Background: There is considerable inter-individual variability in the response to treatment in psychiatric patients. Tools in translational neuroscience, e.g., functional magnetic resonance imaging (fMRI), may be helpful in predicting treatment outcome. Blood-oxygen level-dependent (BOLD) variability, e.g., voxel-wise SD-BOLD, has emerged as an alternative and promising approach for understanding human cognition, but has rarely been considered as a marker of clinical outcome.

Methods: Forty-six patients with social anxiety disorder were scanned with 3T BOLD-fMRI twice (9 weeks apart) prior to treatment. In each scanning session, patients passively viewed facial expressions for 160 seconds. After baseline scanning, patients underwent cognitive behavioral therapy (CBT) for 9 weeks. Multivariate partial least squares (PLS) models were used to link SD-BOLD to treatment outcome (anxiety pre-post change scores), and latent level brain scores were implemented in several subsequent linear regression models.

Results: Treatment outcome varied across patients, but yielded a large effect on anxiety symptom improvement (Cohen’s d = 1.5) on a group-level. Behavioral PLS models strongly revealed lower visual cortex SD-BOLD in patients with more favorable treatment outcomes (first session: ß=.77, Adj-R2 =.58; and second session ß=.78, Adj-R2 =.60). K-fold cross-validation further supported our results, demonstrating a 60-70% reduction in model out-of-sample prediction error when SD-BOLD was included as a predictor.

Conclusions: Our findings provide the first evidence that moment-to-moment variability in neural responses shows translational potential by accurately predicting treatment outcomes in psychiatric patients. If replicated, brain signal variability-based prediction may provide an efficient and viable future tool in clinical psychiatry.