Background: Social anxiety disorder (SAD) is common and often chronic, yet treatment response to first-line interventions such as internet-delivered cognitive behavioral therapy (ICBT) and selective serotonin reuptake inhibitors (SSRIs) remains heterogeneous. Sleep disturbances are prevalent in SAD, and resting-state neural variability has been proposed as a potential marker of neural flexibility. However, whether these two candidate markers can differentiate SAD from healthy controls (HC) or predict treatment outcomes across modalities remains unclear. 

Aims: This study examined whether sleep disturbances and resting-state neural variability (1) differentiate SAD from HC, (2) predict treatment outcomes, and (3) whether the predictive effect of sleep differs between ICBT and SSRI modalities. Methods: Two independent longitudinal samples were included. Sample 1 (46 SAD, 41 HC) received 9-weeks of ICBT and underwent fMRI assessment of resting-state neural variability (standard deviation of the blood-oxygen-level-dependent signal; SDBOLD) in 19 selected brain regions of interest (ROIs) at multiple time points. Sample 2 (24 SAD, 23 HC) participated in a positron emission tomography (PET) study and received 9-weeks of escitalopram (SSRI) treatment. Main analyses included binary logistic regression, linear mixed-effects models (LMMs), Bayesian multivariate models, support vector machines (SVM), and Gaussian mixture modeling (GMM). 

Results: Sleep disturbances significantly distinguished SAD from HC, whereas no single brain region showed group differences in SDBOLD. The 19 ROIs collectively could not predict SAD diagnosis significantly. Neither baseline sleep disturbances (measured by Karolinska Sleep Questionnaire [KSQ]) nor resting-state SDBOLD predicted symptom change with treatment. However, the baseline Insomnia Severity Index (ISI), which specifically assesses insomnia symptoms, significantly predicted ICBT treatment outcome in supplementary analyses. Sleep disturbances at baseline did not predict differential treatment response between ICBT and SSRIs, although both treatments resulted in significant symptom improvement. 

Conclusions: Sleep disturbances measured by KSQ are diagnostically relevant in SAD but did not predict treatment outcomes or inform treatment selection between ICBT and SSRIs in this study. In contrast, ISI did significantly predict ICBT treatment response in supplementary analyses. Resting-state neural variability showed no diagnostic or predictive utility in the ROI-wise analyses.