Sleep qEEG as a non-invasive functional biomarker for Alzheimer’s disease

Abstract

Molecular biomarkers have transformed the biological definition of Alzheimer’s disease (AD), but they do not directly quantify the dynamic consequences of pathology on brain function. Sleep quantitative electroencephalography (qEEG) provides a complementary functional readout. We frame sleep qEEG as an artificial-intelligence problem in which the central contribution is not a single model architecture, but the formulation and validation of several inference tasks on a shared overnight measurement backbone. Across three related studies, polysomnographic recordings were harmonized to four EEG derivations and analyzed separately in the four main sleep stages. First, stage-specific supervised learning detected an AD-related signal with approximately 94% accuracy in internal cross-validation. Second, participant-level qEEG features predicted a shared component of five cerebrospinal-fluid (CSF) oxidative-damage markers, with held-out R2 = 0.625. Third, a biomarker-informed qEEG representation suggested graded heterogeneity within AD. Interpretation of these results rests on a small number of methodological choices: state-conditioned analysis, participant-level validation, leakage control, and the choice of representation. Domain shift, uncertainty, and external calibration become critical when moving beyond internal validation. Sleep qEEG therefore emerges as a promising candidate functional biomarker, but not yet as a deployable stand-alone clinical test.