Wildlife hosts predict the distribution of reported coccidioidomycosis in the western United States

This study demonstrates that the diversity of mammalian reservoirs is the strongest predictor of coccidioidomycosis (Valley fever) endemicity in the western United States, outperforming environmental variables and providing a framework to identify underreported disease risk areas by integrating wildlife distribution data with human case surveillance.

The Gist

Imagine Valley Fever (Coccidioidomycosis) as a sneaky, invisible ghost that lives in the soil of the American West. When the wind blows or the ground is disturbed, this ghost turns into dust and floats into our lungs, making us sick. For a long time, doctors and scientists have tried to map where this ghost lives, but they've mostly been looking at the wrong clues. They've been staring at human hospital records and weather reports, trying to guess where the fungus hides.

This new study is like hiring a detective who knows the local wildlife to solve the mystery.

Here is the story of what they found, broken down simply:

1. The Old Way vs. The New Clue

Previously, scientists thought the fungus was just a "soil dweller" that liked hot, dry weather. They made maps based on rain, temperature, and dirt type. But these maps were often wrong. It's like trying to find a specific type of fish by only looking at the water temperature, without knowing which fish actually live there.

The researchers realized: "If we want to find the fungus, we should look for the animals that carry it."

They discovered that certain small mammals (like kangaroo rats, pocket mice, and other burrowers) act as living incubators for the fungus. The fungus grows inside these animals, and when the animals die and decompose, they seed the soil with the fungus, ready to infect humans.

2. The "Guest List" Analogy

Think of a county (a local area) as a house party.

• The Fungus is the party crasher.
• The Mammals are the guests who invite the crasher in.

The study found that the more "guests" (different types of reservoir mammals) a county has, the more likely it is that the "party crasher" (Valley Fever) will show up.

• The Big Discovery: The number of different mammal species in an area was the single strongest predictor of whether humans would get sick there. It was a much better clue than the weather or the type of soil.
• The Result: If a county has a rich "guest list" of 10 different types of mice and rats, it's almost guaranteed to have Valley Fever. If the guest list is empty, the fungus is likely nowhere to be found.

3. The "Silent Reporters" Problem

Here is where the story gets tricky. The data the scientists used came from human hospital reports. But not everyone tells the truth (or even knows they are sick).

• Some states are like loud, organized secretaries who write down every single case (like Arizona and California).
• Other states are like distracted secretaries who forget to write things down, or don't even have a rule to report it (like Texas or Nevada).

The researchers built a special mathematical model (a "detective's notebook") that could separate real risk from reporting habits.

• They found that in places like southern Nevada, Utah, and parts of Texas, the "mammal guest list" was full, meaning the fungus should be there. But the human hospital reports were very low.
• The Conclusion: These areas are likely under-reporting the disease. The fungus is there, the animals are there, but humans just aren't getting diagnosed or telling the government.

4. Why This Matters

This study changes the game in two big ways:

1. Better Maps: Instead of guessing where the fungus lives based on the weather, we can now look at where the "host animals" live. It's like using a metal detector to find gold instead of just guessing where the ground looks shiny.
2. Saving Lives: By identifying areas where the fungus is likely present but cases aren't being reported, public health officials can go in, test people, and warn communities. They can say, "Hey, even though you haven't heard of many cases in your town, the animals there carry the fungus, so be careful when you dig or work outside."

The Bottom Line

Valley Fever isn't just a weather problem; it's an ecological problem. The fungus rides on the backs of small animals. By tracking the animals, we can find the fungus. And by realizing that some states are bad at reporting the sickness, we can find the hidden pockets of risk that were previously invisible.

It's a reminder that to understand human disease, sometimes you have to look at the mice, the dirt, and the silence of the unreported cases, not just the people in the hospital beds.

Technical Summary

1. Problem Statement

Coccidioidomycosis (Valley fever), caused by the fungal pathogens Coccidioides immitis and C. posadasii, is an emerging environmentally acquired disease with a poorly defined geographic range. Current risk mapping relies heavily on reported human case data, which suffers from significant limitations:

• Surveillance Bias: Reporting varies drastically by state (some states do not report cases to the CDC), and many infections are asymptomatic or misdiagnosed, leading to underreporting (estimated true incidence is up to 30 times higher than reported).
• Ecological Gaps: Traditional models rely on climatic and soil covariates but often overlook the "endozoan hypothesis," which posits that mammalian reservoirs maintain the fungus in the soil and seed it with spores upon death.
• Spatial Autocorrelation: Standard regression approaches struggle with the spatially autocorrelated nature of environmental pathogens.

The study aims to determine if mammalian reservoir distributions improve the prediction of coccidioidomycosis endemicity and to disentangle ecological drivers from surveillance biases.

2. Methodology

The authors developed a hierarchical Bayesian spatial model using Integrated Nested Laplace Approximation (INLA) to predict county-level endemicity across the United States (focusing on the Western U.S.).

• Data Sources:
  • Human Cases: CDC National Notifiable Diseases Surveillance System data for 2022. Endemicity was defined as a binary outcome: $\ge$10 cases per 100,000 population.
  • Mammalian Reservoirs: Species Distribution Models (SDMs) were generated for 22 mammal species previously confirmed as Coccidioides reservoirs (via culture, serology, or sequencing). The number of reservoir species present in each county was summed.
  • Covariates: Bioclimatic data (temperature, precipitation, vapor pressure deficit), soil properties (moisture, bulk density), and land cover data were extracted at the county level.
• Model Structure:
  • Likelihood: Binomial distribution with a logit link.
  • Spatial Component: A Besag-York-Mollie (BYM2) model was used to account for spatial autocorrelation at the county level, combining a spatially structured component (intrinsic conditional autoregressive) and an unstructured i.i.d. component.
  • Hierarchical Effects:
    • Fixed Effects: Population, reservoir species richness, vegetation, soil moisture, soil bulk density, maximum vapor pressure deficit (VPD), and land cover.
    • Random Effects: An i.i.d. state-level random effect was included to capture unobserved heterogeneity in reporting practices across states.
• Model Comparison: Models were compared using the Widely Applicable Information Criterion (WAIC) and Deviance Information Criterion (DIC). Sensitivity analyses were performed by varying the endemicity threshold (5, 10, and 15 cases per 100,000).

3. Key Contributions

• Integration of Host Ecology: The study successfully integrates species distribution models of mammalian reservoirs into epidemiological risk mapping, moving beyond purely abiotic (climate/soil) predictors.
• Decoupling Ecology from Surveillance: By including state-level random effects, the framework distinguishes between true ecological endemicity and artifacts caused by variable reporting infrastructure.
• Identification of Underreporting: The model identifies specific regions where environmental suitability and reservoir presence suggest high risk, but reported cases are low, indicating potential surveillance gaps.

4. Key Results

• Reservoir Richness as Primary Predictor: The number of endemic mammalian reservoir species in a county was the strongest predictor of coccidioidomycosis endemicity.
  • Log-odds ratio (LOR) per standard deviation increase: 1.702 (95% CI: 1.060–2.419).
  • This association remained robust across different endemicity thresholds.
• Limited Role of Abiotic Factors: After accounting for reservoir distributions, environmental variables such as maximum VPD, soil moisture, and land cover were not independently associated with endemicity. This suggests that reservoir presence mediates the effect of these environmental factors.
• Model Performance: The model including mammal reservoir distributions significantly outperformed models without them (Change in WAIC = 32.71), confirming that reservoir data adds substantial predictive power.
• Reporting Heterogeneity: State-level random effects revealed significant variation in reporting:
  • Over-reporting/High Reporting: Arizona, South Dakota, and Minnesota showed significantly higher reporting rates than expected.
  • Under-reporting: New Mexico, Colorado, Texas, and parts of Nevada showed lower reporting rates.
• Predicted vs. Observed Disparities: When comparing environmental-only predictions (based on reservoirs and climate) against observed human cases, the model identified "hidden" endemic zones in southern Nevada, southern/eastern Utah, southern New Mexico, western Texas, and southwestern Colorado. These areas have high predicted environmental suitability but low reported case counts.

5. Significance and Implications

• Validation of the Endozoan Hypothesis: The strong statistical link between mammalian reservoir diversity and human disease endemicity provides robust evidence that wildlife hosts play a critical role in maintaining and spreading Coccidioides.
• Public Health Surveillance: The study highlights critical gaps in current surveillance. States with low reporting rates may be underestimating their disease burden, leading to inadequate resource allocation and public awareness.
• Framework for Future Risk Mapping: The proposed framework offers a transferable method for mapping other environmentally acquired zoonotic diseases. It allows researchers to:
  1. Predict risk in areas with sparse human case data.
  2. Anticipate how shifts in wildlife distributions (due to climate change or land use) might alter disease risk.
  3. Target surveillance efforts to regions where ecological risk is high but reporting is low.

In conclusion, the paper demonstrates that incorporating wildlife reservoir data into spatial models significantly improves the understanding of Valley fever distribution, revealing that ecological drivers are often obscured by surveillance biases.