Agent-based models for food producers’ decision-making about COVID-19 and food safety
Renata Ivanek, Cornell
Abstract:
Food producers are constantly making tough decisions—often in the face of incomplete information and under privacy constraints—to ensure the continued production of food and food safety. In this presentation, I will talk about two agent-based models (‘FInd CoV Control’ and ‘EnABLe’) that we developed to support food producers’ decision-making and contingency planning. FInd CoV Control predicts illnesses and absences from work due to COVID-19 among workers in a food company. It helps companies identify which COVID-19 control strategies would work best in their operation or facility in terms of protected worker health and continued food production. EnABLe recreates the unique environment, equipment, and practices in a food facility and then serves as a digital twin of the facility. It helps the facility monitor the food environment for the presence of harmful bugs to reduce the probability of food contamination with foodborne pathogens. These agent-based models aid the food producers’ decision-making and support the safety and security of the food supply.
Bio:
Dr. Renata Ivanek is a Professor of Epidemiology at the College of Veterinary Medicine, Cornell University. She serves as an Associate Director of the Cornell Institute for Digital Agriculture (CIDA).
Dr. Ivanek’s expertise includes veterinary medicine, epidemiology, and computer modeling. Her research is at the intersection of food and health, and her computer lab develops sustainable approaches for improving food safety, controlling infectious diseases, and optimizing food production systems.
Dr. Ivanek holds a Doctor of Veterinary Medicine degree from the University of Zagreb in Croatia, an M.Sc. in Veterinary Epidemiology from the University of London in the United Kingdom, and a Ph.D. in Comparative Biomedical Sciences from Cornell University.
Summary:
Focus: interaction between food and health within the food supply system
COVID-19 among food industry employees
Very hard-hit
Workers are working in close physical proximity
Climate-controlled and enclosed, which promotes spread
When food facilities (processing, crop picking, etc.) closed or were under-utilized due to COVID, this affected the entire food supply chain
Food processing sector is critical to the whole economy and its workers are critical employees
How do we protect worker health while keeping the food supply chain operational?
FInd CoV Control: decision support tool for food facility managers
Agent-based model of COVID-19 spread within food facilities
Model components
Employees (age, infection status, vaccination status)
Work environment (farm vs facility, shift schedule, agents number, hierarchy and contact network)
Hierarchy: manager, sub-supervisor, workers
Contact between adjacent levels of hierarchy and workers in same shift)
Worker schedule driven by their shift schedule and weekday/weekend
Disease spread dynamics
Used variant of the SEIR model (Susceptible, Exposed, Infected, Recovered) with paths for symptomatic/asymptomatic and accounting for vaccination status
Outcomes
Public health (symptomatic and asymptomatic infections)
Economic (working employees, capacity per shift, expenses)
Validated model against data from pandemic start (no control strategies)
Farm module 2 outbreaks on produce farms
Facility module: 3 outbreaks in dairy, pork and produce processing facilities
Because the pandemic just started the infection and vaccination status of all workers was known so fit was good; major uncertainty was the timing of the start of the outbreak but the predicted shape of the infection curve matched what was seen in the real outbreak.
Online tool for helping people run custom analyses for their facilities with 12 different interventions (vaccines, temperature checks, testing, biosafety)
https://www.foodcovidcontrol.com/FOODCTLConclusions:
Very intensive testing is effective but costly
Less intensive testing is not effective and actually reduces workforce since you send home asymptomatic people without preventing spread
Very intensive physical distancing is cost-effective
Reactive vaccination is too slow
Proactive vaccination is very effective
Need to redo analysis for different facility characteristics, initial conditions (e.g. employee attributes), virus variants
Food safety
Listeria monocytogenes
Exposure via ready-to-eat foods (e.g. fresh fruits/vegetables, ice cream, deli meat) that are not cooked by consumer
Few people get infected, but 1 in 5 people infected die
Cost $4b/year
Once contamination is discovered, it is often traced back to source facility
Facilities employ strategies to avoid food contamination
Environmental monitoring
Testing samples for various Listeria species since they want to look for conditions where these species may thrive
EnABLe model of Listeria spread
Create digital twin of a facility with its physical surfaces
2.5D: several 2D layers (floor, walls, equipment, ceiling)
Equipment, workers and the contact between them
Workers move through space, touch surfaces and equipment
Presence of water is explicitly tracked
1 hour time step
Listeria can grow over time in good conditions or may die off in poor conditions
Information needed:
Floorplan, equipment
Presence of water
Sanitation
Traffic patterns (people, shifts, food products)
Model of Listeria behavior
Introduction process
Growth, transmission and cross-contamination
Reduction due to cleaning
Conclusion:
Facility design strongly affects Listeria spread risk
EnABLe can be used to evaluate different testing/sanitation methods in the same facility
Effective sanitation strategy can effectively reduce/eliminate Listeria contamination even if Listeria enters the facility from the outside (e.g. via produce)
EnABLe can enable contamination root cause analysis