Nonparametric meta-analysis for diagnostic accuracy studies

Area of Science:

• Medical Statistics
• Diagnostic Test Evaluation
• Biostatistics

Background:

• Meta-analysis is crucial for synthesizing diagnostic accuracy studies.
• Sensitivity and specificity are correlated co-primary endpoints in diagnostic meta-analyses.
• Existing methods like bivariate logistic random effects models and copula models have limitations.

Purpose of the Study:

• To propose a new nonparametric approach for meta-analysis of diagnostic accuracy studies.
• To evaluate the performance of the proposed nonparametric method against existing approaches.
• To assess the flexibility and convergence properties of the new method.

Main Methods:

• Development of a novel nonparametric analysis approach for diagnostic accuracy meta-analysis.
• Comparison with the standard bivariate logistic random effects model.
• Comparison with an alternative approach using marginal beta-binomial distributions and copula distributions.
• Simulation studies to evaluate bias, empirical coverage, and mean squared error.
• Application to two real-world meta-analysis examples.

Main Results:

• The empirical coverage for all three approaches was generally below the nominal level.
• The proposed nonparametric model demonstrated superiority in bias, empirical coverage, and mean squared error compared to the standard model.
• The nonparametric model's performance was comparable to the copula model.
• The nonparametric approach offers greater flexibility in handling correlation structures and ensures convergence.

Conclusions:

• The novel nonparametric meta-analysis approach provides a flexible and robust alternative for diagnostic accuracy studies.
• This method addresses limitations of existing models, particularly regarding correlation structures.
• The nonparametric approach shows promising results in simulation studies, often outperforming standard methods.