Leveraging Distributed Brain Signal at Rest to Predict Internalizing Symptoms in Youth: Deriving a Polyneuro Risk Score From the ABCD Study Cohort

Background

The prevalence of internalizing psychopathology rises precipitously from early to mid-adolescence, yet the underlying neural phenotypes that give rise to depression and anxiety during this developmental period remain unclear.

Methods

Youths from the Adolescent Brain Cognitive Development (ABCD) Study (ages 9–10 years at baseline) with a resting-state functional magnetic resonance imaging scan and mental health data were eligible for inclusion. Internalizing subscale scores from the Brief Problem Monitor-Youth Form were combined across 2 years of follow-up to generate a cumulative measure of internalizing symptoms. The total sample (N = 6521) was split into a large discovery dataset and a smaller validation dataset. Brain-behavior associations of resting-state functional connectivity with internalizing symptoms were estimated in the discovery dataset. The weighted contributions of each functional connection were aggregated using multivariate statistics to generate a polyneuro risk score (PNRS). The predictive power of the PNRS was evaluated in the validation dataset.

Results

The PNRS explained 10.73% of the observed variance in internalizing symptom scores in the validation dataset. Model performance peaked when the top 2% functional connections identified in the discovery dataset (ranked by absolute β weight) were retained. The resting-state functional connectivity networks that were implicated most prominently were the default mode, dorsal attention, and cingulo-parietal networks. These findings were significant (p < 1 × 10⁻⁶) as accounted for by permutation testing (n = 7000).

Conclusions

These results suggest that the neural phenotype associated with internalizing symptoms during adolescence is functionally distributed. The PNRS approach is a novel method for capturing relationships between resting-state functional connectivity and behavior.