Source: UNIVERSITY OF VERMONT submitted to
PREDICTING LIVESTOCK DISEASE TRANSMISSION DYNAMICS UNDER ALTERNATE BIOSECURITY RISK MANAGEMENT INTERVENTIONS AND BEHAVIORAL RESPONSES OF LIVESTOCK PRODUCERS
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
NEW
Funding Source
AFRI COMPETITIVE GRANT
Reporting Frequency
Annual
Accession No.
1026829
Grant No.
2021-67015-35236
Project No.
VT-0095CG
Proposal No.
2021-05779
Multistate No.
(N/A)
Program Code
A1222
Project Start Date
Sep 1, 2021
Project End Date
Aug 31, 2026
Grant Year
2021
Project Director
Zia, A.
Recipient Organization
UNIVERSITY OF VERMONT
(N/A)
BURLINGTON,VT 05405
Performing Department
Com Dev & Applied Economics
Non Technical Summary
Transmission of foreign animal diseases, such as foot and mouth disease and African swine fever, could pose a significant risk to US livestock producers. Current approaches US animal health authorities use to manage this risk and the behavioral responses of livestock producers to these policies are not well understood in an integrative way. The overarching goal of this project is to better understand and predict the risk of foreign animal disease transmission among livestock herds in the US. We will test alternative interventions and alternative sequences of interventions, while determining the expected behavioral responses of livestock producers to these interventions. We will test multiple hypotheses with the goal of identifying interventions that improve the alignment of economic incentives, increase the effectiveness of risk communication, and enhance voluntary surveillance of foreign animal disease at minimal cost to taxpayers and livestock producers. Our three research objectives are the following: (1) We will design and test a novel Artificial Intelligence (AI) model to simulate transmission of two foreign animal diseases (foot and mouth disease and African swine fever) in the US livestock sector. (2) We will conduct surveys of livestock producers and implement online experimental games to calibrate the representation of human behavior in the AI model and test alternative policy configurations. (3) We will harness both survey and AI model data to identify combinations of risk management strategies and leverage points in livestock supply chain networks that minimize risk of spreading from livestock diseases. The findings from this project will help to set and calibrate livestock risk management policies (e.g., US Farm Bills), provide targeted guidance to national and regional veterinary health officials, and generate open source knowledge for livestock producers to calculate economic costs for biosecurity risk management interventions at both farm and market levels. This project will provide policy recommendations to livestock producers and policy makers for improving the prevention and control of foreign animal diseases, enhancing biosecurity, and increasing USDA capacity to forecast and respond to outbreaks of high consequence diseases.@font-face{panose-1:2 4 5 3 5 4 6 3 2 4;mso-font-charset:0;mso-generic-mso-font-pitch:variable;mso-font-signature:-536870145 1107305727 0 0 415 0;}p.MsoNormal, li.MsoNormal, div.MsoNormal{mso-style-unhide:no;mso-style-qformat:yes;mso-style-parent:"";margin:0in;mso-pagination:widow-orphan;;mso-fareast-}p{mso-style-priority:99;mso-style-qformat:yes;mso-margin-top-alt:auto;margin-right:0in;mso-margin-bottom-alt:auto;margin-left:0in;mso-pagination:widow-orphan;;mso-fareast-}.MsoChpDefault{mso-style-type:export-only;mso-default-props:yes;mso-ascii-mso-ascii-theme-font:minor-latin;mso-fareast-mso-fareast-theme-font:minor-latin;mso-hansi-mso-hansi-theme-font:minor-latin;mso-bidi-mso-bidi-theme-font:minor-bidi;}div.WordSection1{page:WordSection1;}
Animal Health Component
100%
Research Effort Categories
Basic
50%
Applied
50%
Developmental
0%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
3113399106050%
3113510106050%
Knowledge Area
311 - Animal Diseases;

Subject Of Investigation
3510 - Swine, live animal; 3399 - Beef cattle, general/other;

Field Of Science
1060 - Biology (whole systems);
Goals / Objectives
The overarching research goal of this project is to predict transmission dynamics of two foreign animal diseases (FADs) -- foot-and-mouth disease (FMD) and African swine fever (ASF) -- in a spatially-explicit dynamically-evolving network of livestock producers under alternate biosecurity risk management interventions and sequences of interventions and heterogeneous behavioral responses of livestock producers to these interventions. We propose to test the following three risk management strategies and associated hypotheses:(i) Incentives: Compare baseline policy of unconditional indemnity with conditional indemnity and no indemnity as alternate policy mechanisms for cost effective minimization of FAD risk. We hypothesize (H1.1) that an unconditional indemnity policy reduces the adoption of biosecurity practices, while conditional indemnity policy increases the adoption of biosecurity practices. No indemnity policies will trigger middle of the road willingness to invest in biosecurity. Further, we hypothesize (H1.2) that high-price/high-demand of livestock products increases the proportion of risk taking behaviors, while low-price/low-demand of livestock products increases proportion of risk averse behaviors among livestock producers. Regulators can use price support mechanisms, e.g. price floors and ceilings, to increase biosecurity adoption and buffer livestock markets against price shocks from domestic and international markets.(ii) Communication: Test different risk communication strategies from the perspective of regulators as an approach to nudge higher adoption of biosecurity practices by livestock producers under differential conditions of social ecological uncertainty and network structures. We hypothesize (H2.1) that graphical, as opposed to verbal and numerical communication of FAD outbreak risk induces higher adoption of biosecurity and compliance with risk management protocols. Further (H2.2), communication of information about adoption of biosecurity by others in the network and disease incidence and propagation in the network induces risk averse behaviors.(iii) Control: Compare alternate social psychological approaches to overcome moral hazard and collective action dilemmas associated with risk control strategies during the FAD outbreaks. We hypothesize that (H3.1) higher trust of government and positive attitude towards veterinary science treatments (e.g. vaccinations) induces higher compliance with disease surveillance and control practices (e.g. stop movement orders). Finally, we hypothesize that (H3.2) provision of network-wide FAD risk information to producers increases perceived behavioral control, which in turn increases compliance with risk control actions. For successfully attaining these research goals and testing these research hypotheses, we will implement the following three research objectives:Design and test an ABM for a baseline scenario of transmission of two FAD (FMD and ASF) in the US livestock sector and compare its projections against the InterSpread Plus model.Develop and implement panel surveys and online experimental gaming simulations to calibrate the behavior in the ABM, test risk management strategies and infer heterogeneous behavioral responses of livestock producers.Integrate risk mitigation and behavioral response inferences derived from surveys and games into the ABM and identify combinations of risk management strategies and leverage points in supply chain networks that minimize risk from livestock diseases.
Project Methods
Efforts:1: We will develop a national level Agent Based Model (ABM) that will simulate integrated socio-ecological dynamics of livestock diseases by endogenously accounting for two-way feedbacks and couplings between social systems (e.g. behaviors and policies) and ecological systems (e.g. susceptibility of farm animals to disease vectors). A baseline scenario of the ABM will be compared against the InterSpread Plus model simulations.2: We will develop and implement three online experimental simulation games and one panel survey to test risk mitigation strategies and infer heterogeneous behavioral responses of livestock producers. First experimental game will test incentive design hypotheses. Second experimental game will test risk communication hypotheses. Third experimental game will test risk control hypotheses. Finally, two waves of a national level survey instrument will be implemented and analyzed to quantify social psychological predictors of biosecurity adoption by cattle producers.3. We will integrate risk mitigation & behavioral response inferences derived from surveys and games into ABM and identify combinations of risk mitigation strategies and leverage points in supply chain networks that minimize risk from livestock diseases.Evaluation: Our project is based on the scientific approach of co-production of knowledge that incorporates important stakeholders in various stages of the project for both formative and summative evaluation. Specifically, at the time of project design, we have worked with external stakeholders to develop a timeline of 8 milestones (described below) for evaluating and tracking the project performance vis-a-vis objectives described above. For co-production of knowledge, we are working in partnership with the USDA APHIS CEAH (letters of collaboration from Dr. Amy Delgado, Associate Director, Monitoring and Surveillance and Dr. Columb Rigney, Veterinary Medical Officer, Transboundary Disease Analytics Unit were submitted with the proposal). Further, we will engage a diverse set of 8 to 12 industry experts and scientists, especially veterinarians, to design and implement the proposed ABM, survey and games. We will organize two mediated modeling workshops with public and private sector stakeholders in the second and fourth year of the project at the UVM campus to co-design and test the risk management scenarios in the ABM, games and surveys. Letters of collaboration from Dr. Lisa Becton, Director of Swine Health Information and Research at the National Pork Board and Dr. Michael Sanderson, Professor of Epidemiology and Beef Production at Kansas State University were attached with the proposal, both of whom have agreed to participate in both mediated modeling workshops. Dr. Delgado and Dr. Rigney will also participate in both of these meetings. Two to four livestock producers and policy makers from US congress working on the US Farm Bill will also be invited. Formative feedback and recommendations derived from the mediated modeling workshops will be used to evaluate the progress of the project with respect to the following milestones co-designed with USDA APHIS, industry and academia experts.Milestones: M1: Develop a baseline national level ABM (summer 2022);M2: Compare baseline ABM projections with InterSpread Plus (fall 2022);M3: Experiment 1 implemented (Summer 2023);M4: Experiments 2 and 3 implemented (Summer 2024);M5: Survey wave one (Spring 2022) & wave two (Spring 2024);M6: Integrate experimental data into ABM (summer 2024);M7: Integrate survey data into ABM (summer 2025);M8: Conduct global sensitivity analysis & identify leverage points (summer 2026)

Progress 09/01/23 to 08/31/24

Outputs
Target Audience:Scientists, modelers, and officials interested in obtaining a better understanding of how risk associated with infectious diseases are affected by the structure of the livestock industry and the behaviors of those involved. Further, the followingstakeholders are informed by the outputs and outcomes of this project: veterinary epidemiologists, state animal health officials, and livestock industry professionals. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?Four graduate students and two undergraduate students are being trained in animal disease transmission modeling, including InterSpread Plus, ABM, as well as experimental gaming and survey data collection procedures. Two of these four graduate students have successfully defended their dissertations during summer/fall of 2024. One undergraduate student also successful defended their thesis that focused on the experimental game data analysis. Another undergraduate's thesis is focusing on the routing aspect of trucks between farms. The intent is to better understand how disease spreads via trucks. An enterprise license for InterspreadPlus has been procured for UVM. A virtual machine running Windows has been provisioned for use by project members who do not have direct access to a Windows platform computer. A week long immersive training workshop for the InterSpreadPlus modeling engine was successfully completed by all project participants during year 3 of this project.Three graduate students presented their papers as lead authors at CRWAD 2024 annual conference and two graduate students plan to participate and present in the forthcoming CRWAD 2025 meeting. How have the results been disseminated to communities of interest?In addition to publication ofjournal articles, book chapters, technical reportsand conference presentations, we have continued to interact with our advisory board members. The advisory board members include representatives from USDA APHIS, National Pork Bureau, Livestock Industry and senior academics. Two scientists of USDA APHIS participate in our weekly project meetings. Through these interactions, our team presents the latest research on the ABM, surveys and experimental games to our target audiences. Our meetings with scientists and policy makers have also served the purpose of disseminating information about ABM, games and surveys to the community of veterinary scientists who study the individual parts of our system models. These experts we have worked with have expressed interest in our model and have discussed it with their respective colleagues. We have also engaged UVM extension professionals through offering atraining workshop in 2024 UVM EPIC conference. Students in computer science, sustainable development policy, community development and applied economics have been exposed to ABM design and testing procedures, choice experiment surveys, and game designs. What do you plan to do during the next reporting period to accomplish the goals?In years 4 & 5 of our project, we will continue to develop and test the national scale ABM for simulating ASF and FMD and incorporate policy and behavioral responses from the latest survey and gaming data sets. In the ABM, our implementation of feed mills and their truck routes is expanding to include data on registered medicated feed mills in the US, with their locations publicly available. These routes are being constructed systematically using all available knowledge on the heterogeneity of routing practices gathered from expert interviews, open resources, and academic literature. Special focus is being paid on feed routing and swine shipments as they offer potentially a cost-neutral way of increasing biosecurity for the US system. Swine shipment networks are being implemented building off data-informed inference models developed by researchers at the University of Colorado, Linköping University, and The University of Warwick. However this required data sharing agreements that have been finalized with relevant data owners; and data is being analyzed to parametrize supply chain networks. We also are implementing airborne transmission and seasonality of disease dynamics. Airborne transmission follows a similar network-based approach as the other interactions in that a pre-determined list of neighboring farms within a given cutoff radius will be read into the model at initiation along with their distances. Infections from airborne transmission will be limited to these interactions, rather than following a spatial dilution model or similar. Seasonality is important to incorporate as we begin data calibration from MSHMP, as seasonality currently is seen in disease trends which will make calibration difficult without it. Generally we plan to approach seasonality as changing epidemiological parameters slightly according to a schedule. As we add these additional features and expand our model to represent the national scale, new bottlenecks in the computation will arise. We will be utilizing existing runtime testing techniques and developing our own in order to identify and address them. We have also designed and implemented a second choice experiment survey in spring/summer 2024. From the survey, we are developing 2 additional papers focusing on a discrete choice experiment and an extended integrated theory of planned behavior. We plan for continued development of the experimental game to incorporate more features into the simulated decision mechanism. Participant recruitment using online survey platforms will be utilized to collect behavioral data to inform the agent based modeling initiative. The gaming data has been analyzed within the team in order to scope additional updates to the experimental simulation design, before we scale up our participant recruitment phase. Manuscripts from gaming data are being presented in CRWAD 2025. We plan to recruit upwards of 1000 additional participants for the next data collection phase of our experimental game. Behavioral results from integration of gaming data in ABMs will be analyzed and prepared for publication in journal articles. Machine learning models will be developed for analyzing learning from the experimental game, testing study hypotheses and integration into the ABM. The biosecurity adoption strategies developed by learning agents will be compared under different policy conditions. Additional experiments will compare agents with different lengths of memory, access to different information, and more or less discounting of future rewards. These results can be compared with data from experimental games and surveys to determine the verisimilitude of learned agent strategies. Findings from the ABM experimental simulations, games and surveys will be published in journal articles, conferences as well as policy briefs.

Impacts
What was accomplished under these goals? For objective 1, design and test an ABM, the research team has continued to make significant progress in developing a national scale ABM. Sixpapers from this work were presentedat Conference of Research Workers in Animal Diseases (CRWAD) 2024, the Agricultural Policy Research Group (APRG) conference, 12th International Congress on Environmental Modeling & Software, 3rd International Congress on Environmental Peacebuilding, and the Intelligent Systems Conference (ICRCC). Below, we outline specific progress and details in the development of this ABM. The general approach of our participatory ABM is trifold: First, we model the endemic disease Porcine Reproductive and Respiratory Syndrome Virus (PRRSV as a proxy for African Swine Fever (ASF) in the swine industry to calibrate the ABM with data gathered from multiple sources including incidence data and from the Morison Swine Health Morrison Swine Health Monitoring Program, (MSHMP). Next, we alter the model to simulategame data to equilibrate the model to these changes. Once calibrated with PRRSV data and equilibrated to pre-emptive system-wide biosecurity measures, we introduce ASF to the model to see how the adaptations the system has made to the pre-emptive system changes influences the ability of the United States to resist an ASF outbreak. The current iteration of the ABM is written in C++ using the GPU-accelerated FLAME-GPU library with Python used for additional external features. Use of our model for wide parameter sweeps is facilitated through runtime parameterization and CUDA (Compute Unified Device Architecture) processes. This language and library offer high processing speed by utilizing parallelization of agents, but little debugging capability. As such we have implemented our own debugging and output procedures to accelerate model development via output analysis using Python. Python is also used for external interfacing to and sideloading of additional features. We use Python to create a set of pre- made synthetic feedmill truck routes which we load into the model to avoid the computational cost of routing procedures in real time, and to explore more fully the effect of these routes. Reinforcement learning for ABM producer agents was developed in Python alongside the open-source C++ model. Initial agents trained using Q-Learning were shown to develop behavioral strategies similar to hand-designed strategies based on survey and experimental game data. Additional agents trained using Deep Q-Learning demonstrated less future reward discounting and learned strategies that more efficiently invested in biosecurity. Basic computational architecture of this model was presented in a noteworthy AI conference entitled Intelligent Systems Conference and published in ACM Networks and Systems (Andrew et al. 2024). Implemented advances of our model compared to prior livestock epidemiology models include modeling the base unit of infection and sale as batches of swine of the same age on a farm which can have multiple batches. Modeling the infection dynamics, transfer and sale of batches of pigs provides a meso-scale metapopulation approach allowing for realistic within- farm transmission dynamics, such as sustained infection which is characteristic of PRRSV in the US in recent years, and realistic economic modeling of sales without the prohibitive computational cost of modeling individual swine. Analysis of survey and gaming data set (described under objective 2) has led to identification of behaviroal response strategies under alternate indemnity policy and variable network structure configurations. These strategies are being encodedin ABM to predict PRRSV, ASF and FMD transmission dynamics under alternate biosecurity risk management interventions and sequences of interventions and heterogeneous behavioral responses of livestock producers to these interventions. For objective 2, we have made significant progress in designing and implementingsurveys and gaming experiments. We designed and implemented a survey tailored to infer heterogenous behaviors among swine producers in the US. We have analyzed the collected data resulting in the development of two scientific papers that have been published in high impact journals: Nature Scientific Reports and Preventive Veterinary Medicine. We have also presented the findings at 2major international conferences to a diverse audience includingISVEE17and CRWAD. Another survey using choice experimental approach has been deployed to understand relative attribute weights of biosecurity indemnity policy design choices assigned by different audience groups (producers, extension professionals, service providers etc). Analysis of choice experiment data and its integration with ABM is underway. A pair of advanced experimental games were successfully developed using Unity software and C#' and successfully launched with 1000 players generating approximately 500,000 decision observations during the summer/fall of 2024. Version 1 of the experimental game focused on government policies that would provide indemnity protection against ASF outbreaks as an incentive for purchasing protective biosecurity. We recruited 500 participants to characterize behavioral risk distributions and adaptive learning associated with policy incentives. We then developed version 2 of the experimental game to also include a more pronounced spatial network mechanism of disease spread into the simulation to identify the effect on producer decision making. The network connectivity of a hog production supply chain was varied, which introduced spatial risk into the user's decision calculus. We recruited an additional 500 participants through Amazon Mechanical Turk to participant in the experiment. The network connectivity and spatial risk was also found to have a behavioral impact on defensive biosecurity investment decisions, which will be infused in ABM. This experiment is structured as an outbreak simulation with analogous mechanics to the ABM. This experimental game includes a spatial distribution of configurable animal diseases, allowing the user to make decisions based upon risk related to infection transfer between commercial systems. A full user interface was developed to inform the player of the infectious disease risk along with a series of policy treatments such as government indemnity policies to test their influence on decision-making. As disease percolates throughout the supply chain, non-player farms can report the presence of an infectious disease, which is clearly rendered on the user interface. The disease reports can contain false positives, which provide additional uncertainty within the simulation based experimental game. The player weighs the risk with respect to spatial progression of the disease spread and can choose to invest their simulation dollars on biosecurity investment, which reduces disease transmission probability. The player must balance the risk of disease and the cost of biosecurity investment, with the added complexity of the government policy interventions for indemnifying culled animals. Beta testing of this experimental game inclusive of full behavioral and network data collection has been completed with participants recruited from Amazon Mechanical Turk's online crowdsourcing worker platform. Data from the beta version allowed us to test the data architecture, robustness of software (i.e., bug catching), test the epidemiological dynamics embedded within the experimental gaming framework, and develop statistical analyses to test network and policy level biosecurity). For objective 3, the open source ABM has been configured to incorporate survey and gaming data. Overall, threejournal articles, one book chapter, one UNEP report and eight conference presentations/abstractshave been published. Twoconference papers and one posterhasbeen accepted for presentation at CRWAD 2025.

Publications

  • Type: Peer Reviewed Journal Articles Status: Published Year Published: 2024 Citation: Baye, R. S., Zia, A., Merrill, S. C., Clark, E. M., Koliba, C., & Smith, J. M. (2024). Biosecurity indemnification and attitudes of United States swine producers towards the prevention of an african swine fever outbreak. Preventive Veterinary Medicine, 227, 106193.
  • Type: Peer Reviewed Journal Articles Status: Published Year Published: 2024 Citation: Baye, R. S., Zia, A., Merrill, S. C., Clark, E. M., Smith, J. M., & Koliba, C. (2024). A latent class analysis of biosecurity attitudes and decision-making strategies of swine producers in the United States. Scientific Reports, 14(1), 17427.
  • Type: Peer Reviewed Journal Articles Status: Published Year Published: 2024 Citation: Andrew, K, Zia, A., Rizzo, D. (2024) Integrating Deep Reinforcement Learning into Agent-Based Models for Predicting Farmer Adaptation under Policy and Environmental Variability. K. Arai. (ed.) Intelligent Systems and Applications: Proceedings of the 2024 Intelligent Systems Conference (IntelliSys), Lecture Notes in Networks and Systems (LNNS) 1066, pp. 221238. https://doi.org/10.1007/978-3-031-66428-1_13
  • Type: Book Chapters Status: Published Year Published: 2024 Citation: Zia, A. (2024) Towards the Deployment of Food, Energy and Water Security Early Warning Systems as Convergent Technologies for Building Climate Resilience. PP. 99-118. In Zafar Adeel and Benno B�er (editor) The Water, Energy, and Food Security Nexus in Asia and the Pacific. UNESCO: ISBN UNESCO 978-92-3-100634-0 & Springer Series: Water Security in a New World. https://doi.org/10.1007/978-3-031-29035-0
  • Type: Other Status: Published Year Published: 2024 Citation: Zia, A., Oikonomou, P. (2024) Early Warning and Early Action. PP. 18-32. Digital Technologies for Environmental Peacebuilding: Horizon Scanning of Opportunities & Risks. United Nations Environment Program. ISBN: 978-92-807-4164-3. https://wedocs.unep.org/20.500.11822/45795
  • Type: Theses/Dissertations Status: Awaiting Publication Year Published: 2024 Citation: Baye, Richmond, "African swine fever prevention in the United States: A behavioral and policy perspective" (2024), Graduate College Dissertations and Theses
  • Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Zia, Asim (2024) Harnessing Artificial Intelligence augmented Food, Energy and Water Security Early Warning Systems as Convergent Technologies for Building Peace and Climate Resilience. Third International Conference on Environmental Peacebuilding. June 2024. The Hague, Netherlands.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Zia, Asim; Benjamin Ryan, Halimeh AbuAyyash, Donna M. Rizzo, Scott Merrill, Eric Clark, Maaz Gardezi (2024) Harnessing Choice Experiments to Elicit Preferred Configurations of Trustworthy AI augmented Decision Support Systems (AI-DSS) for Crop Certified Advisors. 12th International Congress on Environmental Modelling and Software. June 2024. East Lansing MI.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Andrew, K, Zia, A., Rizzo, D. (2024) Integrating Deep Reinforcement Learning into Agent-Based Models for Predicting Farmer Adaptation under Policy and Environmental Variability. 2024 Intelligent Systems Conference (IntelliSys), September 2024. Amsterdam, Netherlands
  • Type: Conference Papers and Presentations Status: Accepted Year Published: 2025 Citation: Asim Zia, Richmond Baye, Scott Merrill, Eric Clark, Jackson Dean, Samuel Rosenblatt, Nick Cheney, Laurent Hebert-Dufresne, Julie Smith. (2025) Testing alternate biosecurity policy mechanisms to overcome moral hazard problem in indemnifying cattle producers. Conference of Research Workers in Animal Disease (CRWAD). January 2025. Chicago, IL.
  • Type: Theses/Dissertations Status: Published Year Published: 2024 Citation: Rosenblatt, Samuel Frederick, "Pragmatic interventions against epidemics on networks" (2024). Graduate College Dissertations and Theses. 1956. https://scholarworks.uvm.edu/graddis/1956
  • Type: Theses/Dissertations Status: Published Year Published: 2024 Citation: Cercena, Gian, "Investigating Swine Farm Disease Spread by a Large Agent-Based Model" (2024). UVM Patrick Leahy Honors College Senior Theses. 619. https://scholarworks.uvm.edu/hcoltheses/619
  • Type: Conference Papers and Presentations Status: Accepted Year Published: 2025 Citation: Baye*, Richmond; Asim Zia, Scott Merrill, Eric Clark, Julie Smith (2025) Regional variability in biosecurity investment choices in the United States: the role of sociopsychological and demographic factors. Conference of Research Workers in Animal Disease (CRWAD). January 2025. Chicago, IL.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Baye, R.S., Smith, J., Zia, A. (2024). Unraveling the nexus of farm-level biosecurity and transboundary animal disease prevention: evidence from regret minimization models. International Symposium on Veterinary Epidemiology and Economics. November 2024. Sydney, Australia
  • Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Baye, R.S., (2024). A latent class analysis of biosecurity attitudes of swine producers in the United States. Agricultural Research Policy Group (ARPG-24). January 2024. New Orleans, LA.
  • Type: Conference Papers and Presentations Status: Accepted Year Published: 2025 Citation: Johnbosco U. Osuagwu, Julia M. Smith, Columb Rigney. Evaluating FMD Spread among U.S. Dairy Cattle Premises: Findings from the InterSpread Plus Model. Conference of Research Workers in Animal Disease (CRWAD). January 2025. Chicago, IL.


Progress 09/01/22 to 08/31/23

Outputs
Target Audience:Scientists, modelers, and officials interested in obtaining a better understanding of how risk associated with infectious diseases are affected by the structure of the livestock industry and the behaviors of those involved. Further, the following stakeholders are informed by the outputs and outcomes of this project: veterinary epidemiologists, state animal health officials, and livestock industry professionals. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?Four graduate students and two undergraduate students are being trained in animal disease transmission modeling, including InterSpread Plus, ABM, as well as experimental gaming and survey data collection procedures. One undergraduate thesis will focus on the experimental game data analysis. Another undergraduate's thesis is focusing on the routing aspect of trucks between farms. The intent is to better understand how disease spreads via trucks. An enterprise license for InterspreadPlus has been procured for UVM. A virtual machine running Windows has been provisioned for use by project members who do not have direct access to a Windows platform computer. A week long immersive training workshop for the InterSpreadPlus modeling engine has been scheduled for year 3 of this project. One graduate student was sent as a representative of our team to the World Pork Expo in Des Moines, Iowa to learn more about swine production in the U.S. and to explore some potential new data sharing agreements and collaborations for model development. As a more senior graduate student, this was a learning experience in the professionalism and networking in science that is only experienced as one graduates to a more independent scientist. Additionally this graduate student arranged and went on a personal tour of a swine farm in Ohio to talk about biosecurity and see the system in action. Three graduate students have submitted their papers as lead authors for presentation at CRWAD 2024 annual conference and plan to participate and present in the forthcoming meeting. How have the results been disseminated to communities of interest?In addition to publication of journal articles and conference presentations, we have continued to interact with our advisory board members. The advisory board members include representatives from USDA APHIS, National Pork Bureau, Livestock Industry and senior academics. Two scientists of USDA APHIS participate in our weekly project meetings. Through these interactions, our team presents the latest research on the ABM, surveys and experimental games to our target audiences. Our meetings with scientists and policy makers have also served the purpose of disseminating information about ABM, games and surveys to the community of veterinary scientists who study the individual parts of our system models. These experts we have worked with have expressed interest in our model and have discussed it with their respective colleagues. In addition to the slate of system experts discussed above with which we had more formal meetings, the appearance of our PhD researcher, Samuel F. Rosenblatt at the World Pork Expo in June served to disseminate information about our project to a much wider audience of stakeholders and operatives in the swine industry through informal meetings at the expo. Our second PhD student, Richmond Baye, was invited to present at a high-level conference of agricultural policy makers in Washington DC. Another initiative to promote future result dissemination was with UVMs open-source resource office VERSO (VErmont ReSearch OSPO). We have begun discussions that will guide us to create this as an open source product in the future while still protecting the private data we have been entrusted in our data sharing agreements, and have the VERSO program director Kendall Fortney to help us do this. What do you plan to do during the next reporting period to accomplish the goals?In years 3 and 4 of our project, we will continue to develop and test the national scale ABM for simulating ASF and FMD and incorporate policy and behavioral responses from the latest survey and gaming data sets. The InterspreadPlus (ISP) model will also be analyzed for a global parameter sensitivity analysis to provide a baseline for comparing the ABM simulation results. In the ABM, our implementation of feed mills and their truck routes is expanding to include data on registered medicated feed mills in the US, with their locations publicly available. These routes are being constructed systematically using all available knowledge on the heterogeneity of routing practices gathered from expert interviews, open resources, and academic literature. Special focus is being paid on feed routing and swine shipments as they offer potentially a cost-neutral way of increasing biosecurity for the US system. Swine shipment networks are being implemented building off data-informed inference models developed by researchers at the University of Colorado, Linköping University, and The University of Warwick. However this requires data sharing agreements that will be finalized with relevant data owners. For the development process of the ABM, the implementation of a sequential disease model has begun, whereby the same model will take on new parameters and an additional set of logic to go with ASF at a given introduction time. Finally, there are several aspects of the ABM which we deem as important to the simulation of the system and have begun preliminary research for but have yet to begin implementation. Currently there are several aspects of the model which are probabilistically affected by a producer agent's "biosecurity-level", a single variable meant to represent probability of infection and spread for many types of interaction, but also their likelihood of reacting to events and new information in a way that promotes biosecurity for the system (such as calling a vet). We plan to implement this as three different variables which collectively represent biosecurity. One signaling long term investment in biosecurity, such as installation of air filters in barns, which decreases probability of infection and infectiousness in a certain set of interaction types and which does not decrease over time due to social distancing, another representing short term biosecurity investment which does decrease over time, and a third representing the probability of reacting to events and information. We also are planning on implementing airborne transmission and seasonality of disease dynamics. Airborne transmission will follow a similar network-based approach as the other interactions in that a pre-determined list of neighboring farms within a given cutoff radius will be read into the model at initiation along with their distances. Infections from airborne transmission will be limited to these interactions, rather than following a spatial dilution model or similar. Seasonality is important to incorporate as we begin data calibration from MSHMP, as seasonality currently is seen in disease trends which will make calibration difficult without it. Generally we plan to approach seasonality as changing epidemiological parameters slightly according to a schedule. As we add these additional features and expand our model to represent the national scale, new bottlenecks in the computation will arise. We will be utilizing existing runtime testing techniques and developing our own in order to identify and address them. We have also designed and scheduled the launch of the second choice experiment survey from January 2024 to March 2024. From the survey, we will develop 2 additional papers focusing on a discrete choice experiment and an extended integrated theory of planned behavior. We plan for continued development of the experimental game to incorporate more features into the simulated decision mechanism. Participant recruitment using online survey platforms will be utilized to collect behavioral data to inform the agent based modeling initiative. After the initial alpha test is completed, data will be analyzed within the team in order to scope additional updates to the experimental simulation design, before we scale up our participant recruitment phase. We plan to recruit upwards of 1000 participants for the data collection phase of our experimental game. Behavioral results from collected gaming data will be analyzed and prepared for publication in journal articles. Machine learning models will be developed for analyzing learning from the experimental game, testing study hypotheses and integration into the ABM. The biosecurity adoption strategies developed by learning agents will be compared under different policy conditions. Additional experiments will compare agents with different lengths of memory, access to different information, and more or less discounting of future rewards. These results can be compared with data from experimental games and surveys to determine the verisimilitude of learned agent strategies. Findings from the ABM experimental simulations, games and surveys will be published in journal articles, conferences as well as policy briefs.

Impacts
What was accomplished under these goals? For objective 1, design and test an ABM, the research team has continued to make significant progress in developing a national scale ABM. Three papers from this work have been accepted for presentation at CRWAD 2024 annual conference. Below, we outline specific progress and details in the development of this ABM. The general approach of our participatory ABM is trifold: First, we model the endemic disease PRRSV (Porcine Reproductive and Respiratory Syndrome) as a proxy for ASF (African Swine Fever) in the swine industry to calibrate the ABM with data gathered from multiple sources including incidence data from the Morison Swine Health Morrison Swine Health Monitoring Program. Next, we alter the model to simulate potential interventions, policy changes, and incentive structures and use agent learning informed by experimental game data to equilibrate the model to these changes. Once calibrated with PRRS data and equilibrated to pre-emptive system-wide biosecurity measures, we introduce ASF to the model to see how the adaptations the system has made to the pre-emptive system changes influences the ability of the United States to resist an ASF outbreak. The current iteration of the ABM is written in C++ using the GPU-accelerated FLAME-GPU library with Python used for additional external features. Use of our model for wide parameter sweeps is facilitated through runtime parameterization and CUDA (Compute Unified Device Architecture) processes. This language and library offer high processing speed by utilizing parallelization of agents, but little debugging capability. As such we have implemented our own debugging and output procedures to accelerate model development via output analysis using Python. Python is also used for external interfacing to and sideloading of additional features. We use Python to create a set of pre-made synthetic feedmill truck routes which we load into the model to avoid the computational cost of routing procedures in real time, and to explore more fully the effect of these routes. Reinforcement learning for ABM producer agents was developed in Python alongside the open-source C++ model. Initial agents trained using Q-Learning were shown to develop behavioral strategies similar to hand-designed strategies based on survey and experimental game data. Additional agents trained using Deep Q-Learning demonstrated less future reward discounting and learned strategies that more efficiently invested in biosecurity. Implemented advances of our model compared to prior livestock epidemiology models include modeling the base unit of infection and sale as batches of swine of the same age on a farm which can have multiple batches. Modeling the infection dynamics, transfer and sale of batches of pigs provides a meso-scale metapopulation approach allowing for realistic within-farm transmission dynamics, such as sustained infection which is characteristic of PRRS in the US in recent years, and realistic economic modeling of sales without the prohibitive computational cost of modeling individual swine. The agent-based model we are constructing requires in-depth knowledge of a highly complex and heterogenous system of swine production and commerce in the U.S. In order to bridge the gap between the technical expertise and background knowledge, we have engaged with experts from across the industry in a series of informal interviews and collaborations. Specifically, outside of the expertise of our core team at UVM, we have had a total of eight 1-2 hour interviews with experts in swine production and biosecurity from feedmills to veterinary practice: Dr. Scott Dee, Dr. Jordan Gebhart, Dr. Jason Woodworth, Dr. Lisa Becton, Dr. Stefan Sellman, Dr. Lindsay Beck-Johnson These interviews have informed many aspects of the model which are seemingly small details but where typical assumptions are not viable and can make impactful differences on models which do not include these details. For example we have developed a simulated "pooled testing" procedure with sensitivity and specificity of simulated tests determined according to the number of infectious swine and tested swine on a given premises. Similarly, the heterogeneity of when vet testing occurs and how this information is shared with interested parties is informed by these meetings and implemented. For objective 2, we have made significant progress in designing and implementing a survey and gaming experiments. We designed and implemented a survey tailored to infer heterogenous behaviors among swine producers in the US. We have analyzed the collected data resulting in the development of two scientific papers that are under peer review in high impact journals such as Nature Scientific Reports and Preventive Veterinary Medicine. We have also presented the findings at 3 major international conferences to a diverse audience including ICAS-IX and CRWAD. An advanced experimental game has been in active development using Unity software and coded in C#. This experiment is structured as an outbreak simulation with analogous mechanics to the ABM. This experimental game includes a spatial distribution of configurable animal diseases, allowing the user to make decisions based upon risk related to infection transfer between commercial systems. Supply chains with commercial entities allow for a generalizable disease spread simulation. Diseases and their corresponding epidemiological parameters, like transmission probability, can be instantiated within the simulation and then transferred via explicit through the supply chain's trucking routes to facilities in the spatially explicit system. The trucking routes that connect the producers are configurable along with their capacity to spread disease between commercial systems. A full user interface was developed to inform the player of the infectious disease risk along with a series of policy treatments such as government indemnity policies to test their influence on decision-making. Additional stochasticity is injected into the decision mechanism in the form of infection reports by other simulated farms not controlled by the player. As disease percolates throughout the supply chain, non-player farms can report the presence of an infectious disease, which is clearly rendered on the user interface. The disease reports can contain false positives, which provide additional uncertainty within the simulation based experimental game. The player weighs the risk with respect to spatial progression of the disease spread and can choose to invest their simulation dollars on biosecurity investment, which reduces disease transmission probability. The player must balance the risk of disease and the cost of biosecurity investment, with the added complexity of the government policy interventions for indemnifying culled animals. Beta testing of this experimental game inclusive of full behavioral and network data collection has been completed with participants recruited from Amazon Mechanical Turk's online crowdsourcing worker platform. Data from the beta version allowed us to test the data architecture, robustness of software (i.e., bug catching), test the epidemiological dynamics embedded within the experimental gaming framework, and develop statistical analyses to test network and policy level treatments (e.g., test the effects of conditional indemnity linked to biosecurity for altering the willingness to invest in biosecurity). For objective 3, the open source ABM has been configured to incorporate survey and gaming data. One journal article has been published and one conference paper has been accepted for presentation.

Publications

  • Type: Journal Articles Status: Published Year Published: 2023 Citation: Gabriela Bucini, Eric Michael Clark, Scott C. Merrill, Ollin Langle-Chimal, Asim Zia, Christopher Koliba, Nick Cheney, Serge Wiltshire, Luke Trinity and Julia M Smith (2023) Connecting livestock disease dynamics to human learning and biosecurity decisions. Frontiers in veterinary science. Volume 9. https://doi.org/10.3389/fvets.2022.1067364
  • Type: Conference Papers and Presentations Status: Awaiting Publication Year Published: 2024 Citation: Zia, Asim, Richmond Baye, Scott Merrill, Eric Clark, Gemma Del Rossi, Jackson Dean, Samuel Rosenblatt, Nick Cheney, Laurent Hebert-Dufresne, Julie Smith. Eliciting Behavioral Responses to Alternate Biosecurity Risk Management Strategies in a Stochastic Game. Conference of Research Workers in Animal Disease (CRWAD). January 2024. Chicago, IL.
  • Type: Journal Articles Status: Under Review Year Published: 2023 Citation: Baye, Richmond, Julie Smith, Asim Zia, Eric Clark, Scott Merrill, Chris Koliba. (Under Review) Biosecurity Indemnification and Attitudes of United States Swine Producers towards the Prevention of an African Swine Fever Outbreak. Preventive Veterinary Medicine
  • Type: Journal Articles Status: Under Review Year Published: 2023 Citation: Baye, Richmond, Julie Smith, Asim Zia, Eric Clark, Scott Merrill, Chris Koliba. (Under Review) Enhancing Food Sustainability through Latent Class Analysis of Biosecurity Attitudes and Decision- Making Strategies of Swine Producers in the United States. Nature Scientific Reports.
  • Type: Conference Papers and Presentations Status: Awaiting Publication Year Published: 2024 Citation: Jackson Dean, Eric Clark, Kevin Andrew, Scott Turnbull, Richmond Baye, Samuel F. Rosenblatt, Johnbosco Osuagwu, Asim Zia, Scott Merrill, Julie Smith, Laurent Hebert-Dufresne, Nick Cheney. Reinforcement learning for agent-based modeling of swine producer biosecurity adoption. Conference of Research Workers in Animal Disease (CRWAD). January 2024. Chicago, IL.
  • Type: Conference Papers and Presentations Status: Awaiting Publication Year Published: 2024 Citation: Rosenblatt, Samuel F., Laurent H�bert-Dufresne. Maximizing epidemics with spatial and categorical assortativities in modular geometric contagion networks. Conference of Research Workers in Animal Disease (CRWAD). January 2024. Chicago, IL.
  • Type: Conference Papers and Presentations Status: Awaiting Publication Year Published: 2024 Citation: Baye, Richmond, Julie Smith, Asim Zia, Eric Clark, Scott Merrill, Chris Koliba. Enhancing Food Sustainability through Latent Class Analysis of Biosecurity Attitudes and Decision- Making Strategies of Swine Producers in the United States. Conference of Research Workers in Animal Disease (CRWAD). January 2024. Chicago, IL.
  • Type: Conference Papers and Presentations Status: Awaiting Publication Year Published: 2024 Citation: Zia, Asim, Amy H. Delgado, Gabriela Bucini, Scott C. Merrill, Christopher Koliba, Gemma Del Rossi, Richmond S. Baye, Bo Norby, Julie M. Smith. Testing an Extended Theory of Planned Behavior to Explain Cattle Producers Intent to Comply with FMD Controls. Conference of Research Workers in Animal Disease (CRWAD). January 2024. Chicago, IL.


Progress 09/01/21 to 08/31/22

Outputs
Target Audience:Scientists, modelers, and officials interested in obtaining a better understanding of how risk associated with infectious diseases are affected by the structure of the livestock industry and the behaviors of those involved. Further, the following stakeholders are informed by the outputs and outcomes of this project:veterinary epidemiologists, state animal health officials, and livestock industry professionals. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?Four graduate students and two undergraduate students are being trained in animal disease transmission modeling, including InterSpread Plus, ABM, as well as experiemntal gaming and survey data collection procedures. How have the results been disseminated to communities of interest?In addition to publication of 5 journal articles and 3 conference presentations, we organized a mediated modeling meeting with our advisory board members on October 13 and 14, 2022. The advisory board members included representatives from USDA APHIS, National Pork Bureau, Livestock Industry and senior acamemics. During this meeting, our team presented thelatest research on the ABM, surveys and experimental games. Many members of the advisory board also presented their ongoing research on these topics, as well as reviewed the progress of our project vis a vis our research milestones. Another important part of this mediated modeling meeting was to discuss configuration of processes within the ABM and identification of ABM scenarios for comparison with current models being used by policy makers and the industry professionals. This meeting resulted in concrete recommendations for scenario and hypothesis testing during the forthcoming years 2 and 3 of this project. What do you plan to do during the next reporting period to accomplish the goals?We will continue to develop and test the national scale ABM for simulating ASF and FMD and incoporate policy and behavioral responses from the latest survey and gaming data sets. Findings from the ABM experimental simulations, games and surveys will be published in journal articles,conferences as well as policy briefs.

Impacts
What was accomplished under these goals? For objective 1, design and test an ABM, the research team signed a data and model sharing MoU with USDA APHIS and acquired InterSpread Model configuration for simulation of ASF and FMD. In parellel, we have developed an open source ABM platform in C++ and Java. This ABM simulates disease propagation in Swine Producer networks under alternate policy and behavioral strategies. Findings from this ABM have been recently published in peer reviewed articles and will be presented in the forthcoming 2023 CRWAD conference. For objective 2, develop and implement panel surveys and online experimental gaming simulations, our team has completed collection of wave 1 of a national survey instrument and findings from this survey data will be presented in the forthcoming 2023 CRWAD conference. Two experimental games have been piloted and the results from the gaming data have been published in three journal articles. For objective 3, the open source ABM has been configured to incorporate survey and gaming data. Two journal articles are under review.

Publications

  • Type: Journal Articles Status: Published Year Published: 2022 Citation: Clark EM, Merrill SC, Trinity L, Liu T-L, O'Keefe A, Shrum T, Bucini G, Cheney N, Langle-Chimal OD, Koliba C, Zia A and Smith JM (2022) Comparing behavioral risk assessment strategies for quantifying biosecurity compliance to mitigate animal disease spread. Front. Vet. Sci. 9:962989. doi: 10.3389/fvets.2022.962989
  • Type: Journal Articles Status: Published Year Published: 2022 Citation: Koliba C, Merrill SC, Zia A, Bucini G, Clark E, Shrum TR, Wiltshire S and Smith JM (2022) Assessing strategic, tactical, and operational decision-making and risk in a livestock production chain through experimental simulation platforms. Front. Vet. Sci. 9:962788. doi: 10.3389/fvets.2022.962788
  • Type: Journal Articles Status: Accepted Year Published: 2022 Citation: Tung-Lin Liu, Scott C. Merrill, Aislinn OKeefe, Eric Michael Clark, Ollin D Langle-Chimal, Luke Trinity, Trisha Shrum, Christopher Koliba, Asim Zia, Julia M Smith, Effects of Message Delivery on Cross-cultural Biosecurity Compliance: Insights from Experimental Simulations. Frontiers in veterinary science, in-press (Accepted October 2022). doi: 10.3389/fvets.2022.984945
  • Type: Journal Articles Status: Under Review Year Published: 2022 Citation: Gabriela Bucini, Eric Michael Clark, Scott C. Merrill, Ollin Langle-Chimal, Asim Zia, Christopher Koliba, Nick Cheney, Serge Wiltshire, Luke Trinity and Julia M Smith (In review) Connecting livestock disease dynamics to human learning and biosecurity decisions. Frontiers in veterinary science
  • Type: Journal Articles Status: Under Review Year Published: 2022 Citation: Asim Zia, Amy H. Delgado, Gabriela Bucini, Scott C. Merrill, Christopher Koliba, Gemma Del Rossi, Richmond S. Baye, Bo Norby, Julie M. Smith (Revise and Resubmit) Testing an Extended Theory of Planned Behavior to Explain Cattle Producers Intent to Comply with Foot-and-Mouth Disease (FMD) Control Measures. PLOS Social Epidemiology
  • Type: Conference Papers and Presentations Status: Published Year Published: 2021 Citation: EM Clark, G Bucini, SC Merrill, O Langle-Chimal, C Koliba, L Trinity, N Cheney, T Shrum, A Zia, and JM Smith. Linking experimental games with agent based models to quantify agricultural outbreak dynamics. Conference of Research Workers in Animal Disease (CRWAD). December 2021. Chicago, IL.
  • Type: Conference Papers and Presentations Status: Awaiting Publication Year Published: 2023 Citation: Asim Zia, Eric Clark, Gabriela Bucini, Richmond Baye, Scott Merrill, Christopher Koliba, Nick Cheney, Laurent Hebert-Dufresne, Julie Smith, Discovering Leverage Points for Enhancing Biosecurity in Swine Production Networks Using Agent Based Models . Conference of Research Workers in Animal Disease (CRWAD). January 2023. Chicago, IL.
  • Type: Conference Papers and Presentations Status: Awaiting Publication Year Published: 2023 Citation: Baye, Richmond, Julie Smith, Asim Zia, Eric Clark, Scott Merrill, Chris Koliba. Understanding how indemnity affects biosecurity measures by United States swine producers. Conference of Research Workers in Animal Disease (CRWAD). January 2023. Chicago, IL.