Bayesian meta-analysis of diagnostic tests allowing for imperfect reference standards.

Area of Science:

• Medical diagnostics
• Biostatistics
• Infectious diseases

Background:

• Meta-analyses of rapid diagnostic tests (RDTs) for infectious diseases are increasingly common.
• Phase IV studies in target populations, often in resource-constrained settings, are crucial but face challenges with gold standard diagnostic tests.
• Imperfect reference standards in primary studies can lead to biased meta-analyses of RDT diagnostic accuracy.

Purpose of the Study:

• To extend the standard bivariate model for meta-analysis of diagnostic studies.
• To correct for differing and imperfect reference standards in primary studies.
• To accommodate data from studies using latent class analysis to address the absence of a true gold standard.

Main Methods:

• Utilized Bayesian methods to improve estimates of sensitivity and specificity.
• Incorporated prior information on reference test accuracy.
• Employed the deviance information criterion to detect conflicts between prior information and observed data.
• Applied the model to RDT diagnostic accuracy data for visceral leishmaniasis.

Main Results:

• The proposed Bayesian bivariate model allows for improved diagnostic accuracy estimates.
• Prior information can enhance the reliability of sensitivity and specificity estimates.
• Standard meta-analytic methods underestimated the specificity of the RDT for visceral leishmaniasis in the applied dataset.

Conclusions:

• The extended bivariate model offers a robust approach to meta-analyzing diagnostic accuracy when reference standards are imperfect.
• Bayesian methods and incorporation of prior information provide more accurate estimates of RDT performance.
• This methodology is particularly valuable for RDT evaluation in resource-limited settings.