Most cases of foodborne illness are never reported to public health authorities. But that doesn’t mean those afflicted suffer in silence. Many people turn to social media to complain and even identify food purveyors they believe are responsible.

To tap into this rich source of information, city public health departments have begun mining the tweets and online reviews of those possibly sickened by food.

The City of Chicago Department of Public Health was the first to test the potential of social media in identifying foodborne outbreaks. The department partnered with civic-minded local technologists and the Smart Chicago Collaborative, a nonprofit organization that uses technology to improve the lives of Chicagoans, to develop an application to monitor Twitter for possible food poisoning references. A similar project is under way in New York, where the New York City Department of Health and Mental Hygiene is working with Columbia University technologists and the review website Yelp to comb restaurant patrons’ comments for signs of a budding outbreak.

Public health departments in Chicago and New York are using social media tools such as Twitter and Yelp as sources of consumer comments that might help identify outbreaks of foodborne illness. twitter.com/foodbornechi

A Civic Project

Despite the availability of Chicago’s 311 city services hotline, which fields calls on everything from potholes to food poisoning, many cases of foodborne illness go unreported, explained Bechara Choucair, MD, commissioner of the Chicago Department of Public Health.

“People may not know, or they may not want to,” Choucair said. “We are being more proactive about collecting reports.”

The Smart Chicago Collaborative and local web developers built the application using machine learning to identify potential cases of food poisoning mentioned on Twitter by people in Chicago. The app scans the data and provides the health department with potential leads. Staff members review the leads and send messages to individuals with likely food poisoning, encouraging them to report their illness to the department using an online form.

The Foodborne Chicago app has classified more than 2900 messages on Twitter over the past year as likely related to foodborne illness, according to Choucair. After reviewing these, staff followed up on 330 that fit certain criteria. Of the people they contacted, 87% (288) filed a report about their illness.

The past year has been a learning process for the department and those developing the app. Choucair explained that some tweet authors have been “caught off guard” by receiving a tweet from the department. The technology has also required some tweaks. For example, initially the app might flag a tweet by a professor giving a talk about foodborne illness. The algorithms have since been improved to better identify real complaints.

“We are getting better by the day,” Choucair said.

Bad Reviews

An actual outbreak led to New York’s pilot project testing whether Yelp reviews might signal threats to food safety. Public health officials investigating a 2011 outbreak of gastrointestinal illnesses among a restaurant’s patrons found that some who didn’t file an official complaint reported their experience on Yelp.

Sharon Balter, MD, medical epidemiologist at New York’s public health department, explained that a staff member used her own Yelp account to contact individuals who posted about their experiences at the restaurant under investigation. A few agreed to talk with investigators.

Balter and her colleagues then enlisted Yelp and Luis Gravano, PhD, of Columbia University’s computer science department, and his students for a pilot study to identify outbreaks prospectively based on Yelp reviews. From July 2012 to March 31, 2013, the software program created for the pilot screened 294 000 Yelp reviews (Harrison C et al. MMWR Morb Mortal Wkly Rep. 2014;63[20]:441-445). It cited 893 as requiring evaluation by health department staff. About half were found to be consistent with food poisoning and occurred within a month of the review’s posting or didn’t specify a date (requiring further investigation). Only about 3% of these staff-flagged reviews had been reported through traditional means.

Staff further investigated 129 claims and conducted phone interviews with 27 people who posted illness-related reviews. The interviews enabled staff to identify 3 unreported outbreaks that caused 16 illnesses. Investigations of the restaurants linked to the outbreaks found numerous violations of safe food-handling practices.

Balter said the results suggest that Yelp surveillance may identify small outbreaks of foodborne illness that traditional surveillance techniques miss. “It’s a way of reaching people who don’t know to call the health department,” she said, noting that individuals posting online are likely different from those who call the department to report illnesses.

In fact, her colleagues have used Yelp messages to contact individuals who posted about illnesses that were not recent enough to be included in the pilot to let them know they can report their concerns to the department.

One downside of Yelp outreach is that some users infrequently check their Yelp messages. Others may have chosen not to reply. Chicago and New York have been sharing their experiences and trying to learn from each other, said Choucair. Balter said New York is working with Gravano and his students to use Twitter for surveillance. Choucair noted that the computer code for the Foodborne Chicago app is now available on GitHub, an open source repository that allows other public health departments or civic-minded technologists to access computer code and improve on it.