command mfp with MMRM

[John Dubois]

Hello everyone,
I would like to use the mfp command with an MMRM model (mixed command), but it is not available with this type of model. Is there an extension for this command that can accomplish it?
Thanks in advance for the response.

[Leonardo Guizzetti]

There are two issues to consider here and I don’t additional advice on workarounds. Very often REML is used with mixed models and it’s not appropriate to select models of fixed effects unless you have decided on and chosen all of your random effects. However, some choices in random and residual covariance structures are dependent on the fixed effects structure. Second, often it’s of interest to model time flexibly, but in an MMRM the point is to consider time categorically, so these two views are incompatible.

[John Dubois]

Thank you, Leonardo,
I didn't fully understand your response. For me, time can be used continuously in an MMRM model, and that's actually what I want to do. That's why I wanted to use the mfp function.

[Leonardo Guizzetti]

An MMRM is a specific type of mixed model, where the base model incorporates repeated measures are correlated by means of residual covariance structure, and time is explicitly treated categorically. This is discussed in depth in Molenberghs et al (2004).

You can model time as parametric form (linear, quadratic, etc.) but this falls under a linear mixed model classification. These are the models which would accommodate fractional polynomials or splines, for example. Nevertheless, -mfp- wasn't designed for mixed models in general, and I don't use them enough to offer more specific guidance.

Geert Molenberghs, Herbert Thijs, Ivy Jansen, Caroline Beunckens, Michael G. Kenward, Craig Mallinckrodt, Raymond J. Carroll, Analyzing incomplete longitudinal clinical trial data, Biostatistics, Volume 5, Issue 3, July 2004, Pages 445–464, https://doi.org/10.1093/biostatistics/kxh001

[Joseph Coveney]

I would like to use the mfp command with an MMRM model (mixed command), but it is not available with this type of model.

According to its help file, mfp is available with xtgee and if your mixed model has two levels and the conventional exchangeable covariance structure, then you could use xtgee in lieu of mixed.

[John Dubois]

Hello to you both,
Thank you very much for your help and I will look into it.
Have a nice day!