Progress 02/01/23 to 01/31/24

Outputs
Target Audience:The primary audience of this study is the swine producers or companies, who will be supported by the capacity to automatically integrate previously under-utilized data streams to facilitate their decision-making routine. This process will be emphasized with producers enrolled in the study, and the progress made through thestudywill be reported and shared with other producers to encourage more production systems to adopt the same or similar approach. Swine veterinarians can leverage the results of this study to benchmark the information to other field conditions where they practice and better understand the dynamic interactions and causation of factors impacting swine performance. Academicians can benefit from identifying major risk factors impacting swine performance. It will provide information for follow-up studies or the possibility of utilizingsomeresearch findings within their programs. The study also aims to stimulate researchers to conduct similar studies under different field conditions. Changes/Problems:The Holden Farms was included in the project to replace one of the three proposed initial swine production systems. Multiple data streams are already being collected for this new company, and the data-wrangling algorithms were made to link the informationtothe PROSPER server. As mentioned in the previous year's report, a dedicated server was purchased for storing all data related to the project, with the support of the computer science personnel on building the cybers infrastructure. What opportunities for training and professional development has the project provided?During the second year of the project, the project coordinator (Dr. Edison Magalhaes) continued to be trained on data programming and analysis, with the support of co-PIs from the Department of Statistics and Computer Science. Also, building visualization dashboards was part of the training with the support from a computer science graduate student from one of the co-PIs (Dr Hridesh Rajan). A post-doc was trained by the project coordinator on the inclusion of the third swine production company part of the PROSPER platform. This student received training on data programming for building the algorithms for Holden Farms, and outreach to communicate with the company representatives and validate the algorithms. In addition, personnel from the three swine-producing companies partnering in this project continued to participate in meetings to validate the digital platform developed for their system, training production system partners on data integration and validation. How have the results been disseminated to communities of interest?The results related to year two of the grant were shared with communities of interest in the following avenues: Four oral presentations at swine-related conferences (2at the 55thAmerican Association of Swine Veterinarians Annual Meeting, 1 at the Leman Swine Conference,and 1 at the U.S. Precision Livestock Farming Conference). Two podcasts about the project were recorded targeting swine veterinarians and swine producers(The Swine Health Blackbelt Podcast, andThe PigX Swine Extension Podcast, respectively). Two peer-reviewed papers submitted to scientific journals. Three peer-reviewed papers were published in scientific journals targeting swine stakeholders and livestock data scientists. Three oral presentations were given to the international audience, where two consisted of producers, veterinarians, and animal scientists working in the Brazilian swine industry. The goal was to encourage students and international producers to join the project and disseminate the Precision Swine Health & Production Management Concept. What do you plan to do during the next reporting period to accomplish the goals?The automated report for addressing changes in the database willbe expanded for the other two companies. Data analysis and the implementation of the data analysisresearch objectives proposed in the grant (risk factor, forecasting, and causal inferences analyses) will be applied to the Holden Farms company includedin the project. Data from thethree US swine production systems will be integrated and anonymized, as we plan to analyze it as a single anonymized aggregated file and share the information in the public dashboards.

Impacts
What was accomplished under these goals? The accomplishments for year two, in addition to year one accomplishments mentioned below,included: A third swine company was included in the grant. Holden Farms is a family-owned midwestern swine production system within the top 20 largest swine companies in the U.S. with ~70,000 sows. The data wrangling process and the algorithms for this company started to be included in August 2023 and were concluded in February 2024. Data analysis and implementation of research objectives 2 of the grant are now being investigated. An automated report for addressing changes in the database automated synchronization, reportingthenlogs of events concerning data beingchanged (deleted and inserted), updates, verification, and log process developed. This process was developed for The Hanor Company and is expected to be expanded for the other two companies. In addition to Holden Farms included in the grant, two more swine companies (onein Brazil and one inMexico) joined the project, allowing US producers to have international benchmarking data. Visualization dashboards were built forThe Hanor Company and Iowa Select Farms, available at the PROSPER project tab on the FieldEpi website (www.fieldepi.org). These dashboards are password-protected and are system-specific. Currently,6dashboards are available for each company and are replacing the pdf reports previously proposed, as they are updated once a week, and the producers and veterinarians at the companies can access the information when needed.

Publications

• Type: Other Status: Published Year Published: 2023 Citation: MAGALHAES ES. [Economic and productivity losses during the post-weaning phase: how to measure?], 2023 Swine Talks Conference, by Wisenetix. December 5th 2023, Brazil. (delivered online in Portuguese)
• Type: Journal Articles Status: Published Year Published: 2023 Citation: MAGALH�ES ES, ZHANG D, WANG C, THOMAS P, MOURA CAA, TREVISAN G, HOLTKAMP DJ, RADEMACHER CJ, SILVA GS, LINHARES DCL, 2023. Comparing forecasting models for predicting nursery mortality under field conditions using regression and machine learning algorithms. Smart Agric. Technol. 5, 100280. https://doi.org/10.1016/J.ATECH.2023.100280
• Type: Journal Articles Status: Published Year Published: 2023 Citation: MAGALHAES ES, ZHANG D, WANG C, THOMAS P, MOURA CAA, HOLTKAMP DJ, TREVISAN G, RADEMACHER C, SILVA GS, LINHARES DCL. Field Implementation of Forecasting Models for Predicting Nursery Mortality in a Midwestern US Swine Production System. Animals. 2023; 13(15):2412. https://doi.org/10.3390/ani13152412
• Type: Journal Articles Status: Published Year Published: 2023 Citation: MAGALH�ES ES, ZIMMERMAN JJ, THOMAS P, MOURA CAA, TREVISAN G, SCHWARTZ K, BURROUGH E, HOLTKAMP DJ, WANG C, RADEMACHER C, SILVA GS, LINHARES DCL (2023). Utilizing productivity and health breeding-to-market information along with disease diagnostic data to identify pig mortality risk factors in a US swine production system. Frontiers in Veterinary Sciences. Volume 10: 2023. doi: 10.3389/fvets.2023.1301392.
• Type: Journal Articles Status: Submitted Year Published: 2024 Citation: MAGALHAES ES, ZHANG D, MOURA CAA, TREVISAN G, HOLTKAMP DJ, LOPEZ WA, WANG C, LINHARES DCL, SILVA GS. Development of a pig wean-quality score using machine-learning algorithms to characterize and classify groups with high mortality risk under field conditions. Preventive Veterinary Medicine. [Under Review].
• Type: Journal Articles Status: Submitted Year Published: 2024 Citation: MAGALH�ES ES, ZIMMERMAN JJ, THOMAS P, MOURA CAA, OCONNOR A, ZHANG D, WANG C, TREVISAN G, HOLTKAMP DJ, WANG C, SILVA GS, LINHARES DCL. Measuring the impact of sow farm outbreaks with PRRS virus on the downstream mortality using causal inference methods. PLoS One. 2024. [to be submitted].
• Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: MAGALH�ES ES, ROSERO D, LINHARES DCL. Identification of drivers of mortality, identifying action items, and economic outcomes. 55th Annual Meeting of the American Association of Swine Veterinarians. Feb 24, 2024; Pg. 12-13. (delivered in-person in English). Available at: https://doi.org/10.54846/am2024/s5-4
• Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: MAGALH�ES ES, ROSERO D, LINHARES DCL. Lessons from a consolidate producer`s diagnostic, health, and productivity data. 55th Annual Meeting of the American Association of Swine Veterinarians. Feb 24, 2024; Pg. 16. (delivered in-person in English). Available at: https://doi.org/10.54846/am2024/s2-6
• Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: MAGALH�ES ES, ZHANG D, WANG C, THOMAS P, MOURA CAA, TREVISAN G, HOLTKAMP DJ, RADEMACHER C, SILVA GS, LINHARES DCL. Field implementation of forecasting models for predicting nursery mortality in one Midwestern US swine production system. 2nd US Precision Livestock Farming Conference (USPLF2023). May 22nd, 2023; Pg. 157-163. Knoxville TN, USA. (delivered in-person in English).
• Type: Other Status: Published Year Published: 2023 Citation: MAGALHAES ES. [Risk factors of swine wean-to-finish mortality]. June 28th, 2023. BRF Swine Academy 2023. Curitiba-PR, Brazil. (delivered online in Portuguese).
• Type: Other Status: Published Year Published: 2023 Citation: MAGALHAES ES. The economic and productivity impact of the major post-weaning risk factors, 2023 Leman Conference, Brazilian Symposium (session in Portuguese). Saint Paul, United States. September 2023.

Progress 02/01/22 to 01/31/23

Outputs
Target Audience:Swine producers/companies: swine producers are the main audience of this study andwill be supported by providing the capacity to automatically integratedpreviously under-utilized data stream tosupporttheirdecision-making routine. This process will beemphasized with producers enrolled in the current study, with the progress made through this study being reported and shared with other producers, to recruit more production systems to utilize the same or similar approach. Swine Veterinarians: bysharing the results of the findings in the present study, swine veterinarians will leverage/ benchmark such information to other field conditions where they practice to support decisions and better understandthe dynamic interactions/causation of factors impacting swine performance. Swine researchers: our efforts toward identifying the major risk factors impacting swine performance will provide information toother researchers for follow-up studies or the possibility ofutilizingsome of the research findings within their programs. There is also the opportunity to stimulate other researchers to conduct similar studiesunder different field conditions. Changes/Problems:One of the three companies initially proposed to be includedthis project was has not granted us data sharing permissions. There aretwo other companies in Brazil and one in Mexico who are interested in joining this study and could be included in the grant. We will continue to work to include another US swine production system. We did not use BOA cyber infrastructure, but we purchased a dedicated server for this project, with the support of the software developers and co-PIs from the Computer Science Departmentto build the cyber infrastructure of the server. What opportunities for training and professional development has the project provided?During year one of the project, a significant amount oftraining and professional development was conducted with the project coordinator (Dr. Edison Magalhaes) concerning data programming. The support of co-PIs from the Department of Statistics and Computer Science was crucial in this step, as the development of end-to-end automated data wrangling pipeline was desired. In addition, personnel from the two swine-producing companies partnering in this project participated in all meetings held to validate the digital platform developed for their system, training production system partnerson data integration and validation. Furthermore, the type of analysis conducted in this study was an innovative approach implemented in these companies, thus training their research personnel on utilizing the data from the master table for purposes unrelated to this project. For example, the master table developed for The Hanor Company is being utilized in their nutrition-related internal studies and biosecurity programs. How have the results been disseminated to communities of interest?The results related to year one of the grant were shared with communities of interest in the following avenues: Four oral presentations atswine-related conferences (3 at the American Association of Swine Veterinarians Annual Meeting) and one at the Iowa State University Swine Disease Conference). Two publications in digital magazines targeting swine producers and veterinarians (National Hog Farmer magazineand Benchmark Magazine) Two peer-review papers were submitted to scientific journals targeting swine veterinarians and swine data scientists. Two oral presentations were given to international audiences to encourage international producers to join the project, bringing potentially different perspectives to the consideration of the US swine producers. What do you plan to do during the next reporting period to accomplish the goals?Develop a report of the database automated synchronization, where logs of events are reportedconcerning data being changed (deleted and/or inserted), updates, verification, and log process developed. This process is currently being conducted manually and without a report to notify/alert such events. Develop an automated data connection between our server and the ISF data, as the current data is sent automatically through e-mail, which is thenautomatically imported into SAS. However, we believe this process is not sustainable over time with this company, and data import through API should be pursued. Incorporate at least one more swine production company intothe project. We have two swine companies in Brazil and one in Mexico interested in joining the project and willing to share data. Bringing international producers to the program allows US producers to understand the competitive advantages of international producing at high standards (i.e., achieving great results and/or taking advantage of other data streams). Develop a data wrangling pipeline to receive data from a website where the company personnel upload tables through Excel template spreadsheets, and these tables go to our server automatically to be integrated into a master table. This aims to explore multiple data acquisition approaches because not all farms would have API data capability.

Impacts
What was accomplished under these goals? The sustainability of swine operations depends on farm productivity, which is highly impacted by several factors occurring simultaneously throughout the life cycle of a pig (~ 6 months). Even though swine producers capture most of the data related to these factors, data streams are often disconnected, making it challenging to analyze them collectively to characterize the predictors of swine performance. In this project, we developed the Predictors of Swine Performance (PROSPER) digital platform, a system to capture, integrate, and analyze multiple data streams related to health, management, and environment, such as mixing pigs from different sources, space allowance, and nutrition for swine companies. We conducted this process for two prominent swine producers in the US who provided access to their data, allowing us to build system-specific algorithms for the aforementioned data-wrangling procedures. Once the data was connected, PROSPER generated an automated master table report, including a "background check" of all groups of pigs marketed, making it possible to analyze multiple factors impacting swine performance simultaneously. The preliminary results supported the decision-makers in each company in understanding factors impacting the productivity of pigs under their reality. One example was an analysis of groups marketed between 2021 and 2022 for Iowa Select Farms (the fourth largest swine producer in the US) focused on factors impacting swine wean-to-finish mortality. We identified that, for example, when a bacterial pathogen (Mycoplasma hyopneumoniae) was present in sow farms, it increased their grow-finish mortality by 2%. This result led the decision-makers to strategically plan a pathogen elimination protocol from their farms for 2023-2024. The PROSPER platform was developed in the first year of the project. In contrast to the original proposal, which planned to utilize an existing cyber infrastructure, a server was purchased specifically for this goal. This improvement provided superior data management and security throughout and beyond the project duration. The cyber-infrastructure was developed using data from The Hanor Company, which provided API access to our team. After synchronizing to the Hanor database and transferring data to our server, a safe local copy of the data was created. This backup process continues to be conducted daily, and four different data verification processes are required for all new data, as follows: Insert new records into the local copy database. Modify changed records. Wipe "deleted" records. Log the changes. The deleted records are stored in a separate database within our server for further verification while "changed" records are maintained in the previous and most recent versions for data comparison. Following the data capture and validation steps, SAS algorithms were developed to manage, clean, match, merge, and verify all 90 tables provided by the partnering companies in this project. The algorithms developed for the Hanor company consisted of a script comprised of 30,000 rows of data-programming codes, currently running once a week on the server's local copy database. The process of constructing the Hanor system-specific master table utilized SAS 9.4 to scan dates and unique identifiers across multiple data streams available in the database to connect different information from birth-to-market for all groups of pigs marketed within this company. This information comprised productivity data, health status, medication and vaccination records, diagnostics, carcass performance, environmental reports, and infrastructure information. The processes mentioned above are currently being conducted for one (The Hanor Company) of the three swine companies originally proposed on the grant. The second company, Iowa Select Farms (ISF), still needs to gain the ability to provide data through API connection with our server. However, the information is received weekly through automated e-mails (a total of 74 files are sent through e-mails weekly), sending all required files to the project-specific user. System-specific algorithms using SAS software were developed for this company (ISF) to conduct data importation, management, cleaning, and integration. The third swine company originally proposed in the grant is in the process of inclusion in the project. After the conclusion of the aforementioned steps, two master tables were built (one for each company), containing all retrospective data for all groups of pigs marketed over four years in each company. These master tables are then utilized on the next research objective of the grant, focusing on analyzing the data tables to identify the major drivers of wean-to-finish (W2F) mortality and average daily weight gain (ADWG). Forecasting models for swine nursery mortality were developed in this study with the support of the co-PI`s from the ISU Department of Statistics. For this purpose, the initial model was developed for Iowa Select Farms, where multiple machine-learning and regression models were trained and tested (cross-validated) on two years of nursery closeout data and later tested on nem-incoming data of pigs recently weaned into nursery sites. Two publications were generated from this research objective, with the first one (submitted to the Smart Agricultural Journal) discussing and describing the methodology of comparing the performance of different forecasting models to predict nursery mortality. The second publication (submitted to the USPLF conference and later recommended for publication in a special edition of the Animals Journal concerning the USPLF conference) focused on the field implementation and implications of a forecasting model for a US swine production system. Causal inference models were developed and tested for one of the production systems in this study (The Hanor Company), to measure the causal effect of PRRS virus epidemic groups in the sow farm on the downstream nursery mortality of the progenies, compared to nursery mortality for non-epidemic PRRS groups, based on analysis of an observational dataset. For this purpose, a causal diagram was developed within our group (PI and CO-PIS), which served as the foundation for selecting the set of covariates to be controlled when measuring the causal effect of PRRS virus in sow farms. A doubly robust inverse probability weighting method was utilized in this step, and we observed that the causal effect of PRRS was twice as high as the original effect when conducting univariate analysis. A manuscript is in preparation concerning these results to be submitted yet this year. Strategic collaboration and extension activities were conducted during this first year with the two swine companies enrolled in the project, where we focused on effective dissemination of the knowledge generated by the research objectives once we finished the development of the cyber-infrastructure on data integration. In summary, the work conducted during the first year of the grant successfully developed the analytical capabilities related to the three research objectives initially proposed (risk factor analyses, causal inference, and forecasting) for two swine production systems. More examples of the experiences in implementing technologies and applied knowledge to support producers making data-driven decisions will be collected and shared during the second and third years of the grant. Altogether, the activities conducted in year 1 demonstrated the successful development and implementation of the PROSPER platform for the two aforementioned producers, thus promoting the Precision Animal Agriculture concept. We look forward to continuing to develop this project through years 2-3.

Publications

• Type: Journal Articles Status: Submitted Year Published: 2023 Citation: Magalhaes, E.S., Zhang, D., Wang, C., Thomas, P., Moura, C.A.A., Trevisan, G., Holtkamp, D.J., Rademacher, C.J., Silva, G.S., Linhares, D.C.L. Field implementation of forecasting models for predicting nursery mortality in a Midwestern US swine production system. Animals Journal. Submitted, 2023.
• Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: GEBHARDT, J., MATCHAN, S., ROSS, J., DEKKERS, J., CANAVATE, S., RADEMACHER, C., JOHNSON, C., TOKACH, M., MAGALHAES, E.S., WOODWORTH, J., DEROUCHEY, J. Improving pig survivability through research and industry collaboration. 54th Annual Meeting of the American Association of Swine Veterinarians. Mar 4-7, 2023; Pg. 3-4.
• Type: Journal Articles Status: Submitted Year Published: 2023 Citation: Magalhaes, E.S., Zhang, D., Wang, C., Thomas, P., Moura, C.A.A., Trevisan, G., Holtkamp, D.J., Rademacher, C.J., Silva, G.S., Linhares, D.C.L. Comparing forecasting models for predicting nursery mortality under field conditions using regression and machine learning algorithms. Smart Agricultural Technology. Submitted, 2023.
• Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: MAGALHAES ES, ZHANG D, WANG C, THOMAS P, MOURA CAA, TREVISAN G, ZIMMERMAN JJ, RADEMACHER C, HOLTKAMP DJ, SILVA GS, LINHARES DCL. Forecasting swine nursery mortality under field conditions. 54th Annual Meeting of the American Association of Swine Veterinarians. Mar 4-7, 2023; Pg. 270-271.
• Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: MAGALHAES ES, DONOVAN T, ROSERO D, TREVISAN G, ZIMMERMAN JJ, HOLTKAMP DJ, SILVA GS, LINHARES DCL. Impact of sow farm PED outbreaks on the downstream nursery performance in the absence of PRRS acute herds. 54th Annual Meeting of the American Association of Swine Veterinarians. Mar 4-7, 2023; Pg. 47-48.
• Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: MAGALHAES ES, MOURA CAA, THOMAS P, TREVISAN G, ZIMMERMAN JJ, HOLTKAMP DJ, LINAHRES DCL, SILVA GS. Which factors are relevant to identifying high nursery mortality groups under field conditions? 54th Annual Meeting of the American Association of Swine Veterinarians. Mar 4-7, 2023; Pg. 268-269.
• Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: MAGALHAES ES, THOMAS P, MOURA CAA, TREVISAN G, HOLTKAMP DJ, ZIMMERMAN JJ, RADEMACHER C, SILVA GS, LINHARES DCL. Forecasting nursery mortality and identification of the main drivers before pig placement. 54th Annual Meeting of the American Association of Swine Veterinarians. Mar 4-7, 2023; Pg. 16-17.
• Type: Websites Status: Published Year Published: 2023 Citation: MAGALHAES E S, TREVISAN G, HOLTKAMP D J, WANG C, RADEMACHER C J, SILVA G S, LINHARES D C L. Key findings regarding post-weaning mortality. National Hog Farmer. Feb 07, 2023. Available at: https://www.nationalhogfarmer.com/animal-health/key-findings-regarding-post-weaning-mortality
• Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: MAGALHAES E S, ZHANG D, WANG C, THOMAS P, MOURA C A A, TREVISAN G, ZIMMERMANN J J, RADEMACHER C J, HOLTKAMP D J, SILVA G S, LINHARES D C L; Predicting Groups with High Nursery-age Mortality and Ranking the Most Important Risk Factors. 2022 ISU James D. McKean Swine Disease Conference; Nov 3-4, 2022; Pg. 50; Ames
• Type: Conference Papers and Presentations Status: Other Year Published: 2023 Citation: MAGALHAES ES. [Strategic use of data analysis on the decision-making process in the swine industry]. March 16th, 2023. XVII Encontro Regional Abraves PR 2023. Toledo-PR, Brazil. (delivered in-person in Portuguese).
• Type: Conference Papers and Presentations Status: Other Year Published: 2023 Citation: MAGALHAES ES. The data-driven decision-making process in the modern swine industry. April 11th, 2023. Swine Group UFRGS. Porto-Alegre-RS, Brazil. (delivered online in English).
• Type: Other Status: Published Year Published: 2023 Citation: MAGALHAES E S, TREVISAN G, HOLTKAMP D J, WANG C, RADEMACHER C J, SILVA G S, LINHARES D C L. The importance of integrating multiple data streams in the swine industry. Benchmark Spring Edition 2023 - Published by Pigchamp, Pg 30 - 31.