Jaime, it seems to me a funnel plot using the odds ratio or log odds ratio (Littenberg and Moses D) would be very useful. Also, heterogeneity testing on the same measure would seem to be an essential part of diagnostic meta-analysis. But there is a technical problem if this measure changes as the threshold changes (an asymmetrical SROC or a tilted regression line in S and D space) and it is apparent that different studies used different thresholds. I would be very interested if anyone has solved this problem.
From: Jaime Latour Perez [mailto:[log in to unmask]]
Sent: Sunday, March 30, 2003 7:32 AM
To: [log in to unmask]
Subject: Funnel plot for diagnostic studies
Dear list members:
It is generally accepted that publication bias is more prevalent for studies of diagnostic accuracy than for randomised controlled trials. However, systematic reviews of diagnostic tests usually don’t inform about this aspect, and the studies that report empirical evidence of this kind of bias are rare. I’ve found only one relevant article in Pubmed (Song F et al. Int J Epidemiol 2002 Feb; 3:88-95). Furthermore, J. Deeks in his chapter of systematic reviews of diagnostic tests (Egger et al. Systematic reviews in health care. BMJ Books 2001, p) maintain that “there is no equivalent of the funnel plot” for diagnostic tests.
I realize that the analysis of publication bias of diagnostic tests is more complex than on therapy studies (due to aspects such as heterogeneity, threshold effect, high frequency of 2 by 2 tables with “0” cells, or the mixed effect of poor designed small studies). However, I’m not sure that the plots of individual LRs versus study precision could’nt help us to suspect publication bias in systematic reviews of diagnostic studies.
So I would be grateful if somebody could answer the following question:
The funnel plot (and related methods of publication bias analysis) żis useful for detecting publication bias in a single systematic review of diagnostic tests?
Thank you in advance