Cluster sample size and testing

Specify correlation sizes and power

Include cluster count, size distribution, intracluster correlation and cluster loss in scenarios. The equal-size design-effect expression is illustrative, not a general substitute for a design-specific calculation.

Flow diagram

Cluster sample size and testingF1.S1Define design and primary contrastF1.S2Specify within-cluster correlationF1.S3Model unequal cluster sizesF1.S4Account for dropout and smallsamplesF1.S5Document power scenariosF1.N1Cluster design scenariosScenarioClustersCluster sizedistributionICCDropoutPower orprecisionMain scenarioHigher ICCUnequal sizesCluster lossF1.N2Equal-size cluster design effectThe simple design effect assumes equalcluster size and exchangeable correlation; itis not a universal cluster-trial powercalculation.F1.N3Method and protocol evidenceF1.S0 Start flowCluster sample size and testingF1.S999 End flowEnd flow

Steps

Define design and primary contrast

Specify number of clusters, allocation and target estimand.

Specify within-cluster correlation

Record ICC evidence, outcome variation and plausible uncertainty.

Model unequal cluster sizes

Include expected size distribution, imbalance and informative-size concerns.

Account for dropout and small samples

Separate cluster loss from participant attrition and justify inferential methods.

Document power scenarios

Use a design-compatible formula or simulation with assumptions and provenance.

Cluster sample size and testing

End flow

Tables and supporting information

Cluster design scenarios

Cluster design scenariosScenarioClustersCluster sizedistributionICCDropoutPower orprecisionMain scenarioHigher ICCUnequal sizesCluster loss

No numerical default is supplied; include number of clusters as well as total people.

Equal-size cluster design effect

The simple design effect assumes equal cluster size and exchangeable correlation; it is not a universal cluster-trial power calculation.

Method and protocol evidence

Attach the protocol, methodological sources and versioned software specification supporting these decisions. Record citations and provenance; do not enter participant identifiers.

Sources