Decoding obstructive sleep apnea phenotypes through integrated multimodal sleep EEG

Obstructive sleep apnea (OSA) is a heterogeneous disorder poorly characterized by the apnea-hypopnea index (AHI), with current subtyping overlooking sleep neurophysiology. We aimed to identify objective neurophysiological phenotypes and mechanisms by integrating multimodal sleep EEG with clinical and biomarker data. We analyzed polysomnographic, demographic, and biomarker data from 309 OSA patients, extracting quantitative EEG (qEEG) features spanning spectral power and sleep microstructure. These were fused via a cascaded multimodal transformer, and phenotypes identified by consensus clustering. Three stable, clinically distinct phenotypes emerged, with the ten most discriminatory features being all qEEG-derived. Phenotype 0 (n = 52) was the youngest group and showed intermediate sleep architecture, with a numerical trend toward higher self-reported cardiovascular disease; Phenotype 1 (n = 131) exhibited marked N3 sleep deficit and the most aberrant EEG spectral profile; and Phenotype 2 (n = 126) showed preserved sleep architecture. Mean AHI did not differ significantly, yet the distribution of AHI severity categories differed across phenotypes, highlighting a dissociation between conventional severity metrics and neurophysiological subtypes. We introduce a physiology-driven OSA taxonomy decoupled from AHI, prioritizing brain signatures to enable biologically targeted diagnosis and therapy of OSA.