A vine copula mixed model for trivariate meta-analysis of diagnostic studies accounting for disease prevalence and non-evaluable subjects
share
A recent paper proposed a trivariate generalized linear mixed model (TGLMM) approach to handle non-evaluable index test results under the missing at random (MAR) assumption with an application to the meta-analysis of coronary CT angiography diagnostic accuracy studies. We propose the trivariate vine copula mixed model to handle non-evaluable index test results. The vine copula mixed model includes the TGLMM as a special case and can also operate on the original scale of sensitivity, specificity, and disease prevalence. The performance of the proposed methodology is examined by extensive simulation studies in comparison with the TGLMM approach. Simulation studies showed that the TGLMM approach over-estimate sensitivity and specificity when the univariate random effects are beta distributed. Under the MAR assumption, the vine copula mixed model gives nearly unbiased estimates of test accuracy indices and disease prevalence. After applying the vine copula mixed model approach to re-evaluate the coronary CT angiography meta-analysis, a vine copula mixed model with the sensitivity, specificity, and prevalence on the original scale provided better fit than the TGLMM, which models the sensitivity, specificity and prevalence on a transformed scale.
READ FULL TEXT
