A state transition-based method for quantifying EEG sleep fragmentation

Abstract

Sleep fragmentation is the predominant factor causing excessive daytime sleepiness in diseases such as sleep apnea and periodic leg movement syndrome. The reference standard for quantifying sleep fragmentation is the arousal index (ArI), which is defined as the average number of arousals per hour of sleep. Arousal scoring is tedious and subjective resulting in considerable inter- and intra-rater variability. Moreover, ArI is only weakly correlated with other indicators of sleep fragmentation such as the total sleep time (TST) and the sleep efficiency (SE). This introduces consistency problems, making the ArI difficult to interpret in practice. In this article, we address these issues by proposing a novel measure of sleep fragmentation termed the weighted-transition sleep fragmentation index (χ). This new measure is derived by capturing the different sleep states transitions and assigning weights to them. A significant correlation was found between χ and all other indices of sleep fragmentation (r = 0.72, σ = 0.0001, r = −0.59, σ = 0.001, r = −0.72, σ = 0.0001, respectively, for ArI, TST and SE. These results suggest that χ is an accurate and useful tool for clinical practice.

Appendices

Appendix A: Standard definitions

EEG arousal (ArI) is defined as abrupt shift in EEG frequency, which may include theta, alpha activity and/or frequencies greater than 16 Hz (but not sleep spindles) subjected to the following scoring rules:

• the subject must be asleep for a minimum period of 10 s before declaring an arousal event,

• EEG frequency shift must be sustained for a 3 s duration or more, and,

• EEG arousal from REM sleep requires presence of simultaneous increase in the sub mental EMG amplitude.

Respiratory disturbance index (RDI) is defined as the average number of apnea/hypopnea events per hour of sleep.

Arousal index (ArI) is defined as the average number of arousal events per hour of sleep.

Total time in bed (TTB) is the time spent on the bed from the ‘lights off’ when the recording starts in the night to the ‘lights on’ in the morning when the recoding ends.

Total sleep time (TST) is defined as the actual sleep time (total time spend in REM and NREM sleep).

Sleep efficiency (SE) is defined as the percentage ratio 100X(TST/TTB) %.

Periodic leg movement index (PLMI) is defined as the average number of periodic leg movement events per hour of sleep.

Appendix B: Epworth sleepiness scale [11]

The Epworth sleepiness scale is used to determine the level of daytime sleepiness. It is a set of questionnaire. Each question rates the chance that a person would doze off or fall asleep during different routine daytime situations. Answers to the questions are rated on a scale from 0 to 3, with 0 meaning person would never doze or fall asleep in a given situation, and 3 meaning that there is a very high chance that person would doze or fall asleep in that situation. An overall score of 10 or more is considered sleepy. A score of 18 or more is very sleepy.

Appendix C: Optimization

In the Method Sect. 2.2, we converted the hypnogram data into the hypnogram time series (HTS) by assigning a number (E _k) to each data epoch. The number assigned to each epoch depends on the sleep stage of that epoch. Results shown in the work of this article were generated using the numbers given in the Table 1. These numbers were selected after a careful search process, such that the correlation between the new index χ and traditional indices ArI, TST and SE can be optimized. For this, we defined E _k as a set of numbers using (5)

$$ E_{\text{K}} = \left\{ {\begin{array}{*{20}c} {{\text{SW}},} & {{\text{S}}1,} & {{\text{S}}2,} & {{\text{S}}3,} & {{\text{S}}4,} & {\text{SR}} \\ \end{array} } \right\} $$

(5)

In (5), SW, S1, S2, S3, S4, and SR are the numbers or the values assigned to state, Stage Wake, Stage 1, Stage 2, Stage 3, Stage 4 and Stage REM respectively. To assign the value to SW, S1, S2, S3, S4 and SR we created six sets of number, each for six states, as shown below

$$ {\text{SW}} \in \left\{ 0 \right\},{\text{ S1}} \in \left\{ {0,1,2,3,4} \right\},{\text{S2}} \in \left\{ {0,1,2,3,4,5} \right\},{\text{S3}} \in \left\{ {0,1,2,3,4,5,6} \right\},{\text{S4}} \in \left\{ {0,1,2,3,4,5,6,7} \right\}{\text{and SR}} \in \left\{ {0,1,2,3,4,5,6,7} \right\} $$

A particular sleep state can have any value for its set, however, under a limitation. The limitation is ‘light sleep stages can never have value greater than deeper sleep stage’, for e.g. if S1 = 3 then S2, S3 and S4 will be always ≥3. Similarly if S2 = 4 then S3, S4 will be ≥4. This applies only to S1, S2, S3 and S4 sleep stages. SR can have any value from its set irrespective of values of other sleep stage. By changing the values given to SW, S1, S2, S3, S4 and SR in (5) and following the above rules, we generated different sets of E _K. For each set of E _k, we computed χ index for all the patients and evaluated the correlation between χ and sleep parameters.

Figure 5 shows the correlation plot for all the E _k sets. In the figure, x-axis represents the sets E _k with different sets of values. According to Fig. 5, correlation between χ and ArI is highest (r = 0.76) when E _k = [0, 4, 4, 6, 6, 3], but the correlation between χ and other parameters was low, r = −0.48 and r = −0.62 for TST and SE, respectively. However, these values are still comparable to those between ArI-TST (r = −0.36) and ArI-SE (r = −0.66). With these values, it is difficult to interpret the sleep fragmentation and daytime sleepiness in the clinical practice. In the results presented in this article (Sect. 3), we propose to select E _k = [0, 3, 4, 5, 7, 3]. At these values of sleep states, we sacrifice the correlation between χ-ArI by a small amount (r = 0.72, difference = 0.04, decreased by 5.2%), however the correlation between χ-TST and χ-SE increases by a significant amount (r = −0.59 (increased by 22.9%) and −0.72 (increased by 16.1%), respectively.

Fig. 5

Optimization of the values given to different sleep states. Graph shows the variation in correlation between χ and ARI/TST/SE/ESS when different sets of numbers (E _k) were given to six sleep stages. We can see from the figure that the correlation between χ and sleep parameters depends on the values given to sleep states