296 - Improving Under 12 Pediatric Mental Health Screening By Adding Sensor Data and Offering Immediately Actionable Resources

Publication Number: 3285.296

Ellen w. McGinnis, Wake Forest School of Medicine of Wake Forest Baptist Medical Center, Winston Salem, NC, United States; Bryn C. Loftness, Vermont Complex Systems Institute at the University of Vermont, Burlington, VT, United States; G. James O'Connor, Wake Forest School of Medicine of Wake Forest Baptist Medical Center, Winston-Salem, NC, United States; Jenna G. Cohen, Robert Larner, M.D., College of Medicine at the University of Vermont, Glastonbury, CT, United States; Josh Cherian, Wake Forest School of Medicine of Wake Forest Baptist Medical Center, Winston Salem, NC, United States; Ryan S.. McGinnis, Wake Forest School of Medicine of Wake Forest Baptist Medical Center, Winston-Salem, NC, United States

Background: The American Academy of Pediatrics now recommends universal mental health screening for children to address the escalating pediatric mental health crisis. Current survey-based methods often fail to accurately identify problems due to children's limited ability to articulate symptoms, the unobservable nature of many conditions, and low caregiver mental health literacy. Even after identification, access to evidence-based interventions remains limited-fewer than 30% of diagnosed children receive treatment in many states. Barriers such as long waitlists, economic hardship, stigma, and low mental health literacy contribute to delays, with families often adopting a "wait and see" approach that can worsen symptoms and outcomes.

Objective: To address these gaps, we developed two innovations supported by NIMH (K23-123031) and NSF (2046440):(A) a biobehavioral digital health technology leveraging wearable sensors and machine learning to objectively identify phenotypes of anxiety, depression, and ADHD; and (B) a child mental health toolkit featuring brief, tailored, and actionable digital flashcards co-designed with families and providers to manage common symptoms at home.

Design/Methods: In a sample of 103 children (ages 4-8) during lab visits, we administered brief (7-minute) mood induction tasks while children wore sensors. Caregivers completed gold-standard diagnostic interviews and surveys (e.g., CBCL). In a separate study, 48 caregivers of children (ages 4-11) who recently attended an annual well-visit with elevated PSC-17 scores completed a survey assessing willingness to use evidence-based strategies at home.

Results: Sensor-based models demonstrated strong classification accuracy (AUC = .78-.93) and consistently outperformed CBCL, which under-detected symptoms. Combining sensor data with CBCL attenuated caregiver reporting biases. Caregivers expressed high willingness to implement evidence-based strategies across symptom types, with personalization emerging as a key factor for optimal care.

Conclusion(s): Findings highlight the potential for scalable, family-centered solutions embedded in primary care that integrate objective sensor data with actionable resources to improve early identification and management of pediatric mental health concerns.

Detected Pediatric Mental Health Diagnoses by Assessment Type
Child mental health diagnoses as assessment via gold-standard semi-structured interview (green) compared to sensor-based ML models (purple) and caregiver-report survey (pink) for child ADHD, Anxiety, Depression, Internalizing (Anxiety and/or Depression) and Transdiagnostic (Any Diagnosis).

Patient Caregivers Rate their Willingness to Try Evidence Based Strategies to Manage Child Mental Health
Caregivers of pediatric patients (N=48) in primary care rate 50 evidence based strategies on how willing they are to try them at home to help manage their child's mental health problems. Example of 13 strategies shown here by child problem type (internalizing (blue) vs externalizing (red)).