[R-meta] proportional meta-analysis

Michael Dewey lists at dewey.myzen.co.uk
Mon Feb 5 10:33:15 CET 2018

Dear Ahmed

Comments in line below

On 04/02/2018 17:55, Ahmed Tarek wrote:
> Dear all,
> I am quite the beginner when it comes to R. I have recently started to do a meta-analysis investigating the retention of different dental fissure sealers. The dataset included: study name, length of study (2,3,5, years), total number of teeth initially sealed, number of teeth sealed at the end of study, material used for this study.

Are these independent studies at each lenght of follow-up or do studies 
give informaiton at more than one follow-up duration?

> I divided data according to follow up, and made a proportional meta analysis forest plot comparing the different materials after 2 years (another plot for 3 yr, and another for 5). I made the proportional meta analysis based on the Random effects model using the package “meta”.

I think you are using metafor, not meta?

> 1. If material X was worse than material Y, can I show that the difference was statistically significant? (considering I was comparing 5 different materials at the same time)

Yes, you need to use material type as a moderator assuming each study 
only used one material. If studies used more than one material you need 
something more complex.

> 2. I noticed the heterogeneity results (I^2) to be higher for a certain material even though on the forest plot, the results did not seem that heterogeneous when compared to other groups that reported lower heterogeneity. Is that ok? also the study weights did not necessarily represent study sample size, would that be because I used the DL method?

In random effects models the weights become more similar as the 
heterogeneity increases.

> ies.logit=escalc(xi=cases, ni=total, measure="PLO", data=dat)
> pes.logit=rma(yi,vi,data = ies.logit, method = "DL", weighted = TRUE)
> pes=predict(pes.logit, transf = transf.ilogit, digits = 5)
> print(pes,digits=5); print(pes.logit,digits=4);confint(pes.logit,digits = 2)
> pes.logit.Auto=rma(yi,vi,data=ies.logit,subset=material=="Auto",method = "DL")
> pes.logit.Light=rma(yi,vi,data=ies.logit,subset=material=="BLight",method = "DL")
> pes.logit.F=rma(yi,vi,data=ies.logit,subset=material=="F",method = "DL")
> pes.logit.GI=rma(yi,vi,data=ies.logit,subset=material=="GI",method = "DL")
> pes.Auto=predict(pes.logit.Auto, transf=transf.ilogit, digits = 5)
> pes.Light=predict(pes.logit.Light, transf=transf.ilogit, digits = 5)
> pes.F=predict(pes.logit.F, transf=transf.ilogit, digits = 5)
> pes.GI=predict(pes.logit.GI, transf=transf.ilogit, digits = 5)
> 3. For the 2 year studies, almost 100 studies were included, making the forest plot look a bit crammed, should I use another package, that could make it look better?
> 4. I also could not arrange the materials the way I wanted in the forest plot, they kept being arranged alphabetically, even though I used the command “sort=FALSE”

Did you mean to use the ordered parameter, not sorted?

> I really appreciate the help.
> Regards,
> Ahmed Bedir
> 	[[alternative HTML version deleted]]
> _______________________________________________
> R-sig-meta-analysis mailing list
> R-sig-meta-analysis at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis


More information about the R-sig-meta-analysis mailing list