The Methodology Center Tech Report No. 15-131 Pragmatic Framework 1

The Methodology Center Tech Report No. 15-131 Pragmatic Framework 1

Building health behavior models to guide the development of just-in-time adaptive interventions: A pragmatic framework

Inbal Nahum-Shani

The University of Michigan

Eric B. Hekler

Arizona State University

Donna Spruijt-Metz

The University of Southern California

Technical Report Number 15-131

Copyright 2015, The Pennsylvania State University

ALL RIGHTS RESERVED

Please send questions and comments to Inbal Nahum-Shani,.

The suggested citation for this technical report is

Nahum-Shani, I., Hekler, E. B., & Spruijt-Metz, D. (2015). Building health behavior models to guide the development of just-in-time adaptive interventions: A pragmatic framework (Technical Report No. 15-131). University Park, PA: The Methodology Center, Penn State.

This work was supported by Awards P50DA010075, R01 AA-014851, and U54 EB020404the National Institutes of Health. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Abstract

Advances in wireless devices and mobile technology offer many opportunities for deliveringjust-in-time adaptive interventions (JITAIs)—suites of interventions that adapt over time to an individual’s changing status and circumstanceswith the goal to address the individual’s need for support, whenever this need arises.A major challenge confronting behavioral scientists aiming to develop a JITAI concerns the selection and integration of existing empirical, theoretical and practical evidence into a scientific model that can inform the construction of a JITAI and help identify scientific gaps. The purpose of this paper is to establish a pragmatic framework that can be used to organize existing empirical and theoretical evidence into a useful model for JITAI construction. This framework involves clarifying the conceptual purposeof a JITAI, namely the provision of just-in-time (JIT) support viaadaptation, as well as describing the components of a JITAI and articulating a list of concrete questions to guide the establishment of a useful model for JITAI construction. The proposed framework includes an organizing scheme for translating the relatively static scientific models underlying many health behavior interventions into a more dynamic model that better incorporates the element of time. This framework will help to guide the next generation of empirical work to support the creation of effective JITAIs.

Introduction

Advances in wireless devices and mobile technology offer many opportunities for delivering interventionsat any time,and in a way that accommodatesanindividual’s immediate needs(Riley et al., 2011). The term “Just-In-Time Adaptive Interventions” (JITAIs) (Spruijt-Metz & Nilsen, 2014)is used to describe a suite of interventions that adapt over time to an individual’s changing status and circumstances, with the goal to address the individual’s need for support, whenever this need arises.Recent advances in mobile technology and wearable sensors make these interventions increasingly more feasible and acceptable. For example, mHealth interventions attempting to provide timelysupport are being developed and evaluated for a wide range of health issues and behaviors, such asphysical activity (King et al. 2013; Consolvo et al. 2008), drug abuse (Dennis, Scott, Funk, & Nicholson, 2014), alcohol use (Witkiewitz et al., 2014; Gustafson et al., 2014), smoking cessation (Riley, Obermayer, & Jean-Mary, 2008), obesity/weight management (Patrick et al., 2009), and mental illnesses (Ben-Zeev et al., 2014).

Despite the increased use and appealing nature of JITAIs, a major gap exists between the technological capacity to deliver JITAIs and existing health behavior models. Establishing a scientific model is an important step in constructing behavioral interventions (Collins, Murphy, & Bierman, 2004). Most behavioral interventions are developed based on scientificmodels that articulate key risk and protective factors that are associated with the targetedhealth outcome. These models are often used to construct interventionstoaddressthese key risk and protective factors.However, most existing models largely emphasize and articulate static relationships, focusing on risk and protective factors that change relatively slowly over time, such as demographic factors, psychiatricdiagnoses, and past high-risk behaviors(Spruijt-Metz, Nilsen, & Pavel, 2014). JITAIs, on the other hand,provide support whenever such support is needed, seeking to address risk and protective factors that are dynamic and likely to change(often rapidly) over time, such as mood, location, social interactions and immediate crises in everyday life (Csikszentmihalyi & Rathunde, 1993).

A major challenge confronting behavioral scientists aiming to develop a JITAI concerns the selection and integration of existing evidence into ascientific model that can inform the construction of a JITAI. While the need to develop more dynamic health behavior models has been well-established (Riley et al., 2011), the current literature provides little guidance on the structure and predictions needed from these models to scientifically inform the development of efficacious JITAIs. The choice ofscientific modelsto inform the development of a JITAI should be guided partiallyby the requirements of the JITAI itself, such as providing insights not only on how to intervene, but also when and when notto intervene.

The aim of this paper is to establish a pragmatic framework that can be used to organize existing and new empirical and theoretical evidence into a useful model for JITAI construction. The foundation for this framework is established by clarifying the conceptual purpose of a JITAI, namely the provision of just-in-time (JIT) support via adaptation.After briefly reviewing the key elements in operationalizing JITAIs (Nahum-Shani et al. 2014), we offer a list of concrete questions to guide the process of establishing a useful model for JITAI construction. We conclude by discussing opportunities for future research that can advance the science of JITAIs.The hope is that the proposedframework will help guide the next generation of empirical work to support the creation of effective JITAIs.Table 1 summarizes key terms and definitions.

What is a JITAI?

To clarify the conceptual purpose of JITAIs, we elaborate on the two key concepts that distinguish these interventions from standard intervention designs:just-in-time and adaptive.

Just-In-Time (JIT). The concept JIThas long traditions in various fields. For example, in industrial management and operation research JIT is a philosophy of manufacturing that seeks to “produce the right item, at the right time, in the right quantities” (Canel, Rosen & Anderson, 2000, pp.52). It is based on a management plan that emphasizes continuous improvement and identifies and then eliminates all“waste”— defined as anything that does not add value to the product. In the field of education, the term JIT is rooted in instructional approaches that focus on real-life tasks as the driving force for learning. Because these tasks and the real-life context in which they are performed involve high cognitive load, these approaches emphasize the need to take the limited human-processing capacity into account. Hence, strategies for scaffolding include JIT support, meaning providing the type of support needed, precisely when needed, and only when needed during task performance (see van Merrienboer, Kirschner, & Kester, 2003 for review). Overall, the traditions above, as well as others (e.g., Drews et al., 2007; Frazier, Spekman, & O’Neal, 1988; Karolak& Karolak, 1995), conceptualize JIT as the effective provision of timely support, operationalized by offering the type of support needed, precisely when needed, in a way that minimizeswaste (i.e., defined as anything that does not benefit the person) and accommodates the real-life setting in which support is needed.

We build on the traditions above to suggest that in the context of health behavior interventions,the JIT approachis primarily motivated by the need to effectively assist people whenever they are vulnerable and/or whenever opportunities for positive changesarise (Ben-Zeev et al., 2014; King et al. 2013). Given that vulnerability and opportunity can occur anytime in everyday life (Fletcher, Tobias, & Wisher, 2007; Witkiewitz & Marlatt, 2004), JIT support in this setting can be operationalized by (a) identifying states of vulnerability or opportunity for progress and providing the type of support needed in such states, precisely and only when needed; as well as by (b) ensuring that the person is in a state of receptivity; that is, in a state where s/he can receive, process and use the type of support needed.

State of vulnerability/opportunity.Stress-vulnerability and coping theories (Zubin & Spring, 1977, Lazarus, 1993, Witkiewitz & Marlatt, 2004) conceptualize a vulnerable state as the person’s transient tendency to experience adverse health outcomes or to engage in maladaptive behaviors. A vulnerable state is a function of the interplay between relatively stable factors (e.g., personality traits, socio-econmic status, air polution) and more dynamic situational factorsranging from relatively rare life events (e.g., unemployement), to more transient experiences (e.g., conflict with a coworkers). Here, JIT support can be operationalized by identifying times in which the person is vulnerable and providing the type of support needed, only when needed, in order to break the link between the vulnerable state and adverse health outcomes.

With regard to opportunities for positive change,various learning and motivational theories highlight the importance of concepts such as shaping (i.e., training by reinforcing successively improving approximations of adesired behavior: Bouton, 2007; Ferster & Skinner, 1957) and teachable moments (i.e., a time when a person is more likely to internalize information and takeaction;Fisher, Piazza, & Roane, 2011; Murimi et al., 2014; Leist & Kristofco, 1990). The underlying assumption is that in order to facilitate improvement in some behavioral or cognitive domain, it is important to identify transient oportunities for learning and improvement and provide the type of support needed, only when needed in order to gradually move the person’s actions/cognitions in the desired direction. Here, JIT support can be operationalized by identifying real-life oppotunitites for changeand immediately providing the type of support needed to capitalize on these oportunities, only when such oportunities arise.

State of receptivity.To further minimize waste and accommodate the real-life setting in which support is often needed, it is critical to ensure that the person is in a state where s/he is receptiveto the support needed. Integrating research in the area of supportive communication and ubiquitous computing (e.g., Ford & Ellis, 1998; McIntosh, Seaton, & Jeffrey, 2007; Resnicow, Baranowski, Ahluwalia, & Braithwaite, 1998;Sarker et al., 2014), we define astate of receptivityas a restricted time interval in which the personcan receive, process,anduse the type of support needed. A variety of facets can impact receptivity,and existing work can help guide thinking about this concept.

The dual process theory of supportive communication outcomes (Burleson, 2009), provides a logical foundation for understanding receptivity, suggesting that this construct is a function of the interplay between two key elements. The first element-- the nature of support-- includes features such as the type of supportive content (which might be more or less demanding in terms of reflective processing, depending on facets such as structure, length and complexity), and the presence of cues (i.e., paraverbaland/or nonverbal aspects of support, that trigger heuristics, associations,or sensations relevant to the situation, and hence are less cognitively demanding: see also Evans, 2008; Castelo et al., 2012). The second element -- the recipient’s ability and/or motivation to process the support provided-- can be influenced by relatively stable characteristics (e.g., attachment style, locus of control, age,cognitive complexity, and working memory capacity) and more dynamic/situational factors (e.g., the severity of problem, timing, emotions, location and presence/absence of attention distracters) (Burleson, 2009).

To improve a person’s motivation and ability to process and use the type of support needed, when needed, research in human-computer interaction articulates ways to improve the overall usability and enjoyment of using mHealth and ubiquitous computing interventions.For example, Consolvo, McDonald, and Landay (2009) generated eight design guidelines (e.g., abstract and reflective, unobtrusive, possible to be used in public)thatprovide heuristics for improving receptivity to mHealth interventions by ensuring that support is designed to be conducive in the moment it is needed. Other research has focused on how balance between perceived usefulness/value and perceived burden influences one’s motivation to use the support provided (e.g., Or & Karsh, 2009; Polonsky, Fisher, Hessler, & Edelman, 2014), as well as on heuristics for simplifying supportive content in order to reduce burden (Fogg, 2009).

An important facet to consider with regard to receptivity is the ethics of intervening in a real-life setting (CapronSpruijt-Metz, 2014). Ethical considerations such as privacy, confidentiality, safety and the general welfare of the recipient might lead to the decision not to provide support even though it is needed (Kjeldskov, Skov, Als, & Høegh, 2004). For example, when the person is driving a car, it might not be safe to deliver support; when s/he is in a meeting, support can be disruptive; and when s/he is around other people, providing certain types of support (e.g., feedback) might jeopardize the person’s privacy (De Costa et al., 2010).

Beyond this, a variety of other theories and empirical evidence can help explain receptivity. Although a full discussion of these is beyond the scope of the current manuscript (see King, Currie, & Petersen, 2014; Yatchmenoff, 2005; Staudt, 2007; Naughton, Jamison, & Sutton, 2013), this line of research builds the foundation for future research aiming to identify times at which a person might be more or less receptive to specific types of support.

Adaptation.The discussion above suggests that in the context of health behavior interventions, JIT support can be operationalized by offering the right type of support only when the person is (a) vulnerable or open to positive changes, and (b) receptive to the support needed. This requires a strategy for adapting the type (or dose/modality) and timing of support.

The distinctions between targeted, tailored, personalized, and adaptive interventions are important ones, yet the terms are often used interchangeably in research literature. Moreover, the same terminology often captures different meanings in different fields[1]. To standardize the terminology, we use the term individualization to capture the use of information from the individual to make decisions about when, where and how to intervene. Additionally, we distinguish between individualization that is static, where relatively stable information from the person (e.g., gender, baseline severity of symptoms) is used to make intervention-related decisions (e.g., to offer intervention package A or B); and dynamic, where time-varying information from the person (e.g., changes in psychological distress, response to an intervention, intervention adherence) is used to make intervention decisions repeatedly in the course of the intervention (e.g., changing the type, dosage, or timing of intervention delivery). The term adaptive is used to describe this dynamicform of individualization (Collins et al., 2004).

Building on the above terminology, we conceptualize the JITAI as an intervention design that uses a dynamic form of individualization to operationalize theprovision of JIT support.Specifically, JITAIsoperationalize the individualization of the selection and delivery of intervention options based on ongoing assessments of the individual’s state and ecological context, with the goal to offer the right type of support precisely when, and only when, the person is in a state of vulnerability/opportunity and receptivity.

JITAIs become increasingly possible with the growingavailability of technology such as wearable and ubiquitous computing sensors (e.g., wearable activity monitors, smartwatches, home automation and tracking systems such as a smart thermometers), mobile-phone-based sensing (e.g., accelerometry, GPS, light sensors, microphones), digital footprints (e.g., social media interactions, email, digital calendars), and low-effort self-reporting (e.g., ecological momentary assessment [EMA] and more advanced low-burden opportunities available via technologies like smartwatches). The portability and pervasive nature of these devices make it possible to monitor the individual anytimeand to identify states of vulnerability/opportunity and receptivity at any given moment (Hekler, Klasnja, Traver, & Hendriks, 2013).

Elements of a JITAI

With the conceptual purpose of a JITAI established, we now turn to describing the elements of a JITAI, to help ground our pragmatic framework. A JITAI includes6 key elements: a distal outcome, proximal outcomes, decision points, intervention options, tailoring variables, and decision rules (Nahum-Shani et al., 2014).The distal outcome is the ultimate goal the JITAI is intended to achieve. The proximal outcomes are the short-term goals the intervention is intended to achieve, and are oftenmediators of the distal outcome. Decision points are times at which an intervention option is selected based on currently available information (e.g., at 2pm, every 3 minutes). Intervention options are the array of possible type/dose/timingof support thatmightbe employed at any given decision point. Tailoring variables are baseline and time-varying information that informs which intervention option to offer at each decision point (e.g., levels of urge, location, daily drinking). Finally, decision rules are used to operationalize the individualization by specifying which intervention option to offer to whom and when.

------

Figure 1

------

For example, consider JITAI #1 in Figure 1, where the decision rule is designed to reduce prolonged sitting (distal outcome) among office workers,by encouraging them to take active breaks (proximal outcome). There is a decision point every 5 minutes; the tailoring variable is the current bout of accumulated uninterrupted computer activity; the intervention options are either to recommend movement, or to provide nothing; and the decision rule links information from the individual (tailoring variable) to specific intervention options, by specifying 30 minutesas the cut-point of the tailoring variable that determines whether the individual should be offered either a recommendation or nothing.

To summarize, a JITAI is an intervention design that employs dynamic individualization (i.e., adaptation) to facilitate the provision of JIT support. To construct JITAIs, it is important to clearly understand and articulate the key elements, namely the distal outcome, the proximal outcome(s), decision point(s), tailoring variable(s), intervention options, and decision rule(s). However, such understanding is well-beyond current behavioral theories and empirical evidence.

The technology-science gap

Current health behavior theories and related empirical evidence paint a largely static picture of human behavior, cognition and emotions; they fail to capture the dynamic processes underlying the emergence of a vulnerable state or the adoption and maintenance of healthy behaviors (Spruijt-Metz et al., under review). Even dynamic models that acknowledge the role of episodic factors in health behavior processes (e.g., the dynamic model of relapse; Witkiewitz & Marlatt, 2004) do not specify the temporal nature of each factor in a way that informswhen and how to intervene.Although many health behavioral models acknowledge individual differences in response to treatment, in most cases these models can only inform the most basic form of individualization (i.e., they usesingle time point factors like age, gender, or baseline symptom severity to make intervention decisions) rather than the dynamic individualization required to operationalize JIT support. Finally, existing intervention models often adopt a one-size-fits-all approach to intervention provision (Drotar & Lemanek, 2001; Marcus et al., 2000; Sorensen, Emmons, Hunt, & Johnston, 1998), failing to provide actionable insights with regard to the various elements of individualization articulated above.A major step in building the theoretical and empirical foundation for creating effective JITAIs is to articulate a pragmatic framework guided by specific questions that must be answered to create a JITAI. These specific questions can help develop and refine existing theories and to inform prioritization offuture studies aiming to construct JITAIs. Below, we offer a set of pragmatic questions to guide the process of establishing a scientific model that can inform the construction of JITAIs.