Analysis of cluster randomized trials with repeated cross-sectional binary measurements

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Health Services Research

Background:

• Cluster randomized trials (CRTs) with repeated cross-sectional designs require specific analytical techniques.
• Existing methods for evaluating interventions on dichotomous outcomes in such designs are not fully explored.
• Baseline imbalance can significantly affect intervention impact assessment.

Purpose of the Study:

• To compare analytical methods for evaluating intervention impact in CRTs with repeated cross-sectional designs.
• To address challenges of accounting for clustering and baseline imbalance.
• To identify optimal statistical approaches for analyzing dichotomous outcomes in this trial design.

Main Methods:

• Comparison of cluster-level analyses, clustered Woolf method, generalized estimating equations (GEE) marginal models, multilevel models, and random effects meta-analysis.
• Application of methods to a CRT evaluating Cochrane review evidence implementation in 25 hospital obstetric units.
• Data collection involved 30 pregnancies per unit at baseline and 30 separate pregnancies at follow-up.

Main Results:

• Analyses ignoring baseline showed no intervention effect.
• Substantial baseline imbalance necessitated baseline-adjusted analyses.
• While change-from-baseline analyses showed intervention effects, analysis of covariance (ANCOVA) did not, due to using cluster-level responses.
• ANCOVA approaches are preferable to change-from-baseline analyses, which can be misleading.

Conclusions:

• Analysis of covariance is a preferred method for repeated cross-sectional CRTs.
• Achieving baseline balance through stratification or ensuring large sample sizes is vital.
• Careful consideration of analytical techniques is essential for accurate interpretation of intervention effects in complex trial designs.