Misclassification SIMEX in meta-analysis of accuracy of diagnostic tests

Bivariate random-effects models represent a recommended approach in meta-analysis of the accuracy of a diagnostic test compared to a gold standard. Several techniques have been proposed in the literature to properly correct for the presence of measurement errors that affect the study-specific estimates of accuracy values. Recently, SIMEX, a simulation-based approach developed in the measurement error literature, has been suggested as an accurate and computationally convenient alternative to likelihood-based solutions and has been applied to a continuous approximate normal version of the accuracy measures. Nevertheless, it is preferable to work directly with the observed binary data in terms of true/false positives/negatives. This paper investigates a modified version of SIMEX, called MC-SIMEX, useful to deal with misclassified binary data in a bivariate meta-analysis setting. Attention will be devoted to the estimation of the variance of the misclassification corrected estimates. A series of simulations comparing MC-SIMEX to likelihood-based solutions and to the SIMEX continuous counterpart highlights a satisfactory performance of the proposal, in terms of either accuracy of inferential results and computational feasibility. The methods are applied to a real meta-analysis about the accuracy of the saliva test as a diagnostic tool for SARS-CoV-2 infection.