Bayesian Additive Regression Trees for Group Testing Data

Area of Science:

• Biostatistics
• Epidemiology
• Medical Diagnostics

Background:

• Group testing offers cost savings for low-prevalence disease screening compared to individual testing.
• Estimating individual disease risk from group test data is challenging due to unknown statuses and potential test errors.
• Existing regression methods often assume known covariate effect forms, risking model misspecification.

Purpose of the Study:

• To develop a flexible Bayesian framework for modeling individual disease probability using group testing data.
• To address challenges posed by unknown individual statuses and imperfect assay classifications in group testing.
• To estimate unknown covariate functions and assay accuracy probabilities within any group testing design.

Main Methods:

• Proposed a Bayesian additive regression trees (BART) framework.
• Applied BART to model individual-level disease probability with group testing data.
• Accommodated potentially misclassified test results and unknown covariate effect functions.

Main Results:

• The BART framework provides a flexible approach to group testing data analysis.
• Successfully estimated unknown covariate effects and assay classification probabilities.
• Demonstrated utility across various group testing protocols.

Conclusions:

• The proposed Bayesian additive regression trees method offers a robust and flexible solution for analyzing group testing data.
• This approach enhances the estimation of disease risk and covariate relationships, even with imperfect testing.
• The methods are applicable to diverse group testing scenarios, improving diagnostic accuracy and cost-effectiveness.