Walking is a low-cost, widely accepted physical activity (PA) associated with significant health benefits [1,2]. However, a meta-analysis by Conn et al [3] found that PA interventions showed improvements of only 14.7 minutes/week (about 2 minutes/day), suggesting more potent interventions are needed. Goal setting is a key component of behavioral theories [4,5] and PA interventions [6]. Though step goals are a key modality to increase PA, differing implementation strategies impede definitive conclusions [7]. Also, inconsistencies in reward structures for goal attainment further impair best practices for behavior change. Principles of operant learning and behavioral economics have the potential to concurrently refine goal setting and incentive strategies in the behavioral sciences [8].

Goal setting approaches are often fixed over time, though typically vary from researcher-assigned [9-12] to participant-selected [7,13,14]. If goals include participant input, it is often to set a starting goal from a baseline value [7]. A review of goal setting approaches for weight loss using PA and diet interventions found mixed results, noting the presence of extensive methodological issues and confounders in most studies [15]. Traditionally, fixed or static goals are the same across individuals (eg, walk 10,000 steps or exercise 30 minutes daily), affording the investigator a simple but insensitive threshold to promote behavior change. A major limitation of static goals is the inherent inflexibility to accommodate the myriad influences on day-to-day behavior (eg, illness, major life events, daily, or other cyclical schedules [16]).

Adaptive goals that adjust frequently and uniquely to an individual’s recent performance may be a more realistic approach to developing flexible yet challenging and attainable goals, but the task remains to standardize treatment dose across participants. Recently, intensively adaptive interventions have gained attention [16,17]. Intensively adaptive interventions require intensive (eg, daily) repeated measures of the tailoring variable(s) and target behavior. Concepts from percentile shaping may be one solution to standardize treatment dose [16,18] and through technology can be coupled with intensive repeated measures to produce intensively adaptive interventions [16,17].

Percentile shaping uses a moving window of recent performance (eg, last 9 observations or days) and a rank-order percentile algorithm to produce adaptive goals that can adjust systematically up or down daily, both within and between individuals, and over time. Percentile shaping capitalizes on the natural variation in behavior to produce personalized goals. Percentile shaping also generates inherently specific, measurable goals that can be explicitly rewarded. Only a handful of studies have tested the use of a percentile shaping approach by providing adapting goals to increase physical activity, and none have orthogonally compared goals derived from percentile schedules with immediate versus delayed reinforcement [16,19,20].

Rewarding small changes in behavior over time is important; however, types and dimensions (eg, latency, schedule) of reinforcement for goal attainment vary widely across interventions [15,16,21-23]. Principles of operant psychology [8] and behavioral economics [24] posit a temporal inconsistency in reward structures for healthy versus unhealthy behaviors. Specifically, “less healthy” behaviors tend to deliver an immediate reward (eg, physical comfort for sedentary behavior), with deleterious effects in the future (eg, lowered physical fitness contributing to heart disease). However, “more healthy” behaviors typically require an immediate cost (eg, physical discomfort for intense exercise) and require a sustained effort to experience delayed rewards (eg, improved physical fitness). As the benefits of healthier behaviors are inherently delayed, the immediate benefit of the less healthy response often wins in the economics of behavior. Interventions that attempt to tip the balance of reward in favor of healthy behaviors through immediate rewards are likely to produce more favorable and longer-term behavior change [16,24-26].

The purpose is to present the rationale and methods from the Walking Intervention Through Texting (WalkIT) trial—a 4-month, 2x2 factorial randomized controlled trial (RCT) for inactive, overweight, and obese adults. We used a semi-automated text message system to deliver adaptive versus static goals and immediate versus delayed reinforcement. The primary aim was to evaluate whether adaptive goals and immediate reinforcement resulted in a greater change in objectively measured PA compared to the static intervention and delayed reinforcement groups. Daily step counts were measured by a Fitbit device over the course of the 4-month study to evaluate the primary aim. We hypothesized that participants in the adaptive goals and immediate reinforcement groups would increase their average steps/day more than participants in the static goals or delayed reinforcement groups. Secondary aims were to evaluate the effectiveness of the adaptive goal and immediate reinforcement interventions to improve psychological measures, aerobic fitness, and cardiometabolic risk factors.

The WalkIT trial was a 2x2 factorial RCT conducted over 4 months. Following a 10-day baseline phase to assess usual PA levels measured by Fitbit Zip accelerometers, participants underwent simple randomization using a computerized random number generator for assignment into one of four treatment groups. Main effects of the treatment included Goal Type (adaptive vs static goals) and Reinforcement Type (immediate vs delayed reinforcement). In brief, adaptive goals and immediate reinforcement were based on a percentile-shaping algorithm that adjusted each participant’s goal up and down daily based on their previous nine valid observations (usually the last 9 days) of Fitbit-measured steps. Static goals were set to the recommended 10,000 steps per day and did not change over the course of the study. Participants in the immediate reinforcement group received praise feedback and one reward point each time they met a daily goal, whereas those in the delayed reinforcement group received monthly incentives. All participants received a walking intervention with semi-automated text message–based components. Researchers monitored the text message system for non-standard messages from participants (eg, when a participant asked a question, the research staff was notified), and staff responded through the system.

The Arizona State University Institutional Review Board approved the intervention trial and all the procedures used in data collection. The study is registered as a clinical trial (NCT02053259). See Multimedia Appendix 1 for the CONSORT EHEALTH checklist [27].

Participants were generally healthy, inactive, 18-60 years old, with a body mass index (BMI) of 25-55 kg/m² (see Table 1 for complete list of inclusion and exclusion criteria). Initially, criteria were 18-45 years and BMI 25-45 kg/m², but these were increased due to interest and to meet recruitment goals. Attempts were made to contact those previously ineligible who met the expanded criteria. The criteria for determining an inactive participant was defined in two ways. First, physical activity level was screened online with the International Physical Activity Questionnaire (IPAQ) short form. We addressed the limitations of using the IPAQ (eg, recall and social desirability bias) as a pre-screening method by inviting those with ≤1300 metabolic equivalents (MET)-min/week to make an appointment for the office visit. Then, in the 10-day baseline phase prior to randomization, we provided participants with a masked Fitbit (to limit participant reactivity) and monitored wear and steps/day remotely. Inactive participants, defined as those who did not accumulate ≥10,000 steps/day on ≥5 days/week, were randomized to one treatment group. Participants were required to have basic computer literacy, daily access to a personal computer for study-related software, a mobile phone with short message service (SMS, or text) capabilities, and be willing to receive up to 3-5 text messages per day. These criteria were imperative to receiving the mobile health (mHealth) related intervention components.

Recruitment emails and paper fliers included a brief study overview, notice of compensation for participating, and instructions on how to receive more information and begin the screening process. Local businesses, government agencies, social networking groups, retail outlets, and university departments were contacted to send the email notice and some elected to post physical fliers for their employees or patrons. Focused recruitment of minority populations was conducted through a free online social group advertisement.

Interested participants were directed to a secure online survey system (Qualtrics, LLC) for a pre-screening step, where they found a brief description of the study and completed the eligibility survey. Those determined to be eligible at pre-screening were contacted via phone and email for a telephone follow-up screening. Written (via online survey check box) and verbal informed consent (via the phone) were obtained at the initiation of each screening, respectively. Over the phone, the study was described in detail to participants, who were then offered opportunities to ask questions about participating and asked to clarify responses from their pre-screening responses to further assess eligibility. Qualified individuals were invited to schedule an appointment at the research office to review the study, provide written informed consent, complete baseline measures, and participate in accelerometer training.

Participants were required to reside in Maricopa County, Arizona, and to agree to make two visits to the research office located in Phoenix, Arizona, for pre- and post-intervention measurements. Rolling recruitment occurred February-August 2014 with data collection completed in December 2014. Weather was anticipated to be an influential confounder as the study occurred chiefly in the warmer months and over a monsoon season. Phoenix has a subtropical dry arid desert climate at low latitude (Köppen climate BWh). Wide variation in seasonal temperatures (eg, average high temperatures: July 41.2°C/106.2°F, December 18.9°C/66.0°F), along with monsoons (which include dust storms and flash floods), may drastically limit outdoor activity on very hot or hazardous days.

The 2013 median annual household income in Maricopa County was US $53,596 [28]. Maricopa County’s majority is white, non-Hispanic (58%), with those identifying as Hispanic or Latino (30%), black (8%), and Asian (4%) being the largest racial/ethnic minority groups [28]. Statewide, 93% of households have access to at least one vehicle [28] and since 2/3 of the state population lives in the recruitment area (ie, Maricopa County), this indicates heavy reliance on car travel within our study area.

Figure 1 shows how intervention groups varied by goal type (ie, adaptive vs static goals) and reinforcement type (ie, immediate vs delayed reinforcement) in a 2x2 design. This design allowed all participants to receive a form of the treatment in an effort to reduce attrition and improve compliance over the duration of the study. During the intervention phase, all participants were asked to self-report steps nightly (ie, before midnight) via the study’s interactive text messaging system that acknowledged their report and provided differential feedback based on group status and performance. The factorial design efficiently tested how each component influenced PA without the use of a “no-treatment” control group. All participants were told their ultimate target behavior was 10,000 steps on ≥5 days per week and that daily goals sent via SMS were good for 1 day only.

Eligible participants visited the laboratory twice for about 2 hours each time. The initial visit included the written informed consent, physical activity PAR-Q+, the pre-measures as listed in Table 2, and training on the Fitbit and the texting system. At the second visit, the participant returned the Fitbit, completed post-measures (similar to pre-, except where noted in Table 2 [36-49]), and was debriefed regarding the study purpose. Data collection staff were blinded to treatment allocation at pre- and post-measures. However, it was impossible to blind participants as treatment groups were explained in the consent process.

This study integrates measures of behavior change and physiological outcomes to evaluate intervention strategies and mechanisms that improve health through adoption of walking behaviors over 4 months in an inactive, overweight/obese adult sample. The study examines the effects of two experimental factors: (1) percentile shaping to produce performance-based adaptive goals, versus typical static goals of 10,000 steps/day, and (2) reward structure using principles of behavioral economics (ie, US $1 per daily goal achieved, obtained immediately as goal achievement is reported), versus a delayed incentive. The group with a combination of static goals and delayed reinforcement approximates procedures found in practical settings (eg, a physician offering a PA brochure, recommending 10,000 steps/day, and giving a pedometer to a patient) with the difference being a predetermined monthly reward for continuing with the study—a common approach in many research studies [16,20,51,52].

Our factorial study design allows examination of the independent and joint effects of these components and explores the promise of percentile shaping and small immediate rewards to optimize behavioral interventions. Our work will contribute to the field by testing specific methodologies that link behavior change theory to practical applications. The limited body of research on shaping to improve PA shows complementary results, even with differing methodological approaches [16,19,20]. Further, employing mHealth strategies, such as SMS for treatment delivery and reward mechanisms, and wearable activity monitoring with feedback and wireless upload, enhance the overall treatment and theoretical fidelity [53] while capitalizing on the omnipresence of mobile technology.

Potential limitations of this study include limited generalizability due to convenience sampling, although random allocation to the treatment group improves internal validity and reduces selection bias. Inclusion criteria may also limit generalizability as only generally healthy persons with a BMI classification as overweight or obese were included. Further limitations include a 4-month intervention length, which may not be long enough for some individuals to fully adopt successful walking routines. Also, without a post-intervention period follow-up, we will not be able to determine behavioral maintenance.

Strengths include the relatively large sample size, especially considering the extensive laboratory visits (approximately 2 hours each). The intensive repeated measures design is important for monitoring PA behavior to provide continual performance-based feedback via percentile shaping. We also included a large number of pre-menopausal women in the physiological measures, which is important due to underrepresentation in studies that limit inclusion to men and post-menopausal women when assessments involve biomarkers such as biochemical assays and arterial stiffness. Increasing time spent in PA is independently beneficial to health [54] especially for inactive populations and regardless of weight status [55-57]. We aim to better elucidate the link between behavior change and mediators of physiological and cardiometabolic health markers.

The Walking Interventions Through Texting (WalkIT) trial will further the understanding of theory-based intervention components to increase the PA of men and women who are healthy, insufficiently active, and are overweight or obese. With the overwhelming number of options interventionists have to use in health promotion, it is useful to look mechanistically at specific intervention components to optimize the treatment with economical, scalable mHealth methods. Though many studies have investigated walking interventions through a variety of methods, WalkIT is among the first directing the focus to the individual components of combined goal setting and reward structures in a factorial design to increase walking. The WalkIT trial is expected to produce results useful to future research interventions and perhaps industry initiatives, primarily focused on mHealth, goal setting, and those looking to promote behavior change through performance-based incentives.