Food distribution modeling to accelerate foodborne disease outbreak investigations

Food safety procedures, such as the recommendations recently published by the FDA, are critical to reducing foodborne illness. However there is no way to completely eliminate the risk of receiving contaminated food. When prevention efforts fail, rapid identification of the source product is essential. The medical and economic losses incurred grow with the duration of the outbreak. To identify the contaminated product, public health investigators reconstruct the relevant food distribution network. The time required for such an investigation ranges from days to weeks. Accelerating this process will reduce the number of people sickened and restore consumer confidence in our food system. In this presentation, we describe a model-based technology to accelerating the investigation that begins with pre-computing the spatiotemporal distribution of retail products using data warehoused in retail inventory systems. We show through the application of a maximum likelihood method that it is possible to determine the most probable contaminated product based on a small number of geo-coded case reports. The method was evaluated using real German data. In this study, we applied and validated our predictive analytics proving product identification rates of 90% or higher in as few as 10-20 case reports. We will describe how correlations between existing distributions can predict contaminated food sources most easily or least easily comparing the results obtained from the model. This exploratory project may someday spot hard-to-find patterns in vastly varied kinds of data; analyze and integrate information real-time in a context-dependent way; and deal with the ambiguity found in real-world environments.