My dataset contains participants who have been treated with intervention or placebo, the outcome is salivaflow (continous) and is measured at 0, 4 and 12 months. The outcome variable ($Y_{i}$) is the dependent variable, using a multilevel repeated measures random effects model with participants as the random effect factor, and time (with 3 levels, incl. baseline) as fixed effect factors based on a restricted maximum likelihood (REML) model.

This statistical model will hold all between-time comparisons for all assessment points up 12 months from baseline (incl. baseline) and thus allows for evaluation of the average effect, as well as the trajectory over time from baseline to 12 months follow-up.

Does the following encompass the above?

model <- lmer(UnstimSalivaFlow ~ GROUP * MONTH + (1|PtID), data = data)

Also, I wish the output to be as change from baseline in group_intervention and in group_placebo seperately and reported as least square means including p-values for group.


1 Answer 1


The model you wrote assumes that

  1. the residual error is the same at all timepoints (unlikely, usually goes up over time)
  2. all timepoints are equally correlated (unlikely, usually more correlated the closer together), i.e. this assumes a compound symmetric correlation matrix for the resdiuals
  3. correlation and residual error are the same in all treatment groups (might or might not be the case)
  4. normal residuals are appropriate for all visits (would be severely violated, if you had any inclusion criterion for the study that was applied at month 0, such as value must be > X at month 0 to be randomized); for the baseline you could avoid this assumption by making it a covariate (you would add the month 0 value as a main effect, as well as the interaction with MONTH)

To relax these assumptions, you could use the mixed model for repeated measures (MMRM), which is e.g. described here as part of this set of case studies in modeling in drug development (it uses the mmrm R package).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.