Modeling the spillover risk of highly pathogenic avian influenza from wild birds to cattle in Denmark: A data-driven risk assessment framework

Here is an explanation of the research paper, translated into simple, everyday language with some creative analogies.

🦆 The Big Picture: A "Bird Flu" Weather Forecast for Cows

Imagine a severe flu virus (Highly Pathogenic Avian Influenza, or HPAIV) that usually lives in wild birds. In early 2024, this virus jumped from birds to cows in the United States, causing a massive outbreak.

Now, scientists in Denmark are worried. Denmark is a major stopover for migrating birds (like a busy airport for geese and ducks), and it has a huge number of dairy cows. But unlike the U.S., Denmark doesn't actively test cows for this virus yet. They are flying blind.

The Goal: The researchers built a "Risk Radar" to predict when and where this bird flu might jump from wild birds to Danish cows, so farmers and officials can be ready.

🛠️ How They Built the "Risk Radar" (The Method)

Since they didn't have data from Denmark yet, they used a clever trick: They used the U.S. as a practice run.

The Blueprint: They looked at the U.S. outbreaks. They knew exactly when and where the virus jumped from birds to cows there.
The Translation: They took that U.S. data and "translated" it to Denmark. They asked: "If the virus behaves the same way here as it did in the U.S., where would it hit Denmark?"
The Ingredients: To make the prediction, they mixed three main ingredients:
- Bird Traffic (eBird): How many wild birds are flying over a specific area? (Think of this as the "crowd size").
- Cow Density: How many cows are living in that area? (Think of this as the "target size").
- Virus Activity (Bird Flu Radar): Is the virus currently active in the wild bird population? (Think of this as the "storm warning").

🌧️ The Two Ways to Look at the Risk

The scientists ran the model two different ways, like looking at the same storm through two different lenses:

1. The "Bird Behavior" Lens (Frequency-Dependent)

The Analogy: Imagine a bird landing at a farm to drink water. It doesn't care if there are 10 cows or 1,000 cows; it just visits the farm once.
The Result: The risk is highest where wild birds hang out the most.
Where? Along the coastlines, near lakes, and near the German border. These are the "bird highways."

2. The "Cow Density" Lens (Density-Dependent)

The Analogy: Imagine the virus is a fire. If you have a huge stack of hay (lots of cows) in one spot, the fire spreads faster there, even if fewer birds visit.
The Result: The risk shifts toward areas with huge numbers of cows, regardless of how many birds are there.
Where? Inland areas with massive dairy farms, particularly in the west and south.

📅 When is the Danger Highest?

The model acts like a seasonal weather forecast.

The "Storm Season": The risk is highest from December to March.
Why? This is when wild birds are migrating south for the winter. It's like a massive parade of birds passing through Denmark, carrying the virus.
The "Quiet Season": In the summer, the risk drops significantly because the birds are far away.

📍 Where Should Denmark Look?

The map shows that the "hotspots" are:

The Coastlines: Where birds land and rest.
The German Border: A major entry point for migrating birds.
The "Sweet Spot": Some areas, like Northern Jutland, are risky under both models. These are the places where officials should definitely start testing cows first.

⚠️ The "Unknowns" (Why it's not 100% perfect)

The scientists are honest about the limitations. It's like trying to predict a hurricane with a map from last year:

Under-reporting: Maybe the virus jumped to cows in the U.S. more often than we know, meaning the risk is actually higher than calculated.
Indoor vs. Outdoor: Danish cows spend most of their time inside barns, while U.S. cows might be outside more. If cows are inside, they might be safer from birds.
The "Starling" Problem: The model mostly looks at ducks and geese. But what about starlings? They fly into barns and might carry the virus inside. The model might be missing this "backdoor" entry.

💡 The Bottom Line

This paper doesn't say, "The virus will jump to cows tomorrow." Instead, it says:

"Based on how the virus moved in the U.S., here is our best guess for where and when it might jump in Denmark. Don't wait for an outbreak to happen. Start watching the coastal areas and the German border during the winter migration season."

It's a data-driven early warning system designed to help Denmark catch the virus early, protect their dairy industry, and keep the food supply safe.

Here is a detailed technical summary of the paper "Modeling the spillover risk of highly pathogenic avian influenza from wild birds to cattle in Denmark: A data-driven risk assessment framework."

1. Problem Statement

Since early 2024, Highly Pathogenic Avian Influenza Virus (HPAIV) H5N1 (clade 2.3.4.4b) has spilled over from wild birds to dairy cattle in the United States, infecting over 1,000 herds. This event poses significant threats to animal and public health. However, Denmark lacks active surveillance for HPAIV in cattle (relying only on passive, suspicion-based reporting) and has no specific tools to assess spillover risk from wild birds to cattle. Given Denmark's location along major migratory flyways and its high density of dairy cattle (12.6 head/km²), there is an urgent need to quantify the risk of cross-species transmission to inform targeted surveillance and preparedness strategies.

2. Methodology

The authors developed a quantitative, data-driven risk assessment model to estimate the weekly probability of HPAIV introduction from wild birds to cattle at a grid-cell level.

Data Sources

The model integrates five key datasets, harmonized to a spatial resolution of 9 × 9 km and a temporal resolution of weekly intervals:

eBird Status and Trends: Provides relative abundance indices for 54 Anseriformes (waterfowl) species.
Gridded Livestock of the World (GLW) & Danish Central Husbandry Register (CHR): Provides cattle density data for the U.S. (calibration) and Denmark (prediction).
Bird Flu Radar: Provides a weekly indicator of HPAIV presence in European wild birds (based on a scenario tree model).
World Animal Health Information System (WAHIS): Provides confirmed spillover event data from the U.S. (used for calibration).

Model Framework

The core model estimates the weekly probability ( $P$ ) of spillover using a frequency-dependent transmission assumption (Eq. 1), which posits that contact rates depend on bird behavior and farm attractiveness rather than cattle density within a grid cell:
$P = 1 - e^{-\frac{\beta \cdot Prev \cdot A \cdot dt}{N_c}}$
Where:

$\beta$ : Transmission rate parameter.
$Prev$ : Prevalence of infected wild birds.
$A$ : Wild bird abundance.
$N_c$ : Number of cattle in the grid cell.
$dt$ : Time interval (1 week).

Calibration & Extrapolation:

Calibration: The model was calibrated using observed U.S. spillover events (Nevada and Arizona, 2024) to estimate the composite parameter $\beta_{US} \cdot Prev_{US}$ .
Extrapolation: The calibrated parameter was applied to Danish data. A scaling parameter ( $r$ $r$ ) was introduced to account for differences in prevalence and transmission rates between the U.S. and Denmark.
- $r = 1$ during outbreak periods (assuming comparable infection pressure).
- $r = 0.05$ during non-outbreak periods (reflecting lower prevalence).

Sensitivity and Structural Analysis

Parameter Uncertainty: Tested variations in the scaling parameter ( $r$ ) and the Bird Flu Radar threshold probability.
Structural Sensitivity: Compared the frequency-dependent model against a density-dependent model (where contact rates increase with cattle population size) to assess robustness.

3. Key Contributions

First Quantitative Risk Model for Denmark: Developed the first data-driven framework specifically estimating HPAIV spillover risk from wild birds to Danish cattle.
Cross-Species Calibration: Successfully calibrated a transmission model using U.S. spillover data and extrapolated it to a European context, addressing the lack of local empirical spillover data.
Dual-Transmission Assumption: Explicitly compared frequency-dependent vs. density-dependent transmission structures to highlight how assumptions about contact patterns alter spatial risk prioritization.
Operational Tool: Created a framework that can be updated in real-time using Bird Flu Radar and eBird data to guide surveillance.

4. Key Results

Temporal Risk

Peak Periods: The highest risk occurs during December to March, aligning with wild bird migration seasons.
Magnitude: Under the baseline frequency-dependent model, the expected number of spillover events in Denmark is below 0.35 cases per week.
Annual Estimates:
- 2024: Estimated 5.44 expected spillover cases (95% CI: 0.90–16.9).
- First 8 months of 2025: Estimated 1.92 expected cases.
Sensitivity: The timing of high-risk weeks (Weeks 1–13) remained stable regardless of parameter variations, though the absolute number of cases scaled linearly with the parameter $r$ .

Spatial Risk

Frequency-Dependent Model: High-risk areas are concentrated along the Danish coastline, near inland water bodies, and near the German border. Risk is driven by high wild bird abundance relative to cattle.
Density-Dependent Model: Shifts high-risk areas toward regions with high cattle densities (e.g., western and southern Jutland), while coastal areas drop in relative risk.
Robust Areas: Both models consistently identify Northern Jutland and areas near the German border as high-risk, suggesting these are robust targets for surveillance.

Sensitivity Analysis Findings

Scaling Parameter ( $r$ ): Doubling $r$ roughly doubles the expected case count, but does not change the spatial or temporal patterns.
Threshold Probability: Lowering the threshold for detecting HPAIV in wild birds increases the predicted risk and slightly extends the high-risk season.
Model Structure: The choice between frequency- and density-dependent assumptions significantly alters the spatial prioritization but not the temporal pattern.

5. Significance and Implications

Surveillance Guidance: The model provides a scientific basis for targeted surveillance. Authorities can prioritize milk testing or serological screening in coastal regions and near the German border, particularly during winter and early spring.
Early Warning: The framework serves as an early warning system, allowing stakeholders to anticipate periods of elevated risk before clinical cases appear.
Policy Support: By quantifying the risk, the study supports the design of biosecurity measures and resource allocation, even in the absence of confirmed local spillover events.
Scalability: The modeling framework is flexible and can be adapted for other countries or regions facing similar cross-species transmission risks.

Limitations Noted:
The model relies on U.S. calibration data and assumes comparable transmission dynamics. It does not account for potential underreporting in the U.S., the exclusion of seabirds (gulls/terns) in the Bird Flu Radar tool, or the specific biosecurity differences between indoor Danish farming and the outdoor U.S. systems where initial spillovers occurred. Future work should focus on validating these predictions with targeted surveillance data.