Researchers use Sequential, Multiple Assignment, Randomized Trial (SMART) designs to investigate the effectiveness of intervention components (or combinations of them) at multiple stages. At the beginning of each stage, participants may be assigned their next intervention through randomization (e.g., 50% receive treatment A, 50% receive treatment B). But, depending on the SMART, not all patients are randomized to a new intervention at every stage. In some SMARTs, only some participants (e.g., those most likely to benefit from additional intervention) are re-randomized. For example, in the SMART shown below, only those who did not respond to their assigned first-stage treatment are randomized to intervention D or intervention E.

In a SMART with this design, participants follow different treatment pathways determined by randomization and their response status. Each pair of pathways defines an adaptive intervention (above). For example, pathway A-C (available only to responders) combined with pathway A-D (available only to non-responders) represents one adaptive intervention (AI1). Pathway A-C combined with pathway A-E represents another adaptive intervention (AI2). We call these AIs “embedded” because their designs are readily apparent in the design of the SMART.

A common primary aim in a SMART is to compare the relative effectiveness of these embedded adaptive interventions. Our collection of downloadable and free-to-use software includes programs for conducting this type of analysis with data arising from a SMART.