Mixed Model for Repeated Measures (MMRM)

This week our statisticians describe the mixed model for repeated measures (MMRM).

🎓️ In our last post we learned about comparing the means of groups with ANOVA.

❓️But how can we handle group comparisons when we have several measurements from the same patient?

Challenge:

 ⚠️ An important assumption for ANOVA is the independence of the measurements. When we have repeated measurements from the same patient at different time points, this assumption is no longer fulfilled.

Statistical Method:

📊 One possible method is ANOVA for repeated measurements.

📊 A better option is using a mixed model for repeated measures (MMRM). One advantage: When considering a lot of measurements from different time points, we often face the problem of missing data. This does not matter when using an MMRM, since the model can handle this issue! 💪

📊 The structure of the covariance matrix is relevant for the model. In many applications measurements from time points which are close together have a higher correlation than measurements from time points which are far away from each other. The model is most flexible when assuming an unstructured covariance matrix.