Skip to content

[R-meta] Proportional meta-analysis

5 messages · Ahmed Tarek, Viechtbauer Wolfgang (STAT), Michael Dewey

#
Dear All,

In a previous email I have been explaining about a proportional meta analysis I have been performing. I am working on comparing the retention rates of different materials. I first took the approach of grouping different materials with similar follow up times , and calculating a summary proportion of all materials, however I later found out that performing a moderator analysis ( in this case ?materials?) would only compare the materials against the summary proportion (?) , which is not my aim. I aim to see how each material performed, and compare directly between them. For example, I have materials A,B,C,D and I wish to compare the retention between A and B, to see if they are significantly different. 

So I am thinking I should calculate a summary proportion for each material alone, and then perhaps conduct a chi square test? to see if the summary means of each material is different? 

I would really appreciate any input. 
Attached is the summary proportion of all materials reporting retention after 3 years. For my paper, I want to highlight that the GI material is significantly worse than the Auto material. Could I still do that with moderator analysis?

Regards,
Ahmed Bedir

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://stat.ethz.ch/pipermail/r-sig-meta-analysis/attachments/20180209/3bf0dbc9/attachment-0001.html>

-------------- next part --------------
A non-text attachment was scrubbed...
Name: forestplot_meta2yearsnew5Novembernoout.png
Type: image/png
Size: 350537 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-sig-meta-analysis/attachments/20180209/3bf0dbc9/attachment-0001.png>
1 day later
#
Dear list colleagues

before replying to this you may like to examine the identical question 
(and possible answers) posted elsewhere

https://stats.stackexchange.com/questions/327794/proportional-mata-analysis

so as to avoid duplicating effort.

Michael
On 09/02/2018 19:46, Ahmed Tarek wrote:

  
    
  
1 day later
#
Dear Ahmed,

See my comment below.
But this is exactly what a moderator analysis will do. Fit a single model to all outcomes, using 'materials' as a moderator. One level of the 'materials' factor will become the reference level to which all other levels are compared. Using contrasts, you can then also compare all other levels against each other.

Best,
Wolfgang
3 days later
#
Dear Wolfgang, 

Thank you very much for your reply. I did the subgroup meta-analysis, and it did indeed show me the relationship between the material (intercept) and the other materials. I then simply stated that the pooled estimates previously calculated where signifiant or not when compared to the intercept. 



Should I also report the numbers in the first column? ?estimate?. Are they of releveance? because I don?t seem to know how to interpret them. 

Regards,
Ahmed Bedir
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://stat.ethz.ch/pipermail/r-sig-meta-analysis/attachments/20180215/4677783c/attachment-0001.html>

-------------- next part --------------
A non-text attachment was scrubbed...
Name: Screen Shot 2018-02-10 at 11.13.11 PM.png
Type: image/png
Size: 75544 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-sig-meta-analysis/attachments/20180215/4677783c/attachment-0001.png>
#
Dear Ahmed

They are indeed of relevance as they represent the estimated effect of 
each material. So Light-polymerised sealant is on average 0.25 higher 
than your reference category with confidence interval between 0.07 and 
0.43. And so on.
On 15/02/2018 18:50, Ahmed Tarek wrote: