Bivariate random-effects meta-analysis models for diagnostic test accuracy studies using arcsine-based...

Area of Science:

• Biostatistics
• Medical Informatics
• Diagnostic Accuracy Research

Background:

• Diagnostic tests are crucial for patient classification.
• Meta-analysis synthesizes diagnostic test accuracy, often using bivariate models due to sensitivity/specificity correlation.
• The standard bivariate linear mixed model (LMM) is widely used but can be improved.

Purpose of the Study:

• To propose and evaluate novel variance-stabilizing transformations for bivariate meta-analysis of diagnostic test accuracy studies.
• To enhance the performance of the standard bivariate linear mixed model (LMM).

Main Methods:

• Introduced two variance-stabilizing transformations: arcsine square root and Freeman-Tukey double arcsine.
• Extended the standard bivariate linear mixed model (LMM).
• Conducted simulation studies comparing proposed methods against the standard LMM using performance metrics like bias, root mean square error, and coverage probability.

Main Results:

• The proposed methods demonstrated superior performance compared to the standard LMM across most simulation scenarios.
• Improvements were observed in terms of reduced bias, lower root mean square error, and better coverage probability.
• The enhanced methods performed well even when simulation data adhered to the standard LMM assumptions.

Conclusions:

• The proposed variance-stabilizing transformations offer a more robust approach for bivariate meta-analysis of diagnostic test accuracy.
• These methods provide improved statistical properties, leading to more reliable synthesis of test performance.
• The study illustrates the practical application of these enhanced methods using real-world diagnostic data.