Identifying dementia could be easier by using a routine sleep EEG to incorporate dementia screening techniques.
Dementia classification algorithms had promising results in dementia screening when using sleep electroencephalograms (EEGs), according to a study published in Sleep. The researchers believe this is evidence that sleep can be used as a window into neurodegenerative diseases.
In the United States, 11% of people 65 years and older have dementia, and 16% have mild cognitive impairment (MCI). However, dementia is largely undiagnosed in this population, as symptoms are difficult to evaluate. The researchers of the present study aimed to determine if features from machine learning models and sleep EEGs would be able to discriminate patients with dementia and MCI from patients who are cognitively normal (CN).
This cross-sectional study included participants who had a polysomnogram (PSG) at the Sleep Laboratory at Massachusetts General Hospital from 2009 to 2019. Diagnostic, full-night titration, and split-night titration were the 3 major types of sleep tests included in the dataset. Participant demographics, medications, clinical notes, and encounter diagnoses were extracted. Patients were excluded if their encounter diagnosis contained family history as a keyword.
Total resting time, total sleep time, duration of sleep stages, percent of time spent in sleep stages, sleep efficiency index, sleep onset latency, time to first sleep stage, and wake after sleep onset were the sleep architecture features obtained from hypnograms. All participants were matched in each group by sex and age.
There were 10,785 PSGs from 8044 participants that were included in this study, with the median (IQR) age of the participants at 63 (56-70) years and 57% of the participants identified as men. There were 339 participants with 449 PSGs in the dementia group, 514 participants with 672 PSGs in the MCI group, and 7263 participants with 9663 PSGs in the CN group. There were differences in participants with mood disorders, diabetes, alcoholism, psychotic disorders, and anxiety disorders between the dementia and MCI groups vs the CN group.
There were 499 features that had significant odds ratios in dementia vs CN, 177 of which were associated with dementia and 322 of which were associated with CN. There were 386 features that had significant odds ratios for MCI vs CN, 95 of which were associated with MCI and 291 of which were associated with CN.
The 3 machine learning methods had similar performance in differentiating between the dementia, MCI, and CN groups. Support vector machines had the best performance of the 3, with an area under receiver operating characteristic curve (AUROC) of 0.78 and area under precision-recall curve (AUPRC) of 0.22. A logistic regression model was able to best discriminate between MCI and CN, with an AUROC of 0.73 and AUPRC of 0.18. The logistic regression model also performed the best for differentiating dementia/MCI vs CN, with an AUROC of 0.76 and AUPRC of 0.32.
There were some limitations to this study. PSGs from sleep clinics were used for this study, which makes generalizability questionable and needing further study. Clinical diagnoses and cognitive screening tests were used to label dementia and MCI groups and did not include biomarker or neuropathologic confirmation. Also, participants in the MCI and dementia groups likely had underlying neurodegenerative diseases that have EEG signatures.
The researchers concluded that “brain activity during sleep has the potential to detect features associated with dementia and contains information that help inform individual-level clinical decision making.”
Reference
Ye EM, Sun H, Krishnamurthy PV, et al. Dementia detection from brain activity during sleep. Sleep. Published online November 30, 2022. doi:10.1093/sleep/zsac286