Levendowski, Daniel J.

Abstract Introduction This study investigated the impact of visual editing of sleep recordings for use in predicting neurodegenerative disorder (NDD) risk. Methods All participants underwent in-home Sleep Profiler studies. Forty-four were diagnosed with mild cognitive impartment (MCI = 40) or Alzheimer’s disease (AD = 4)(MCI/AD: 55% women, 72.6+/-8.6 years), 26 with REM sleep behavior disorder (RBD: women = 27%, age = 67+/-9.8), and 29 were controls (NC: women = 69%, age = 63+/-11.5). A 4-class machine learning algorithm used age plus eight sleep biomarkers to assign NDD risk probabilities. Records with CG-assigned probabilities >70% were labeled Probably-Normal, and from 45-70% Likely-Normal. Similar thresholds were used to assign Likely- or Probably-Abnormal. NDD risk characterizations were generated based on auto-scoring (“auto”), and again after the recordings were visually edited for accuracy (“edited”). Results Comparing the auto- vs. edited-NDD classifications in the MCI/AD group, the same NDD risk classifications were observed in 91% (39/43), with 2 cases shifting from Likely-Normal to Probably-Normal, one shifted from Likely-AD to Likely-Normal with AD Indications, and one changed from Probably-AD to Likely-AD. The same NDD risk classifications were observed in 81% (21/26) of the RBD cases, 3 shifted from Probably-Abnormal to Likely-Abnormal, one shifted from Likely-Normal to Probably-Normal, and one changed from Likely-Abnormal to Likely-Normal. The same NDD risk classifications were observed in 86% of the NC group, with 4 cases shifting from Likely-Normal to Probably-Normal. Conclusions These findings suggest that the likelihood of a gross misclassification of NDD risk resulting from auto-detected sleep biomarkers is relatively low with inconsistencies directed toward slightly increased NDD risk classifications.

Abstract Introduction This study investigated differences in opinions about neurodegenerative disorders (NDD) in countries with large populations and expanded access to healthcare. Methods A 5-point Likert scale survey assessed opinions of 30–59-year-olds, pertaining to their parents and NDD conditions. 100 surveys, translated into native languages, were acquired in eight developed countries. Positive (somewhat- or strongly-agree) vs. other responses, sex, and country were submitted to logistic regression and Chi-squared analyses. Results “It is important for me to know if my parent has a risk for developing a NDD condition in the future, even if nothing can be currently done to prevent or cure those diseases” had higher positive responses in Australia (92%) compared to Germany (74%) and the USA (78%). “I believe that knowing about my parent’s risk for developing a NDD condition will help me assist in the planning of their future” had higher number of positive responses in Australia (91%) versus Japan (70%), Germany (76%), Italy (80%) and UK (80%). “If my parent(s) was found to be at risk for a NDD condition, I would want to learn more about medications and adaptive therapies that could delay onset” was responded to more positively in Australia (92%), Canada (78%), France (87%), Germany (78%), Italy (88%), UK (88%), and USA (81%) versus Japan (60%). Conclusions Australians responded more positively to questions related to care of a parent with a NDD disorder. These findings suggest cultural differences that could influence education and support systems requirements for sleep medicine providers and NDD patients.

AbstractBackgroundIn this pilot study, the diagnostic agreement for sleep biomarker‐based neurodegenerative disorder (NDD) risk probabilities was evaluated in patients with Alzheimer’s disease dementia (AD), Lewy body disease (LBD), mild cognitive impairment (MCI), and controls (CG) with a Mini‐Mental State Exam scores ≥28.MethodsSleep biomarkers recorded with the Sleep Profiler (Advanced Brain Monitoring, Inc.) were used as inputs to a 4‐class machine learning algorithm trained to assign NDD risk probabilities to AD=27, LBD=19, isolated REM sleep behavior disorder=15, and CG=58. Input variables included age, time‐REM, non‐REM hypertonia, autonomic‐activation index, spindle‐duration, atypical‐N3, time‐supine, sleep‐efficiency, relative‐theta, and theta/alpha. Distributions of NDD risk were then assigned to AD=37, LBD=19, MCI=71, and CG=106 records.Records with CG output probabilities ≥70% were labeled Probably‐Normal, and from 45‐70% Likely‐Normal with indications of a NDD added when the LBD, AD or prodromal synucleinopathy (pSYN) risks exceeded approximately 25%. Similar thresholds were used to assign Likely‐ or Probably‐ LBD, AD, pSYN, or Mixed based on the group probabilities. Likely‐Normal with indications consistent with the diagnosis trended toward agreement.ResultsThe proportions of each group with classifications consistent with the diagnosis or trended toward agreement were CG=86%, MCI=75%, LBD=79%, and AD=87%.In the CG group, NDD risk assignments were Probably‐Normal=68% and Likely‐Normal with NDD indications=18%, Likely‐NDD=10%, and Probable‐NDD=4%.The MCI group was assigned Probable‐NDD=38% (AD=26%, Mixed=10%, LBD/pSYN=2%), Likely‐NDD=17% (AD=7%, Mixed=6%, LBD=3%, pSYN=1%), Likely‐Normal with indications=20% (AD=9%, pSYN=7%, Mixed=4%) and Probably‐Normal=25%The LBD group’s assigned risks consistent with the diagnoses included Probable‐NDD=74% and Likely‐Normal with pSYN indications=6% but included inconsistencies from Probable‐AD=5%, Likely‐Normal with AD indications=5%, and Probably‐Normal=11%.AD group’s consistent risk classifications included Probable‐NDD=60%, Likely‐NDD=11%, Likely‐Normal with AD/Mixed indications=13% and countered by inconsistencies from Probable‐LBD/pSYN=8%, Likely‐Normal with pSYN indications=3%, and Probably‐Normal=5%.ConclusionsConsistencies between the diagnoses and assigned NDD risks agreements exceeded 75% in all four groups. Studies are underway to confirm the test‐retest reliability and/or evaluate NDD risk probabilities in other patient populations.Support: NIH grants R44AG050326, R44AG054256, RO1AG060477, R33AG064271, ULRR024150, R34AG056639.

Abstract Biomarkers that aid in early detection of neurodegeneration are needed to enable early symptomatic treatment and enable identification of people who may benefit from neuroprotective interventions. Increasing evidence suggests that sleep biomarkers may be useful, given the bi-directional relationship between sleep and neurodegeneration and the prominence of sleep disturbances and altered sleep architectural characteristics in several neurodegenerative disorders. This study aimed to demonstrate that sleep can accurately characterize specific neurodegenerative disorders (NDD). A four-class machine-learning algorithm was trained using age and nine sleep biomarkers from patients with clinically-diagnosed manifest and prodromal NDDs, including Alzheimer’s disease dementia (AD = 27), Lewy body dementia (LBD = 18), and isolated REM sleep behavior disorder (iRBD = 15), as well as a control group (CG = 58). The algorithm was validated in a total of 381 recordings, which included the training data set plus an additional AD = 10, iRBD = 18, Parkinson disease without dementia (PD = 29), mild cognitive impairment (MCI = 78) and CG = 128. Test-retest consistency was then assessed in LBD = 10, AD = 9, and CG = 46. The agreement between the NDD profiles and their respective clinical diagnoses exceeded 75% for the AD, LBD, and CG, and improved when NDD participants classified Likely Normal with NDD indications consistent with their clinical diagnosis were considered. Profiles for iRBD, PD and MCI participants were consistent with the heterogeneity of disease severities, with the majority of overt disagreements explained by normal sleep characterization in 27% of iRBD, 21% of PD, and 26% of MCI participants. For test-retest assignments, the same or similar NDD profiles were obtained for 88% of LBD, 86% in AD, and 98% of CG participants. The potential utility for NDD subtyping based on sleep biomarkers demonstrates promise and requires further prospective development and validation in larger NDD cohorts.

Abstract Introduction n this cross-validation study, the distributions of sleep biomarker-based neurodegenerative disorder (NDD) risk were evaluated across a spectrum of NDD patients and presumed controls which were independent of the model development dataset. Methods Variables that included age, time-REM, non-REM hypertonia, autonomic-activation index, spindle-duration, atypical-N3, time-supine, sleep-efficiency, relative-theta, and theta/alpha obtained with Sleep Profiler were used to train a 4-class machine learning algorithm that generated probabilities of NDD risk for the groups: Alzheimer’s dementia (AD), Lewy body disorder (LBD), prodromal synucleinopathy (pSYN based on REM sleep behavior disorder patients), and controls (CG). Records with CG output probabilities >70% were labeled Probably-normal, and from 45-70% Likely-normal with indications of a NDD added when the LBD, AD or pSYN risks exceeded approximately 25%. Similar thresholds were used to assign Likely- or Probable- LBD, AD, pSYN or Mixed based on the group probabilities. This cross-validation study was designed to use datasets that were independent of those used to train the classification algorithm. Thus, NDD risk distributions were evaluated in patients with mild cognitive impairment (MCI=71), Parkinson’s disease (PD=16), progressive supranuclear palsy (PSP=21) and control subjects (NC=41) with Mini-Mental State Exam scores>28 and/or normal for MCI by Petersen criteria. Results In the MCI group, NDD risk assignments were Probable-AD=26%, -Mixed=10%, -LBD=1%, Likely-AD= 7%, -Mixed=6%, -LBD=3%, -pSYN=1%, Likely-Normal indications-AD=9%, -pSYN=7%, -Mixed=4%, and Probably-Normal=26% In the PSP group, Probable-LBD=43%, -Mixed=29% and -pSYN=4%, Likely-Mixed=10%, Likely-Normal indications AD=10%, and Probably-Normal=4%. In the PD group, Probable-LBD=13%, -pSYN=13%, -AD=6%, Likely-Mixed=19%, -pSYN=6%, Likely-Normal Indications-pSYN=19%, -AD=12%, Likely-Normal=6% and Probably-Normal=6% In the NC group, 83% were assigned minimal NDD risk with Probably-Normal=70%, Likely-Normal plus indications=13%, but with Likely NDD=13% and Probable NDD=4%. Conclusion NDD risks were disbursed in a manner consistent with the range of severities expected In MCI and PD groups, with 25% exhibiting relatively normal sleep patterns. Over 70% of the PSP patients exhibited probable NDD risk, with patterns similar to those with LBD. The NC NDD risk distributions were consistent with those assigned to the CG used for model development. Support (if any) NIH grants R44AG050326, R44AG054256, RO1AG060477, R33AG064271, ULRR024150, R34AG056639.

Abstract Introduction In this pilot study, the diagnostic agreement for sleep biomarker-based neurodegenerative disorder (NDD) risk probabilities was evaluated in patients with Alzheimer’s disease dementia (AD), Lewy Body disease (LBD), and isolated REM sleep behavior disorder (iRBD) and controls with a Mini-Mental State Exam scores>28. Methods Sleep biomarkers recorded with the Sleep Profiler (Advanced Brain Monitoring, Inc) were used as inputs to a 4-class machine learning algorithm trained to assign NDD risk probabilities to AD=27, LBD=19, iRBD=15 and controls (CG=58). Input variables included age, time-REM, non-REM hypertonia, autonomic-activation index, spindle-duration, atypical-N3, time-supine, sleep-efficiency, relative-theta, and theta/alpha. Records with CG output probabilities >70% were labeled Probably-normal, and from 45-70% Likely-normal with indications of a NDD added when the LBD, AD or prodromal synucleinopathy (pSYN) risks exceeded approximately 25%. Similar thresholds were used to assign Likely- or Probable- LBD, AD, or pSYN based each of the group probabilities. Classification accuracies were evaluated in AD=37, LBD=19, iRBD=19 and CG=61 records. Longitudinal test-retest reliabilities were assessed for CG (n=45, range 8-56 months), and after six-months for LBD (n=10) and AD (n=9). Results In the AD group, NDD risk assignments were consistent with diagnoses=85%, with Probable-NDD=59%, Likely-NDD=11%, Normal plus AD/Mixed indications=15%, but with 5%=Probably-normal. In the LBD group, assigned risks were consistent with the diagnosis=79%, with Probable-NDD=74%, Normal with pSYN indications=5%, but with 11%=Probably-normal. In the iRBD group, Probable-NDD=42%, Normal plus pSYN indications=26% and Probably-normal=32%. In the CG group, 89% were assigned minimal NDD risks with Probably-normal=67%, Normal plus indications=22%, but with Likely NDD=8% and Probable NDD=3%. The AD test-retest reliability showed the same NDD classifications=78%, with slightly increasing=11% and decreased=11%. In the LBD group, the same NDD classifications were achieved=80%, with trending equivalence=10%, and decreasing severity=20%. The CG test-retest reliability showed the same NDD classifications=76%, with trending equivalence=22%, and increased severity=2%. Conclusion Consistencies between the diagnoses and assigned NDD risks and test-retest classification agreements were approximately 80% in the AD, LBD, and CG groups. In the iRBD group, the NDD risks were disbursed consistent with the range of expected severities. These results needed to be cross-validated in independent NDD datasets. Support (if any) NIH grants R44AG050326, R44AG054256, RO1AG060477, R33AG064271, ULRR024150, R34AG056639.

Abstract Introduction In this pilot study, sleep biomarker-based neurodegenerative disorder (NDD) risk was assessed in a clinical cohort of patients diagnosed with varying severity of obstructive sleep apnea (OSA). Methods Consecutive in-home multi-night Sleep Profiler records from a community sleep clinic were screened for age >54 years old. Cases with comorbid restless leg syndrome, hypersomnia, missing apnea-hypopnea indexes (AHI), AHI< 5 or single-night recordings were excluded. The 67 remaining records were 40% female, ages 63 + 7.4 years and up to 84 years old. NDD risk was assigned using a machine-learning algorithm combining age and nine sleep biomarkers to assign group probabilities: for Alzheimer’s disease, Lewy Body disease, prodromal synucleinopathy and controls (CG). Sleep biomarkers included: time-REM, non-REM hypertonia, autonomic-activation index, spindle-duration, atypical-N3, time-supine, sleep-efficiency, relative-theta, and theta/alpha. Those with CG probability score >70% were classified as “Probably-normal”. Those with CG probability between 45 - 70% were classified as “Likely-normal”. Insomnia severity >14 and a PHQ 9 > 5 were used as cutoffs for clinical insomnia and depression. Mann-Whitney U-tests were used to evaluate group differences. Results This cohort had 37% mild, 34% moderate, and 29% severe OSA. Self-reported comorbid insomnia and depression were present in 43%. Seventy-three percent of patients were classified as “Probably-normal”, 16% “Likely-normal”, 11% “Likely-NDD” or “Probable-NDD”. The “Probably-normal” group was younger than the rest of the cohort (62 + 6.2 years vs. 68 + 8.6 years, p< 0.01). Of those classified as “Probably-normal” (n=49), OSA was mild in 16, moderate in 17 and severe in 16. Mean AHI trended higher in the “Probably-normal” group than in those with elevated NDD risk (29 + 23.9 vs. 19 + 13.6 events/h, p=0.085). Conclusion Approximately 89% of the clinical cohort was classified as “Probably-normal” or “Likely-normal”. NDD risk was age-dependent. Interactions between OSA severity and NDD risk were not apparent in these findings, possibly related to group-wide effects. NDD risk, as measured by this model, did not appear to be adversely biased by untreated OSA. These results suggest a need for longitudinal studies to compare NDD risk in untreated OSA versus controls without sleep disordered breathing. Support (if any) NIH grants R44AG050326, R44AG054256, RO1AG060477, R33AG064271, ULRR024150, and R34AG056639.

AbstractBackgroundDementia with Lewy bodies (DLB) and Alzheimer’s disease (AD) are challenging to differentiate. Mobile health devices that measure motor and sleep domains between clinic visits may improve differential diagnosis.MethodParticipants with probable AD and probable DLB wore ActiGraph triaxial wGT3X‐BT actigraphy devices on their nondominant wrists for 2 weeks and multichannel Sleep ProfilerX8TM EEG devices on their foreheads for 2 consecutive nights.3 Actigraphy data were processed and scored using GGIR and activity counts in R. The Sleep Profiler characterizes abnormal slow‐wave sleep and non‐REM hypertonia (NRH).ResultAt baseline, participants with DLB (n = 9) had higher UPDRS scores and a higher likelihood of psychosis than participants with AD (n = 8; see Table 1). Using univariate Wilcoxon Rank‐Sum tests, we found that at baseline participants with DLB had lower daily activity counts, higher % time in NRH, and lower % time in REM sleep than participants with AD (unadjusted p‐values < 0.05; See Figure 1).In a recent 6‐month followup, activity levels were lower on average than baseline activity levels for both DLB (n = 5) and AD (n = 5). at 6‐months, the differences in % sleep in NRH increased between DLB participants (n = 5) and AD participants (n = 3) as average % NRH increase 6% in DLB and decreased 1% in AD (see Table 2). These 6‐month differences are not statistically significant, but data collection are ongoing.An exploration of methods for processing and scoring actigraphy data revealed that processing can result in wide range of findings. As a result, we have focused on actigraphy outcomes which are more suitable to older participants with dementia.ConclusionParticipants with DLB demonstrated lower baseline physical activity and higher rates of baseline sleep disturbances than participants with AD; sleep differences between DLB and AD appeared to increase over 6 months.Actigraphy and sleep‐monitoring devices are well tolerated in participants with dementia and they can reliably collect valid, useful activity and sleep data in between clinic visits.

AbstractBackgroundEEG slowing, a biomarker of cognitive decline in Alzheimer’s disease (AD), has been reported in wake or sleep EEG [1,2,3]. However, the relation between wake and sleep EEG and their comparative sensitivity to early stage of cognitive decline is less understood. In this study we compared sleep and wake EEG biomarkers using a common set of endpoints.MethodsParticipants included individuals with mild cognitive impairment (MCI, n = 30, ages 53‐88), Alzheimer’s disease dementia (ADem, n = 15, ages 60‐84), and controls with normal cognition (HC, n = 55, ages 50‐84). Twenty‐channel resting‐state EEG with eyes‐closed was collected during 5‐minutes of wakefulness (STAT X24). Sleep EEG recordings from AF7‐AF8 were collected in‐home and staged using Sleep ProfilerTM. Power spectral densities (PSD) and the derived Theta‐Alpha ratios were computed for wake EEG sites (counterpart bipolar montage), with the wake FP1‐FP2 compared by stage to the sleep EEG at AF7‐AF8. Significant differences in Theta‐Alpha ratios between MCI or ADem vs. HC groups were identified using independent t‐test (p<0.05) with the normalized effect size measured using Hedges‐g (E.S.).ResultsCorrelation coefficients between wake Alpha, Theta power and non‐REM sleep were: r = 0.83, 0.75 for stage N1, r = 0.71, 0.66 for stage N2, and r = 0.55, 0.50 for stage N3. Based on the frontal Theta‐Alpha ratios, the ADem group showed significant EEG slowing during both wake (E.S. = 0.73) and sleep (E.S. = 1.65, 1.31, 0.61, 1.71 for N1, N2, N3, and REM) as compared to the HC, while MCI group only exhibited EEG slowing during sleep (E.S. = 0.56, 0.77, 0.51 for N1, N2 and REM). Temporal (T5‐T6) wake EEG manifested significant slowing in both ADem (E.S. = 1.21) and MCI (E.S. = 0.56) groups. Compared to the HC, ADem patients exhibited significantly less sleep time (E.S. = 0.91, p = 0.002) and less stage REM (E.S. = 1.0, p = 0.001).ConclusionsCognitive decline in the AD spectrum is associated with common signatures in both sleep and wake EEG. However, the magnitude of these effects and their spatial distribution may be different in sleep versus wakefulness. These findings support a multi modal approach to neurophysiological assessment using both sleep and wakefulness data.