ABSTRACT
ABSTRACT I review the cluster-randomized trial (CRT), a study design well suited to comparative effectiveness research, in that it often allows treatment under the study protocol to closely mimic the natural implementation
of the study arms in practice. In addition, CRTs are often far cheaper per participant than traditional individually randomized trials (IRTs) and can be used in settings where individual randomization may be impossible, or would muddy, not clarify, the effect of the study arms. However, CRTs offer less power per participant so that an assessment of the trade-off of cost versus power may be necessary if both designs are plausible for the question under study. Alternatively, an increased sample size may be needed for a CRT to equal the power of an IRT for a given effect size. CRTs require more complex methods of data analysis than traditional trials, and power or sample size estimation is more complex and has received less methodological attention than traditional trials. I motivate the design and explain its suitability for effectiveness questions. I then provide several examples of specific CRTs, along with rationales for why a CRT was better than or necessary in preference to an IRT. Next, I provide heuristics for understanding the above assertions and discuss analysis methods and power/sample size methods for CRTs.
