Keeping continuous diagnostic data continuous: Application of Bayesian latent class models in veterinary research.

Area of Science:

• Veterinary epidemiology
• Statistical modeling
• Diagnostic test evaluation

Background:

• Bayesian finite mixture models (Bayesian latent class models) are increasingly used for diagnostic tests without a gold standard.
• Most veterinary literature uses these models for dichotomized outcomes.
• Continuous test outcomes, like ELISA S/P ratios, are underutilized in these models despite retaining more information.

Purpose of the Study:

• To provide a practical guide for using Bayesian finite mixture models with continuous diagnostic outcomes.
• To demonstrate the benefits of modeling continuous data over dichotomization for veterinary diagnostic tests.
• To illustrate estimation of true prevalence and evaluation of test accuracy.

Main Methods:

• Revisiting Bayesian finite mixture models for continuous diagnostic outcomes.
• Analyzing synthetic and literature datasets using these models.
• Comparing continuous outcome modeling with dichotomized approaches.

Main Results:

• Continuous Bayesian finite mixture models accurately estimate true prevalence, sensitivity, and specificity.
• Modeling continuous outcomes allows for defining optimal cut-offs post-testing based on specific needs.
• Continuous data enables individual animal-level test interpretation, unlike dichotomization.

Conclusions:

• Dichotomization is not mandatory when using Bayesian latent class analysis for diagnostic test data.
• Bayesian latent class analysis with continuous outcomes should be preferred for veterinary diagnostic tests yielding continuous results.
• This approach maximizes information utilization and enhances diagnostic test evaluation accuracy.