Comments
Abstract
Rationale: Abnormal patterns of sleep and wakefulness exist in mechanically ventilated patients.
Objectives: In this study (SLEEWE [Effect of Sleep Disruption on the Outcome of Weaning from Mechanical Ventilation]), we aimed to investigate polysomnographic indexes as well as a continuous index for evaluating sleep depth, the odds ratio product (ORP), to determine whether abnormal sleep or wakefulness is associated with the outcome of spontaneous breathing trials (SBTs).
Methods: Mechanically ventilated patients from three sites were enrolled if an SBT was planned the following day. EEG was recorded using a portable sleep diagnostic device 15 hours before the SBT. The ORP was calculated from the power of four EEG frequency bands relative to each other, ranging from full wakefulness (2.5) to deep sleep (0). The correlation between the right and left hemispheres’ ORP (R/L ORP) was calculated.
Measurements and Main Results: Among 44 patients enrolled, 37 had technically adequate signals: 11 (30%) passed the SBT and were extubated, 8 (21%) passed the SBT but were not deemed to be clinically ready for extubation, and 18 (49%) failed the SBT. Pathological wakefulness or atypical sleep were highly prevalent, but the distribution of classical sleep stages was similar between groups. The mean ORP and the proportion of time in which the ORP was >2.2 were higher in extubated patients compared with the other groups (P < 0.05). R/L ORP was significantly lower in patients who failed the SBT, and the area under the receiver operating characteristic curve of R/L ORP to predict failure was 0.91 (95% confidence interval, 0.75–0.98).
Conclusions: Patients who pass an SBT and are extubated reach higher levels of wakefulness as indicated by the ORP, suggesting abnormal wakefulness in others. The hemispheric ORP correlation is much poorer in patients who fail an SBT.
Critically ill patients can develop EEG abnormalities in the ICU. The impact of these abnormalities at the time of liberation from mechanical ventilation is poorly established. We conducted standard polysomnography and calculated the odds ratio product, which is a continuous index used to evaluate sleep depth, 15 hours before a spontaneous breathing trial (SBT) in patients deemed ready to attempt liberation from mechanical ventilation.
Abnormal patterns of sleep and wakefulness were highly prevalent, but sleep scoring by conventional criteria did not differ between patients with successful and failed SBT. By contrast, the level of wakefulness, as assessed by the odds ratio product, was significantly higher in patients with successful SBT. Poor correlation between sleep depth in right versus left hemispheres predicted SBT failure.
Patients under mechanical ventilation in the ICU present a variety of EEG abnormalities during both wakefulness and sleep (1–3). Excessive sleep fragmentation, reduced REM sleep, and loss of normal circadian rhythm are consistent across studies, suggesting frequent sleep deprivation (3–5). The EEG during behaviorally confirmed wakefulness is often abnormal in ICU patients, with an increase in slow wave activity (seen during sleep in outpatients) and a decrease in the activity of higher frequencies that characterize wakefulness (1, 3, 6, 7). This pattern has been called pathological wakefulness. Patients with pathological wakefulness also often show atypical EEG patterns during sleep, with a marked reduction in EEG spindles and K complexes that help differentiate different stages of non-REM sleep; this is referred to as atypical sleep (6). Accordingly, in many ICU patients, it is difficult to distinguish between wakefulness and sleep from the EEG alone using the standard rules (7). The high prevalence of sleep loss/disruption in these patients and the fact that similar EEG changes are observed with experimental sleep deprivation (8, 9) suggest that sleep loss is an important contributing factor to these EEG abnormalities (10).
A recent study found that weaning time is longer in difficult-to-wean patients who have atypical sleep than in those who display normal sleep patterns (11). The reasons for failing a spontaneous breathing trial (SBT) are multifactorial (12, 13). Sleep deprivation may be a risk factor for weaning failure because it can reduce ventilatory responses to hypoxemia (14) and decrease respiratory muscle endurance (15); impair immune responses (16, 17), cardiovascular responses (18), neuroendocrine and metabolic function (19, 20), and neurocognitive function (21); and increase the incidence of delirium (22, 23).
In this study we evaluated the EEGs of mechanically ventilated ICU patients during a 15-hour period preceding an SBT. We hypothesized that patients with atypical sleep or pathological wakefulness were more likely to fail an SBT. In addition to conventional scoring, we used a digital scoring system (24, 25) that produces a number of EEG markers (odds ratio product [ORP] and spindle characteristics) that are relevant to identify pathological wakefulness and atypical sleep as well as possible cerebral pathology (24). The ORP is a continuous index of sleep depth that ranges from 0 (very deep sleep) to 2.5 (full wakefulness). The ORP is derived from the relation of powers in different EEG frequencies to each other (24). The ORP is highly correlated with arousability (24) and is therefore a valid index of sleep depth. One of its advantages is that it can distinguish between different levels of wakefulness, as the range of wakefulness extends from 2.5 (full wakefulness) to ≈1.8 (epochs are still scored as wakeful but contain some sleep features). This would make it particularly useful for identifying pathological wakefulness in which the wakeful EEG contains some sleep features. In such cases, the ORP during wakefulness would be closer to 1.8 than to 2.5.
The rationale for this study is that over a prolonged period of observation (15 h), an individual who is neither sleep deprived nor pathologically obtunded should have both sleep periods and ORP levels close to 2.5 present over the period of recording.
We conducted a prospective multicenter physiological study (SLEEWE [Effect of Sleep Disruption on the Outcome of Weaning from Mechanical Ventilation]) between January 2016 and May 2017 in three ICUs of three hospitals affiliated with the University of Toronto. The study was approved by the local research ethics board (#15-142) and was registered with ClinicalTrials.gov (Identifier: NCT02464735). Patients and/or next of kin gave consent before being included.
Intubated, mechanically ventilated patients were eligible for inclusion when an SBT was planned by the clinical team for the next day. The exclusion criterion was impaired consciousness with a Glasgow Coma Scale ≤ 8T.
Sleep was monitored from 5:00 p.m. to 8:00 a.m using a portable monitor (Alice PDx diagnostic system; Philips Respironics) that included two central EEG electrodes, right and left electrooculography electrodes, submental electromyography electrodes, and electrocardiography electrodes. Sleep assessment was performed offline using manual and digital techniques. These methods are described in detail in the online supplement. Briefly, sleep recordings were manually scored, first quickly after the recording and second by a sleep specialist with experience in ICU tracings (T.K. and X.D.) who was blinded to patient’s status. The 2007 American Academy of Sleep Medicine (AASM) rules were applied (26). When typical wakeful and sleep EEG patterns were absent, sleep was scored using the alternative classification, including pathological wakefulness and atypical sleep (6). The ORP was continuously quantified (by M.Y., who was also blinded to patient status and weaning outcome; see the online supplement for details). For six of the 37 patients, the EEG signals were technically unacceptable for this type of analysis. In the remaining 31 patients, the following ORP-derived indices were calculated:
| • | Average ORP over the entire 15 hours of total recording time. | ||||
| • | Percentage of total recording time with ORP >1.5, >2.0, and >2.2. | ||||
| • | Intraclass correlation coefficient (ICC) between the ORP in the right and left hemispheres (R/L ORP). Normally, sleep depth changes in parallel in both hemispheres and the ICC for right versus left ORP across the night is typically between 0.9 and 1.0. Lower values indicate regional differences in sleep depth, which would suggest disruption of the normal processes that coordinate sleep throughout the brain. | ||||
In addition to ORP-related variables, we also calculated the density of spindles (number per minute when the ORP was <1.5 with no REM, indicating likely non-REM sleep).
A daily screening was performed each afternoon and patients who were anticipated to undergo an SBT the morning after were included. The morning after the sleep assessment, if the patient met the readiness-to-wean criteria, an SBT was performed. The criteria to undergo an SBT the following day were oxygen saturation as measured by pulse oximetry (SpO2) ≥ 92% on fraction of inspired oxygen (FiO2) ≤ 0.5 and positive end-expiratory pressure ≤8 cm H2O, and low/no doses of vasopressors. SBTs are standard practice in all three ICUs (27), and were performed on the ventilator with no pressure assist of any kind (28). The SBT lasted for up to 60 minutes. Success or failure of the SBT was determined by the clinical team based on predefined criteria (29). Likewise, the decision to extubate after a successful SBT was made by the ICU team independently of the study.
Demographic data, comorbidities, admission diagnosis, Sequential Organ Failure Assessment (SOFA) score upon admission and enrollment, duration of mechanical ventilation upon and after enrollment, and duration of ICU and hospital stays were recorded. Blood pressure, heart rate, respiratory rate, SpO2, blood gases, analgesic and sedative medications, mode of ventilation, and ventilator settings were recorded at the time polysomnography (PSG) was started, at 8:00 a.m. the next day, and during the SBT. Neurological function was assessed twice daily using the Richmond Agitation Sedation Scale (RASS) and confusion assessment method for the ICU (CAM-ICU) scores.
PSG was set up at 5:00 p.m. A trained investigator positioned the electrodes and checked for correct recording with a laptop equipped with dedicated software (Sleepware G3; Philips Respironics). The following morning, the SBT was performed, usually between 8:00 and 9:00 a.m.
Continuous variables are presented with means and SD, whereas categorical variables are summarized using proportions and 95% confidence intervals (CIs). The normality of the distribution was checked by using the Kolmogorov-Smirnov test. We initially sought to compare patients who passed the SBT and those who failed it, but we decided to separate the patients into three groups: those who failed the SBT, those who passed but were not extubated, and those who passed the SBT and were extubated (SBT failure, SBT success without extubation, and SBT success with extubation, respectively). The groups were discriminated based on 1) a recent large observational study on weaning from mechanical ventilation, which found that less than 60% of patients who pass an SBT are extubated on the same day (30, 31); 2) the distinction between the SBT, which detects the ability to be separated from the ventilator, and the extubation criteria (32); and 3) clinical practice in the three centers. The main comparisons were made among the three groups (failed SBT, successful SBT with extubation, and successful without extubation). In the online supplement, we also report results from comparisons between patients who passed the SBT and those who failed the SBT. For sample size calculation, we assumed a success/failure rate for the SBT of 55%/45% and planned to have a minimum of 15 patients per group. We also anticipated a dropout rate of approximately 20% due to technical problems, and thus planned to enroll 42 patients from the three sites.
Proportions were compared using Fisher’s exact test. Continuous variables were compared by ANOVA, paired test, or unpaired test as appropriate. Receiver operating characteristic (ROC) curves were constructed to evaluate the ability of the R/L ORP to predict SBT success. Sensitivity, specificity, and area under the ROC curve (AUC-ROC) were calculated. A P value < 0.05 was considered significant.
A total of 44 patients were enrolled and 37 had PSG recordings of acceptable quality. The patients’ characteristics are shown in Table 1. The most common reason for intubation was acute respiratory failure (49%). At enrollment, the patients had been ventilated for 6 ± 4 days and had a SOFA score of 8 ± 4. On the day of PSG, RASS was 0 ± 1 and delirium was present in five patients (14%).
| Failed SBT (n = 18) | Passed SBT | Overall P Value | ||
|---|---|---|---|---|
| No Extubation (n = 8) | Extubation (n = 11) | |||
| At ICU admission | ||||
| Male, n (%) | 11 (61) | 7 (88) | 6 (55) | 0.29 |
| Body mass index, kg ⋅ m−2 | 30 ± 10 | 26 ± 4 | 31 ± 5 | 0.47 |
| APACHE II | 21 ± 9 | 17 ± 10 | 26 ± 7 | 0.10 |
| Main reason for intubation, n (%) | ||||
| Acute respiratory failure | 11 (61) | 3 (37) | 4 (36) | 0.84 |
| Acute respiratory distress syndrome | 7 (39) | 3 (37) | 4 (36) | 0.99 |
| Coma | 2 (11) | 3 (26) | 2 (18) | 0.29 |
| Cardiac arrest | 1 (6) | 0 (0) | 1 (9) | 0.62 |
| Postsurgery | 2 (11) | 2 (37) | 2 (18) | 0.55 |
| Other | 2 (11) | 0 (0) | 2 (18) | 0.45 |
| At time of enrollment | ||||
| Length of ICU stay, d | 4.4 ± 3.2 | 5.0 ± 2.5 | 10.4 ± 8.6* | 0.01 |
| SOFA score | 6 ± 3 | 8 ± 3 | 7 ± 3 | 0.32 |
| Treatment regimens, n (%) | ||||
| Continuous sedative infusion | 7 (39) | 2 (37) | 8 (73) | 0.08 |
| Continuous analgesic infusion | 5 (28) | 5 (63) | 5 (45) | 0.23 |
| Neurologic assessment | ||||
| RASS | 1 ± 1 | 0 ± 2 | 0 ± 1 | 0.26 |
| CAM-ICU positive, n (%) | 0 (0) | 2 (25) | 3 (27) | 0.06 |
| Arterial blood gases | ||||
| PaCO2, mm Hg | 44 ± 12 | 39 ± 9 | 35 ± 6 | 0.10 |
| PaO2/FiO2 ratio | 249 ± 73 | 219 ± 79 | 282 ± 90 | 0.36 |
| At time of SBT | ||||
| Systolic arterial pressure, mm Hg | 133 ± 22 | 141 ± 22 | 126 ± 22 | 0.47 |
| Diastolic arterial pressure, mm Hg | 65 ± 14 | 68 ± 14 | 65 ± 6 | 0.88 |
| Heart rate, min−1 | 90 ± 18 | 84 ± 12 | 88 ± 15 | 0.75 |
| Respiratory rate, min−1 | 25 ± 7 | 21 ± 7 | 21 ± 4 | 0.29 |
| SpO2, % | 96 ± 3 | 97 ± 2 | 96 ± 2 | 0.51 |
SBT was successful in 19 patients (51%), with 11 being extubated and 8 deemed not ready for extubation by the clinical team, and was unsuccessful in 18 (49%). The reasons for SBT failure and for not extubating those who passed are given in Table E1 in the online supplement.
Patients who failed the SBT had a shorter duration of ICU stay than their counterparts (Tables 1 and E2). RASS scores were similar among all groups and delirium was slightly but not significantly higher in patients who passed the SBT (5/19 vs. 0/18). Clinical variables at the time of the SBT did not differ.
Sleep duration and sleep quality during the night before an SBT, based on conventional and alternative assessments of the PSG recordings, are presented in Tables 2 and E3. Pathological wakefulness and atypical sleep were frequent but did not significantly differ across groups. Pathological wakefulness and atypical sleep, as previously defined (6), were present in 39% and 55% in patients who failed the SBT, 50% and 50% in patients who passed the SBT and were not extubated, and 27% and 27% in patients who passed the SBT, respectively. As a consequence, only 61% of the patients could be scored according to classical stages. Total sleep time based on this analysis was found to be shorter in patients who failed the SBT. When it was scorable, the distribution of stage 3, REM sleep, and fragmentation index did not differ between groups.
| Failed SBT (n = 18) | Passed SBT | Overall P Value | ||
|---|---|---|---|---|
| No Extubation (n = 8) | Extubation (n = 11) | |||
| Sleep quantity (conventional criteria) | ||||
| Duration of polysomnography, min | 753 ± 219 | 781 ± 116 | 875 ± 22 | 0.16 |
| Total sleep time, min | 187 ± 125 | 366 ± 189 | 260 ± 195 | 0.05 |
| Total sleep time, % | 22 ± 16 | 46 ± 20 | 30 ± 22 | 0.02 |
| Wakefulness, min | 455 ± 182 | 385 ± 165 | 530 ± 242 | 0.31 |
| Classical scoring possible | 8 | 4 | 7 | |
| Sleep stage 1, % | 18 ± 9 | 14 ± 17 | 15 ± 9 | 0.84 |
| Sleep stage 2, % | 51 ± 21 | 57 ± 21 | 59 ± 17 | 0.76 |
| Sleep stage 3, % | 26 ± 27 | 27 ± 31 | 20 ± 15 | 0.85 |
| REM stage, % | 3 ± 4 | 2 ± 2 | 6 ± 6 | 0.29 |
| Arousal and microawakening, h−1 | 34 ± 12 | 26 ± 19 | 36 ± 18 | 0.57 |
| Sleep quality (alternative criteria) | ||||
| Pathological wakefulness, n (%) | 7 (39) | 4 (50) | 3 (27) | 0.59 |
| Atypical sleep, n (%) | 10 (56) | 4 (50) | 3 (27) | 0.32 |
| Abnormal sleep EEG pattern, n (%) | 9 (50) | 4 (50) | 4 (36) | 0.75 |
| Absence of spindles, n (%) | 12 (66) | 4 (50) | 4 (36) | 0.27 |
| Digitally derived indices | ||||
| n | 15 | 7 | 9 | |
| Average ORP | 1.12 ± 0.40* | 0.91 ± 0.32* | 1.5 ± 0.40 | 0.02 |
| Time ORP > 2.2, % total recording time | 3.8 ± 6.2* | 6 ± 13 | 18 ± 18 | 0.03 |
| Time ORP > 2.0, % total recording time | 9.1 ± 11.4* | 8 ± 15* | 31 ± 26 | 0.01 |
| Time ORP > 1.5, % total recording time | 29 ± 27* | 18 ± 18* | 55 ± 28 | 0.02 |
| R/L ORP ICC | 0.54 ± 0.26*† | 0.80 ± 0.15 | 0.80 ± 0.16 | <0.01 |
| Awakening index, h−1 | 9 ± 7 | 7 ± 6 | 19 ± 19 | 0.07 |
| Spindle density (min−1) | 0.30 ± 0.34 | 0.21 ± 0.32 | 0.59 ± 0.56 | 0.15 |
ORP analysis was possible in only 31 of the 37 enrolled patients. Figures 1 and E1 show EEG tracings in seven 30-second epochs, representing different levels of wakefulness and sleep in a patient with normal EEG patterns; note that the average ORP in each reflects the visual differences between the seven strips. There was no correlation between the mean ORP in the first third of the recording and the RASS measured at the start of the recording (Figure E2).
Figure 1. EEG traces (C3) representing progression from full wakefulness (A) to deep sleep (stage N3, G) in a patient from this study who had normal EEG patterns, and the corresponding average ORP values (average of the 10 3-s values in the 30-s epoch). Note that the top three panels (A–C) meet the guidelines of wakeful epochs even though their visual appearance differs substantially. The ORP values reflect the gradual transition from full wakefulness (A) to stage 1 (D). Note also that E and F are both scored stage N2 despite the marked difference in appearance. Again, the ORP reflects the gradual deepening of sleep in stage N2. Arrows point to EEG spindles. Horizontal bars in D identify sections with a wakeful pattern (high-frequency dominance). A more detailed legend is available in the online supplement (Figure E2). ORP = odds ratio product.
[More] [Minimize]ORP-derived indices before the SBT showed significant differences between groups (Table 2). Patients who were successfully extubated had higher average ORP during total recording time and more time with ORP > 2.0 and > 2.2 than patients in the other two groups (Table 2). Figures 2 and E3 show the probability of a successful SBT in relation to time spent above specified ORP levels. There was no significant difference among groups in spindle density.
Figure 2. Probability of a successful SBT according to the percentage of TRT spent with ORP > 2.2. Patients were sorted according to the percentage of TRT spent with ORP > 1.5, > 2.0, and > 2.2. Each series was divided into three approximately equal aliquots (lowest 10, middle 10, and highest 11). The percentage of TRT spent with ORP > 1.5 ranged from 0% to 98%. When the threshold was raised to percentage of time spent above 2.2 (full wakefulness), the probability of success was 0% in the lower third (% < 0.8), 20% in the middle third (% 0.8–4.0), and 65% in the highest third (% > 4.0). See also Figure E3 for more details. ORP = odds ratio product; SBT = spontaneous breathing trial; TRT = total recording time.
[More] [Minimize]R/L ORP ranged from 0 to 0.97 (0.68 ± 0.24). Figure 3 shows three examples spanning the entire spectrum, and Figure 4 shows examples of EEG tracings with a discrepancy between the right and left ORP. Average R/L ORP was significantly lower in patients who failed the SBT as compared with their counterparts (Figures 5 and E4). When comparing SBT failure versus all SBT success, R/L ORP was 0.54 ± 0.26 versus 0.80 ± 0.15, respectively (P = 0.006). The AUC-ROC of R/L ORP to predict failure of SBT was 0.91 (95% CI, 0.75–0.98). An R/L ORP > 0.70 predicted a successful SBT with a sensitivity of 85% (95% CI, 56–98%) and a specificity of 88% (95% CI, 62–98%). Interestingly, there was a nonlinear relation between R/L ORP and percentage of time spent above specified ORP levels. Figure 6 shows the relation between time > 2.2 and R/L ORP.
Figure 3. Top: scatter plots of the relation between the ORP in the right and left hemispheres across total recording times in three patients. Each dot represents the results in a single 30-second epoch. (A) From the patient with the best correlation (ICC = 0.97). (B) From a patient with a very poor correlation (ICC = 0.12). (C) From a patient with an intermediate correlation (ICC = 0.62). The bottom panels show the time course of the difference between the two sides. Gray lines are consecutive epoch-by-epoch differences. Black lines are the 10-minute moving average of the individual epoch differences. Arrows identify the individual 30-second epochs shown in Figure 4. ICC = intraclass correlation coefficient; ORP = odds ratio product.
[More] [Minimize]
Figure 4. Tracings illustrating the visual appearance of the left and right central EEG derivations (C3 and C4) in five epochs with large differences between the ORP values of the two sides. ORP values are listed to the left of the panels. A–D are from the patient in Figure 3B, and their locations are indicated by the arrows in the scatter plot of Figure 3B. E is from the patient in Figure 3C. Note that except when the difference between the two ORP values is very large (A and B), it is very difficult to visually appreciate the presence of a major discrepancy between the right and left signals. The epoch in A could not be scored. ORP = odds ratio product.
[More] [Minimize]
Figure 5. ICC between the ORP values of the right and left hemispheres in patients who failed an SBT, who passed and were not extubated, and who passed and were extubated. ICC = intraclass correlation coefficient; ORP = odds ratio product; SBT = spontaneous breathing trial.
[More] [Minimize]
Figure 6. Relation between the percentage of total recording time spent with ORP > 2.2 and ICC between ORP values in the right and left hemispheres (R/L ORP ICC). When the percentage of total recording time was 0, it was changed to 0.1 to allow logarithmic regression. Patients who spent >11% of the time with ORP > 2.2 (n = 7) had invariably high R/L ORP ICCs (0.89 ± 0.08), whereas patients who had R/L ORP ICCs < 0.70 (n = 14; ICC = 0.49 ± 0.19) invariably had very little time with ORP > 2.2 (percentage of time > 2.2 = 2.7% ± 3.4%). ICC = intraclass correlation coefficient; ORP = odds ratio product.
[More] [Minimize]In this study we investigated quality and quantity of sleep using conventional sleep scoring guidelines and an index (ORP) that measures where the brain state lies on a continuous scale between full wakefulness and deep sleep, in the 15 hours preceding an SBT in patients clinically deemed ready for attempting to terminate mechanical ventilation. The high prevalence of stages referred to as pathological wakefulness and atypical sleep made the classical scoring of sleep of limited value, and the distribution of sleep stages did not differ. By contrast, the degree of wakefulness was clearly lower in patients who failed the SBT based on ORP assessment. The two main findings are that the likelihood of success of SBT and extubation is highly correlated with the fraction of monitoring time spent in full wakefulness (ORP > 2.2), and that a poor correlation between sleep depth in the right and left brain hemispheres predicts SBT failure. The group of patients who passed the SBT and were extubated was the only one characterized by both high hemispheric correlation and full wakefulness.
Identification of pathological wakefulness, the presence of sleep features in the EEG during confirmed wakefulness, has been technically difficult. So far, identification of this state has required simultaneous direct observation of the patient (to establish behavioral wakefulness) and monitoring of the EEG for the presence of slow activity typically associated with sleep (6, 7). Because direct observation over extended periods is not feasible, it is not possible to determine whether such a pattern, if present, represents the situation throughout wakefulness or transient sleepiness during the few minutes of observation, which may be normal. Furthermore, identification of excessive slow activity in the EEG requires specialized expertise that is not readily available in the ICU.
The ORP can distinguish between different levels of wakefulness (Figures 1A–1C) and has been observed to decrease during wakefulness after sleep restriction (33) and during sleep deprivation studies (34). Full wakefulness is typically associated with ORP > 2.2 (Figure 1A). In this study, 23 of 31 patients (74%) spent <10 minutes out of 15 hours with an ORP > 2.2. To put this in perspective, a previous study reported that ambulatory patients spent ≈10% of an 8-hour nocturnal study with an ORP > 2.2 (24). Obviously, if a normal ambulatory subject were monitored for 15 hours, the percentage of time spent with an ORP > 2.2 would be close to 50% because the balance of time (7 h) would be mostly wakeful time. Only four patients (14%) approached this level of full wakefulness and all passed the SBT. Thus, the vast majority of patients in this study had some degree of obtundation or abnormal or incomplete wakefulness most of the time they were awake.
Although the EEG pattern of pathological wakefulness is consistent with sleep deprivation (1), it may also be observed with other encephalopathies (35). This possibility is less likely, however, given that the patients were deemed to be ready for termination of ventilation and their RASS scores consistently showed an awake state.
The current study demonstrates that a successful SBT and subsequent extubation are directly correlated with time spent with full wakefulness (Figure 2). Yet, the reasons for SBT failure were primarily respiratory failure and desaturation (Table E1). Assuming that an abnormal wakefulness is related to sleep deprivation, and given that sleep deprivation is known to depress ventilatory responses to CO2 and hypoxia, and reduce respiratory muscle endurance (14, 15), one could speculate that sleep deprivation contributed to SBT failure through failure to respond to hypoxemia/hypercapnia, resulting in desaturation without distress (eight of the 18 patients who failed; Table E1), or through impaired diaphragm endurance (15), which would result in respiratory distress with or without desaturation (eight of the 18 patients who failed). An adequate response to the load requires intact responses to CO2 and hypoxia, and reasonable respiratory muscle endurance.
A significant proportion of the patients who passed the SBT (8/19, 42%) were not extubated, because they were deemed not ready by the clinical team. This finding is in line with a recent epidemiological study conducted in France, in which only 58% of the patients who passed the SBT were actually extubated (30). In fact, the finding that these patients had more abnormal wakefulness than those who were extubated in our study (lower ORP levels; Table 2) suggests that the decision to delay extubation had biological grounds.
Recently, Thille and colleagues reported a longer duration of weaning in patients with atypical sleep (11). Our results add support to their finding that abnormal EEG patterns influence clinical outcome in ICU patients. However, when we used the same techniques they used to identify patients with atypical sleep, we found no differences among the three patient groups (Table 2, sleep quality by alternative criteria).
There are several reasons why the alternate methods Thille and colleagues used were not discriminating in our study. First, they studied patients who had already failed an SBT. Their patients may have had more severe abnormalities capable of being identified by less sensitive techniques. Second, the main diagnostic feature of atypical sleep is a visual absence of spindles and K complexes (6). Agreement between manual scorers in spindle detection is poor (36, 37). Furthermore, determining that spindles are completely absent is problematic as it requires careful inspection of each epoch in the recording. Accordingly, the lack of significant differences in the number of patients with absent spindles in our study (Table 2, sleep quality) and the presence of such differences in Thille and colleagues’s study may reflect differences in manual scoring (11). Finally, as in most sleep studies in the ICU, they selected nonsedated patients who had been off sedation for several days.
We used an automatic validated (S. Warby, Ph.D., written personal communication, 2018 Dec 13) spindle detector. Despite a stated absence of spindles by visual inspection in more than half the patients (Table 2), none of the patients had a complete absence of spindles by digital analysis. Spindle density was not significantly different among the three groups (P = 0.15, Table 2). However, it was significantly higher in the extubated patients than in the other two groups combined (0.59 ± 0.56 min−1 vs. 0.27 ± 0.33 min−1, P = 0.03). It must be noted that spindle density was markedly depressed in all three groups relative to the values obtained with the same digital detector in non-ICU patients (2.65 ± 1.62 min−1 per EEG channel). The highest spindle density in the current study was 1.56 min−1, well below the average in non-ICU patients. Given that spindles are involved in memory consolidation (38), it is tempting to speculate that spindle suppression in ICU patients is a mechanism aimed at reducing memory of the unpleasant experiences encountered in this environment. Whether such suppression is protective against future psychopathology is debatable (39, 40).
This is the first time that agreement in sleep depth between the right and left hemispheres has been examined in ICU patients. This correlation has been observed in hundreds of non-ICU PSG studies both in normal subjects and in patients with chronic sleep disorders (M.Y., unpublished results). The R/L ORP ICC in non-ICU patients is only rarely below 0.70. Accordingly, the finding that the R/L ORP was <0.7 in nearly half of the patients (Figure 5) is remarkable and highly significant. Moreover, the facts that the R/L ORP predicted the success or failure of an SBT (AUC-ROC = 0.91), and that patients with values < 0.7 spent little/no time with ORP > 2.2, whereas in all patients who spent >11% of the time with normal wakefulness the R/L ORP was normal, further emphasize the importance of this finding and suggest that it is a feature that develops with extreme pathological wakefulness.
The finding that the R/L correlation may be relevant to the success of an SBT was coincidental. When the ORP was introduced in the clinical sleep laboratory of the author who developed it (M.Y.), it was noted that in some patients there was, at times, a marked difference between the ORP in the left hemisphere and that in the right hemisphere. M.Y. developed this tool and added it to a battery of new EEG biomarkers he developed to help identify their significance (e.g., the α intrusion index and spindle characteristics). Other than the current findings, there is no literature regarding its association with clinical disorders, but its use in research is just beginning.
Although it is not possible at present to determine why SBT failure and a poor R/L correlation are associated, the finding that a poor correlation is associated with severe pathological wakefulness (Figure 6) suggests a possible link. As discussed above, pathological wakefulness in the ICU setting is likely the result of sleep deprivation. Sleep deprivation may increase the risk of SBT failure through its negative effect on ventilatory responses and respiratory muscle endurance. To the extent that a poor R/L correlation reflects more severe sleep deprivation, SBT failure when the R/L correlation is poor may simply be a reflection of more severe respiratory control abnormalities.
A poor R/L correlation is a form of regional differences in sleep (i.e., some parts of the brain are asleep while others are awake). This form of sleep, often called unihemispheric sleep, is widely used by dolphins and related mammals (41), as well as by birds (42), when operating under physiological conditions that require long periods without sleep. It is possible that this primitive adaptive mechanism is reactivated in humans under conditions in which natural sleep is deemed by the individual to be unsafe.
When discrepancies are present, a spectral analysis typically shows one EEG having slightly higher power in the β frequency (>14 Hz) and lower power in the slow frequency (<7 Hz). As illustrated in Figure 4, which shows tracings from some of the greatest outliers in the ORP scatter plots (arrows in Figure 3), the difference in the visual appearance of the two EEG signals when discrepancies exist is too subtle to detect by the naked eye unless it is very large. Accordingly, such an abnormality can only be detected through digital analysis.
Together, the current study and the previous study by Thille and colleagues (11) clearly indicate that EEG abnormalities are an important risk factor for failure to wean. Given that patients who fail weaning contribute disproportionately to the costs of ICU care and to morbidity and mortality rates (30), studies are clearly needed to determine the cause of these abnormalities (e.g., sleep deprivation, metabolic factors, and drugs) and how to prevent them.
Our study is hypothesis generating, and the current findings suggest that EEG monitoring throughout an ICU admission could allow early detection of pathological wakefulness so that measures can be taken to mitigate its progression. Unless research studies point to other etiologies, we suggest that the appearance of pathological wakefulness strongly suggests sleep deprivation, and measures should be taken to ensure that the patient gets adequate sleep.
Visual inspection of an EEG is not sufficient to detect the EEG abnormalities of pathological wakefulness (Figure 4).
The fact that all patients with severe pathological wakefulness, including those with the most severe form (<2% time with ORP > 2.2 and R/L ORP < 0.7), scored 0 ± 1 on the RASS indicates that RASS scores are quite insensitive for detecting pathological wakefulness.
This study is the first to report the use of a new method that allows continuous measuring of sleep depth in the ICU. In addition, this study was conducted in three ICUs from three different hospitals. PSG- and ORP-derived indices were assessed off-line by sleep specialists (T.K., X.D., and M.Y.) who were blinded to the patients’ conditions and SBT outcomes. This study also has limitations, including the intrinsic limitations of the classical sleep scoring process. Assessment of the hemispheric EEG correlation with the R/L ORP was not correlated with a specific neurological investigation, but there were no clinical grounds to suspect the presence of primary brain disease.
Our findings indicate that quantifying abnormal wakefulness and correlating it with hemispheric EEG abnormalities is feasible and potentially helpful at the bedside to identify patients who are not ready to be weaned from the ventilator. They also underline the need for studies to determine the reasons for these EEG abnormalities and how to avoid them. Time spent during full or normal wakefulness (as assessed by the ORP) was greater in patients who passed an SBT and were extubated, and hemispheric EEG correlation was much poorer in patients who failed an SBT.
The authors thank the nurses and respiratory therapists at the three ICUs who were essential for the success of this project, as well as the research coordinators at the three sites, especially Kurtis Salway, Gyan Sandhu, Jennifer Hodder, and Sumesh Shah. They also thank Orla Smith, Jenny Gu, and Carolyn Campbell for their help in the overall organization of this study; Unmesh Edke for enrolling patients; and the attending physicians for supporting the study. They also thank Philips for providing the devices used in this study.
| 1. | Cooper AB, Thornley KS, Young GB, Slutsky AS, Stewart TE, Hanly PJ. Sleep in critically ill patients requiring mechanical ventilation. Chest 2000;117:809–818. |
| 2. | Gabor JY, Cooper AB, Crombach SA, Lee B, Kadikar N, Bettger HE, et al. Contribution of the intensive care unit environment to sleep disruption in mechanically ventilated patients and healthy subjects. Am J Respir Crit Care Med 2003;167:708–715. |
| 3. | Freedman NS, Gazendam J, Levan L, Pack AI, Schwab RJ. Abnormal sleep/wake cycles and the effect of environmental noise on sleep disruption in the intensive care unit. Am J Respir Crit Care Med 2001;163:451–457. |
| 4. | Elliott R, McKinley S, Cistulli P, Fien M. Characterisation of sleep in intensive care using 24-hour polysomnography: an observational study. Crit Care 2013;17:R46. |
| 5. | Friese RS, Diaz-Arrastia R, McBride D, Frankel H, Gentilello LM. Quantity and quality of sleep in the surgical intensive care unit: are our patients sleeping? J Trauma 2007;63:1210–1214. |
| 6. | Drouot X, Roche-Campo F, Thille AW, Cabello B, Galia F, Margarit L, et al. A new classification for sleep analysis in critically ill patients. Sleep Med 2012;13:7–14. |
| 7. | Watson PL, Pandharipande P, Gehlbach BK, Thompson JL, Shintani AK, Dittus BS, et al. Atypical sleep in ventilated patients: empirical electroencephalography findings and the path toward revised ICU sleep scoring criteria. Crit Care Med 2013;41:1958–1967. |
| 8. | Naitoh P, Kales A, Kollar EJ, Smith JC, Jacobson A. Electroencephalographic activity after prolonged sleep loss. Electroencephalogr Clin Neurophysiol 1969;27:2–11. |
| 9. | Olbrich E, Landolt HP, Achermann P. Effect of prolonged wakefulness on electroencephalographic oscillatory activity during sleep. J Sleep Res 2014;23:253–260. |
| 10. | Younes M. To sleep: perchance to ditch the ventilator. Eur Respir J 2018;51:pii: 1800624. |
| 11. | Thille AW, Reynaud F, Marie D, Barrau S, Rousseau L, Rault C, et al. Impact of sleep alterations on weaning duration in mechanically ventilated patients: a prospective study. Eur Respir J 2018;51:pii: 1702465. |
| 12. | McConville JF, Kress JP. Weaning patients from the ventilator. N Engl J Med 2012;367:2233–2239. |
| 13. | Dres M, Demoule A. Diaphragm dysfunction during weaning from mechanical ventilation: an underestimated phenomenon with clinical implications. Crit Care 2018;22:73. |
| 14. | White DP, Douglas NJ, Pickett CK, Zwillich CW, Weil JV. Sleep deprivation and the control of ventilation. Am Rev Respir Dis 1983;128:984–986. |
| 15. | Chen HI, Tang YR. Sleep loss impairs inspiratory muscle endurance. Am Rev Respir Dis 1989;140:907–909. |
| 16. | Bryant PA, Trinder J, Curtis N. Sick and tired: does sleep have a vital role in the immune system? Nat Rev Immunol 2004;4:457–467. |
| 17. | Faraut B, Boudjeltia KZ, Vanhamme L, Kerkhofs M. Immune, inflammatory and cardiovascular consequences of sleep restriction and recovery. Sleep Med Rev 2012;16:137–149. |
| 18. | Khan MS, Aouad R. The effects of insomnia and sleep loss on cardiovascular disease. Sleep Med Clin 2017;12:167–177. |
| 19. | Davies SK, Ang JE, Revell VL, Holmes B, Mann A, Robertson FP, et al. Effect of sleep deprivation on the human metabolome. Proc Natl Acad Sci USA 2014;111:10761–10766. |
| 20. | Spiegel K, Leproult R, Van Cauter E. Impact of sleep debt on metabolic and endocrine function. Lancet 1999;354:1435–1439. |
| 21. | Bonnet MH. Acute sleep deprivation. In: Kryger MH, Roth T, Dement WC, editors. Principles and practice of sleep medicine. Philadelphia, PA: Elsevier Saunders; 2005. pp. 51–66. |
| 22. | Babkoff H, Sing HC, Thorne DR, Genser SG, Hegge FW. Perceptual distortions and hallucinations reported during the course of sleep deprivation. Percept Mot Skills 1989;68:787–798. |
| 23. | Roche Campo F, Drouot X, Thille AW, Galia F, Cabello B, d’Ortho M-P, et al. Poor sleep quality is associated with late noninvasive ventilation failure in patients with acute hypercapnic respiratory failure. Crit Care Med 2010;38:477–485. |
| 24. | Younes M, Ostrowski M, Soiferman M, Younes H, Younes M, Raneri J, et al. Odds ratio product of sleep EEG as a continuous measure of sleep state. Sleep 2015;38:641–654. |
| 25. | Malhotra A, Younes M, Kuna ST, Benca R, Kushida CA, Walsh J, et al. Performance of an automated polysomnography scoring system versus computer-assisted manual scoring. Sleep 2013;36:573–582. |
| 26. | Berry RB, Brooks R, Gamaldo C, Harding SM, Lloyd RM, Quan SF, et al. AASM scoring manual updates for 2017 (version 2.4). J Clin Sleep Med 2017;13:665–666. |
| 27. | Goligher EC, Detsky ME, Sklar MC, Campbell VT, Greco P, Amaral ACKB, et al. Rethinking inspiratory pressure augmentation in spontaneous breathing trials. Chest 2017;151:1399–1400. |
| 28. | Sklar MC, Burns K, Rittayamai N, Lanys A, Rauseo M, Chen L, et al. Effort to breathe with various spontaneous breathing trial techniques: a physiologic meta-analysis. Am J Respir Crit Care Med 2017;195:1477–1485. |
| 29. | Boles J-M, Bion J, Connors A, Herridge M, Marsh B, Melot C, et al. Weaning from mechanical ventilation. Eur Respir J 2007;29:1033–1056. |
| 30. | Béduneau G, Pham T, Schortgen F, Piquilloud L, Zogheib E, Jonas M, et al.; WIND (Weaning according to a New Definition) Study Group and the REVA (Réseau Européen de Recherche en Ventilation Artificielle) Network. Epidemiology of weaning outcome according to a new definition: the WIND Study. Am J Respir Crit Care Med 2017;195:772–783. |
| 31. | MacIntyre N. Another look at outcomes from mechanical ventilation. Am J Respir Crit Care Med 2017;195:710–711. |
| 32. | Thille AW, Richard J-CM, Brochard L. The decision to extubate in the intensive care unit. Am J Respir Crit Care Med 2013;187:1294–1302. |
| 33. | Schweitzer PK, Griffin KS, Younes M, Walsh JK. Assessment of sleep depth and propensity during sleep restriction using the odds-ratio-product [abstract]. Sleep 2018;41(Suppl 1):A57–A58. |
| 34. | Kuna ST, Tanayapong P, Maislin G, Staley B, Pack FM, Pack AI, et al. Odds ratio product: a measure of sleep homeostasis following prolonged wakefulness [abstract]. Sleep 2018;41(Suppl 1):A83. |
| 35. | Sutter R, Kaplan PW, Valença M, De Marchis GM. EEG for diagnosis and prognosis of acute nonhypoxic encephalopathy: history and current evidence. J Clin Neurophysiol 2015;32:456–464. |
| 36. | Warby SC, Wendt SL, Welinder P, Munk EGS, Carrillo O, Sorensen HBD, et al. Sleep-spindle detection: crowdsourcing and evaluating performance of experts, non-experts and automated methods. Nat Methods 2014;11:385–392. |
| 37. | Wendt SL, Welinder P, Sorensen HBD, Peppard PE, Jennum P, Perona P, et al. Inter-expert and intra-expert reliability in sleep spindle scoring. Clin Neurophysiol 2015;126:1548–1556. |
| 38. | Clawson BC, Durkin J, Aton SJ. Form and function of sleep spindles across the lifespan. Neural Plast 2016;2016:6936381. |
| 39. | Schelling G, Stoll C, Haller M, Briegel J, Manert W, Hummel T, et al. Health-related quality of life and posttraumatic stress disorder in survivors of the acute respiratory distress syndrome. Crit Care Med 1998;26:651–659. |
| 40. | Jones C, Griffiths RD, Humphris G, Skirrow PM. Memory, delusions, and the development of acute posttraumatic stress disorder-related symptoms after intensive care. Crit Care Med 2001;29:573–580. |
| 41. | Lyamin OI, Manger PR, Ridgway SH, Mukhametov LM, Siegel JM. Cetacean sleep: an unusual form of mammalian sleep. Neurosci Biobehav Rev 2008;32:1451–1484. |
| 42. | Rattenborg NC, Lima SL, Amlaner CJ. Half-awake to the risk of predation. Nature 1999;397:397–398. |
*L.B. is Deputy Editor of AJRCCM. His participation complies with American Thoracic Society requirements for recusal from review and decisions for authored works.
M.D. was supported by the French Intensive Care Society (SRLF Bourse de Mobilité 2015), the 2015 Short Term Fellowship Program of the European Respiratory Society, the 2015 Bernhard Dräger Award for Advanced Treatment of Acute Respiratory Failure from the European Society of Intensive Care Medicine, the Assistance Publique Hôpitaux de Paris, the Fondation pour la Recherche Médicale (FDM 20150734498), and MitacsGlobalink Sorbonne Universités. L.B. holds the Keenan Chair in Critical Care and Acute Respiratory Failure.
Author Contributions: M.D., N.R., R.L., and L.B. designed the study. M.D. coordinated the study. M.D., I.T., D.L.G., T.P., D.J., E.C., S.M., M.E.W., and L.B. were responsible for patient screening, enrollment, and follow-up. M.D., M.Y., X.D., and L.B. analyzed the data. M.Y., T.K., and X.D. scored the polysomnography. M.D., M.Y., and L.B. wrote the manuscript. All authors had full access to all of the study data, contributed to drafting the manuscript or critically revised it for important intellectual content, approved the final version of the manuscript, and take responsibility for the integrity of the data and the accuracy of the data analysis.
This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org.
Originally Published in Press as DOI: 10.1164/rccm.201811-2119OC on March 1, 2019
Author disclosures are available with the text of this article at www.atsjournals.org.
