Simulation and identification of foodborne outbreaks in a large supermarket consumer purchase dataset

Abstract

Foodborne outbreaks represent a significant public health burden. Outbreak investigations are often challenging and time-consuming, and most outbreak vehicles remain unidentified. The development of alternative investigative strategies is therefore needed. Automated analysis of Consumer Purchase Data (CPD) gathered by retailers represents one such alternative strategy. CPD-aided investigations do not require trawling questionnaires to create a hypothesis and can provide analytical measures of association by direct data analysis. Here, we used anonymized CPD from 920,384 customers enrolled in Norway’s largest supermarket loyalty program to simulate foodborne outbreaks across a range of different parameters and scenarios. We then applied a logistic regression model to calculate an odds ratio for each of the different possible food vehicles. By this method, we were able to identify outbreak vehicles with a 90% accuracy within a median of 6 recorded case-patients. The outbreak vehicle identification rate declined significantly when using data from only one of two retailers involved in a simulated outbreak. Performance was also reduced in simulations that restricted analysis from product ID to the product group levels accessible by trawling questionnaires. Our results show that—assuming agreements are in place with major retailers—CPD collection and analysis can solve foodborne outbreaks originating from supermarkets both more rapidly and accurately than than questionnaire-based methods and might provide a significant enhancement to current outbreak investigation methods.

Simulation and identification of foodborne outbreaks in a large supermarket consumer purchase dataset