Advancing pediatric mental health
intervention through multimodal
digital phenotyping

Chapter 1 established technical feasibility and research context for ChAMP, a smartphone-based platform for collecting multimodal signals associated with childhood anxiety, ADHD, and depression during brief structured behavioral tasks. In 101 children ages four to eight, machine learning classifiers achieved 70 to 73 percent balanced accuracy, comparable to established parent-report instruments. The chapter also introduced an ethical framework for deploying technologies such as ChAMP, offering guidance for real-world implementation.

Chapter 2 asked which components of a maximally instrumented wearable assessment are actually necessary for deployment. In a subset of the same ChAMP cohort, body-region-restricted classifiers reached AUCs of 0.61 to 0.92 across five diagnostic targets (with discrimination significant for three after correction), and identified up to 1.2 to 3.3-fold more clinician-confirmed cases than parent-report CBCL alone. An implementation burden score framework formalized the burden-versus-performance tradeoff and shows near-optimal accuracy is reachable at substantially reduced sensor load. The chapter also reported the first validation of trapezius-derived far-field ECG in any population: under a quality gate met by 73 to 77 percent of recordings during low-to-moderate motion, trapezius-derived heart rate matched chest ECG within 1.9 to 3.3 bpm across four low-to-moderate-motion task contexts (and 5.4 bpm during active play). The trapezius was presented as an emerging sensing location for pediatric populations, less invasive than chest electrodes and more reliable than wrist photoplethysmography (PPG).

Chapter 3 investigated the family-facing experiences and infrastructure needed to turn screening results into care. In a survey of 290 caregivers (251 concerned caregivers contributing to the treatment-seeking analysis), mental health literacy predicted professional treatment-seeking after controlling for personal mental health diagnosis and seven demographic covariates. Information-seeking source breadth and clinical search vocabulary emerged as the strongest independent predictors of taking action: what families most lack is often not information but the frameworks to act on it. Convergent evidence from customer discovery with 250+ stakeholders and an IRB-covered prototype evaluation across caregivers, pediatricians, and therapists confirmed that families want concrete coping strategies, clear next steps, and interventions they can personalize and easily access. The chapter reported both the caregiver help-seeking gaps and the early-stage co-design built to address them.