[R-meta] [meta] package query: Forest plot displaying the mean weighted effect size with and without the outliers(meta.cont)

[Ishtiaq Ahmed]

Dear Gerta,
Thank you for your reply. The data frame shared is not created in R just to make the code simple and reproducible. My original dataset has outliers identified via find.outliers(), influential  study analysis so, I wanted to remove from my data frame and than perform a meta-analysis to see if the effect size changes. My original data is very large and I am not able able to share it here so, I created a fake data in R and decide to remove 2 studies (by calling them outliers(just to create an example). Now I want to add the effect size to original forest plot.
P.S: I am sorry to mention in my previous mail that it is auto-generated data in R.
Regards
Ishtiaq


From: Dr. Gerta R�cker<mailto:gerta.ruecker using uniklinik-freiburg.de>
Sent: Thursday, 11 May 2023 11:35
To: R Special Interest Group for Meta-Analysis<mailto:r-sig-meta-analysis using r-project.org>
Cc: Ishtiaq Ahmed<mailto:Ishtiaq.Ahmed using vub.be>
Subject: AW: [meta] package query: Forest plot displaying the mean weighted effect size with and without the outliers(meta.cont)

Dear Ishtiaq Ahmed,

Independently of the technical question (Guido may comment on), I would like to question your approach. In my view, it does not make sense to declare two of (only) five studies as "outliers", the more so, as their confidence intervals are at least partly overlapping with those of the other three studies, and they also are not attributed much less weight (sample size) than the others. For me, the forest plot doesn't seem to justify classifying these as outliers.

It is trivial that excluding studies due to large effect sizes reduces heterogeneity and shifts the pooled effect. However, the aim of meta-analysis is not to minimize heterogeneity, but to accept and account for it. Excluding studies from analysis, also in sensitivity analyses, should be based on risk of bias, if at all, not on effect sizes.

Best,
Gerta



-----Urspr�ngliche Nachricht-----
Von: Ishtiaq Ahmed via R-sig-meta-analysis <r-sig-meta-analysis using r-project.org>
Gesendet: Donnerstag, 11. Mai 2023 10:53
An: r-sig-meta-analysis using r-project.org
Cc: Ishtiaq Ahmed <Ishtiaq.Ahmed using vub.be>
Betreff: [R-meta] [meta] package query: Forest plot displaying the mean weighted effect size with and without the outliers(meta.cont)

Dear all,
I hope you all are doing well. I am conducting a meta-analysis and am trying to produce a forest plot displaying the mean weighted effect size with and without the outliers. I looked on so many websites and tried a lot of syntaxes, however, didn't really find anything about what I am looking for.
                                 Assume there are 5 studies in the meta-analysis. Outlier analyses showed that Studies 3 and 4 are outliers leading to an overestimation of the mean weighted ES. Accordingly, I calculate a mean ES with (k = 5) and without the outlier (k = 3). I would like to display the studies sorted according to their ES and report both Mean ESs in the forest plot. I would appreciate any help and/or advice.
Study3 Study4
Mean ES (k = 5) Mean ES (k = 3)

My Reproducible code:
Ishtiaq <- data.frame( nexp = c(20, 30, 25, 18, 22),
 Pain_int_mean = c(-0.5, -0.8, -0.6, -1.2, -0.7),
 Pain_int_sd = c(0.2, 0.3, 0.25, 0.18, 0.22),
 ncon = c(18, 28, 23, 15, 20),
Pain_con_mean = c(0.3, 0.2, 0.5, 0.4, 0.6),
 Pain_con_sd = c(0.15, 0.25, 0.2, 0.18, 0.22),
 study_label = c("Study 1", "Study 2", "Study 3", "Study 4", "Study 5"))

# meta-analysis 1
Pain_meta <- metacont(nexp, Pain_int_mean, Pain_int_sd, ncon, Pain_con_mean, Pain_con_sd, studlab = study_label, data = Ishtiaq, sm = 'SMD', method.smd = "Hedges", method.tau = "REML")

# create forest plot
 forest.meta(Pain_meta, sortvar = TE, studlab = TRUE, prediction = TRUE, test.overall.random = T, random = T, fixed = F ,label.right = "Favours control", col.label.right = "black", label.left = "Favours experimental", col.label.left = "black", col.square = "darkcyan", col.diamond = "darkblue", col.diamond.lines = "cyan4", col.predict = "red", print.tau2 = F, mlab = "RE Model for All Studies (k = 5)")
#Remove studies with TE >5
Pain_meta_out_removed <- abs(Pain_meta$TE) > 5
## Then, create the meta-analysis with outliers removed
Pain_meta_no_out = metacont(nexp, Pain_int_mean, Pain_int_sd, ncon, Pain_con_mean, Pain_con_sd, studlab = study_label, data = Ishtiaq[!Pain_meta_out_removed, ], sm = 'SMD', method.smd = "Hedges", method.tau = "REML", random = TRUE, fixed = FALSE)
#add the effect size to the forest plot
????????

The addpoly works in metafor() package but I like the plots produced by forest.meta()  and it is easy to handle So, I want to ask if there is any option in the meta() package to add a polygon for ES with and without outliers?

Thanks in advance.

Best regards,

Ishtiaq Ahmed

        [[alternative HTML version deleted]]

_______________________________________________
R-sig-meta-analysis mailing list @ R-sig-meta-analysis using r-project.org
To manage your subscription to this mailing list, go to:
https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis


	[[alternative HTML version deleted]]