INTRODUCTION: Sleep abnormalities are highly prevalent in patients with neurodegenerative disease, often appearing in the pre-clinical stage and may reflect the underlying neuropathology long before cognitive decline is detected. Changes in sleep spindles during non-rapid eye movement (NREM) sleep have been associated with cognitive decline in Parkinson’s disease1, and reduced slow wave sleep has been associated with increased beta amyloid concentrations in cerebrospinal fluid in cognitively normal elderly2. These sleep architecture characteristics of NREM sleep are believed to be associated with the metabolic clearance system of the brain. Increased orexin levels in patients with Neurodegenerative disease (NDD) have been associated with prolonged sleep latency, reduced sleep efficiency, and REM sleep impairment3. Additionally, severe obstructive sleep apnea (OSA), which causes sleep discontinuity, has been associated with a higher risk of NDD4. This is the first report investigating sleep biomarkers (i.e., architecture and continuity) in NDD patients using data acquired in-home with a self-applied acquisition system.
METHODS: Subjects: Under IRB approval, two-night overnight EEG studies were obtained from patients diagnosed with NDD (Mild Cognitive Impairment (MCI), Alzheimer’s disease (AD) or Lewy Body Dementia (DLB) and elderly, healthy controls (HC) (Table 1). In order to match the ages of the NDD group, a subset of HC studies (n=37) selected from a database acquired at Washington University Knights Alzheimer’s Disease Research Center were included in the analysis. Analyses were performed on the HC cohort that included 31 males (69 + 10.1 years) and 26 females (70 + 8.1 years), and the NDD cohort included 24 males (71 + 7.9 years) and 12 females (72 + 9.0 years).
Sleep Parameters: The Sleep Profiler used in this study was a battery-powered recorder designed to acquire 3 frontopolar EEG signals between AF7-AF8, AF7-Fpz, and AF8-Fpz. Power spectra from the delta (1–3.5 Hz), DeltaC (delta power corrected for ocular activity), theta (4–6.5 Hz), alpha (8–12 Hz), sigma (12–16 Hz), beta (18–28 Hz), and EMG bands (40-128 Hz) and ocular activity were extracted and applied to previously validated algorithms that detect cortical arousals, sleep spindles and stage each 30-s epoch as awake, non-REM (NREM stages N1, N2 or N3 or, rapid eye movement (REM) sleep.5 Studies were auto-scored and manually reviewed for quality. Studies were excluded when at least one night of data was unavailable due to poor data quality that impacted the distributions of sleep architecture. Additionally, HC were excluded due to age or sleep patterns associated with moderate/severe obstructive sleep apnea.
Statistics: Sleep biomarkers from the HC, MCI and AD groups were submitted for analysis with student t-tests. The NDD data were combined and submitted along with the HC data for stepwise analysis in order to identify discriminating biomarkers. A linear discriminant function analysis (DFA) was then applied using a total of 12 variables selected by either stepwise analysis, significance of t-test results, or previously reported as sleep biomarkers of NDD.
RESULTS: Figure 1 shows the distributions of variables used in the DFA classifier for the HC, MCI and AD groups. The resulting DFA classifier, after leave-one-out-cross-validation, provided an overall accuracy of 83.33% (sensitivity and specificity = 83.33%, negative predictive value = 88.24%, positive predictive value 76.92%). The ROC curve findings for these pilot data which are presented in Figure 2 suggest excellent diagnostic accuracy.