Background: Systematic reviews that “compare” the accuracy of 2 or more tests often include different sets of studies for each test.
Purpose: To investigate the availability of direct comparative studies of test accuracy and to assess whether summary estimates of accuracy differ between meta-analyses of noncomparative and comparative studies.
Data Sources: Systematic reviews in any language from the Database of Abstracts of Reviews of Effects and the Cochrane Database of Systematic Reviews from 1994 to October 2012.
Study Selection: 1 of 2 assessors selected reviews that evaluated at least 2 tests and identified meta-analyses that included both noncomparative studies and comparative studies.
Data Extraction: 1 of 3 assessors extracted data about review and study characteristics and test performance.
Data Synthesis: 248 reviews compared test accuracy; of the 6915 studies, 2113 (31%) were comparative. Thirty-six reviews (with 52 meta-analyses) had adequate studies to compare results of noncomparative and comparative studies by using a hierarchical summary receiver-operating characteristic meta-regression model for each test comparison. In 10 meta-analyses, noncomparative studies ranked tests in the opposite order of comparative studies. A total of 25 meta-analyses showed more than a 2-fold discrepancy in the relative diagnostic odds ratio between noncomparative and comparative studies. Differences in accuracy estimates between noncomparative and comparative studies were greater than expected by chance (P < 0.001).
Limitation: A paucity of comparative studies limited exploration of direction in bias.
Conclusion: Evidence derived from noncomparative studies often differs from that derived from comparative studies. Robustly designed studies in which all patients receive all tests or are randomly assigned to receive one or other of the tests should be more routinely undertaken and are preferred for evidence to guide test selection.
Primary Funding Source: National Institute for Health Research (United Kingdom).