Why Use a SMART Design to Build an Adaptive Intervention?

Adaptive Interventions

blank SMART diagramAdaptive interventions have four critical components.

  1. Sequence of decisions regarding patient care – Most interventions require decisions such as, “If the patient is unresponsive to the initial treatment, what treatment should we provide next?” or “Once the patient has stabilized, what treatment is needed to prevent relapse?”
  2. The set of treatment options at each decision point – For example, if a patient is unresponsive to a drug, should the dosage be increased, should the drug be discontinued, or should counseling be increased? All of these are treatment options.
  3. Tailoring variables – These are the factors used to trigger a change in the treatment. These can be things like early signs of nonresponse, manifestation of side effects, or environmental or social characteristics. The idea is to identify the variables that best indicate when the appropriate treatment has changed.
  4. A sequence of decision rules – This links the first three components. There should be one decision rule per decision. The tailoring variables provide information about which of the treatment options is most appropriate for the patient at the time of the decision.
A series of decision rules must be established to guide treatment based on individual characteristics.

Adaptive interventions allow researchers

  • to help patients who do not respond to initial treatment,
  • to respond if the effectiveness of an intervention wanes over time due to changes in the patient’s situation or response,
  • to prioritize when the patient possesses comorbid conditions (e.g., depression and alcoholism),
  • to address relapses (as are common when treating substance use),
  • to decrease burden and/or cost of the intervention when a patient is stable, and
  • to respond when patients do not adhere to a treatment.

Why Conduct a SMART?

SMART designs allow for the testing of multiple potential adaptive interventions along with the proposed tailoring variables.

Sequential, multiple assignment, randomized trials (SMARTs) provide data that enables the development of high-quality adaptive interventions. In a SMART there is a separate stage for each of the critical decisions involved in the adaptive intervention. At each stage, all participants are randomly assigned to a treatment option. By randomizing participants multiple times, scientists can assess the effectiveness of each stage. So, several adaptive interventions are embedded within each SMART design for testing. This allows testing of the tailoring variables and the intervention components in the same trial, and it allows clinicians to develop the best decision rules based on research rather than a priori decisions. Read about a SMART that tested the level of the tailoring variable and the treatment options for the construction of an adaptive intervention for relapsing alcoholics.

For more information about SMART designs see Lei et al. (2012).


SMART vs Other Experimental Designs

Aside from the advantages described above, SMART designs also allow comparisons of different treatment options within the context of what happens in later stages. Other than a SMART design, a scientist has two alternatives for testing and building adaptive interventions: multiple, one-stage-at-a-time, randomized trials; or a randomized trial that compares fully-formed adaptive interventions.

Multiple, One-Stage-at-a-Time, Randomized Trials

Imagine that you have three primary research questions:

  • Which intervention should I provide initially?
  • Which intervention should I provide to responsive patients?
  • Which intervention should I provide to non-responsive patients?

By identifying the initial treatment first, and then separately identifying the best treatments for responders and non-responders, you could create an adaptive intervention, but you would lack some of the information provided in a SMART design.

Compared to multiple, one-stage-at-a-time, randomized trials, SMART designs provide

  • better ability to compare the impact of a sequence of treatments, rather than examining each piece individually,
  • increased ability to test tailoring variables, and
  • decreased impact of cohort effects (because non-responders in a SMART are more representative of all non-responders to initial treatment, including both highly motivated and less motivated individuals).

Randomized Trial of Fully-Formed Adaptive Interventions

By studying prior research, you may decide to start all patients on the least intensive treatment and increase treatment when necessary. This would also not provide you all the information that a SMART design could.

Compared to a randomized trial that compares fully-formed adaptive interventions, SMART provides the ability to understand what is working (and not working) within the intervention. SMART allows you to compare options including dose, type, and timing of treatments. Also, secondary analysis of data from a SMART allows researchers to assess tailoring variables.

For more information about the relative merits of SMART, see Lei et al. (2012).

Read an example of SMART intervention for relapsing alcoholics.

See our recommended reading for SMART.

Interested in other designs for behavioral interventions? Read about our work on optimizing behavioral interventions.


Lei, H., Nahum-Shani, I., Lynch, K., Oslin, D., & Murphy, S. A. (2012). A “SMART” design for building individualized treatment sequences. Annual Review of Clinical Psychology, 8, 14.1-14.28. doi: 10.1146/annurev-clinpsy-032511-143152 PMC Journal- In process


Last updated: May 11, 2020