INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The main reason for treating obstructive sleep apnea (OSA) is to control excessive daytime sleepiness. However, the correlations between measures of OSA severity (either respiratory or EEG) and assessments of daytime sleepiness or vigilance are not close (1). This reduces the value of the sleep study in identifying those patients who will gain symptomatic response to a treatment such as nasal continuous positive airway pressure (CPAP). There could be many reasons for this poor correlation. For example, there is considerable night-to-night variation in sleep apnea severity (8), and a single hospital sleep study cannot fully represent what happens normally, whereas the symptoms are a cumulative result of this variable sleep fragmentation. Patients may also vary their nocturnal sleep time, or take extra daytime sleep, to compensate for their fragmented sleep. There may also be considerable variation in the extent to which a patient is forced to wake up with each respiratory event to reestablish adequate airflow across the pharynx (13, 14). Rees and colleagues (15) have shown that not all apneas end in visible EEG arousals, and that the proportion of such "invisible" events varies between individuals.

A recently developed EEG analysis technique, based on neural net processing, allows a quantitative second-by-second description of sleep "depth" (16), and thus the pattern and extent of sleep fragmentation around a respiratory event can be explored in greater detail. Variation in the size and nature of arousals in OSA may explain some of the differences in the degree of daytime symptoms that do not appear to be described adequately by the simple indices of disease severity in current use.

This report describes the results of using this technique to explore differences in the magnitude of arousal that terminate respiratory events in sleep apnea, in particular, interindividual variations, any differences caused by posture, and whether the events are hypopneas or apneas. This newer technique is also compared with the more traditional approach of measuring the length of visible arousals on the EEG (17). Supine posture worsens OSA (18) and increases the pharyngeal closing pressure by about 2 cm H₂O (21) and may therefore increase the degree of pharyngeal muscle tone needed to reopen the pharynx. In a similar way, apneas may need more pharyngeal activation than do hypopneas, if only to overcome the surface tension forces that hold wet mucosal surfaces together (21).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subjects

Thirty all-night polysomnographic tracings obtained from previous research studies (27, 28) on patients with severe OSA (apnea/hypopnea index [AHI] > 30/h) were reviewed in order to identify patients who had several apneas and hypopneas in the same posture, all within an hour, and all in non-REM sleep. In addition they were reviewed to identify patients who had either apneas or hypopneas in both supine and decubitus posture, again within an hour and in non-REM sleep: apneas and hypopneas were not mixed within a subject. The original research studies were approved by the local research ethics committee.

Techniques

The polysomnographic recordings (MPA II; Oxford Medical, Abingdon, UK) had been performed in our sleep laboratory after a simplified diagnostic study using the Visi-Lab system (Stowood Scientific Instruments, Oxford, UK). Recordings consisted of electroencephalogram (EEG) (Fp1 or 2 versus contralateral mastoid), electrooculogram (EOG), and the chin electromyogram (EMG). A frontal EEG lead was used to improve recognition of brief arousals (29). Airflow at the nose and mouth were recorded using thermistors, ribcage and abdominal movements were monitored by inductance plethysmography, oxygen saturation (Sa_O2) was recorded by oximetry, and posture was identified from video recordings.

Sleep stage was defined according to standard criteria (30). All tracings were reanalyzed, apneas were defined as oronasal airflow falling for 10 s or more to less than 20% of the preceding few breaths, and hypopneas were similarly defined but with a 50% threshold. No Sa_O2 criterion was used to define hypopneas. Microarousals associated with these respiratory events were defined according to standard criteria (17).

The neural network EEG analysis has been fully described elsewhere (16). Briefly, the EEG was digitized to 8-bit accuracy at 128 Hz and processed using a neural network trained on a frequency-domain representation of the EEG, which is calculated every second on sequential 1-s segments (Questar; Oxford Medical Ltd.). This representation is based on an autoregressive model of the EEG that characterizes the dominant frequencies in the EEG for each second. Consensus scored segments of EEG corresponding to wakefulness, light/REM sleep, and deep sleep make up the training database for the network. The neural network outputs are probability estimates of these three states and thus allow interpolation between states. Unlike conventional sleep staging, the EEG is quantified on a continuous scale that is not linearly related to conventional sleep stages but tracks the whole sleep/wake continuum. A neural network output of +100 indicates wakefulness, a value of zero indicates REM/light sleep, and a value of 100 indicates Stage 4 sleep.

Analysis

Each patient's sleep study was reviewed to identify periods containing apneas or hypopneas, while supine or decubitus, occurring within a period of an hour and in non-REM sleep. A minimum of nine respiratory events per situation (maximum, 32; mean, 18) were identified and analyzed using the neural net processing and microarousal scoring. The period analyzed centered around the end of the apnea or hypopnea (identified as the sudden rise in nasal flow) and extended from 15 s before the break point to 30 s after. For each patient and situation, apnea or hypopnea, supine or decubitus, the neural net patterns from the contributory events were time-averaged across the 45 s to provide an average description of the EEG changes. The degree of sleep disturbance described by this time-averaged neural net tracing was then quantified by calculating the standard deviation (SD) of the 45 points (45 at 1 per second).

In addition, the lengths of the events classified as microarousals according to the rules of the American Sleep Disorders Association (ASDA) (17) were measured (to a maximum of 30 s). An ASDA brief arousal in non-REM sleep is defined as a speeding of the EEG lasting more than 3 s. Events lasting less than 3 s were assigned a length of zero.

The contribution to the variance in the SD values obtained from apnea versus hypopnea (10 patients), decubitus versus supine (12 patients), and also between subjects for each group was compared using analysis of variance (ANOVA; SAS statistical software; SAS Institute, Cary, NC) (31). The lengths of the ASDA microarousals for each situation were similarly analyzed.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Out of the 30 polysomnograms reviewed we could find only 10 where both apneas and hypopneas occurred in the same posture, all within an hour, and in non-REM sleep: we could find only 12 where apneas (or hypopneas) occurred both supine and decubitus, all within an hour, and in non-REM sleep. Six patients were common to both sets of data. Representative examples of the time-averaged data for eight of the subjects, apnea versus hypopnea (four patients), supine versus decubitus (four patients), are shown in Figures 1 and 2. The neural net SD data across the 45 s for all subjects are shown in Figure 3; apnea versus hypopnea mean data, 27.6 (SD, 8.0) and 21.9 (SD, 9.9); supine versus decubitus, 26.1 (SD, 11.9) and 26.8 (SD, 8.8). The length of the ASDA-defined arousals for each situation are shown in Figure 4; apnea versus hypopnea mean data, 12.6 s (SD, 1.7) and 9.9 s (SD, 2.4); supine versus decubitus, 11.5 s (SD, 3.5) and 11.7 s (SD, 2.2). The ANOVA results of these data are shown in Table 1. There were large, highly significant, interindividual variations in both patient groups in the degree of sleep disturbance measured using the neural net technique, but not when using the length of the visually evident arousal. With both techniques there was a significant difference between apneas and hypopneas, with apneas being more sleep-disturbing. There was no significant difference between respiratory events occurring supine versus decubitus using either technique, neural net or visual arousal length.

View larger version (34K): [in this window] [in a new window]   Figure 1.   Neural network tracings from four individual patients experiencing apneas (left panels) and hypopneas (right panels). Each tracing starts 15 s before apnea or hypopnea break and extends to 30 s after. Each tracing is the time-averaged result of several individual apneas or hypopneas (mean = 18 events). The vertical axis extends from 100 (in deep non-REM sleep) to +100 (fully awake), and the error bars are each point's SEM.

View larger version (34K): [in this window] [in a new window]   Figure 2.   Neural network tracings from four individual patients experiencing obstructive events supine (left panels) and decubitus (right panels). Each tracing starts 15 s before apnea or hypopnea break and extends to 30 s after. Each tracing is the time-averaged result of several individual obstructive events (mean = 18 events). The vertical axis extends from 100 (in deep non-REM sleep) to +100 (fully awake) and the error bars are each point's SEM.

View larger version (16K): [in this window] [in a new window]   Figure 3.   Comparison of the degree of EEG disturbance (based on the neural network analysis) for all subjects comparing apneas versus hypopneas (10 patients) (left panel ), 27.6 (SD, 11.9), 21.9 (SD, 9.9), respectively, and supine versus decubitus (12 patients) (right panel ), 26.1 (SD, 11.9) and 26.8 (SD, 8.8), respectively. The higher the standard deviation the more EEG disturbance around the obstructive event.

View larger version (14K): [in this window] [in a new window]   Figure 4.   Comparison of the length of EEG arousal (ASDA microarousal length) for all subjects comparing apneas versus hypopneas (10 patients) (left panel ), 12.6 s (SD, 1.7) and 9.9 s (SD, 2.4), respectively, and supine versus decubitus (12 patients) (right panel ), 11.5 s (SD, 3.5) and 11.7 s (SD, 2.2), respectively.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

This study has shown that there are significant interindividual variations in the degree of EEG disturbance resulting from obstructive respiratory events. This could be shown using an objective EEG analysis technique using neural network processing (that does not use a threshold to define arousal) but was not shown by the conventional approach measuring the visible length of the arousal. In addition, using either technique, hypopneas were significantly less sleep-disturbing than were apneas, but, contrary to our hypothesis, it did not seem that supine events were more disturbing than decubitus events.

This difference between apneas and hypopneas, as defined by thermistor tracings, may relate to surface forces (21). Issa and Sullivan (21) showed that the CPAP pressure needed to open an already closed airway in sleeping patients with OSA was greater than the pressure at which the airway closed again during pressure reduction. They suggested that once the pharynx has collapsed, then surface tension forces from the apposing wet mucosal surfaces require an extra dilating force to overcome them (22): during hypopnea the mucosal surfaces are not apposed. Alternatively, the occurrence of hypopneas rather than apneas may simply be due to a variation in collapsibility at that time, thus a smaller return of pharyngeal tone, and hence sleep disturbance, would be needed to reopen the airway. In this study we were careful to minimize this effect by comparing apneas and hypopneas from each subject in the same posture, sleep state, and time of night. Hence there is unlikely to have been any substantial difference in the tendency to collapse.

The failure to show any difference between the degree of arousal supine versus decubitus was not what we had expected. Again, Issa and Sullivan (21) showed that on average the pressure required to hold open the pharynx was about 2 cm H₂O higher supine than decubitus, presumably because of the extra collapsing forces related to jaw and tongue weight. However, it may be that in patients with severe OSA, where lateral pharyngeal wall compression may be more important than tongue mass (32), gravitational direction may become less critical. In patients with severe OSA there is little difference in the AHI decubitus versus supine (20), in contrast to the patients with less severe OSA (19).

There are certain assumptions made in the analysis of the EEG by the neural network technique in order to arrive at a single figure for sleep depth (16). The spectrum of awake to deeply asleep is represented from +100 to 100, respectively. Although this is a linear scale it should not be thought that a certain rise in the level (say 30 points) means a similar degree of lightening of sleep depth, e.g., from 90 to 60 and from 50 to 20. Quantification of sleep depth by any current technique cannot be linearly related to brain state since we do not understand the neurophysiology of sleep and the arousal process. Thus our results might have been biased by differences in the starting level of sleep depth in the different situations. However, this was not the case since there were no significant differences between the neural net starting sleep depth, using the first 10 s of each 45-s analysis period, depending upon whether the subsequent event was an apnea/hypopnea, 10 (SD, 28) versus 9 (SD, 23), p > 0.5; or supine/decubitus, 13 (SD, 27) versus 6 (SD, 27), p = 0.4. There was a significant correlation overall between sleep depth in the first 10 s and the SD across the whole analysis period (45 s), r = 0.56, p < 0.001. This is not surprising since the deeper the initial sleep state, the greater the range of the sleep/wake spectrum that could be crossed during arousal.

The average length of arousals measured using ASDA criteria was more than 10 s, and greater than 90% of the respiratory events produced arousals exceeding 3 s. This is longer, and a greater percentage, than found by others, and it may relate to our use of a frontal electrode rather than the more conventional C3 or C4 positioning (15). EEG arousals are more sensitively detected using frontal electrodes (29), and it would be reasonable to suggest that they would thus also be a little longer. The similarity between the results obtained from conventional measurement of arousal length and the neural network processing when comparing apneas and hypopneas is reassuring. The neural net technique measures variation in the EEG, even if it does not rise above a threshold that would be recognized visually as an arousal. Thus, the neural network technique would be expected to give different results under certain circumstances, and indeed the overall correlation between the two techniques (neural network SD versus ASDA length) is only 0.27 (95% confidence interval, 0.03 to +0.53). Thus, the two techniques measure different aspects of the arousal process, and this probably explains why the results looking at interindividual variation were different.

These data offer possible explanations for some of the poor correlations seen between sleep study measures of sleep apnea severity (e.g., AHI and microarousal scoring) and daytime measures of performance (e.g., sleep propensity and vigilance tests). Such correlations rarely exceed 0.6 (1, 28), even in the best studies on groups of patients with a wide range of sleep apnea severity. Thus, well over half of the variance in daytime function is left unaccounted for. It is not yet clear how much of this might be ascribed to the differences in arousal magnitude identified in this study. However, a different study from our laboratory looking at EEG variability across the whole night (using the same neural net analysis) has found this to be an additional independent predictor of daytime sleepiness, even after allowing for both respiratory and EEG (ASDA-defined arousal) measures of OSA severity (28). These two results both suggest that increasing the sophistication of the EEG analysis during disturbed sleep may help to identify which obstructive events are pathophysiologically important. Further studies using other measures of arousal "size" are needed.

Bonnet (33) showed in studies on normal subjects that the size of auditory-induced arousals across a night apparently do not influence the degree of objective sleepiness or vigilance the next day. However, some of the subjective symptom scales indicated more effect from larger arousals. He subjected 11 normal subjects to arousals every 2 min throughout two sequential nights on three separate occasions. On one pair of nights the stimulus was applied until subjects gave a verbal response, on another pair of nights all that had to be elicited was a quarter-turn body movement, and on another pair of nights only EEG speeding was required (length unspecified). Compared with the control night, all the three arousal "strengths" led to a similar reduction in a single morning sleep latency test, but there was more mood disturbance and subjective sleepiness after the larger arousals. However, the smallest recurrent arousals, requiring only EEG speeding, were equally disruptive to normal sleep architecture, removing all Stage 4 sleep. This suggests that these EEG arousals were quite long and perhaps longer than many of the microarousals resulting from OSA (15), and therefore Bonnet's experiment may not have fully encompassed the spread of arousal size reported here.

We have shown that all arousals caused by obstructive respiratory events at night are not equal, and that there is much interindividual variation between patients in the degree to which each event disturbs the EEG. Contrary to expectation, there seemed to be no difference in the degree of EEG disturbance supine versus decubitus. However, apneas are more disturbing than hypopneas, and thus it may not be appropriate to regard them as equal when assessing overall sleep apnea severity and possible effects on daytime symptoms. The development of methods to reliably identify which obstructive events are most disruptive to sleep may improve the ability of sleep studies to identify those patients with symptomatic OSA who are likely to respond to nasal CPAP.