Data-Driven Methods for Predicting ADHD Diagnosis and Related Impairment: The Potential of a Machine Learning Approach.

Abstract

Current diagnostic criteria for ADHD include several symptoms that highly overlap in conceptual meaning and interpretation. Additionally, inadequate sensitivity and specificity of current screening tools have hampered clinicians' ability to identify those at risk for related outcomes. Using machine learning techniques, the current study aimed to propose a novel algorithm incorporating key ADHD symptoms to predict concurrent and future (i.e., five years later) ADHD diagnosis and related impairment levels. Participants were 399 children with and without ADHD; multiple informant measures of ADHD symptoms, global impairment, academic performance, and social skills were included as part of an accelerated longitudinal design. Results suggested eight symptoms as most important in predicting impairment outcomes five years later: (1) Has difficulty sustaining attention in tasks or play activities, (2) Does not follow through on instructions and fails to finish work, (3) Has difficulty organizing tasks and activities, (4) Avoids tasks (e.g., schoolwork, homework) that require sustained mental effort, (5) Is often easily distracted, (6) Is often forgetful in daily activities, (7) Fidgets with hands or feet or squirms in seat, and (8) Interrupts/intrudes on others. The algorithm comprising this abbreviated list of symptoms performed just as well as or significantly better than one comprising all 18 symptoms in predicting future global impairment and academic performance, but not social skills. It also predicted concurrent and future ADHD diagnosis with 81-93% accuracy. Continued development of screening tools will be key to ensuring access to clinical services for youth at risk for ADHD.