Background: In meta-analyses of diagnostic test accuracy, results are often presented as separate analyses of sensitivity and specificity, but this ignores the correlation between these parameters. More complex approaches such as the bivariate approach or hierarchical summary receiver operating characteristic (HSROC) curve model do account for the correlation, but are more difficult to implement and interpret.
Objectives: To demonstrate how diagnostic accuracy may be more simply summarised using a single parameter rather than sensitivity and specificity. Summarising the diagnostic accuracy of a test in a study with a single parameter permits the use of standard meta-analysis techniques for pooling results across studies, making interpretation simpler. Four parameters will be considered: diagnostic odds ratio, diagnostic standardised mean difference, Lehmann’s index and Youden’s index.
Methods: The four parameters will be described, and shown to be related to the underlying distribution of the diagnostic test. It will be shown how all four parameters may easily be estimated using simple regression models. The four new parameters will be compared, and compared to existing methods, using commonly cited examples of diagnostic test accuracy meta-analyses.
Results: As will be seen from various examples, the four single parameter methods offer a generally robust alternative to the HSROC model. Summary results and conclusions can vary substantially between methods, demonstrating that summaries of diagnostic accuracy may be sensitive to assumptions about the underlying distribution of the diagnostic test.
Conclusions: Summarising diagnostic accuracy using one of the proposed single parameter methods offers an alternative approach to more complex methods such as the HSROC model. Results of these analyses can easily be presented on a forest plot, making interpretation of results simpler and making interpretation of heterogeneity and comparisons between different diagnostic tests straightforward.