Estimating effective sample sizes using standard errors from appropriate analyses of cluster-randomized trials.

Area of Science:

• Biostatistics
• Clinical Trials Methodology

Background:

• Cluster-randomised trials (CRTs) often yield less information than individually randomised trials due to within-cluster correlation, leading to a lower effective sample size (ESS).
• Ignoring clustering in analyses can result in biased estimates and overly precise statistical results (e.g., small p-values, narrow confidence intervals).
• Estimating ESS using intra-cluster correlation coefficients (ICCs) is common, but ICCs are frequently unknown and must be assumed.

Purpose of the Study:

• To present and evaluate two novel methods for estimating the ESS in CRTs.
• To provide practical tools for analysing CRT data without requiring pre-specified ICCs.

Main Methods:

• Two distinct approaches were developed to estimate ESS using standard errors from cluster-aware analyses.
• Method 1 scales counts by the ratio of variances from cluster-aware versus simple analyses.
• Method 2 employs an optimization procedure to adjust event proportions or group means.

Main Results:

• Both presented methods successfully prevent over-precision in analyses of group-level summary data from CRTs.
• The second method additionally corrects for potential bias introduced by ignoring clustering in the analysis.
• Comparisons were made using data from three example CRTs against analyses that did not account for clustering.

Conclusions:

• The proposed methods offer valuable tools for interpreting and synthesizing evidence from CRTs.
• These approaches are applicable even when ICCs are not reported, provided appropriate cluster-aware analyses are performed.