Grant Number: 1 R01 MH123831-01.
Co-PIs: Leanne Tamm, Ph.D. & Jeff Epstein, PhD
Collaborators: Jon Dudley, Ph.D., Gowtham Atluri, Ph.D., Mekibib Atluri, Ph.D., Sarah Karalunas, Ph.D., Bonnie Nagel, Ph.D., & BJ Casey, Ph.D.
Attention-Deficit/Hyperactivity Disorder (ADHD) is a highly heterogeneous disorder, with multifactorial etiological risk factors, diverse expressions of symptoms, comorbidities, and long-term trajectories. An approach to parsing such heterogeneity is to move beyond symptom ratings toward clinically meaningful phenotypic measures that have well-theorized relations with neurobiological systems. This approach serves as the basis of the NIH Research Domain Criteria (RDoC) framework. In the proposed study, we will explore attention in an attempt to understand heterogeneity within children with ADHD. Reaction time variability (RTV), an index of attention, is the cognitive correlate that typically demonstrates the largest effect size when comparing ADHD to non-ADHD children. However, while RTV is considered a robust correlate of ADHD, its etiology is unclear and individuals with ADHD themselves vary considerably on indices of RTV. Thus, first establishing the neurobiological basis for RTV and then exploring if it can be used to understand heterogeneity in ADHD is critical. The Adolescent Brain Cognitive Development (ABCD) study provides an unparalleled opportunity to examine disordered attention, as indicated by RTV, in a large sample of children recruited at ages 9 to 10 and followed longitudinally. ABCD measures include attentional tasks, diagnostic interviews, and extensive neuroimaging. At baseline, 1079 children in ABCD met diagnostic criteria for ADHD. We propose to utilize machine learning to explore the neurobiological basis of RTV using the entire ABCD neuroimaging sample (n=9,598). We will also explore heterogeneity within ADHD by identifying groups of individuals diagnosed with ADHD who are characterized by unique RTV and neuroimaging profiles. To establish the validity of these profiles, we will examine their association with functioning. Machine learning focuses on learning statistical functions from multidimensional data sets to make generalizable predictions about individuals; it allows for inferences at the level of the individual and is sensitive to subtly distributed differences. Thus, it is an ideal approach for deriving subject-level biomarkers. The first aim is to determine which neuroimaging data are associated with each reaction time variable derived from Gaussian, ex-Gaussian, and drift diffusion models. The second aim is to explore corresponding developmental trends in RTV and neuroimaging data. The third aim is to a) identify groups of ADHD subjects with similar attentional profiles and, b) explore the neurobiological signature of these attentional profiles using the data we derived in aim 1. The fourth aim is to examine clinical correlates of empirically-determined attentional profiles. Conceivably, identifying mechanistic biomarkers of disordered attention reflected by RTV could refine pharmacological, cognitive, and behavioral interventions; this could lead to a higher probability of success for treatments directed toward that particular mechanism for individuals within specific ADHD subgroups. This work could also be relevant for disordered attention in other disorders characterized by high levels of RTV (e.g., Autism).