[R-meta] Issue with netmeta package

[Guido Schwarzer]

Am 09.08.17 um 19:21 schrieb faniapk at gmail.com:

> Dear Chrysostomos.
>
> You error arises from the configuration of netmeta::pairwise(): you 
> need to give your data in contrast form (one row for each comparison). 
> Your 4-arm trial should contribute six rows in the dataframe, not one:
>
> Topiramate 50mg(route1) vs Topiramate 50mg(route2)
> Topiramate 50mg(route1) vs Topiramate 50mg(route3)
> [...]

Dear Chrysostomos,

Disadvantage of this valid approach is that the network consists of 
three (or more likely four) Topiramate 50mg treatments:
- Topiramate 50mg(route1)
- Topiramate 50mg(route2)
- Topiramate 50mg(route3)
- Topiramate 50mg - if any other study includes Topiramate 50mg and does 
not provide information on the route


I see two alternatives:
1) Conducting a meta-regression
2) Pooling of Topiramate 50mg groups

Ad 1)

At the moment, meta-regression is not implemented in netmeta. You could 
use R package metafor for this with route as a covariate / moderator. 
However, I am not sure how to include the Topiramate 50mg group from an 
additional study without information on the route. You would have to 
assume a certain value for the covariate.

Ad 2)

In order to include a single Topiramate 50mg group in the network 
meta-analysis, results from the three treatment groups have to be 
combined. Different ways to combine or split treatment groups have been 
described by us in a paper recently accepted by Research Synthesis 
Methods (Rücker et al., 2017). In your case with binary outcomes, 
pooling of treatment groups would simply mean to add the number of 
events and number of participants (see Method 1 in Rücker et al., 2017).


Note, in your specific example, the above discussion is rather 
theoretical as you have 0 events in all four treatment groups. 
Therefore, even merging the three Topiramate 50mg groups (Method 1 in 
Rücker et al., 2017) would result in 0 events and 33 participants 
compared to 0 events and 11 participants in the Topiramate 100mg group. 
In this study without any events, risk ratio and odds ratio are not 
defined. Obviously, a continuity correction could be used to include the 
study in the analysis, however, the resulting risk ratio / odds ratio 
would largely depend on the type of continuity correction. Personally, I 
do not see meta-regression as a reasonable approach with such scarce data.

Best wishes,
Guido

Reference:
Rücker G, Cates CJ, Schwarzer G. Methods for including information from 
multi-arm trials in pairwise meta-analysis. Res Synth Methods 2017, Jul 31.
http://onlinelibrary.wiley.com/doi/10.1002/jrsm.1259/full
--
Dr. Guido Schwarzer (sc at imbi.uni-freiburg.de)
Institute for Medical Biometry and Statistics
Stefan-Meier-Str. 26, D-79104 Freiburg | Phone: +49 (0)761 203 6668
http://www.imbi.uni-freiburg.de        | Fax:   +49 (0)761 203 6680