How to calculate standardized mean difference using ipdmetan (two-stage IPD meta-analysis)?

[Mansour Alshehri]

Hi everyone,

I'm running a two-stage IPD meta-analysis by using ipdmetan:
ipdmetan, study(StudyID) random(reml) title(RMS-EO-AP) texts(110) forest(favours(Control # LBP)): xtmixed RMS i0.Group c.Age i.Sex c.BMI if Eyes==3 & Direction==1

I would like to ask about how to calculate the standardized mean difference as when I added esize(hedgesg):
ipdmetan, study(StudyID) random(reml) esize(hedgesg) title(RMS-EO-AP) texts(110) forest(favours(Control # LBP)): xtmixed RMS i0.Group c.Age i.Sex c.BMI if Eyes==3 & Direction==1
I received an error message "option hedgesg is incompatible with command-based syntax (Syntax 1)".

Thanks

[David Fisher]

Hi Mansour,

Firstly: for the benefit of other readers, I should just point out that ipdmetan is a user-contributed command, available from SSC.

The option esize(hedgesg) is, I think, taken from the documentation for the Stata 16 meta suite, which is a separate command with incompatible options.

If you wish to label your effects as, e.g. "Hedges' g", you can do this with the option effect("Hedges' g"). However, this is only a label, for the purposes of presentation; it does not affect the actual results. ipdmetan simply takes the effect sizes from the fitted model, in this case xtmixed. These will, I presume, be adjusted coefficients describing the change in RMS between two values of Group. Whether these can accurately be described as "standardized mean differences" is outside my sphere of knowledge.

Hedges's g and other "mean difference" options are mainly used with aggregate (i.e. non-IPD) with user-written metan or Stata 16 meta. However, ipdmetan does allow you to analyze IPD as if it were aggregated, by calculating the mean and SD per group and then applying an aggregate-like analysis. So, for a Hedges SMD, you could code:

ipdmetan RMS Group if Eyes==3 & Direction==1, study(StudyID) random(reml) hedgesg effect("Hedges' g") title(RMS-EO-AP) texts(110) forest(favours(Control # LBP))

However, you cannot now adjust for your covariates (age, sex, BMI) or apply any mixed structure to your data.

I hope that helps to clarify matters!

Best wishes,

David.

[Mansour Alshehri]

Dear Dr Fisher

Many thanks for your comment.

I used xtmixed because I want to adjust some factors such as age, sex and BMI, and then I pooled the adjusted coefficients differences between groups and visualized the results with forest plots using ipdmetan. However, in our IPD data, RMS and some other outcomes were measured with two different units/scales. So, it would be better to use SMD but this may be difficult as also I want to adjust for confounding variables.

I just did an analysis for each study separately to obtain AD using xtmixed and extracted the marginal means, SE and sample size for each group. I converted SE to SD via "SE multiplying by the square root of the sample size for each group") and then used meta to calculate SMD for the effect size with hedgesg method (it was the default)and it work. I wonder if what I have done here is correct/fine.

Regards,
Mansour,