Statistical analysis and handling of missing data in cluster randomized trials: a systematic review

Area of Science:

• Biostatistics
• Health Research Methodology
• Clinical Trials

Background:

• Cluster randomized trials (CRTs) are essential for health research when individual randomization is infeasible or contamination is a concern.
• CRTs face challenges with missing data and accounting for clustering in primary analyses.
• This review evaluates methods for handling missing data and statistical analysis in CRTs.

Purpose of the Study:

• To systematically review and evaluate approaches for handling missing data in CRTs.
• To assess statistical analysis methods concerning primary outcomes in CRTs.
• To identify current practices and suggest improvements for data handling in CRTs.

Main Methods:

• Systematic search of CRTs published between August 2013 and July 2014 across PubMed, Web of Science, and PsycINFO.
• Independent review of 86 CRTs to assess missing data extent and handling methods.
• Evaluation of primary analysis for cluster or individual level accounting.

Main Results:

• 93% of CRTs reported missing outcome data, with a median of 19% missing.
• Complete case analysis was the most common method (55%) for handling missing data.
• Only 16% reported sensitivity analyses for missing data, and 78% accounted for clustering in primary analysis.

Conclusions:

• Missing data is prevalent in CRTs, and current handling methods are often suboptimal.
• Appropriate missing data methods and sensitivity analyses are crucial to reduce bias and increase statistical power.
• Researchers should employ valid missing data techniques and explore robustness through sensitivity analyses with varied assumptions.