Machine Learning-Derived Sleep EEG Brain Age Index Predicts Dementia Risk

Overview

A novel machine learning model quantified brain age from sleep EEG microstructures, producing a brain age index (BAI) that predicts incident dementia risk in community-dwelling adults. Analysis of five longitudinal cohorts demonstrated that an older BAI relative to chronological age is associated with higher dementia incidence, independent of traditional risk factors.

Background

Sleep disturbances are recognized as early markers and modifiable risk factors for dementia, but conventional macrolevel sleep metrics have shown inconsistent associations with cognitive decline. Sleep EEG microstructures directly reflect neural processes and offer a more nuanced assessment of brain aging. Prior studies linked cognitive impairment to specific EEG patterns, yet summarizing these complex features remains challenging. The brain age index (BAI), derived from machine learning integration of multiple EEG features, offers a personalized biomarker to detect prodromal dementia changes.

Data Highlights

Key Findings

• The sleep EEG-based brain age index (BAI) integrates multiple microstructural EEG features into a single age-like metric.
• An older BAI relative to chronological age is significantly associated with increased risk of incident dementia in community-dwelling populations.
• The association between BAI and dementia risk was consistent across five diverse longitudinal cohorts with different dementia ascertainment methods.
• Key EEG features contributing to BAI include spindle density, spindle–slow oscillation coupling, and spectral power across multiple frequency bands.
• The BAI provides a personalized, interpretable biomarker that captures complex sleep EEG patterns beyond traditional macrolevel sleep metrics.

Clinical Implications

The sleep EEG-derived brain age index offers clinicians a novel tool to identify individuals at elevated risk for dementia before clinical symptoms emerge. Incorporating BAI assessment into sleep studies could enhance early detection and enable targeted interventions to modify dementia risk. This personalized approach may improve prognostication and guide preventive strategies in aging populations.

Conclusion

Machine learning-based quantification of sleep EEG microstructures into a brain age index robustly predicts incident dementia risk, supporting its potential as a digital prodromal biomarker. Future work should explore integration into clinical practice for early dementia risk stratification.

References

1. Sun et al. 2021 -- Machine learning model for sleep EEG brain age
2. Prior clinical cross-sectional study associating BAI with dementia
3. MESA, ARIC, FHS-OS, MrOS, SOF cohort studies