Continuous-data diagnostic tests for paratuberculosis as a multistage disease

Area of Science:

• Veterinary Epidemiology
• Diagnostic Test Evaluation
• Bayesian Modeling

Background:

• Multistage diseases require nuanced diagnostic approaches.
• Traditional diagnostic test evaluation (sensitivity, specificity) may not fully capture uncertainty in complex disease progression.
• Paratuberculosis (Johne's disease) serves as a model for a multistage infectious disease.

Purpose of the Study:

• To develop a generalizable method for interpreting continuous-data diagnostic tests in multistage diseases.
• To apply this method to paratuberculosis using fecal culture and indirect ELISA data.
• To directly incorporate test uncertainty into individual diagnostic results.

Main Methods:

• Utilized data from a Danish research project on paratuberculosis.
• Linked fecal culture testing with indirect ELISA results.
• Employed Bayesian networks with log-transformed optical densities, adjusting for covariates (parity, age at first calving, days in milk) to estimate probabilities for three infection states (non-infected, two infection stages).

Main Results:

• Successfully obtained probabilities for each infection stage based on adjusted optical density values.
• Demonstrated a method where test uncertainty is applied directly to individual results.
• This approach contrasts with aggregating uncertainty into population-level sensitivity and specificity.

Conclusions:

• The proposed Bayesian network approach offers a robust method for interpreting continuous diagnostic test data in multistage diseases.
• Directly accounting for test uncertainty at the individual level enhances diagnostic accuracy for diseases like paratuberculosis.
• This method provides a more precise assessment of infection status compared to traditional approaches.