Cluster-randomized parallel trial
Should intact groups rather than individuals be assigned to interventions?
When to use this project
Randomize hospitals, practices, schools or other clusters when the intervention is delivered collectively or individual assignment would cause substantial contamination.
How it differs
Randomization occurs at cluster level while outcomes may be measured in individuals. Observations within a cluster are dependent; many individuals cannot compensate automatically for too few clusters.
Decisions before you start
- Define cluster and participant eligibility, consent and recruitment timing relative to allocation.
- Choose a cluster-average or participant-average estimand and compatible weighting.
- Plan cluster count, variable cluster sizes, intracluster correlation and small-sample inference.
How the flows work together
Each flow below links to its diagram, steps and tables. Flow order organizes the project; it does not imply that all work is strictly sequential. Data assessment, analysis planning and protocol refinement may inform one another.
1. Cluster trial overview
Overview · Question through clustered trial reporting
Use this overview to state the purpose and link to detailed decisions. Keep a versioned protocol and decision register, and add evidence references for the rationale.
2. Cluster and participant recruitment
Selection · Track both levels of enrollment
Clusters and people are different units. Recruitment after allocation can introduce selection if knowledge of treatment affects entry; document timing, masking and consent at both levels.
3. Cluster allocation and delivery
Conduct · Specify assignment contamination and implementation
Randomize at the declared cluster level and retain stratification or restrictions. Contamination and fidelity concern intervention delivery; they do not justify reclassifying randomized assignment after observing outcomes.
4. Endpoints and estimands
Design · Define cluster and individual effects
Choose whether the average effect weights clusters equally or participants equally. Unequal and informative cluster sizes can make these different targets.
5. Cluster sample size and testing
Statistics · 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.
6. Follow-up and data populations
Data · Track cluster losses and participants
Loss of a cluster and loss of a participant have different consequences. Preserve both identifiers and reconcile counts within their own units; do not add them together.
7. Clustered analysis plan
Statistics · Specify multilevel estimation and sensitivity
Use a model or cluster-level analysis compatible with the target weighting and randomization. With few clusters, justify degrees of freedom or small-sample corrections rather than relying on large individual counts.
Equations and statistical interpretation
These explanations describe notation and assumptions. Flow does not calculate these values.
Equal-size cluster design effect
m is the common cluster size and ρ is the intracluster correlation coefficient for the specified outcome.
This is an illustrative variance inflation under a simple equal-size, exchangeable-correlation setting. It is not a complete power calculation for unequal clusters, few clusters, repeated periods or complex estimands.
Published example
Huang et al. (2013), Targeted versus Universal Decolonization to Prevent ICU Infection
Identify the hospital randomization unit, intervention groups and patient outcomes. Examine how the analysis accounts for cluster dependence.
This publication illustrates a study design, not an endorsement of every method or a reproduction of its results. Follow its protocol, supplement and update history when studying the example.
Hypothetical scenario
A health system wants to compare a ward-wide infection-prevention package with usual practice.
- Cluster trial overview: Define clusters and the intended population, then approve and register the study.
- Cluster and participant recruitment: Recruit wards and specify patient inclusion before or after allocation, addressing recruitment bias.
- Cluster allocation and delivery: Randomize wards and track contamination and intervention fidelity.
- Endpoints and estimands: Define the outcome level and target weighting across wards and patients.
- Cluster sample size and testing: Evaluate power over plausible correlation, cluster-size and dropout scenarios.
- Follow-up and data populations: Track ward loss separately from patient loss and preserve multilevel identifiers.
- Clustered analysis plan: Analyze with cluster-aware uncertainty and sensitivity to weighting, missingness and small cluster counts.
Deliver a populated protocol, decision log, data specification, and planned outputs. Enter numerical findings only after analysis and verification. Keep decisions made before seeing results distinguishable from later amendments.
Sources
- CONSORT 2025: reporting randomized trials
- SPIRIT 2025: randomized trial protocols
- CONSORT extension for cluster randomized trials (2012)
- ICH E9(R1): estimands and sensitivity analysis (2019)
- WHO: clinical trials and prospective registration
- ISRCTN: what the clinical study registry records
- ClinicalTrials.gov: about the registry
Source check: 2026-09-17. Record the version of each reporting or appraisal tool used in your actual protocol.
Compare all Clinical trials project types · Open editor in a new tab