fbpx
Sign up now!
Don't show this again
Sweepstakes Rules
Download the report!Continue to Site >
or wait 7 secs

Thank you for confirming your subscription!

(And remember, if ever you want to change your email preferences or unsubscribe, just click on the links at the bottom of any email.)

We’re glad you’re enjoying Pig Health Today.
Access is free but you’ll need to register to view more content.
Already registered? Sign In
Tap to download the app
X
Share
X

REPORTS

Collect articles and features into your own report to read later, print or share with others

Create a New Report

Favorites

Read Later

Create a new report

Report title (required) Brief description (optional)
CREATE
X
NEXT
PORK POULTRY
follow us


You must be logged in to edit your profile.

Favorites Read Later My Reports PHT Special Reports
Pig Health Today is equipped with some amazing (and free) tools for organizing and sharing content, as well as creating your own magazines and special reports. To access them, please register today.
Sponsored by Zoetis

Pig Health Today | Sponsored by Zoetis

.

Novel machine-learning tool can predict PRRSV outbreaks and biosecurity effectiveness

In spite of advanced biosecurity protocols, the porcine reproductive and respiratory syndrome virus (PRRSV) continues to plague global pork producers with unpredictable, seasonal outbreaks. In the US, it ranks as the pork industry’s most costly disease at $1 billion annually. But imagine the improved outcome if a pig production company, producer or veterinarian could predict a PRRSV outbreak along with the key and ranked biosecurity measures to reduce the risk.

Those are the objectives that Gustavo Machado, DVM, PhD, assistant professor in the department of population health and pathobiology, College of Veterinary Medicine, North Carolina State University, is working on with his research group.

“Biosecurity and risk assessments are complicated efforts, and it can be overwhelming to decide where to place your focus,” Machado told participants of Zoetis’ 2020 Peer Circle webinar series. “So, we are collecting information and trying to make sense of which biosecurity measures are having the greatest impact on a farm’s PRRSV rates.”

The complexity of biosecurity makes it a prime candidate to apply machine-learning methodology and mathematical models to unlock the possibilities to make short-term predictions. This led Machado and his research group to embark on a project using machine learning to identify and rank the most relevant biosecurity practices associated with PRRSV outbreaks, as well as to enhance knowledge about transmission and benchmark key biosecurity practices within and between swine production systems over time.

The ultimate goal is to develop an easy-to-use, economically effective biosecurity tool, available to all producers and to other scientists, which will give the industry the capacity to greatly curtail PRRSV introductions. The tool is in development as open source and will be included as a biosecurity module available online.

Building the base

The project began in 2019, with funding from USDA and the Swine Health Information Center in collaboration with Daniel Linhares, DVM, PhD, associate professor of veterinary diagnostics and production animal medicine, Iowa State University. Participants included a couple of hundred sow-farm sites from across the US.

For the process, a baseline survey collects information on 40 biosecurity measures identified as being relevant to a PRRSV outbreak. “We had nearly 100 questions but whittled it down to make it more manageable for the farm. Veterinarians and producers as less likely to answer hundreds of questions,” Machado said, “and we want to make the data practical and easy to interpret but still be accurate.” More details about this phase can be found online in a recent publication.

The producer or farm manager enters the information directly into the database through a web-based app. It is set up to collect longitudinal data — or cohort data — repeating observations of the same variables over time. “We collect longitudinal data on farms between different states in the US to determine the chances of a PRRSV outbreak,” he said. “We follow farms for some time and provide instantaneous risk assessments for PRRSV and a ranked list of key on-farm biosecurity practices to be targeted.”

The program uses 80% of the data for machine learning — to train the model — and 20% to validate the model. “The algorithms run in the background and produce results all in one place, in our tool named Rapid Access Biosecurity (RAB) app — RABapp,” Machado noted.

There are two main products for participants:

  • One is a benchmarking report of the overall participants, which is oriented toward swine veterinarians. “We can take the 40 biosecurity measures and plot how each farm is performing,” Machado said. “Producers and veterinarians can use the benchmark to see ‘How do I compare? What am I doing correctly? What is my neighbor doing correctly or not?’”
  • The other product is farm-level benchmarking, which provides the predictive probability of the farm having an outbreak in a new PRRS season. In the US, the season runs from late December to July. Not only can the report identify the biosecurity risks making the farm vulnerable but also rank the areas where changes are needed. “We really want to provide farm-specific recommendations in a rapid fashion,” he said.

The program also can rank sites within a production company to identify which ones are at a high or low risk of an outbreak and where to focus biosecurity efforts. This is particularly helpful in determining a farm’s chance to remain PRRS-negative for the coming period.

From a geographic viewpoint, the data can be split to determine where farms fall — PRRS-positive or PRRS-negative — to reveal what’s prevalent through the population within an area.

So far, the project has shown the key features that are most prevalent for PRRSV outbreaks are:

  • Annual employee turnover of the site. “We see a spike in PRRS cases when annual turnover reaches the 50% mark,” Machado said.
  • Distance from main road, which was set as 3 miles — the closer the site, the higher the risk. “We looked at virus sequences and nearby farm and rarely do we see the same virus beyond 3 miles,” he added.
  • Farm location/neighborhood concentration. Naturally, the more populated and varied the surrounding production sites, the higher the risk.

Of course, some things you can’t modify or suggest changes, and all three of the risks just cited are things a farm cannot change. So, then the priority is to find interventions to minimize the outbreak risk.

It’s important to note that as more production sites access the program and the RABapp, those biosecurity rankings will change. “Machine learning is data hungry, so the more data we have the better the analysis gets, and it changes what’s important,” he added.

Next steps

To date, the program is more than 80% accurate at classifying a farm at high, mid or low risk for a new PRRSV outbreak, but the work continues. “We’re going to include more algorithms to increase the performance of the model,” Machado said. “Specifically, include the time effect of variations to improve accuracy and model stability.”

Another action underway is to develop a threshold for high/low risk categories, as well as improve the description of key features in each risk category.

The app and easy-to-use web interface are currently available to US breeding farms. The next step is to reach more producers, including those in Canada. “We are beginning a pilot project in Japan and Europe,” Machado said. “In 2021, we are going into Brazil and create a more generic tool. It will have different outcomes and limitation.”

The researchers also plan to open the program up to test it on classical swine fever and Seneca Valley virus. Eventually, this platform could be expanded to address any number of swine diseases, including transboundary diseases.

There are some limitations to the current project, as Machado pointed out:

  • Right now, the focus is only on breeding farms. It is not available for downstream farms (i.e. grow-finish farms).
  • Other risk factors are not directly considered.
  • Co-infections are not considered.

The program does not consider any PRRSV interventions, such as vaccine use, primarily because the focus is on biosecurity.

“We do have PRRSV strains entered into the model, so we’ll include that in the future,” he added. “Longer term, we hope to expand this concept and app to any disease, depending on the mode of transmission.”




Posted on August 15, 2021

tags: , ,
RELATED NEWS
  • Developing data-driven biosecurity programs

    Pablo Moreno, DVM, Swine Veterinary and Production International Service LLC, recently spoke with Pig Health Today about implementing best biosecurity practices.

  • Internal biosecurity essential to endemic disease control on pig farms

    Establishing effective internal biosecurity protocols is critical to breaking the circular spread of influenza and other pathogens between sow farms and growing sites, according to Montserrat Torremorell, DVM, PhD, University of Minnesota.

  • Researchers create algorithm to predict PEDV outbreaks

    Researchers from North Carolina State University have developed an algorithm that could give pig farms advance notice of porcine epidemic diarrhea virus (PEDV) outbreaks.




You must be logged in to edit your profile.

Share It
Looking at individual “pieces of the puzzle” means farm owners can evaluate the pros and cons of Mycoplasma elimination programs and come to a practical solution, says veterinarian David A. Baumert.

Click an icon to share this information with your industry contacts.
Google Translate is provided on this website as a reference tool. However, Poultry Health Today and its sponsor and affiliates do not guarantee in any way the accuracy of the translated content and are not responsible for any event resulting from the use of the translation provided by Google. By choosing a language other than English from the Google Translate menu, the user agrees to withhold all liability and/or damage that may occur to the user by depending on or using the translation by Google.