Including Tau2 estimates within forest plots produced by IPDMETAN and/or METAN

[Alex Wright-Hughes]

Hi there,

I'd like to output and include estimates of Tau2 within a forest plot of a random effects meta-analysis with a number of subgroups.

I am using the following code:

ipdmetan, study(StudyName) rfdist re(reml,hksj) by(Int,force) npts forest( astext(65) textsize(90) boxsca(30) effect(Std mean difference) xlabel(-1.5 -1 -0.5 0 0.5 1 1.5, format(%5.1fc)) cirange(-1.5  1.5)  range(-1.5  1.5) graphregion(color(white)) ciopts(lcolor(black) lw(thin)) favours(Favours intervention # Favours control) hetinfo(tausq  isq  pvalue ) extraline(yes)) saving(AD3m.dta,replace): reg scoreH BLRandTrtID0 BL_score i.BLGender0 BLAge  if Timepoint == 2

And the resulting forest plot is attached. This seems to provide what I want but the presented estimates of Tau2 are not consistent with the estimated values (see subgroup labelled 3).
The estimates of tau are not included in the STATA window output (I think because of the subgroups) but according to the below commands these are:

matrix list r(bystats)

          1:          3:          4:          5:          8:
          reml_hksj   reml_hksj   reml_hksj   reml_hksj   reml_hksj
    eff   .14305875  -.34760726  -.02324482  -.08602197  -.07699708
 se_eff   .28624069   .39736611   .19202394   .06818284   .03510469
eff_lci  -.41796269  -5.3966224  -.39960483  -.95236711  -.22804039
eff_uci   .70408019   4.7014079   .35311519   .78032316   .07404623
   npts          34         117          74          97         355
   crit    1.959964   12.706205    1.959964   12.706205   4.3026527
      z   .49978482           .  -.12105169           .           .
      t           .  -.87477833           .  -1.2616367  -2.1933556
     df           .           1           .           1           2
 pvalue    .6172266   .54245854    .9036501   .42667879   .15955421
tausq           0   .24239245           0           0           0
tsq_lci           .           0           .           0           0
tsq_uci           .   40.146095           .   3.6930345   .36962275
  rflci           .           .           .           .  -.52304451
  rfuci           .           .           .           .   .36905035

Can anyone able to advise what is going on here?
Is it also:
- possible to include the CI for Tau2 within the forest plot?
- edit the presentation of the pooled effect estimate to present the CI with/without the HKSJ adjustment within the same plot (much like the presentation of prediction intervals), or otherwise improve presentation when the HKSJ adjusted CIs are so wide as is the case in subgroup 3?

Many thanks for any help!

Alex

[David Fisher]

Hi Alex,
Many apologies for the late response, I hope this will still be helpful to you.
I have now released updates to ipdmetan and metan which should resolve this issue. Please update both packages (via SSC); the latest versions are 4.03 (ipdmetan) and v4.06 (metan).
Best wishes,
David.