A simple and robust method for multivariate meta-analysis of diagnostic test accuracy.

Area of Science:

• Medical Statistics
• Diagnostic Test Evaluation
• Meta-Analysis Methodology

Background:

• Meta-analyses of diagnostic test accuracy often combine diverse study designs (case-control, cohort).
• Existing bivariate random-effects models fail to distinguish study types, neglecting valuable prevalence data from cohort studies.
• Trivariate generalized linear mixed-effects models, while accounting for prevalence, impose restrictive assumptions on correlation structures and data distribution.

Purpose of the Study:

• To evaluate existing meta-analysis models under assumption violations.
• To propose a novel, robust method for meta-analysis of diagnostic test accuracy.
• To fully leverage information from both case-control and cohort studies in meta-analyses.

Main Methods:

• Evaluation of standard random-effects models against assumption violations.
• Development of a new, assumption-free method for joint analysis of diagnostic accuracy indices.
• Simulation studies to compare the performance of the proposed method against existing techniques.

Main Results:

• The proposed method demonstrates superior robustness to model misspecifications compared to current approaches.
• Existing models show performance degradation when their underlying assumptions are violated.
• The new method effectively utilizes information from both case-control and cohort studies.

Conclusions:

• A simple and robust meta-analysis method is presented for diagnostic test accuracy.
• This method provides valid joint inferences without relying on strong, often violated, statistical assumptions.
• The proposed approach is recommended for meta-analyses involving mixed study designs, enhancing accuracy and reliability.