Sharp and sleepy: evidence for dissociation between sleep pressure and nocturnal performance


Pr P. Philip, GENPPHASS, CHU Pellegrin, Place Amélie Raba-Léon, 33076 Bordeaux Cedex, France. Tel.: +33 5 57 82 01 72; fax: +33 5 56 79 48 06; e-mail:


While sleep restriction decreases performance, not all individuals are equal with regard to sensitivity to sleep loss. We tested the hypothesis that performance could be independent of sleep pressure as defined by EEG alpha–theta power. Twenty healthy subjects (10 vulnerable and 10 resistant) underwent sleep deprivation for 25 h. Subjects had to rate their sleepiness (Karolinska Sleepiness Scale) and to perform a 10-min psychomotor vigilance task (PVT) every 2 h (20:00–08:00 hours). Sleep pressure was measured by EEG power spectral analysis (alpha–theta band 6.0–9.0 Hz). Initial performance, EEG spectral power and KSS score were equal in both groups (anova, NS). The performance of vulnerable subjects significantly increased during the night (ranova, P < 0.01), whereas resistant subjects globally sustained their performance. Homeostatic pressure and subjective sleepiness significantly increased during the night (ranova, P < 0.01) identically in both categories (ranova, NS). Resistant subjects sustained their reaction time independently of the increase in homeostatic pressure. The phenotypic determinants of vulnerability to extended wakefulness remain unknown.


The demands of our modern societies are such that many workers or drivers need to sustain wakefulness throughout the night. Extended wakefulness through the night is consistently associated with an increase in subjective sleepiness and with decreased neurobehavioral performance (Philip and Akerstedt, 2006). Unfortunately, subjective sleepiness is not the best individual predictor of nocturnal performance decrements (Leproult et al., 2003).

Variations in both subjective and objective alertness during extended wakefulness vary under the dual regulation of a circadian rhythm and a homeostatic process. In addition, waking systems could play a significant role in the nocturnal maintenance of performance (Dagan and Doljansky, 2006). Sleep homeostasis is related to the duration of wakefulness and is one of the main regulators of sleep–wake patterns (Dijk et al., 1990). EEG power during wake (alpha–theta waves) (Aeschbach et al., 1999; Cajochen et al., 2000) or sleep (delta power) (Borbely, 1982) has been used to monitor sleep pressure. While both of these measures seem to mirror sleep homeostasis, they are not significantly correlated (Leproult et al., 2003). Sleep restriction or extended wakefulness increase sleep pressure and decrease performance in consequence, possibly because of sleep state instability (Doran et al., 2001).

While there is no doubt that sleep pressure is responsible for performance decrements and accidents in the population at large (Connor et al., 2002), not all individuals are equally affected by sleep loss, some being more resistant than others (Philip et al., 2004). Stability of responses suggests that individual differences in neurobehavioral deficits from sleep loss constitute a trait differential vulnerability (Van Dongen et al., 2004, 2005). Night-to-night stability of the EEG spectrum in sleep implies the existence of inter-individual trait-like differences in sleep EEG (Tan et al., 2001). Trait variability seems particularly strong for slow wave sleep and for quantitatively assessed delta power in non-REM sleep EEG (Tucker et al., 2007).

It is important to explore individual differences in subjective and objective vulnerability to minimize safety risks and maximize performance in shift work (Van Dongen, 2006). The identification of predictors for vulnerability to sleep loss would have important implications in multiple domains (health, safety, productivity, etc.). To date, no strong evidence for reliable sleep variable predictors has yet emerged (Leproult et al., 2003; Tucker et al., 2007; Van Dongen et al., 2004).

Recent findings show that circadian markers do not differentiate resistant from vulnerable subjects (Leproult et al., 2003; Van Dongen et al., 2004). Leproult et al. (2003) examined the inter-relationship between subjective, objective (i.e. cognitive performance) alertness and EEG measures of alertness during 27 h of continuous wakefulness. They concluded that sleep deprivation has highly reproducible, but independent, effects on brain mechanisms controlling subjective and objective alertness. Subjective alertness and cognitive performance varied over the same time course but were quantitatively uncorrelated. Moreover, objective alertness was temporally associated with an increase in EEG delta activity, although the magnitudes of these two effects were not significantly related.

Additionally, self-evaluation of sleepiness and performance are not correlated in either resistant or vulnerable subjects (Van Dongen et al., 2004). Moreover, vulnerable and resistant subjects do not differ in terms of sleep needs (Van Dongen et al., 2004). Subjective alertness or sleep trait characteristics do not seem to be adequate measures to predict individual cognitive performance impairment.

To find a predictor to cognitive impairment induced by sleep deprivation, we designed a protocol to analyze the relationship between sleep pressure (assessed by the gold standard technique, i.e. EEG spectral analysis) and performance in groups of subjects unequally affected by sleep deprivation (i.e. vulnerable versus resistant).

Materials and methods


Twenty male subjects were divided into two groups according to their response to sleep deprivation. Vulnerable subjects increased their 10% slowest reaction times by more than 300% over the night and resistant subjects increased their 10% slowest reaction times by less than 100% over the night.

The 10 healthy resistant subjects had a mean age of 20.8 ± 1.5 years (range 19–23) and a BMI of 22.3 ± 1.3 (range 19.9–24.4). The 10 healthy vulnerable subjects had a mean age of 21.0 ± 1.7 years (range 18–24) and a BMI of 22.1 ± 2.2 (range 20.2–27.8). Mean age and BMI did not differ between the groups (t = −0.275, NS and t = 0.252, NS respectively). All subjects provided written informed consent, and the study was approved by the local ethics committee.

Inclusion criteria

Subjects underwent a clinical evaluation and an interview with a sleep specialist. We used actimetry recordings (Shinkoda et al., 1998) to quantify the usual sleep duration and sleep efficiency. A nocturnal polygraphy was performed to eliminate any subjects suffering from sleep disorders. In addition, each individual had 5 days of actimetry prior to inclusion. All subjects presenting sleep efficiency lower than 85% were excluded from the study. We decided to include exclusively males in this study to avoid a source of variability in the results due to the gender factor. Effectively, many female-related characteristics (timing of menstrual cycle, method of contraception, etc.) are known to interfere with sleep, alertness and cognitive performance.

Five-day actimetry results prior to inclusion were: resistant (total sleep time 439.9 ± 48.4 min; sleep efficiency 92.0 ± 3.7%); vulnerable (total sleep time 453.9 ± 43.0 min; sleep efficiency 92.1 ± 5.3%).

The selected participants were instructed to maintain a regular sleep–wake schedule (controlled by actimetry) for 3 days before the study period.


To investigate the relationship between homeostatic drive and performance, we selected subjects as vulnerable (more than 300% increase in 10% slowest reaction times over the night) or as resistant (less than 100% increase in 10% slowest reaction times over the night) to sleep loss on the basis of their performance on a psychomotor vigilance task (PVT) during a first selection night.

Subjects were then tested in the laboratory at 2-h intervals for 12 h continuously from 20:00–08:00 hours after a usual sleep night at home (from 23:00–07:00 hours). The waking EEG was continuously monitored during the 12-h period of the study. No stimulant use of any kind was allowed during the study.

Visual Analogue Scale and Karolinska Sleepiness Scale

Subjects rated their subjective sleepiness assessment on a 100-mm Visual Analogue Scale (VAS) from ‘excellent’ to ‘very poor’ and on the Karolinska Sleepiness Scale (KSS) (Akerstedt and Gillberg, 1990) every 2 h.

Psychomotor vigilance task

They performed a 10-min simple reaction time test (SRTT) on a PALM personal organizer (Gillberg et al., 1994; Philip et al., 1999). A black square was displayed 100 times on the screen at randomized (2–7 s) intervals over 10 min. The subject’s task was to respond to the stimuli by pressing a key to turn the square off. If no response was given within 1.75 s, a new interval was started. Pressing the key before the square was displayed, or within 100 ms, caused the response to be discarded and a warning to be displayed. The software that controls the internal clock yields data with at least 0.01 ms resolution (the keyboard is sampled at CPU frequency divided by the number of instructions needed for sampling). Another part of the program calculates the 10% slowest responses (Dinges et al., 1997) during the 10-min task. The 10% slowest responses were used as the dependent variable. It is therefore possible to measure sensitive variations of responses which are noted during an increased instability of sleep–wake status, as observed during sleep deprivation in humans (Doran et al., 2001). Compared with a simple mean, the technique extracts the slowest reactions which are characteristic of episodes of sleepiness but does not take into account the mean reaction time and the fastest reaction times, which are more influenced by age than by sleepiness.


Sleep pressure was measured by EEG power spectral analysis on the alpha–theta band (6.0–9.0 Hz). Every hour from 20:00–08:00 hours, the waking EEG signal was recorded during a 4-min eyes-open session of the Karolinska Drowsiness test (Akerstedt and Gillberg, 1990). The participants had to look at a picture on the ceiling and avoid movement to obtain 20-s epochs without eye blinking, sleep, artifacts due to body movements, slow eye movements or sweating. A trained technician checked the EEG quality on line. EEG (FP1/FP2, F3/F4, C3/C4, P3/P4, O1/O2), electromyogram and electro-oculogram were recorded on a Coherence Polysomnography system (Deltamed®, Paris, France). Signals were digitized at a sampling rate of 256 Hz (according to Coherence software specifications) and filtered with a digital filter at a cut-off frequency of 70 Hz. EEG signals of all derivations were subjected to spectral analysis by a fast Fourier transform (Coherence software). Power spectra were computed for consecutive 4-s epochs, providing a frequency resolution of 0.25 Hz. Absolute amplitudes were added per 0.25 Hz narrow bands for frequencies between 0.5 and 48.0 Hz. Values obtained were averaged for each 20-s epoch selected.

Statistical analysis

Subjects’ results prior to inclusion [age, BMI, sleep duration (five nights) and sleep efficiency] are presented as mean and standard deviation. They were compared between groups using Student’s t-test.

Actimetry results the night before the experiment (sleep duration and sleep efficiency) are presented as mean and standard deviation. They were compared between groups using Student’s t-test. Initial performance, sleepiness and EEG (at 20 h) were compared between groups using a one-factor anova. The effect of the group (resistant versus vulnerable) and time of testing on performance, sleepiness (KSS) and EEG were evaluated by anova for repeated measurements (ranova). To correct for sphericity, all P-values derived from ranova were based on Huynh–Feldt’s corrected degrees of freedom. When normality was not respected (tested by Skewness and Kurtosis), the data were transformed into a z-score. The 10% slowest were normalized as a percentage (100% represents the mean 10% slowest reaction time at 20:00 hours). EEG was z-transformed.



The night before sleep deprivation, actimetric recording indicated an estimated mean total sleep time of 436.0 ± 29.7 min for resistant subjects and 420.4 ± 36.4 min for vulnerable subjects, and there was no difference between the groups (t = 1.048, NS). Sleep efficiency was 91.4 ± 5.2% for resistant subjects and 88.6 ± 7.0 for vulnerable subjects, and there was no difference between the groups (t = 1.016, NS).


Initial performance (10% slowest 20:00 hours) did not differ between both groups (anova; F = 0.214, NS) and was normally distributed. A two-factor repeated measures anova with category (vulnerable versus resistant subjects) and session (time of testing) as main explanatory factors showed a significant effect for session (ranova; F = 19.238, P < 0.01), category (ranova; F = 10.008, P < 0.01) and an interaction between session and category (ranova; F = 6.251, P < 0.01). Figure 1 shows the different increase in performance during the night in the two groups.

Figure 1.

 Ten per cent slowest SRTT expressed as percentage (reference value at 20:00 hours = 100%) mean, resistant versus vulnerable, with standard errors.

Karolinska Sleepiness Scale

Initial KSS values (20:00 hours) did not differ between both groups (anova; F = 0.340, NS). A two-factor repeated measures anova with category (vulnerable versus resistant subjects) and session (time of testing) as main explanatory factors showed no significant effect for category (ranova; F = 1.489, NS), but a significant effect for session (ranova; F = 74.795, P < 0.01). Figure 2 shows a similar increase in sleepiness on the KSS during the night in the two groups.

Figure 2.

 Karolinska Sleepiness Scale scores, resistant versus vulnerable, with standard errors.


Initial EEG values (20:00 hours) did not differ between the groups (anova; F = 0.314, NS). For all derivations (FP1–F3/FP2–F4, F3–C3/F4–C4, C3–P3/C4–P4, P3–O1/P4–O2), a two-factor repeated measures anova with category (vulnerable versus resistant subjects) and session (time of testing) as main explanatory factors showed no significant effect for category (ranova; NS), but a significant effect for session (FP1–F3/FP2–F4, F3–C3/F4–C4, C3–P3/C4–P4, P3–O1/P4–O2, ranova; F = 4.075, P < 0.01; F = 4.607, P < 0.01; F = 4.294, P < 0.01; F = 2.683, NS). Figure 3 shows a similar increase in EEG spectral power during the night in the two groups.

Figure 3.

 EEG FP–F derivation, alpha–theta band (6.0–9.0 Hz), resistant versus vulnerable, with standard errors.


This study is the first to confirm Leproult et al.’s (2003) results of dissociation between objective performance and sleep pressure as measured objectively by alpha–theta EEG. In a previous study (Philip et al., 2004), we did not include objective measurements of sleep pressure and were unable to demonstrate clearly any dissociation between performance and sleep pressure owing to a lack of EEG recordings, even though in normal subjects there is a strong correlation between self-perception of sleepiness and EEG spectral changes (Akerstedt and Gillberg, 1990).

In agreement with previous works (Carskadon and Dement, 1981; Frey et al., 2004; Herscovitch and Broughton, 1981; Philip et al., 2004; Sangal et al., 1999; Van Dongen et al., 2003), this study confirms a dissociation between objective performance and sleep pressure as measured by subjective sleepiness.

We confirm in this new study that performance is dissociated from sleep pressure as assessed by EEG in resistant and vulnerable subjects. With similar sleep duration the day before the experiment, subjects showed great inter-individual variability in performance under sleep deprivation. This goes in line with the results of Van Dongen et al., (2004) showing a lack of relationship between sleep needs and resistance to sleep loss. We also observed greater inter-individual variability in vulnerable than in resistant subjects as a function of hours of wakefulness. This variability could be explained by the state instability hypothesis (Doran et al., 2001).

During sleep deprivation there was no topographic EEG difference (left–right) between resistant and vulnerable subjects, which could explain a differential EEG activation between the two groups.

These findings confirm the absence of relationship between sleep pressure during prolonged wakefulness and performance. Moreover, vulnerable and resistant subjects do not differ in terms of sleep needs (Van Dongen et al., 2004). Therefore, individual variability in response to sleep loss seems to be independent of inter-individual variations in sleep homeostasis (Taillard et al., 2003).

The physiological determinants of this resistance remain unknown but could be related to alerting neural pathways. Further research on resistant and vulnerable subjects is needed to analyze inter-individual variability and better understand sleep physiology, especially the domain of individual differences in sleep and wakefulness. A recent study (Retey et al., 2006) suggests that adenosinergic mechanisms contribute to individual differences in the vulnerability to sleep deprivation-induced changes in neurobehavioral function and the regional distribution of EEG power relevant for sleep regulation and attentional processes. By identifying predictors of vulnerability to sleep loss, it would be possible to anticipate and prevent cognitive impairment after sleep loss. By contrast, a study (Philip et al., 2006) comparing caffeine versus napping in sleepy drivers showed that subjects may benefit from a short nap while experiencing no performance restoration with caffeine. These results suggest that adenosinergic mechanisms (which are strongly blocked by caffeine) do not account for vulnerability to sleep in all subjects and that other neural mechanisms could explain these performance decrements.


We thank Racha Rachedi for selecting participants, collecting and monitoring data, and Victor Bibène for collecting and monitoring data.