Featured Article: Analyzing Data from a SMART to Prevent Alcohol Abuse Relapse

April 6, 2017:Ishani

Response to substance abuse treatment can look very different between individuals and even within individuals at different points in time. Sequential, multiple assignment, randomized trials (SMARTs) are being used to develop interventions that adapt based on individual needs and circumstances. New methods for data analysis show promise for improving intervention developers’ ability to tailor an intervention even more specifically to an in individual’s needs for a broad range of health issues, including substance use. In a recent article in the journal Addiction, Methodology Center researchers Inbal (Billie) Nahum-Shani, Daniel Almirall, and their collaborators demonstrate the utility of Q-learning, a method developed in computer science, for the analysis of data from a SMART to prevent relapse among individuals with alcohol use disorders. Q-learning helped the authors identify a subset of individuals who appeared to be responding to treatment, but who needed additional treatment to maintain progress.

The authors analyzed data from 250 participants in the Extending Treatment Effectiveness of Naltrexone (ExTEND) trial (D. Oslin, P.I.; NIAAA; R01 AA014851). ExTEND was a 24-week SMART that examined how to build an adaptive intervention to prevent relapse among people with alcohol use disorders using the drug naltrexone. Naltrexone is promising for treating alcohol dependence, but there is a broad range of responses to the drug. The researchers used data from the ExTEND SMART to construct an adaptive intervention that is tailored even further based on an individual’s response to the initial treatment.

The resulting adaptive intervention recommends additional treatment for a subset of participants who, though initially classified as responders to Naltrexone, are likely to benefit from a more intense maintenance intervention. Q-learning is similar to moderated regression analysis, but it is suitable for examining whether and how certain covariates are useful in developing or improving an adaptive intervention. Lead author Inbal “Billie” Nahum-Shani said, “Q-learning can help us identify new ways to tailor treatments beyond the tailoring variables we typically include in a SMART by design. The goal of this paper is to provide an accessible overview of this method to investigators in the area of substance use disorders and to demonstrate how it can help advance the science of adaptive interventions in this important field.”

By using Q-learning with data from a SMART, researchers can build empirically validated adaptive interventions for a broad array of health problems.

Open the article. (Journal access required.)


Nahum‐Shani, I., Ertefaie, A., Lu, X. L., Lynch, K. G., McKay, J. R., Oslin, D. W., & Almirall, D. (2017). A SMART data analysis method for constructing adaptive treatment strategies for substance use disorders. Addiction.