A new computer model that uses machine learning and Google's unidentified search and location data and aggregates of logged in users was significantly more accurate in identifying potentially unsafe restaurants when compared to existing methods of consumer complaints and routine inspections of according to research by Google and Harvard TH Chan School of Public Health. The results indicate that the model can help identify time lags in food security in near real time.
"Foodborne illness is common, expensive, and lands thousands of Americans in emergency rooms every year. This new technique, developed by Google, can help local restaurants and health departments find problems faster before they become major public health problems, "said author Ashish Jha, KT Li Professor of Global Health at Harvard Chan School and director of the Harvard Global Health Institute.
The study will be published online on November 6, 2018 in npj Digital Medicine.
Foodborne illness is a persistent problem in the United States and in current methods by restaurants and local health departments to determine an outbreak based primarily on consumer complaints or routine inspections. These methods can be slow and cumbersome, usually resulting in delayed responses and increased spread of the disease.
To counter these shortcomings, Google researchers developed a machine-learned model and worked with Harvard to test it in Chicago and Las Vegas. The model works by first classifying search queries that may indicate foodborne illness, such as "stomach cramps" or "diarrhea." The model then uses data from the smartphone's location history and smartphone aggregates from people who chose to save it to determine which restaurants people who searched those terms recently visited.
The health departments of each city have received a list of restaurants that have been identified by the model as potential sources of foodborne illness. The city then dispatched the health inspectors to these restaurants, although health inspectors did not know whether their inspection was motivated by this new model or by traditional methods. During the study period, health departments continued to follow their usual inspection procedures as well.
In Chicago, where the model was deployed between November 2016 and March 2017, the model generated 71 inspections. The study found that the rate of unsafe restaurants among those detected by the model was 52.1% compared to 39.4% among inspections triggered by a complaints-based system. Researchers have noted that Chicago has one of the most advanced monitoring programs in the country and already employs social media mining techniques, but this new model has proven to be more accurate in identifying restaurants that had food safety violations.
In Las Vegas, the model was implemented between May and August 2016. Compared with routine health inspections, the rate of identification of unsafe restaurants was higher.
When researchers compared the model with routine health department inspections in Las Vegas and Chicago, they found that the overall rate in both cities of unsafe restaurants detected by the model was 52.3%, while the overall restaurant detection rate unsafe via routine inspections in the two cities was 22.7%.
Interestingly, the study showed that in 38% of all cases identified by this model, the restaurant potentially causing foodborne illness was not the latest one visited by the person researching symptom-related keywords. The authors said that this is important because previous surveys have shown that people tend to blame the last restaurant they visited and therefore can file a complaint to the wrong restaurant. However, clinically, foodborne illness can take up to 48 hours or more to become symptomatic after someone has been exposed, the authors said.
The new model has outperformed inspections based on complaints and routine inspections in terms of accuracy, scale, and latency (the time spent between people getting sick and the outbreak being identified). The researchers noted that the model would be better leveraged as a complement to existing methods used by health departments and restaurants, allowing them to prioritize inspections better and conduct internal food safety assessments. More proactive and timely responses to incidents can mean better public health outcomes. In addition, the model can be valuable to small and medium-sized restaurants that can not afford to the security operations personnel to apply advanced food safety monitoring and data analysis techniques.
"In this study, we have only scratched the surface of what is possible in the field of machine-acquired epidemiology." I like the analogy with the work of Dr. John Snow, the father of modern epidemiology, who in 1854 had to leave. from central London by asking people where they got the water to find out the source of a cholera outbreak. We can use online data to make epidemiological observations almost in real time with the potential to significantly improve public health timeliness and cost- efficient, "said Evgeniy Gabrilovich, a senior research scientist at Google and co-author of the study.