Subgroup analysis

[nociale]

Does rbmi already support subgroup analysis?

What is the best way of performing subgroup analysis with rbmi?

If one would like to basically perform a (baseline) subgroup analysis (let's see by gender) of the primary analysis. Is that correct to consider a three-way interaction treatment-visit-gender in the imputation step and a two way interaction 'treatment-gender' in the 'analysis' (ANCOVA) step? Would, in this case, the output from analysis() contain the treatment effect with each subgroup (and would then these estimates be pooled in the pool() function)?

Once this is solved, should we add an example in the advanced vignette?

[wolbersm]

Hi @nociale

Yes, I think so:
If we call the subgrouping variable s, then I think this would mean:
Imputation model: y~ <usual adjustment for baseline value etc.> + visit*treatment*s [triple interaction]
Analysis model: y[visit] ~ <baseline value + possible other baseline covariates> + treatment*s

I also think that your for modifying regarding analysis() and pool() function make sense.

One additional topic is that subgroup analyses often include interaction test p-values for the treatment*s interactions and in case s has >2 levels, the resulting test statistics is a chi-squared statistics rather than a Z statistics (estimate/SE) which would need some extra thought regarding pooling.