resolution.”
Moreover, not all absenteeism happens at work or school. Being sick makes people change their plans and behaviors in any number of ways. Reservation cancellations are one of those ways that Brownstein and Nsoesie decided to try to capture.
As they reported in the Journal of Medical Internet Research, their team gathered data on table availability from OpenTable, an online restaurant reservation service, in 10 distinct locations in the U.S. and Mexico.
They then compared that information with influenza surveillance data before and during the 2013 flu season from state public health agencies, Google Flu Trends and the Pan American Health Organization, looking for any correlations between jumps in table availability and reports of flu-like illness.
“Our goal was to see whether an increase in cancellations could be an indication that something was happening in the community,” explains Nsoesie, a research fellow in Brownstein’s group. “OpenTable doesn’t provide those data directly, but we were able to look for changes in table availability shortly before selected lunch and dinner times.”
As a proof of concept, the comparison worked well. “We found that for some locations there was a strong correlation between table availability and reports of flu-like illness,” says Nsoesie. For instance, there were strong dinnertime correlations for Baltimore and Miami, as well as for Atlanta around lunch.
Surprisingly, though, the correlation didn’t hold up for cities in more northern climates like Boston. That gap may be due at least in part to limitations in the data Brownstein and Nsoesie had to work with.
“This is all very preliminary, and there is a lot that needs to be done to explore the data further,” Nsoesie says. “We only gathered data on table availability and controlled for the number of participating restaurants in each location, but that doesn’t tell us anything about how many active OpenTable users there are per city.”
“There may be other factors, like weather generally, that affect reservations in northern climates,” Brownstein adds. “The baseline level of OpenTable activity may also be more stable in the areas where we did find correlations.”
Brownstein and Nsoesie note that there are several additional things they’d like to try to shore up their analysis, such as surveys to ask directly why people chose to cancel a reservation, collecting data covering a longer period of time and carrying out comparisons with other sources of flu surveillance data.
With such information, they may be able to better understand the correlations they found and maybe reveal new ones.
“This is all very preliminary, and how the correlations we’ve found translate into real-world results isn’t completely clear yet,” Nsoesie cautions. “But our results are promising, and we’d like to investigate them further.”
