1
$\begingroup$

I am currently conducting a meta-analysis on four different outcome measures (A, B, C, and D). However, I am struggling with the interpretation of the (contour-enhanced) funnel plots. I examined the risk of publication bias by (1) visually inspecting the funnel plots, (2) Egger's regression test for asymmetry, and (3) Duval and Tweedie's Trim and Fill method.

Egger's regression test showed a significant result (i.e., indicating asymmetry) for outcome A, C, and D. However, when the trim and fill method is applied, only positive effect sizes are imputed (not negative correlations as would be expected if publication bias was present). Can we conclude from this that the asymmetry is likely to stem from factors other than publication bias?

Below are the funnel plots and contour-enhanced funnel plots for the four outcomes: Funnel plots

Contour-enhanced funnel plots (white area indicates non-significant results w/ p < 0.05)

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ I am sorry but I think the Egger test is correct in most cases. The funnel plots are funny, in the sense that you would expect less variability in effect in more precise studies, and this is not true for most plots. In addition, you can see like an oblique-shaped pattern suggesting small study effects. I guess you are not using dichotomous outcomes but continuous ones. In such cases the problem may have to be the way variability (ie variance) is computed in some trials. $\endgroup$
    – Giuseppe Biondi-Zoccai
    Commented Jun 5, 2020 at 14:15

1 Answer 1

Reset to default
3
$\begingroup$

The only one which would cause me to think there might be a problem would be D where the Peters' plot suggests that any missing studies would be in the area of statistical significance (the darker areas) since the two negative ones which they would balance are in that area.

As to the cause you are correct that Peters and colleagues do indeed suggests that such a finding indicates either some other type of small-study effect like study quality or publication bias based on some characteristic other than statistical significance.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.