Purpose: We are developing a new software module for the open-source software platform FoodRiskLabs, which is designed to support risk assessors and decision makers in regular risk assessments and outbreak investigations. Specifically, the new software components support the identification of a disease-causing “contaminated” product by comparing the products sales distribution pattern with the spatial distribution pattern of human infections.
Methods: The new software module extends the collection of FoodRiskLabs tools which are based on the open-source data analytics platform KNIME. It includes the likelihood-based approach introduced by Kaufman et al. (2014), and provides support for parallelized or cloud-based execution. Further, a new Monte Carlo simulation-based algorithm has been developed that allows to identify the minimal set of products containing the “guilty” product within a user-defined confidence limit.
Results: The software features were tested on artificial outbreak scenarios generated from real world sales data. It was used to study the performance effects of three influencing factors using the extended likelihood-based approach. These analyses confirmed that for the given scenario settings the number of products under suspicion and the spatial granularity of the available data strongly influence the algorithm’s performance.
Significance: The new FoodRiskLabs extension will be made freely available for download and joint development as an open-source community resource (https://foodrisklabs.bfr.bund.de). It provides a scalable software infrastructure enabling food authorities (and private sector stakeholders) to include food sales data into their outbreak investigations. Further research is necessary to address remaining open questions with respect to the algorithm’s performance in cases where the underlying assumptions on the product sales data are not fulfilled.