Figure 4 presents a comparison of point estimates and confidence interval widths of summary sensitivity and specificity from univariate random effects meta-analyses using the exact binomial likelihood vs. using a normal approximation (both models fit using maximum likelihood estimation). The scatter plots idicate that point estimates using the exact binomial likelihood are generally larger compared to those using a normal approximation. A similar effect is observed for the confidence interval widths (the width is larger for estimation using the exact binomial likelihood). Please see the text for additional discussion of the reasons for these differences.

Figure 4Comparison of point estimates and confidence interval widths of summary sensitivity and specificity (logit scale) from univariate random effects meta-analyses using the exact binomial likelihood versus using a normal approximation (both models fit using MLE)

Scatter plot of estimated logit-transformed sensitivity, specificity and their corresponding confidence interval widths from univariate random effects meta-analyses using the exact binomial likelihood versus using an approximate normal likelihood to describe within-study variability CI = confidence interval width; MLE = maximum likelihood estimation.

From: Results

Cover of An Empirical Assessment of Bivariate Methods for Meta-Analysis of Test Accuracy
An Empirical Assessment of Bivariate Methods for Meta-Analysis of Test Accuracy [Internet].
Dahabreh IJ, Trikalinos TA, Lau J, et al.

NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.