Using Wearable Devices and Machine Learning to Forecast Preschool Tantrums and Identify Clinically Significant Variants. - Project Summary
Mood and behavior problems emerging in the first few years of life often persist across later developmental
stages and into adulthood, resulting in significant impairment and societal costs. However, the emerging signs
of psychopathology are difficult to differentiate from normative misbehavior in early childhood, creating a “when
to worry” problem for caregivers and providers. Specifically, the cardinal behavioral manifestation of early
psychopathology, the temper tantrum (e.g., screaming, stamping, hitting), is both a transdiagnostic symptom of
myriad disorders and a normative response to frustration young children commonly exhibit. It is unknown why
and how clinically significant vs. normative tantrums differ due to a paucity of research capturing the complex,
real-time, bio-behavioral changes occurring within both the child and caregiver, prior to and during tantrums.
Research investigating the characteristics of tantrums occurring in the home environment, at multiple levels of
analysis, has the potential to differentiate clinical vs. normative tantrum variants, and identify a precursor phase
to tantrums that could be translated into future interventions. The Specific Aims of the proposed study are to
discriminate children with and without psychopathology based on the characteristics of their tantrums, and
accurately forecast future tantrums using real-time data. To accomplish these aims, the study team has
developed and successfully piloted a custom smart-watch app designed to precisely denote the onset and
offset of tantrums in real time and synchronize with an array of wearable and contactless devices measuring
heart rate, respiration, movement, and changes in vocal features. Sixty caregiver-child dyads, 50% of whom
meet criteria for a DSM 5 disorder, will be recruited. Tantrums and bio-behavioral signals will be continuously
recorded in the home for one month as caregivers and children live their normal lives. Conventional statistical
modeling and cutting-edge machine learning will be used to classify the presence or absence of
psychopathology in children, predict the severity level of the following day’s tantrums, and anticipate an
individual tantrum before it occurs. This project, if successful, would produce first-of-its-kind data yielding a
new understanding of the complex temporal and bio-behavioral processes underlying clinical vs. normative
tantrums and algorithms designed to predict tantrums before they occur. These products are potentially highly
significant as they will allow the field to pivot to developing next-generation, home-based, automated systems
to assist in diagnosing and treating mental illness earlier in the lifespan.
