Sensitivity Analysis for Publication Bias in Diagnostic Meta-Analysis of Sparsity Using the Copas t-Statistic...

Area of Science:

• Biostatistics
• Medical Informatics
• Diagnostic Test Evaluation

Background:

• Publication bias (PB) threatens meta-analysis validity by overestimating diagnostic accuracy.
• Existing methods often use normal approximations, unsuitable for sparse diagnostic data.
• The bivariate binomial model offers better properties for diagnostic meta-analysis.

Purpose of the Study:

• To develop a novel method for addressing publication bias in diagnostic meta-analysis.
• To adapt existing bias modeling techniques for the bivariate binomial framework.
• To provide a reliable approach for synthesizing diagnostic accuracy, accounting for unpublished studies.

Main Methods:

• Extended the Copas t-statistic model to address publication bias within the bivariate binomial model.
• Proposed likelihood conditional on published and estimation strategies.
• Utilized exact within-study models for improved finite sample properties.

Main Results:

• The proposed method offers an interpretable approach to address publication bias on the summary receiver operating characteristic curve.
• Demonstrated practicability on real-world diagnostic meta-analyses.
• Simulation studies evaluated the performance and robustness of the new method.

Conclusions:

• The novel method effectively addresses publication bias in diagnostic meta-analysis, particularly with sparse data.
• This approach enhances the reliability of synthesized diagnostic accuracy.
• The method provides a valuable tool for researchers conducting meta-analyses of diagnostic studies.