Using virtual agents to increase physical activity in young children with the virtual fitness buddy ecosystem: Study protocol for a cluster randomized trial

Abstract

Background:

Designing and implementing a truly self-determined physical activity (PA) intervention has required excessive amounts of labor and expenses that, until recently, have made it prohibitively costly to implement in the field at scale.

Methods:

Guided by self-determination theory, and harnessing the power of consumer-grade interactive technologies, we developed the Virtual Fitness Buddy (VFB) Ecosystem. Designed to foster intrinsic motivation toward adopting PA as a lifestyle change in 6–10-year-old children, the Ecosystem features a mixed-reality kiosk which houses a personalized virtual pet for each user. Each time a child visits the kiosk, the pet (a mid-sized dog) automatically detects its owner based on the data from a child’s Fitbit, assists the child in setting daily PA goals and provides tailored feedback on the child’s PA progress. The pet alerts parents in real-time by sending text messages and relaying the parents’ response to the child, so that parents and children can remain connected about the child’s PA progress even when they are physically apart. We aim to implement the kiosk in 12 afterschool sites, plus use 12 additional sites as controls, where children can still set and view progress toward their PA goals without access to a virtual pet.

Conclusion:

The VFB Ecosystem represents a new generation of technology-mediated health interventions for children to promote sustainable PA lifestyle changes. Because the VFB Ecosystem is a cost- and labor-effective solution that integrates consumer-grade technology with low barriers for continued use, it has the potential for rapid diffusion and widespread public health impact.

1. Introduction

To encourage physical activity (PA) among younger children, especially those within underserved communities, a number of technology-mediated health interventions, such as exergames, have been developed to integrate game-like fun into children’s PA interventions [1–7]. However, these interventions have typically hinged on extrinsic rewards and have shown limited success in motivating children to be physically active in the long term after the extrinsic rewards cease [1,2,8,9]. To promote enduring PA behaviors in children by leveraging the power of intrinsic motivation for PA, recent research has been informed by theoretical tenets from self-determination theory (SDT) [10] to design interventions that are autonomous, personalized, and provide social support from parents and virtual agents [11].

SDT suggests that the key to altering individuals’ long-term behaviors lies in fostering intrinsic motivation, or the drive toward an activity because of its inherent pleasures and satisfactions. Three basic, universal, and cross-developmental psychological needs facilitate intrinsic motivation. These include autonomy, or the desire to make one’s own choices; competence, or the desire to feel capable and confident in engaging in an activity; and relatedness, or a desire to feel connected to others [10,12]. SDT logic suggests that health interventions for lasting PA behavior change in children should: (a) encourage children to set autonomous PA goals, (b) promote individually tailored PA competency, and (c) offer meaningful social support for adhering to the PA intervention.

Although simple in concept, the resource-intensive nature of designing and implementing a truly self-determined PA intervention requires the ability to reliably track individual users’ progress on autonomously-driven health goals, as well as a way to provide user-tailored feedback [13,14]. The labor and costs associated with all of the above, until recently, have made it prohibitively expensive to implement in the field at scale [15–17]. However, harnessing the power of consumer-grade interactive technologies can provide a resource-efficient solution to overcome many of the practical challenges associated with SDT-based PA interventions [11,18,19].

In line with SDT and previous research demonstrating the feasibility of this approach [11], we developed the Virtual Fitness Buddy (VFB) Ecosystem, a PA intervention that encourages children aged 6–10 to interact with a personalized virtual agent using a mixed-reality kiosk system. Guided by SDT’s mechanisms of change, the virtual agent, in the form of a pet, helps children set self-determined PA goals (autonomy), provides immediate, tailored feedback on those goals (competence), and offers positive reinforcement and social support when PA goals are met (relatedness). In addition, the virtual pet keeps parents and children connected when they are physically distant so that parents can provide social support to children and communicate about children’s PA goal progress together (further strengthening relatedness). By facilitating intrinsic motivation for PA, this intervention’s primary aim is to help children learn to embrace moderate-to-vigorous PA as a long-term lifestyle change while being supported by their parents and virtual pet within the Ecosystem. More specifically, we aim to test the Ecosystem’s efficacy as an effective, accessible tool for facilitating children’s intrinsic motivation toward PA as well as their long-term moderate-to-vigorous PA.

2. Methods

2.1. Overview

The VFB Ecosystem is a kiosk-based health intervention designed to foster children’s intrinsic motivation toward adopting PA as a long-term lifestyle change, rather than as a means to attain short-term extrinsic rewards (e.g., points). The kiosk features a mid-sized virtual pet (i.e., dog) that is programmed to recognize its owner based on the data from a child’s Fitbit, so that each child interacts with a unique, personalized virtual pet. Each time a child visits the kiosk, the pet automatically detects its owner, assists the child in setting daily PA goals, evaluates goal attainment, and provides accurate, detailed, and tailored feedback and encouragement on the child’s PA progress. In the meantime, the kiosk automatically records and uploads the PA data stored in each child’s Fitbit. When interacting with the child, the pet alerts parents in real-time by sending text messages, and relaying the parents’ response and encouragements to the child, so that parents and children can remain connected even when they are physically apart.

Children had the capability of engaging with the kiosk as much or as little as they wanted, engaging with the intervention in a largely self-determined and self-regulated way. The feasibility of this technology-based self-determination approach was tested in previous research, which demonstrated that 6–10 year old children in after-school programs (A) participated in a kiosk-based intervention, (B) wore their Fitbits for most days throughout the study period, and (C) set and met PA goals using the kiosk [11]. In the present study, kiosks were deployed for 6 months, but PA was tracked for a total of 12 months to assess the long-term impact of the intervention on PA.

Addressing limitations of previous PA interventions, the Ecosystem enabled the implementation of a long-term, truly self-determined PA intervention for young children by (a) reducing the need for in-person interactions with the researcher, (b) fostering children’s PA autonomy and competence each time they set and achieve a PA goal, and (c) encouraging a supportive relationship between children, parents, and their virtual pets.

2.2. Study design

We formed a partnership with a local metropolitan youth organization, the YMCA, to carry out the project within YMCA after-school programs. We chose to implement the intervention in an afterschool-setting due to its ability to provide a long-term, reliable, and secure location for children to access and engage with the intervention kiosk each day as much, or as little as they wanted. We aimed to implement the treatment in 12 after-school sites, plus use 12 additional sites as control. Implementation in the 24 sites is planned across three years, with 19 sites completed so far. An unblinded 2-arm, clustered randomized, match-paired, prospective design was used to investigate the impact of the VFB Ecosystem against a control condition that received a computer system that allowed children to set PA goals and receive feedback but without the virtual pet. The trajectories of PA in children residing in the Atlanta Metropolitan Area were observed. Blinding of researchers to the conditions was not possible given the visibility of the kiosks housing the VFB Ecosystem. However, children and parents were blinded to the alternative treatment. A clustered randomized design was necessary because it was not possible to randomize children at the individual level without contamination of outcomes from children receiving visibly different treatment assignments within sites. The afterschool sites were match-paired within cohorts based on the percentages of their students who are eligible for free lunches under the National School Lunch Program, sponsored by the United States Department of Agriculture. The percent of children eligible to receive free lunches is a widely used metric for a school’s average socioeconomic status, as it is a strong predictor of the percent of families in poverty [20–22]. All study procedures were approved by the university’s institutional review board.

2.3. Participant recruitment and eligibility criteria

Flyers and in-person information sessions were employed for recruitment. At each site, parents interested in participating completed an eligibility form to ensure they (a) had at least one child between the ages of 6–10 enrolled in the after-school program who (b) could participate in low to moderate PA without assistance and (c) attended the after-school program most days of each week throughout the school year. Parents were also asked to give researchers permission to view and record their child’s Fitbit data for the duration of the study.

2.4. Intervention components

2.4.1. Orientation session

Researchers hosted an intervention orientation at each site to enroll interested families into the study. During this session, researchers explained the study procedures, obtained informed consent from parents and assent from children, distributed and set up Fitbit devices, and obtained baseline measurements from parents and children.

2.4.2. Accelerometer

To obtain daily PA information from children, we used wrist-worn electronic activity trackers from Fitbit Inc. These devices estimate steps and active minutes (at 3 intensity levels) through the analysis of data from an internal 3-axis accelerometer. Periodically, the data are requested through Bluetooth by internet-connected devices, such as the end-user’s laptop or mobile phone, and then automatically uploaded to the Fitbit online platform, where it is then available for download by an authorized user or third-party application through the application-programming interface (API). The VFB accesses “intraday” data, which provides 1-min granularity to PA data, in addition to the default aggregated daily sum.

The Fitbit integrates with the VFB system through an intermediate server. The server contains the authorization keys for each Fitbit account and is thus able to serve as a proxy to access fitness data. This means that parents only need to authorize access to the Fitbit information once, and this access can be used at all kiosks and on the website. Once the server is authorized, any VFB kiosk can access the Fitbit data.

Parents also had access to their child’s PA data using the official Fitbit application available on their smartphone. Parents were encouraged to use the Fitbit application to check their child’s PA progress and upload their child’s Fitbit data to the VFB servers during the post-intervention months of the study (months 6–12).

2.4.3. Intervention delivery

The intervention was delivered by a kiosk-based system that varied in content depending on which condition participants were assigned to. In both conditions, the kiosk was an interactive, child-friendly computer that was freely accessible to children participating in the intervention at their after-school program. Participating children in both conditions could use the kiosk to set autonomous PA goals (satisfying SDT’s autonomy need), review their progress on these goals (satisfying SDT’s competence need), and receive outside social support from parents who were notified of this progress (satisfying SDT’s relatedness need). Participants interacting with the treatment kiosk were also afforded an additional avenue for immediate relatedness satisfaction through the addition of a (A) virtual pet and (B) parental text-messaging system, both of which were housed within treatment kiosk.

Both the treatment and control group kiosks included a wireless communication device (a small USB-connected dongle, provided with the Fitbit) for detecting in-range Fitbit devices (the detection range is approximately 90 ft), and for downloading data from Fitbit devices, which functioned well within 30 ft. The communication protocol for these devices is known (see https://bitbucket.org/benallard/galileo/), and based on this, a custom software application was created to identify participants and upload their encrypted Fitbit data to Fitbit servers. This enabled, from the perspective of the participant, a seamless interface to access their PA data during each visit, where they only needed to approach the kiosk for it to automatically detect their Fitbits and download data. As downloading a single participant’s data could take a few minutes, this process took place in the background, whenever the participant was in range, or when they approached the kiosk and selected their name from a list of detected participants.

2.4.4. VFB treatment kiosk

For the treatment group, the VFB kiosk consisted of a large-screen (50 in.) television monitor mounted above a cabinet with wheels (Fig. 1). The cabinet housed a desktop computer, a touch-screen display mounted on the front as the input interface, and a Microsoft Kinect sensor (Kinect for Xbox One) mounted below the touch-screen. Six of these Kiosks were custom-made for the project, and were designed to be accessible to children, with the touch-screen and Kinect sensor mounted lower to accommodate shorter heights.

Fig. 1.

Virtual fitness buddy kiosk.

2.4.4.1. VFB software.

The software for the VFB was created using the Unity 3D game engine. The virtual environment for the simulation was modeled as an outdoor dog park, where participants could visit and play with their pet, which was a mid-sized dog in this trial. These pets were created and customized by each participant. The customization consisted of choosing one of eight different breeds of dogs, a name, tag, and collar. This was done at the start of the intervention.

The virtual pet included several indicators of its health, tied to each user’s PA progress as measured by the child’s Fitbit device, and to virtual properties that moderated game performance, for example, the virtual pet’s response speed, which made it more adept at tricks based on the user’s moderate-vigorous PA level. As children engaged in PA and met more self-determined goals, the pet would become visually fitter, gradually appearing happier, as well as more muscular and toned. Other intervention objectives were tied to game access. For example, to encourage participants to set and meet PA goals, a virtual currency was implemented, and earned by meeting self-determined goals. The virtual currency could then be used to buy accessories for their pets and unlock games.

Another design objective was for the kiosk to be used frequently, but with a short duration (< 5 min) during each interaction. This helped resolve two potential issues of the Ecosystem: 1) A relatively large number of participants could access the kiosk each day, though not simultaneously, and 2) children’s daily screen-time exposure would be minimal. To achieve this, the virtual pet was given a daily amount of quantified “energy.” When children interacted with the virtual pet, the pet would consume this energy. Once fully consumed, the pet would no longer be able to play games with the child, and the interaction would automatically terminate. As further incentive to engage in PA, daily energy would increase as the participant’s average daily activity increased. Effectively, the more PA the child engaged in (taken as a daily average over the past week), the more energy the pet would have to play. Notably, even if the pet was out of energy, children were still able to set PA goals and review their progress on those goals as often as they wanted.

All interactions between the child and the pet were motion-based, in which children had to use full body gestures to play with their pets. This was enabled by the Microsoft Kinect sensor embedded in the kiosk to track the participant’s movements. Body motion was then translated to the motion of an on-screen avatar, which the participant could use to interact with the pet during games. Games included teaching the virtual pet a range of tricks (e.g. sit, roll over), or playing fetch with various items that could be purchased with credits (e.g., ball or frisbee). Children could also play basketball with their pet, and an “item-blaster” game, which was a time-based projectile game, where the pet retrieved items from popped balloons. The motion-based interactions with the pet were designed to realistically represent the human-pet relationships in the offline world while leveraging the creativity of the virtual world wherein the pet is able to play any game that the child desires.

2.4.4.2. Parent text-message system.

Parents were integrated into the Ecosystem through two mechanisms. First, parents would receive simple-messaging-service (SMS) notifications through their smartphone when certain events occurred. These included when goals were met or when the Fitbit had a low battery and needed to be charged. Each notification message also included a secure link to the participant’s data dashboard, which could be used to see PA progress. A second mechanism allowed parents to send messages to their child, such as a congratulatory message upon meeting a goal. Parents could do this at any time, and participants would receive the message the next time they accessed the kiosk as an on-screen pop-up.

2.4.5. Control kiosk

Control sites received a simpler kiosk setup consisting of only a laptop. As with the treatment-site participants, the control-site kiosks featured proprietary software that enabled participants to upload PA data from their Fitbits, see their active minutes and steps, and set goals. To the extent possible, these interfaces were identical to those of the treatment-site, except for the lack of the virtual pet and the ability to connect children with their parents.

2.4.6. Researcher involvement

The implementation of the kiosks at both the treatment- and control-sites was designed to be largely self-sufficient in that it only required that after-school site directors turn on the kiosks each day and let the children freely interact with it as they wished. Although this process was designed to require minimal researcher involvement, researchers made regular quality check site visits to (a) maximize study protocol compliance among after-school site directors and children, (b) troubleshoot technology issues arising with Fitbits and kiosks, and (c) address parent questions in person (as opposed to via email).

2.5. Measurements

PA from the Fitbit was measured continuously for one year, and intervention engagement from the kiosk was measured continuously for the 6 months during which the kiosk was deployed. At four points throughout the year (baseline, and 3-months, 6-months, and 9-months after the intervention’s start), researchers visited after-school sites to assess children’s body composition (height, weight, body fat percentage, and waist circumference), deploy ActiGraph activity monitors to assess PA for one week for cross-validation against the Fitbits, and administer self-report surveys to children in order to assess self-reported psychosocial variables related to PA. Parents completed self-report surveys at each time-point from home. For children ages 6–7, we reduced the number of survey items and implemented a previously validated pictorial version of the survey [23]. Items on the pictorial version verbally described two characters: one who exemplified the survey item, and one who did not. Children were asked to select the character who, and to what extent (two levels), was more like them. Survey items and choices were audio recorded for children who could not read the survey.

2.6. Primary outcomes

2.6.1. Moderate to vigorous PA

2.6.1.1. ActiGraph.

At each of the four measurement periods, up to 20 children at each after-school program were fitted with ActiGraph GT9X accelerometers attached to an elastic belt and positioned at the mid-axillary line of the right hip. If a site had less than 20 children, all children at the site willing to wear the device received one. Children were verbally instructed to wear the device during all waking hours, except for water-based activities, for 8 continuous days. Parents received detailed written instructions and received text message reminders during the wear-period and were asked to record the times the device was worn and reasons for non-wear.

ActiGraph data were collected in 10-s epochs to accurately classify the intensity of activities performed during short bursts, as is typical of children in the study’s age range. Vector magnitudes during each epoch were classified into intensity categories (i.e., light, moderate, vigorous) using the age-appropriate cut-points developed by Evenson and colleagues [24]. The duration of device wear was estimated using the Choi algorithm [25], where periods with consecutive zero vector magnitudes of 90-min or longer were classified as non-wear. Only days with a minimum of 10 h of wear were considered as valid estimates of daily activity and a minimum of four days (three weekdays and one weekend day) of valid data were required for inclusion in analyses [26].

2.6.1.2. Accelerometer.

Each child was given a Fitbit Flex (Cohort 1) or Fitbit Ace 2 (Cohort 2) activity tracker to wear daily for one year. Children were instructed to wear the waterproof Fitbit on their non-dominant wrist except for device charging and bathing activities. Daily Fitbit wear times were estimated using a modification of the Choi algorithm applied to the ActiGraph data, using consecutive zero step counts instead of vector magnitudes, to identify periods of probable non-wear. Activity metrics obtained for each minute of wear were steps and time spent in sedentary, lightly active, fairly active, and very active intensity categories. The activity categories are based on proprietary Fitbit algorithms but approximately correspond to light, moderate, and vigorous intensities. Step estimates from waist-worn Fitbit devices were strongly correlated (r = 0.85–0.96) with ActiGraph step estimates obtained during school hours in a sample of 9–10 year old students [27] but have also been reported to produce higher step estimates (161 steps/h) than the ActiGraph in similarly aged students, with larger step differences reported at higher activity intensities [28].

2.6.2. Moderators

Guided by the theoretical tenets of SDT, we assessed several critical moderators of the VFB kiosk’s ability to influence children’s PA. These moderators were focused on assessing children’s PA autonomy, competence, and relatedness using both event data measured by the kiosk, as well as self-reported measures [29].

2.6.3. Kiosk measures of intrinsic motivation

For both the treatment and control groups, we assessed children’s PA autonomy by examining the PA goals they set throughout the 6-month intervention period. Coupled with PA data garnered from their Fitbits, we assessed the number of PA goals children set and achieved throughout the intervention period as a measure of children’s PA competence. For children in the treatment group, we assessed relatedness according to their separate connectedness with parents and their virtual pets. To assess connectedness with parents, we summed the number of text messages exchanged between the parents and children using the kiosk. To assess connectedness with their virtual pets, we summed the number of minutes children spent visiting their virtual pets throughout the intervention period. For children in the control group, we assessed connectedness to parents using responses to self-report surveys (described in the next section).

2.6.4. Self-reported intrinsic motivation

Intrinsic motivation was assessed via adapted self-report items from previously validated scales assessing 8–10 year old children’s attraction to PA [30], PA self-efficacy [31,32], PA self-regulation [32,33], PA psychological need satisfaction [29], as well as PA support from peers [34], parents [35,36], and, in the treatment group, the virtual pet [34,37]. For children aged 6–7, we adapted items from the above instruments to assess children’s PA attraction, self-regulation, as well as support from peers, and, if in the treatment group, the virtual pet using a previously validated pictorial survey procedure [23]. As an additional measure of PA support from parents, we also included measures in parents’ self-report surveys to assess their perceived efforts to support their child’s PA [38].

2.7. Covariates

2.7.1. Child’s body composition

Children’s height, weight, percent body fat, and waist circumference were obtained at the baseline, 3-month, 6-month, and 9-month measurement periods. All measures were taken in light clothing and with shoes and socks removed. Height was measured to the nearest 0.1 cm using a Hopkins Road Rod portable stadiometer (Caledonia, MI). Body weight (to the nearest 0.1 kg) and percent body fat (to 0.1%) were measured using a Tanita BF-689 Children’s Body Fat Monitor (Arlington Heights, IL). Waist circumference was measured twice in line with the umbilicus to the nearest 0.1 cm using a constant tension tape measure. If the first two measurements differed by >0.5 cm, a third measurement was taken and the average of the two closest readings used for analysis.

2.7.2. Demographics

In parent’s self-report surveys at baseline, we also collected information about the parents’ age, gender, race/ethnicity, and socioeconomic status, as well as information about their child’s age, gender, and race/ethnicity.

2.8. Compensation schedule

Participating families were compensated up to $220 for completion of all study components. Compensation for surveys increased by $10 at each timepoint, beginning at $20 and ending at $50 for a total of $140. Participants also were able to earn $10 four times throughout the study simply for remaining enrolled, for a total of $40. Additionally, families of children who were selected to wear an ActiGraph activity monitor at each time point were awarded an additional $10 each time they returned the ActiGraph device, for a total of $40.

2.9. Analytic plan and power analysis

2.9.1. Sample size and power

Under our cluster randomized design, decisions regarding the number of YMCA sites, and the number of children within each site to be recruited must balance costs at the site and participant levels. For participants, costs include the Fitbit and ActiGraph devices used to measure PA, monetary compensation, and time it takes for the data management team to monitor compliance with study protocols and manage the resulting data for each participant. Costs at the site level include the kiosks at the treatment sites and laptops at the control sites. This equipment also requires physical space, and must be both powered and continuously connected to the internet, requiring coordination with information technology personnel at sites to ensure that data are transmitted to the study server.

However, the primary limitation on the number of sites that may be recruited beyond their availability is the personnel resources of the research team required to recruit, enroll, and inform parents during the launching of this project. Given the practical constraints of time, money, and personnel, a maximum of 12 sites per cohort, and a minimum of 20 participants per site were feasible targets [11]. Power to detect treatment effects in clustered randomized designs depends more on the numbers of groups than on the number of participants per group, so the reduction in the number of children per YMCA site from 50 in the original protocol will not have a large impact on power as confirmed in the following.

This study was powered for the primary outcome variable, the main effect of treatment on hours of moderate-to-vigorous intensity PA. Under match-pairing of groups, the standard error of the difference between the treatment means is a function of the variance components for subjects $({\sigma }_{\epsilon }^2)$, and sites within matched-pairs $({\sigma }_{g\left(p\right)}^2)$ as well as the correlation among repeated measures. From preliminary studies, the variance among subjects is ${\sigma }_{\epsilon }^2=0.983$ in the control group and ${\sigma }_{\epsilon }^2=0.877$ in the treatment group; to be conservative we take the variance among subjects to be the larger of these two. We have no information on the variance between sites ${\sigma }_{g\left(p\right)}^2$. However, under effective match-pairing, we expect ${\sigma }_{g\left(p\right)}^2$ to be smaller than ${\sigma }_{\epsilon }^2$. In addition, we do not have any information regarding the expected correlation among repeated measures, so we considered power under a range of possible correlations among repeated measures 12 weeks apart under an exponential correlation function model.

Table 1 gives the differences between treatment means (effect sizes) of PA that can be detected for 12 replicate sites within each treatment and 50 subjects per site under 80% power and a significance level of 0.05. The results suggest that even under the largest variance between sites that we might expect, we should be able to detect a difference of 0.80 h/day in PA, a value less than the 1.09 h/day detected in a pilot study.

Table 1.

Comparison of effect sizes that may be detected with 80% power as a function of the correlation among repeated measures and the variance between groups for the original study design with 50 participants per site and the modified design with only 20 participants per site.

Original Design				Modified Design
	Variance Between Groups				Variance Between Groups
Correlation	0.1	0.5	1.0	Correlation	0.1	0.5	1.0
0.0	0.258	0.563	0.793	0.0	0.270	0.568	0.797
0.5	0.268	0.567	0.797	0.5	0.294	0.580	0.806
0.9	0.292	0.579	0.805	0.9	0.346	0.608	0.826

2.10. Statistical analyses

The primary goal for our data analyses is to describe the impact of the VFB Ecosystem on the trajectory of minutes of moderate to vigorous activity (MVPA) at baseline, 3-, 6-, and 9-months as measured by the Fitbit and cross-validated with the ActiGraph. Here, a linear mixed effects model will be applied including fixed effects for treatment (VFB vs. control), time, and treatment by time interaction, and random effects for participant and matched-pair. The inclusion of matched-pair as a random effect ensures that sites are treated as the experimental units instead of the individual subjects. To take correlations among the unequally-space repeated measures into account, exponential and Matérn-class correlation functions and unstructured correlation matrices will be considered [39]. Gender, age, race/ethnicity, numbers of parents, and BMI will be considered as potential covariates in the mixed-effects models.

Linear mixed effects models will be used to describe the impact of the VFB Ecosystem on motivation, PA self-efficacy, and perceived social support from baseline, 3, 6, and 9-month assessments. In addition, a multivariate version of the linear mixed model will be constructed using a Kronecker product covariance structure [40,41] to describe the impact of the VFB ecosystem on the correlated responses of MVPA, motivation, self-efficacy, social support. Partial correlations among the outcome variables will provide insight into the degree to which motivation, PA self-efficacy and perceived support may promote PA. For example, a strong positive correlation between MVPA and social support would suggest that strengthening the latter may promote positive change in MVPA. Mediating effects of children’s perceived social support from the parents and the virtual pet, perceived PA self-efficacy, and intrinsic motivations for PA will also be assessed. Bayesian Network Models (BNM) [42] will be used to describe the joint trajectories of the dimensions of MVPA, motivation, self-efficacy and social support over time.

2.11. Trial status

Implementation of the randomized trial began in August of 2018 and is ongoing. The trial is currently in the second year of data collection with the second of three participant cohorts.

3. Discussion

Leveraging the features of emerging technology, this trial introduces a novel, SDT-driven approach to encourage 6–10 year old children to adopt PA as a persistent lifestyle. The Virtual Fitness Buddy Ecosystem allows users to set autonomous PA goals, receive tailored feedback on their progress toward those goals, and share their achievements with parents, even when parents and children are physically apart.

3.1. Challenges faced by traditional PA interventions

Traditional PA interventions have been limited in their ability to promote enduring behavior change for several reasons. Emerging evidence suggests one reason for this lies in interventions’ focus on providing young people with extrinsic rewards for PA, which serve as an effective motivator only in the short term [1,2,8,9]. SDT logic suggests that a critical determinant of enacting long-term behavior change is rooted in individuals’ intrinsic motivation to achieve a self-determined goal [10]. Yet due to the cost- and resource-intensive nature of implementing tailored, truly self-determined interventions for each user, existing SDT-based interventions have been forced to implement designs that feature externally determined goals set by researchers and often imprecise feedback on children’s progress toward those goals. Additionally, for practical reasons, the inclusion of parents as PA social support pillars has been limited to cases in which parents can be physically present during the intervention. All of these elements have made it difficult to support and satisfy the three psychological needs of autonomy, competence, and relatedness, which SDT purports as necessary for promoting lasting behavior change.

3.2. Theoretical implications of the ecosystem

The novel VFB Ecosystem overcomes many of the previous challenges associated with implementing truly SDT-based health interventions for children. For instance, in attempts to increase children’s PA, past interventions have often implemented researcher-designed health programs (i.e., externally set PA goals) and offered minimal, if any, tailored feedback on children’s individual PA progress throughout the program [15–17,43]. Yet in order for PA interventions to result in lasting behavior change, SDT logic suggests that they need to be autonomously driven by individual users, and should be accompanied by feedback that is customized to the user’s individual progress [13,14]. Guided by this logic, the Ecosystem was designed as a novel instrument for offering a truly self-determined PA health program to children.

The Ecosystem’s use of interactive technology allows users to set their own PA goals (appealing to autonomy), provides tailored feedback on their progress toward those goals (appealing to competence), and promotes social support between children and parents, even when parents are not physically present, as well as between children and their virtual pets (appealing to relatedness). Beyond providing a rigorous test of SDT’s mechanisms of change, which suggest that appealing to the three psychological needs can facilitate intrinsic motivation toward PA, the current trial aims to advance understanding of the extent to which virtual agents can serve as a reliable source of social support for users, especially children who may not experience PA social support at home. The question of whether virtual agents can provide sufficient social support for children’s PA extends earlier scholarship on whether computers and agents can interact with users and build social relationships [44,45].

Daily PA activities recorded over the course of the year will also allow us to examine the effect of the novelty of the technology. Studies reveal that for many interactive platforms, there is a “novelty effect” wherein interactive elements are highly engaging at first but wane as users habituate to the novelty [46]. Developing platforms that sustain interest is a major challenge for interactive health interventions, and this project will be among the first to investigate engagement with interactive health interventions for children over an extended length of time. In the Ecosystem, the wellbeing of the virtual agent is connected to the child’s wellbeing: as children set and meet PA goals, the virtual agent becomes fitter, more responsive, and playful. Over time, children may develop perceptions of a social relationship with the virtual agent, similar to human-pet relationships in the physical world. These elements promoting sustained engagement may counteract elements that lead to decay in the novelty, and ultimately the internalization of PA benefits. Once internalization is achieved, and PA becomes a daily routine and a norm for children and parents, PA change may be sustained without the assistance of the Ecosystem.

3.3. Practical implications of the Ecosystem

The VFB Ecosystem overcomes several practical challenges associated with implementing health interventions in the field at scale. First, the Ecosystem was designed to be largely self-sufficient, requiring only minimal interaction between researchers administering the intervention and after-school site adult supervisors facilitating access to the kiosks. The value of a cost- and labor-effective health intervention at scale is notable and should not be understated. Additionally, the present trial uses emerging interactive technology to connect young users and their parents even when they are not together physically. By nesting a technology-mediated intervention within the context of the parent-child relationship, the Ecosystem encourages families to construct and share a communal goal toward PA promotion, rather than the child working toward a PA goal in isolation.

More importantly, the effectiveness of the Ecosystem’s design provides important insights for the future development of PA interventions using technology to unobtrusively track users’ PA. The Ecosystem’s use of consumer-grade activity monitors allows for the tracking of continuous PA during an academic year in a diverse sample of elementary school aged children. The activity monitors have the potential to inform current understanding of intra-day patterns of activity during school days, weekends, and during intersessions. Concurrent measurement of PA with the waist-worn ActiGraph and wrist-worn Fitbit will allow a comprehensive comparison of the wrist-worn Fitbit activity monitor with the well-validated waist-worn ActiGraph. Prior comparisons of Fitbit and ActiGraph estimates in children have been restricted to earlier waist-worn Fitbit devices during school hours and/or for a short period of time.

3.4. Challenges/limitations

Although the present trial overcomes many barriers to implementing SDT-based PA interventions in the field at scale, there are several unique challenges to this trial, both at the user-level and the site-level. At the user level, ensuring children consistently wear their activity monitors is paramount. Without their Fitbit devices, children were unable to fully engage with either the treatment or control kiosk, making Fitbit wear a critical component of the intervention. We have addressed this issue by attempting to keep Fitbit wear at the forefront of families’ minds, by visiting after-school sites periodically and talking with children to encourage wear, sending text-message wear reminders to parents, and giving children the ability to use the kiosk to send a text-message to their parent indicating that they forgot their Fitbit. An additional challenge lies in the extent to which children are able to meaningfully set and modify PA goals. For most children enrolled in the intervention thus far, learning to use the kiosk and interact with their pet (if applicable) is straightforward. However, especially for young children, determining a feasible PA goal and continually updating that PA goal based on PA progress as displayed by the kiosk requires some degree of trial and error. Nevertheless, once they become acclimated to viewing and interpreting their PA progress, we find that children of all ages are able meaningfully set and modify their PA goals.

To be effectively implemented at scale at the site level, this trial requires basic technology and space infrastructure, such as reliable wireless internet access and modest storage space for kiosk safekeeping. Despite being mostly self-sufficient, regular staff engagement at each individual after-school site is also necessary, as children have appeared to be more interested in engaging with the kiosk if adults are excited about children using it. The importance of staff buy-in is underscored by the fact that the kiosk is designed to have one child visit at a time. Given that most after-school programs are scheduled around group-based activities, limiting children’s access to the kiosk on an individual basis requires some acclimation for children, and active involvement of staff is critical in facilitating this.

4. Conclusion

To encourage PA among children aged 6–10, we developed the Virtual Fitness Buddy Ecosystem, a novel, comprehensive system of consumer-grade digital devices with the goal of facilitating children’s enduring intrinsic motivation for moderate-to-vigorous intensity PA. Capitalizing on the power of PA social support that parents can provide to children, the Ecosystem allows parents to stay involved in the intervention and support their children’s PA goals via their smartphone. Additionally, the virtual pet, housed by the Ecosystem kiosk is designed to mimic human-pet relationships in the physical world, guiding children to set and meet PA goals and offering tailored feedback on those goals. Using technology to amplify the power of existing social relationships, the Ecosystem is a cost- and labor-effective PA intervention with the potential for widespread public health impact.