ANCOVA with multiple imputation and fixed effects? Is there such a thing?

[Lisa Quadt]

Hiya everyone,

I am analysing data from an RCT, where have 2 groups (active vs control), measured anxiety levels and psychophysiological values before therapy (T1), and 1-week post-intervention (T2), and then measured anxiety levels again 3 months post-intervention (T3). We analyse the data on an intention-to-treat basis, and the primary outcome was anxiety at T3. For this, I ran mi estimate mixed with anxiety levels at T2 and T3 (depvar 'outcome'), group (active vs control, 'intervention'), and time (T2/T3) as fixed effects, baseline anxiety as covariate, and individual participants (ID) as random effects.

mi estimate: mixed outcome T1 b1.Intervention##b2.time || ID:,  cov(exchangeable)

This all worked well. However, for the psychophys measures, I only have T1 and T2 data, so I am struggling to adapt this model to these variables where no T3 data exists. As I see it, I have 2 options. One is to run an ANCOVA with mi estimate regress to assess between-group effects at T2, controlling for baseline scores:

mi estimate: regress T2_psychophys T1_psychophys Intervention

But then I won't have individual participants as random effects anymore. The second option is to run the same model as before, but without the covariate, but will this give me an accurate measure of between-group effects at T2 if the outcome variable includes T1 and T2?

mi estimate: mixed outcome b1.Intervention##b2.time || ID:,  cov(exchangeable)

Or maybe there is a third option I am not aware of? I'd be grateful for any advice, I feel like I am missing something. Thanks in advance!