summary forest plot for meta-analysis

[Nicoletta Riva]

Hello,

I am doing a study-level meta-analysis with different outcomes. In addition to the standard forest plots, I would like to produce a summary forest plot, by removing individual studies and just keeping the summary effect for each outcome, and then diplay all the summary effects (obtained with random effect model) in a single graph. Is there any command that could do this in STATA?

Thanks in advance

Nicoletta

[Nerea Becerra]

Hi Nicoletta,
Did you manage to represent all the summary effect of each outcome in the same graph?


Thank you in advance.

Nerea

[Marcos Almeida]

Just wondering: if the outcomes are different amongst the studies, how can they be pooled in a meta-analysis as "summary" ?

[Nicoletta Riva]

Hello,

@Nerea: no, I did not directly manage with my version of Stata (v.12). I inputed the results of the metanalyses in a new dataset and plotted them using eclplot (clarifying of course on the left hand side which were the ouctomes of interest).

@Marcos: yes, you are right, the aim was not to pool the different outcomes, but only to show the results of different analysis. I am sorry if my initial question was not clear.
There are a number of papers in famous medical journals with this type of graph. Probably it is an easy way to graphically convey the main findings to the readers (even though from a purely statistical point of view might not be correct).

Thanks again

Nicoletta

[David Fisher]

Hi Nicoletta,

With user-written metan (or my own package, admetan, which works similarly but is more recent), this is the way I would approach this:

Firstly, I would try to have all the individual study observations, for all outcomes, in the same dataset. (You may already have done this!) (If the outcomes are very different, e.g. binary vs continuous, then you may need to convert them all to generic effect sizes and standard errors.)

Then, you can approach the problem as if it were a subgroup meta-analysis, where the different outcomes are the subgroups. That is, use the by() option; and use the summaryonly option to force the display of the summary diamonds only, not the individual studies.

By default, metan (or admetan) will attempt to meta-analyse all the data (i.e for all outcomes!) as well. You obviously do not want this; so use the option nooverall to suppress it.

Best wishes,

David.