Article type
Year
Abstract
Background: Comparisons of the diagnostic accuracy of competing tests may be based on summary curves from hierarchical summary receiver operating characteristic (HSROC) meta-regression models. However, the degree of asymmetry (shape) of the curves may not be reliably estimated, especially when the number of studies is small. Furthermore, a common shape is often presumed for different tests evaluated in a comparative meta-analysis.
Objectives: To assess the asymmetry of SROC curves and the effect on relative diagnostic accuracy when comparing tests.
Methods: Systematic reviews and meta-analyses of test accuracy in the Database of Abstracts of Reviews of Effects published between 1994 and October 2012 were identified. Using the HSROC model, we first investigated the shape of the SROC curve in a meta-analysis for each test before performing comparative meta-analyses for each test comparison. The effect of assuming common asymmetry for SROC curves of different tests was explored by fitting different HSROC meta-regression models to each test comparison. We assessed asymmetry statistically by using likelihood ratio tests and also compared summary findings from the meta-analyses.
Results: We included 57 reviews that evaluated the accuracy of two tests and provided sufficient data for meta-analyses. In meta-analyses of individual tests, the degree of asymmetry of SROC curves typically decreased as the number of included studies increased. Although there was statistical evidence (P ≤ 0.05) of differences between tests in the asymmetry of SROC curves for 16 (34%) of the 47 test comparisons where models converged, differences in estimates of relative test performance and their precision between models were generally small.
Conclusions: Evidence of asymmetry in meta-analyses with few studies is likely to be a chance finding. The assumption of common asymmetry can be appropriate when comparing the SROC curves of different tests, especially when there are few studies in the meta-analysis.
Objectives: To assess the asymmetry of SROC curves and the effect on relative diagnostic accuracy when comparing tests.
Methods: Systematic reviews and meta-analyses of test accuracy in the Database of Abstracts of Reviews of Effects published between 1994 and October 2012 were identified. Using the HSROC model, we first investigated the shape of the SROC curve in a meta-analysis for each test before performing comparative meta-analyses for each test comparison. The effect of assuming common asymmetry for SROC curves of different tests was explored by fitting different HSROC meta-regression models to each test comparison. We assessed asymmetry statistically by using likelihood ratio tests and also compared summary findings from the meta-analyses.
Results: We included 57 reviews that evaluated the accuracy of two tests and provided sufficient data for meta-analyses. In meta-analyses of individual tests, the degree of asymmetry of SROC curves typically decreased as the number of included studies increased. Although there was statistical evidence (P ≤ 0.05) of differences between tests in the asymmetry of SROC curves for 16 (34%) of the 47 test comparisons where models converged, differences in estimates of relative test performance and their precision between models were generally small.
Conclusions: Evidence of asymmetry in meta-analyses with few studies is likely to be a chance finding. The assumption of common asymmetry can be appropriate when comparing the SROC curves of different tests, especially when there are few studies in the meta-analysis.