Objective Measurement of Hyperactivity in Children: LemurDx Project

Globally, Attention-Deficit Hyperactivity Disorder (ADHD) affects approximately 5% children and adolescents [ref]. ADHD is a neurodevelopmental syndrome that often leads to increased inattentiveness, impulsivity, and hyperactivity. Children with ADHD start showing these signs as early as four years of age. In school-age children, 55% of all ADHD cases show hyperactivity symptoms [ref]. The current standard of measurement of hyperactivity in children depends on subjective reports via questionnaires from parents or teachers. These questionnaires are convenient as they save time, money, and effort. However, research has also shown very low inter-rater reliability for these surveys between parents and teachers [ref]. The inherent subjectivity of these tests and physicians' lack of contextual awareness often lead to misdiagnoses. This is problematic as overdiagnosis leads to unnecessary treatment, and underdiagnosis can lead to delayed treatment. Thus, there is a need to add some objectivity to the diagnosis and measurement of hyperactivity.

Our initial step was to objectively measure hyperactivity using passive mobile sensing ([J.5] LemurDx). In collaboration with the University of Pittsburgh Medical Center, we collected data from 61 children (25 diagnosed with hyperactivity) who wore a smartwatch for up to 7 days without altering their daily routines. However, in such an unconstrained setting, simply quantifying children's motion levels is not sufficient due to the lack of contextual understanding; for instance, children may exhibit high levels of movement during exercise regardless of their hyperactivity status. To address this, we developed models that estimate the child's activity context using multimodal sensor data, which serves as prior information to assess hyperactivity severity based on motion levels. Our analysis demonstrated that this context-aware filtering approach effectively quantifies children's hyperactivity. Additionally, we enhanced the sensing pipeline to generate a risk score in real time, correlating with the child’s current hyperactivity level based on the incoming data. We conducted interviews with doctors and designed a user interface to support their diagnosis, which is currently under development.

We plan to deploy a real-time monitoring system to assist parents or caregivers of children with hyperactivity. Based on the risk score of LemurDx, this assistant will notify parents when specific patterns indicate a potential onset of heightened hyperactivity. Recognizing the challenges parents and caregivers face, such as predicting and managing hyperactive episodes and balancing these demands with daily responsibilities, this assistant will equip them with actionable insights derived from real-time data, enabling proactive management of hyperactivity. Here, to enhance the model's accuracy post-deployment, we intend to implement a feature that allows parents and caregivers to provide feedback or additional contextual information about the child’s activities. This assistant is designed to foster a more supportive environment that promotes the child's long-term development.

We thank Sam Shaaban from NuRelm, Inc. and his team for their support in advancing the deployment of our system.