From mixed effects modeling to spike and slab variable selection: A Bayesian regression model for group testing data.

Area of Science:

• Epidemiology
• Biostatistics
• Infectious Disease

Background:

• Group testing reduces costs and time for disease screening by testing pooled biospecimens.
• Group testing introduces data complexity, obscuring individual disease status and complicating surveillance.
• Individual participants may have varied testing involvement, further complicating data analysis.

Purpose of the Study:

• To propose a flexible Bayesian generalized linear mixed model for disease surveillance using group testing data.
• To estimate unknown assay accuracy probabilities within the group testing framework.
• To account for heterogeneity in covariate effects across different population subgroups.

Main Methods:

• Developed a Bayesian generalized linear mixed model adaptable to any group testing protocol.
• Incorporated spike and slab priors for model selection of fixed and random effects.
• Estimated assay accuracy and addressed potential heterogeneity in covariate effects.

Main Results:

• The proposed model effectively handles complex group testing data, including imperfect test results.
• It accurately estimates assay accuracy probabilities and accounts for population subgroup variations.
• Demonstrated utility through numerical studies and application to chlamydia surveillance data.

Conclusions:

• The Bayesian model provides a robust framework for disease surveillance with group testing data.
• It enhances accuracy by estimating assay performance and addressing subgroup heterogeneity.
• This methodology improves the reliability of infectious disease surveillance, particularly for conditions like chlamydia.