*Geek Box: Cluster Randomisation
When you read about ‘randomisation’ in a study, more often than not this is referring to randomising of individuals – often in a computer-generated 1:1 fashion – to an intervention or control group. However, there are circumstances where the nature of the intended intervention mean that randomisation of groups, or clusters, may be more attractive.
For example, in the present study it was the General Practices that were randomised, and therefore all participants enrolled in the study at a GP randomised to the intervention would be in the intervention group, while all participants enrolling at a GP randomised to the control group would be control participants. Some biomedical purists do not look favourably on cluster randomisation, as they are deemed to compromise on precision.
However, ‘compromise’ is the key word here: what is being gained is pragmatism, as generally the cluster is within the community, and the trial is therefore being conducted under more “real world” conditions. Cluster-randomisation also helps avoid what is known as ‘contamination’, which is where there is potential for an intervention group and placebo group to overlap and have contact. For example, in this study, if individuals were randomised from the same GP practice, there could be contact between participants in the intervention group and control group in the practice. Cluster-randomisation also captures a more whole-population approach, when often many RCTs may lack applicability to the wider population.
By conducting a cluster RCT, it is therefore possible to know that an intervention could be scaled. So, it depends on the research question, the type of intervention, and the context, and there may be times when randomising larger level units is appropriate.