The MMRM (Mixed Model for Repeated Measures) is likely the most widely used model in clinical trials within the industry. Those from academic settings may be more familiar with its alternative name, the Linear Mixed-Effects Model. Given the repeated measures nature of clinical trials—where samples from a single patient are collected multiple times throughout the study—the MMRM is typically planned for primary time points when analyzing continuous and ordinal primary variables, as well as for secondary and tertiary variables in most clinical trials.

Below is a typical example of SAS codes found in the Statistical Programming Protocol for the MMRM:

proc mixed data = data_in METHOD=REML; 

  class treatment visit  other_factors_to_consider subjid;

  model change = treatment visit  treatment*visit other_factors_to_consider

              base / s ddfm= kr;

  Repeated visit / type = UN subject = subjid;

run;

Points to Remember:

1. REML (Restricted Maximum Likelihood) is used instead of Maximum Likelihood Estimation (MLE) as it focuses on estimating variance components while reducing bias.

2. The CLASS statement is used to specify categorical variables (classification variables), which can be either character or numeric.

3. Additional factors that may influence the results—such as geographical region or disease history—can be included in the model. In the above code, I have represented these as other_factors_to_consider.

   • The primary outcome of interest is typically the change from baseline, represented as change.
   • The baseline value (base) is also commonly included in the model as an independent variable (x) to account for the final result.

4. The "s" option in the MODEL statement (/ s ddfm=kr) stands for “solution”, requesting that the fixed-effects parameter estimates be produced.

   • DDFM specifies the method for computing denominator degrees of freedom, and "kr" refers to Kenward-Roger, a method proposed by Kenward and Roger in 1997.

5. The REPEATED statement specifies visit as the repeated effect.

   • If the model fails to converge, try changing the TYPE option to TOEP.
   • If it still does not converge, switch the model option to TYPE=CS.
   • The SUBJECT option identifies the subject in the mixed model. In this dataset, subjid represents each participant.