Mobile and Wireless Technologies in Health Behavior and the Potential for Intensively Adaptive Interventions

Abstract

Recent advances in mobile and wireless technologies have made real-time assessments of health behaviors and their influences possible with minimal respondent burden. These tech-enabled real-time assessments provide the basis for intensively adaptive interventions (IAIs). Evidence of such studies that adjust interventions based on real-time inputs is beginning to emerge. Although IAIs are promising, the development of intensively adaptive algorithms generate new research questions, and the intensive longitudinal data produced by IAIs require new methodologies and analytic approaches. Research considerations and future directions for IAIs in health behavior research are provided.

Introduction

Assessments of real-time and real-world health behaviors and their influences are rooted in various research traditions including ambulatory monitoring, diary studies, and prompted self-reports [1]. The repeated sampling of behaviors and experiences in real-time, Ecological Momentary Assessment (EMA), has developed into a mature methodology that provides important findings not possible from retrospective reports [2*]. Recent advances in mobile/wireless technologies and the integration of these data with data from other digital sources have enabled unprecedented capabilities to intensively assess health behaviors “in the wild.” Ninety percent of U.S. adults own a cell phone, and nearly 60% own a smartphone (Pew Research Center; URL: http://www.pewinternet.org/fact-sheets/mobile-technology-fact-sheet), providing a ubiquitous personal device to deliver EMA as well as to sense location, activity, and other relevant health behavior variables.

Passive wireless sensor capabilities also are developing rapidly, both in the scientific and commercial sectors. Accelerometry for physical activity assessment has been the most established sensor technology for research purposes [3], but sensors are now available to assess environmental exposure, location, physical activity, sleep, social interactions, and images of the person and environment [4]. Wireless sensors for physiological parameters such as blood pressure, heart rate, and respiration rate are sufficiently high quality for approval by the U.S. Food and Drug Administration for use in hospital settings [5]. Recent advances in nanosensing and biosensors have led to the development of implantable biosensors [6] and nanosensors nested in a gum base to monitor biological activity [7].

A recent development is the integration of several physiological and behavioral sensors into a “smartwatch” that it is conveniently wearable for long durations [8]. Although the quality of measurement remains unknown, these smartwatches include activity monitoring, heart rate, and blood pressure with future plans to add glucose, blood pressure and other physiological parameters. The quality of these commercial measures remains unclear, and validation with standardized measures needs to be actively pursued by the research community.

These mobile and wireless health (mHealth) technologies have developed at an exponential pace in recent years, but the integration and translation of these cutting-edge technologies into rigorously evaluated health research and healthcare tools have not kept pace. Remarkable technological advances have been made in the last decade, and with sufficient research and evaluation, these technologies provide the potential to advance research, prevent disease, enhance diagnostics, improve treatment, reduce health disparities, increase access to health services, and lower healthcare costs in ways previously unimaginable (see Figure 1).

Figure 1.

Continuum of mHealth tools

Leveraging Real-Time Assessment for Intensively Adaptive Interventions

The rapidly expanding repertoire of real-time inputs relevant to health behaviors provides the potential for intensively adaptive interventions (IAIs). Earlier technological advances such as Internet-delivered interventions were critical to the development and implementation of tailored interventions in which baseline information is used to match the intervention to the individual [9]. Adaptive interventions adjust the intervention not only at baseline but also at various points throughout the intervention process [10*]. The Sequential Multiple-Assignment Randomized Trial (SMART) compares the decision points of adaptive interventions in which the next course of intervention is based on the prior intervention response [11].

These adaptive interventions typically make intervention adjustments after a few weeks or months of intervention. In contrast, IAIs make these adjustments every few days, hours, minutes, or even seconds utilizing real-time inputs. An IAI adjusts the timing and content of the intervention at any time point based on the response to previous intervention outputs as well as intrapersonal state and social/environmental context. IAIs have been made possible by the rapid proliferation of mobile and wireless technologies, which allow real-time and intensive assessment as well as intervention delivery throughout the day.

IAIs have also been described as Just-in-Time Adaptive Interventions (JITAIs) and Ecological Momentary Interventions (EMIs). The term “JITAI” has computer science origins and denotes that these IAIs are not delivered in “real-time”, i.e., immediately in response to the input, but instead “just-in-time” after a meaningful series of inputs that can be obtained to select an intervention with appropriate content for the appropriate time and place [12]. The term EMI derives from EMA and denotes that mobile devices can be used not only to prospectively obtain self-reports throughout the day, but also to deliver interventions based on these self-reports and other mobile inputs [13*,14]. The review of 27 EMIs [13], however, included very few interventions that were intensively adaptive [13], a finding consistent with our own review of mobile interventions [15*]. Therefore, although we believe that JITAIs and EMIs have similar connotations, we have used IAI to clearly describe an intervention that adapts rapidly (within a day or two at most, but typically within hours) in response to real-time inputs from sensors and EMAs. Using the first author's prior work for illustration, a text message smoking cessation intervention included in the Heron and Smyth EMI review [13] that adjusts text content and frequency based on time of day and quit date [16] would not be considered an IAI because the adjustments were not based on real-time inputs. In contrast, a special purpose device for scheduled gradual reduction of smoking that adjusted the interval to the next prompted cigarette based on when the user smoked in response to the previous prompt [17] would be considered an IAI despite its dated technology and algorithms.

Studies of IAIs for health behavior interventions are beginning to appear in the literature. In a small study, Adams and colleagues evaluated an adaptive intervention that generated personalized daily physical activity (step) goals and micro-incentives adjusted based on previous goal attainment [18*]. Adaptive intervention participants increased their average steps per day (2728 steps/day increase) significantly more than those in a static goal condition (1598 steps/day increase) over the six-month treatment period.

Hermens and colleagues have utilized electromyogram sensors in a wearable garment to assess upper back muscle tension in chronic pain patients and provide adaptive feedback (personalized coaching system, PCS) via the patient's smartphone when insufficient muscle relaxation occurs [19]. Evaluations of interventions using this type of PCS showed an improvement in musculoskeletal pain, but no significant differences between the PCS group and those who received traditional care [20-22].

Future Research in IAI

IAIs hold considerable promise to improve behavioral interventions. A compelling argument can be made that dynamic and intensively adapting interventions delivered frequently throughout the day in the context of where the behavior occurs and in response to a wide array of inputs on current context, intrapersonal states, and responses to prior intervention outputs should be more effective than static, non-adapting interventions, but this remains an empirical question. Tailored interventions, for example, appear to be less effective over the long-term or for less complex behaviors [23,24]. Therefore, although there is considerable interest in IAIs, research is needed to determine if, for whom, and in what context IAIs produce better outcomes than non-adaptive interventions.

IAIs generate new research questions not previously considered in non-adaptive interventions, but also provide a platform to answer these research questions. For example, we have noted in previous publications that current health behavior theories are ill-equipped for guiding IAI interventions, but that IAIs also provide a novel platform for developing and testing new theories and conceptual models for IAI development [15,25].

An important set of new IAI research questions involves the Goldilocks dilemma (too much, too little, or just right) of IAI inputs and outputs. How frequently do inputs need to be obtained for the IAI to effectively adapt? Too many EMA prompts throughout the day could result in missing data and premature termination of the intervention, but too few EMA prompts could obscure important temporal patterns that could influence the intervention or result in missed opportunities to deliver an intervention in a relevant context. For sensors capable of collecting data nearly continuously, what are meaningful segments for the purposes of adapting the intervention?

Regarding IAI outputs, mHealth intervention developers essentially guess at how often an individual should be prompted or intervened upon, attempting to achieve an adequate intervention intensity to produce the desired outcome without being overly intrusive and risking treatment dropout. The effective range of frequency and duration of these intervention outputs each day likely varies between individuals as well as within individuals over time, but to date, there is minimal empirical guidance on how to set these parameters, either initially or adaptively over the course of treatment. Tailoring intensity based on user preferences at baseline is an inadequate method for setting intervention intensity since users may prefer an intensity too low to be effective, or too high without first experiencing how intervention intensity affects their daily routine [26]. Research on the timing and intensity of mobile intensively adaptive health interventions is clearly needed.

Other important empirical questions related to the timing of IAIs involve reactive versus preemptive interventions. Thus far, IAIs have used reactive algorithms that determine the timing and content of the intervention outputs based on prior inputs, but preemptive interventions may be preferable for some health behavior problems. For example, falls can be accurately detected via passive sensor technologies, either those worn on the individual or installed in the home [27], but such technology is of much greater value if it can predict and preempt falls before they occur [28]. Model predictive control strategies may be particularly appropriate for the computational modeling of preemptive interventions [29].

IAIs generate not only new research questions but also the need for new research methods to appropriately study IAIs. The intensive longitudinal data generated by IAIs require approaches that are capable of modeling patterns and variability over time such as Time-Varying Effect Models [30*] and Mixed-Effect Location Scale Models [31]. IAIs are not only multi-component but also multi-contextual and multi-timed; therefore, randomized controlled trials (RCTs) comparing adaptive versus static interventions are necessary to establish efficacy but inadequate to optimize IAIs. Traditional RCTs cannot keep pace with the development of new technologies that can improve IAIs [32], and RCTs of multi-component interventions provide little guidance for developing an optimal health behavior intervention [33*]. There are a number of relatively new methods such as Multiphase Optimization Research Trials (MOST) [34] that isolate components and their sequencing to optimize multi-component interventions. Recent extensions of optimization trials such as micro-randomized trials [35] may be particularly well suited for evaluating the components of IAIs delivered via mobile/wireless platforms. Computational modeling simulations also have the potential to provide guidance on when and in which contexts IAIs should intervene. For example, Savage and colleagues used a control systems engineering approach to simulate an adaptive intervention to obtain optimal prenatal weight gain [36*].

Conclusion and Summary

IAIs offer considerable promise to leverage rapidly evolving technologies to improve health and change behavior. The capabilities to assess factors that might influence intervention effects at any given time based on real-time inputs obtained from integrated mobile/wireless devices provide enormous possibilities for optimizing behavioral interventions. Despite this promise, IAIs remain largely untested. While many mobile interventions characterized as EMIs have been evaluated, very few intensively adapt their interventions based on real-time data input. Newer methods and analytics provide the tools to evaluate IAIs in a more iterative and component-based manner that will, in turn, generate a more cumulative science of IAIs to better explain, predict, and promote behavior change.

Highlights.

Highlights for “Mobile and Wireless Technologies in Health Behavior and the Potential for Intensively Adaptive Interventions”

• Advances in technology provide opportunities for intensively adaptive interventions (IAIs).

• Despite considerable interest, IAIs are largely untested.

• IAIs generate new research questions like the optimal intensity of intervention inputs/outputs.