Consider this example. In the design of the prototypical SMART shown in the figure below, only non-responders to stage 1 treatment are re-randomized to modest versus vigorous augmentation. In trials of this type, we refer to the response/non-response variable as an “embedded tailoring variable” because it is embedded in the trial design and used to restrict who gets re-randomized in stage 2. Note that as a result of this restriction, this example SMART includes four embedded adaptive interventions. Click here to learn more about this related concept.
SMARTs like this are used when it is known that modest augmentation or vigorous augmentation should not be considered (i.e., one or the other is not a viable stage 2 treatment option) for responders to first-stage intervention.
Note that not all SMARTs require an embedded tailoring variable.