What is it?
In a cluster randomised controlled trial (RCT) rather than randomising individual participants as in traditional RCTs, groups (or clusters) of participants are randomised to either a control or intervention arm. Examples of clusters include villages, hospitals or schools. Cluster RCTs are also known as group randomised, field, community-based or place-based trials.
The main advantage of cluster RCTs is to reduce the potential for ‘contamination’ between treatment groups since all participants in the same randomised cluster receive the same care. This is particularly useful for interventions involving media campaigns, organisational changes and legislation. Cluster RCTs may be considered appropriate for medicines when trial procedures or information given to those in the one arm may influence (or ‘contaminate’) treatment in the other arm. With less possibility of interference in the ‘usual care’ arm of a trial, the results may better reflect clinical practice.
Why is it useful?
- Control on cross-contamination: contamination between and across individuals within the study is limited because everyone in the same cluster receives the same intervention or control.
- Measures effect of a medicine at population level: a cluster RCT may provide evidence on relative effectiveness by measuring the overall effect of a medicine at the population level, encompassing both the direct effect of a medicine on an individual and also indirect effects due to clinician training and health promotion activities or changes in epidemiological factors such as herd immunity.
- Reduced costs: may be less expensive than traditional RCTs because of efficiencies when recruiting a whole group.
When is it suitable?
- While this design is traditionally used for interventions targeted at clusters or groups of people or at health professionals, such as media campaigns, training, or at wider organisation changes, it may also be suitable for medicines if the trial procedures or information given during one arm has the potential to influence or contaminate treatment in the other arm.
What are the limitations?
- Prone to bias: cluster RCTs are prone to confounding. In traditional RCTs, which randomise individual participants to treatments, the randomisation minimises selection bias by distributing confounding factors (both known and unknown) between groups. However, in cluster trials there may be similarities within clusters (and differences between different clusters) that cannot be addressed through randomisation.
- Concealment of treatment allocation not possible: this can lead to biased recruitment in cluster trials. Alternatively, participants can differentially refuse consent to participate in the trial, which can be another source of selection bias and also lead to dilution bias. Measurements of outcomes can nevertheless be made blind to treatment allocation, in which case it is important to use intention-to-treat analysis.
- Large sample size needed: the unit of randomisation is a group or cluster of people, so a larger number of participants are needed to achieve the same statistical power than for a traditional RCT.
- Issues obtaining consent: informed consent may be an issue as there may be several levels at which consent may need to be obtained.
- Complex analysis required: analysis of cluster RCT data must take into account the clustered nature of the data. Failing to do so will result in the risk of a type 1 error (erroneously concluding there was a statistically significant difference). Participants within the same cluster may be more similar than those in different clusters, leading to a correlation of observations within the clusters. This feature of the data needs to be accounted for in the analysis.