I am testing differences in gene expression across four time-points and between treatment and control. In total, 50 individuals were recruited for the Treatment Group and 58 recruited for the Control Group.
The way the study was (or should, let's say) design was to measure the same individuals in the four time-points and then perform repeated measures analysis (using a LMER for example - one can then account for confounding variables).
Now, what was done was repeated measures only in the treatment group, although very poor (only 9 out of 50 have measures for the four moments, but 26 have for 3). Each of the 58 patients were measured in one time point - which means for each time-point, I have around ~70-75% missing values for each control.
My approach (apart from point all the reasons why the experimental design was poorly drawn) would be to use lmer, with a random intercept:
lmer(Exp ~ TimePoint*Group + (1|Patients)
However, with the amount of missing data I have in my data, and considering the Control group have considerable more missing data (thus, making it MNAR missing data), I am a bit wary to follow said approach. Do you think lmer is still a valid approach in this scenario?
Thanks
