Design Home-Based Healthcare Systems for Sleep Disorders

Abstract

Sleep plays a critical role in physical and mental health development. Insufficient sleep impairs children’s brain development, social emotional and behavioral function, and learning capacity. In this dissertation, we investigated the sleep patterns of 4th- and 5th-grade students in the Z-factor study which was a part of the Sleep to Enhance Participation in STEM (STEPS) project for elementary schools. The study included 257 students enrolled in a Southwestern US school district who participated in Z-factor during the Spring 2016-17 and Fall 2017-18 semesters and met the study inclusion criteria. Participants underwent 5 to 7 days of continuous wrist actigraphy and completed an online sleep diary as part of course curriculum. We found approximately two-thirds of the students slept less than 9 hours per night. The sleep midpoint time on weekends was about 1 hour later than on weekdays (02:56 HH:MM vs 01:51, p < 0.01). The sleep duration of 4th-graders was about 10 minutes longer than 5th-graders (528.72 vs 519.26 minutes, p < 0.03). A larger percentage of 5th-grade students slept less than 9 hours compared to 4th-grade students (76.29% vs 63.75%, p < 0.05). Boys’ had shorter sleep duration than girls (518.96 vs 531.10 minutes, p < 0.01), and a larger percentage of boys obtained less than 9 hours of sleep compared to girls (76.98 vs 60.31%, , p < 0.01). Boys have slightly lower sleep efficiency than girls (85.97 vs 87.15%, p < 0.01). Insufficient sleep and irregular sleep-wake behaviors are also common among adult population. Long term insufficient sleep can cause cardiovascular diseases, diabetics, etc. However, there are hundreds and thousands undiagnosed sleep disorders patients in the US. Sleep-disordered breathing (SDB) is a common type of sleep disorders. Therefore, an easy-to-use SDB screening tool is required. Machine learning provides a powerful set of tools for analyzing clinical data. It can automatically recognize patterns in clinical data and predict on future unseen data. In this dissertation, we developed a machine learning based SDB screening tool incorporating nocturnal pulse oximetry with demographic, anatomic, and clinical data. We used a general population database to train and test the tool. The tool is compatible with 6 categories of apnea-hypopnea index (AHI) with 4% oxyhemoglobin desaturation threshold, ≥ 5, 10, 15, 20, 25, and 30 events/h. The area under receiver operating characteristic curve (AUROC) was 0.904, 0.912, 0.913, 0.926, 0.930, and 0.954, respectively, with models of AHI ≥ 5, 10, 15, 20, 25, and 30 events/h thresholds. The sensitivities of all neural network models were higher than 95%. The AHI ≥ 30 events/h model had the maximum sensitivity: 98.31% (95% CI: 95.01%–100%). In sleep disorder diagnosis, the frequency of cortical arousals is an indicator of sleep quality. Additionally, cortical arousals are used to identify hypopneic events. However, it is inconvenient to record electroencephalogram (EEG) data during home sleep. Fortunately, most cortical arousal events are associated with autonomic nervous system activities that could be observed on an electrocardiography (ECG) signal. ECG data has lower noise and is easier to record at home than EEG. In recent years, deep learning algorithms have shown promise in biomedical data analysis. In this dissertation, we developed a deep learning-based algorithm that uses a single- lead ECG to detect cortical arousal. The algorithm is suitable for use in home sleep tests, long-term in-home healthcare, and emergency care. This study included 1547 polysomnography records that met study inclusion criteria and were selected from the Multi-Ethnic Study of Atherosclerosis (MESA) database for model training, validation, and testing. We developed a deep learning model assembled convolution neural networks and recurrent neural networks which: (1) accepted varying length time series data; (2) directly extracted features from the raw ECG signal; (3) captured long-range dependencies in the time series data; and (4) predicted arousal in one second resolution. We evaluated the model on a test set (n=311). In gross sequence level evaluation, the model produced a 0.621 area under precision-recall curve (AUPRC) and 0.928 area under receiver operating characteristic curve (AUROC). In event level evaluation, the number of ground truth arousals and number of predicted arousals were strongly correlated (r = 0.81, p < 0.0001 , 95% confidence interval: 0.77-0.85).