OBBI SIG Talks with Daniel Almirall about Experimental Designs for Adaptive Interventions
Ahnalee Brincks, PhD, Optimization of Behavioral and Biobehavioral Interventions Special Interest Group (OBBI SIG) Communication Chair
The Optimization of Behavioral and Biobehavioral Interventions Special Interest Group (OBBI SIG) recently interviewed Daniel Almirall, Ph.D., Associate Professor in the University of Michigan Institute for Social Research. Dr. Almirall is a leading statistician in the area of experimental design for adaptive interventions, with an emphasis in sequential multiple assignment randomized trials (SMARTs). He is currently co-PI on a 5-year R01 from the National Institutes of Health focused on the development of analysis methods for repeated outcome measures arising from a SMART.
OBBI SIG: On your website you describe adaptive interventions as “sequences of individually tailored decision rules that specify whether, how, and when to alter the intensity, type, or delivery of treatment at critical decision points in the medical care process.” How did you become interested in adaptive interventions?
Dr. Almirall: When I was a doctoral student (2000-2007), behavioral medicine scientists were just starting to talk about adaptive interventions. I was working on a special case of these involving predetermined sequences of treatments, but I knew that it was unlikely we would know what sequence of interventions was best for a person using baseline characteristics alone. What I really wanted to be doing was research on adaptive interventions, in which intervention decisions could change based on the changing characteristics/context of the participant. After finishing my PhD, I accepted a position at Duke University as an assistant professor of biostatistics. Ultimately, I left that job to devote 100 percent of my time to developing expertise in adaptive interventions through an advanced post-doctoral training experience at the University of Michigan. The past seven years in this research area have been really fun.
OBBI SIG: You have been involved in the design and implementation of numerous SMARTs across a range of behavioral interventions. What do you see as essential to the success of a SMART study?
Dr. Almirall: My answer to this question may surprise you because it is not specific to SMART. I think there are two things that are essential to the success of any trial about adaptive interventions. The first, which is most important, is whether there is a strong rationale for the study design. Here, I feel strongly that the scientific questions should drive everything, including whether to use a SMART design at all. Second, it is critical to understand the distinction between adaptive intervention design (i.e., what we do when we have our clinician hats on) and study design (i.e., what we do when we have our scientist hats on). It is surprising to me how often these two ideas are mixed up. For example, as part of the adaptive intervention, it is important to have a plan for preventing or addressing treatment drop-out (e.g., what will you do if your participant doesn’t show up to the clinic, including failing to provide information needed to determine response or non-response?); and it is important to have a separate plan—this one is part of the study design not the adaptive intervention design—for preventing or addressing study drop-out (e.g., to obtain research outcomes regardless of treatment drop-out). This distinction is critical to the success of any behavioral intervention trial (SMART or not).
OBBI SIG: What would you recommend to an investigator who is interested in learning more about adaptive interventions and related experimental designs?
Dr. Almirall: We now have numerous manuscripts in the literature that define adaptive interventions and how they help guide clinical practice, and that is the first thing a new investigator should understand. The second step is to determine the scientific questions that need to be addressed regarding adaptive interventions in the investigator’s scientific area. Then, she or he should develop an understanding of the various experimental designs that can be used to optimize an adaptive intervention. One is a SMART, but there are others. I am currently finishing a chapter for Linda Collins’ forthcoming book (more information here) that discusses three types of experimental designs available for adaptive interventions.
OBBI SIG: What are you most excited about in your current research?
Dr. Almirall: I am currently very excited about trial designs for optimizing ‘multilevel adaptive interventions’, specifically cluster-randomized SMARTs. A few years ago, my colleague Amy Kilbourne and I designed the first-ever cluster-randomized SMART to learn how best to intervene adaptively with community-based mental health clinics to encourage them to adopt evidence-based practices for mood disorders. The intervention target was the clinic, and if the clinic was not showing change after six months, an alternate intervention was offered. This led us to think about how to design a SMART where clinics were randomized and re-randomized; yet with outcomes at the patient level. This first study was just the tip of the methodological iceberg: behavioral intervention scientists likely will be interested in studies that address scientific questions about how best to adapt intervention based on both clinic and patient levels (e.g., hybrid type-2 implementation study designs). Currently, there is no methodological work in this area. I am extremely excited about working out the design and statistical methods that will lead to better science in this area.