Mental health conditions are the main contributor to the substantial increase in childhood disability in the last decade [1], with most having their onset in childhood or adolescence [2,3]. Recent estimates suggest a 10% prevalence of mental disorders in children and adolescents in Great Britain [4] and 12% in Europe [5], whereas approximately 1 in every 4 to 5 youth in the United States meets criteria for a mental disorder with severe impairment across their lifetime [6,7]. This realization is fueling calls for interventions in childhood to avert the development of long-term disability [8-10]. Research in prevention science showcases the feasibility of such interventions in child populations: prevention programs develop key cognitive and emotional protective factors—such as emotion regulation (ER) or coping strategies—which, in turn, can reduce the incidence of mental health disorders in later life [11-17]. There are a variety of types of prevention programs, from universal interventions that are designed to be used with all children to indicated interventions that are targeting those already presenting with early signs of serious disorders [12]. Similar to therapeutic settings, existing prevention programs rely predominantly on in-person training. As a result, these interventions struggle with the challenges of cost, reach, and intervention fidelity [18-22].

Although existing programs are relatively successful in targeting children within the captive audience context of schools [21-25], a principal challenge remains in extending this support into the day-to-day contexts in which protective competencies are applied, practiced, and developed [22]. The current model relies on parents to deliver such at-home interventions and requires extensive training to do so effectively: For example, a shortened version of the Incredible Years program [26,27] still required 12 to 24 weeks of parent training in groups of 6 to 10 parents for 2.5 hours, once a week. Other programs, such as the seminal Perry Preschool program, were even more intensive, comprising a 2-year program of 2.5 hours of interactive academic instruction daily for children at school, coupled with 1.5-hour weekly home visits by trained staff [28]. Such approaches experience low enrollment rates, and the lack of continued engagement with interventions beyond formal delivery classroom context is also a common limitation [18-20]. These difficulties in bridging the formal school and informal home contexts are crucial in prevention science: family interactions are a strong mediating factor for developing resilience and impacting core socioemotional competencies, especially for younger children [29-33]. Moreover, lack of consistency of at-home and at-school support diminishes the effects of prevention programs [22,34].

New delivery mechanisms and intervention approaches are sorely needed to address these issues [8]. Digital mental health interventions are increasingly seen as having the potential to deliver on these aims, revolutionizing when, how, where, and to whom interventions can be delivered [10,35-37]. Although the interest in technology-enabled mental health continues to soar—especially in the context of treatment for adult populations—a consistent set of challenges has, however, emerged around ensuring uptake and long-term engagement of digital interventions [10,38]. Reliance on didactic and information delivery models, limited use of user-centered design, and lack of immediately perceived benefits leading to low motivation are commonly cited reasons [39-41]. These difficulties are likely to be exacerbated for prevention interventions for children, but surprisingly little research has investigated it empirically [42,43]. As such, it is not clear if and how technology could be used to facilitate transfer of such learning from school into families; or to enable new types of interventions that would empower parents and children to further develop protective competencies independently of formal training programs.

This work investigates a proof-of-concept prototype of a newly proposed intervention delivery mechanism within the context of (1) universal prevention programs [25] for children aged 6 to 10 years and their families and (2) ER as a specific instance of a psychological protective factor. We chose ER as it is a fundamental life skill, with effects on life outcomes comparable in size to those of IQ or family social status [44,45]. Research shows that these effects are wide-reaching: if ER is poorly developed, it leads to increased incidence of both internalizing and externalizing mental health disorders [46-49] and is associated with societal problems such as criminal behavior [50], low personal well-being [44], and academic underachievement [51]. Moreover, existing intervention research shows that ER is difficult to develop without detailed in situ guidance and support [26,52-54]; and parenting strategies play a key role in shaping child emotional coping and regulatory skills [55-63]. This is particularly important within underprivileged families: prior research repeatedly shows that children from these populations are at risk of low self-regulation competencies at an early age [64,65], and the gap further widens over the school years [66].

The data reported here build on an iterative user-centered design process, which led to the development of a novel intervention prototype described in the next section. Within the 2-year-long development phase (reported in full elsewhere [67]), we worked with children, parents, and prevention science experts to codesign a proof-of-concept technology platform to support children in developing ER skills. Theoretically, the intervention is grounded both in basic models of ER [68], as well as close collaboration with developers of evidence-based interventions (Second Step), while also deeply involving children and families in codesign to ensure the intervention fits into their daily lives [39,40,69]. In effect, the designed prototype attempts to fuse the understanding of evidence-based methods from prevention science (what works), human-computer interaction (what is technically feasible and acceptable to users), as well as insights into the everyday practices of families within the social context we designed for (what people actually do). This iterative design process has led to a novel situated intervention model: the intervention is delivered through an interactive object (smart toy) sent home with the child, without any prior training necessary for either the child or their carer (see the next section for design details and logic model).

The aim of this qualitative study was to provide an initial validation of the feasibility of core fundamental principles underpinning the proposed novel intervention model [70], which was developed in previous research [67]. Specifically, the intervention model assumes that (1) children will be naturally compelled to keep interacting with the intervention without external guidance; (2) it will become incorporated into their everyday emotion regulatory practices, even without any formal training; and finally (3) the intervention would be perceived as acceptable to parents. Given the novel nature of the proposed delivery mechanisms, it is crucial to test whether these principles are fulfilled by the current prototype before more expansive investigations take place.

Data from exploratory deployments reported in our previous study [67] are promising; however, these are limited by short post hoc interviews with children, no information from parents, no objective log data, and only very short deployment times (median 3 days). This study builds on these preliminary findings using a range of data-collection methods (pre- and postinterviews with parents and children, log data analysis, and child photo diaries) to investigate (1) engagement and acceptability of the device in the homes during 1-week deployments and (2) subjective indicators of effects on emotion regulatory practices (whether positive or negative), as reported by parents and children.

The prototype takes the form of a hand-crafted plush toy (see Figure 1 and the study by Slovak et al [67] for the design process), which was designed to travel home with the child from school and support in-the-moment soothing. The toy is introduced to the child as an anxious creature that needs kind attention from humans, such as soft stroking and hugging. Embedded electronics enable the prototype to produce vibration patterns that simulate a heartbeat (ranging from frantic to slow and steady). When picked up, the toy emits a frantic heartbeat that slows down if the child uses calm stroking movements, as registered by the embedded sensors (see Figure 2). If the toy is soothed for long enough, the prototype transitions into a purring vibration indicating a calm, contented state. For a full description of the physical design, interactive features, and a more detailed logic model, see Multimedia Appendix 1. We included 29 publications in Multimedia Appendix 1 [65,68,71-99].

The logic model underlying the intervention is assumed to operate on 3 levels building on each other: Level 1 pertains to directly providing in-the-moment soothing support to children in naturally occurring emotional moments when they would attempt to calm down. The prototype’s physical and interaction design was aimed to tap into a number of known regulatory factors, grounded theoretically in Gross’ extended process model of ER [68]. Specifically, we designed the prototype interaction with the aim to impact 2 separate stages: the attentional deployment stage [71-75], by shifting children’s attention from the emotion-eliciting situation toward interacting with the toy and the response modulation stage, by facilitating downregulation through pleasant tactile interaction analogously to the mechanisms assumed to underpin emotion regulatory effects of human-animal interaction [76-81].

Level 2 is concerned with mechanisms that facilitate children’s long-term engagement with the intervention, building on the positive subjective experience of in-the-moment soothing. The framing of the toy as an anxious creature in need of assistance is the hypothesized key driver: we assume that this framing will not only support conveying the benefits resulting from extrinsic ER [77,82,83], but also facilitate the creation of a sense of relationship and responsibility for the well-being of the creature, similar to the long-term engagement seen with child-oriented robots [75] or products such as Tamagotchi [84-86].

Finally, level 3 is assumed to emerge from repeated experience of soothing interactions over time, leading to a shift in children’s ER practices and implicit beliefs about emotion. Specifically, we hypothesize that repeated interactions with the toy will result in the establishment of more adaptive ER patterns and shift children’s implicit beliefs about the controllability of emotion [87,88], a well-known target for intervention [89-93]. As these effects are expected to arise only through ongoing long-term interactions and, thus, rely strongly on appropriation in situ, we did not expect to see any indicative data for these proposed mechanisms within this pilot study; however, these will be crucial for long-term effect of the smart toy intervention. This study aimed to provide pilot indicative data pertaining to levels 1 and 2.

The goals of this early feasibility study were to investigate the engagement and acceptability of the device in the homes during 1-week deployments with children aged 6 to 10 years and their families and also subjective indicators of effects on emotion regulatory practices (whether positive or negative), as reported by the parents and children. Together, the aim was to collect indicative qualitative data pertaining to level 1 and 2 of the underlying logic model: we were interested to see if children would find the individual interactions comforting, whether they would sustain engagement over the week periods (and what role any emerging relationship with the toy might play here), and whether the toy will become embedded into their everyday activities, including being explicitly used for ER.

As we were interested in studying natural appropriation in situ, children were given a prototype, a discovery book that presented the simple narrative and suggested playful activities, and a simple digital camera to keep at home for 7 to 8 days. We gathered data through a number of channels. The main sources were (1) semistructured interviews with parents and children before and right after the deployment; (2) any photos the children took during the deployment, which also served as ticket to talk about their experiences during the week; and (3) automatically collected logs by the prototype, which recorded all touch interactions throughout the weekly deployment.

The discovery book contained some information about the creature’s background and various activities the child could fill in on their own or with the help of their parents, such as photo challenges around the toy, and an emoji diary where they could use emoji stickers to keep track of how they and their creature were feeling on each day of the deployment. In designing the discovery book and activities, our aim was to facilitate children’s engagement with the toy in a playful manner as well as complement the interview data with a richer understanding of how families experienced having the toy at home. As such, the discovery book was as much a research tool as a part of the intervention (implicitly providing the narrative and suggested activities).

The study was funded by a personal fellowship and University College London (UCL) and received ethical approval from UCL’s ethics committee (3923/005).

The prototypes and accompanying materials were deployed in waves to 10 families of 11 children (3 girls, 8 boys; aged 6-10 years) from August to November 2018. One additional family had been recruited, but experienced a malfunctioning prototype and has not been included in the dataset. Participants were recruited from 3 communities in the United Kingdom through a range of methods, including online advertisements, in-person recruiting in 2 schools which had served as recruitment sites for previous phases of the project, and snowball sampling. The majority of participating families (7/10, 70%) lived in an area falling within the 20% most deprived in England (measured according to English Indices of Multiple Deprivation [100]), with the remaining 30% (3/10) living in areas falling within deciles 3 to 5. Recruitment was stopped based on data saturation [101]: the interview data collected were highly consistent across families, with only limited new insights emerging by the tenth interview, within the context of a pilot study.

All engagements with families were conducted by the first author, who holds an MSc in developmental psychology. The researcher visited families who had orally agreed to take part to obtain consent from parents and assent from children, conducted a semistructured interview with at least 1 parent, and gave children the toy, discovery book, and a simple digital camera to keep at home for 7 to 8 days (1 deployment was extended for a day because of a technical failure and 2 more for scheduling reasons). The first semistructured interview with parents focused on families’ existing emotion regulatory practices, perceived challenges to ER, and parents’ expectations from the week-long deployment. After 3 or 4 days, the researcher visited families again to change the toy’s battery. On the last day of the deployment, the researcher visited the families to pick up the toy and materials and interview each child and at least 1 of their parents individually (see Multimedia Appendix 2 for the interview guides). The interview sessions (approximately 1 hour) were conducted in person in participants’ homes. After the end of the interview, parents completed a brief demographic questionnaire with items on age, race, ethnicity, education level, current employment status, marital status, and housing situation. Engagement with the toy was tracked automatically by the toy throughout the deployment, by registering and logging every interaction with a timestamp.

The semistructured interview conducted at the end of the deployment included questions designed to elicit participants’ views and experiences of using the toy as well as their expectations of long-term outcomes if they were to keep it for longer. During the interviews, the photos children took and the completed activities in their discovery books were used as prompts to ask families about the child’s engagement with the toy. The interview sessions (approximately 1 hour for the parent interview and 30 min for the child interview) were conducted in person in participants’ homes when the researcher visited the families to collect the toy and accompanying materials. Families were offered £50 compensation for their time. All interviews throughout the development were audio-recorded, with permission from the parents and children; the researcher also collected simple field notes about who was present during the visits and also detailed any additional observations that seemed important but would not be captured by the audio recordings .

The study included 11 children from 10 families as a pair of siblings received 1 toy each during the same week (female children n=3; female parents n=11; mean age of children 7.1 years [SD 1.22, range 6-10]; mean age of parents 37 years [SD 5.36, range 28-44]). For a more detailed description of participants’ demographic characteristics, refer to Multimedia Appendix 3). One additional family had a malfunctioning prototype and has been removed from the main analysis. Table 1 includes individual information for age, gender, and other deployment-related information for each of the children. We had no attrition; all participants finished all phases of the study.

In this section, we are reporting on the interaction data automatically collected by the toy during the deployments. As outlined in the Data Analysis section, we classify any given minute as active only if the toy logged at least 20 different sensor interactions during that 1-min interval. This is to avoid counting accidental touches, or just moving the toy from one place to another.

Overall, the log data support the qualitative observations, showing sustained engagement throughout the deployment: the families used each toy, on average, for 74.9 active min per day (median 60.5; SD 64.1; see Figure 4 for box plots for individual children). We did observe that, overall, the average interaction times per day were longer for the first 3 days of the deployments compared with the last 3 days—but even then, the average active engagement was 43.8 min per day (median 30.5, SD 35.7). This might indicate that the engagement was stronger in the first few days because of novelty effects, and the children’s interest in the toy started to wane toward the end of the deployment. Another plausible explanation that would be in line with interview data is that in the first days, children and other family members interacted more with it as they were exploring the features, whereas in the last 3 days, the children already knew how the toy worked and used it as and when they needed it. Long-term deployments are needed to understand the stability of engagement beyond the first week.

As expected, we observed a stronger engagement on weekends and holidays when most children would interact frequently with the toy throughout the day, whereas on school days, children interacted with it the most early in the morning (before school) and in the afternoon. To illustrate this, Figure 5 visualizes the weekly active minutes for child 7, selected as a typical example: child 7’s overall active minutes length is close to the median of the dataset and also qualitatively typical to the interaction patterns we observed for other children. In this case, comparing the data on a weekend day (Sat 22nd) and on a school day (Wed 26th) exemplifies how the active times have been influenced by school times: with the child having frequent interactions with the toy from the morning up to the evening on the weekend day, while briefly engaging with the toy in the morning before school and throughout the afternoon after their return on the school day. The log data also seem to confirm participants’ reports that children would at times interact with the toy around bedtime to relax. In some cases, interactions were also registered at nighttime, suggesting that children had the toy in bed with them; because of the inherent limitations of the log data in terms of interpretability, we cannot ascertain if these touch traces represent intentional (eg, children waking up in the middle of the night and stroking the toy) or accidental interactions.

When all interaction data are aggregated, the most frequently activated sensor was that of the back (35%), followed by the gyroscope (26%), feet (20%), and ears (20%). The large percentile of back sensor activation is consistent with the patterns of interaction reported in participants’ interviews, as hugging and stroking the toy’s back—both of which would activate the back sensor—were reported as children’s preferred soothing interactions. Although the percentile of gyroscope activation was higher than we expected, considering it consistently happened alongside the activation of the back sensor, it does not seem likely that it indicates shaking or rough handling by the children.

The aim of this qualitative in situ study was to investigate the engagement, acceptability, and initial subjective indicators of emotion regulatory effects for a proof-of-concept intervention model, as instantiated in a smart toy prototype. The novelty of the proposed approach was to deliver at-home interventions through an interactive object that becomes incorporated into child’s everyday interactions to provide in-the-moment regulatory support, without any explicit training necessary for the child or the parent.

The fundamental assumptions underpinning the logic model of such situated and child-led intervention was that (1) children would be naturally compelled to keep interacting with the intervention without external guidance; (2) it would become incorporated into their everyday emotion regulatory practices, even without any formal training; and finally (3) the intervention will be perceived as acceptable to parents. The qualitative findings described above suggest that all 3 conditions were satisfied: all children reported sustained engagement with the prototype, without any externally imposed conditions and have been consistently labeling such interactions as subjectively pleasing. Both parents and children further described the observed emotion regulatory effects of child-toy interaction under a variety of contexts (eg, self-soothing after an interpersonal conflict, reduction in subjective anxiety levels, relaxation support, and coping with pain). Finally, all children and 9/10 parents were keen on keeping the prototype for longer, suggesting a high acceptability and suitability with respect to social practices in the home.

The qualitative findings also provided some indicative support for the hypothesized mechanisms underpinning the first 2 levels of the logic model: level 1 as facilitating in-the-moment regulatory support (relying on attentional deployment and response modulation) and level 2 as scaffolding ongoing engagement (through the creation of an emotional attachment to the toy).

For level 1, the experiences described by both parents and children supported the in-the-moment regulatory effects: the children described the moments of holding the prototypes as happy and calming, and some have reported to deliberately seek the interaction to calm down. Interestingly, half of the parents have described similar soothing experiences themselves, suggesting that the effects might be consistent across a wider age range, as could be expected given the reliance on fundamental emotion regulatory mechanisms [68,71,73,76,77]. Although it is impossible to disentangle the assumed attentional deployment and response modulation mechanisms based on the retrospective interview data, the stories captured in the interviews provide some support for the hypothesis that physiological effects arise from a combination of tactile stimulation (eg, “I just put it to my chest and it worked” type of quotes common across the dataset) and more conscious focus on changing the creature’s emotions.

Similarly, the hypothesized level 2 mechanisms have received indicative support in the interview dataset. All children referred to the prototype as if it were alive, attributing a range of human-like mental states to the toy, together with an associated range of caring behaviors (eg, making a bed or custom-made clothes to help it feel warm, making sure it is not stressed, and controlling how others interact so as to not hurt it). Combined with the sadness associated with the end of deployment, these observations suggest that the prototype was successful in generating an emotional attachment, which appeared to facilitate the continued engagement. These relationship-building effects appear analogous to those observed with other animal-like robots in other contexts: see Turkle et al [103] for a critical analysis of the mechanisms behind such computational devices presenting themselves as relational artifacts.

The study data do not provide indications of any effects on longitudinal shifts in emotion regulatory practices (level 3) because of the short-term deployment and lack of baseline and follow-up measurements. Further efficacy research, including in situ studies (such as randomized wait-listed designs in schools), is needed to understand the effects of the existing prototype on child ER practices and mindsets. Interesting research directions also include questions around the impact of associated materials (such as the discovery book) on the intervention effectiveness.

To the best of our knowledge, the proposed intervention model is unique in prevention science as it suggests an intervention delivery method that becomes fully embedded in children’s everyday lives, does not require any explicit training, and is relying on in-the-moment experiential support rather than information delivery. It draws inspiration from the large body of research on animal-assisted interventions (see Crossman [104] for a review), which has suggested promising outcomes in a number of populations. These include increased social interaction among children with autism spectrum disorder [105], increased social behaviors and reduced agitation and aggression among persons with dementia [106], reduction in symptoms among patients with depression [107], and increased emotional well-being such as reduced anxiety and fear [108]. A related area of work is focused on social assistive robots [84,105-108], which are designed to act as pet surrogates, such as the robotic seal Paro [109]. A majority of such socially assistive robotics (SAR) interventions has so far, however, focused on occasional use by older adults, particularly those suffering from dementia [109-114].

The design of SAR with typically developing children has been limited to educational interventions outside of mental health domain [115-117]. Despite the reported promising outcomes of SAR interventions in other contexts, no studies to date explored the use of SARs as part of prevention interventions (for ER or other protective factors) with typically developing children, and only 1 recent study [80] has explored the effects of interacting with Paro robot on children’s mood, anxiety, and arousal after exposure to a lab-based, stress-inducing task: interaction with the robot resulted in greater increases in positive mood than any of the control conditions but did not have a significant effect on negative mood, anxiety, or arousal.

More broadly, this proof-of-concept prototype can be seen as illustrative of a conceptual shift in how early prevention interventions might be created and delivered with technology: the notion of situated interventions and child-led rather than parent-driven approach.

The goal of a situated intervention refers to designing programs that will allow the families to draw on—and learn from—specific lived experiences as part of the intervention. This goes beyond purely just-in-time intervention delivery such as reminders or activity suggestions [37,118]: the purpose is to flip the existing intervention model that is based on information delivery and didactic learning (eg, at an in-person workshop or classroom lesson) to be applied later toward a model where the intervention directly supports both children and parents to learn from the daily emotional challenges they encounter. As with the example prototype discussed here, successful situated interventions would aim to embed intervention delivery as an implicit part of everyday situations—such as those of stress, anxiety, or sadness in the case of the toy presented in this study. The goal is then to utilize these everyday moments as an opportunity for ongoing, iterative training, rather than having to rely on vignettes, role-plays, or the recollection of past experience as is common now [18,22,34]. Psychologically, the notion of situated interventions thus corresponds to the need for in-the-moment scaffolding of experiential learning that underpins all socioemotional competencies [22,51,119-121] but has been pragmatically impossible to date.

The second key shift toward child-led interventions argues for the potential of repositioning the child as the immediate recipient of some or all aspects of the technology-enabled intervention. In the current prevention science models, the child is either seen as a captive audience within the in-school programs or as a secondary actor who is impacted by parental training. The reasons for this are understandable: the existing interventions could not rely on young children to drive the intervention as it is, for example, unlikely that a child aged 6 years would be able to teach their parents new parenting strategies as a workshop coach might, or directly engage (or want to engage) with a written text on a leaflet sent home. The ongoing, in-the-moment scaffolding facilitated by situated, technology-enabled interventions could address both of these issues and reposition the child as the main actor of the intervention, both in terms of who is driving the intervention transfer to home as well as who is to be engaged with the intervention once it is there.

One strength of the study was the emphasis on in situ unstructured deployments, which provided ecologically valid data about possible appropriation in families. Most parents were from underprivileged neighborhoods, and many were in difficult personal situations; we have avoided tapping into the proverbial worried well and instead worked with a population who could be expected to strongly benefit from ER interventions [122-124]. The detailed interviews then provided a holistic understanding of how the prototypes have been used and the impact they might have on the family life. Another strength was including the interview data from both parents and children (in addition to photographs collected by participants during the week), triangulating the evidence across all stakeholders.

The data have been promising in terms of observed engagement and acceptability, which were high across all 10 families recruited into the study. This consistency—together with analogous positive effects from earlier deployment [67]—is particularly promising in view of the commonly high attrition rates and nonengagement for technology-enabled mental health interventions [10,38-41]. However, there may have been some self-selection recruitment effects: the families have explicitly opted into the study and, thus, might be more likely to respond positively than the general population. Further studies should investigate the engagement rates when deployed, for example, as part of school-based approaches and with reduced researchers’ engagement (eg, questionnaire rather than interview methods).

An expected limitation of a pilot qualitative study is the lack of definitive data on psychological effects. Although participants’ reports suggest that they experienced subjectively significant changes to their everyday emotion regulatory practices, more rigorous studies are necessary to understand the strength of psychological effects and whether these would scale up. In particular, it is not yet clear if these would lead to long-term changes, and whether the magnitude would lead to a clinically significant change in emotion-coping mechanisms and strategies [45]. As such, the lack of data on the presumed level 3 effects is the most important gap. It will require not only rigorous efficacy study designs to estimate the current effects but also likely further iterative codesign development (with parents, children, and prevention science experts) to strengthen the intervention impact. The qualitative pilot data from this and previous publication [67] provide a good starting point for such future work.

This is the first known study investigation of the use of object-enabled intervention delivery to support ER in situ. To understand the feasibility of such novel intervention mechanism, this qualitative study examined its appropriation and engagement by 11 children from low-socioeconomic status families over the period of 1 week. Triangulating both parental and child interviews, the data provide a holistic picture of how the prototype was incorporated into the family life. The strong engagement and qualitative indications of effects are promising—children were able to use the prototype without any training and incorporated it into their ER practices during daily challenges. Future work is needed to extend these indicative data with larger studies examining the psychological efficacy of the proposed intervention. More broadly, our findings suggest the potential of a technology-enabled shift in how prevention interventions are designed and delivered: empowering children and parents through child-led, situated interventions, where participants learn through actionable support directly within family life, as opposed to didactic in-person workshops and a subsequent skills application.