Mobile Health

There are three main approaches that can be currently used to optimize JITAIs. As Collins et al. (2005) (link) note, behavioral interventions have traditionally been optimized using a series of randomized controlled trials (RCTs). As an experimental method, RCTs are designed to assess whether, on average, the intervention package as a whole had an effect on the behavior of interest. However, RCTs are not designed to investigate which components of an intervention are efficacious, when they are efficacious, or what psychosocial or contextual factors influenced their efficacy. In secondary analyses of RCT data, random assignment facilitates the assessment of treatment-effect moderation with respect to baseline characteristics (e.g., age), which provides information about static factors that influence efficacy of the intervention package (e.g., Hekler et al., 2013 (link)). However, this is insufficient for JITAI development, where we must evaluate what time-varying factors influence the efficacy of different components in order to understand when and in what contexts a particular intervention option should be delivered. If there is sufficient variability across time in receipt of a component due to non-adherence or implementation problems (e.g., participants’ medication adherence goes up and down over time), then RCT data can be used to investigate time-varying moderation and effects of time-varying components. However, as is well known, in such secondary analyses, baseline randomization offers no protection against causal confounding, and results are subject to bias. Thus, important questions pertaining to JITAI optimization—when a particular intervention component should be delivered, what factors at the time of delivery affect whether the intervention component will have the desired effect, and so on—are poorly addressed by data from standard RCTs.
As an alternative to RCTs, there has been a resurgence of interest in the use of single-case experimental designs (SCEDs) to develop and evaluate mHealth interventions (Dallery & Raiff, 2014 ; Dallery, Cassidy, & Raiff, 2013 (link)). SCEDs enable highly efficient preliminary efficacy testing of an intervention component, since each participant acts as his or her own control. However, in their traditional forms (i.e., reversal, multiple-baseline, and changing-criterion designs), SCEDs are of little help for determining the time or context in which a certain intervention option is most efficacious. This is because SCEDs often do not clearly articulate decision points for intervention-component delivery or systematically examine moderators of observed effects.
To overcome the limitations of RCTs in guiding intervention design, Collins and colleagues (Collins, Chakraborty, Murphy, & Strecher, 2009 (link); Collins et al., 2005 (link)) proposed the use of factorial experiments as a part of the Multiphase Optimization Strategy (MOST) for multi-component interventions. Traditional factorial designs can be used to assess the effects of each individual intervention component and key interactions of interest, enabling researchers to choose, based on empirical evidence, which intervention components to include in an intervention package and at what dose. However, traditional factorial designs do not allow the determination of times when it is most effective to deliver each intervention option. Nor do these designs allow researchers to investigate what time-varying factors moderate the relative effect of different time-varying intervention components. These are key questions for JITAI development. Micro-randomized trials overcome these limitations of traditional factorial designs, and for JITAI development, they can be incorporated into MOST as an alternative experimental design in the early stages of intervention development.