Altered resting-state EEG microstate topographies and transition dynamics in generalized anxiety disorder

Generalized anxiety disorder (GAD) has been associated with abnormalities in large-scale brain networks, but resting-state electroencephalography (rs-EEG) microstates (MS) alterations remain unclear. We investigated spatial and temporal MS abnormalities in GAD and whether MS-derived features could distinguish patients with GAD from healthy controls (HC).

Resting-state EEG data from 104 patients with GAD and 89 HC were analyzed using a five-class MS solution. MS duration, occurrence, coverage, and transition probabilities were compared between groups. Topographic analysis of variance (TANOVA) tested group differences in MS map configurations. Ten supervised machine-learning models were evaluated using MS-derived features, and SHAP (SHapley Additive exPlanations) analysis was used to interpret the best-performing model.

TANOVA showed significant group differences in the topographic configurations of MS-A, MS-D, and MS-E, but not MS-B or MS-C. Significant Group × Microstate Class interactions were observed for duration, occurrence, and coverage, although Bonferroni-corrected class-specific comparisons were nonsignificant. Patients with GAD showed reduced bidirectional transition probabilities between MS-B and MS-C after Bonferroni correction. XGBoost achieved the best overall classification performance, but performance declined substantially in the held-out test set, indicating limited out-of-sample generalizability.

GAD was associated with potential alterations in the spatial organization of specific EEG MS and with reduced bidirectional transitions between MS-B and MS-C. Although MS analysis may help characterize resting-state scalp-level electrophysiological organization in GAD, the classification findings remain preliminary because of limited held-out performance and require validation in larger independent cohorts before any clinical application.