Data-driven profiles of behavior in pediatric medical disorders.

Abstract

Behavioral impairment is comorbid with pediatric medical conditions and impacts academic, social-emotional, and medical outcomes. In prior work, we applied graph-theory analysis to parent-report measures of behavior to derive multidimensional profiles in a multi-site database of children with psychiatric disorders and healthy controls (comprised of participants from Children's National Hospital, Georgetown University, and Kennedy Krieger Institute), and identified three unique profiles characterized by relative weaknesses in (a) metacognition, (b) emotion regulation, and (c) inhibition. In this study, we also found broadly the same behavioral profiles within a large (N = 466) cross-sectional clinical database collected at Children's National Hospital from 2014 to 2018 comprised of children with pediatric medical conditions affecting the central nervous system. A support vector machine (SVM) classification derived from the psychiatric sample was then applied to the medical sample and had high (but not perfect) accuracy, suggesting subtle differences in profile composition between medical and nonmedical populations, particularly within the Inhibit subgroup. These findings lend further support to the existence of three transdiagnostic profiles, representing unique targets for personalized intervention. However, findings also highlight that the etiology of behavior problems (psychiatric versus medical) may matter.