Application and Expansion of an Algorithm Predicting Attention-Deficit/Hyperactivity Disorder and Impairment in a Predominantly White Sample

Current assessment protocols for attention-deficit/hyperactivity disorder (ADHD) focus heavily on a set of highly overlapping symptoms, with well-validated factors like cognitive disengagement syndrome (CDS), executive function (EF), age, sex, and race and ethnicity generally being ignored. Using machine learning techniques, the current study aimed to validate recent findings proposing a subset of ADHD symptoms that, together, predict ADHD diagnosis, severity, and impairment level better than the full symptom list, while also testing whether the inclusion of the factors listed above could further increase accuracy. Parents of 1,922 children (50.1% male) aged 6-17 years completed rating scales of ADHD, CDS, EF, and impairment. Results suggested nine symptoms as most important in predicting outcomes: (a) has difficulty sustaining attention in tasks or play activities; (b) does not follow through on instructions and fails to finish work; (c) avoids tasks (e.g., schoolwork, homework) that require sustained mental effort; (d) is often easily distracted; (e) has difficulty organizing tasks and activities; (f) is often forgetful in daily activities; (g) fidgets with hands or feet or squirms in seat; (h) interrupts/intrudes on others; and (i) shifts around excessively or feels restless or hemmed in. The abbreviated algorithm achieved accuracy rates that did not significantly differ compared to an algorithm comprising all 18 symptoms in predicting impairment, while also demonstrating excellent discriminative ability in predicting ADHD diagnosis. Adding CDS and EF to the abbreviated algorithm further improved the prediction of global impairment. Continued refinement of screening tools will be key to ensuring access to clinical services for youth at risk for ADHD. General Scientific Summary This study suggests that nine core symptoms of attention-deficit/hyperactivity disorder (ADHD), when incorporated into a machine learning algorithm, can predict ADHD diagnosis and impairment with accuracy equal to that of all 18 symptoms. The inclusion of executive function and cognitive disengagement syndrome in the algorithm appeared to further improve accuracy, although age, sex, and parent race and ethnicity did not. Deployment of this algorithm into schools and pediatrician offices could provide the best and most efficient means to simultaneously screen for the risk of ADHD and impairment.