The role of good quality sleep in health is well known [1, 2]. However, many critically ill patients treated in the intensive care unit(ICU) experience sleep disruption [36]. Intensive care patients may experience normal or near-normal total sleeptime (TST) but have been reported to have reduced slow wave (SW) and rapid eye movement(REM) sleep, largely because of frequent arousals. The factors causing sleep disruptionin intensive care patients are not fully understood; many extrinsic and intrinsicfactors have been proposed, such as noise [7], light, inflammatory mediators [6], sedative and opioid medications [4] and mechanical ventilator settings [8]. Polysomnography (PSG) recording is supported by reports from ICU patients ontheir inability to sleep well in ICU [9, 10]. Patients often attribute disturbances to their sleep to noise levels [10].

Twenty-four hour PSG studies in ICU have been few and there are no recent studies; therehave been fewer studies still examining the prevalence of sleep disruptive factorssimultaneously with PSG or investigations of ICU patient perceptions of the quality ofsleep and sleep disruptions in conjunction with PSG recording. The primary aim of thisobservational study was to provide a current assessment of the quality and quantity ofcritically ill patients' sleep while they were treated in ICU. The secondary aims wereto explore the prevalence of the main factors that potentially affect sleep in ICU, thatis the environmental sound pressure and illuminance levels ICU patients are exposed to,and the frequency of treatment and care patients receive. We also assessed self-reportedsleep quality while patients were treated in ICU and the hospital ward and patients'perceptions of sleep disruptions in ICU.

Materials and methods

Study setting and sample

This study was conducted in a 36-bed adult general ICU in a 600-bed hospital inSydney, Australia. The ICU provided specialty services such as cardiac, spinal, burnsand renal and was a closed unit with an accredited intensivist responsible for themanagement of all patients. Ward rounds were conducted by the intensivist twice a daywhen treatment goals/plans were reviewed, for example sedation medicationprescriptions were adjusted according to the individual patient needs (routinely,calm and interactive). The registered nurse (RN) to patient ratio was 1:1 formechanically ventilated patients and 1:2 for patients requiring high dependency care.There was no distinct area for the care of high dependency level patients;mechanically ventilated patients were cared for alongside patients of lower acuity.The RN performed all the nursing care for the patient. The main practices associatedwith sleep promotion during the time in which the study was conducted were offering anight-time hypnotic (that is, temazepam) and earplugs/eye shades to the patients anddimming the main lights at night.

Patients were included if aged >16 years and likely to be treated in ICU for >24hours and able to give informed consent on their own behalf. The ability to provideconsent was assessed by the following means: i) subjective advice by the bedsidenurse; cognition was discussed including orientation to time and place and ability tofollow simple instructions; ii) the patient was approached and asked to state theirname (or mouth the words); if the patient was able to perform this instruction afurther check was made; iii) the patient's understanding and ability to followinstructions was checked by asking them to nod when the correct colour card was heldup from a selection of three. Exclusion criteria included a history of sleepdisorders, psychiatric illness requiring medication, a known diagnosis of dementia orcentral neurological impairment confirmed by radiological scan. Human Research EthicsCommittee approval for the study was provided by the Health Service and theUniversity of Technology Sydney. Patients provided informed consent with writtenconfirmation by their next of kin in cases where the patient was unable to sign theconsent form. Data were collected from January to December, 2009 and September 2010to April 2011. Screening for eligibility was performed on weekdays.

Data collection

On enrolment, patients rated their sleep quality prior to hospitalization using a 1to 10 scale (10 = excellent) and patients or their proxy completed the InsomniaSeverity Index (ISI) [11]. The ISI comprises seven items based on the symptoms and severity of sleepdisturbance; each item is scored from 0 (not at all) to 4 (extremely). The total ISIscore ranges from 0 to 28 with a cutoff for clinical insomnia of 15. Concurrentvalidity with a sleep diary has been reported at r = 0.65 [11].

Patients were monitored for one 24-hour period using a portable PSG device, eitherPS-2™ (Compumedics, Melbourne, Australia) or ALICE LE™ (PhilipsRespironics, Amsterdam, Netherlands). Recording began and finished between 1000 and1700 hours. Electroencephalograph (EEG) (O1/M2, C4/M1), electromyograph (EMG),electrooculograph (EOG) (right and left) and electrocardiograph (ECG) (lead II) wererecorded. Patients' skin was prepared according to standard techniques. Gold cup EEGelectrodes were placed at O1/M2 and C4/M1 according to the International 10-20 System [12]. Two EOG electrodes were used for right and left eye movements. The EMGelectrodes were located over the right and left masseter (facial) muscles. Electrodeapplication was performed by the authors (RE and MF) who were both trained in thetechnique. Electrode impedance was maintained <9,000 ohms. Visual checks wereperformed hourly and electrodes were replaced if impedance values approached 9,000ohms or when the patient was disturbed for routine repositioning.

Sound and illuminance levels were recorded simultaneously with PSG using theintegrated sound pressure level meter (model 2250, Brüel and Kjaer™,Nærum, Denmark) and illuminance level meter (T-10, Konica Minolta™, Osaka,Japan). Continuous equivalent sound pressure levels (Leq) in 'A' weighted decibelsand peak sound pressure levels (Lpeak) in 'C' weighted decibels were logged everysecond. More detailed information about the protocol for sound pressure levelrecording and configuration of patient rooms is reported elsewhere [13]. Illuminance level (in lux) was recorded, using a sensor placed close tothe patient's head, once per minute. The bedside nurse was requested to log an eventwhenever the patient received treatment or care using a specially designed MicrosoftAccess™ (Microsoft, Redmond, WA, USA) form listing a number of events on acomputer located within reach. The event log contained the following items: clinicalassessment; tracheal suctioning; pressure area care; physiotherapy; mouth/eye care;blood test (sampling); wash; non-invasive blood pressure; eating and drinking;dressing; pain; line insertion; X-ray; clinical crisis; agitation/anxiety/confusion;electrode replacement and other (for example placing an extra blanket on thepatient).

On completion of PSG recording patients rated their previous night's sleep using theRichards Campbell Sleep Questionnaire (RCSQ) [14]. The RCSQ contains five 100mm visual analogue scales (VAS): sleep depth,latency, awakenings, time awake and quality of sleep (higher scores indicate bettersleep). The RCSQ was pilot tested in a medical ICU [15] and validated with overnight PSG in medical ICU patients [14]. In our study, patients who were unable to write were assisted; thepatient used their current communication strategy to indicate where the investigatorshould mark the VAS.

One to two nights after transfer to the ward, patients rated their sleep on the wardusing the RCSQ and reported on sleep-disturbing factors in ICU using the Sleep inIntensive Care Questionnaire (SICQ) [10]. The SICQ was developed to determine the perceived effect of the ICUenvironment on sleep. It contains seven questions (rated 1 to 10), including overallsleep quality at home, sources of perceived sleep disruption and sources ofnoise.

Demographic and clinical data were collected from the patient's record. The AcutePhysiology and Chronic Health Evaluation (APACHE) III modified diagnostic codes [16] were used to classify diagnoses. The APACHE II severity of illness scoreon admission and the Modified Sequential Organ Failure Assessment (SOFA) [17] at the time of enrolment were calculated to assess severity ofillness.

Data analysis

Polysomnography recordings were scored manually in 30-second epochs, by threequalified sleep technologists using standard Rechtschaffen and Kales [18] (R and K) criteria. TST was defined as the time spent in all sleep stages.The percentage of time in each sleep stage and sleep during the day (0600 to 2100hours) was calculated. Arousals were defined using American Academy of Sleep Medicinecriteria [19] and the number per hour of sleep (arousal index) reported. Morphineequivalent doses of opioid medications [20] and midazolam equivalent doses of benzodiazepine [21] were calculated to summarise the doses of opioid and sedative medicationsadministered.

The software packages PASW™ (version 18; SPSS Inc, Chicago, IL, USA) andMicrosoft Excel (2007) were used to analyse the data. Means and medians were used todescribe continuous data and frequencies and percentages were used for categoricaldata. Interrater reliability for scoring the PSG data by sleep technologists wasperformed on 16 (30%) recordings using Cohen's Kappa statistic. The Mann-Whitney Utest was used to detect differences between the TST and arousal indices for patientswho received mechanical ventilation during PSG and those who did not and to comparearousal indices for patients who received benzodiazepine medications or propofol andthose who did not. The correlation between arousal indices and peak sounds (>80dB(A)) was explored using Pearson's r. Wilcoxon signed ranks test was used to comparepatients' self-reported quality of sleep in ICU and the ward.


Sample characteristics

During the study twenty-four hour PSG data were analysed for 53 patients and 47patients were followed up on the ward. Of 266 eligible patients 57 were enrolled; ofthose not enrolled 74 declined participation and 135 were transferred to the wardbefore they could be invited to participate. Two patients requested removal of PSGafter recording began, palliation was initiated for another and data for another wasunable to be analysed. Figure 1 provides a flow diagram of thenumber of eligible patients and the number invited to participate and enrolled. Thecharacteristics of patients enrolled in the study are provided in Table 1. Admission diagnoses were mostly non-operative (66%), meanAPACHE II score was 18.70 (SD: 8.23) and the mean SOFA score was 4.04 (SD: 2.53).Patients were interactive and calm (mean Vancouver Interaction Scale (VICS) [22] score: 27.06 (SD: 3.80) (equivalent to 0 or -1 on the Richmond AgitationSedation Scale (RASS) [23]).

Figure 1
figure 1

Prevalence of eligible patients. Number of patients admitted toICU during the time period in which the study was conducted, eligible for thestudy, enrolled and completing 24-hour polysomnography (PSG) monitoring andsleep questionnaires on the hospital ward. aEnrolment ofeligible patients was limited by the availability of only one PSG and oneresearcher for the entire 24-hour recording period.

Table 1 Patient characteristics

Twenty-one percent of patients reported a pre-hospital ISI score of ≥15,indicating moderate to severe clinical insomnia, however, median pre-hospitalisationsleep quality on the SICQ was 8.00 out of 10.00. There was no difference between themean RCSQ score in ICU and on the ward (P = 0.61). Noise was rated thehighest sleep-disturbing factor (Table 2).

Table 2 Sleep outcomes: subjective reports

Sleep, sound and light

Median TST was five hours (Figure 2). The median duration ofsleep without waking was 00:03 (hh:mm). Unusual sleep stage transitions were noted(for example progression from stage 1 to REM sleep). Daytime sleep comprised 41% ofTST and EEG delta wave activity was observed when patients were interactive andapparently awake. The majority of sleep was stages 1 and 2 (19 and 73 %,respectively). SW and REM sleep was scant (median (IQR) 0 (0 to 1) and 0 (0 to 6)respectively). Arousals were frequent (median arousal index: 27) (Table 3). Interrater reliability of the sleep technologists' PSGanalysis for sleep/awake was 0.58 to 0.68 (Table 4).

Figure 2
figure 2

Boxplot of total sleep time (TST) (hh:mm:ss) for the sample.There was a large variation the TST for the sample.

Table 3 Sleep outcomes: PSG-derived data, sleep time and stages
Table 4 Interrater reliability (Cohen's Kappa statistic) for Rechtschaffen and Kalesscoring for sleep technologists (16 recordings)

Sound pressure levels were high (mean Leq 53.95 (SD: 2.33) dB(A) during the day and50.20 (SD: 3.21) dB(A) at night). There were a median 416/h sound peaks >80 dB(C)during the day and 90/h at night. The correlation between arousal indices and numberof sound peaks >80 dB(A) was weak during the day (r = 0.13) and night (r = 0.19);neither was statistically significant. Median illuminance levels were lower at night(1.74 lux) than during the day (74.20 lux). The average number of care events was1.74/h; the lowest was between 0200 and 0500 hours (<1.00/h). No patients wore earplugs or eye shades during sleep monitoring.

Sleep mechanical ventilation and medications

Pressure support was the ventilation mode used for 26 patients (six patients receivedpressure control ventilation (PCV) for periods during pressure support), one patientreceived synchronised intermittent ventilation (SIMV) and another patient receivedPCV. Twelve patients were extubated (that is, had their endotracheal tube removed)during PSG monitoring. The median TST of patients who received mechanical ventilationduring PSG recording and those who did not was 05:14 (IQR: 03:36 to 07:57) vs. 03:57(IQR: 01:39 vs. 06:47; P = 0.049) and median arousal indices (20.50 IQR:11.50 to 32.5 vs. 29.50 IQR: 22.00 to 40.87, P = 0.018). The differences forother sleep parameters did not reach statistical significance for patients whoreceived mechanical ventilation during PSG recording and those who did not (forexample, stage 1: 22.87 (24.13) vs. 27.54 (19.85) %, stage 2: 69.99 (24.02) vs. 67.30(20.17) % and REM: 4.47 (5.90) vs. 2.38 (4.91) %).

The mean equivalent dose of morphine was 12.46 (SD: 24.42) mcg/kg/h and equivalentdose of midazolam was 2.26 (SD: 11.60) mcg/kg/h. Six patients received temazepam(four received 20mg and two 10mg) for night sedation. There was a difference in themedian arousal indices for patients who received benzodiazepine medication/propofoland those who did not (22.00 IQR: 11.25 to 31.75 vs. 30.00 IQR: 20.75 to 40.75; P = 0.019). The differences for other sleep parameters did not reach statisticalsignificance for patients who received benzodiazepine medication/propofol and thosewho did not (for example, stage 1: 26.41 (20.74) vs. 23.71 (23.75) %, stage 2: 67.42(19.38) vs. 69.97 (24.74) % and REM: 2.85 (4.76) vs. 4.14 (6.16) %).


This study provides a current characterisation of the quality and quantity of sleep inintensive care patients, together with factors that potentially affect the sleep ofpatients in ICU. Sleep quality was poor; over 90% of patients' sleep was stage 1 and 2and TST was below that experienced by healthy adults. These results indicate that therehas been no improvement since studies published between 1976 and 2003 [3, 5, 2426], despite improvements in ICU design, technology and health care personneltraining. Sleep fragmentation and unconventional architecture was evident with multipleshort, non-contiguous sleep periods and numerous arousals. Arousals were frequent, aswell as non-sequential stage changes [4, 27]. A large percentage of TST was during daytime hours

Sound levels were elevated and exceeded World Health Organization (WHO) standards forhospitals, that is Leq should not >35 dB(A) in patient areas [28]. In our study, Leq was >49 dB(A). Continuous equivalent sound pressure levelswere 10 dB(A) lower than in many studies in ICU [5, 29, 30] but similar to others [26, 31]. The frequency of sound peaks >80 dB(C) was notable (range: 31 to 1,436/h)and high in comparison with other studies. Kahn et al. [32] reported a mean of 56/h before a noise reduction program and 40/h afterwards,while the mean reported by Meyer et al. [33] was 60/h. It is somewhat surprising that there was no correlation betweenarousal indices and the number of sound peaks >80 dB(A), although it is possible thatpatients had become accustomed to high sound levels. Illuminance levels were appropriateat night (median <2 lux). However, daytime illuminance levels (74 lux) may have beentoo low to encourage a 24-hour circadian rhythm. Study of endogenous melatonin secretionin healthy participants indicates that illuminance levels of <100 lux may not besufficiently bright to suppress melatonin secretion in some individuals [34]. Thus patients in our study may have had high melatonin levels during the daythat contributed to the proportion of daytime sleep. The mean number of treatment/careevents was 40 (1.7/h). This contrasts with hourly rates reported by others (for example3, 7 and 6/h) [6, 25, 26]. Under-recording by the bedside nurses may have occurred in our study. Whilean attempt was made to record care events/treatment in a synchronous manner, this proveddifficult in practice. Hence we are unable to draw conclusions about the relationshipbetween such events and arousals.

Medications administered and considered to potentially affect sleep noted during PSGrecording were mainly opioids and sedative medications. A large proportion (60%) ofpatients received opioid medications, potentially affecting their sleep. SWS has beenfound to be reduced by opioids, with a concomitant increase in stage 2 sleep [35]. A large percentage of patients in our study received benzodiazepines orpropofol (53%) (however, only lightly sedated patients were enrolled; the mean VICSscore was 27.06 (SD: 3.80)). This may also have contributed to the high proportion ofstage 2 and lack of REM sleep [36]. However, it is notable that patients who received benzodiazepine medicationor propofol had fewer arousals than patients who did not. Despite the apparentsuppression of SWS and REM, sleep may have been more consolidated when sedativemedications were administered. The somewhat surprising findings that TST was higher andthere were fewer arousal indices in patients who received mechanical ventilation duringPSG recording may be partly attributable to the larger sedative medication doses theyreceived (mean: 3.49 (SD: 15.64) vs. 0.40 (SD: 0.75) mcg/kg/h).

Patients' self-reports of sleep quality in ICU were poor but comparable to previousstudies. The mean RCSQ score from patients ready for transfer to the ward in the sameICU (n = 222) was 47.18 mm [37]. In a study investigating the utility of the RCSQ and concordance of nurseand patient sleep assessment in ICU, the mean RCSQ score was 45.50 mm [38].

Patient perceptions of sources of sleep disruption were similar to previous studies.Noise was rated the most disruptive, as has been reported elsewhere [26]. In the development of the SICQ, ratings of disruptive activities were lowerand noise (4.5) was less disruptive than vital signs (5.5) and phlebotomy (5.5) [10]. However, the SICQ was first reported in 1999; non-invasive vital signsmonitoring and blood sampling in the current study would be expected to be lessdisruptive than techniques in use in the 1990s.

Study strengths and limitations

This study is the largest of its kind to be conducted using PSG with simultaneousdata collection for factors known to affect sleep in ICU and is the first to presentdata from the Australian context. In addition, other investigators have rarelycollected data on the patients' perception of sleep quality and potentialsleep-disrupting factors in conjunction with PSG recording. This subjectiveinformation is vital in corroborating objective data, particularly as sleep is asubjective experience.

A limitation that became evident during the study was the difficulty in interpretingthe PSG data using conventional R and K analysis. One patient's EEG waveform wasaffected with 'alpha intrusion' (alpha wave activity superimposed on delta waves),which made analysis impossible (the likely explanation being the administration ofantipsychotic medications after enrolment/PSG recording began). Observation of EEGdelta wave activity in some patients who appeared to be awake has been noted in otherstudies [4, 5, 39]. Challenges in scoring ICU patients' sleep data were recently reported byDrouot et al. [39] in their analysis of PSG data from two studies conducted previously. Theinvestigators noted the presence of EEG delta wave activity during apparentwakefulness (the presence of EMG, EOG and limb activity) and a lack of K complexesand sleep spindles preventing classification of stage 2. However, interrater andintrarater scorer reliability checks were not performed in that study.

Another limitation is the effect of benzodiazepine medications on the interpretationof sleep parameters. It is known that benzodiazepine medication increases EEG betawave activity and reduces EEG delta wave activity [40]. Benzodiazepine medication also increases EEG spindle activity [41] (though this appears not to have been the case in our study). The combinedeffect of the opioids and sedative medication on EEG activity likely affected theresults. However, since these medications are essential adjuncts in the treatment andcomfort of many ICU patients, excluding patients who received them would haveseverely limited recruitment and the generalisability of our results. Otherlimitations were the presence of factors such as the use of different modes ofmechanical ventilation and health conditions such as systemic inflammatory response,but such factors are common in ICU patients and their exclusion would also limit theapplicability of the results to the ICU patient population.

Patient enrolment was limited by a number of factors. The availability of PSGmonitoring equipment and the principal researcher were the primary factors forpatients being discharged before they were approached to participate. For this study,we had access to only two people trained in sleep recordings (RE and MF) and oneportable sleep-monitoring device, creating practical limitations to the number andfrequency of studies performed. In addition, there was often limited opportunity inwhich patients were cognitively able to agree to participate before they weredischarged to the hospital ward. A large proportion of patients approached about thestudy declined to participate; many considered the application of monitoring as aset-back in their condition regardless of the reason for it. This may limit thegeneralisability of the results to a sub-set of patients treated in the studyICU.

Interrater reliability of the R and K analysis was moderate, 0.56 for sleeptechnologists one and two and 0.51 for sleep technologists two and three but lowerthan Kappa values for sleep technologists in sleep investigation units (for example0.72 [42]). Interrater reliability for each stage was also lower than in sleeplaboratory studies. The reliability of PSG analysis in ICU patients has beeninfrequently reported. In one study of critically ill, non-ventilated traumapatients' night-time sleep, interrater reliability for one overnight recording, wasreported as Kappa = 0.82 [43]. However, in a comparison by Ambrogio et al. of four methods ofanalysing ICU patients' PSG data, interrater reliability was considerably lower thanin the present study (Kappa = 0.19) [44].


The quantity and quality of patients' sleep while in this ICU using both objective andsubjective assessment methods were found to be poor. Given the similarity between thesleep outcomes and prevalence of potential sleep-disturbing factors in our study andprevious studies, sleep disruption clearly remains a substantial problem for many ICUpatients. Importantly, the current study has added to evolving knowledge of thechallenges of using PSG in ICU and analysing the data. An alternative objective sleepassessment method is required for ICU patients in order to further our understanding ofsleep disruption in this vulnerable patient population and to test clinicalinterventions for their well-being and recovery.

Key messages

·Sleep in ICU patients is highly fragmented with concomitant deficiencies in SW andREM sleep.

·Use of PSG in ICU patients and analysing the data are challenging.

·There is a need to develop alternative methods to conventional PSG staging tomeasure sleep in ICU patients and develop interventions which will improve sleep.