INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Disorders of sleep-related breathing are generally associated with observable abnormalities in the sleep electroencephalogram (EEG). Frequent brief arousals from sleep represent a hallmark disturbance of the EEG in the disorders of obstructive sleep apnea (OSA) and upper airway resistance syndrome (UARS) (1). These brief arousals, also termed "microarousals" (see the American Sleep Disorders Association [ASDA] criteria [4]), most often occur at the termination of each abnormal respiratory event, and are thought to be a major contributing cause of the excessive daytime sleepiness (EDS) common in these and other sleep disorders marked by multiple microarousals (5, 6). Much controversy continues, however, about whether polysomnographically observable sleep-related disturbance episodes that do not result in visually apparent arousals negatively affect sleep quality. We hypothesize that visually undetectable EEG alterations may occur during breathing disturbances in the absence of arousal. If so, these alterations may be clinically relevant.

A few studies have reported quantitative changes in the EEG associated with arousal from apnea (7, 8) when visually apparent changes were observed. Similar quantitative EEG changes may also occur when no visually discernible arousal activity is present. These changes may reflect alterations in brain function as the consequence of a respiratory event, which may negatively affect sleep independently of the arousal phenomenon or as a manifestation of "subarousals." Such alterations could account for part of the incomplete correlation that has been observed between measures of sleep architecture (including arousals) and measures of daytime sleepiness (9, 10).

To further characterize nonvisually detectable changes in the sleep EEG during respiratory events, we analyzed segments of the EEG accompanying periods of increased airflow resistance in subjects with UARS. In this condition, abnormal decrements in esophageal pressure (Pes), which reflects intrathoracic pressure and increased inspiratory effort, are frequently associated with arousals from sleep (so-called Pes events). These excessive decrements in Pes pressure are interpreted as being due to increases in upper airway resistance (1). Additionally, evaluation of polysomnographic data from patients with UARS seen in the Stanford Sleep Disorders Clinic reveals that not all Pes events culminate in an obvious electroencephalographic arousal as defined by traditional criteria.

In view of the now well-established finding that UARS plays a significant role in daytime sleepiness (1, 2, 11), we hypothesized that quantifiable alterations occur in the EEG during UARS-related events before the arousal period both in events terminating in an arousal and in those not so terminating (nonarousal events). Similarly, we also expected that quantifiable EEG changes would occur during the period of usual arousal when no arousal was detected. To our knowledge, this has neither been previously investigated in UARS nor carefully explored in OSA. This report outlines our findings.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subjects, Data Collection, and Data Selection

We evaluated 15 diagnostic polysomnograms from patients in whom UARS was subsequently diagnosed at the Stanford Sleep Disorders Clinic. Polysomnographic data were collected in the standardized fashion, including EEG tracings from leads C3-A2, C4-A1, O1-A2, and O2-A1 (bandpass filtered at 0.3 to 30 Hz); electromyogram (EMG) recordings from the chin and leg; electrocardiographic data (modified lead V2); nasooral airflow (by thermocouple measurement), thoracic and abdominal expansion (with piezoelectric bands); breath-sound intensity with a microphone (anterior neck); Pes (with a water-filled catheter placed behind the left ventricle); and arterial oxygen saturation Sa_O2 via pulse oximetry (Nellcor Inc., Oakland, CA). All physiologic signals were collected via Grass analogue amplifiers and were stored digitally on magnetooptical media, using the SensorMedics (Sensormedics Corporation, Yorba Linda, CA) or Sandman (Nellcor Puritan Bennett, Ottawa, ON, Canada) sleep data storage systems. The sampling rate of the recorded EEG was 100 Hz with the SensorMedics system and 128 Hz with the Sandman system.

Our clinical experience has shown that Pes events are most prevalent in Stage 2, non-rapid eye movement (N-REM) sleep. On the basis of this observation, and because Stage 2 represents the predominant stage of sleep and is generally increased in UARS patients, we selected for study only events occurring in Stage 2 sleep. Sleep stages and states were defined according to the criteria of Rechtschaffen and Kales (12) and all 30-s Stage 2 sleep epochs were identified.

Respiratory events were detected on the basis of a respiratory pattern previously observed to be associated with abnormal breathing during sleep in UARS. These events have been termed Pes crescendos (1). Pes crescendos were defined by consecutively increasing negative inspiratory Pes values, without apnea or hypopnea, followed abruptly by a series of breaths of reduced negative Pes values. To meet the definition, the crescendos had to last for at least 10 s (typically at least three breaths), and the Pes of the breath following the final breath of the crescendo had to be less than 75% of the Pes of the previous breath.

The point of Pes reversal was defined as the point of the peak end-inspiratory Pes of the last breath of the crescendo (i.e., the breath containing the largest Pes value), and was used as a common reference point for the timing of respiratory events and associated EEG signals.

All respiratory events occurring during Stage 2 sleep and meeting the foregoing criteria were captured through use of a cursor-based graphic computer interface in conjunction with a software program that subsequently identified the most negative digitized Pes data point for the identified breath. The period captured included the 26 s before the end of the Pes crescendo and the 26 s after the end of the crescendo. Thus, the total period analyzed for each event was 52 s. Events were discarded if, through visual inspection, a breathing artifact, excessive EMG artifact, or other artifact (obscuring most frequencies) was present in the EEG.

The remaining artifact-free events were subcategorized into two groups: (1) Pes crescendos with associated ASDA-defined arousals (AR) (4); and (2) Pes crescendos without associated ASDA-defined arousals (N-AR). Two experienced researchers independently conducted this classification. In the approximately 15% of cases in which there was a disagreement, the events were rated by a third scorer. The events were then classified according to the majority view, and only then were subjected to spectral analysis.

Data Analysis

Fast Fourier transformations (FFTs) were performed on 4-s, nonoverlapping "windows" (using Hamming windows) of the EEG (C3/A2) data for the periods defined earlier (26 s) before and after Pes reversal. Division of the overall 52-s analysis period by 4-s windows produced 13 consecutive windows per Pes event. Window 7 contained the 2 s before and 2 s after a Pes reversal. FFT calculations were performed on the standard frequency bands of 0.5 to 2.0 Hz (low delta), 2.0 to 4.0 Hz (high delta), 4.0 to 8.0 Hz (theta), 8.0 to 12.0 Hz (alpha), 12.0 to 16.0 Hz (sigma), and 16.0 to 20.0 Hz (beta).

Mean Pes values were calculated for each 4-s window. These were determined by measuring the difference between the Pes at peak inspiratory effort and the Pes at the previous expiration. The Pes values thus represent breath-related effort rather than absolute intrinsic resistance values.

All of the AR segments for an individual subject were averaged to produce mean values for each of the EEG frequency bands, and for Pes, for each 4-s window. The same procedure was applied to the N-AR segments.

Statistical Analysis

Investigation of the distribution characteristics of the EEG variables indicated that the data were suitable for analysis of variance (ANOVA) following a logarithmic transformation. A series of two-way repeated measures ANOVAs was used to assess the effect of time and event type (AR versus N-AR) on data from each frequency band. The time factor consisted of comparison of Window 1 (baseline) with Window 6 (2 to 6 s before the event), Window 7 (2 s before to 2 s after the event), and Window 8 (2 to 6 s after the event). Bonferroni's corrections were made for multiple comparisons. Differences were considered significant if they had values of p < 0.01 after the correction.

Separate ANOVAs, rather than a multivariate ANOVA, were conducted for each frequency band, since the data of interest related to changes within particular frequency bands rather than to overall changes in activation across all frequencies. For each analysis, investigation of the time × event-type interaction term revealed whether the pattern of change in EEG power over time differed as a function of the presence of an AR. Cross-correlation analysis was then used to compare the mean EEG power values within each band for the AR and N-AR data. The same analysis was performed with the Pes data. This form of analysis allows the determination of the overall correlation between two time series. It operates by determining the relationship between two series when they are coincident in time and then when each is lagged by a number of intervals relative to the other. Significant correlations with zero lag indicate that the two series are strongly related, with changes in each series occurring at the same time. Significant correlations at a negative or positive lag indicates that the two series are most strongly related when one is shifted in time relative to the other.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Population

The anthropometric data from the subjects studied are shown in Table 1. Subjects seen during a 6-mo period were randomly selected from a large database of UARS patient charts and polysomnographic recordings maintained at the Stanford Sleep Disorders Clinic.

EEG Findings

An example of an AR segment is presented in Figure 1, and an example of an N-AR segment in Figure 2. The values (mean ± SE) for all EEG bands and Pes are presented for AR and N-AR segments in Figure 3. The Number of AR segments per subject ranged from five to 45 (mean ± SEM = 16.5 ± 3.1). The number of N-AR segments ranged from six to 53 (21.8 ± 3.3). The numbers of each event type per subject during Stage 2 N-REM sleep are shown in Table 1.

View larger version (35K): [in this window] [in a new window]   Figure 1.   Sixty seconds of data extracted from polysomnographic data and showing a representative example of a Pes event consisting of a crescendo followed by an abrupt reversal. In this case the reversal is associated with an ASDA-defined arousal (AR).

View larger version (32K): [in this window] [in a new window]   Figure 2.   Sixty seconds of data showing a Pes crescendo with reversal not associated with an ASDA-defined arousal (N-AR). Data are from the same subject as in Figure 1.

View larger version (22K): [in this window] [in a new window]   Figure 3.   Grand mean data from all 15 subjects for the 13 4-s time windows investigated in EEG tracings. EEG data represent the mean absolute power within each band; Pes data represent the maximum Pes swings seen within each 4-s window. Data from AR and N-AR conditions are overlayed in each panel of the figure. Error bars represent the SEM.

Baseline versus the 4-s epoch preceding Pes reversals. There was a significant main effect of time for the low delta (0.5 to 2Hz) (F[1,14] = 26.8) and high delta (2-4 Hz) bands (F[1,14] = 18.6). In both cases there was significantly more activation in Window 6 than in the baseline. No frequencies showed any main effect of event type or time × event-type interactions. Pes values showed a significant main effect of time (F[1,14] = 65.4), with no effect of event type or time × event-type interaction.

Baseline versus the 4-s epoch surrounding Pes reversals. All frequency bands other than sigma (12 to 14Hz) showed a significant effect of time, indicating a general increase in activation associated with events. This was particularly prominent in the low frequency bands. The F ratios with (1,14) degrees of freedom were: low delta: 205.2; high delta: 188.2; theta: 69.9; alpha: 29.4; and beta: 57.8.

No frequency bands showed a main effect of event type. Only activity in the beta band showed a significant interaction between time and event-type, indicating a larger increase in AR than in N-AR events (F[1,14] = 18.8).

Pes values again showed only a significant main effect of time (F[1,14] = 71.3).

Baseline versus the 4-s epoch following Pes reversals. All frequency bands showed a significant effect of time, indicating a general increase in activation associated with events. In this case the effect was particularly prominent in the beta band. The F ratios with (1,14) degrees of freedom were: low delta: 48.7; high delta: 12.2; theta: 19.4; alpha: 68.1; sigma: 18.9; and beta: 139.1

The alpha and beta bands showed a significant main effect of event type, with more activation with arousal events (alpha [F(1,14) = 7.8]; beta [F(1,14) = 20.4]).

The alpha (F[1,14] = 23.5), sigma (F[1,14] = 16.2), and beta bands (F[1,14] = 54.1) showed significant time × event-type interactions, indicating a greater increase in AR events than in N-AR events relative to baseline.

Pes values showed no main or interaction effects, indicating that for both AR and N-AR events the values in Window 8 had returned to baseline. However, it is clear that whereas the values for Pes in N-AR events remained near baseline, the values for AR events decreased to below baseline levels in Window 9 and remained reduced until Window 13. A post hoc analysis was conducted in which the AR and N-AR means of Pes for Windows 9 through 13 were compared. This indicated that N-AR values for Pes were significantly increased relative to the AR values (t[14] = 3.641, p = 0.003).

Cross-correlation comparison of ASDA-defined and non-ASDA-defined arousals. The pattern of activation for AR and N-AR events was compared across all 13 time windows for each EEG band using cross-correlation analysis. For all frequency bands, the two event types showed substantial cross-correlation that fell outside the 95% confidence limits for chance. The sigma and beta bands showed the strongest relationship, with the arousal time series offset by one window after the N-AR series (sigma: 0.763; beta: 0.902), although in both cases significant cross-correlations were also seen with zero lag (sigma: 0.623; beta 0.816). All other bands clearly showed the strongest cross-correlations without a lag between the series (low delta: 0.966; high delta: 0.907; theta: 0.951; alpha: 0.885).

The two Pes time series had a cross-correlation coefficient of 0.913 at zero lag.

Summary of Results

Pes reversals were preceded by substantial increases in delta activation regardless of whether or not they were eventually associated with an AR. Delta power was increased in 97% of AR events and 95% of N-AR events. In the period surrounding Pes reversals, all EEG frequencies showed a significant increase in activation, with only beta activity showing a significantly larger increase in AR than in N-AR events. In the period following Pes reversals, alpha, sigma, and beta activity showed more of an increase in AR events; all other frequencies showed an increase in activation relative to baseline that was not different for AR and N-AR events. Pes values were significantly increased in the windows before and at the point of Pes reversal in the case of both AR and N-AR events. The N-AR values returned to baseline immediately after Pes reversal, but there was a further significant and sustained reduction if an AR occurred.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The most common complaint of patients with UARS is excessive EDS, often despite presumably adequate amounts of sleep. Reduced sleep effectiveness is the expected cause of EDS in UARS. This reduction in sleep effectiveness has been postulated to result from abnormalities in overall sleep architecture, and more specifically from sleep fragmentation represented by brief EEG arousals at the end of abnormal respiratory events. These events are documented by the presence of successively larger negative inspiratory values of Pes over the course of a few or many consecutive breaths, with abrupt reductions in negative pressure at the termination of each event. We have termed these events Pes crescendo events (3, 13).

The coincidence of EEG arousals and Pes events is well documented, and was noted in our original reports (1, 13). However, a percentage of Pes events terminate without coincident EEG activity meeting criteria for ASDA-defined arousals. We have previously reported this finding (14), and have observed the frequent occurrence of bursts of hypersynchronous delta wave activity with and without AR activity. Following Pes events in the present study the use of EEG spectral analysis facilitated the investigation of visually unobservable changes that may occur at the C3 scalp site. The ability to time-lock events to a single Pes data point, and to average EEG data over a number of events, greatly enhances our ability to uncover even small but consistent EEG changes. We selected only artifact-free Pes events occurring in Stage 2 sleep for this study, in order to control for potential stage-dependent differences in EEG activity and respiratory effort.

Our analysis of EEG activity associated with Pes events occurring during Stage 2 sleep revealed an overall change in the cortical EEG, which was consistent across subjects and was independent of the presence of an ASDA-defined arousal. The most striking effect was seen in the delta frequency bands, which were significantly increased in power over the period from 6 s to 2 s before Pes reversal, with this increase continuing through the event and into the 2 s to 6 s period beyond Pes reversal. As reported, this delta activation was not only highly significant when mean values were compared, but it essentially occurred in every event (97% of AR and 95% of N-AR events). This general finding may be consistent with the findings of Kay and colleagues (15), who reported a strong positive relationship between the degree of upper airway resistance and delta band activity during stage 2 and slow-wave sleep in normal subjects without sleep pathology. However, all frequency bands showed a significant increase relative to baseline in the time windows surrounding and following Pes reversal. Thus, although there was more activation in the high-frequency bands when visually discernible arousals were produced, there was still increased power in most frequency bands even when no visually observable arousals were produced. This is highlighted by the significant zero-lag cross-correlations between AR and N-AR events for all frequency bands. The pattern of abrupt increases in the delta-band power near the termination of Pes crescendos with or without associated ASDA-defined arousals may be indicative of a compensatory central nervous system mechanism for promoting the continuation of sleep  a mechanism that is successful in the case of N-AR events but unsuccessful in AR events.

Other groups have reported findings that complement the present data. Basner and coworkers (16) reported that the presentation of auditory tones during an apnea leads to a reduction in apnea length. Although this was often due to a tone-induced arousal, the effect was also seen in a substantial proportion of trials (22 of 124 trials) in the absence of an arousal. Carley and colleagues (17) found that the presentation of auditory tones to subjects in NREM sleep led to a respiratory arousal (increase in tidal volume and decrease in inspiratory duration) irrespective of whether a scoreable arousal was present in response to the tone. Martin and associates (18) presented tones to subjects at an intensity sufficient to produce changes in blood pressure or heart rate but below the level required to produce ASDA-defined arousals. Alpha-frequency power was shown to be increased after these tones despite the lack of visual arousal criteria. One night of such stimuli led to small but significant increases in daytime sleepiness (multiple sleep latency test and maintenance of wakefulness test) (18).

Other physiologic conditions or stimuli are known to be associated with visually observable hypersynchronous delta waves recorded in the central EEG leads, including somnambulism, in which bursts of delta waves are manifest at the initiation of the episode (19, 20). Additionally, auditory stimulation has been shown to induce delta-wave bursts when not of sufficient intensity to induce complete arousals (6). Moreover, in recent studies, experimentally induced inspiratory occlusions produced EEG changes in subjects who did not display arousals from Stage 2 sleep. These responses were characterized by low-frequency, high-amplitude components in the averaged EEG response in the first second following the occlusions (21).

In our study, Pes values before, during, and immediately after the Pes reversal points were essentially identical regardless of whether or not a visually discernible arousal was present. The major difference occurred in the period following the reversal. In N-AR events the Pes values returned to baseline after the reversal. In AR events the Pes values decreased further and then established a stable level, more negative than the baseline level. It is probable that at baseline (26 s before the arousal), Pes values are already increased due to increased respiratory effort, which is associated with the establishment of stable sleep and may be higher than in a drowsy state. When a non-visually discernible arousal occurs, the respiratory effort returns to the previous baseline level. On the other hand, when alpha and/or beta activity is present in a sufficient degree, lower Pes values are attained. This may be due to wakefulness mechanisms modulating upper airway muscle tone (25) and leading to a larger airway caliber that translates into a lower Pes value.

One fact is clear from the present data. Even if Pes reversal occurs with substantial delta EEG activation, the presence of a visually observable arousal results in a significantly smaller inspiratory effort, for a longer period, than does a reversal not associated with a visually scored arousal. This must have an impact on the disease process in UARs. Pes reversals not associated with arousal may be more detrimental, with the possibility of induction of oscillatory phenomena that have been implicated as one of the elements leading to increasing reductions in airway caliber and the eventual development of apnea (17).

In summary, the results of this study suggest that multiple changes occur in EEG activity during and surrounding the termination of a Pes crescendo event in individuals with UARS, whether or not these changes are visually observable on the EEG tracing. Quantitative polysomnographic analysis may provide better correlates to the daytime symptoms of the tiredness, fatigue, low energy, and/or excessive daytime sleepiness experienced by these patients.