INTRODUCTION

Top Abstract Introduction Methods Results Discussion References

AROUSAL FROM non-rapid-eye-movement (NREM) sleep during obstructive sleep apnea (OSA) appears to occur when the progressive increase in inspiratory effort reaches a threshold that is independent of the particular combination of hypoxia and hypercapnia stimulating ventilatory effort (5, 15, 19, 28, 29). Thus the maximum level of inspiratory effort preceding arousal serves as an index of the arousal threshold to airway occlusion. It has been appreciated that the mean arousal threshold varies between different patients and can be influenced by central nervous system depressants (2, 3), sleep fragmentation (4, 9, 22), and therapy of sleep apnea (8). However, the arousal threshold also shows within-night variability in a single patient (2, 18, 20, 28, 29). The nature and causes of this variability are not well understood. Change in sleep stage is one likely cause (2, 17, 30). However, there is still considerable variability in the arousal threshold within a given stage of sleep. For example, patients with OSA exhibit variability in arousal threshold over the night in NREM sleep, even though the majority of time is spent in sleep stages 1 and 2. Although the division of NREM sleep into only four stages is a crude separation, there is no alternative well-established electroencephalogram (EEG) correlate of the depth of sleep. Inasmuch as stages 3 and 4 NREM sleep are associated with increased delta wave activity and a higher arousal threshold (2, 17, 30), one might expect the delta power obtained from spectral analysis of the EEG to be an index of the depth of sleep. In fact, one study of the arousal threshold to acoustic stimuli in rats found a correlation between the threshold and the delta power of the EEG (21). However, although the delta power decreases over the night in normal subjects (7, 11), the arousal threshold to airway occlusion may actually increase (20). Even so, the delta power might provide a marker for fluctuations in the level of arousability within a given cycle of NREM sleep. In normal subjects the delta power increases after the onset of a cycle of NREM sleep and then decreases during periods of rapid-eye-movement (REM) sleep (7, 11). Thus fluctuations in EEG delta power also provide a continuous marker of the cycles of sleep. The goals of this study were to describe the within-night variability in the maximum esophageal pressure deflection preceding apnea termination (DP_max) as an index of the arousal threshold and then to correlate these changes with fluctuations in the delta power of the central EEG.

	METHODS

Top Abstract Introduction Methods Results Discussion References

Nine patients with severe OSA, defined as an apnea + hypopnea index >40 events/h on a previous study, were studied on a single night in the sleep laboratory. The project was approved by the Human Studies Subcommittee of our hospital. Written informed consent was obtained from all subjects before they participated in the study.

Sleep was monitored using two pairs of EEG leads (C₄-A₁ and O₂-A₁), two pairs of electrooculogram leads, and chin electromyogram leads using standard techniques (23). An electrocardiogram lead was also monitored. The arterial O₂ saturation was also measured continuously using pulse oximetry (model 3700, Ohmeda, Boulder, CO). Airflow was detected using a mask with a built-in pneumotachograph worn over the nose and mouth and kept in place with head straps. Deflections in esophageal pressure during occluded efforts were determined using a soft fluid-filled catheter (12). The catheter, a 5-Fr polyurethane pediatric feeding tube (Sherwood Medical, St. Louis, MO), was inserted through one nostril and swallowed into the esophagus. The tip was positioned 34-36 cm from the nares to obtain tracings with the smallest amount of cardiac artifact. Before insertion of the catheter, 1-2 ml of 4% lidocaine was dripped into one nostril to minimize discomfort during catheter insertion. The application of lidocaine occurred 1 h before sleep monitoring began. The catheter was well tolerated without additional anesthesia. The catheter was connected to a disposable Transpac II pressure transducer (Abbot Critical Care, North Chicago, IL). The system was calibrated with a water manometer. The transducer system contained a valve that allowed a slow infusion of fluid (normal saline) at 3-4 ml/h to maintain catheter patency.

All measured variables were recorded on a 12-channel polygraph (model 78D, Grass Instrument, Quincy, MA) using a paper speed of 10 mm/s. The central EEG amplifier was calibrated so that 50 µV resulted in a pen deflection of 1 cm. Low- and high-frequency filter settings (one-half amplitude frequency) of 0.3 and 100 Hz, respectively, were used as well as a 60-cycle notch filter. The data were also acquired digitally by recording the output of the polygraph amplifier (J6 output jack) at a sampling rate of 100 Hz using an analog-to-digital board (Dataq Instrument, Akron, OH) and stored on an IBM-compatible computer. Before patient recording, calibration signals were also digitally acquired for the central EEG channel. The digitally acquired signal was multiplied by an appropriate factor for conversion to microvolts.

Data analysis. Sleep was staged in 30-s epochs from the polygraph tracings according to standard criteria (23). Total sleep time was the minutes of stages 1-4 and REM sleep. The sleep period time was total sleep time plus wakefulness after sleep onset, i.e., the time after sleep onset but before the final awakening. We defined an arousal on the basis of criteria proposed by the American Sleep Disorders Association (1). An obstructive apnea was defined as an absence of flow signal from the mask covering the nose and mouth for 10 s associated with continued evidence of respiratory effort (esophageal pressure deflection). A hypopnea was defined as a reduction in airflow to <50% of baseline lasting for 10 s. Obstructive hypopnea was defined as a hypopnea associated with an increase in esophageal pressure deflection.

EEG signal processing. Spectral analysis of the central EEG was performed by analyzing the data in sequential nonoverlapping 1-s windows using the autoregressive model (6, 14). The advantage of this method over the conventional Fourier transform is that it is not critically dependent on data at the borders of the time window being analyzed. The delta power was defined as the power from 1 to 4 Hz. Using this method, we computed a delta power for each second (delta power-s) during the entire night. An average delta power was then determined for each sequential 30 s of the data file by averaging the delta power-s values for that time interval. The digital data (EEG, electrooculogram, electromyogram, airflow, and esophageal pressure) were manually viewed against elapsed time using a computer program (Windaq, Dataq Instrument). This allowed precise timing of all events (apnea onset/cessation, arousals, artifacts). The delta power-s values during arousals at apnea termination, during artifacts in EEG signal, and during isolated K complexes were manually excluded from the 30-s averages. Each value of DP_max (apneas and hypopneas) was paired with the 30-s delta power value with which it was associated. For example, the DP_max of the apnea ending at 85 s was paired with the average delta power for 61-90 s. When there was no event for a given 30 s, a value of DP_max was assigned by linear interpolation. When short periods of wakefulness intervened, delta power and DP_max values were assigned by linear interpolation. When longer periods of wakefulness (>10 min) separated periods of sleep, the subsequent sleep was time shifted to eliminate this gap in sleep. We thus obtained a value of DP_max and delta power for each 30-s time interval of sleep.

Analysis of delta power and DP_max vs. time (epoch) data. Delta power (30-s average) and DP_max were plotted against time (epoch number). Periodicity was analyzed by performing spectral analysis of the delta power vs. time and DP_max vs. time data using the fast Fourier transform method and phase shift and correlation by cross-correlation analysis (Matlab, Math Works, Natick, MA). Because a prerequisite for spectral and cross-correlation analyses is stationarity, the delta power and DP_max time series were detrended (subtracting a linear fit of the data) as part of these computations (16). From the results of the fast Fourier analysis, the periods (1/frequency) of the major peak in the power spectral density of DP_max vs. time and delta power vs. time were identified for each subject. The significance of these spectral peaks was determined with the aid of the F distribution (13). The lag number of the maximum correlation in the cross-correlation analysis was also identified for each subject. A maximum correlation at a lag number of zero means no phase shift. The maximum correlation assessed the relationship between DP_max and delta power values (detrended data) during REM and NREM sleep. Correlation by linear regression was also performed between DP_max and delta power in NREM sleep for the actual and detrended data.

	RESULTS

Top Abstract Introduction Methods Results Discussion References

The patients were 46.8 ± 7.7 (SD) yr of age and 163.9 ± 134% of ideal body weight (24). The amount and architecture of recorded sleep are summarized in Table 1. The pattern of sleep disturbance is typical for patients with OSA, showing almost no stage 3/4 sleep, reduced REM sleep, and a high arousal index. The patients as expected had a very high apnea + hypopnea index. About 90% of all apnea terminations were associated with evidence of cortical arousal. The mean values of DP_max and delta power (30-s averages) and their coefficients of variation are shown for NREM and REM sleep in Table 2. There was considerable variability of DP_max during the night within NREM sleep, which was composed almost entirely of stages 1 and 2 sleep. The mean values of delta power in NREM and REM sleep for the subject group are also shown in Table 2. Mean DP_max and delta power were significantly lower in REM sleep.

The DP_max and delta power (30-s averages) fluctuated in cycles during the night in all patients. A plot of the values for one subject is shown in Fig. 1. In this subject, delta power and DP_max varied synchronously and in phase over the night. The amplitude of the delta power tended to decrease over the night in this subject. Plots of the power spectral density of DP_max and delta power for this same subject are shown in Fig. 2. The major spectral peak of both signals occurred at a frequency of 0.0078 cycles/min (this corresponds to a period of 128 min). A plot of the cross correlation between DP_max and delta power (detrended data) for this subject (Fig. 3) shows that the maximum correlation (r = 0.56, P < 0.001) occurred at a lag of 0 epochs (no phase shift).

View larger version (37K): [in this window] [in a new window]   Fig. 1.   Maximum esophageal pressure deflection before apnea termination (DPmax) and mean electroencephalogram (EEG) delta power for 1 patient with obstructive sleep apnea plotted against epoch number (30-s epochs). Horizontal bars, periods of rapid-eye-movement sleep. DPmax and delta power varied cyclically and synchronously during night.

View larger version (18K): [in this window] [in a new window]   Fig. 2.   Power spectral density (PSD) of DPmax and delta power plotted against frequency for patient whose data are illustrated in Fig. 1. Major spectral peaks for DPmax and delta power were at a frequency of 0.0078 cycles/min, corresponding to a period of 128 min.

View larger version (12K): [in this window] [in a new window]   Fig. 3.   Cross-correlation plot of DPmax vs. delta power for patient whose data are illustrated in Figs. 1 and 2. Correlation of DPmax vs. delta power is plotted vs. lag number (in epochs). A positive lag number means DPmax is shifted ahead of delta power. Maximum correlation (r = 0.56) occurred at a lag number of 0 epochs (no phase shift).

The major power spectral density peaks of DP_max and delta power occurred at the same frequency in all subjects and were highly significant (P < 0.001). The periods corresponding to these frequencies ranged from 68 to 146 min with a mean for the group of 117.6 ± 8.8 min. The maximum correlation between the DP_max and delta power time series (cross-correlation analysis) occurred at a lag of zero epochs (no phase shift).

The correlations between DP_max and delta power (detrended data) for NREM and REM sleep for each subject are shown in Table 3. All these correlations were highly significant (P < 0.001). The ² values between the DP_max and delta power time series (detrended data) evaluated at their major periodicity are also shown in Table 3. The values reveal a high coherence (>0.7) between fluctuations in DP_max and delta power.

Significant correlations between DP_max and delta power were also present if the data for REM sleep were excluded. The correlation between DP_max and delta power during NREM sleep (actual and detrended data) for each subject is shown in Fig. 4. All these correlations were significant (P < 0.001). For the entire group, the mean correlations between DP_max and delta power in NREM sleep were 0.41 ± 0.04 (actual) and 0.44 ± 0.03 (detrended data).

View larger version (31K): [in this window] [in a new window]   Fig. 4.   Correlation coefficients between DPmax and delta power in non-rapid-eye-movement (NREM) sleep for each subject (actual and detrended data). All coefficients were statistically significant (P < 0.001).

The mean time course of DP_max and delta power within cycles of NREM sleep is illustrated in Fig. 5. Here, mean values of DP_max and delta power are plotted for each one-fifth of cumulative sleep within an NREM cycle. The values of DP_max and delta power for each NREM cycle were expressed as a percentage of the corresponding mean value for that cycle. The data shown represent the average of 35 cycles of NREM sleep in the 9 patients. DP_max and delta power initially increased, reaching peak values near midcycle, and then decreased slightly toward the end of the cycle.

View larger version (19K): [in this window] [in a new window]   Fig. 5.   Average values of DPmax and delta power for each one-fifth of NREM sleep cycles. DPmax and delta power are expressed as a percentage of respective mean value for NREM sleep cycle in which they occurred before averaging. Mean DPmax and delta power values at 3rd and 4th segments were significantly higher than respective values at 1st segment of cycles.

Although DP_max and delta power values for the entire night in NREM sleep were significantly correlated, their overnight trends differed. When each subject's data were expressed as a fraction of the all-night mean, separation into the night by quarters of cumulative NREM sleep showed that the delta power decreased significantly during the night (Fig. 6), whereas the DP_max showed a trend for an increase for the group as a whole through the third quarter. Although all patients showed a decrease in delta power, only six showed a definite increase in DP_max during the night in NREM sleep.

View larger version (20K): [in this window] [in a new window]   Fig. 6.   Mean values of delta power (Pwr; left) and DPmax (right) for each one-fourth of cumulative NREM sleep. For each subject, mean values for each one-fourth are expressed as a fraction of mean for night. Group averages of these one-fourth values are plotted. Delta power significantly decreased during night, whereas DPmax (NREM sleep) showed a trend for an increase through 3rd quarter of cumulative NREM sleep (P = 0.09).

	DISCUSSION

Top Abstract Introduction Methods Results Discussion References

This study demonstrates that, in patients with severe OSA, DP_max varies over the night in a cyclical fashion. We also found that, despite almost continuous apnea, the EEG delta power also varied in cycles throughout the night. The cyclic variation in DP_max was synchronous, with similar variation in the delta power of the central EEG. This variation was not simply secondary to a reduction in delta power and DP_max during periods of REM sleep. Analysis of cycles of NREM sleep showed that DP_max and delta power initially increased, reaching a maximum value on average at or after midcycle. Further analysis showed that the major period of the cycles of DP_max and delta power is approximately what one sees in the normal sleep cycles. We also found a significant correlation between the delta power and the DP_max for the entire night and for NREM sleep. A higher DP_max was associated with a higher delta power of the EEG. However, the EEG delta power decreased across the night (as in normal subjects), whereas the mean DP_max during NREM sleep tended to increase slightly overnight in most patients.

Studies in patients with OSA (19, 28, 29) and in normal subjects using mask occlusion (5) or induced hypoxia-hypercapnia (15) have found that arousal from NREM sleep occurs once a given level of inspiratory effort is reached independent of the values of hypoxia and hypercapnia stimulating ventilatory effort at the time of arousal. Central nervous system depressants (2, 3) or prior sleep fragmentation (4, 9, 22) increases the level of inspiratory effort associated with arousal. The level of effort triggering arousal is then an index of the arousal threshold (arousability). As might be expected, the arousal threshold (or maximum inspiratory effort before arousal) in a given individual varies during the night (18, 20, 28, 29). In fact, the above studies of the effects of hyperoxia and hypercapnia on arousal (5, 15, 19) controlled for this within-night variation by randomly alternating conditions over a night so that equivalent states of arousability would be present in each condition. However, little is known about the nature and mechanisms responsible for the within-night variation in the arousal threshold to respiratory stimuli.

One obvious explanation for variations in arousability would be variations in the stage of sleep. In normal subjects the arousal threshold to acoustic stimuli is higher in stages 3 and 4 than in the lighter stages of NREM sleep (30). Studies of mask occlusion in normal subjects (2) or inspiratory resistive loading (17) have found a higher arousal threshold in sleep stages 3 and 4 than in sleep stages 1 and 2. However, there are several difficulties with using traditional sleep staging as a marker of the depth of sleep in patients with severe OSA. First, separation of NREM sleep into only four stages would allow for only a very crude characterization of sleep depth. Second, the NREM sleep of patients with severe OSA usually comprises only stages 1 and 2. Even applying the usual rules for sleep staging may be problematic in these patients, inasmuch as only short periods of sleep are spaced between periods of arousal and ventilation. An alternative is to use a continuous characteristic of the EEG as a marker for sleep stage and perhaps the depth of sleep. The delta power of the EEG is a continuous marker of the sleep cycles, increasing at the start of each NREM cycle and falling with the onset of REM sleep. Inasmuch as the delta power is highest in stage 3/4 sleep, one might also hypothesize that a higher delta power would be associated with a higher arousal threshold. In fact, Neckelman and Ursin (21) found a correlation between the arousal threshold to acoustic stimuli and the EEG delta power in rats. This led us to correlate changes in the EEG delta power with variation in DP_max. In our study we also found a significant correlation between the delta power of the EEG and the arousal threshold to airway occlusion (as reflected by DP_max) in NREM sleep. However, correlation does not prove causality. Furthermore, although the cyclic variation in DP_max and the EEG delta power were synchronous, the overnight trends differed. Thus the delta power, rather than being an absolute marker for the depth of sleep (arousal threshold) across the entire night, is a marker for the time course of arousability within a cycle of NREM sleep. As a given cycle of sleep progresses, DP_max and delta power increase, reach a plateau, and then fall at or just before the start of the subsequent REM cycle.

Despite severe sleep fragmentation and a virtual absence of slow-wave sleep in our patients, we demonstrated that an underlying rhythm of cycles of delta activity was still present. The delta power also decreased during the night as in normal subjects. In contrast, DP_max in NREM sleep for the group showed an increasing trend at least through the third quarter of cumulative NREM sleep. However, at least two of our patients had a downward trend in DP_max. O'Donnell and co-workers (22) did not find a progressive within-night increase in the maximum suction pressure triggering arousal in a study of airway occlusion in dogs after a previous night of sleep deprivation. Montserrat and co-workers (20) found that the maximum tension-time index of the diaphragm before apnea termination (another index of the arousal threshold) increased from the first to the final portion of the night in NREM sleep. In this study, 10 consecutive apneas from the earliest and latest portions of the night showing stable NREM sleep (and not contiguous with REM periods) were analyzed. This result is not necessarily inconsistent with our findings. The earliest period undoubtedly occurred during an upswing of a cycle of DP_max and the final period prior to a downswing before a REM period. In that study two esophageal balloons were utilized, whereas our study employed a small fluid-filled catheter. The latter system is much more comfortable and causes minimal changes in sleep architecture (10). Thus our patients may have adapted to the monitoring system better and slept more deeply in the earlier portions of the night. However, we still found a trend for a small overnight increase in DP_max during NREM sleep in most of our patients. Our findings do suggest that evaluations of trends in the arousal threshold (or maximum respiratory effort preceding apnea termination) during the night must consider the cyclic behavior of this threshold. Comparisons of small segments of the night must therefore be performed with caution.

In our spectral analysis of DP_max vs. time during the night, we included values during REM as well as NREM sleep. There is no evidence to support DP_max as an index of the arousal threshold during REM sleep. Although respiratory effort progressively increases near apnea termination in NREM sleep, this is often not the case in REM sleep (22). We included the REM values so that DP_max would be a continuous function and the period of the fluctuations would reflect the actual time course during the night. However, one might object that our finding of cyclical variation in DP_max and EEG delta power was simply the result of the low values of DP_max and delta power associated with periods of REM sleep. This is not the case, inasmuch as we also analyzed cycles of NREM alone and demonstrated a significant correlation between DP_max and the EEG delta power in NREM sleep as well as a synchronous variation in both during cycles of NREM sleep.

To obtain continuous values of DP_max and delta power throughout the night, we had to deal with the problem of intervening periods of wakefulness. As discussed in METHODS, we added interpolated values of DP_max and delta power for short periods of intervening wakefulness and deleted longer periods by time shifting the data. The latter procedure could alter the major periodicity of the time series. However, although on average 12.5% of the sleep period time was composed of epochs scored as wakefulness, 77% of these periods of wakefulness were <5 min in duration and 88% were <10 min in duration. Values for these segments were added by interpolation. In only three patients were major data shifts (15 min) required (1 shift each in 2 subjects and 2 shifts in 1 subject). If we eliminated these patients, the mean periodicity for the remaining group differed only slightly from the entire group. In addition, inasmuch as such shifts were performed on the delta power and the DP_max data, the ability of our analysis to detect synchronous variation in these two signals was not altered, even in these three subjects.

Other interpretations of our results are certainly possible. One could argue that the cyclic variation in the delta power in our patients was not a marker for events controlling arousability but simply an effect of variations in the arousal threshold. That is, a higher arousal threshold produced a longer apnea, which produced an opportunity for deeper sleep (greater delta power) to develop. Svanborg and Guilleminault (27) demonstrated an increase in delta power from the initial to the final portion of apnea. In their study the delta power of the EEG started to rise on average ~13 s after apnea onset. The EEG delta power of the final 2 s of the apnea was double that of the initial 2 s. However, we do not believe that the cyclic variation in the delta power was simply secondary to a cyclic variation in the arousal threshold. First, this logic cannot explain the progressive fall in the amplitude of delta power over the night (without a decrease in the arousal threshold). Second, when we repeated the calculations averaging over NREM cycles (Fig. 5) using only the delta power from events from 20-25 s, we still noted the same cyclic behavior in delta power increasing over the initial part of the cycle, reaching a maximum value, and then decreasing slightly at the end of the cycles. Thus, although the delta power may increase during apnea, other factors in addition to event length are responsible for the cyclic variation in delta power.

In summary, we found that DP_max varied in cycles during the night in a group of patients with severe OSA. During each cycle of NREM sleep, DP_max initially increased and reached peak values on average near or after midcycle. Inasmuch as DP_max is an index of the arousal threshold in NREM sleep, this suggests a cyclic variation in arousability. Despite severe sleep fragmentation, we also found a cyclic variation in the EEG delta power. DP_max varied synchronously with the delta power of the EEG: a higher DP_max was associated with a higher value of delta power. Although the overnight trends in DP_max and the EEG delta power differed, their synchronous variation during the night and within cycles of NREM sleep suggests that they have some common controlling influences. We propose that the mechanisms controlling the cycles of sleep are responsible for the cyclic variability of the arousal threshold to airway occlusion.