Cluster randomised trials

Rather than randomise each subject individually it is sometime necessary to randomise at a group level by, for example, therapist, general practice or school. Thus, the patients treated by an individual therapist, or patients within a general practice, or children within a school become part of a cluster. That cluster might be randomised to a treatment or control group. Subjects within a cluster tend to be correlated and this should be accounted for in the design and analysis of the trial. If the “cluster effect” is not accounted for the power is the study is likely to be reduced with misleadingly narrow confidence intervals and reduced P values.

Resources:

• Campbell MJ, Donner A, Klar N. Developments in cluster randomizes trials and Statistics in Medicine. Statistics in Medicine 2007; 26(1): 2-19.

• Bland JM, Kerry SM. Trials randomised in clusters. BMJ 1997; 315(7108): 600

• Kerry SM, Bland JM. Analysis of a trial randomised in clusters. BMJ 1998; 316(7124): 54

Examples of published studies

• Cooper P, Chico M, Vaca M, Moncayo A, Bland JM, Mafla E, Sanchez F, Rodrigues L, Strachan D, Griffin G. Effect of albendazole treatments on the prevalence of atopy in children living in communities endemic for geohelminth parasites: a cluster-randomised trial. The Lancet 2006; 367(9522): 1598-1603.