Hi experts, i am a university student working on my thesis.. I have a dichotomous outcome that i am interested to perform Egger's test to find publication bias but while i was searching online i realise that its mainly used for the continuous outcomes.
Q1: Am i supposed to perform Egger's test and funnel plot for dichotomous outcome ?
Q2: If yes, what data do i need and which software (free) is available because I'm using revman and i understand that i can't perform egger's test on revman
1. Visual checking for asymmetry in funnel plot can be used for both continuous and dichotomous outcomes, see: https://training.cochrane.org/handbook/current/chapter-13#section-13-3-5-2 and https://handbook-5-1.cochrane.org/chapter_10/10_4_3_1_recommendations_on_testing_for_funnel_plot_asymmetry.htm. On the other hand, while Egger test can theoretically be used for dichotomous outcomes, the output may be suboptimal under certain circumstances.
2. Please check this list of free meta-analysis tools: https://toptipbio.com/free-meta-analysis-software/. Special mention goes to JASP which I know can go beyond what RevMan can do in this regard (which is funnel plot visualization) while still being as nearly as intuitive (no need for programming language competency).
The answer is a little more complicated than yes or no, as often happens in real life! See this helpful page http://openmetaanalysis.github.io/methods.html – scroll down to the section called "Assessing quality across a group of studies (PRISMA Items 15)". It has some useful advice.
Also have a look at Rücker's test, which would seem to apply in your case (but given the constraints mentioned in the openMeta web page) https://pubmed.ncbi.nlm.nih.gov/?term=17592831%2C19836925%5Buid%5D
You can do the Rückers test in R, using the package metasens – see https://cran.r-project.org/web/packages/metasens/. This may not be great news if you don't use R, however the help materials for metasens are good, and you can install RStudio to run R, which takes 90% of the pain out of getting started with it.
Finally, the whole business of measuring heterogeneity has been called into question – see Article The Dilemma of Heterogeneity Tests in Meta-Analysis: A Chall...
Sure, you can do it (given that you have enough studies - literature suggests at least 10). You will just regress the log odds ratio or log risk ratio on standard error (weighted by inverse variance).
Looking for freely available software, you can check the metabias command in the meta library in R. You can find more information here
https://ebmh.bmj.com/content/ebmental/22/4/153.full.pdf
Perhaps, there are other options that I am not aware of.
Be careful that Egger's test explores for small study effects which is used as a proxy for publication bias (but there may be other reasons why smaller studies show systematically larger effects compared to larger studies) and that even if you do not find statistically significant results, this is not a proof that you do not have small study effects especially if you have few studies.
A1: Regardless of the dichotomous or continuous outcome, both the graphical (funnel plot) and statistical tests (Egger's test/Begg's test) are necessitated to detect the publication bias.
A2:
- Data you need: Effect size and standard error.
- Free software: Meta-essential, OpenMeta[Analyst], JASP, jamovi
Note: I partially disagree with what you found over the internet. Egger's test is recommended for both dichotomous or continuous outcomes when there is no or less heterogeneity. If heterogeneity is too high, then this statistical test is highly likely to fail to detect true publication bias.
Hope you find this answer helpful.
Thanks.
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Writing