How do authors investigate selective publication in diagnostic test accuracy reviews?

Article type
Authors
van Enst WA1, Ochodo EA2, Leeflang MM2, Scholten RJPM1, Hooft L1
1Dutch Cochrane Centre, Academic Medical Center, Amsterdam, The Netherlands
2Department of Clinical Epidemiology, Biostatistics and Bioinformatics, Academic Medical Center, The Netherlands
Abstract
Background: Systematic reviews can be misleading when the results are affected by selective publication. For intervention reviews it is advocated to investigate the presence of publication bias graphically by funnel plots or by the use of statistical tests. These methods, however, appear to be less useful for investigating publication bias in diagnostic test accuracy (DTA) reviews and to date it’s not clear how thesemethods could be applied or interpreted in the DTA setting.

Objective: To explore what methods authors use to investigate publication bias in DTA reviews and how they interpret the results.

Methods: We have searched MEDLINE for DTA reviews published between September 2011 and January 2012. We have extracted methods that were applied to investigate publication bias (graphically or statistically), the results thereof and the author’s conclusion.

Results: We included 113 reviews whereof 44 investigated publication bias: seven explored publication bias graphically, 11 performed a statistical test and 26 did both. Funnel plots addressed the diagnostic odds ratio in 22 cases, sensitivity/specificity in 3 cases, and 8 addressed other parameters. The statistical methods to investigate publication bias were Egger’s test (n = 17), Deeks’ test (n = 10) and Begg’s test (n = 4), while multiple or other methods were used in six reviews. High risk of publication bias was identified by graph and test in four cases, only by graph in one case, only by test in seven cases and in three cases graphs and tests gave conflicting results.

Conclusions: Little is known about the actual presence and the potential impact of publication bias in DTA reviews. Statistical methods to test for publication bias in diagnostic meta-analyses have their limitations, though they are frequently applied (39%). More guidance and empirical studies on the use and interpretation of these tests are needed.