Joint meta-analysis of two diagnostic tests accounting for within and between studies dependence

Area of Science:

• Biostatistics
• Medical Statistics
• Diagnostic Test Accuracy Research

Background:

• Meta-analysis of diagnostic test accuracy predominantly focuses on single tests.
• Recent advancements include multinomial generalized linear mixed models for paired test designs.
• Existing methods may not fully capture complex dependencies in paired diagnostic test data.

Purpose of the Study:

• To propose a novel statistical model for the joint meta-analysis of two diagnostic tests in paired studies.
• To account for within-study and between-study dependencies in diagnostic test accuracy meta-analysis.
• To enable the derivation of summary receiver operating characteristic curves on the original scale of latent proportions.

Main Methods:

• A novel joint meta-analysis model assuming independent multinomial distributions for test result combinations.
• Utilizing a one-truncated D-vine copula for the random effects distribution of latent proportions, allowing for tail dependence and asymmetry.
• The proposed model generalizes existing multinomial generalized linear mixed models.

Main Results:

• The model successfully accounts for within-study dependence from paired test application.
• It allows for modeling between-study dependencies in diagnostic accuracy meta-analysis.
• Demonstrated utility through simulation studies and a real-world meta-analysis of Down's syndrome screening tests.

Conclusions:

• The proposed D-vine copula-based joint meta-analysis model offers a flexible and powerful approach for paired diagnostic test accuracy studies.
• This methodology enhances the analysis of diagnostic test performance by capturing complex dependencies.
• It provides a valuable tool for deriving summary receiver operating characteristic curves and improving meta-analytic insights.