What are the disadvantages of cluster randomization

It is not possible to analyse a cluster-randomized trial as though individual patients had been randomized. The difficulty arises because the experimental unit is different from the observation unit, and conventional statistical methods are predicated on these being one and the same. In an individually randomized trial, one can consider each patient's outcome to be independent of that for the other patients in the trial. In a cluster-randomized trial, that is not possible, as patients within a cluster are more likely to have similar outcomes than a random sample of patients from all the clusters. The trial analysis must take this into account, as must sample size calculations. The consequence is that a cluster-randomized trial requires more (often many more) patients than an individually randomized trial; in other words a cluster-randomized trial of N patients will have less power than an individually randomized trial of N patients. In fact, the power of the trial is determined by the number and size of the clusters; the smaller and more numerous the clusters, the less the impact on power. This fact makes it important to use these designs only when they are absolutely essential.

0 0

Post a comment