Andrew S. P. Lim

Study Objectives: We investigate a Mamba-based deep learning approach for sleep staging on signals from ANNE One (Sibel Health, Chicago, IL), a non-intrusive dual-module wireless wearable system measuring chest electrocardiography, triaxial accelerometry, chest temperature, and finger photoplethysmography and finger temperature. Methods: We obtained wearable sensor recordings from 357 adults undergoing concurrent polysomnography at a tertiary care sleep lab. Each polysomnography recording was manually scored, and these annotations served as ground truth labels for training and evaluation of our models. Polysomnography and wearable sensor data were automatically aligned using their electrocardiography channels with manual confirmation by visual inspection. We trained a Mamba-based recurrent neural network architecture on these recordings. Ensembling of model variants with similar architectures was performed. Results: After ensembling, the model attains a 3-class (wake, non-rapid eye movement sleep, rapid eye movement sleep) balanced accuracy of 84.02 per cent, F1 score of 84.23 per cent, Cohen’s κ of 72.89 per cent, and a Matthews correlation coefficient (MCC) score of 73.00 per cent; a 4-class (wake, light NREM [N1/N2], deep NREM [N3], REM) balanced accuracy of 75.30 per cent, F1 score of 74.10 per cent, Cohen’s κ of 61.51 per cent, and MCC score of 61.95 per cent; a 5-class (wake, N1, N2, N3, REM) balanced accuracy of 65.11 per cent, F1 score of 66.15 per cent, Cohen’s κ of 53.23 per cent, MCC score of 54.38 per cent. Conclusions: Our Mamba-based deep learning model can successfully infer major sleep stages from the ANNE One, a wearable system without electroencephalography, and can be applied to data from adults attending a tertiary care sleep clinic.

Study Objectives: Wrist accelerometry is widely used for inferring sleep-wake state. Previous works demonstrated poor wake detection, without cross-device generalizability and validation in different age range and sleep disorders. We developed a robust deep learning model for to detect sleep-wakefulness from triaxial accelerometry and evaluated its validity across three devices and in a large adult population spanning a wide range of ages with and without sleep disorders. Methods: We collected wrist accelerometry simultaneous to polysomnography (PSG) in 453 adults undergoing clinical sleep testing at a tertiary care sleep laboratory, using three devices. We extracted features in 30-second epochs and trained a 3-class model to detect wake, sleep, and sleep with arousals, which was then collapsed into wake vs. sleep using a decision tree. To enhance wake detection, the model was specifically trained on randomly selected subjects with low sleep efficiency and/or high arousal index from one device recording and then tested on the remaining recordings. Results: The model showed high performance with F1 Score of 0.86, sensitivity (sleep) of 0.87, and specificity (wakefulness) of 0.78, and significant and moderate correlation to PSG in predicting total sleep time (R=0.69) and sleep efficiency (R=0.63). Model performance was robust to the presence of sleep disorders, including sleep apnea and periodic limb movements in sleep, and was consistent across all three models of accelerometer. Conclusions: We present a deep model to detect sleep-wakefulness from actigraphy in adults with relative robustness to the presence of sleep disorders and generalizability across diverse commonly used wrist accelerometers.

Objectives: We present and evaluate a Mamba-based deep-learning model for diagnosis and event-level characterization of sleep disordered breathing based on signals from the ANNE One, a non-intrusive dual-module wireless wearable system measuring chest electrocardiography, triaxial accelerometry, chest and finger temperature, and finger phototplethysmography. Methods: We obtained concurrent PSG and wearable sensor recordings from 384 adults attending a tertiary care sleep laboratory. Respiratory events in the PSG were manually annotated in accordance with AASM guidelines. Wearable sensor and PSG recordings were automatically aligned based on the ECG signal, alignment confirmed by visual inspection, and PSG-derived respiratory event labels were used to train and evaluate a deep sequential neural network based on the Mamba architecture. Results: In 57 recordings in our test set (mean age 56, mean AHI 10.8, 43.86\% female) the model-predicted AHI was highly correlated with that derived form the PSG labels (R=0.95, p=8.3e-30, men absolute error 2.83). This performance did not vary with age or sex. At a threshold of AHI$>$5, the model had a sensitivity of 0.96, specificity of 0.87, and kappa of 0.82, and at a threshold of AHI$>$15, the model had a sensitivity of 0.86, specificity of 0.98, and kappa of 0.85. At the level of 30-sec epochs, the model had a sensitivity of 0.93 and specificity of 0.95, with a kappa of 0.68 regarding whether any given epoch contained a respiratory event. Conclusions: Applied to data from the ANNE One, a Mamba-based deep learning model can accurately predict AHI and identify SDB at clinically relevant thresholds, achieves good epoch- and event-level identification of individual respiratory events, and shows promise at physiological characterization of these events including event type (central vs. other) and event duration.

As the global burden of chronic kidney disease (CKD) continues to rise, it is important to identify factors that may support preservation of kidney function with age. Circadian rhythmicity declines with age, and evidence suggests that circadian rhythms influence renal physiology, yet their role in long-term kidney health remains underexplored. We tested the hypothesis that lower circadian rhythmicity would be associated with an elevated risk of kidney function decline. About 1315 older adults participating in the Rush Memory and Aging Project, a prospective community-based cohort study of the chronic conditions of aging, were studied. The median age of participants was 81.5 years [IQR: 71.8-91.2], 76.3% were female, and 93.7% identified as Caucasian. The stability of the circadian activity rhythm was quantified from up to 10 consecutive days of wrist activity using interdaily stability (IS) metric. Lower baseline IS was associated with lower eGFR (estimate = -1.9 ml/min/1.73 m² per 1SD lower IS, SE 0.4, p = 5.0 × 10^-6). Among those without baseline CKD (eGFR < 60 ml/min/1.73 m²), lower IS was associated with a ~ 25% higher future risk of developing CKD (HR 1.26, 95% CI 1.11-1.43) over a 5-years period, approaching the increased hazard associated with diabetes (HR 1.37). Effects were independent of known CKD risk factors including age, sex, obesity, hypertension, and diabetes, as well as multiple measures of sleep quality. Reduced circadian rhythmicity may therefore represent a novel modifiable risk factor for age-related decline in kidney function and CKD.

Sleep fragmentation is common in older adults and is associated with cognitive impairment and dementia, as well as key histopathological correlates of dementia, including small vessel disease and cerebral infarcts. Vascular and blood–brain barrier dysfunction are thought to contribute to cognitive decline and dementia. Pericytes, a key vascular cell type, may play a key role. In model organisms, sleep disruption is associated with pericyte dysfunction and blood–brain barrier breakdown. Recent advances in single-nucleus RNA sequencing (snRNAseq) technology have identified two transcriptionally distinct subtypes of pericytes: extracellular matrix protein-expressing M-pericytes and solute carrier-expressing T-pericytes. However, the relationship between sleep, pericyte biology and cognition in humans remains unclear. We tested the hypothesis that differences in the composition of brain pericyte subpopulations, as inferred from marker gene expression, may link sleep fragmentation and cognitive decline. We leveraged two published human brain snRNAseq datasets to identify specific marker genes for M- and T-type pericytes. We then used post-mortem bulk RNAseq data from the dorsolateral prefrontal cortex (n = 1092) and lateral orbitofrontal cortex (n = 495) to quantify expression of these marker genes in older adults in two longitudinal cohort studies: the Religious Orders Study and Rush Memory and Aging Project. We derived trajectories of global cognitive function from participants’ ante-mortem annual cognitive assessments, while sleep fragmentation was derived from ante-mortem wrist-actigraphy recordings from a subset of 572 participants. We used multivariate linear regression to relate pericyte marker gene expression to sleep fragmentation and cognitive decline in the decade preceding death. In the dorsolateral prefrontal cortex, greater average sleep fragmentation was associated with greater expression of M-pericyte marker genes [estimate = +3.65 × 10−1, standard error (SE) = 1.61 × 10−1, P = 0.024] but not T-pericyte marker genes. Dorsolateral prefrontal cortex expression of M-pericyte (estimate = −8.30 × 10−3, SE = 3.37 × 10−3, P = 0.014) but not T-pericyte marker genes was associated with more rapid cognitive decline in the 10 years prior to death. In the lateral orbitofrontal cortex, greater sleep fragmentation was also associated with greater composite M-pericyte gene expression (estimate = + 4.48 × 10−1, SE = 1.92 × 10−1, P = 0.02), which in turn was associated with faster cognitive decline in the decade preceding death (estimate = −1.30 × 10−2, SE = 4.55 × 10−3, P = 0.0044). These findings identify a potential role of M-pericytes in linking sleep fragmentation and cognitive trajectories in older adults. Additionally, our findings highlight the importance of vascular mechanisms in linking disrupted sleep to dementia.

Background: Cardiovascular diseases are closely linked to cognitive health. Subclinical cardiovascular functional changes, such as cardiac autonomic dysfunction, precede cardiovascular diseases and improve risk stratification. Continuous monitoring of heart rate or pulse rate is a commonly used approach to evaluate cardiac cycle and autonomic regulation. We investigated whether the complexity of pulse rate is associated with longitudinal cognitive decline in older adults. Methods: Overnight pulse oximetry data were collected from 503 participants (mean age=82±7 [SD] years, 76% female). We used a previously established distribution entropy algorithm to extract the complexity of pulse rate as a proxy for subclinical cardiovascular function. Participants completed a standardized cognitive test battery during the same visit of pulse oximetry and at least 1 follow‐up visit. Linear mixed‐effects models were conducted to test whether distribution entropy is associated with longitudinal changes in global cognition and separately, in 5 cognitive domains. Results: Greater distribution entropy (ie, better complexity) was associated with a slower decline in global cognition; the effect of 1‐SD increase in distribution entropy was equivalent to being approximately 3 years younger. No associations were observed between conventional time‐ or frequency‐domain pulse rate variability measures and cognitive changes. Conclusions: Higher complexity of pulse rate is linked with slower cognitive decline in older adults. Future studies should test whether complexity is also associated with future risks of neurodegenerative disorders, such as dementia, and further elucidate the causal directions.

Hemodynamic homeostasis is essential for adapting the heart rate (HR) to postural and physiological changes during daily activities. Traditional HR monitoring, such as 24 hour (h) Holter monitoring, provides important information on homeostasis during daily living. However, this approach lacks concurrent activity recording, limiting insights into hemodynamic adaptation and our ability to interpret changes in HR. To address this, we utilized a novel wearable sensor system (ANNE^@Sibel) to capture time-locked HR and daily activity (i.e., lying, sitting, standing, walking) data in 105 community-dwelling older adults. We developed custom tools to extract 24 h time-locked measurements and introduced a “heart rate response score” (HRRS), based on root Jensen–Shannon divergence, to quantify HR changes relative to activity. As expected, we found a progressive HR increase with more vigorous activities, though individual responses varied widely, highlighting heterogeneous HR adaptations. The HRRS (mean: 0.38 ± 0.14; min: −0.11; max: 0.74) summarized person-specific HR changes and was correlated with several clinical measures, including systolic blood pressure changes during postural transitions (r = 0.325, p = 0.003), orthostatic hypotension status, and calcium channel blocker medication use. These findings demonstrate the potential of unobtrusive sensors in remote phenotyping as a means of providing valuable physiological and behavioral data to enhance the quantitative description of aging phenotypes. This approach could enhance personalized medicine by informing targeted interventions based on hemodynamic adaptations during everyday activities.

Background: Cardiovascular diseases may be linked to dementia pathogenesis. The cardio‐autonomic control (CAC) plays an important role in the development of cardiovascular diseases and may be implicated by dementia pathophysiology. We tested whether the CAC proxied by the complexity of beat‐to‐beat pulse rate intervals (PRI) predicted longitudinal cognitive decline in older adults.Method: Continuous overnight pulse oximetry (photoplethysmography) data from 594 participants (mean age = 82.76±7.11 at oximetry assessment; 76% female) in the Rush Memory and Aging projects were studied. All participants completed cognitive assessments at oximetry baseline. Eighty‐four participants who had at least one follow‐up cognitive assessment were included in the analyses. After quality control, the feet of pulse waves in the photoplethysmography data were detected, and the consecutive foot‐to‐foot internals (i.e., PRI) time‐series within each 10‐min window were analyzed using the distribution entropy that evaluated the complexity of PRI (i.e., greater distribution entropy means more complex dynamics). To reduce behavioral and circadian effects, we used the mean distribution entropy values during 12AM‐3AM for each individual. Mixed‐effects models were constructed to associate distribution entropy to longitudinal cognitive changes, adjusting for age, sex, and education.Result: At baseline, distribution entropy was normally distributed (mean = 0.33±0.14). On average, the global cognition score was 0.12 units (SD = 0.60). This score declined at an annual rate of 0.15 units (SE = 0.04; p = 0.0001). The distribution entropy was not associated with global cognition at baseline, but greater distribution entropy predicted slower cognitive decline. Specifically, a 1‐SD increase in distribution entropy correlated with a drop of 0.11 (SE = 0.04; p = 0.005) in the rate of cognitive decline.Conclusion: The nighttime CAC (potentially during sleep) is linked with longitudinal cognitive profiles in older adults. As the CAC is regulated by an internal body clock—the circadian system, in which a dysfunction is involved in Alzheimer’s pathogenesis, it is possible that the CAC underlies the connections of cardiovascular disorders and circadian dysfunction with Alzheimer’s disease and its clinical outcomes. Future studies are warranted to test these hypotheses. Further studies should also investigate whether CAC at different times of the night has different correlations with cognitive changes and whether CAC predicts Alzheimer’s disease.

Background: Actigraphy derived measures including amplitude, regularity, and variability of daily rhythm, have been shown to predict incident Alzheimer’s dementia. Here we developed an integrated actigraphy biomarker (IAB) for Alzheimer’ risk and investigated whether the IAB predicted the conversion from mild cognitive impairment (MCI) to Alzheimer’ dementia.Method: We studied 1195 participants (age 80.8±7.2yrs [SD]) from the Rush Memory and Aging Project who had finished baseline actigraphy assessment (∼10 days), were free from Alzheimer’s dementia at actigraphy baseline and had been followed for up to 15 years with annual cognitive assessment and clinical diagnoses. Ten sleep/circadian related features were derived from baseline actigraphy recordings and were fed into a random forest survival model for prediction of time and incident Alzheimer’s dementia. IAB was derived from the model as the relative risk. Cox proportional hazards and logistic regression models were used to evaluate the performance of IAB in predicting incident Alzheimer’s dementia (all 1195 participants) or MCI (858 without MCI at baseline) and predicting the conversion from MCI to Alzheimer’s dementia. Demographic variables including age, sex, and education were controlled in all Cox and logistic regression models.Result: Total 287 participants developed Alzheimer’s dementia during the follow‐up. The derived IAB was 0.6 SD larger in the participants developed Alzheimer’s dementia as compared with the controls. Larger IAB was associated with increased risk of AD with a hazard ratio (HR) = 1.63 (95% CI = 1.46‐1.81, P<0.0001) for 1‐SD increase in IAB. IAB did not predict incident MCI (P = 0.8). Within participants with MCI (337 at baseline and 308 developed), larger IAB was associated with a higher risk for AD, i.e., HR = 1.34 for 1 SD increase (95% CI = 1.18‐1.51, P<0.0001). The logistic model using the cutoff of 3 years for the MCI‐Alzheimer’s dementia conversion gives the odd ratio = 1.56 for 1 SD increase of IAB, and results in AUC = 0.68, with a sensitivity = 0.68 and specificity = 0.61.Conclusion: Derived actigraphy biomarker was predictive of Alzheimer’s risk at preclinical stages, and the conversion from MCI. Actigraphy provides useful information for early prediction and detection of AD thought its performance needs to be improved.