Longitudinal resting-state EEG–based modeling predicts phenoconversion and delineates heterogeneity in isolated REM sleep behavior disorder

Isolated REM sleep behavior disorder is a high-risk prodromal syndrome for Lewy body diseases but highly heterogeneous. Neurophysiologic markers resolving this heterogeneity and predicting phenoconversion remain limited. We applied Subtype and Stage Inference (SuStaIn) modeling to longitudinal resting-state EEG from a prospective cohort of 285 participants. SuStaIn identified two subtypes: Subtype 1, a posterior beta–dominant, largely non-progressive phenotype associated with higher educational attainment and absence of phenoconversion, and Subtype 2, a frontal slow-wave–dominant phenotype characterized by progressive delta/theta increases and cognitive decline. Longitudinal validation demonstrated substantial subtype shifting in Subtype 1 and stage advancement predominantly in Subtype 2. In Kaplan–Meier analysis, subtype shifting and accelerated annual stage advancement (ΔStage ≥ 2) identified individuals at exceptionally high phenoconversion risk with survival approaching 0–15% by 7.5 years. These findings extend SuStaIn beyond cross-sectional inference by incorporating longitudinal subtype-shifting and stage-advancement features to stratify prodromal heterogeneity and predict neurodegenerative progression.