Edited by: Sairam Parthasarathy, University of Arizona, United States
Reviewed by: Kristen Knutson, Northwestern Medicine, United States; Andrea Romigi, Istituto Neurologico Mediterraneo Neuromed (IRCCS), Italy
This article was submitted to Sleep Disorders, a section of the journal Frontiers in Neurology
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Outcomes following critical illness and discharge from the Intensive Care Unit (ICU) can range from full recovery to varying degrees of disability. There is more and more evidence that sleep deprivation during the ICU stay has both negative short-term effects, such as poor comprehension of discharge instructions and delirium, and lasting serious consequences, such as cognitive impairment, that are of key interest and importance to patients and their families, clinicians, hospitals, and payers (
The purpose of an evaluative measure is to describe a construct of interest in a specific population, and to measure the extent of change in the construct over time. Sleep is a multi-dimensional construct, composed of dimensions such as total sleep time, percent of sleep stages, frequency of awakenings or arousals, expectations, global perceptions, sleep movements, tiredness upon awakening, daytime energy levels, and functional impairments. Various measures of sleep in ICU patients exist, but they do not all measure exactly the same dimensions, and we do not expect them to demonstrate 100% agreement. For example, the Richards Campbell Sleep Questionnaire (RCSQ) (
In this systematic literature review we identified measures of sleep in critically ill adults hospitalized in the ICU and discussed the dimensions of sleep that they measure. We focused on publications from 2000 to 2019 because the ICU environment and care delivery has significantly changed, and earlier studies may not be relevant to the current ICU setting. We evaluated the strengths and weaknesses of the measures, and their clinical and research usefulness. As an essential step, we assessed their measurement properties based on the criteria described by McDowell (
A systematic database search was performed in February 2020. We conducted a search on PubMed/MEDLINE, CINAHL, and Cochrane Library with the following combination of MESH terms/ keywords: sleep AND (critical care OR intensive care OR ICU). The inclusion criteria were: (1) primary sources published from 2000 through 2019; (2) systematic or focused reviews 2000–2019, (3) written in English and electronically available in full-text format; and (4) measured sleep in the ICU using at least one method. The authors evaluated the titles and abstracts of all potentially useful studies based on the inclusion criteria and identified articles for a full-text review. Additional relevant studies, such as those referenced by reviews, were further included. The reviewers reached a consensus on which original research studies were to be included in the review. If an article described a measure developed prior to 2000 we reviewed the original publication.
The electronic database search initially identified 1,096 studies (CINAHL 167, PubMed/Medline 926, Cochrane 3). After removal of duplicates, the titles and abstracts of 1,015 articles were examined, resulting in selection of 81 articles for full-text reading. After exclusion of articles that did not meet the inclusion criteria, and adding articles from reference lists, a total of 62 studies were included in this review.
Search strategies for measures of sleep in critically ill patients. **Due to overlaps, some studies used more than one sleep assessment; *Total number is 62.
All included literature by category.
Cooper et al. ( |
Raymond et al. ( |
Olson et al. ( |
Richards et al. ( |
Richards et al. ( |
Bourne et al. ( |
Ibrahim et al. ( |
Richards et al. ( |
Freedman et al. ( |
Beecroft et al. ( |
Richardson et al. ( |
Richardson et al. ( |
Parthasarathy and Tobin ( |
Chen et al. ( |
Beecroft et al. ( |
Frisk and Nordström ( |
Gabor et al. ( |
van der Kooi et al. ( |
Dennis et al. ( |
Ugras et al. ( |
Hardin et al. ( |
Hamze et al. ( |
Litton et al. ( |
Richardson et al. ( |
Alexopoulou et al. ( |
Kamdar et al. ( |
Aitken et al. ( |
Toublanc et al. ( |
Ambrogio et al. ( |
Naik et al. ( |
Nicolas et al. ( |
|
Beecroft et al. ( |
Hsu et al. ( |
Bourne et al. ( |
|
Drouot et al. ( |
Scotto et al. ( |
||
Kondili et al. ( |
Li et al. ( |
||
Gehlbach et al. ( |
Kamdar et al. ( |
||
Watson et al. ( |
Jones and Dawson ( |
||
Cordoba-Izquierdo et al. ( |
Van Rompaey et al. ( |
||
Alexopoulou et al. ( |
Bihari et al. ( |
||
Su et al. ( |
Little et al. ( |
||
Elliott et al. ( |
Zhang et al. ( |
||
Elliott et al. ( |
Elliott et al. ( |
||
Knauert et al. ( |
Kamdar et al. ( |
||
Alexopoulou et al. ( |
Su et al. ( |
||
Vacas et al. ( |
Maidl et al. ( |
||
Huttmann et al. ( |
Hata et al. ( |
||
Boyko et al. ( |
Storti et al. ( |
||
Boyko et al. ( |
Ugras et al. ( |
||
Demoule et al. ( |
Demoule et al. ( |
||
Menear et al. ( |
|||
Aitken et al. ( |
|||
Rood et al. ( |
|||
Louis et al. ( |
Polysomnography (PSG), a multi-parametric recording of the biophysiological changes based on electroencephalographic (EEG) activity, combined with concurrent polygraphic monitoring of electrooculogram (EOG), electromyogram (EMG), electrocardiogram (ECG), as well as other parameters that occur during sleep, has long been regarded as the gold standard for objectively measuring quality and quantity of sleep for comparatively healthy populations outside of the ICU (
In addition to the 4 types of multi-parametric devices, alternative portable brain function monitors involve the use of processed EEG, such as the Bispectral Index (BIS) (
The original Rechtschaffen and Kales sleep scoring manual (R&K rules) (
The R&K and AASM criteria were developed for recording and scoring sleep in typical healthy individuals, without neuropathology or psychoactive medication use, in the controlled environment of a sleep laboratory or the usual home sleep environment. Applying these standard criteria for recording and scoring sleep to critically ill patients is challenging. Typical ICU environments are noisy, and treatments often are invasive and intensive. Sedatives, analgesics, the stress response, mechanical ventilation, and neuropathology may result in atypical brain waves, muscle tension, eye and body movements. Multiple illness-related factors often are associated with atypical biophysiological sleep activity, e.g., sleep fragmentation with frequent arousals and awakenings, disorganized circadian rhythms, and disrupted sleep architecture (increased stage N1 and N2 sleep, and decreased stage N3 and REM sleep) (
Twenty-five studies (
Most of the included studies used portable, unattended PSG for at least 8 consecutive hours, and only about half of the studies (13/25) used PSG ≥24 h. Five studies (
Over half (15/25) of the studies used standard sleep scoring (R&K or AASM). Although the R&K method showed good to excellent interobserver reliability for assessing sleep in ambulatory individuals (Cohen κ range 0.68–0.82) (
In another study, Drouot et al. (
On the basis of Drouot's work and Young's EEG classification (
Scoring polysomnography and EEG-derived data for sleep in critically ill patients.
Modified delta (mDelta) criteria ( |
• Decreased amplitude of delta waves associated with aging. |
mDelta criteria consisted of a frequency criterion of <4 Hz and an amplitude criterion of >50 μV (peak to peak). | Compared to published normal values, all chemically paralyzed patients in this study had increased delta activity, whether scoring was traditional R&K or mDelta. |
Spectral analysis ( |
• Spectral analysis is an automated method that quantifies EEG activity across the EEG spectrum. |
Sleep scoring by 3 manual methods; (1) R&K, (2) sleep-wake organization pattern, and (3) visual detection of burst suppression; and 1 computer-based method: spectral analysis of EEG signals with FFT | Reproducibility for spectral analysis was better than manual methods (R&K and sleep-wake organization pattern) ( |
Atypical sleep and pathologic wakefulness ( |
Sleep cannot be classified with standard criteria in one third of mechanically ventilated, non-sedated, and conscious ICU patients. | • To add 2 new states: atypical sleep and pathologic wakefulness; Quantitative assessment of sleep/wake EEG patterns: EEG peak frequency, EEG reactivity, EEG power spectra. |
Atypical sleep was predicted with a sensitivity of 100% and a specificity of 97% in non-sedated conscious ICU patients by using this method. |
Revised scoring system incorporating frequently seen atypical characteristics ( |
• Pathologic wakefulness: any EEG frequency other than alpha or beta with behavioral characteristics of wakefulness. |
High interrater reliability (weighted κ = 0.80 [0.48,0.89]) |
A number of processed EEG-based brain function monitors were originally developed to monitor sedation during anesthesia, and some have undergone limited testing as potential measures of the sleep/wake state in critically ill patients. An advantage of most of these monitors in the ICU, vs. PSG, is that a technician does not need to be in attendance to ensure a good recording, and replacement of sensors do not require a skilled technologist as with PSG (
Polysomnography remains the gold standard for evaluating physiological sleep in ICU patients. However, there are a number of challenges, such as technical difficulties (placement and maintenance of electrodes, data interpretation), acceptability by patients, family, and clinical staff (i.e., patients' discomfort, severity of illness and ventilator status, and interference of complex treatment and patient transfer, etc.), as well as additional expense. The greatest challenge to date has been lack of reliability for scoring sleep due to atypical EEG findings often found in ICU patients. Recently, several investigators have addressed this challenge by developing and validating ICU specific scoring rules. We recommend that investigators should always report PSG recording and scoring methods and justify their choices. Compared to PSG, other portable EEG-based monitors are more feasible in ICU patients, but their validity as alternatives to PSG in the ICU setting require further testing.
The actigraph is a motion sensor detector (accelerometer) similar in size to a wristwatch that is used to assess motor activity. The device can be used to determine physiological sleep or waking during each set epoch by counting activity within a defined threshold (
Relatively few investigators have studied the measurement attributes of actigraphy in critically ill patients. We found nine studies: (
Beecroft et al. (
In another study, van der Kooi et al. (
A recently published article concluded that actigraphy is an objective and relatively reliable measure compared to nurse observations (
An actigraph is usually applied to the wrist, but the ankle can be used. Kamdar et al. (
Three experimental studies (
Actigraphy in critically ill patients: design, sample, performance, feasibility, responsiveness, and recommendations.
Raymond et al. ( |
Observational |
MicroMini Motionlogger Actigraph (Ambulatory Monitoring Inc.). 1-min epochs, Hi-PIM |
Actigraphy total sleep time mean = 332 min (sd 105) and # awakenings mean = 25.8 (sd 9.5) vs. questionnaire total sleep time mean = 391 min (sd 142) and # awakenings mean = 3.8 (sd 7.5) | N/A | Actigraphy underestimated time slept and overestimated awakenings compared to patient questionnaire. |
Bourne et al. ( |
RCT |
Actiwatch (Cambridge Neurotechnology) |
Placebo group SEI actigraphy = 0.75, BSI = 0.26, RCSQ = 0.50, nurse observation = 0.50 |
Not responsive—no between group differences in SEI in melatonin group vs. placebo group, BSI difference, but NS ( |
Actigraphy overestimated sleep efficiency, compared to other measures. |
Beecroft et al. ( |
Observational |
Actiwatch Model AW-64 |
Actigraphy analysis conducted using 4 different activity count thresholds |
N/A | Actigraphy overestimated total sleep time and sleep efficiency compared to PSG |
Chen et al. ( |
RCT (valerian acupressure vs. usual care) |
Actigraph GT1M, ActiGraph, LLC, ActiWeb software |
Large differences between TST with actigraphy vs. observation; for example, 2.3 h TST by observation vs. 7.3 h by actigraphy at baseline |
Responsive. Actigraphy showed a significant within group increase in TST and a reduction in wake minutes in the experimental group. | Actigraphy may have overestimated sleep time, as 7.3 h slept is higher than most other studies. |
van der Kooi et al. ( |
Observational |
Actiwatch (Cambridge Neuro-technology) |
The median sensitivity of actigraphy to detect sleep was 94%; median specificity for detection of awakenings was 19% | N/A | Limitations—sample size ( |
Hamze et al. ( |
Descriptive |
Actisleep (Actigraph Corporation), Actilife version 5 software |
529 care interventions were recorded by the nurses, but only 21 awakenings were scored by the actigraph. |
N/A | The actigraph had a transparent film over it that may have affected results. |
Kamdar et al. ( |
Prospective observational |
Actiwatch Spectrum (Philips Respironics) |
0 movement in 83% of epochs (ankle) and 64% of epochs (wrist); likely overestimated sleep |
N/A | Sleep differed based on placement. |
Naik et al. ( |
Cross-sectional |
Actigraph SOMNOwatch (SOMNOmedics GMbH) RCSQ |
Mean TST actigraphy 6.3 h (sd 1.7), RCSQ = 51.6 (sd 13.5) | N/A | ICU TST higher than other studies, actigraphy may have overestimated sleep, also younger subjects in this study (mean = 36.8 years, sd 12.7) |
Hsu et al. ( |
Experimental with back massage vs. usual care |
Actiwatch 2 (Philips Respironics) |
Actigraph TST = 5.9 h |
Responsive. |
Limitations: no data on sedatives or correlation of actigraphy, VSH, and observation measures |
An actigraph is a non-invasive device used to measure objective sleep quality and has been regarded as an acceptable substitute for PSG due to its lower cost and user-friendliness. Actigraphy is easier to tolerate than multiple PSG leads and provides objective data that is somewhat consistent with PSG. In addition, actigraphy shows moderate responsiveness to interventions as evidenced by improvement in the expected direction and consistency of response with other outcome measures in two of the three clinical trials. The primary weakness of actigraphy is that sleep/wake determination is based on movement, or lack thereof, and ICU patients have reduced movement regardless of their sleep-wake status.
In general, in critically ill patients, actigraphy tends to show higher total sleep time, better sleep efficiency, and more nighttime awakenings compared with PSG, and more overall awakenings compared to nurse assessment and patient questionnaires. However, we identified only nine published articles during the past 20 years, and only five of the nine studies evaluated measurement properties. In general sample sizes were small, and often data were collected for only one night. Also, important information was often lacking, such as PSG scoring reliability and method for dealing with atypical EEG waveforms. Up to 1/3 of PSG data is unable to be reliably scored using standard criteria. Therefore, research on actigraphy in critically ill patients is needed with larger sample sizes, longer durations, and specific sensors and software tested for critically ill individuals who often have low mobility states and often receive sedatives and analgesics. If PSG is used as the comparison, investigators should provide detailed discussion on how atypical EEG waveforms were scored and reliability of scoring. In addition, the ICU population is quite heterogeneous, and the exclusion criteria for reliable actigraphy requires further discussion and consensus. In general, specificity for identifying wake in actigraphy is lower than expected, and sedation, analgesia, and immobility are likely to influence specificity.
Structured observation, also known as systematic observation, is a method for collecting data in which researchers (or clinicians) gather data without direct involvement of participants. Coding of the data is done using previously determined specific behavioral actions. Specific criteria for the behaviors are developed and validated. Data are most often collected by clinicians or research assistants, who have been trained and verified as competent to identify the behaviors. Interrater reliability (consistency between data collectors on coding the behavior) is important, especially in a setting where multiple clinicians or researchers collect data. The observations may be continuous over a specified period of time or completed at specific intervals. Sometimes the data are captured via video, and later scored or coded by humans, or more recently by technology, using specific criteria. Structured observational measures have been used extensively in other populations, for example non-verbal children and older adults with dementia, to measure or identify various behaviors, such as pain.
A few clinician observation sleep tools have been developed and used to identify sleep in ICU patients. These tools identify sleep by structured observations conducted by staff nurses, often in the course of their clinical care, or the tools are used to collect information on patient's sleep retrospectively from the clinicians.
In this review we located 7 studies that used a clinician observed sleep measures (
The SOT, developed and validated by Edwards and Schuring, was designed for nurses in the ICU to assess patient's sleep and wake states at 15-min intervals (
The SOT was used by Litton and colleagues to assess sleep disruption in the ICU in a large prospective multi-site observational study (
Other clinician observation measurements for sleep in the ICU use visual cues such as eye closure and not moving to determine sleep duration (
Observation of sleep in critically ill patients: performance, feasibility, responsiveness, missing data, and recommendations.
SOT ( |
6 trained nurse observers |
Yes | Yes, sleep, as measured by the SOT, changed in the expected direction. Patients were 1.6 times more likely to be asleep during the intervention compared to the control ( |
Not reported | Limitation: nurses were not blinded to intervention, or the light and sound measures, and lack of blinding may have affected responsiveness |
Number of hours of observed sleep—by bedside nurses ( |
Observational criteria for sleep were eyes closed, decreased motor activity, lack of interaction with the environment, and lack of purposeful activity. Validity of measure, training of observers, and IRR were not mentioned. | Yes | No. Placebo TST = 240 min (range 75–331.3) vs. Melatonin TST = 243.4 min (range 0–344) | Not reported | Recommend training of nurse observers, competency assessment, and assessment of IRR prior to, and quarterly during data collection |
Investigator-developed single item ordinal scales: (1) hours slept, and (2) comparison with normal sleep |
Reliability and validity not discussed |
Yes | N/A. No intervention | Not reported | Strength: neither the nurse nor patient was aware of the other's rating |
Hours slept and number of awakenings at the end of shift—by bedside nurses |
TST (hours)—Observation 5.35, PSG 3.10, Actigraphy 4.43 |
Yes | N/A, no intervention | Not reported | Nurses reported better sleep than measured by either PSG or actigraphy |
SOT ( |
The SOT was not compared with other measures in this study. |
Yes | Yes, the SOT was responsive. The results showed a change in sleep, in the expected direction, during the Quite Time intervention, compared to pre/post-test. | Not reported | Limitation: the nurses collecting the sleep data were unblinded to experimental condition |
SOT ( |
The SOT was not compared with other measures in this study. | Unclear-large amount of missing data—reasons not discussed. | N/A No intervention | Sleep data missing in 163 (33%) of sample | Recommend using behavioral assessment of sleep combined with actigraphy, PSG, and other technologies to improve sleep/wake identification in objective measures. |
Bedside nurses documented in the electronic medical record categorical data: no sleep, minimal sleep, moderate sleep, majority sleep, or slept all night ( |
Validity—RCSQ Questions 1–5 and nurse observation were significantly correlated (Spearman's rank correlation = 0.39–0.50, |
Yes | N/A, no intervention | Not reported | Recommend that nurses assess and document sleep quality and quantity as part of routine clinical care |
Clinician observed sleep is particularly appealing for ICU patients who cannot provide information on perception of their sleep. Nurses are at the forefront of patient care, and they can provide important information on sleep while they are assessing other vital signs. Nurse observed sleep tools have the potential to be integrated into routine clinical practice, similar to pain assessments. In general, nurse observed sleep duration has shown good validity compared to other methods. For example, the SOT agreement with PSG-identified sleep was 81.9% (
There are several methodological weaknesses in the literature and caveats regarding clinician observation of sleep. In studies to date, there is insufficient discussion of nurse observer training, agreement among the nurse observers, and discussion of any problems with missing data. These weaknesses may affect responsiveness in future clinical trials using these observational methods. Other potential limitations to structured observations are the potential to accidentally awaken the patient during the observation, blinding of nurses to intervention condition, and issues with feasibility such as insufficient nursing time for the observations. In other populations, trained research assistants often collect observational data. Investigators might consider research assistants, instead of nurses, for collecting observational data on sleep, especially when the nature of the intervention prevents blinding of the nurses.
Perception of sleep quality is an important dimension of sleep that may not be captured by objective measures. Decades of research have shown differences between sleep state perception and objectively measured sleep in a number of clinical sleep populations, most notably insomnia sufferers (
The RCSQ is a five-item visual analog scale, measuring five domains of sleep, including sleep latency, sleep efficiency, sleep depth, number of awakenings, and overall sleep quality (
The Verran Synder-Halpern Sleep Scale (VSH) is a visual analog scale (9–15 items, depending on version) that was originally developed and validated for measuring perception of sleep in healthy adults. It has subsequently been used and validated for sleep measurement in critically ill patients in several studies (
Storti and colleagues developed a 9-item questionnaire, the Coronary Care Unit Questionnaire (CCUQ) to assess sleep in the coronary care unit (
In a recent study, Rood et al., conducted a large validation study (
Patient questionnaires for measuring sleep in critically ill patients: performance, feasibility, responsiveness, missing data, and recommendations.
RCSQ ( |
The RCSQ is a 5-item questionnaire for patients to evaluate sleep depth, sleep latency (time to fall asleep), number of awakenings, sleep efficiency, and sleep quality. Each response is recorded on a 100 mm visual analog scale, with higher scores indicating better sleep and the total score representing overall perception of sleep quality. Internal consistency reliability was 0.90 and principal components factor analysis revealed a single factor (Eigenvalue = 3.61, percent variance 72.2). The RCSQ was significantly correlated with PSG variables, and total score accounted for about 33% of the variance in sleep efficiency index by PSG ( |
ICU patients, total |
|
VSH sleep scale ( |
The VSH sleep scale consists of visual analog items measuring perception of sleep the preceding night. Reliability coefficient was 0.82 (theta) in original 8-item scale, with 2-factors, disturbance and effectiveness; correlation with items on other validated sleep scales ranged from |
ICU patients, total |
|
CCUQ ( |
The CCUQ was designed to evaluate sleep quality in the coronary care environment. It measures factors that impact sleep quality, such as bed quality, light, and noise and consists of 9 items, 1–5 points each, in Portuguese, with total scores ranging from 18 to 90 points, and higher scores indicating better sleep. In a validation study ( |
The CCUQ scale is unique because it measures factors that impact sleep in the ICU. It shows acceptable reliability and significant correlation with sleep efficiency as measured by PSG. | |
NRS—Sleep ( |
The NRS—Sleep is a single item numeric rating scale for ICU patients. The validation study was conducted in two phases. In the first phase, 468 ICU patients were enrolled, and 194 assessed sleep quality using the RCSQ and the NRS—Sleep. The NRS—Sleep significantly correlated with the RCSQ ( |
The NRS is comparable to the RCSQ to assess sleep quality and is a feasible method to monitor sleep in everyday clinical practice. |
|
Investigator developed/modified tools ( |
A variety of tools have been created or adapted, and used to collect information from ICU patients on sleep history, sleep disturbing factors, and sleep quality and quantity. |
ICU patients, total |
Patient perception of sleep quality is an important dimension for sleep clinicians and investigators to monitor, and it is of key importance to patients and their families. Similar to the findings from the recent review by Jeffs and Darbyshire (
This review evaluated the measurement properties, feasibility, and responsiveness of existing instruments used to evaluate sleep in patients hospitalized in the ICU. An extensive search strategy resulted in 62 articles. We divided the instruments into 4 groups based on the dimensions of sleep they measured: (1) physiological sleep measured by polysomnography and other EEG-based methods, (2) actigraphy, (3) clinician observation, and (4) patient perception of sleep using questionnaires.
Sleep is multi-dimensional, composed of dimensions such as total sleep time, awakenings, expectations, global perceptions, movements, tiredness upon awakening, daytime energy, and function. Various measures of sleep in ICU patients exist, but they do not all measure exactly the same dimensions. Traditional measurement science specifies that while we expect correlation between dimensions of a construct, we would not expect the different dimensions to demonstrate 100% agreement. In general, we want to emphasize that PSG, other EEG-based methods, actigraphy, clinician observation, and patient perceptions provide complementary, but somewhat different information on sleep quality in critically ill patients. Given the multiple dimensions of sleep in critically ill patients, we highly recommend using multiple measures of sleep, especially in clinical trials. Clinical trialists should carefully consider sensitivity of outcome variables derived from each of the various sleep measurement methods, especially when choosing primary outcome variables for use in clinical trials. In general, awakenings are difficult to reliably capture in methods other than polysomnography.
Physiological sleep measured by PSG provides precise, objective information on sleep latency, sleep continuity, percent of sleep stages, sleep duration, and other objective sleep parameters. In general, it has excellent validity for recording physiological sleep. Although labor intensive, it is certainly feasible, as evidenced by the relatively large number of studies that have used PSG to study sleep in critically ill patients. The main drawback of PSG is reliability of scoring using standard AASM criteria due to the absence of stage N2 markers, polymorphic delta, burst suppression, use of sedating medications, electrical interference in the ICU, shivering, and other abnormalities or underlying illnesses. Several investigators have developed and validated alternative scoring methods for critically ill patients, but most studies to date have not used these new scoring methods. We recommend that investigators report and justify PSG scoring methods, and report scoring interrater reliability.
The traditional stage scoring of polysomnographic records provides basic information of sleep macroarchitecture, however, this method may be insufficient to detect sleep abnormalities in ICU patients who suffer from critical illnesses, external stimuli (e.g., psychotropic medications, ventilation, light, noise, treatment, and care), as well as potentially undiagnosed sleep disorders. Studies that have examined sleep microstructures in other populations provide insight into better understanding sleep abnormalities in ICU patients. For example, there can be significant arousal-related phasic events, even when the macrostructure of sleep appears to be normal (
Another challenge for observational PSG studies in critically ill patients is the selection of a control group for comparison. In all cases, we recommend that comparison groups should be matched on age, gender, and any other relevant factors. Investigators should also consider matching on relevant pre-morbid factors, such as reported sleep quality and overall health. Other important considerations, depending on the study aims, are mechanical ventilation and mode of ventilation, sleep promotion protocols, and sedating medications.
Compared to PSG, few studies of other portable EEG-based monitors have been conducted. While collecting data using EEG-based monitors is less labor-intensive than PSG, their validity in the ICU setting requires further testing. We recommend, when possible, they be used along with other methods to provide valuable validity data.
In this review, we identified only 9 actigraphy studies in critically ill patients, most had small sample sizes, only about ½ evaluated measurement properties, and reliability of PSG scoring was infrequently mentioned. Overall, actigraphy tended toward more total sleep time, higher sleep efficiency, and more frequent nighttime awakenings compared to PSG, and more overall awakenings compared to nurse assessment and patient questionnaires. Additional research on actigraphy in critically ill patients is needed with larger sample sizes, longer durations, and specific sensors and settings for low mobility states. In addition, the exclusion criteria for reliable actigraphy in the ICU population requires further discussion and consensus.
Systematic clinician observation for sleep and wake states by nurses or other trained personnel is a good choice, especially for those ICU patients who are unable to self-report. The clinician observation method requires that coding of sleep or wake data is based on the presence or absence of specific behaviors, and that the data collectors have been trained and verified as competent. Similar to PSG scoring, interrater reliability (consistency between data collectors on coding the behavior) is important. We recommend that clinicians and investigators use the SOT because it was excellent agreement with PSG-identified sleep. Nurse observed sleep tools have the potential to be integrated into routine clinical practice, similar to pain assessments. Unfortunately, in studies to date, there is insufficient attention paid to nurse observer training, agreement among the nurse observers, and discussion of missing data. Other potential limitations of systematic observation methods are the potential to accidentally awaken the patient during the observation and issues with feasibility such as insufficient nursing time. In clinical trials, it is important that those collecting the outcome data are blinded to group assignment, which may preclude nurses from collecting observational data in some clinical trials. Investigators might consider research assistants for sleep observations when the nature of the intervention prevents blinding.
Sleep questionnaires measure patients' perceptions of their sleep quality. A limitation for all self-assessment tools is that patients have to be alert, oriented, and able to respond and provide feedback. However, perception is an important dimension of sleep that may not be totally captured by other objective measures. Perception of ICU sleep is influenced by many factors, including usual home sleep quality and patterns, and expectations. We recommend the RCSQ for sleep assessment in ICU patients, based on its reliability and validity, and feasibility. The RCSQ also has the advantage of several validated translated versions for non-English speakers. We recommend that the NRS—Sleep be incorporated into routine ICU assessment.
Future directions for ICU sleep research might include new methods for identifying sleep using machine learning to analyze the multitude of data already continuously collected in ICUs, such as heart rate, blood pressure, and oxygen desaturation to identify sleep and wake, and perhaps NREM and REM sleep. Accuracy of these methods might be improved by adding additional sleep-specific devices, such as the EOG and EMG. Another idea to improve the feasibility and accuracy of observation of sleep and wake, or replace it, is face recognition technology.
In conclusion, there is ample evidence that sleep deprivation during the ICU stay has negative short-term effects, and serious lasting consequences that are of key importance to patients. Measuring the impact of interventions to improve sleep and prevent sleep deprivation requires reliable and valid sleep measures, and investigators have made good progress developing, testing, and applying these measures in the ICU. We recommend future large, multi-site intervention studies that measure multiple dimensions of sleep, and provide additional evidence on instrument reliability, validity, feasibility, and responsiveness. We also encourage testing new technologies to augment existing measures to improve their feasibility and accuracy.
KR, Y-yW, JJ, and LY planned the manuscript. KR, Y-yW, JJ, and LY wrote the manuscript and carried out the subsequent revisions. Y-yW provided primary contributions to polysomnography part. JJ provided primary contributions to actigraphy part. LY provided primary contributions to subjective measurements part. KR provided substantial contributions to the whole process of manuscripts writing and revisions. All the co-authors provided substantial contributions to the first manuscript draft and subsequent revised versions.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.