Studies with zero events being excluded in subgroup analysis despite continuity correction

[Irene White]

Hi,
I'm doing a meta-analysis with metaprop on a group of studies in which many had zero events. I tried using a Freeman-Tukey double arcsine transformation but kept getting pooled estimates of 1.0 (95% CI: 1.0 to 1.0). A continuity correction is generating more reasonable estimates while including all studies, but when I try to do subgroup analyses, many of those studies are being excluded. Does anyone know how to fix this problem?

[Sebastian Koehler]

HI,

I have the same problem. Doing individual analyses with metaprop and cc per subgroup works well, but when using the by-subcommand, studies with zero events are excluded. Any suggestions much appreciated!

Best wishes,
Seb

[Leonardo Guizzetti]

I'm not familiar with with the user-contributed command -metaprop-, but two thoughts come to mind. One is that you may simply add 0.5 or whatever continuity correction value you want to the group's with exactly zero events, then proceed with the need of the cc option.

Second, if you have access to Stata 16, you may use the built-in meta-analysis suite. See -meta meta-.

Third, perhaps other user-contributed meta-analysis commands exist for binary outcomes that do properly handle continuity corrects.