What is a SMART?

The Sequential, Multiple Assignment, Randomized Trial (SMART) is an experimental design used to construct an optimized adaptive intervention.

In the SMART experimental design, study participants progress through multiple stages. At each stage, a participant may be randomized to one of two or more intervention options. Each stage of randomization helps researchers compare the relative effectiveness of multiple intervention options.

Why use a SMART?

Researchers in behavioral science and other fields may need to answer a number of scientific questions before they recommend an adaptive intervention or evaluate its effectiveness. SMARTs are for researchers who want to answer multiple questions at multiple stages as a means of maximizing the potential impact of their adaptive intervention.

See if the SMART experimental design is right for your project

What types of questions can a SMART answer?

Intervention sequencing questions.

What intervention option should we offer first?

Among individuals who do not respond to first stage treatment, what is the effect of augmenting treatment compared to the effect of intensifying treatment?

Among individuals who respond to first stage treatment, what is the best frequency for continuing treatment?

Temporal interaction effect questions.

Are there positive or negative interaction effects between the intervention options delivered sequentially?

Are there later-stage treatment options that compromise the benefits of earlier treatment options?

Which first-stage treatment yields insights that inform a more optimal second stage treatment decision?

Intervention tailoring questions.

Should the first-stage intervention differ based on comorbid health conditions, history of prior treatment, or other factors known prior to starting treatment?

How should changes in patients’ health metrics determine intervention options in stage 2 and beyond?

What should trigger the transition from stage one to stage two—adherence, therapeutic response, or some combination of the two?

Questions about defining response.

What is the best way to define response and non-response to first-stage treatment?

How stringent or lenient should we make the cutoff that defines response and non-response to first-stage treatment?

How often should individuals be monitored in an adaptive intervention?

An optimized adaptive intervention…

Makes accommodations for a heterogenous target population.

Capitalizes on between-person heterogeneity to provide the best intervention option for each individual.

Takes advantage of positive synergies between sequential intervention options.

Avoids negative interactions between sequential intervention options.

Is feasible to deliver in real-world practice settings.

Is acceptable to the people delivering the intervention and those receiving it.

Is replicable by future interventionists and future behavioral scientists.

What does it mean to construct an optimized adaptive intervention?

We think of the questions listed above as optimization questions because their answers can help researchers construct a high-quality adaptive intervention that has the potential to improve health outcomes in the long-term for a greater number of individuals.

These questions are fundamentally different from those addressed by confirmatory randomized trials, which focus on confirming/evaluating how well a treatment works.

Learn more about the optimization of adaptive interventions.


A SMART for Constructing a Mobile-technology Assisted Adaptive Intervention for Weight Loss

This SMART is designed to construct an adaptive intervention that integrates mobile technology in the treatment of obese and overweight adults.

Optimization Questions

To determine the effects of first-stage intervention options.

Does offering an App alone initially lead to weight loss by month 6 that is noninferior to offering App + Coaching?

To determine the effects of second-stage intervention options.

Is it more beneficial, in terms of weight loss at 6 months, to offer vigorous or modest augmentation to those who did not respond to the initial intervention?

To identify moderators.

Does the effect of first stage intervention options vary by baseline information (e.g., gender, BMI); does the effect of second-stage intervention options vary by baseline and time varying moderators (socioeconomic status, self-efficacy, extent of self-monitoring)?

To compare adaptive interventions.

Is it better, in terms of weight loss at 6 months, to (a) start with the App alone, and then augment modestly for non-responders and continue (stay the same) for responders, or (b) start with App+Coaching, and then augment vigorously for non-responders and continue for responders?

SMART Design Schematic