An Examination of Computational Learning Mechanisms Underlying Aberrant Food Approach in Youth with ARFID

/ABSTRACT The K23 application proposes to examine learning and neurocognitive mechanisms underlying food approach in avoidant/restrictive food intake disorder (ARFID) using computational methods and will position the applicant, Marita Cooper, Ph.D., to transition to research independence with expertise using computational modeling to examine mechanisms of restrictive eating disorders (ED) in youth. ARFID is the most prevalent ED in childhood, with sequelae including malnutrition, delayed growth, cardiac complications, and death. Early data suggest youth with ARFID exhibit executive functioning deficits, including weak central coherence and poor response inhibition. These data underly the hypothesis that youth with ARFID may be slow to learn food approach, impacting food intake, and requiring more exposure to novel foods for learning to occur. Knowledge of neurocognition and learning in ARFID is in its infancy, yet computational modeling offers an innovative approach to probe underlying processes and identify target mechanisms related to aberrant food approach. Two approaches with utility in other EDs, active inference and reinforcement learning, have not been applied to ARFID. The study will examine learning mechanisms and neurocognition of aberrant food approach in ARFID. We will recruit 99 youth (66 with ARFID, 33 controls) ages 8-18, matched on age and sex. Participants will complete a three-armed bandit task, assessing learning mechanisms (via food and neutral stimuli), and a meal-based buffet task assessing food approach (macronutrient and caloric intake). We will assess neurocognition, ED symptoms, and approach/ avoidance. Aim 1 hypothesizes that youth with ARFID will exhibit poorer performance (under both neutral and food conditions) than healthy controls and that worse performance will relate to overall intake during the buffet task. Aim 2 follows participants naturalistically, repeating assessments at 6- and 12-month follow-up. We will examine whether baseline performance predicts improvement in ARFID symptoms at follow-up. Aim 3 will compare whether active inference or reinforcement learning models best fit participant learning behavior. The project will be an important major step in developing a data-driven model of ARFID, providing critical information about potential drivers of aberrant food approach. The proposed project will support expert mentorship and training for Dr. Cooper including 1) learning and neurocognitive development in youth; 2) conducting and managing longitudinal research in clinical samples; and 3) practical skills in computational modeling transferrable to future research. The resources of Children’s Hospital of Philadelphia and University of Pennsylvania and an expert team of mentors (with expertise in mechanisms of ARFID/restrictive ED, development, clinical research, and computational modeling) provide an outstanding context to launch Dr. Cooper’s career. Project findings are consistent with the NIMH strategic goal to identify validated targets for intervention and will inform a competitive R01 application examining computational learning mechanisms in a transdiagnostic sample of youth with restrictive ED. Project Number: 1K23MH138754-01A1 | Fiscal Year: 2026 | NIH Institute/Center: National Institute of Mental Health (NIMH) | Principal Investigator: Marita Cooper | Institution: CHILDREN'S HOSP OF PHILADELPHIA, PHILADELPHIA, PA | Award Amount: $170,024 | Activity Code: K23 | Study Section: Special Emphasis Panel[ZRG1 BP-S (02)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11230775