Publication bias for a small number of studies

[Andrea Ciarrocchi]

I performed a meta-analysis involving only four RCTs. What is the most correct way to deal with publication bias? I used Begg test but I know it is unreliable when only a small number of studies are included. Even funnel plots (standard error by log odds ratio) seem quite unuseful. What do you suggest?

My study compares the rate of events (yes vs no) in treated vs not treated patients.
Thank you in advance.

Andrea Ciarrocchi

[Blair T. Johnson]

I think all of the publication bias strategies work better to the extent that you have more studies, so you are stuck if you are trying a quantitative tactic.

In the bigger picture,  you could examine your individual effect sizes to see whether published studies are all significant or whether null effects entered the literature. If null effects have been published, then it would seem much harder to conclude that there is publication bias.

Many Cochrane Collaboration meta-analyses are quite small, so you might also see how they are handling the issue.