What is a cluster randomised trial?

A cluster randomised trial (CRT), also known as a group randomised trial or community randomised trial, is a randomised controlled trial in which groups of individuals (clusters eg school, clinic, village) are randomised.

Well-known textbooks on CRTs

Example CRTs

The CRT design has been used in many settings and for many common areas of health research, but most commonly in health services research where health care providers are randomised, to evaluate infectious disease programs, and in developing countries where whole communities are randomised.

Examples of clusters that have been selected as the unit of randomisation in a CRT design include:

General practices
Health facilities/clinics
Wards (within hospitals)
Nursing homes
Communities
Schools

Examples of interventions that have been evaluated using a CRT design include:

Health treatment and prevention programs
Vaccination programs
Behavioural and organisational change interventions
School-based interventions

Why do a CRT?

CRTs can be harder to design, require more subjects than individually randomised trials, and are often more prone to biases. However, reasons why we might choose to conduct a CRT rather than an individually-randomised trial include that the intervention is implemented at the cluster level, there are practical and/or ethical difficulties in randomising at individual level (although randomising clusters to avoid consent is not acceptable), to avoid issues of contamination, or to estimate indirect effects, for example in vaccination trials. Below are various examples of trials that have been conducted for these reasons:

The intervention is implemented at the cluster level or it is logistically easier or more ethical to administer to groups of individuals.

To avoid issues of contamination, e.g. for infectious diseases, or to estimate indirect effects, for example in vaccination trials.

Correlation in CRTS

In comparison to trials in which individual randomisation is used, outcomes from individuals in the same cluster will tend to be more similar to each other than a random sample for the whole group and this must be accounted for in both the design and analysis of CRTs. The intracluster correlation (ICC) is a measure of the between-cluster variation. It can also be thought of as a measure of the homogeneity of individuals within a cluster. There are three main interpretations of the ICC:

  • The proportion of the total variance due to between-cluster variation

  • The correlation between members of the same cluster (more generally referred to as the intraclass correlation)

The most widely used interpretation in CRTs is the proportion of variance. For more information, see:

Even low ICC values should be taken into account in the design and analysis of a CRT. For a paper with some example ICC values, see the following:

Types of design

The standard design, the parallel two-arm design, randomises clusters to one of two arms or conditions, eg treatment or control, and measures outcomes in individuals in both arms on the same follow-up measurement schedule. Depending on the research question, the same individuals may be followed up over time (a cohort design) or different individuals may be sampled at different time points (a cross-sectional design). When the parallel CRT design is augmented by the addition of baseline measures before randomisation, this is referred to as the parallel cluster randomised trial with before and after observations (CRT-BA).For resources, see resources on the parallel-arm CRT design.

Alternatives to the parallel two-arm design include the crossover design and the stepped wedge design, which is a modification of the crossover design. In the crossover design, clusters are randomised to a treatment sequence, and in the stepped wedge design all clusters start in the control condition and eventually end in the treatment condition, where the treatment start time is randomly allocated. See what is a SW-CRT? and resources on the stepped wedge CRT design for more information on this.