Bivariate meta-analysis of sensitivity and specificity with sparse data: a generalized linear mixed model approach

Contrary to the approach of Reitsma et al., our approach does not require and to be large and ; such that the variances of the estimated logit transformed Se and Sp for each study can be approximated by and . [...]the generalized linear mixed model approach does not require an ad hoc continuity correction when the number of true positives, true negatives, false positives, or false negatives is zero in a study. The correlation between sensitivity and specificity is estimated as suggesting moderate negative correlation between Se and Sp, with 95% confidence interval of -0.78, -0.05 by assuming normality on Fisher's z transformation of that is, normality on , and using the delta method in SAS NLMIXED to compute the variance of Fisher's z. Simulation studies were conducted to evaluate the performance of the generalized vs. general linear random effects models for bivariate analysis of Se and Sp.