How to Conduct a Meta-Analysis of Proportions in R
Вставка
- Опубліковано 3 лип 2024
- ***This tutorial has been published in a peer-reviewed journal. Check it out on my ResearchGate page: www.researchgate.net/publicat....
***The preprint of the written version of the tutorial has been posted on my ResearchGate page: www.researchgate.net/publicat....
***Please note: I have updated the code I used in the video. Check out the newest version of the code on my Github page: github.com/wnk4242/metaprops. On this page, you can download both the code and the dataset I used in the video.
Meta-analysis of proportions is a little bit different than other types of meta-analysis- it includes studies that do not use control groups. This video tutorial is an original work I produced. It is the first tutorial out there that provides a step-by-step guide showing you how to conduct a full meta-analysis of proportions using R.
Contents
0:16 Introduction
1:46 The dataset
3:07 Setting working directory
3:44 Loading packages and data
4:20 Pooling effect sizes
6:40 Interpretation of the results
8:45 Identifying outliers and influential cases
11:41 Removing outliers
12:24 Forest plots
14:11 Moderator analysis (subgroup analysis)
17:37 The fishy R-squared values
20:36 Two computational options for tau-squared values
23:39 Meta-analytic scatterplots
24:55 Meta-regression
26:39 Publication bias
If you have any questions, feel free to leave a comment below or email me at naike.wang@jjay.cuny.edu. - Наука та технологія
You explain so well and calmly, this is amazing! Thank you so much 🙏
Very well done indeed!!! Amazing tutorial
Came from Zhihu too. Really inspiring!
Hello Naike, thanks for sharing this information. It helps a lot
Thanks! Very helpful!
Thank you, this is very helpful!
Really nice work
many many thanks.
great contribution
many thanks
Dear Naike,
This is very helpful! Thank you very much for making such a nice tutorial which is 100% reproducible.
Thanks a lot for the video. 谢谢 哈哈 来自祖国的问候~~
Great !!!!!!
Awesome video and article! Is there any indication on when it will be fully published as I can only seem to find the pre-print version? Thanks!
Thank you, the tutorial was very helpful. I was wondering if the arcsine transformation you use is the same as the double arcsine transformation Barendregt et al. (2013) refered to?
Thank you very much!! One question: for the meta-regression, if we wanted to calculate the OR with CI, would we just exponentiate the estimate and its CI?
Dear Naike, Thank you very much for this fantastic tutorial. It is helping me so much as I am conducting a meta-analysis on prevalence of a rare pulmonary desease. Do you have any other tutorial like this on meta-analysis of RCT with quantitative data? Or code available in you github? I would be happy to have such help. Thank you. Marius
Fantastic video and files, thank you! Is there any way to take the data from decimal form and convert from proportions to percentages in the Forest Plots?
Hi, Eric. Sorry for the late response. I don't check my video very often. You can do this by using the "pscale" argument in the forest () function of the meta package. It is used to rescale proportions. pscale=n means that proportions are expressed as events per n observations. So, if you set pscale=1 (which is what I did in the video), 15/15398 will be shown as 0.00097415 in the forest plot. So, if you don't want to rescale proportions, use pscale=1. If you set pscale=10, the result will become 0.0097415. If you set pscale=100 (meaning events per 100 observations), it will become 0.097415. If you really want to get rid of the decimal, you have to set pscale=1000 or 10000 depending on your need. Remember, if you only change pscale, your graph will become too narrow or too wide, so you will also need to respecify the "xlim“ argument. Use this generic formula: xlim=c(0, x) in which x is a number that is a little bit bigger than the biggest 95% CI upper limit. For instance, in the example data, the biggest 95% CI upper limit is 0.00302 (Pi, 2012), so I would set xlim=c(0, 0.005). If you change pscale=1 to pscale=1000, you should also change xlim=c(0, 0.005) to xlim=c(0, 50). Don't forget to change the x-axis label after you rescale the proportions. For example, if you set pscale=100, you should use xlab = "Percentage (%)". Does this make sense to you? I hope it is helpful. The best way to learn it is to play with the code and example data yourself. I recently published the written version of the tutorial on my personal site (find the link in the description). I have improved the code in the written tutorial, so I think it's going to be very helpful to you.
does anyone know what do you do if one of the proportions is 1 and you cannot use the logit transformation
Dear Naike, this was really helpful. Great tutorial indeed. Learned a lot. I followed your tutorial and was able to reproduce the whole thing!! However, just a little bit of problem I found, probably nothing significant for you, was unable to get rid of printing of 'Common effect model' in the forest plot picture. Both the 'Common effect model' and 'Random effect model' label showing in my Forest plot graph, but can see only 'random effect model' label in your your forest plot graph though. Any suggestion would be highly appreciated. Thank you.
I managed to remove the common effects model by using the argument common = FALSE in the metaprop command.
Good video but please upload this in 1080p as the = ~ and - are indistinguishable.
Dear Naike, Thank you for the wonderful video. I have a question. I am following your steps for conducting a meta-analysis, however, I do not have subgroups in my sample. I am unable to generate a forrest plot using the code in your video. I tried using the written version of your tutorial, where you have given the code for forrest plot without subgroups which is:
pes.summary=metaprop(cases, total, authoryear, data=dat, sm="PLO",method.tau="DL", method.ci="NAsm")forest(pes.summary,xlim=c(0,4),pscale=1000,rightcols=FALSE,leftcols=c("studlab", "event", "n", "effect", "ci"),leftlabs=c("Study", "Cases", "Total", "Prevalence", "95% C.I."),xlab="Prevalence of CC", smlab="",weight.study="random", squaresize=0.5, col.square="navy",col.square.lines="navy",col.diamond="maroon", col.diamond.lines="maroon", pooled.totals=FALSE,comb.fixed=FALSE,fs.hetstat=10,print.tau2=TRUE,print.Q=TRUE,print.pval.Q=TRUE,print.I2=TRUE,digits=2)
the following error pops up:
Error: unexpected symbol in "pes.summary=metaprop(cases, total, authoryear, data=dat, sm="PLO")forest"
please help. Im stuck since days
i keep getting this message [could not find function "forest"] when i try to create r Forest plot with metafor. Any idea how to solve this error?
This is a great tutorial. Thanks for uploading it. I am currently working on a meta-analysis of proportions and there is just one single thing that does not work with the funnel plot code. Every time I try to run the code: funnel(pes.logit, yaxis="sei"), I get this error message: Error in funnel(pes.logit, yaxis = "sei") :
unused argument (yaxis = "sei")
I’ve been trying to use multiple versions of the code but still does work. Have you had to deal with this error before? How can you fix it? ty
Amazing video , we are going meta analysis of prevalence studies and when I used your script I get errors. I am a first time user. Is there a easy way
Dear Naike, thank you for your help. I am currently involved in met-analysis of heritability, which is similar to proportion in principle. Can I use this for the meta-analysis of heritability reports? I was also using this script from metaprop, but the forest plot is producing two vertical lines, i.e, both fixed and random effects are displayed though the random effect model is chosen.
NOTE:The data are fake
m
Hi Nigus. As long as your data is proportional and your meta-analysis is non-comparative (i.e. one-arm), you can use the code. Thank you！
Dear Naike, thank you for your response. It is a great help for my work.
To stop the forest plot from producing one of the two vertical lines, add comb.fixed=FALSE or comb.random=FALSE in the forest() function, depending on which vertical line (fixed-effect vs. random-effects) you would like to plot.
Wow, great, comb.fixed= FALSE worked for me. I can see only one vertical line for the random effects model.
Glad to hear that!
Before this worked and i was able to produce forest plots with no problem. Now when i try to build a forest plot with subgroup analysis, i get this error: argument is of length zero. Anyone please?
I have addressed this issue in my published version of this tutorial. Please check it out on my researchgate: www.researchgate.net/publication/375451196_Conducting_Meta-analyses_of_Proportions_in_R.