Chronic Wasting Disease Surveillance Infrastructure for North America
Strategic Priority
Conservation & Science - Fish and Wildlife Health (“One Health”)
Project Description
Chronic wasting disease (CWD) is a fatal disease of cervids with significant ecological and economic impacts. Our current multi-state and province effort, Surveillance Optimization Project for Chronic Wasting Disease (SOP4CWD), has engineered freely available data infrastructure and data modeling capacities that unlock the the potential of “big data” science in planning disease surveillance for 24 US states and 1 Canadian province. However, CWD impacts additional cervid species and locations within North America, therefore we propose to expand the project to an international-scale system designed to provide data-driven decision support for surveillance of CWD for any interested wildlife management agency or organization in Mexico, USA, Tribal nations, and Canada. While CWD is the primary focus of the project, the proposed infrastructure and tools will be designed so they may be readily adapted for use in surveilling other wildlife diseases and/or cervid species. Shared outcomes of the project include the ability to take a continental approach in the management of diseases that threaten our wildlife legacy and human societies.
Project Facts
- Organization Name: Cornell University
- Organization Status: Other
- State: New York
- Obligation: $274,723
- Start Date: 01-01-2023
- End Date: 12-31-2023
Results
Rather than going at it alone, we decided to try a novel approach to CWD surveillance. We aimed to build a surveillance network to help wildlife agencies use the best available local and regional data to guide their annual surveillance planning. We called this network the SOP4CWD.
Led by Krysten Schuler of the Cornell Wildlife Health Lab and Sonja Christensen at Michigan State, we kicked off the project in early 2020 with an in-person meeting of stakeholders. We brought together an interdisciplinary team of modelers, wildlife agency professionals, and regulators to discuss the need of the perfect surveillance system.
We immediately realized the need for standardized terminology around CWD. The MSU team is publishing a glossary of CWD terminology to help modelers, managers, and regulators communicate with each other and their stakeholders.
The project has snowballed. In only three years, we have accumulated partners at 26 wildlife agencies and 7 research institutions coast to coast. We keep these partners in the loop via monthly project updates, which go out to a listserv of about 100 recipients.
To undertake a project this size, we first needed an organizational infrastructure. Specifically, we had to spell out the guidelines for agency participation when publishing new science using shared data.
Our first development was the data use agreement or DUA.
This DUA says, “We have a lot of different players here with many differing priorities, but everyone involved pledges to be collaborative and ethical.”
To participate, an agency collects tissues, sends them off for testing, collates their data into electronic format, signs the dua, then pushes that data into the shared storage location for later use in models.
That shared location is called the Data Warehouse. Developed in partnership with DJ Case explicitly for SOP4CWD, this system is a state-of-the-art database with user privacies, privileges, and passwords.
As a quick note - only agencies can access the Warehouse – the public can’t see any of this. The warehouse currently stores 5 data types relevant to CWD surveillance and the models.
The first type is the CWD testing data – all the records your agency collects on the CWD status of individuals from hunter harvest, roadkills, clinical suspects, etc. Key variables include species, age, sex, date and location of harvest, and CWD status.
The second type is basic farmed cervid data.
While the next step of the project is to incorporate fine-scale captive data collected by ag departments, the current status of the Warehouse only includes rudimentary non-wild cervid data, such as the number and types of facilities in each area in each year.
Similarly, the next data types are meat processors and taxidermy data from businesses dealing with cervid species. These data involve basic attributes such as counts of facilities in business in each area each year.
Recall that Sop aims to develop a surveillance framework for CWD in wild cervid herds. To do that, w need some data on the free-ranging populations of interest. The Warehouse stores demographic data such as fertility rates, harvest rates, abundance estimates, etc.
Finally, the warehouse holds expense data, defined loosely at the cost of conducting surveillance at your agency.
Once all these data are uploaded into the warehouse, they can be used in state-specific models.
This brings us to the models themselves.
The Warehouse is not just a storage facility; it has automated code to inform models using the data to produce visualizations to support agency decision-making. The first visualization doesn’t come from a model but the data itself. A benefit is that you can see across state or even international boundaries. Already you can use these outputs to form a more complete picture.
Indeed, you can use the functions in the warehouse to draw these maps by season-year, age/sex segment, and state.
You can also zoom out to see the extent of data in the Warehouse. Again, you can draw this regional map based on season, age/sex segment, etc.
Similarly, there are capacities to make maps of cervid facility data, taxidermy data, demographic data, and other various data stored in the warehouse.
I will now move on to how the Warehouse generates outputs using automated models.
The first model is the Hazard model, which quantifies the risk that CWD will be introduced somewhere across a vast jurisdiction.
The model considers known hazards within the state of interest, including potential sources of CWD introduction, such as cervid facilities and meat processors. The model next considers hazards posed by conditions or activities in neighboring jurisdictions. It further considers an assessment of risk obtained from the expert opinion of agency biologists versed in CWD. Because each of these has to do with human activity, they collapse into the anthropogenic risk score. The model further accounts for the potential that CWD will be introduced through the natural movements of cervids and converts these data into a demographic risk score. Finally, the model meshes the anthropogenic with the demographic layers to set surveillance quotas for the upcoming sampling season.
The quotas are point-based to give the greatest flexibility to the agency during collection season.
Our next model is a compartment-type epidemiological model that describes CWD behavior produced by the demographic dynamics of a live cervid population. We call it the SLEI model, short for Susceptible, Latent, Exposed, and Infectious.
The main output is R0, a summary of disease potential in a hypothetically healthy population. A R0 value above one means the introduction of the disease will lead to spreading through the live population. A R0 value below one means the introduction of the disease will likely be squashed out through population dynamics alone.
Our last model is a knapsack algorithm that we use to optimize sampling across a state or province.
The algorithm is conceptually simple. Suppose you are stuffing your pack for a hike and need to decide what to put in it. You can take an infinite set of items, but you only want the set that maximizes the benefit of the pack while staying within its volume.
Regarding CWD surveillance, we have costs, the volume of our pack, and benefits, or the usefulness of the items in the pack to our journey.
On the cost side, we have the actual cost of sampling - the transportation, the diagnostic fees, the salaries of the agency reps, etc. We can use differentials across the state to determine the cost landscape.
The benefits of sampling require a slug of additional complicated considerations, such as the location of confirmed CWD, sampling intensity, hazards, outbreak risks, regulations, etc. We came up with 35 variables that might contribute to CWD.
We collated data from two dozen agencies, including ¾ of a million deer statuses, to produce a dataset with 14000 unique county/season year combinations.
SOP4CWD continues to grow. We are currently adding additional models to the warehouse.
The first is the sample size model, which lets users compute target sizes under specific biological scenarios. One such scenario is when segments of deer congregate on the landscape at certain times per year.
A second is the habitat risk, which depicts the likelihood that a culled deer will test positive for CWD. This model predicts disease by age/sex segment and home range.
A third is an agent-based model that identifies high risk areas for CWD introduction. Our collaborator Dr. Carlos will present this model later in the conference.
We see no reason to stop growing this system. In fact, why not add more models?
