Abstract
Intra-individual variability in reaction times (IIVRT), which generally occurs as a result of episodic long reaction times (RTs), is a marker for impaired attention. Multiple functional neuroimaging studies have attempted to discern neurofunctional correlates of IIVRT, but few use models that account for trial-level IIVRT. Neurofunctional correlates of IIVRT differ depending on the method applied, and few studies have used multiple methods in the same sample. This study utilized Stop-Signal Task functional magnetic resonance imaging (fMRI) data from 8,066 children (9-10 years old) in the Adolescent Brain Cognitive Development (ABCD) study. IIVRT was modeled using multiple methods, including converting RTs to z-scores, variance time course modeling, and a novel machine-learning technique (i.e., hidden Markov model) to compute the probability of a trial reflecting good or poor attentional states. Across all three methods, lower IIVRT was associated with greater activation in the default mode network (DMN), while higher IIRVT was associated with greater activation in the dorsal attention network (DAN). Although all models yielded similar neural correlates, z-score modeling demonstrated the strongest effect sizes in task-related networks. Our findings are congruent with previous work in adults and demonstrate the reproducibility and developmental stability of the neural correlates of trial-level IIVRT. Higher effect sizes for brain-IIVRT associations using the z-score method suggest that this approach is a simple and promising candidate for investigating neural mechanisms related to IIVRT.