Cluster randomised controlled trials

This section contains the following:


A cluster randomized trial involves social units or clusters of individuals, rather than individuals themselves, being randomised to different intervention groups.  In such circumstances, individual patient level data are still collected.  The units of randomization in these studies are diverse ranging from families, households, worksites, and hospitals to entire neighbourhoods or communities.  The primary difference between a cluster and patient randomized trial is that participants within one cluster may be more alike (correlated) than participants in another cluster.  The measure of this correlation is known as the intraclass correlation.  As the correlation increases, a trial requires an increasing number of participants compared to a patient randomised trial (see Sample size justification ). See further reading and examples for more details.  Whilst many characteristics of cluster randomized trials are in common with patient randomized trials, the adoption of a cluster randomized design has some implications for the design, conduct and analysis of the trial.  Thus, most sections of this protocol support tool apply directly to cluster randomized trials.  Special consideration needs to be given to cluster trials in the following sections (the specific issues are outlined in the relevant subsections):

Back to top

Illustrative example - From the Lancet

Populations served by single health centres were randomised to receive improved services for the treatment of sexually transmitted diseases.  (Lancet 1995; 346: 530-6)

Back to top

Illustrative example - from the American Journal of Epidemiology

Families were randomised to evaluate the effectiveness of treated nasal tissues versus placebo tissues upon the incidence of respiratory illness. (American Journal of Epidemiology 1988; 128: 1162-72)

Back to top

Illustrative example - WELLINGTON trial

Farms were randomised to receive a community health worker intervention to improve tuberculosis control in the agricultural sector of the Winelands Health District, South Africa. (WELLINGTON trial - go to protocol)

Back to top

Further reading

Sally M Kerry and J Martin Bland. Statistics notes: Sample size in cluster randomisation. BMJ 1998; 316: 549.

Sally M Kerry and J Martin Bland. Analysis of a trial randomised in clusters. BMJ 1998; 316: 54.

Ukoumunne OC, Gulliford MC, Chin S, Sterne JAC, Burney PGJ. Methods for evaluating area-wide and organisation-based interventions in health and health care: a systematic review. Health Technol Assess 1999; 3(5) .

Donner A, Klar N. Design and analysis of cluster randomized trials in health research. London: Arnold, 2000.

Back to top

This page was last updated 4th June 2004.