Digital farm-to-facility food safety testing optimization

Effective food safety product testing in the produce industry is limited by a history of both academic studies and customer requirements focusing on single, or limited, points in the supply chain. This project proposes to build on previous work simulating both in-field and packing-house pathogen product testing to create an integrated production, harvesting, processing, and packing model to define optimum food safety testing schemes for produce. To do this, the project will first build a Field-to-Facility model of leafy green produce safety testing using spreadsheet- and flowchart-based computer simulation. These will incorporate a range of alternative testing plans, potential processing impacts on pathogen risk, and contamination scenarios. Second, the project will generalize that model to incorporate the additional important example commodities tomatoes, apples, cilantro, and jalapenos. – each with different hazard profiles and risk management options determined though literature review and site visits to growers and processors. Third, the project will simulate many iterations of these supply chains tracking the variability and uncertainty in the ability of specific testing schemes to identify and reject produce contaminated by different hazard profiles. These results will determine recommendations for optimized field-to-facility food safety product testing.

Optimum food safety testing in the produce industry is limited by inconsistent requirements for product testing, legacy approaches focusing on single points in the supply chain, and inability of testing schemes to bound a contamination event. This project will create an integrated production, harvesting, processing, and packing model to define optimum food safety product testing schemes for produce. This project will address a research gap on how and when sampling and testing could, or could not, be used to detect and manage both point-source and systematic contamination as part of a Field-to-Facility food safety management system. First the project will build a Field-to-Facility computer model of leafy green produce safety testing, incorporating a range of alternative testing plans, potential processing impacts on pathogen risk, and contamination scenarios - defined in collaboration with academic and industry partners. Scenarios are based around a nominal 100,000 lb production lot of leafy greens, harvested in 10,000 lb sublots, subject to value-addition and packing, and tested by multiples of representative N60 composite samples taken at different supply chain points. Second, the model will represent a variety of higher-risk commodities with distinct risk profiles and risk-management options, specifically tomatoes, apples, jalapenos, and cilantro. This adaptation will involve site visits to major production regions in FL, WA, MI, SC, and CA to qualitatively assess the supply chain, literature review to estimate parameters, and discussions with industry partners to define additional product testing plans. Third, the project will optimize testing across the supply chain of each commodity. This is done by simulating 1,000s of variability and uncertainty iterations of field contamination, testing, and interventions, and calculating residual pathogens in the system. Then using these results for formal sensitivity analysis and optimization to define which sampling plans best reduce food safety risk across a range of contamination scenarios. The goal of this project is to affect real change in produce food safety testing though industry, academic, and regulatory use of both the model and published results to advocate for optimized testing in practice.

1. Build a Field-to-Facility generic supply chain model of produce safety testing. 

2. Adapt the supply chain and collect parameters to represent a variety of higher-risk commodities with distinct risk profiles and risk-management options. 

3. Optimize testing across the supply chain of each commodity incorporating representative testing programs at primary production, harvesting, receiving, processing, and packing and assessing their impact to manage safety.

Study 1: Leafy Greens This study indicates that sampling is less impactful at reducing the endpoint adulterant cells when effective systems-based interventions are in place. The model showed that the sampling plans in this study have limited power to detect contamination levels like the ones that caused a foodborne disease outbreak in 2018. The model showed that conducting sampling too early in the system may not be as beneficial compared to sampling closer to a contamination event, as shown by the Holding intervention. Sampling plans should focus on locations where contamination is likely to be high enough for powerful sampling, at preharvest, harvest, and receiving, as this would allow sampling to detect the incoming product that presents high levels of contamination. Effective interventions reduce contamination during processing, making finished product sampling and customer sampling have negligible effects. In addition, finished product sampling could better inform a contamination event if all interventions were to fail. This study suggests that interventions are effective at reducing incoming contamination. Producers and buyers should focus on implementing suitable food safety interventions as primary preventive controls. Once interventions are implemented, sampling plans can be used as a tool to detect high-level contamination or unappreciated/untreated sources of contamination. 

Study 2: Tomatoes This study builds upon the findings from the study of leafy greens. This study aimed to identify possible sampling locations for tomatoes along the farm-to-packinghouse process. In addition, we also wanted to implement sampling plans based on scientific recommendations that would be feasible for growers to implement. The findings suggest that the best sampling locations are at harvest and receiving, and this aligns with the findings for the leafy green study. Locations where contamination levels are predicted to be present consistently and at higher levels are the sampling locations that result in the higher detection power. When the clustering level is higher (0.1% cluster) the best sampling location was predicted to be at packed product, showing that when contamination is high and highly clustered, the processing steps provide enough crosscontamination to make detection more powerful towards the end of the system. However, the model shows that for the 0.15% cluster contamination event, relative efficacy was a maximum of 12.3% compared to the 98.1% reduction achieved when contamination was random uniform (widespread 100% cluster). The model shows that for random uniform (widespread 100%) and the 10% cluster the best sampling plans are those that test the larger amount of product (60 and 20 tomatoes), while when contamination is highly clustered (1% or 0.1% cluster), the best sampling plans are those that sample more individual tomatoes (60 tomatoes, and 60 tomato mash). While some very large sampling plans (60 tomatoes and 60 tomato mash) typically showed large reductions in cells reaching the system endpoint, these specific plans are likely too large to be practical. More feasible plans like the 2 and 6 tomatoes sampling plans, which match total mass to 375 and 1,500 g masses collected with ICMSF plans, are underpowered, especially under highly clustered scenarios. 

Study 3: Cilantro Simulation results of FDA BAM Chapter 19B and 19C protocols showed that the number of samples matters for C. cayetanensis detection in cilantro. This study identified important contamination thresholds for C. cayetanensis detection. At levels above 1.58 oocysts per liter for water testing, and 0.92 oocysts per gram for produce testing, a single 10-L or 25-g sample will reliably detect C. cayetanensis contamination. This model provides the industry with information on the performance of the current testing methods for C. cayetanensis under different contamination levels, and testing scenarios. Similar results were observed after scenario analysis. Results showed that increased number of samples and daily testing have higher efficacy at detecting C. cayetanensis during a harvest season, whereas a single sample during the harvest season will have a hard time detecting a random contamination event. Overall analysis showed that testing both water and produce has a higher efficacy at detecting C. cayetanensis than only testing for water or produce. This is more evident when random irrigation with contaminated water occurs. Based on the simulation results, it would be ideal to increase the number of samples for both water and produce testing to increase the detection of C. cayetanensis, especially when lower contamination is expected. However, expenses associated with C. cayetanensis testing will limit the number of samples required for improved detection.