A deep transfer learning approach for sleep stage classification and sleep apnea detection using wrist-worn consumer sleep technologies.

Obstructive sleep apnea (OSA) is a common, underdiagnosed sleep-related breathing disorder with serious health implications Objective - We propose a deep transfer learning approach for sleep stage classification and sleep apnea (SA) detection using wrist-worn consumer sleep technologies (CST). Methods - Our model is based on a deep convolutional neural network (DNN) utilizing accelerometers and photo-plethysmography signals from nocturnal recordings. The DNN was trained and tested on internal datasets that include raw data from clinical and wrist-worn devices; external validation was performed on a hold-out test dataset containing raw data from a wrist-worn CST. Results - Training on clinical data improves performance significantly, and feature enrichment through a sleep stage stream gives only minor improvements. Raw data input outperforms feature-based input in CST datasets. The system generalizes well but performs slightly worse on wearable device data compared to clinical data. However, it excels in detecting events during REM sleep and is associated with arousal and oxygen desaturation. We found; cases that were significantly underestimated were characterized by fewer of such event associations. Conclusion - This study showcases the potential of using CSTs as alternate screening solution for undiagnosed cases of OSA. Significance - This work is significant for its development of a deep transfer learning approach using wrist-worn consumer sleep technologies, offering comprehensive validation for data utilization, and learning techniques, ultimately improving sleep apnea detection across diverse devices.

Fully Automated Detection of Isolated Rapid-Eye-Movement Sleep Behavior Disorder Using Actigraphy.

Isolated rapid-eye-movement (REM) sleep behavior disorder (iRBD) is caused by motor disinhibition during REM sleep and is a strong early predictor of Parkinson's disease. However, screening questionnaires for iRBD lack specificity due to other sleep disorders that mimic the symptoms. Nocturnal wrist actigraphy has shown promise in detecting iRBD by measuring sleep-related motor activity, but it relies on sleep diary-defined sleep periods, which are not always available. Our aim was to precisely detect iRBD using actigraphy alone by combining two actigraphy-based markers of iRBD - abnormal nighttime activity and 24-hour rhythm disruption. In a sample of 42 iRBD patients and 42 controls (21 clinical controls with other sleep disorders and 21 community controls) from the Stanford Sleep Clinic, the nighttime actigraphy model was optimized using automated detection of sleep periods. Using a subset of 38 iRBD patients with daytime data and 110 age-, sex-, and body-mass-index-matched controls from the UK Biobank, the 24-hour rhythm actigraphy model was optimized. Both nighttime and 24-hour rhythm features were found to distinguish iRBD from controls. To improve the accuracy of iRBD detection, we fused the nighttime and 24-hour rhythm disruption classifiers using logistic regression, which achieved a sensitivity of 78.9%, a specificity of 96.4%, and an AUC of 0.954. This study preliminarily validates a fully automated method for detecting iRBD using actigraphy in a general population.Clinical relevance- Actigraphy-based iRBD detection has potential for large-scale screening of iRBD in the general population.

Remote Estimation of Peripheral Oxygen Saturation and Pulse Rate From Facial Analysis Using a Smartphone Camera.

Vital sign monitoring is an invaluable tool for healthcare professionals, both in the hospital and at home. Traditional measurement devices provide accurate readings but require physical contact with the patient which often is unsuitable, furthermore contact-based devices have been reported to fail by loosing contact due to movement as severe events occur, therefore, a contactless method is necessary.We hypothesize that, in ideal scenarios, it is possible to estimate both SpO2 and pulse rate using only facial video recorded with a smartphone's front-facing camera. To test this hypothesis, a dataset of 10 healthy subjects performing various breathing patterns while being recorded with a smartphone camera was collected during ideal lighting conditions.Using advanced image and signal processing methods to acquire remote photoplethysmography (rPPG) estimates from a patient's forehead, our proposed method can achieve SpO2 estimation results with Arms = 1.34% (accuracy RMS) and MAE ± STD = 1.26 ± 0.68% (mean average error) across a SpO2 range of 92% to 99% (percentage point SpO2) and pulse rate estimation results with Arms = 3.91 bpm (beats per minute) and MAE ± STD = 3.24±2.11 bpm across a pulse rate range of 60 bpm to 90 bpm. We conclude from these results, that remote vital sign estimation using facial videos recorded entirely with a smartphone camera is possible.

SViT: A Spectral Vision Transformer for the Detection of REM Sleep Behavior Disorder.

REM sleep behavior disorder (RBD) is a parasomnia with dream enactment and presence of REM sleep without atonia (RSWA). RBD diagnosed manually via polysomnography (PSG) scoring, which is time intensive. Isolated RBD (iRBD) is also associated with a high probability of conversion to Parkinson's disease. Diagnosis of iRBD is largely based on clinical evaluation and subjective PSG ratings of REM sleep without atonia. Here we show the first application of a novel spectral vision transformer (SViT) to PSG signals for detection of RBD and compare the results to the more conventional convolutional neural network architecture. The vision-based deep learning models were applied to scalograms (30 or 300 s windows) of the PSG data (EEG, EMG and EOG) and the predictions interpreted. A total of 153 RBD (96 iRBD and 57 RBD with PD) and 190 controls were included in the study and 5-fold bagged ensemble was used. Model outputs were analyzed per-patient (averaged), with regards to sleep stage, and the SViT was interpreted using integrated gradients. Models had a similar per-epoch test F1 score. However, the vision transformer had the best per-patient performance, with an F1 score 0.87. Training the SViT on channel subsets, it achieved an F1 score of 0.93 on a combination of EEG and EOG. EMG is thought to have the highest diagnostic yield, but interpretation of our model showed that high relevance was placed on EEG and EOG, indicating these channels could be included for diagnosing RBD.

Quantifying physiological stability in the general ward using continuous vital signs monitoring: the circadian kernel density estimator.

Technological advances seen in recent years have introduced the possibility of changing the way hospitalized patients are monitored by abolishing the traditional track-and-trigger systems and implementing continuous monitoring using wearable biosensors. However, this new monitoring paradigm raise demand for novel ways of analyzing the data streams in real time. The aim of this study was to design a stability index using kernel density estimation (KDE) fitted to observations of physiological stability incorporating the patients' circadian rhythm. Continuous vital sign data was obtained from two observational studies with 491 postoperative patients and 200 patients with acute exacerbation of chronic obstructive pulmonary disease. We defined physiological stability as the last 24 h prior to discharge. We evaluated the model against periods of eight hours prior to events defined either as severe adverse events (SAE) or as a total score in the early warning score (EWS) protocol of ≥ 6, ≥ 8, or ≥ 10. The results found good discriminative properties between stable physiology and EWS-events (area under the receiver operating characteristics curve (AUROC): 0.772-0.993), but lower for the SAEs (AUROC: 0.594-0.611). The time of early warning for the EWS events were 2.8-5.5 h and 2.5 h for the SAEs. The results showed that for severe deviations in the vital signs, the circadian KDE model can alert multiple hours prior to deviations being noticed by the staff. Furthermore, the model shows good generalizability to another cohort and could be a simple way of continuously assessing patient deterioration in the general ward.

Deviations in continuously monitored electrodermal activity before severe clinical complications: a clinical prospective observational explorative cohort study.

Monitoring of high-risk patients in hospital wards is crucial in identifying and preventing clinical deterioration. Sympathetic nervous system activity measured continuously and non-invasively by Electrodermal activity (EDA) may relate to complications, but the clinical use remains untested. The aim of this study was to explore associations between deviations of EDA and subsequent serious adverse events (SAE). Patients admitted to general wards after major abdominal cancer surgery or with acute exacerbation of chronic obstructive pulmonary disease were continuously EDA-monitored for up to 5 days. We used time-perspectives consisting of 1, 3, 6, and 12 h of data prior to first SAE or from start of monitoring. We constructed 648 different EDA-derived features to assess EDA. The primary outcome was any SAE and secondary outcomes were respiratory, infectious, and cardiovascular SAEs. Associations were evaluated using logistic regressions with adjustment for relevant confounders. We included 714 patients and found a total of 192 statistically significant associations between EDA-derived features and clinical outcomes. 79% of these associations were EDA-derived features of absolute and relative increases in EDA and 14% were EDA-derived features with normalized EDA above a threshold. The highest F1-scores for primary outcome with the four time-perspectives were 20.7-32.8%, with precision ranging 34.9-38.6%, recall 14.7-29.4%, and specificity 83.1-91.4%. We identified statistically significant associations between specific deviations of EDA and subsequent SAE, and patterns of EDA may be developed to be considered indicators of upcoming clinical deterioration in high-risk patients.

MSED: A Multi-Modal Sleep Event Detection Model for Clinical Sleep Analysis.

Clinical sleep analysis require manual analysis of sleep patterns for correct diagnosis of sleep disorders. However, several studies have shown significant variability in manual scoring of clinically relevant discrete sleep events, such as arousals, leg movements, and sleep disordered breathing (apneas and hypopneas). We investigated whether an automatic method could be used for event detection and if a model trained on all events (joint model) performed better than corresponding event-specific models (single-event models). We trained a deep neural network event detection model on 1653 individual recordings and tested the optimized model on 1000 separate hold-out recordings. F1 scores for the optimized joint detection model were 0.70, 0.63, and 0.62 for arousals, leg movements, and sleep disordered breathing, respectively, compared to 0.65, 0.61, and 0.60 for the optimized single-event models. Index values computed from detected events correlated positively with manual annotations (r2 = 0.73, r2 = 0.77, r2 = 0.78, respectively). We furthermore quantified model accuracy based on temporal difference metrics, which improved overall by using the joint model compared to single-event models. Our automatic model jointly detects arousals, leg movements and sleep disordered breathing events with high correlation with human annotations. Finally, we benchmark against previous state-of-the-art multi-event detection models and found an overall increase in F1 score with our proposed model despite a 97.5% reduction in model size.

Wireless Single-Lead versus Standard 12-Lead ECG, for ST-Segment Deviation during Adenosine Cardiac Stress Scintigraphy.

Wearable wireless electrocardiographic (ECG) monitoring is well-proven for arrythmia detection, but ischemia detection accuracy is not well-described. We aimed to assess the agreement of ST-segment deviation from single- versus 12-lead ECG and their accuracy for the detection of reversible ischemia. Bias and limits of agreement (LoA) were calculated between maximum deviations in ST segments from single- and 12-lead ECG during 82Rb PET-myocardial cardiac stress scintigraphy. Sensitivity and specificity for reversible anterior-lateral myocardial ischemia detection were assessed for both ECG methods, using perfusion imaging results as a reference. Out of 110 patients included, 93 were analyzed. The maximum difference between single- and 12-lead ECG was seen in II (-0.019 mV). The widest LoA was seen in V5, with an upper LoA of 0.145 mV (0.118 to 0.172) and a lower LoA of -0.155 mV (-0.182 to -0.128). Ischemia was seen in 24 patients. Single-lead and 12-lead ECG both had poor accuracy for the detection of reversible anterolateral ischemia during the test: single-lead ECG had a sensitivity of 8.3% (1.0-27.0%) and specificity of 89.9% (80.2-95.8%), and 12-lead ECG a sensitivity of 12.5% (3.0-34.4%) and a specificity of 91.3% (82.0-96.7%). In conclusion, agreement was within predefined acceptable criteria for ST deviations, and both methods had high specificity but poor sensitivity for the detection of anterolateral reversible ischemia. Additional studies must confirm these results and their clinical relevance, especially in the light of the poor sensitivity for detecting reversible anterolateral cardiac ischemia.

Continuous monitoring is superior to manual measurements in detecting vital sign deviations in patients with COVID-19.

BACKGROUND: Patients admitted to the emergency care setting with COVID-19-infection can suffer from sudden clinical deterioration, but the extent of deviating vital signs in this group is still unclear. Wireless technology monitors patient vital signs continuously and might detect deviations earlier than intermittent measurements. The aim of this study was to determine frequency and duration of vital sign deviations using continuous monitoring compared to manual measurements. A secondary analysis was to compare deviations in patients admitted to ICU or having fatal outcome vs. those that were not. METHODS: Two wireless sensors continuously monitored (CM) respiratory rate (RR), heart rate (HR), and peripheral arterial oxygen saturation (SpO2 ). Frequency and duration of vital sign deviations were compared with point measurements performed by clinical staff according to regional guidelines, the National Early Warning Score (NEWS). RESULTS: SpO2 < 92% for more than 60 min was detected in 92% of the patients with CM vs. 40% with NEWS (p < .00001). RR > 24 breaths per minute for more than 5 min were detected in 70% with CM vs. 33% using NEWS (p = .0001). HR ≥ 111 for more than 60 min was seen in 51% with CM and 22% with NEWS (p = .0002). Patients admitted to ICU or having fatal outcome had longer durations of RR > 24 brpm (p = .01), RR > 21 brpm (p = .01), SpO2 < 80% (p = .01), and SpO2 < 85% (p = .02) compared to patients that were not. CONCLUSION: Episodes of desaturation and tachypnea in hospitalized patients with COVID-19 infection are common and often not detected by routine measurements.

Automatic scoring of drug-induced sleep endoscopy for obstructive sleep apnea using deep learning.

BACKGROUND: Treatment of obstructive sleep apnea is crucial for long term health and reduced economic burden. For those considered for surgery, drug-induced sleep endoscopy (DISE) is a method to characterize location and pattern of sleep-related upper airway collapse. According to the VOTE classification system, four upper airway sites of collapse are characterized: velum (V), oropharynx (O), tongue (T), and epiglottis (E). The degree of obstruction per site is classified as 0 (no obstruction), 1 (partial obstruction), or 2 (complete obstruction). Here we propose a deep learning approach for automatic scoring of VOTE obstruction degrees from DISE videos. METHODS: We included 281 DISE videos with varying durations (6 s-16 min) from two sleep clinics: Copenhagen University Hospital and Stanford University Hospital. Examinations were split into 5-s clips, each receiving annotations of 0, 1, 2, or X (site not visible) for each site (V, O, T, and E), which was used to train a deep learning model. Predicted VOTE obstruction degrees per examination was obtained by taking the highest predicted degree per site across 5-s clips, which was evaluated against VOTE degrees annotated by surgeons. RESULTS: Mean F1 score of 70% was obtained across all DISE examinations (V: 85%, O: 72%, T: 57%, E: 65%). For each site, sensitivity was highest for degree 2 and lowest for degree 0. No bias in performance was observed between videos from different clinicians/hospitals. CONCLUSIONS: This study demonstrates that automating scoring of DISE examinations show high validity and feasibility in degree of upper airway collapse.

Continuous Glucose Monitoring Reveals Perioperative Hypoglycemia in Most Patients With Diabetes Undergoing Major Surgery: A Prospective Cohort Study.

OBJECTIVE: To investigate the frequency and duration of hypo- and hyperglycemia, assessed by continuous glucose monitoring (CGM) during and after major surgery, in departments with implemented diabetes care protocols. SUMMARY BACKGROUND DATA: Inadequate glycemic control in the perioperative period is associated with serious adverse events, but monitoring currently relies on point blood glucose measurements, which may underreport glucose excursions. METHODS: Adult patients without (A) or with diabetes [non-insulin-treated type 2 (B), insulin-treated type 2 (C) or type 1 (D)] undergoing major surgery were monitored using CGM (Dexcom G6), with an electrochemical sensor in the interstitial fluid, during surgery and for up to 10 days postoperatively. Patients and health care staff were blinded to CGM values, and glucose management adhered to the standard diabetes care protocol. Thirty-day postoperative serious adverse events were recorded. The primary outcome was duration of hypoglycemia (glucose <70 mg/dL). Clinicaltrials.gov: NCT04473001. RESULTS: Seventy patients were included, with a median observation time of 4.0 days. CGM was recorded in median 96% of the observation time. The median daily duration of hypoglycemia was 2.5 minutes without significant difference between the 4 groups (A-D). Hypoglycemic events lasting ≥15 minutes occurred in 43% of all patients and 70% of patients with type 1 diabetes. Patients with type 1 diabetes spent a median of 40% of the monitoring time in the normoglycemic range 70 to 180 mg/dL and 27% in the hyperglycemic range >250 mg/dL. Duration of preceding hypo- and hyperglycemia tended to be longer in patients with serious adverse events, compared with patients without events, but these were exploratory analyses. CONCLUSIONS: Significant duration of both hypo- and hyperglycemia was detected in high proportions of patients, particularly in patients with diabetes, despite protocolized perioperative diabetes management.

A Flexible Deep Learning Architecture for Temporal Sleep Stage Classification Using Accelerometry and Photoplethysmography.

Wrist-worn consumer sleep technologies (CST) that contain accelerometers (ACC) and photoplethysmography (PPG) are increasingly common and hold great potential to function as out-of-clinic (OOC) sleep monitoring systems. However, very few validation studies exist because raw data from CSTs are rarely made accessible for external use. We present a deep neural network (DNN) with a strong temporal core, inspired by U-Net, that can process multivariate time series inputs with different dimensionality to predict sleep stages (wake, light-, deep-, and REM sleep) using ACC and PPG signals from nocturnal recordings. The DNN was trained and tested on 3 internal datasets, comprising raw data both from clinical and wrist-worn devices from 301 recordings (PSG-PPG: 266, Wrist-worn PPG: 35). External validation was performed on a hold-out test dataset containing 35 recordings comprising only raw data from a wrist-worn CST. An accuracy = 0.71 ± 0.09, 0.76 ± 0.07, 0.73 ± 0.06, and κ = 0.58 ± 0.13, 0.64 ± 0.09, 0.59 ± 0.09 was achieved on the internal test sets. Our experiments show that spectral preprocessing yields superior performance when compared to surrogate-, feature-, raw data-based preparation. Combining both modalities produce the overall best performance, although PPG proved to be the most impactful and was the only modality capable of detecting REM sleep well. Including ACC improved model precision to wake and sleep metric estimation. Increasing input segment size improved performance consistently; the best performance was achieved using 1024 epochs (∼8.5 hrs.). An accuracy = 0.69 ± 0.13 and κ = 0.58 ± 0.18 was achieved on the hold-out test dataset, proving the generalizability and robustness of our approach to raw data collected with a wrist-worn CST.

Ambulatory Detection of Isolated Rapid-Eye-Movement Sleep Behavior Disorder Combining Actigraphy and Questionnaire.

BACKGROUND: Isolated rapid-eye-movement sleep behavior disorder (iRBD) is in most cases a prodrome of neurodegenerative synucleinopathies, affecting 1% to 2% of middle-aged and older adults; however, accurate ambulatory diagnostic methods are not available. Questionnaires lack specificity in nonclinical populations. Wrist actigraphy can detect characteristic features in individuals with RBD; however, high-frequency actigraphy has been rarely used. OBJECTIVE: The aim was to develop a machine learning classifier using high-frequency (1-second resolution) actigraphy and a short patient survey for detecting iRBD with high accuracy and precision. METHODS: The method involved analysis of home actigraphy data (for seven nights and more) and a nine-item questionnaire (RBD Innsbruck inventory and three synucleinopathy prodromes of subjective hyposmia, constipation, and orthostatic dizziness) in a data set comprising 42 patients with iRBD, 21 sleep clinic patients with other sleep disorders, and 21 community controls. RESULTS: The actigraphy classifier achieved 95.2% (95% confidence interval [CI]: 88.3-98.7) sensitivity and 90.9% (95% CI: 82.1-95.8) precision. The questionnaire classifier achieved 90.6% accuracy and 92.7% precision, exceeding the performance of the Innsbruck RBD Inventory and prodromal questionnaire alone. Concordant predictions between actigraphy and questionnaire reached a specificity and precision of 100% (95% CI: 95.7-100.0) with 88.1% sensitivity (95% CI: 79.2-94.1) and outperformed any combination of actigraphy and a single question on RBD or prodromal symptoms. CONCLUSIONS: Actigraphy detected iRBD with high accuracy in a mixed clinical and community cohort. This cost-effective fully remote procedure can be used to diagnose iRBD in specialty outpatient settings and has potential for large-scale screening of iRBD in the general population. © 2022 The Authors. Movement Disorders published by Wiley Periodicals LLC on behalf of International Parkinson and Movement Disorder Society.

Agreement between standard and continuous wireless vital sign measurements after major abdominal surgery: a clinical comparison study.

Objective. Continuous wireless monitoring outside the post-anesthesia or intensive care units may enable early detection of patient deterioration, but good accuracy of measurements is required. We aimed to assess the agreement between vital signs recorded by standard and novel wireless devices in postoperative patients.Approach. In 20 patients admitted to the post-anesthesia care unit, we compared heart rate (HR), respiratory rate (RR), peripheral oxygen saturation (SpO2), and systolic and diastolic blood pressure (SBP and DBP) as paired data. The primary outcome measure was the agreement between standard wired and wireless monitoring, assessed by mean bias and 95% limits of agreement (LoA). LoA was considered acceptable for HR and PR, if within ±5 beats min-1(bpm), while RR, SpO2, and BP were deemed acceptable if within ±3 breaths min-1(brpm), ±3%-points, and ±10 mmHg, respectively.Main results.The mean bias between standard versus wireless monitoring was -0.85 bpm (LoA -6.2 to 4.5 bpm) for HR, -1.3 mmHg (LoA -19 to 17 mmHg) for standard versus wireless SBP, 2.9 mmHg (LoA -17 to 22) for standard versus wireless DBP, and 1.7% (LoA -1.4 mmHg to 4.8 mmHg) for SpO2, comparing standard versus wireless monitoring. The mean bias of arterial blood gas analysis versus wireless SpO2measurements was 0.02% (LoA -0.02% to 0.06%), while the mean bias of direct observation of RR compared to wireless measurements was 0.0 brpm (LoA -2.6 brpm to 2.6 brpm). 80% of all values compared were within predefined clinical limits.Significance.The agreement between wired and wireless HR, RR, and PR recordings in postoperative patients was acceptable, whereas the agreement for SpO2recordings (standard versus wireless) was borderline. Standard wired and wireless BP measurements may be used interchangeably in the clinical setting.

Forecasting of Continuous Vital Sign Using Multivariate Auto-Regressive Models.

This project assessed the use of multivariate auto-regressive (MAR) models to create forecasts of continuous vital signs in hospitalized patients. A total of 20 hours continuous (1/60Hz) heart rate and respiration rate from eight postoperative patients, where used to fit a centered MAR model for forecasting in windows of 15 minutes. The model was fitted using Markov Chain Monte Carlo sampling, and the model was evaluated on data from five additional patients. The results demonstrate an average RMSE in the forecast window of 11.4 (SD: 7.30) beats per minute for heart rate and 3.3 (SD:1.3) breaths per minute for respiration rate. These results indicate potential for forecasting vital signs in a clinical setting.

End-to-end Deep Learning of Polysomnograms for Classification of REM Sleep Behavior Disorder.

Rapid eye movement (REM) sleep behavior disorder (RBD) is parasomnia and a prodromal manifestation of Parkinson's disease. The current diagnostic method relies on manual scoring of polysomnograms (PSGs), a procedure that is time and effort intensive, subject to interscorer variability, and requires high level of expertise. Here, we present an automatic and interpretable diagnostic tool for RBD that analyzes PSGs using end-to-end deep neural networks. We optimized hierarchical attention networks in a 5-fold cross validation directly to classify RBD from PSG data recorded in 143 participants with RBD and 147 age-and sex-matched controls. An ensemble model using logistic regression was implemented to fuse decisions from networks trained in various signal combinations. We interpreted the networks using gradient SHAP that attribute relevance of input signals to model decisions. The ensemble model achieved a sensitivity of 91.4 % and a specificity of 86.3 %. Interpretation showed that electroencephalography (EEG) and leg electromyography (EMG) exhibited most patterns with high relevance. This study validates a robust diagnostic tool for RBD and proposes an interpretable and fully automatic framework for end-to-end modeling of other sleep disorders from PSG data. Clinical relevance- This study presents a novel diagnostic tool for RBD that considers neurophysiologic biomarkers in multiple modalities.

Detection of Cheyne-Stokes Breathing using a transformer-based neural network.

Annotation of sleep disordered breathing, including Cheyne-Stokes Breathing (CSB), is an expensive and time-consuming process for the clinician. To solve the problem, this paper presents a deep learning-based algorithm for automatic sample-wise detection of CSB in nocturnal polysomnographic (PSG) recordings. 523 PSG recordings were retrieved from four different sleep cohorts and subsequently scored for CSB by three certified sleep technicians. The data was pre-processed and 16 time domain features were extracted and passed into a neural network inspired by the transformer unit. Finally, the network output was post-processed to achieve physiologically meaningful predictions. The algorithm reached a F1-score of 0.76, close to the certified sleep technicians showing that it is possible to automatically detect CSB with the proposed model. The algorithm had difficulties distinguishing between severe obstructive sleep apnea and CSB but this was not dissimilar to technician performance. In conclusion, the proposed algorithm showed promising results and a confirmation of the performance could make it relevant as a screening tool in a clinical setting.

Prediction of Serious Adverse Events from Nighttime Vital Signs Values.

The period directly following surgery is critical for patients as they are at risk of infections and other types of complications, often summarized as severe adverse events (SAE). We hypothesize that impending complications might alter the circadian rhythm and, therefore, be detectable during the night before. We propose a SMOTE-enhanced XGBoost prediction model that classifies nighttime vital signs depending on whether they precede a serious adverse event or come from a patient that does not have a complication at all, based on data from 450 postoperative patients. The approach showed respectable results, producing a ROC-AUC score of 0.65 and an accuracy of 0.75. These findings demonstrate the need for further investigation.

Age estimation from sleep studies using deep learning predicts life expectancy.

Sleep disturbances increase with age and are predictors of mortality. Here, we present deep neural networks that estimate age and mortality risk through polysomnograms (PSGs). Aging was modeled using 2500 PSGs and tested in 10,699 PSGs from men and women in seven different cohorts aged between 20 and 90. Ages were estimated with a mean absolute error of 5.8 ± 1.6 years, while basic sleep scoring measures had an error of 14.9 ± 6.29 years. After controlling for demographics, sleep, and health covariates, each 10-year increment in age estimate error (AEE) was associated with increased all-cause mortality rate of 29% (95% confidence interval: 20-39%). An increase from -10 to +10 years in AEE translates to an estimated decreased life expectancy of 8.7 years (95% confidence interval: 6.1-11.4 years). Greater AEE was mostly reflected in increased sleep fragmentation, suggesting this is an important biomarker of future health independent of sleep apnea.

Prediction of serious outcomes based on continuous vital sign monitoring of high-risk patients.

Continuous monitoring of high-risk patients and early prediction of severe outcomes is crucial to prevent avoidable deaths. Current clinical monitoring is primarily based on intermittent observation of vital signs and the early warning scores (EWS). The drawback is lack of time series dynamics and correlations among vital signs. This study presents an approach to real-time outcome prediction based on machine learning from continuous recording of vital signs. Systolic blood pressure, diastolic blood pressure, heart rate, pulse rate, respiration rate and peripheral blood oxygen saturation were continuously acquired by wearable devices from 292 post-operative high-risk patients. The outcomes from serious complications were evaluated based on review of patients' medical record. The descriptive statistics of vital signs and patient demographic information were used as features. Four machine learning models K-Nearest-Neighbors (KNN), Decision Trees (DT), Random Forest (RF), and Boosted Ensemble (BE) were trained and tested. In static evaluation, all four models had comparable prediction performance to that of the state of the art. In dynamic evaluation, the models trained from the static evaluation were tested with continuous data. RF and BE obtained the lower false positive rate (FPR) of 0.073 and 0.055 on no-outcome patients respectively. The four models KNN, DT, RF and BE had area under receiver operating characteristic curve (AUROC) of 0.62, 0.64, 0.65 and 0.64 respectively on outcome patients. RF was found to be optimal model with lower FPR on no-outcome patients and a higher AUROC on outcome patients. These findings are encouraging and indicate that additional investigations must focus on validating performance in a clinical setting before deployment of the real-time outcome prediction.