Multiple large-scale efforts are underway that share a goal of revising psychological diagnostic nosology to better align with mechanistic features, including cognitive and neural development. Latent grouping approaches may be helpful. However, to date, evidence for replicable and clinically useful cognitive or neuroimaging-based subgroups is weak. Here, we apply common clustering approaches (k-means and latent profile analysis) to (a) cognitive features, including measures of executive functioning and computational reaction time parameters, and (b) structural brain imaging features in the Adolescent Brain Cognitive Development cohort. Using a randomization approach for determining optimal group number and testing within-sample replication, we seek to identify a reproducible grouping structure. To address questions of generalizability to a clinical population, we next determine whether groups can be reproduced in a second, independent sample (Oregon Attention-Deficit/Hyperactivity Disorder-1000) with more than 50% of children diagnosed with attention-deficit/hyperactivity disorder. Finally, we test both concurrent and predictive clinical validity. Results demonstrated that computational reaction time parameters can yield reproducible groups in both population-based and attention-deficit/hyperactivity disorder clinical samples with at least some evidence for clinical validity. Diffusion tensor imaging-measured fractional anisotropy features yielded the strongest evidence of reproducible groups among structural neuroimaging features, although generalizability and clinical validation were not as strong. Overall, clustering approaches offer an important way to leverage rich, high-dimensional data sets. Computational cognitive parameters and diffusion tensor imaging-measured fractional anisotropy may be an appropriate focus of future studies, and similar methods could be applied to other feature sets. (PsycInfo Database Record (c) 2026 APA, all rights reserved).
Karalunas et al. (2026) studied this question.