Cluster Randomized Trials with a Pretest and Posttest: Equivalence of Three-, Two- and One-Level Analyses, and Sample...

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Trials

Background:

• Cluster randomized trials (CRTs) are essential when individual randomization is infeasible or risks treatment contamination.
• CRTs assign entire groups (clusters) to interventions, necessitating specific analytical approaches.
• Pretest-posttest designs with quantitative outcomes are common in CRTs.

Purpose of the Study:

• To demonstrate the analytical equivalence of four distinct statistical methods for analyzing CRT data with pretest and post-test measures.
• To compare these methods using a real-world CRT example.
• To derive a simplified sample size calculation method based on the established equivalences.

Main Methods:

• Comparison of a three-level mixed-effects model (cluster, person, time) with unstructured covariance matrices.
• Evaluation of a two-level mixed-effects model (cluster, person) using change from baseline.
• Analysis of a two-level mixed-effects model (cluster, time) with cluster means.
• Assessment of a one-level analysis of cluster means of change from baseline.
• Equivalence demonstration between constrained mixed models and pretest-covariate methods.

Main Results:

• Four different analytical approaches for cluster randomized trials with pretest-posttest data were found to be statistically equivalent.
• The study illustrated these equivalences using a mental health CRT in children.
• A straightforward method for calculating sample sizes in CRTs with baseline measurements was developed.

Conclusions:

• The equivalence of these analytical methods provides flexibility in analyzing cluster randomized trial data.
• The simplified sample size calculation method facilitates more efficient study planning.
• These findings contribute to robust statistical practices in cluster randomized trial research.