Sleep Arousal and Cardiovascular Dynamics

Sleep arousal conventionally refers to any temporary intrusions of wakefulness into sleep. Arousals are usually considered as a part of normal sleep and rarely result in complete awakening. However, once their frequency increases, they may affect the sleep architecture and lead to sleep fragmentation, resulting in fatigue, poor executive functioning and excessive daytime sleepiness. In the electroencephalogram, arousals mostly appear as a shift of power in frequency to values greater than 16 Hz lasting 3-15 seconds. The general objective of this thesis was to investigate on the nature of sleep arousal and study arousal interaction and association with cardiovascular dynamics. At the first step of this research, an algorithm was developed and evaluated for automatic detection of sleep arousal. The polysomnographic (PSG) data of 9 subjects were analysed and 32 features were derived from a range of biosignals. The extracted features were used to develop kNN classifier model in to differentiate arousal from non-arousal events. The developed algorithm can detect arousal events with the average sensitivity and accuracy of 79% and 95.5%, respectively. The second aim was to investigate cardiovascular dynamics once an arousal occurs. Overnight continuous systolic and diastolic blood pressure (SBP and DSP), spectral components of heart rate variability (HRV) and the pulse transit time of 10 subjects (average arousal number of 51.5 +/- 21.1 per person) were analysed before and after arousal occurrence. Whether each cardiovascular variable increases or decreases was evaluated in different types of arousals through slpoe index (SI). The analysis indicated a post-arousal SBP and DBP elevation and PTT dropping. High frequency component of HRV (HF) dropped at arousal onset whilst low frequency (LF) component shifted. HRV spectral components extracted from ECG, lead I alongside with PTT were utilised for sleep staging in 22 healthy and insomnia subjects using linear and non-linear classifier models. Obtained result shows that developed model by DW-kNN classifier could detect sleep stages with mean accuracy of 73.4% +/- 6.4. An empirical curve fitting model for overnight continuous blood pressure estimation was developed and evaluated using the first and second derivatives of fingertip PPG (VPG, APG) along with ECG. The VPG-based model could estimate systolic and diastolic blood pressure with mean error of 3:96 mmHg with standard deviation of 1.41 mmHg and DBP with 6:88 mmHg with standard deviation of 3.03 mmHg. The QT and RR time intervals are two cardiac variables which represent beat to beat variability and ventricular repolarisation, respectively. PSG dataset of 2659 men aged older than 65 (MrOS Sleep Study) was analysed to compare on RR and QT interval variability pre- and post-arousal onset. The cardiac interval gradients were developed to monitor instantaneous changes pre-and post-onset. Analysis of gradients demonstrated that both RR and QT are likely to start shortening several second prior to onset by average probability of 73% and 64%. The QT/RR linear correlation was significantly rising after arousal inducing regardless of arousal type and associated pathological events (Rpost = 0.218 vs Rpre = 0.047). ANOVA test and Tukey’s honest post-hoc analysis indicated a significant difference between cardiac intervals variability between respiratory, movements and spontaneous arousals. In addition, respiratory disturbance index (RDI) as a measure of sleep apnoea severity was reversely correlated with both QT (RVarQT = -0.251, p 1:1 ms) and greater frequency of sleep arousal, less physical activity and medical history of several cardiovascular disease. Similarly participants in quartile DRR> - 8:8 were likelier to be obese with less physical activity, medical history of COPD and stroke and suffered from severer degree of sleep apnoea (RDI = 28:7 20:4 vs RDI = 25:5 +/- 17:6, p < 0:001). Kaplan-Meier analysis showed that the distribution DRR at arousal onset was significantly associated with cardiovascular (CV) mortality (p < 0:001). Cox proportional hazard regression models also indicated the effect of arousal duration in prediction of CV mortality, where longer arousals had more prognostic value for CV mortality than shorter arousals.Thesis (Ph.D.) -- University of Adelaide, School of Electrical and Electronic Engineering, 202