meta analysis of proportion - cumulative, subgroup analysis

[Alex Yancheva]

Hello,
I would really appreciate your advice. I am working on a meta analysis of proportions: my question is why I am not able to do cumulative analysis? I would like to do cumulative analysis on different treatment options which I would like to be ordered by duration - in descending order.

metaprop df_compete_c df_complete_t, ftt random label (namevar=study_id) power(2), cumulative(rx, by(duration) descending)

the answer I get is that cumulative is not recognised
I also tried
metaprop, subgroup(rx duration)
metaprop, cumulative(duration)
and I received a message that varlist is required.
I have been reading for the last few days how I can do this but I am able to find answer anywhere.
Many thanks

[Leonardo Guizzetti]

Welcome to the Statalist. Please spend some time reviewing the FAQ when posting questions.

-metaprop- is a user-contributed command available from SSC, as you are asked to explain. You cannot perform a cumulative meta-analysis because it simply isn't supported. In general, what commands can and cannot do are documented in their help file. See -help metaprop- for more details on it's specific capabilities.

Fortunately, a cumulative meta-analysis is not so difficult to perform "by hand", even if it might be tedious. Here's some code to get you started to show you some technique to perform the necessary looping. You can consider how you want to collect or graph said results, but that's beyond the scope of what I can provide right now. You could consider -frame post- , for example, as a way to push results into a frame for each iteration if using Stata 17 (or the older -postfile- for earlier Stata versions).

* use some example data
webuse bcg, clear
gen denc = nposc + nnegc
gen byte latgroup = lat > 36
keep trial nposc denc latgroup year

* start here
* First, define your own sort order.
* Let's define order to be by chronological order, within groups of latitude (> 36 or <= 36 degrees)
bysort latgroup (year) : gen sgorder = _n

* Loop over subgroups
forval sg = 0/1 {
  * loop over number of studies within subgroup
  summ sgorder if latgroup==`sg', meanonly
  local n_studies `r(max)'
  forval order = 1/`n_studies'{
    * do the meta-analysis with all studies up to an including the current one.
    metaprop nposc denc if latgroup==`sg' & sgorder<=`order', nograph
  }
}

[Tiago Pereira]

Alex, two aspects:

1. You can check -metaprop_one-. type:

findit metaprop_one

I found some options in metaprop_one that are not available in metaprop

2. There is some literature against the double-arcsine transformation (ftt option in your command). Check if the issue discussed in the following paper applies to your case: https://pubmed.ncbi.nlm.nih.gov/30945438/

[Alex Yancheva]

Thank you both very much.