Applications of Item Response Theory to Clinical ADHD Research: Analysis of the Hierarchical Structure of ADHD Symptoms and Increased Precision of Treatment Effect Estimation

Attention-deficit/hyperactivity disorder (ADHD) is a neurodevelopmental disorder with childhood onset that confers greater risk for many negative outcomes including future psychopathology, increased risk for substance abuse, poor self-esteem, and poor social functioning. Identification of targeted treatment approaches requires not only accurate measurement of the construct under study, but also an analytic method that can appropriately identify the effect of treatment, while accounting for the noise of measurement error. This study aimed to address the evident gap in our ability to predict an individual’s response to treatment by first identifying a best-fitting model for ADHD symptoms and then using this best fitting model to estimate treatment effect with increased precision.A sample of 1,612 children and adolescents ages 6 to 17 was used to test the best-fitting model for ADHD. Results from the confirmatory model suggest that a modified bifactor model had better fit (BIC = 47902.44, M2(1346) = 2944.68, RMSEA = .03) compared to the unidimensional and correlated factor models and best minimized mean dimension differences across gender and age. In this modified bifactor model, impulsivity items loaded only on the primary dimension, while inattention and hyperactivity items loaded on both the primary dimension and two separate specific dimensions.In the analysis of treatment effect, two randomized controlled clinical trials of pharmacotherapy for children and adolescents with ADHD were evaluated; DMPH compared to DMPH + guanfacine and atomoxetine compared to placebo. Item parameters generated in study one were used to estimate the treatment models. In the IRT analyses, atomoxetine produced significant reductions in the inattention and the general dimension, however there was more variability in children’s response to atomoxetine as measured by the general dimension. When comparing DMPH and combination treatment, combination treatment was superior in the treatment of the general dimension. Examining variability in average scores provided by the IRT analysis, it appeared that combination treatment was more effective in consistently reducing symptoms across children on the inattention dimension, while DMPH was more effective in consistently reducing symptoms across children on the general dimension. Therefore, creating conditionally independent dimensions clearly allowed for more precise modeling of treatment effect. The field of psychiatry, and more broadly treatment research, could benefit substantially from continued use of IRT models