Predicting Transdiagnostic Social Impairments in Childhood using Connectome-based Predictive Modeling

Background: Social impairments are core features of multiple neurodevelopmental disorders. Previous neuroimaging studies have focused on elucidating associations between brain function and social impairments within disorders but have not predicted these impairments from brain connectivity in a transdiagnostic manner, across several diagnostic categories. This study used a machine learning approach to examine functional connectivity that predicts elevated social impairments in a transdiagnostic sample of youths. We hypothesized that predictive edges would be from brain regions involved in social cognition. Methods: Connectome based predictive modeling (CPM) was used to build a transdiagnostic model of social impairments as measured by the Social Responsiveness Scale (SRS-2, raw score >75). We used functional connectivity data during a social movie-watching task from the Healthy Brain Network data (N=144, mean age=11.68 (3.52), 32% male). The average number of diagnoses was 3.4 (SD = 1.82, range = 0-11), including ASD (40.9%), ADHD (79%), mood disorders (15.9%), and anxiety disorders (43%). A similar transdiagnostic sample high SRS-2 scores (n=41) was used for replication. Results: SRS2 scores were predicted from functional connectivity data using both 10-fold cross-validation (median q2=0.32, r=0.57, p<.001) and leave-one-group-out cross-validation (median q2s>0.04, rs>0.36, ps<.001). Predictive connections were widely distributed across the brain but were rooted in the social brain, the subcortex, and the salience network. The model successfully predicted SRS-2 scores in the replication sample (r=0.33, p<.035, df=39). Conclusions: We identified connectivity patterns predictive of social impairments in a transdiagnostic sample. These networks may inform novel targeted interventions for social impairments across traditional diagnostic categories.