I have heard Dr. Collins say that the SMART is NOT an adaptive trial design. I don’t understand this, can you explain?
The SMART is NOT an adaptive trial design. Part of the confusion may be due to subtle differences in the way the terms “design” and “adaptive” are used.
We’ll go through the explanation step by step.
- Let’s draw a distinction between intervention design and trial design. Intervention design refers to the approach to conducting an intervention—what makes up the treatment, how it is delivered, etc. Trial design refers to the approach to conducting an experiment—what experimental conditions are included, how random assignment is to be conducted, etc.
- Let’s draw a distinction between fixed and adaptive. The word “adaptive” can appropriately be applied in many different contexts, including trial design and intervention design. The word adaptive simply means that a priori rules are in place for making strategic changes in approach as needed “on the fly.” In the literature on MOST, we have typically used the word “fixed” to indicate the opposite of adaptive, that is, an approach that does not (or at least is not intended to) change over time.
- Intervention designs may be either fixed or adaptive. In a fixed intervention design, the treatment strategy calls for all participants to receive the same treatment (any variability is accidental). By contrast, “an adaptive intervention is a sequence of pre-specified decision rules that can be used to guide whether, how, or when—and based on which measures—to alter an intervention or intervention component (e.g., treatment type, duration, frequency or amount) at critical decision points during the course of care” (Almirall, Nahum-Shani, Wang, & Kasari, 2018). In other words, in an adaptive intervention the treatment is varied strategically “on the fly.” Here the purpose of adaptation is to obtain a good outcome for all participants, using the finite resources at hand. A good adaptive intervention design provides participants with the type and amount of treatment they require and no more, so that finite resources are not wasted on those who do not need them and instead are diverted to participants who will benefit. (For more about adaptive interventions, see the Almirall et al. chapter in Collins & Kugler, 2018.)
- Experimental designs may be either fixed or adaptive. The terms fixed and adaptive have the same general meaning but are used a bit differently in this context as compared to how they are used when describing intervention designs. In a fixed experimental design, the entire design is specified a priori with the idea that there will be no changes for the duration of the experiment. By contrast, in an adaptive trial design, aspects of the experimental design, such as how random assignment is conducted, may be adapted over time in response to emerging results, using a Bayesian approach. For example, random assignment procedures could be changed such that the probability of being assigned to a certain conditions increased while other assignment probabilities decreased. In other words, in an adaptive trial design, aspects of the experimental design are varied strategically “on the fly.” Here the purpose of adaptation is to obtain the highest-quality scientific information possible within the finite resources available to conduct the trial.
- Now let’s highlight one more distinction, the distinction between optimization trial designs and evaluation trial designs. Optimization trial designs are used when seeking empirical information about the performance of individual components of an intervention, for the ultimate purpose of intervention optimization. Evaluation trial designs are used when directly comparing an intervention as a package to a control or to one or more different interventions. As the names imply, optimization trial designs are used in the optimization phase of MOST, and evaluation trial designs are used in the evaluation phase of MOST.
- OK, now keep all that in mind as you review the following:
- The SMART is an optimization trial design. It is used in the optimization of certain types of adaptive interventions (for more about this, see Chapter 7 in Collins, 2018).
- The SMART is considered a FIXED trial design. When a SMART is conducted, all aspects of the experimental design are determined before the experiment starts, and are not subject to change as the experiment proceeds. It is true that in the sequential randomization in a SMART, a subsequent randomization may be conditional on the observed value of a tailoring variable. However, every aspect of this, including the tailoring variable, decision rule, and randomization probability, are fixed in advance and are not subject to any updating as the trial proceeds.
- To date, adaptive clinical trials are all RCTs, so they would be used in the evaluation phase of MOST. To our knowledge there are not (yet) any adaptive optimization trial designs.
To sum up: The SMART is a fixed optimization trial design. It is used in the optimization of certain types of adaptive interventions. By contrast, the adaptive clinical trial is a Bayesian evaluation trial design.
All Frequently Asked Questions
Do you see MOST intersecting with implementation science, and if so, how?
Can I expect reviewers of grant proposals to understand MOST? If not, how should I handle this in the application? There is not enough room to provide a lot of background.
I work in the prevention field, in which the horizon on the outcome is very far away. How can I optimize under these circumstances? It seems like it would take decades.
Should I apply for funding to support all three phases of MOST?
How can I find out what other studies using MOST have been conducted in my area?
How can I obtain the background I need to write a grant proposal involving MOST?
Whether you are looking for additional support as you prepare a grant proposal involving MOST or practical information helpful in managing your optimization trial, this section provides resources for a deeper dive into intervention optimization.
REDCap with Most
The goal of this manual is to show how one might setup a REDCap project to support a research study with multiple conditions, such as factorial experiments common in the Multiphase Optimization Strategy (MOST) framework.
Establishing a conceptual model and understanding optimization
Sometimes the conceptual model is not a model of a health behavior per se, but a model of maintaining treatment fidelity, promoting adherence or compliance, or the like. The conceptual model is explained in more...
Common misconceptions about factorial experiments
A factorial experiment is essentially an RCT with a lot of experimental conditions, and therefore is extremely difficult to power.
Informal introduction to factorial experimental designs
The purpose of this page is to clarify some concepts, notation, and terminology related to factorial experimental designs, and to compare and contrast factorial experiments to randomized controlled trials (RCTs). A more in-depth introduction can...
LET’S STAY IN TOUCH
Join the cadio Mailing List
Keep up to date with the latest news, events, online courses, and resources from cadio.