| Summary: | I read Gepner and colleagues’ article with interest. In their most common form, randomized trials compare groups of persons receiving different interventions studied simultaneously. Each participant receives 1 intervention selected at random at enrollment. The probabilities of being allocated to each treatment are usually equal (for example, an allocation ratio of 1:1). That all participants have the same probability of receiving each treatment is implied. That principle is violated if the allocation ratio is changed (for example, from 1:1 to 1:2) partway through the trial. Although all participants are randomly assigned to 1 of the same set of possible treatments, bias can arise because the sets of participants recruited in the 2 phases of the trial may differ—for example, those recruited in the second part may be healthier than those in the first part. The correct approach is to separately analyze the data for each part of the trial and then combine the results
|
|---|