Pattern mixture models for linear mixed models?

[Karin Lindqvist]

Hi, I am running linear mixed models on a dataset with substantial missing data, and is therefore interested in conducting a pattern mixture model. I have weekly measurements on the outcome variable for 10 weeks, and then 2 follow up occasions. Thus, data can be missing for all time points except baseline, in any order. I have found functions where I can let stata describe patterns of missingness, but only across variables (i. e. different variables in relation to each other) but I have not found it for long format (I have outcome PHQ-9 in long format for time 0 1 2 3… etc). I found the package, rctmiss which seems to be what I am looking for, but it can not model linear mixed models, but only regression models, from what I see… Does anyone know of a way of doing this? Best Karin

[Erik Ruzek]

I have not done this myself as I have rarely worked with data that is not plausibly missing at random (MAR), given covariates and appropriate approaches for MAR (i.e., multiple imputation, full information maximum likelihood, etc.). But there was a presentation from the Stata UK conference back in 2016 by Welch and colleagues on pattern mixture longitudinal modeling. They also have a Stata Journal article describing their twofold program (search twofold) See also this 2019 article by Iddrisu and Gumedze, which utilizes mimix (see the SJ article for it by Cro et al.).