Connectome-Based Predictive Models Optimized for Sleep Differentiate Patients with Depression from Psychiatrically Healthy Controls

Biol Psychiatry Cogn Neurosci Neuroimaging. 2025 Oct 16:S2451-9022(25)00303-9. doi: 10.1016/j.bpsc.2025.10.002. Online ahead of print.

ABSTRACT

BACKGROUND: It is unknown whether brain-based predictive models derived from sleep features are useful for the clinical diagnosis of Major Depressive Disorder (MDD).

METHODS: Using resting-state fMRI data from ABCD (Curated Data Release 3.0), we trained a connectome-based predictive model (CPM) on 35,778 pairwise connections (Pearson’s r) from 2349 (234 participants with at least 1 psychiatric disorder, 2112 controls) participants aged 11-12 to predict sleep duration (measured from FitBit). Linear regression models were used to compare the predicted values from these CPMs with self-reported sleep duration and diagnostic group status in an independent cohort of 78 participants (57 MDD, 21 controls) aged 14-18.

RESULTS: The ABCD-based CPM predicted self-reported sleep duration in the independent cohort of MDD participants (partial r=0.332, p=0.009). Even though self-reported sleep duration did not significantly differ between diagnostic groups (t=0.13, p=0.90), the ABCD-based CPM successfully distinguished between diagnostic groups (partial r=0.334, p<0.001), and CPM-predicted sleep durations correlated with depression symptom severity (partial r=0.294, p<0.001). These diagnostic group differences were driven primarily by patterns of hypoconnectivity between various resting-state networks (including the default mode, frontoparietal, motor, subcortical, and visual associative networks).

CONCLUSIONS: CPMs trained to predict objective sleep duration are robust and generalizable. Intrinsic functional connectivity differences between clinically depressed and psychiatrically healthy adolescents are detectible by CPMs optimized for sleep prediction, underscoring the shared neural bases between sleep health and depression. Future work will test whether sleep-based CPMs are predictive of clinical course and if they generalize to other disorders beyond depression.

PMID:41109569 | DOI:10.1016/j.bpsc.2025.10.002