Abstract
This study assessed the prognostic value of probabilistic sleep staging in early PD by quantifying sleep stage uncertainty and mixing. Data from two longitudinal cohorts (DeNoPa and ICEBERG) included 280 patients with early PD (mean duration 21.1 months) and 158 matched controls. Sleep stages were scored automatically using U-Sleep, which provides for each 30-s epoch a probabilistic distribution of wakefulness (W), NREM stage N1, N2, N3, and REM sleep. Uncertainty (percentage of sleep stages with less than 80% certainty) and mixed stages were quantified. PD patients showed higher sleep stage uncertainty and wake-N3 mixing, especially those with REM sleep behavior disorder. Greater W-N3 mixing was associated with worse motor and cognitive scores at baseline and with faster motor decline over time. EEG features of this mixing suggested unstable deep sleep. These findings support the value of automated sleep analysis to detect microstructural sleep disruptions linked to PD progression.