The mixed model for repeated measures for cluster randomized trials: a simulation study investigating bias and type I...

Area of Science:

• Biostatistics
• Clinical Trials Methodology

Background:

• Cluster randomized trials (CRTs) are essential for interventions where individual randomization is impractical.
• The mixed model for repeated measures (MMRM) is widely used for longitudinal continuous outcomes in individually randomized trials.
• MMRM offers advantages in avoiding model misspecification and ensuring unbiasedness for missing data (missing completely at random or missing at random).

Purpose of the Study:

• To extend the mixed model for repeated measures (MMRM) to cluster randomized trials (CRTs).
• To evaluate the statistical properties of the extended MMRM-CRT using simulation when data are missing at random.
• To demonstrate the MMRM-CRT with a real-world example of a cardiovascular disease prevention trial.

Main Methods:

• Extended MMRM by incorporating a random intercept for the cluster.
• Conducted a simulation experiment varying key parameters: number of clusters, cluster size, intra-cluster correlation, and missingness patterns.
• Simulated continuous outcomes at baseline and three post-intervention time points.
• Applied the MMRM-CRT to a cluster randomized trial in diabetes patients for cardiovascular disease prevention.

Main Results:

• Estimates of treatment effects were unbiased under both complete and missing-at-random data scenarios.
• Variance components demonstrated largely unbiased estimates.
• Type I error rates were generally nominal, with minor inflation (up to 0.081) in specific simulation cases.

Conclusions:

• The MMRM for CRTs is appropriate for longitudinal continuous outcomes, contrary to some assertions about its limitations with multiple repeated measures.
• The MMRM-CRT is a statistically sound and recommended analytic strategy for cluster randomized trials with longitudinal data.