Introduction to Sequential Multiple Assignment Randomized Trials (SMART) for the Development of Adaptive Interventions: Two Case Studies in Autism

Friday, May 16, 2014: 1:30 PM
Marquis BC (Marriott Marquis Atlanta)
D. Almirall and S. A. Murphy, University of Michigan, Ann Arbor, MI
Background:  An adaptive intervention is a sequence of individually-tailored treatment decision rules that can help guide behavioral or pharmacological strategies, or their combination, for improving outcomes in the treatment of autism. Adaptive interventions realize this goal by flexibly tailoring treatments to individuals with autism when they need it most and in the most appropriate dose, thereby reducing inappropriate variance in treatment delivery and increasing the total number of individuals who benefit from treatment. Once developed, adaptive interventions can be used as treatment guidelines in public health or clinical practice. The sequential multiple assignment randomized trial (SMART), a new type of research or study design, permits researchers to collect high-quality, experimental data that can be used to develop an optimal adaptive intervention. SMARTs realize this goal by evaluating what are the best treatment components and tailoring variables that lead to an optimal adaptive intervention.


In recent years, there has been a surge of interest in a variety of scientific disciplines in developing and evaluating adaptive interventions, specifically, using the SMART study design. This includes two SMART studies aimed at improving social communication in children with autism. However, despite this overwhelming interest, adaptive interventions and SMART are still new to many autism researchers. The objective of this talk is to fill this education gap concerning adaptive interventions and SMART and to encourage new ideas in the science of adaptive interventions in autism.


We will introduce adaptive interventions and discuss why they are important. We will also introduce the SMART study design, and briefly discuss SMART study design principles, including common choices for primary and secondary aims, and allay concerns that SMART designs necessarily require prohibitively large sample sizes.  These ideas will be illustrated using two SMART case-studies in autism. Both SMARTs were designed to develop and evaluate adaptive interventions for improving spoken communication in children with autism who are minimally verbal.


Adaptive interventions provide a framework to guide the individualization (personalization) of treatment for individuals with autism. SMARTs can be used to examine critical questions in the treatment of autism, such as, “Among children with autism who are minimally verbal, is it best to begin behavioral treatment with discrete trials training (DTT) or with joint-attention social play (JASP)?”; “Among children who do not respond successfully to 12 weeks of behavioral treatment, what is the effect of a treatment that combines DTT and JASP principles (DTT+JASP)?”; and “Who are the types of children more likely to benefit from combined DTT+JASP versus continuing longer with either DTT or JASP?”.


Adaptive interventions have great potential for improving outcomes in individuals with autism, by capitalizing on individual heterogeneity in response to treatment. However, the science of adaptive interventions is still in its infancy in autism. SMART studies have the potential to improve our scientific understanding of how to develop the best possible adaptive intervention.