BAYESIAN GROUP TESTING WITH DILUTION EFFECTS.

Area of Science:

• Statistics
• Epidemiology
• Computational Biology

Background:

• Public health surveillance requires efficient, large-scale testing capacity, especially for infectious diseases like COVID-19.
• Existing group testing methods often struggle with dilution effects and non-binary outcomes.
• There is a need for adaptable testing strategies under varying prevalence and conditions.

Approach:

• Developed a Bayesian framework for group testing incorporating dilution effects using lattice-based models.
• Extended the framework to accommodate general test response distributions beyond binary outcomes.
• Introduced the Bayesian halving algorithm and analogous look-ahead rules for efficient test selection.

Key Points:

• The Bayesian approach effectively handles dilution effects in group testing.
• Bayesian halving demonstrates optimal convergence properties for classification.
• Group testing offers substantial savings in the number of tests compared to individual testing, even at moderate prevalence.
• A trade-off exists between the number of tests and the number of testing stages, alongside increased variability.

Conclusions:

• The developed Bayesian framework provides a robust method for group testing with dilution effects.
• The Bayesian halving algorithm and look-ahead rules offer efficient strategies for disease surveillance.
• A web-based calculator and high-performance computing methods aid in decision-making for pooling strategies.