Multi-stakeholder Perspectives on Mental Health Screening Tools for Children

Pediatric mental health is a growing concern around the world, affecting children’s social-emotional development and increasing the risk of poor behavioral outcomes later in life. However, obtaining a behavioral diagnosis in early childhood is challenging due to lack of access to resources, low parental mental health literacy, and children’s dependence on several stakeholders to coordinate care for them. While app-based, at-home screening tools could offer a scalable and convenient diagnostic solution for families, stakeholder perspectives on their utility and usability remain to be examined. This work reports on a survey of child mental health practitioners and interviews with parents to illustrate existing barriers to care that stakeholders encounter, the perceived benefits of app-based screening tools in meeting their needs, and the challenges in scaling up these tools. We identify where stakeholders agree or disagree, delineate key design tensions, and provide recommendations for the development of future screening technologies.


INTRODUCTION
Over 7 million children in the United States, or 16.5% of the population under 18 years of age, are impacted by mental health disorders [90].Poor mental health in childhood is associated with a decline in quality of life [73], impaired social-emotional development [18,24], and poor learning outcomes [61].Childhood psychopathology also predicts persistent risk of mental disorders into adulthood and across the lifespan [69], the subsequent onset of substance use disorders [47], and a negative impact on employment status and income as an adult [11].Additionally, pediatric mental health exerts an enormous economic burden on society [11,25].
While this has led to growing interest in diagnosing and treating childhood mental disorders, over half of the children with mental health issues fail to receive treatment [77,90].The mental health treatment gap in pediatric populations stems from a range of systemic issues, including an under-resourced mental health system [51], socioeconomic and racial disparities [21,55], the stigma associated with mental disorders [5], and logistical or fnancial burdens of accessing care [17].Children specifcally also cannot seek help for themselves and require an adult primary caregiver to identify potential problems and further have the institutional knowledge and resources to address them [58].However, parents and/or family caregivers may not be able to distinguish between normative behavior and signs of non-normative development in young children, creating a "when to worry" problem for families [86].Evidencebased clinical assessments for young children are also difcult and time-consuming to administer and have relatively poor diagnostic accuracy and specifcity [37,57].
Digital mental health tools have been recognized as alternatives or supplements to the existing mental health care landscape.A range of technologies have been developed to support families track and manage children's development and behavioral symptoms (e.g., [43,46,48,59,64]).Digital interventions have been proposed for depression, anxiety, attention-defcit/hyperactivity disorder (ADHD), autism spectrum disorder, etc. in adolescents and young adults [35].Researchers have highlighted the importance of co-designing digital health interventions with users to increase their uptake and continued use, user satisfaction, and the likelihood of success [14,28,82].This has led to several design explorations and user-centered development exercises in the human-computer interaction (HCI) and computer-supported cooperative work (CSCW) communities around child mental health (e.g., [14,52,71,83]).
However, the design space of mental health screening technologies for children, especially toddlers or preschoolers, has been relatively under-explored.Prior research has focused on understanding technology use practices [29] and decision-making experiences [50] among parents of infants and toddlers.Concerning physical health, the development of caregivers' experiential knowledge [75] and tensions around parenting children with chronic medical needs [74] have been explored previously.Researchers have further looked into how technology can support new parents [30,31,48,85], but there remains a lack of understanding regarding how sociotechnical systems can support caregivers of children with behavioral concerns identify and seek help for these concerns.Similarly, although there has been prior work exploring tools for parent-clinician communication in the context of children's health (eg., [43,64]), it is unclear what data child mental health practitioners would want to collect using at-home sensing technologies.
In this work, we attempt to fll these gaps by examining the perspectives of two key stakeholders -child mental health professionals/clinicians and parents/family caregivers -toward using mental health screening or diagnostic applications for young children 1 .Specifcally, we focus on investigating the perceived utility and drawbacks of app-based, at-home screening tools that utilize behavioral data to predict mental health outcomes in preschool-aged children.Our motivation to concentrate on this specifc class of screening technologies stems from prior work demonstrating the increasing use of mobile applications to support parenting, especially in the early years of a child's life [85].Previous HCI research has also made signifcant advances in developing app-based screening tools for young children using a variety of sensing modalities (e.g., [36,43,44,48]), but comparatively little is known about stakeholders' perspectives towards such tools.
Within the space of app-based screening tools, our work attempts to answer the following research question: What are the unmet needs and gaps in current caregiving practices of parents and clinicians, and to what extent can these be addressed by at-home screening technologies?Recognizing the needs of both stakeholder groups is imperative for researchers and technology developers to create tools that are benefcial and amenable to users and can be seamlessly integrated with existing diagnostic practices.We therefore also investigate diferences in opinion among parents and clinicians concerning child mental health screening tools.
We analyze free-form survey responses from sixty mental health professionals collected by the authors of a prior work that introduced an app-based screening tool for young children [44].Our study is, to the best of our knowledge, the frst to undertake a systematic qualitative inquiry using this dataset to derive thematic fndings.We complement this with original data from interviews with 26 parents of preschool-aged children to gain insight into their lived experience identifying and managing behavioral issues.
Thematic analysis of survey and interview data reveals a lack of scafolding to support parental help-seeking, a dearth of resources for tracking emotional development, and limitations in data availability that impact pediatric mental health screening.Stakeholders collectively outlined various benefts of app-based screening tools, including increased access to screening, the potential to augment existing practices, making diagnosis more frictionless for children, and alleviating parental pressure.At the same time, they identifed challenges in deploying such tools, including integrating them with traditional services, building trust, and minimizing potential harm.Our work also delineates how clinicians and parents have diferent values and sensitivities concerning the utility and drawbacks of home-based screening tools.We further analyze stakeholders' perspectives through Tatar's design tensions framework [81] to identify conficting requirements that must be balanced against each other to build highly scalable and usable mental health screening tools.We ofer concrete design recommendations for developers of screening technologies to achieve this goal.

RELATED WORK 2.1 Technologies for Development Tracking in Early Childhood
There has been long and sustained interest in the HCI community to develop tools that support parents in tracking their child's development in the frst few years of life.For example, Baby Steps encouraged parents to frequently record their child's developmental progress and share it with their pediatrician [48].The system was further extended to a Twitter-based service [78] as well as a text message-based developmental screening tool [79].babyTRACKS allowed parents to record short textual descriptions of developmental milestones and receive expert-curated, crowd-based percentiles to compare their child's development relative to others [12].Spe-cialTime analyzed parent-child dialogues to provide parents with real-time feedback on dyadic interaction patterns while the family underwent parent-child interaction therapy [36].Research has additionally looked into collaboration opportunities between parents and healthcare providers for identifying relevant data to track [83].
Other tools have focused on communication and information sharing among various stakeholders involved in caring for young children.For example, CRAFT allowed parents of children with developmental disabilities to record videos of problem instances that were later reviewed by behavior analysts [64].GeniAuti supported parents of autistic children in recording challenging behaviors based on a clinical data collection form, which experts could then use to make recommendations [43].In light of the signifcant progress made by the above technologies toward identifying and communicating mental health concerns, our work aims to systematically investigate multi-stakeholder perspectives on the perceived utility and impact of such screening tools in meeting their current needs.children while interacting with an emotional robot to diferentiate between children with autism spectrum disorder and those developing typically [15].Building blocks embedded with inertial sensors have been used to sense play styles and predict behavior problems in young children [89].Researchers have also used accelerometers placed on the wrist and ankle to classify children with ADHD vs. healthy controls during a computerized continuous performance task [32] or over the course of a regular school day [63].
Among game-based screening tools for mental health issues in children, EarlyScreen used video-based behavioral markers to predict emotion regulation-related disorders [44].Jiang et al. used a combination of interactive devices and wearables to diagnose symptoms of ADHD in 7-to 13-year-old children as they completed gamifed tasks on a large touch-screen system [42].Other researchers have used gameplay data to detect ADHD, anxiety, and depressive disorders [34,72].Song et al. developed a narrative game to test various dimensions of cognitive control among children and adolescents [76].Smartphone-based games have also been developed for in-the-wild detection of autism spectrum disorders in children [22,84].In addition to games that have been specifcally developed for mental health assessment, Mandryk et al. argue that in-the-wild data collected while playing commercial, of-the-shelf games that are primarily designed for entertainment purposes can be valuable digital biomarkers for mental health [54].Prior work has also explored associations between older players' mental health and their choice of games and genres [67] as well as in-game behaviors [13].

Designing Screening Tools to Support Caregivers
While there is limited research on caregivers' use of, and expectations from, technologies for behavioral screening, we can draw insight into caregiver motivations from prior work exploring these themes in the context of tools for infant tracking.Marx and Steeves observed that parents are often coerced into tracking or "surveilling" their children using technology in order to keep them safe while promoting parental convenience [56].This is brought on by both intrinsic and external factors -Lupton notes that mothers are inherently acutely aware of their parenting responsibilities and additionally face societal pressure to "conform to the ideal of the 'good mother' " [53].Infant tracking technologies can indeed support parents in fulflling these responsibilities and help alleviate pressure -prior research shows how parents use tools both to confrm hypotheses about their child's development that arise from their own lived experience and to minimize relying solely on their intuition to identify potential concerns [88].However, technologies that are designed to help parents overcome their anxiety have also been shown to exacerbate it instead, often by providing continuous and unprecedented access to healthrelated data [45,65,88].Over-reliance on technology may also lead to a false sense of security [88] or complacency and lack of vigilance among parents [10].It may take away from an embodied parenting experience that relies on knowledge and intuition gained from lived experience [53].Therefore, technologists have called for tools that support families by improving parents' confdence in fulflling their new responsibilities, rather than replacing parental intuition [31].One such framework is proposed by Kaziunas et al., who advocate for a 'caring-through-data' approach where data from tracking technologies is used to empower users and foster empathy and communication [45].
Most of the above research, however, focuses on supporting parents in tracking physical health and development, ignoring potential concerns about behavioral and social-emotional needs.We aim to fll this gap by examining the perspectives of caregivers toward behavioral screening tools, focusing on app-based, at-home screening technologies that can potentially supplement existing mental health services.

METHODS
As described previously, we focus our work on understanding the perspectives of two user groups -mental health practitioners and parents/caregivers -toward app-based behavioral screening tools for young children.Stakeholders' perspectives on screening tools were gathered via a practitioner survey undertaken by prior work and original interviews with parents, which are described in Sections 3.1 and 3.2 respectively.

Study 1: Analyzing Survey Responses from Child Mental Health Practitioners
For our frst study, we qualitatively analyzed survey responses from 60 child clinicians and other mental health practitioners collected by the authors of [44] to examine the perspectives of mental health professionals toward screening apps.While this prior work conducted a quantitative examination of Likert scale responses from clinicians, the present study qualitatively analyzed clinicians' freeform, open-ended comments to derive themes related to the use of digital screening tools, providing a richer picture of how child clinicians in the United States perceive the potential adoption of these tools.We reproduce relevant details about the study methods below.
3.1.1Participants.Participants were recruited from the mailing list for the Society of Clinical Child and Adolescent Psychology division of the American Psychological Association [9].While the survey was open to mental health practitioners who worked/had worked with children of any age, we limited our analysis to respondents who have experience working with children under the age of 5 years (since we specifcally focus on mental health screening tools for young children in this work).This subset contained 25 clinicians; 22 self-identifed as female and 3 as male.
The respondents' age, training, experience, and theoretical outlooks are summarized in Figure 1.Most practitioners held a Doctorate in psychology and had been working as a clinician for over 5 years after earning their highest degree.Their therapy practices were guided primarily by cognitive-behavioral theories, followed by a moderate amount of interpersonal and systems frameworks.Practitioners also reported that they used a developmental lens and applied behavior analysis, and relied heavily on evidence-based assessments to inform evidence-based treatments while also acknowledging their shortcomings.
Clinicians had experience working with children from a wide range of racial and ethnic backgrounds, including Hispanic and/or non-Hispanic white families, African American/Black parents, Asian, South American, and Middle-Eastern immigrant families, and other communities of color.Client families ranged from low or very low socioeconomic status to working-class or middle/upper-middle income.Parents' ages and education levels also varied, and many clinicians saw children from the child welfare or foster care systems, multi-generational homes, and families with international and domestic adoptions.

Survey Context:
The EarlyScreen Application.The clinician survey specifcally asked participants to comment on EarlyScreen, an at-home behavioral screening app for preschool-aged children developed by the authors of [44].Briefy, EarlyScreen is a tabletbased app that predicts the child's risk for broad externalizing disorders, including irritability, temper loss, and ADHD.Children play a game where they are introduced to a baker who seeks their help choosing cakes for customers.Customers provide predetermined positive or negative reactions to the child's choices (see Figure 2), with repeated negative feedback meant to induce frustration.The app records children's facial expressions and bodily movements in response to this frustration and uses this data to predict psychopathological risk.For more details on the screening algorithm, we refer readers to the original publication (i.e., [44]).
Survey participants read a textual description of the EarlyScreen application (reproduced in Appendix A) consisting of its purpose and intended use, how it worked, and the predictive performance reported in [44] before answering questions about their perspectives towards the tool.Methodologically, the survey's approach is based on design probes [87] and provocations [2], utilizing an example screening tool to support refection and elicit reactions from participants.

Survey Procedures.
The survey begins by asking about clinicians' backgrounds and current therapy practices without introducing participants to the EarlyScreen probe.Respondents were then invited to describe the gaps and pain points in their current diagnostic practice, with the survey asking "[w]hat sources of data would you like to have access to as part of your diagnostic process but are often unable to access?".Screening technology was still not mentioned at this stage in order to avoid biasing participant responses and instead focus on better understanding their existing needs.
Finally, respondents were presented with a textual description of the EarlyScreen application (see Appendix A).Based on this description, clinicians were invited to answer the following questions: (i) "What do you like most about the EarlyScreen application?",(ii) "What do you like least about the EarlyScreen application?",(iii) "What potential concerns do you have about using or recommending EarlyScreen in addition to your current practices?",and (iv) "What features would you like to see in EarlyScreen that are not described here?".All questions were optional and no word limits were imposed.

Study 2: Interviewing Parents of Preschool-aged Children
In addition to clinician perspectives from the survey described in Section 3.1, we sought to understand the views of another key stakeholder group towards mental health technologies for young children, viz., parents/family caregivers of preschool-aged children.
To this end, we conducted semi-structured interviews with parents to gather additional data to answer our research question.

3.2.1
Participants.We recruited 26 caregivers through word-ofmouth and online advertisements within the community as well as playgroups/parent support groups in the area.In order to gather a wide range of parent perspectives, we did not explicitly select parents with existing concerns about their child's behavior or mental health or control for parental mental health literacy.Instead, we sought to interview caregivers from diverse families including single-and dual-income households, rural and urban communities, immigrant and student parents, etc. Table 1 describes the participant demographics in more detail.In most cases, one parent (or another caregiver, e.g., a grandparent) was interviewed one-on-one by an author.In one case, both parents of a child were independently eligible and chose to participate together -they are identifed as P4a and P4b.Parents were sometimes interviewed in the presence of the child but the child did not participate in the interview.One parent (P23) was excluded from the analysis due to data recording errors.

3.2.2
The EarlyScreen Application as a Speculative Probe.To maintain consistency with the clinician survey as well as to create a shared understanding for participants with varying levels of familiarity with digital screening tools, we used a video demo of the EarlyScreen application [44] as a speculative probe during our parent interviews.We draw inspiration from both [44] as well as other prior work that uses designed applications and artifacts to evoke users' feedback, thus leveraging the speculative nature of cultural probes [16,60,68].A scenario-based video of EarlyScreen was preferable to using a low-or medium-fdelity prototype of the application as an interactive probe to provoke participants' thoughts for two reasons.First, the direct users of an app like EarlyScreen would Second, video probes could be shared more easily if participants chose to complete the interviews remotely.The one-minute video demo (available as Supplementary Material) described EarlyScreen and how it could be used by parents to help identify whether their child might be at risk for irritability and mood disorders.

Interview Procedures.
All parents provided written informed consent before participating.We started the interview by inviting parents to share general concerns about their child's early development and behavior.We asked about caregivers' biggest worries pertaining to their child ("Do you ever worry about your child?If yes, what is your biggest worry?").Parents were not explicitly probed about mental health concerns, rather, they spoke openly about different concerns and family circumstances.If no behavioral concerns were mentioned, parents were asked a single follow-up question ("Any concerns about their mood or behavior?").As listed in Table 1, 15 of the interviewees ultimately expressed behavioral concerns, including but not limited to defant behavior, temper tantrums, trouble sharing or socializing with peers, separation anxiety, etc. Parents were then asked about their current help-seeking practices ("Where do you look for help to ease these worries?") and the information gaps that caregivers wanted to fll ("If you had a crystal ball to learn something about your child and ease your worries, what would it be?What information will you fnd helpful but just don't have access to?").Screening technologies were not mentioned at this stage and participants were invited to talk about any information they would fnd helpful.
Following this open-ended conversation, participants were presented with a video demo of EarlyScreen as a speculative probe.Parents were then asked if they would use an app like EarlyScreen ("If EarlyScreen was widely available and easy to access, would you use it to test your child at home?Why or why not?") and/or what concerns they had about it ("What worries would you have about using such an app?").Finally, participants were asked what they would like to see in screening applications in general that would better support their parenting experience ("What else would you like to see in apps like EarlyScreen that is not described here that would support you as a parent?").
All interviews were conducted in English with questions phrased at an elementary-school reading level, and participants were excluded if they could not read, speak, or understand English.Interviews took up to 30 minutes to complete and interviewees received a $15 gift card.Interviews were conducted in person or via video conferencing and the audio was recorded and transcribed for further analysis.The study was approved by the Institutional Review Board at the University of Massachusetts Amherst and permission was sought from playgroup facilitators where applicable.

Thematic Analysis.
We analyzed survey responses collected by the authors of [44] as well as transcripts from our parent interviews using an inductive approach to identify salient themes, similar to prior work such as [3] and [26].One author performed rigorous initial (open) coding [20] of the survey and interview data and developed a set of preliminary codes for both sources that were guided by our research question.A codebook was generated using the NVivo qualitative data analysis software [39] and the extracted codes was iteratively refned through discussion among authors.The survey and interview codes were collectively grouped into themes refecting (i) participants' concerns stemming from their lived experience caring for young children, (ii) their perceived utility of screening tools, and (iii) the challenges of integrating such tools into their existing practices.The rationale for deriving themes using codes from both stakeholder groups was to present a complementary and holistic picture of perspectives toward screening tools (while certain individual themes may still relate to a single stakeholder group).The analysis and resulting themes were reviewed by all authors.We illustrate these themes in Section 4 and list sample codes for each theme in Appendix B.

Diferences in Stakeholder
Opinions.While a combined thematic analysis of survey and interview codes ofers a comprehensive understanding of multi-stakeholder perspectives towards app-based screening tools, clinicians and parents may also sometimes have diferent or opposing views.We investigated these diferences by comparing survey and interview codes to distill the implications of some unique and/or opposing viewpoints.We used NVivo to classify each instance of a code, marking whether it was derived from clinician or parent responses.We compared the frequency of occurrence of codes within each group and discussed the codes and fndings among the authors to identify salient topics that are elucidated upon in Section 5.

Design Tensions and Recommendations.
We further analyze our fndings through a design tensions lens [81] in Section 6.1, identifying conficting stakeholder values that must be grappled with when designing pediatric mental health screening tools.Briefy, Tatar's design tensions framework conceptualizes the design process as an optimization, or "goal balancing", endeavor in the presence of a "multiplicity of perspectives" or stakeholder values.Tatar emphasizes that such conficting values cannot often be fully resolved, but "only handled via compromise" [81].Prior HCI research has used the framework, or its variants, to study or synthesize design requirements in a range of domains, including health promotion applications [49], learning analytics dashboards [80], online behavioral therapy experiences [41], and wearable technology to support children with ADHD [19].As designing app-based assessments for scaling behavioral health screening outside clinical settings inherently necessitates balancing several benefts and drawbacks, we found the design tensions framework suited to our analysis.Section 6.1 highlights three major design tensions emerging from our analysis.Finally, we integrated our fndings to ofer recommendations for balancing the identifed design tensions and designing usable and scalable screening technologies in Section 6.2.

Positionality Statement
This work stems from the authors' long-standing research interests in technology and mental health.The authors have previously developed tools for a range of mental health and well-being applications, including for children's behavioral health.Some authors have experienced mental health issues and/or have others they care for who have experienced mental health hardships.Three of the authors have experience working with preschool-aged children and their families in clinical settings as registered behavior technicians or clinical psychology trainees, while one author has experience working in special education classrooms with children with autism spectrum disorder.Three of the authors are parents of children who are preschool-aged or older.Half the authors identify as women and the other half as men.The authors' self-described racial identities include, in alphabetical order, Asian (Bangladeshi origin), Asian/Indian, Black Latina, Latinx of Indigenous (American) and Chinese descent, and white.All authors reside in the United States.

FINDINGS: EXISTING CONCERNS AND HOW SCREENING TOOLS CAN ADDRESS THEM
In this section, we describe our fndings highlighting the lived experience of child mental health practitioners and parents of preschoolaged children in terms of their concerns about children's behavioral health, gaps in current screening practices, and perceived benefts and challenges in deploying and using at-home mental health screening apps to support their clinical or parenting practices.As described in Section 3.3, we fuse insights from both stakeholder groups to derive broad themes highlighting their collective perspectives.Quotes from clinicians are identifed by the pseudonyms C1 through C25 and those from parents by P1 through P25.As a result, parents' help-seeking behavior is largely driven by their intuitions ("if we feel something is of, then we talk to the doctors" (P4b)) as well as their willingness and comfort in approaching others for help ("it's like, I don't know, should I ask for help maybe?"(P12)) instead of the more standardized approaches utilized in physical medicine.P13 also emphasized that the lack of concrete information prevented parents from successfully advocating for their child: "[In many public programs] there's some resources available but generally not enough to go all the way around.And so there's like a gatekeeping process where they basically try to determine who should really get these resources, and navigating that process can be really, really difcult." (P13) In terms of overcoming these challenges, most parents reported reaching out to family and friends, doctors, teachers, or community support programs and organizations to discuss their concerns and obtain more information about their child's development, in addition to accessing information from books or the internet.However, they expressed frustration at being subjected to an information overload ("the amount of information out there is overwhelming and it's hard to parse through it to fnd what's actually useful" (P10)) or not being able to receive confrmation of typical development even if they do not have existing concerns ("for the most part I think it's just the typical kid frustration for them.[But] it'd be good to know, it's always good." (P16)).
4.1.2Limited resources available for tracking and supporting emotional development.In addition to the lack of a social-emotional development scafold to help identify concerns, stakeholders emphasized the paucity of resources for supporting families in navigating existing concerns.P15 described struggling to fnd the right strategies to support her child who was prone to emotional outbursts and a short temper: "I have a pretty good resource connection because I do have an education background.So I developmentally know what should be happening and who I should talk to -doctor-wise or other teachers and behavior specialist-wise, but nothing's quite clicked yet." (P15) While some parents struggled to fnd actionable ideas, others would seek reassurance from healthcare providers as they navigated behavioral concerns ("what I want to hear is like, "hey this is fne . . . it can be fxed by doing this"."(P12)).However, practitioners themselves struggled to accurately diagnose or track children's symptoms over time, with C12 stating, "I often work in settings that have a dearth of resources available for both families and clinicians . . .While there are many free narrowband measures available, there are few reliable broadband measures that give the same level of detail as a BASC or CBCL." (C12).Authors' note: BASC (Behavior Assessment System for Children; [70]) and CBCL (Child Behavior Checklist; [1]) are both commonly used proprietary broadband assessment scales for children.Similarly, clinicians also expressed the need for "measures that can be administered at higher frequency (e.g., daily)" (C2) in order to continually track children's behavior.

4.1.3
Behavioral screening is based on incomplete data.Another common challenge encountered by mental health practitioners working with young children was the limited availability of data required for diagnosis and treatment.Clinicians reported wanting access to clinical interview data from both parents instead of just one (C9) and from biological parents in case of adoption or foster care scenarios (C19).They also expressed a need for school and community data, including classroom observations, teacher interviews, school records, and information about how the child interacts with other adults (C1, C7, C11, C13, C14, C15, C19, C24).Further, the diagnostic process often lacks access to behavioral data (C8, C22) and real-world observations outside the clinic (C10, C17).
Additionally, behavioral diagnoses rely on brief observations of the child in a clinical setting.P1 explains how this misses important context and thus decreases the reliability of diagnosis: "Anyone that is trying to know the kid should not spend a couple of hours in a meeting to understand him.So a couple of days, sometimes a couple of weeks, might get you like, a day-to-day or an overall idea of how these kids behave.But in a meeting, with kids that you do not know -for instance a psychologist that spent like an hour with a kid -the kid will be shy, he does not know her/him.And he will not express his feelings as a grown-up would.It will take weeks." (P1) These limitations mean that even families who are able to access mental health services often fail to receive the highest level of care due to missing data or limited contextual information.

Potential Benefts of Screening Tools
4.2.1 Increasing access to mental health services.Both clinicians and parents felt that app-based, at-home screening tools had the potential to "increase access to care for underserved communities" (C2) and reach numerous families as long as they "have a phone/tablet and WiFi" (C7).Stakeholders appreciated the convenience of use (C2) and noted that app-based screening could be particularly "effective for low socioeconomic status families" (C4) and that it was "easy and convenient enough, and doesn't seem super invasive" (P17).
Additionally, parents valued screening tools as a frst step towards seeking professional mental healthcare.P13 described wanting to test their child at home by saying, "I would say if we gave him one of these [tools] and it was like, "Oh we think your kid, [there's] a high percentage chance he has this", yeah, we'll probably follow up with a doctor, like, pronto.Yes, that would be enough to get action for sure. . . .We'll probably give it a try because we'd be really interested to see." (P13) 4.2.2Augmenting existing data and clinical practices.Clinicians appreciated that app-based screening tools were able to "provide a diferent type of information to that [clinicians are] able to collect" (C17), and specifcally highlighted "the passive ability to gather information" (C15) as a key beneft.Parents also felt that such tools could supplement regular healthcare appointments by providing additional insight into the child's development: "I just think [screening tools] are something good because maybe with [these tools] you can get some preliminary information.Which maybe a doctor can't identify when you are at a checkup.So it can be something which is in addition to a routine checkup of a child." (P5) Parents similarly noted that "having a huge database of information to compare one person to, could potentially result in a more reliable prediction or diagnosis than just one health care provider" (P6) and were interested in using screening tools "in conjunction with, like, an actual provider" (P6).
Clinicians also listed additional data streams that could be collected, or functions that could be supported, via app-based screening tools.C7 wanted to see future tools that could provide "the ability [for] the mental health professional to observe the child while engaged in the task and/or the ability to provide live support to the parent/caregiver during/after the task" (C7).Others wanted to use such tools to measure parent-child interaction (C1, C2), caregivers' frustration tolerance (C12), or longitudinal progress (C6, C7).Parents and clinicians also felt that game-based or other naturalistic modes of screening would be particularly benefcial for children who would feel uncomfortable with traditional assessments.C12 explained, "It is sometimes difcult to observe disruptive behaviors when children come to a clinic and meet individuals with whom they are unfamiliar.This application would potentially remove demand characteristics at play within a clinic and capture the child behaving in their natural environment." (C12)

Making
Parents agreed that such tools would be very helpful to children, "more than the formal appointments that make kids sometimes scared" (P11).P13 also said his child would "not sit there and have a long detailed conversation with you about how he feels about these things.He's quickly going to get bored" (P13), and engaging, naturalistic assessments would be a useful alternative to test such children.

4.2.4
Supporting the parenting experience.Parents articulated how screening tools would support them through their own struggles by taking some pressure of of the parenting experience.P15 explained, "As a baby, for [child], I did [an app with] diferent things you can do with your baby for the frst year . . .And I found that particularly helpful for me because I had my own mental health stuf going on at that time with postpartum.So not having to think about it but felt like, "All right, still getting stuf done, I'm still doing a good job.I'm still getting answers".I think that's where [screening apps] would fall under".(P15) In addition to letting parents automatically engage in the screening process without explicit efort, screening apps could also help address the "when-to-worry" problem ("if there was something related to behavior, which might be unusual, which we observe . . .And if that is something which is widely observed -we would probably get to know it from such [apps].(P5)).Screening tools can also provide parents additional insight in terms of "something to go to the doctor with . . .useful information for the follow-up" (P13).

Challenges in Deploying Screening Tools
4.3.1 Integrating with traditional mental health services and enabling opportunities for follow-up.While stakeholders valued athome screening tools as complementary to existing mental health services, some participants also expressed concerns that such tools might "take away from scientifcally validated in-person assessments in sterile clinic environment" (C1) or "override clinical training/judgment" (C9).Over-reliance on app-based tools could also disproportionately afect populations who are less likely to be able to access traditional services.P6 commented, "I'm worried that technology which is cheap and broadly available is disproportionately used by lower-income individuals and individuals that have had difculty accessing traditional services, in lieu of improving accessibility to those services." (P6) C10 similarly felt that "a low-income or otherwise busy family who cannot make a diagnostic interview is also unlikely to follow-up for care" (C10) and said they would want to "see data that suggests the app actually leads to clinical follow-up in high-risk populations" (C10).Others suggested that the app itself could include a "treatment related intervention" (C19) or actionable next steps for parents on "what to do about it, if there's problems that come up" (P18).

Building stakeholder trust in screening technologies.
Another challenge in deploying screening tools for young children lies in establishing stakeholder trust in such technologies.Some participants shared their distrust of mental health technology due to negative perceptions of existing commercial products ("I've heard a lot of horror stories about [redacted online therapy service provider].And like, it's honestly kind of scared me of [of] this sort of tech in general" (P6)).Parents were also concerned about the implications of apps inferring sensitive characteristics of their child's health, and how it might impact their future.P6 added, "I'm worried that if there's something forming psychological profles of children, that's going to follow them their entire lives.I could imagine that it will leak into the education systems, like, somebody who tests poorly on this device is going to get red-shirted." (P6) Stakeholders also had questions about child and parental consent ("Will children always seek parental consent before using the app?" (C10)) and concerns about the lack of situational context (C4, C15) as well as the accuracy of screening tools ("machine learning can be imperfect, and can misidentify or miss at-risk children" (C10)).Moreover, parents stated that they would need explanations for the tool's outcomes to be able to take it seriously: "For any app to gain trust, like, if I give him a game and the only thing the app says is "something is wrong with your child", I wouldn't accept that.Make me understand why you are right." (P4b) Other parents echoed this sentiment, saying feedback in layperson terms ("something like, you know, "your child spent a particularly long time doing this, or reacted this way" . . .whatever you can do to communicate what it is you think you found" (P13)) would help build trust in the tool.

4.3.3
Minimizing potential for harm.Lastly, participants emphasized the need to account for safety, privacy, and ethical considerations when deploying screening technologies.In terms of child safety, parents were worried that tools could be "too involved" (P13) or misleading for the child ("it might be hard for me to explain [outcomes] if there is no lesson in it" (P3)).Other parents were wary of introducing or increasing their child's screen time or device usage (multiple parents), or enabling unrestricted access to the Internet (P15), by using app-based tools.Stakeholders also felt that children might eventually get bored or frustrated with repeated use of screening tools ("I think that you'd have to change [the game] up a lot because I think they would get bored" (P14)), or that they may cause undue stress for children and/or parents.P1 elucidated the latter point, saying, "I think in the US, there is too much of a focus on diagnosing kids, categorizing them . . .I don't think that is necessary . . .And even if the kids start showing any slight symptoms, of any issue, they try to always put way too much pressure on the parents and on [the child].That might not be, uh, motivating for people to go seek help".(P1) Other parents seconded this, while P3 added that "if [the tool] gives like a lot of false positives, then it just becomes more stressful" (P3).In addition to ensuring that tools are accurate, engaging, and safe for children to use, developers also need to focus on data storage and privacy protections.Stakeholders emphasized the need for "really rigorous data protections, including not harvesting and storing anything that isn't totally vital" (P6) and guaranteeing "confdentiality" (C2), and "privacy" (C4, C13, P7).

FINDINGS: DIFFERENCES IN PERSPECTIVES ACROSS STAKEHOLDER GROUPS
While Section 4 highlights the broad collective perspectives of various stakeholders toward app-based screening tools and illustrates how clinicians and parents largely agree on many facets, it is important to also understand where the two groups may have opposing views.Below, we describe some key diferences emerging from our analysis of clinician and parent responses.

Intentionally Inducing Frustration
A characteristic feature of the EarlyScreen app [44], which has been used as a design probe in this work, is that it induces frustration in children in order to record how they respond to negative emotional stimuli.While this approach may not be a necessary component of all app-based screening tools, it is worth discussing stakeholders' views toward intentionally inducing frustration.Clinicians in particular were wary of this practice, noting concerns about "managing a child's reaction to the frustration scenario remotely or leaving the parent/caregiver to deal with whatever happens without live support" (C7).While clinicians acknowledged that "this [frustration] often happens naturally and this certainly occurs when implementing practices like exposure with response prevention" (C12) in routine behavioral healthcare settings, they were nevertheless concerned that "parents may be very wary of using [EarlyScreen] as it induces frustration without clinical supervision" (C8).Interestingly, parents did not share this concern -while they noted that repeated frustration could result in children not completing or repeating the task frequently (P15, P17, P24), parents were highly interested in the possibility of understanding how their child would react to deliberate and controlled frustration.For instance, P4b felt that this approach "has a metric" as opposed to "right now, where if he wants something and we don't give him that, then comes the frustration or anger or tantrum" (P4b).P15, who mentioned concerns about her child's irritability, said, "I like the idea of it purposely being frustrating to see how they would react, and feel like [child] would be a great candidate for that".(P15) Parents further viewed the frustration-inducing game as an opportunity to discuss emotions with children and help them develop emotion regulation capabilities.P22 mentioned that such tools can "help [child] learn or get the feelings out" (P22), while P9 said, "Children tend to be more frustrated and angry when things don't work out in their support . . .I would like to create a feeling of patience in them.So if this app helps reach that, and helps in assessing my child, that will be great".(P9)

Apps and Screen-based Interaction Format
Parents were overwhelmingly more apprehensive about the screenbased assessment medium in EarlyScreen, with over half of them raising concerns about screen time, device usage, or potentially unrestricted access to the Internet while completing the screening.P8 said she probably wouldn't use EarlyScreen, adding: "Me and a lot of the parents I know are super against screen time as much as possible.So we really limit it, even if it's like a game or like [educational content] and stuf, I prefer books and paper media".(P8) P3 shared these concerns about screen time and further added, "I'm okay, or more okay, with TV and [gaming] consoles.I'm less comfortable with kids playing with phones and iPads because those devices access the internet very easily.So they are more exposed to potentially harmful information.Also, there are a lot of micro-transactions in iPad games and iPhone games.It's very easy for kids to purchase something by accident.And also I think a lot of the iPad games and the phone games are just designed to be very addictive compared to TV shows".(P3) Parents also felt that the novelty of screen-based tasks might encourage children to ask to do them more often than is necessary, especially if their screen time was usually very limited (P2, P4, P9, P11, P21).Multiple parents (P3, P8, P9) said they would be more amenable to home-based screening if it did not involve a screen, while one (P7) wanted more data on how using the app might impact children in the long term.
In contrast, none of the clinicians surveyed expressed any concerns about screen/device exposure.Clinicians were more excited about the gamifed format of the assessment, the potential ease of installation and use, and the increased access to screening aforded by this format (see Section 4.2.1).

Integration Within Healthcare Settings
Another important distinction between clinicians and parents was their perspective on the integration of app-based screening tools into the traditional mental health care ecosystem.While both groups cautioned against an over-reliance on such tools at the cost of clinical or parental judgment or improving access to health services, parents still saw value in at-home screening tools that were situated outside healthcare settings.P16 explained how screening tools were valuable for making sure there was indeed nothing to be concerned about ("Better to know more information than not, I think." (P16)).
Parents also felt that at-home screening tools would enable them to track children's socio-emotional development relative to peers, help identify concerns early and ascertain the level of risk associated with them, and empower them in communicating with health care providers.P20 hoped they would provide "a little more info before you go to see someone outside of the home" (P20).
Relatedly, while clinicians were concerned about the potential lack of follow-up, most parents said that they would immediately seek help from a medical professional if an at-home assessment identifed a possible disorder or concern: "I would say if we gave him one of these things and it was like, "Oh we think your kid -a high percentage chance he has this", we'll probably follow up with a doctor like pronto.Yes, that would be enough to get action for sure".(P13)

Design Tensions in Early Childhood Mental Health Screening Tools
As evidenced by our fndings in Sections 4 and 5, designing appbased mental health screening tools for children inherently calls for balancing several key considerations from diferent stakeholders.We examine these factors through the lens of Tatar's design tensions framework [81], identifying three conficting pairs of requirements that are elucidated upon here.Note that these tensions should not be viewed as pitting stakeholder groups against each other (parents vs. clinicians) but instead as an expansive set of design considerations emerging from both agreements and disagreements in a value-sensitive multi-stakeholder analysis.We make design recommendations for balancing these design tensions in Section 6.2.The design tensions and recommendations are summarized in Table 2.
6.1.1Need for rich observational data vs. limiting burden and exposure.Both mental health practitioners and parents expressed a desire for screening tools to capture behavioral data in a variety of contexts and often longitudinally.This has three clear benefts -frst, families can obtain more reliable and trustworthy predictions through repeat testing, minimizing the impact of temporary contextual factors on assessment outcomes.Second, it allows both parents and clinicians to monitor behavior progression over time and in response to interventions.Finally, clinicians can obtain rich, naturalistic data that would not be available in an outpatient clinical setting.In addition to longitudinal data, parents expressed an interest in observing children's response to controlled frustration in EarlyScreen, motivating future tools that can deliver regulated measures of stimuli and record the resulting behavior.However, the repeated use of screening tools comes at a cost.Parents may want to avoid exposing their child to tools repeatedly (e.g., in order to minimize screen time) and families may fnd recurring assessments time-consuming or burdensome.Longitudinal use also requires considering practice efects and keeping the child engaged over time.Inducing negative afect to observe responses necessitates deliberation on the quantity and quality of stimulus to avoid adverse efects for the child and to ensure that families can support any outcomes in the moment without clinical intervention.Screening tools therefore need to balance the reliability and utility of data against usability concerns.
6.1.2Frictionless screening vs. over-reliance on tools.App-based and gamifed screening tools can be more child-friendly than traditional methods of assessment, especially for children who fear doctors, get easily distracted, or are reserved when interacting with strangers.Similarly, they can increase parental convenience and empower parents by providing useful information about their child's wellbeing and their development relative to peers [88].Lastly, athome screening tools can provide parents with a frst-step diagnosis, spur them toward seeking clinical follow-up, and help communicate their concerns to a clinician.
Nevertheless, both clinicians and parents expressed concerns about overly relying on the outcomes of screening tools.It can be detrimental to the child's health if families are unable to follow up on screening outcomes and get the required help.Screening technologies should also not result in loss of parental intuition or reliance on their experiential knowledge, and providers should not let them override their clinical judgment [10].Tools should instead provide stakeholders with clear pathways to support a complementary caregiving process [45].
6.1.3Scaling beyond the clinic vs. minimizing contamination and harm.Screening tools can dramatically reduce barriers to obtaining a behavioral diagnosis for many families, eliminating waiting times and making screening more afordable and accessible [8].However, they may potentially take away from the sterile examination setting of a clinic, making screening subject to a wide range of factors (e.g., child's mood and surroundings, parental infuence, presence of siblings and pets, device variability, etc.) that are difcult to standardize across children and families in home settings [6,91].Furthermore, screening tools would need to account for privacy and data storage considerations [33] -traditional healthcare services often have existing mechanisms in place to address these concerns.

Design Implications and Recommendations
Based on our fndings and the design tensions identifed in the previous subsection, we now discuss design opportunities for building scalable mental health screening technologies for young children.

Explore Novel Assessment Mediums.
A major concern for parents in potentially using app-based screening tools was the addition of screen time into their children's lives.We call upon designers to address this by (i) exploring app-based assessment modalities that avoid media exposure or utilize passive sensing (e.g., collect to empower parents and support their caregiving practices, instead of promoting an over-reliance on predicted outcomes at the cost of parental or clinical judgment [45].Mental health screening tools, in particular, should delineate clear next steps, including (i) informing parents of potential risks and normative comparisons across peers [62], (ii) providing consolidated information to help them articulate concerns when seeking help [30], (iii) connecting them to sources of support within and outside the traditional behavioral care system, (iv) facilitating relationship-building and information sharing across care teams [71], and (v) providing reassurance at a time of distress and supporting them in navigating their own mental health [4].

Controlling for Context and Measuring
Variability.Screening tools that collect assessment data outside controlled clinical settings also need to account for contextual variability and potential contamination due to environmental factors.One approach to limit variability is to observe the child during a particular, focused activity (such as the frustration-inducing game in EarlyScreen).Screening algorithms can also explicitly record and utilize variability such as behavioral context [7] and parent-child interactions [36] in order to make predictions.This would also serve to provide additional metrics for clinicians who subsequently review the data to assess the child.

Contributions to HCI and Child Mental Health Research
As described in Section 2, there has been a growing interest in the HCI and digital health communities to augment traditional mental health services using behavioral screening tools to identify problem behaviors.However, building useful and scalable screening tools, especially for young children, necessitates understanding the unmet needs in the current mental health care landscape from the perspective of various participants in the caregiving process as well as determining to what extent these needs can be fulflled by at-home screening technologies.
Our work is the frst to investigate these questions, focusing specifcally on app-based, at-home screening tools for behavioral screening in preschool-aged children.It makes the following contributions to the broader HCI community interested in developing screening tools for young children: (i) we present empirical fndings on the attitudes of mental health practitioners and parents concerning at-home screening tools, (ii) we highlight nuances in stakeholder views by disentangling the agreements, conficting priorities, and overarching design tensions emerging from our multi-stakeholder inquiry, and (iii) we ofer concrete recommendations for designing future child mental health screening tools, distilling insights from both our own analyses as well as prior research.

Limitations and Future Work
Our work examined the perspectives of two key stakeholders in the child mental health care system -clinicians and parentsto inform the design of scalable screening technologies for children.Our studies focused on child and adolescent mental health practitioners across the United States as well as a predominantly White/European-American/Asian sample of parents of preschoolaged children.As such, further research is required to investigate whether our fndings generalize to other geographical and demographic contexts.The perspectives of children themselves as users of these screening tools were also not examined in this work due to the nature of our inquiry.We encourage future studies to involve children as stakeholders in the design of screening technologies through age-appropriate participatory design approaches [23,38,40].Lastly, our decision to use EarlyScreen as a speculative probe may have also biased participant responses towards a particular class of screening tools (app-based, data-driven assessments with active participation).However, we felt it was necessary to utilize an example screening tool as a provocation in order to both elicit feedback of a speculative nature [2,16] and to familiarize participants who may not have used or even come across such tools before.This was confrmed during the parent interviews, where most parents reported being unfamiliar with at-home behavioral screening technology.Future research could elicit stakeholder feedback on other kinds of screening tools, and triangulate them against the fndings presented here.

CONCLUSION
This work sought to understand the perspectives of mental health practitioners and parents of preschool-aged children towards digital mental health screening technologies.We reported on a survey of clinicians and interviews with parents in the United States, focusing on uncovering the barriers they currently face in identifying and seeking help for behavioral issues in young children and how screening tools could support their caregiving practices.Our work uncovered three key tensions relevant to the design of screening tools: (i) the need for rich observational data vs. limiting burden and exposure, (ii) frictionless screening vs. over-reliance on tools, and (iii) scaling beyond the clinic vs. minimizing contamination and harm.Finally, we ofered recommendations for developers of future tools based on insights from multiple stakeholders and delineated opportunities for future research.
Below is a reproduction of the description of EarlyScreen that was presented to mental health practitioners in the survey described in Section 3.1.
We will now describe EarlyScreen, a prototype digital mental health screening tool that has been designed to help parents, caregivers, and clinicians identify specifc disorders common in early childhood.EarlyScreen is an additional at-home tool meant to complement existing clinical practices.
One possible use for EarlyScreen is to make the diagnostic intake procedure more efcient and convenient for clinicians and families, particularly low-resource, low-income families.For example, EarlyScreen could be used with a family on the waitlist who urgently needs to pivot to treatment, a low-income family who has difculty attending multiple in-person intake appointments, or a clinician looking for diagnostic information from a modality other than questionnaires or observation.It could also be used as an additional tool to track changes over the course of therapy.
• EarlyScreen is a smartphone-or tablet-based "game" that can be played by preschool-aged children that is modeled after existing iPad games.• The game will induce frustration in children using a clinically-validated paradigm and record facial videos during the process using the tablet's front camera.(In the current iteration, the game involves providing negative feedback on children's choices -see link for more information).• Facial expressions and head and eye movements are extracted from the captured video and used by machine learning models to predict: neural activation within the prefrontal cortex, a region of the brain involved in emotion regulation.-the child's score on a series of clinically-validated questionnaires to screen for externalizing disorders and ADHD.
In lab-based tests with 76 participants, a prototype of EarlyScreen could correctly identify 75% of the children exhibiting abnormally low neural activation in the prefrontal cortex during frustration and 77% of children exhibiting normal levels of neural activation.
(For context, the Child Behavior Checklist -a wellvalidated screening tool -correctly identifes 66% of children with exhibiting problematic behavior).
EarlyScreen could also correctly identify 72% of children who scored above clinical thresholds on the CBCL Externalizing disorders, MAP-DB temper loss, and ADHD Inattention and Hyperactivity scales and 76% of children who were below the clinical thresholds.

Figure 1 :
Figure 1: Age, training, experience, and therapy approaches of clinicians who participated in the survey.

Figure 2 :
Figure 2: The EarlyScreen app [44] consists of a bakery game where children select cakes for each customer.The customers provide predetermined positive or negative feedback.Images courtesy of original authors.
be young children, who were not part of this study since we chose to focus on the perspectives of adult stakeholders at this point.
Parents lack scafolding for identifying concerns and seeking help.A common theme that emerged from discussions with parents, both those with and without specifc concerns about their child's behavior, was not knowing what to expect in terms of behavioral development at a certain age and when to seek help.P4b described the need for a developmental equivalent to growth charts, saying, "As a parent, you will be given those developmental charts.So developmentally, like by three months, what he is expected to do.By one year, he will be able to stand with support . . .For psychological or mental things, we don't have that chart.For frustration, social behavior, or tantrums -those are not well documented.There is nothing about what to expect at what age for psychological development." (P4b) screening more frictionless for children.Another important beneft of app-based screening tools, stakeholders noted, was that they could make mental health screening more engaging, natural, and frictionless for young children.Participants thought game-based screening would be "creative and engaging for young children" (C6), especially since tablet-based games with colors, matching objects, pretend play, and cartoon characters would be relevant to children's existing interests ("I think [child would] love it.It's adorable and . . . he watches Daniel Tiger [children's show].And he loves the Baker Aker [character].So it would totally be in his wheelhouse" (P10)).

Table 2 :
Design Tensions and Recommendations for Building Mental Health Screening Tools for Children 6.2.4 Enable Pathways for Future Care.Prior work has discussed the importance of child tracking technologies to use the data they

Table 3
lists a few sample codes associated with each theme, grouped by the topics explored in Section 4.

Table 3 :
Sample Theme-Code Mapping