Cluster and Stepped-Wedge Trials

Overview

Cluster randomised trials assign intact groups such as clinics, wards, schools or communities. They are used when an intervention is delivered at group level, individual randomisation would contaminate care or implementation decisions are inherently organisational. Stepped-wedge trials randomise the order in which clusters cross from control to intervention over calendar periods.

The study question, design, participant pathway, measurements and analysis must be specified as one coherent system. A recognised design label is useful only when the operational details preserve the comparison or prediction that the study is intended to support.[1–3]

Why this matters

Cluster assignment creates correlated outcomes and multiple levels of eligibility, consent and exposure. In stepped-wedge designs, intervention is entangled with calendar time unless models and schedules separate them. Recruiting individuals after cluster allocation can introduce selection if recruiters or participants know the cluster condition.

Core concepts

The cluster is the allocation unit

The intervention may act at cluster, provider or patient levels, but randomisation occurs once per cluster. The effective information is therefore less than the number of individuals suggests because outcomes within clusters are correlated.

Cluster eligibility and participant eligibility differ

Clusters may be recruited and randomised before individuals are identified. Post-randomisation recruitment can differ by condition, especially if staff know assignment. Whenever possible, identify or recruit participants before allocation or use objective population-wide data.

Intracluster correlation affects design and analysis

The intracluster correlation coefficient describes similarity within clusters. Sample-size inflation also depends on cluster size and its variability. Analysis should account for clustering even if the observed ICC is small.

Stepped wedges add time and transition

All clusters begin in control and cross over according to randomised sequences. Secular trends, learning, anticipation and incomplete implementation can mimic effects. The number of clusters, periods and observations per cluster determines precision.

Consent and intervention exposure are multilevel

Cluster gatekeepers may agree to organisational participation but cannot always replace individual consent for personal data or procedures. Participants may be exposed indirectly to a system intervention without individual choice, requiring proportionate ethical justification.

When to use this approach

DECISION GUIDE

Choosing cluster or stepped-wedge randomisation

• Is intervention delivery genuinely cluster level or contamination unavoidable?
• How many clusters and periods are feasible?
• Can individuals be identified before allocation?
• Are calendar trends and implementation transitions measurable?
• Will every cluster receiving intervention later be ethically or politically important?

Working through the key design decisions

Decision 1: Is intervention delivery genuinely cluster level or contamination unavoidable?. This point should be resolved before recruitment, extraction or assignment begins. For cluster and stepped-wedge trials, the protocol should state the operational rule, who applies it, when it is applied and what source documents demonstrate compliance. The investigators should also consider how a different choice would change the eligible population, comparison, timing, measurement or interpretation. A pilot review of realistic records or participant pathways can reveal ambiguities before they become irreversible exclusions or post hoc analytical choices. The decision should be reviewed across sites and subgroups because a rule that works in one clinic may operate differently where referral, resources or disease spectrum differ. Any permitted exception should be explicit, dated and reported with its frequency and likely consequence. The corresponding estimand or target contrast should be written in words, and the planned estimator should be checked against the sampling, clustering, matching or censoring mechanism. Sensitivity analyses should target plausible departures from the design assumptions rather than provide an undirected collection of alternative models.

Decision 2: How many clusters and periods are feasible?. This point should be resolved before recruitment, extraction or assignment begins. For cluster and stepped-wedge trials, the protocol should state the operational rule, who applies it, when it is applied and what source documents demonstrate compliance. The investigators should also consider how a different choice would change the eligible population, comparison, timing, measurement or interpretation. A pilot review of realistic records or participant pathways can reveal ambiguities before they become irreversible exclusions or post hoc analytical choices. The decision should be reviewed across sites and subgroups because a rule that works in one clinic may operate differently where referral, resources or disease spectrum differ. Any permitted exception should be explicit, dated and reported with its frequency and likely consequence. The corresponding estimand or target contrast should be written in words, and the planned estimator should be checked against the sampling, clustering, matching or censoring mechanism. Sensitivity analyses should target plausible departures from the design assumptions rather than provide an undirected collection of alternative models.

Decision 3: Can individuals be identified before allocation?. This point should be resolved before recruitment, extraction or assignment begins. For cluster and stepped-wedge trials, the protocol should state the operational rule, who applies it, when it is applied and what source documents demonstrate compliance. The investigators should also consider how a different choice would change the eligible population, comparison, timing, measurement or interpretation. A pilot review of realistic records or participant pathways can reveal ambiguities before they become irreversible exclusions or post hoc analytical choices. The decision should be reviewed across sites and subgroups because a rule that works in one clinic may operate differently where referral, resources or disease spectrum differ. Any permitted exception should be explicit, dated and reported with its frequency and likely consequence. The corresponding estimand or target contrast should be written in words, and the planned estimator should be checked against the sampling, clustering, matching or censoring mechanism. Sensitivity analyses should target plausible departures from the design assumptions rather than provide an undirected collection of alternative models.

Decision 4: Are calendar trends and implementation transitions measurable?. This point should be resolved before recruitment, extraction or assignment begins. For cluster and stepped-wedge trials, the protocol should state the operational rule, who applies it, when it is applied and what source documents demonstrate compliance. The investigators should also consider how a different choice would change the eligible population, comparison, timing, measurement or interpretation. A pilot review of realistic records or participant pathways can reveal ambiguities before they become irreversible exclusions or post hoc analytical choices. The decision should be reviewed across sites and subgroups because a rule that works in one clinic may operate differently where referral, resources or disease spectrum differ. Any permitted exception should be explicit, dated and reported with its frequency and likely consequence. The corresponding estimand or target contrast should be written in words, and the planned estimator should be checked against the sampling, clustering, matching or censoring mechanism. Sensitivity analyses should target plausible departures from the design assumptions rather than provide an undirected collection of alternative models.

Decision 5: Will every cluster receiving intervention later be ethically or politically important?. This point should be resolved before recruitment, extraction or assignment begins. For cluster and stepped-wedge trials, the protocol should state the operational rule, who applies it, when it is applied and what source documents demonstrate compliance. The investigators should also consider how a different choice would change the eligible population, comparison, timing, measurement or interpretation. A pilot review of realistic records or participant pathways can reveal ambiguities before they become irreversible exclusions or post hoc analytical choices. The decision should be reviewed across sites and subgroups because a rule that works in one clinic may operate differently where referral, resources or disease spectrum differ. Any permitted exception should be explicit, dated and reported with its frequency and likely consequence. The corresponding estimand or target contrast should be written in words, and the planned estimator should be checked against the sampling, clustering, matching or censoring mechanism. Sensitivity analyses should target plausible departures from the design assumptions rather than provide an undirected collection of alternative models.

The approach should be chosen because its structure answers the research question, not because the available dataset happens to resemble a familiar table. The investigator should state the target population, time origin, comparison, outcome and intended measure before collecting or extracting data.

Design and participant pathway

DESIGN PROFILE

Cluster and stepped-wedge randomised trials

Core structure: Randomise intact groups to fixed strategies or to a sequence of crossover times, then analyse outcomes with cluster and time structure.

Best suited for: Service, professional, school, community and policy interventions delivered to groups.

Principal threats: Few clusters, baseline imbalance, recruitment after allocation, contamination, secular trends, unequal cluster size and inappropriate individual-level analysis.

A parallel cluster trial has clearer separation from calendar time and may finish sooner. A stepped wedge can facilitate staged implementation but is not automatically more ethical or efficient; participants exposed later still serve as controls, and secular changes complicate inference. Transition periods may need exclusion or separate coding when implementation is incomplete.

Protocol specification

PROTOCOL CHECKLIST

Decisions to prespecify

• Define cluster and individual eligibility.
• Specify sequence generation, stratification and allocation timing.
• Estimate ICC, cluster-size distribution and design effect.
• Plan recruitment relative to cluster assignment.
• Define periods, transitions and calendar-time adjustment.
• Specify cluster-aware analysis and degrees of freedom.

Protocol item 1: Define cluster and individual eligibility. This item requires a reproducible definition rather than a broad intention. The study manual should identify the responsible role, data field or procedural record, permissible time window, quality check and response to deviations. Where judgement is unavoidable, assessors should receive examples and escalation rules, and a sample of decisions should be reviewed for consistency. The analysis dataset should retain enough provenance to distinguish planned exclusions, measurement failure, participant withdrawal and genuinely absent information. Because advanced designs often use weights, risk sets, clusters, thresholds or time-varying states, the derivation should be independently reproduced from raw dates and identifiers. Simulation or mock-data testing before database lock can verify that the prespecified estimator answers the stated question under the intended design.

Protocol item 2: Specify sequence generation, stratification and allocation timing. This item requires a reproducible definition rather than a broad intention. The study manual should identify the responsible role, data field or procedural record, permissible time window, quality check and response to deviations. Where judgement is unavoidable, assessors should receive examples and escalation rules, and a sample of decisions should be reviewed for consistency. The analysis dataset should retain enough provenance to distinguish planned exclusions, measurement failure, participant withdrawal and genuinely absent information. Because advanced designs often use weights, risk sets, clusters, thresholds or time-varying states, the derivation should be independently reproduced from raw dates and identifiers. Simulation or mock-data testing before database lock can verify that the prespecified estimator answers the stated question under the intended design.

Protocol item 3: Estimate ICC, cluster-size distribution and design effect. This item requires a reproducible definition rather than a broad intention. The study manual should identify the responsible role, data field or procedural record, permissible time window, quality check and response to deviations. Where judgement is unavoidable, assessors should receive examples and escalation rules, and a sample of decisions should be reviewed for consistency. The analysis dataset should retain enough provenance to distinguish planned exclusions, measurement failure, participant withdrawal and genuinely absent information. Because advanced designs often use weights, risk sets, clusters, thresholds or time-varying states, the derivation should be independently reproduced from raw dates and identifiers. Simulation or mock-data testing before database lock can verify that the prespecified estimator answers the stated question under the intended design.

Protocol item 4: Plan recruitment relative to cluster assignment. This item requires a reproducible definition rather than a broad intention. The study manual should identify the responsible role, data field or procedural record, permissible time window, quality check and response to deviations. Where judgement is unavoidable, assessors should receive examples and escalation rules, and a sample of decisions should be reviewed for consistency. The analysis dataset should retain enough provenance to distinguish planned exclusions, measurement failure, participant withdrawal and genuinely absent information. Because advanced designs often use weights, risk sets, clusters, thresholds or time-varying states, the derivation should be independently reproduced from raw dates and identifiers. Simulation or mock-data testing before database lock can verify that the prespecified estimator answers the stated question under the intended design.

Protocol item 5: Define periods, transitions and calendar-time adjustment. This item requires a reproducible definition rather than a broad intention. The study manual should identify the responsible role, data field or procedural record, permissible time window, quality check and response to deviations. Where judgement is unavoidable, assessors should receive examples and escalation rules, and a sample of decisions should be reviewed for consistency. The analysis dataset should retain enough provenance to distinguish planned exclusions, measurement failure, participant withdrawal and genuinely absent information. Because advanced designs often use weights, risk sets, clusters, thresholds or time-varying states, the derivation should be independently reproduced from raw dates and identifiers. Simulation or mock-data testing before database lock can verify that the prespecified estimator answers the stated question under the intended design.

Protocol item 6: Specify cluster-aware analysis and degrees of freedom. This item requires a reproducible definition rather than a broad intention. The study manual should identify the responsible role, data field or procedural record, permissible time window, quality check and response to deviations. Where judgement is unavoidable, assessors should receive examples and escalation rules, and a sample of decisions should be reviewed for consistency. The analysis dataset should retain enough provenance to distinguish planned exclusions, measurement failure, participant withdrawal and genuinely absent information. Because advanced designs often use weights, risk sets, clusters, thresholds or time-varying states, the derivation should be independently reproduced from raw dates and identifiers. Simulation or mock-data testing before database lock can verify that the prespecified estimator answers the stated question under the intended design.

A protocol should make these decisions before outcome-dependent choices become tempting. It should also distinguish the design population from the analysis population, document exclusions after enrolment or sampling, and explain how missing or indeterminate observations will be handled. Any deviation should be dated, justified and retained in the study record.[2–4]

Measurements, timing and analysis interface

Outcome collection should be standardised across clusters and periods. Routine data can reduce recruitment bias but may vary in completeness by clinic. Implementation measures should record when intervention components became operational rather than assuming the assigned crossover date equals full exposure. Cluster-level withdrawals and closures require explicit handling.

The analysis should reflect how participants or clusters entered the study, how observations are related, and how the effect or performance measure was defined. Statistical adjustment cannot repair a comparison that was invalid at the design stage. Conversely, a sound design can still be undermined by an analysis that ignores matching, clustering, censoring, sampling fractions or prespecified thresholds.

Worked example

WORKED EXAMPLE

Selecting a parallel cluster design

Because navigation can begin immediately at half of the clinics and the health system can tolerate concurrent implementation, the team selects a parallel cluster trial. Clinics are stratified by size and urban location before central randomisation. All eligible deliveries are identified from routine records, limiting post-allocation recruitment. The sample-size calculation uses expected ICC and unequal clinic sizes; mixed models include clinic effects and prespecified baseline covariates.

The example should be interpreted in the population and time frame created by the design. The numerical result does not by itself establish causation, transportability or clinical usefulness; those conclusions require additional assumptions and evidence.

Appraising methodological quality

APPRAISAL CHECKLIST

Questions for critical appraisal

• Justification: Was cluster assignment necessary?
• Clusters: Were number, recruitment and baseline characteristics adequate?
• Individuals: Could knowledge of allocation affect identification or consent?
• Time: Were calendar and transition effects handled in stepped-wedge designs?
• Analysis: Were clustering, small numbers and unequal sizes respected?

A reporting guideline can improve transparency, but it is not a substitute for sound design. Appraisal should ask whether the reported methods could generate an unbiased, clinically meaningful answer—not merely whether every checklist item has been mentioned.[3–6]

Suggested reporting

Reporting should identify the design in the title or abstract when appropriate, describe the source and flow of participants, define all key variables, report missing and indeterminate observations, and present estimates with uncertainty. The direction and denominator of every comparison must be explicit.

Ethical and governance considerations

Estimands in cluster trialsMMed extension

Cluster trials require clarity about the population of clusters and individuals, treatment conditions, endpoint, intercurrent events and population summary. Individual-average and cluster-average effects can differ when cluster size is informative. Weighting and model choice should match the decision target.

Common mistakes

Knowledge check

Why does ICC matter?

Answer and reasoning

It reflects within-cluster similarity, reducing the independent information contributed by additional people in the same cluster.

What confounds intervention in a stepped wedge?

Answer and reasoning

Calendar time, because later periods contain more intervention exposure and may also have secular outcome changes.

Can a gatekeeper provide all consent?

Answer and reasoning

Not necessarily. Organisational permission and individual consent address different levels of participation and data use.

Why are few large clusters problematic?

Answer and reasoning

Randomisation balance and variance estimation depend heavily on the number of independent clusters, not just total participants.

Key take-home messages

SUMMARY

Key take-home messages

• Randomise clusters only when the intervention or contamination requires it.
• Protect recruitment from knowledge of cluster assignment.
• Plan with ICC, cluster numbers and size variation.
• Model calendar time and implementation in stepped wedges.
• Address consent and estimands at cluster and individual levels.

References

1. Campbell MK, Piaggio G, Elbourne DR, Altman DG; CONSORT Group. Consort 2010 statement: extension to cluster randomised trials. BMJ. 2012;345:e5661. https://doi.org/10.1136/bmj.e5661
2. Hemming K, Taljaard M, McKenzie JE, et al. Reporting of stepped wedge cluster randomised trials: extension of the CONSORT 2010 statement. BMJ. 2018;363:k1614. https://doi.org/10.1136/bmj.k1614
3. Hopewell S, Chan AW, Collins GS, et al. CONSORT 2025 statement: updated guideline for reporting randomised trials. BMJ. 2025;389:e081123. https://doi.org/10.1136/bmj-2024-081123
4. Butcher NJ, Monsour A, Mew EJ, et al. SPIRIT 2025 statement: updated guideline for protocols of randomised trials. BMJ. 2025;389:e081477. https://doi.org/10.1136/bmj-2024-081477
5. Hayes RJ, Moulton LH. Cluster randomised trials. 2nd ed. Boca Raton: CRC Press; 2017.
6. Hemming K, Haines TP, Chilton PJ, Girling AJ, Lilford RJ. The stepped wedge cluster randomised trial: rationale, design, analysis, and reporting. BMJ. 2015;350:h391. https://doi.org/10.1136/bmj.h391