Assessing Risk of Health Outcomes From Brain Activity in Sleep: A Retrospective Cohort Study

Abstract

Background and objectives: Patterns of electrical activity in the brain (EEG) during sleep are sensitive to various health conditions even at subclinical stages. The objective of this study was to estimate sleep EEG-predicted incidence of future neurologic, cardiovascular, psychiatric, and mortality outcomes.

Methods: This is a retrospective cohort study with 2 data sets. The Massachusetts General Hospital (MGH) sleep data set is a clinic-based cohort, used for model development. The Sleep Heart Health Study (SHHS) is a community-based cohort, used as the external validation cohort. Exposure is good, average, or poor sleep defined by quartiles of sleep EEG-predicted risk. The outcomes include ischemic stroke, intracranial hemorrhage, mild cognitive impairment, dementia, atrial fibrillation, myocardial infarction, type 2 diabetes, hypertension, bipolar disorder, depression, and mortality. Diagnoses were based on diagnosis codes, brain imaging reports, medications, cognitive scores, and hospital records. We used the Cox survival model with death as the competing risk.

Results: There were 8673 participants from MGH and 5650 from SHHS. For all outcomes, the model-predicted 10-year risk was within the 95% confidence interval of the ground truth, indicating good prediction performance. When comparing participants with poor, average, and good sleep, except for atrial fibrillation, all other 10-year risk ratios were significant. The model-predicted 10-year risk ratio closely matched the observed event rate in the external validation cohort.

Discussion: The incidence of health outcomes can be predicted by brain activity during sleep. The findings strengthen the concept of sleep as an accessible biological window into unfavorable brain and general health outcomes.

Figure 1. Cumulative Incidence of the 11 Outcomes in Female (Left of Each Subplot) and Male (Right of Each Subplot) Participants

Each subplot shows the cumulative incidence for one outcome. Cumulative incidence is the proportion of a population at risk that develops the outcome over a specified time. For each outcome, there are 3 cumulative incidence curves: poor sleep (risk score is within the upper 75th percentile, red), average sleep (risk score between the 25th and 75th percentiles, black), and good sleep (risk score lower than the 25th percentile, blue). The dashed lines represent ground truth risks from a nonparametric estimator. Note the different y-axis maximum values. The shaded areas indicate 95% confidence intervals. To ensure these estimates reflect out-of-sample performance, each curve is the average of the 5 curves from the testing sets in cross-validation.

Figure 2. Ground Truth (Black) and Model-Predicted 10-Year Risk (Red) for the Average Sleep Group for Each Outcome, Stratified by Sex: Female (A) and Male (B)

Numbers indicate the 10-year risk as percentage. Error bars indicate the 95% confidence interval. AFib = atrial fibrillation; BD = bipolar disorder; Dem = dementia; Dep = depression; HTN = hypertension; ICH = intracranial hemorrhage; IS = ischemic stroke; MCI = mild cognitive impairment; MI = myocardial infarction; T2D = type 2 diabetes.

Figure 3. Two Examples of Sleep Hypnograms and EEG Spectrograms With High and Low 10-Year Risk of Developing Dementia

(A) A 62-year-old woman with a relatively high predicted 10-year risk. The top panel is the hypnogram; REM sleep is indicated in red. The bottom panel is the spectrogram of the EEG averaged from 2 central channels (C3-M2 and C4-M1). The x-axis is time of the day. The y-axis is frequency in Hz. The color indicates power spectral density on a log scale, in decibels (dB), where higher values are closer to red and lower values are closer to blue. (B) The bottom example is from a 57-year-old woman who has a relatively low predicted 10-year risk. These 2 example participants were selected so that they have similar age around 60 years, same sex, BMI around 30 kg/m², and no medications taken on the night of sleep recording while having different 10-year risks. (C) The top 2 features that contribute most to the difference in dementia risk between panel A and panel B. The contribution is defined as the difference in the feature value time model coefficient. The 2 features are delta-to-alpha ratio at NREM sleep at the frontal channel and central channel, which reflects the relative amount of slow wave, implying sleep depth. The one with high dementia risk (panel A) has a lower relative amount of slow wave, i.e., lighter sleep depth; and the one with lower dementia risk (panel B) has a higher relative amount of slow wave, i.e., deeper sleep depth.