AAP Grand Rounds • August 2011 15 O ver the last century, the randomized clinical trial has become the gold standard for interventional studies. Typically individuals are randomly assigned to an ex- perimental group or a control group. What if it was impractical or unethical to randomize individuals to a treatment group? his is where the cluster randomized trial design comes in. Cluster randomized trials randomize groups, rather than individual subjects, to receive diferent interventions. For example, in this issue one such trial examined the efect of hand hygiene on inluenza infection rates and absenteeism in elementary schools in Cairo, Egypt. Since it is diicult to have each student in a school gain expo- sure to diferent hand hygiene practices without some overlap between the individuals, it made most sense to randomize entire schools to the intervention, rather than each individual. 1 he units of randomization for these studies are diverse, and ex- amples include classrooms, grades, physician practices, hospitals, or even entire communities. Because of the unique design and reporting issues these studies present, an updated version of the CONSORT statement for reporting randomized trials was pub- lished in 2001, providing standards for the reporting of cluster randomized trials. 2,3 he cluster design afords several attractive features for large- scale intervention or prevention trials, including administrative eiciency, a decreased risk of the control group receiving some or all of the intervention (ie, contamination), and potential for improved subject adherence. 4 hese trials are a particularly ap- pealing option to address group or environmental risk factors that are challenging to study at the individual level. Although the beneits of cluster randomized trials are allur- ing, there is a loss of statistical eiciency compared with studies that randomize the same number of individual subjects. 4 his loss stems from the fact that the groups that are randomized are not always the same as the groups that are analyzed, so standard statistical methods are not applicable. 5 For example, clinics may be randomized to receive one or another intervention, but the analysis is of the individuals seen within those clinics. Since members of a group are more likely to have factors in common as a result of interacting as a unit (called nesting), they cannot be treated as completely independent individuals and observa- tions within a group are likely to be correlated. Other statistical measures, eg, intracluster correlation or variance inlation factor, must be employed in the statistical analyses of cluster trials. Cluster randomized trials also present ethical challenges. In randomized trials, informed consent must be obtained to fully inform study participants about the nature of the experiment and the process of random assignment of the study interven- tion. When studying large groups of people, it is oten impracti- cal to obtain each individual’s consent. Frequently, this issue is addressed by having a surrogate policymaker give consent on behalf of the group of people they are representing. Regardless of how consent is obtained, it becomes even more important in cluster randomized trials to report how consent was obtained at the individual and group level, who the decision makers are, and how those delegated oicials were selected. 4 Cluster randomized trials can be administratively eicient studies to explore lifestyle interventions or disease prevention strategies. Because of the statistical and ethical challenges of designing a study with intact social groups as the unit of ran- domization, it is important for investigators to state the rationale for adapting a cluster randomized design, how clustering efects were accounted for in sample size calculations and the statistical analysis, and how both clusters and individuals lowed through- out the entire study. 3 References 1. Clamp M, et al. BMJ. 1998;316:1576-1579 2. Moher D, et al. Lancet. 2001;357:1191-1194 3. Campbell MK, et al. BMJ. 2004;328:702-708 4. Donner A, et al. Am J Public Health. 2004;94:416-422 5. Donner A. J Royal Statist Soc: Series C (Applied Statistics). 1998;47:95-113 WEIGHING THE EVIDENCE Cluster Randomized Trials By Todd A. Florin, MD, The Children’s Hospital of Philadelphia, and Samir S. Shah, MD, Cincinnati Children’s Hospital Medical Center