A Novel Quantitative Arousal-Associated EEG-Metric to Predict Severity of Respiratory Distress in Obstructive Sleep Apnea Patients

Abstract

Respiratory arousals (RA) on polysomnography (PSG) are an important predictor of obstructive sleep apnea (OSA) disease severity. Additionally, recent reports suggest that more global indices of desaturation such as the hypoxic burden, namely the area under the curve (AUC) of the oxygen saturation (SaO2) PSG trace may better depict the desaturation burden in OSA. Here we investigated possible associations between a new metric, namely the AUC of the respiratory arousal electroencephalographic (EEG) recording, and already established parameters as the apnea/hypopnea index (AHI), arousal index and hypoxic burden in patients with OSA. In this data-driven study, polysomnographic data from 102 patients with OSAS were assessed (32 female; 70 male; mean value of age: 52 years; mean value of Body-Mass-Index-BMI: 31 kg/m²). The marked arousals from the pooled EEG signal (C3 and C4) were smoothed and the AUC was estimated. We used a support vector regressor (SVR) analysis to predict AHI, arousal index and hypoxic burden as captured by the PSG. The SVR with the arousal-AUC metric could quite reliably predict the AHI with a high correlation coefficient (0,58 in the training set, 0,65 in the testing set and 0,64 overall), as well as the hypoxic burden (0,62 in the training set, 0,58 in the testing set and 0,59 overall) and the arousal index (0,58 in the training set, 0,67 in the testing set and 0,66 overall). This novel arousal-AUC metric may predict AHI, hypoxic burden and arousal index with a quite high correlation coefficient and therefore could be used as an additional quantitative surrogate marker in the description of obstructive sleep apnea disease severity.

Keywords: AHI; ODI; area under the curve of arousal; arousal index; hypoxic burden; polysomnography; sleep apnea; support vector regressor.

FIGURE 1

The raw EEG signal (in blue) for a duration of 30 seconds is shown in the top plot with a significant respiratory arousal. In the bottom blot (in red) the smoothed version of the EEG signal is shown to estimate the area under the curve.

FIGURE 2

Pipeline figure to show the procedure from the raw EEG data, estimating the area under the curve (AUC) and then followed by the linear as well as the SVM regression.

FIGURE 3

Shows the linear regression results between the arousal-AUC and SpO2-AUC for patients with AHI 15–30/h of TST. The green margins indicate the standard deviation for the correlation and the black dots indicate each subject in this group. r = 0.280/p = 0.056/n = 47.

FIGURE 4

Shows the linear regression results between the arousal-AUC and SpO2-AUC for patients with AHI >30/h of TST. The orange margins indicate the standard deviation for the correlation and the black dots indicate each subject in this group. r = 0.404/p = 0.002/n = 55.

FIGURE 5

Shows the correlation coefficient between the arousal-AUC and SpO2-AUC for all arousals of every individual of the two separate groups (Group A: AHI 15–30/h of TST; Group B: AHI >30/h of TST).

FIGURE 6

Shows the linear regression results between the AHI and SpO2-AUC for patients with AHI 15–30/h of TST. The green margins indicate the standard deviation for the correlation and the black dots indicate each subject in this group. r = -0,0580/p = 0,6987/n = 47.

FIGURE 7

Shows the linear regression results between the AHI and SpO2-AUC for patients with AHI >30/h of TST. The orange margins indicate the standard deviation for the correlation and the black dots indicate each subject in this group. r = -0,0480/p = 0.728/n = 55.

FIGURE 8

Box plot with the distribution of a data set. The x axis shows the division into training set, test set and overall testing. The y-axis shows the SVM correlation coefficient.