Table 11Comparative test performance: Differences in the summary logit-TPRs or logit-FPRs and corresponding standard errors or standard deviations of the posterior distributions

AnalysisTPRFPR
Difference in logitspost. SDDiff or SEDiffOdds ratioDifference in logitspost. SDDiff or SEDiffOdds ratio
Analyses in the 11 paired studies
Separate meta-analyses of the two tests
 a. Bivariate, normal likelihood0.038 (−0.543, 0.619)0.29651.04 (0.58, 1.86)−0.440 (−0.947, 0.066)0.25840.64 (0.39, 1.07)
 b. Bivariate, binomial likelihood0.112 (−0.632, 0.889)0.3827*1.12 (0.53, 2.43)−0.452 (−1.086, 0.168)0.3166*0.64 (0.34, 1.18)
Joint meta-analyses of the two tests
 c. Normal likelihood, using within-study correlation−0.006 (−0.309, 0.296)0.15440.99 (0.73, 1.34)0.480 (0.810,0.150)0.16840.62 (0.44, 0.86)
 d. Normal likelihood, ignoring within-study correlation0.051 (−0.374, 0.476)0.21691.05 (0.69, 1.61)0.427 (0.761,0.093)0.17040.65 (0.47, 0.91)
 e. Multinomial likelihood (uses within-study correlations)0.000 (−0.374, 0.447)0.2087*1.00 (0.69, 1.56)−0.443 (−0.956, 0.057)0.2533*0.64 (0.38, 1.06)
 f. Binomial likelihood (ignores within-study correlation)−0.061 (−0.432, 0.408)0.2161*0.94 (0.65, 1.50)−0.444 (−1.058, 0.094)0.2807*0.64 (0.35, 1.10)
Analyses in all 30 studies
Separate meta-analyses of the two tests
 a. Bivariate, normal likelihood0.045 (−0.414, 0.504)0.23411.05 (0.66, 1.66)0.446 (0.889,0.003)0.22610.64 (0.41, 1.00)
 b. Bivariate, binomial likelihood0.124 (−0.452, 0.724)0.2976*1.13 (0.64, 2.06)−0.451 (−1.006, 0.076)0.2730*0.64 (0.37, 1.08)
Joint meta-analyses of the two tests
 c. Normal likelihood, using within-study correlation−0.127 (−0.316, 0.062)0.09650.88 (0.73, 1.06)0.474 (0.810,0.138)0.17330.62 (0.44, 0.87)
 d. Normal likelihood, ignoring within-study correlation0.004 (−0.345, 0.354)0.17831.00 (0.71, 1.42)0.432 (0.751,0.112)0.16310.65 (0.47, 0.89)
 e. Multinomial likelihood (uses within-study correlations)−0.007 (−0.398, 0.385)0.1989*0.99 (0.67, 1.47)−0.446 (−0.939, 0.026)0.2415*0.64 (0.39, 1.03)
 f. Binomial likelihood (ignores within-study correlation)0.042 (−0.411, 0.505)0.2323*1.04 (0.66, 1.66)−0.444 (−0.953, 0.028)0.2468*0.64 (0.39, 1.03)

FPR = false-positive rate (1−specificity); SE = standard error; TPR = true-positive rate (sensitivity). The joint meta-analyses of the two tests use the unstructured variant of the between-study covariance matrix as per equation (24).

*

Standard deviation of the posterior.

Bold italic font: Statistically significant difference (p<0.05) or posterior probability >0.975 that the difference is less than 0.

From: Analysis of the Example

Cover of Evaluating Practices and Developing Tools for Comparative Effectiveness Reviews of Diagnostic Test Accuracy
Evaluating Practices and Developing Tools for Comparative Effectiveness Reviews of Diagnostic Test Accuracy: Methods for the Joint Meta-Analysis of Multiple Tests [Internet].
Trikalinos TA, Hoaglin DC, Small KM, et al.

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