RAPID PUBLICATION

The Cumulative Cost of Additional Wakefulness: Dose-Response Effects on

Neurobehavioral Functions and Sleep Physiology From Chronic Sleep Restriction
and Total Sleep Deprivation
Hans P.A. Van Dongen, PhD;1 Greg Maislin, MS, MA;1 Janet M. Mullington, PhD;2 David F. Dinges, PhD1

1Unit for Experimental Psychiatry, Division of Sleep and Chronobiology, Department of Psychiatry, and Center for Sleep and Respiratory Neurobi-
ology, University of Pennsylvania School of Medicine; 2Beth Israel Deaconess Medical Center and Harvard Medical School

Objectives: To inform the debate over whether human sleep can be sleep deprivation showed that the latter resulted in disproportionately
chronically reduced without consequences, we conducted a dose- large waking neurobehavioral and sleep δ power responses relative to

Downloaded from https://academic.oup.com/sleep/article/26/2/117/2709164 by guest on 26 November 2023

response chronic sleep restriction experiment in which waking neurobe- how much sleep was lost. A statistical model revealed that, regardless of
havioral and sleep physiological functions were monitored and compared the mode of sleep deprivation, lapses in behavioral alertness were near-
to those for total sleep deprivation. linearly related to the cumulative duration of wakefulness in excess of
Design: The chronic sleep restriction experiment involved randomization 15.84 h (s.e. 0.73 h).
to one of three sleep doses (4 h, 6 h, or 8 h time in bed per night), which Conclusions: Since chronic restriction of sleep to 6 h or less per night
were maintained for 14 consecutive days. The total sleep deprivation produced cognitive performance deficits equivalent to up to 2 nights of
experiment involved 3 nights without sleep (0 h time in bed). Each study total sleep deprivation, it appears that even relatively moderate sleep
also involved 3 baseline (pre-deprivation) days and 3 recovery days. restriction can seriously impair waking neurobehavioral functions in
Setting: Both experiments were conducted under standardized laborato- healthy adults. Sleepiness ratings suggest that subjects were largely
ry conditions with continuous behavioral, physiological and medical mon- unaware of these increasing cognitive deficits, which may explain why the
itoring. impact of chronic sleep restriction on waking cognitive functions is often
Participants: A total of n = 48 healthy adults (ages 21–38) participated in assumed to be benign. Physiological sleep responses to chronic restric-
the experiments. tion did not mirror waking neurobehavioral responses, but cumulative
Interventions: Nocturnal sleep periods were restricted to 8 h, 6 h or 4 h wakefulness in excess of a 15.84 h predicted performance lapses across
per day for 14 days, or to 0 h for 3 days. All other sleep was prohibited. all four experimental conditions. This suggests that sleep debt is perhaps
Results: Chronic restriction of sleep periods to 4 h or 6 h per night over best understood as resulting in additional wakefulness that has a neuro-
14 consecutive days resulted in significant cumulative, dose-dependent biological “cost” which accumulates over time.
deficits in cognitive performance on all tasks. Subjective sleepiness rat- Key Words: chronic sleep restriction, partial sleep deprivation, total sleep
ings showed an acute response to sleep restriction but only small further deprivation, cognitive performance, subjective sleepiness, cumulative
increases on subsequent days, and did not significantly differentiate the 6 deficits, sleep debt, wake extension, core sleep, sleep need
h and 4 h conditions. Polysomnographic variables and δ power in the non- Citation: Van Dongen HPA, Maislin G, Mullington JM, Dinges DF. The
REM sleep EEG—a putative marker of sleep homeostasis—displayed an cumulative cost of additional wakefulness: dose-response effects on neu-
acute response to sleep restriction with negligible further changes across robehavioral functions and sleep physiology from chronic sleep restriction
the 14 restricted nights. Comparison of chronic sleep restriction to total and total sleep deprivation. SLEEP 2003;2:117-126.

INTRODUCTION tions that require high-level cognitive performance at critical times in

potentially lethal situations (e.g., health care, military operations, space
THE DEBATE OVER DAILY SLEEP NEED IN MODERN HUMANS flight).
IS LONGSTANDING. Although it is well established that sleep cannot The scientific debate over the consequences of chronic sleep restric-
be completely eliminated without waking neurobehavioral conse- tion has centered on theoretical concepts such as sleep debt,4,5 sleep ten-
quences,1,2 less is known about the effects of the relatively common dency,5 and core sleep versus optional sleep.6,7 These theoretical con-
practice of chronically reducing time for sleep during the work week3 or cepts have not resolved the issue,8 however, due to ambiguities in the
for even longer periods. Questions of whether there are increased wak- constructs and a lack of reliable scientific evidence on whether humans
ing performance deficits from chronic sleep reduction have substantial can maintain stable behavioral alertness and cognitive functions when
relevance to many human endeavors, especially those requiring activi- daily sleep is reduced across many days. Experimental reports on the
ties 24 h a day, 7 days a week (e.g., industrial production, transportation, effects of long-term chronic sleep restriction have bordered on the anec-
public safety). They have also become of increasing concern for opera- dotal, lacking adequate sample sizes and control groups. Most have
failed to ensure that subjects maintained the assigned sleep/wake sched-
ules; used infrequent, confounded and/or insensitive measures of sleep
Disclosure Statement and waking; and lacked sophisticated time series analyses (for reviews
No significant financial interest/other relationship to disclose. see refs. 9,10). More systematic studies evaluating the cumulative
effects of restricting sleep to between 4 h and 6 h per night for up to a
Submitted for publication January 2003 week have yielded conflicting results. Those that concluded that there
Accepted for publication January 2003 were few if any detrimental effects of chronic sleep restriction on day-
Address correspondence to: Hans P.A. Van Dongen, PhD, Division of Sleep time cognitive performance also failed to keep subjects in the laborato-
and Chronobiology, Department of Psychiatry, University of Pennsylvania ry under controlled conditions to ensure that they obtained only the sleep
School of Medicine, 1019 Blockley Hall, 423 Guardian Drive, Philadelphia, PA permitted and that they took no stimulants (e.g., caffeine).11–14 In con-
19104-6021, USA, Tel: 215-573-5866; Fax: 215-573-6410; trast, those that precisely controlled sleep time and dietary intake by
E-mail: [email protected]
SLEEP, Vol. 26, No. 2, 2003 117 The Cumulative Cost of Additional Wakefulness—Van Dongen et al
monitoring subjects throughout the study, found objective evidence of robehavioral assessments of cognitive performance, mood and symptom
cumulative neurobehavioral effects from sleep restriction to between 4 h complaints every 2 h. Between test bouts they were allowed to read,
and 6 h per night.10,15,16 No study of chronic sleep restriction between 4 watch movies, and interact with laboratory staff to help them stay awake,
h and 6 h per night has yet been published in which precise experimen- but no vigorous activities (e.g., exercise) were permitted. In the chronic
tal control of sleep periods and waking activities was maintained for sleep restriction conditions, normal daylight entered the laboratory but
more than a week. It has remained unclear at what rates neurobehavioral light exposure was relatively low (approximately less than 100 lux). In
deficits accumulate under chronic sleep restriction, and whether they can the 0 h sleep condition, the laboratory was maintained in less than 50 lux
reach impairment levels comparable to those found for total sleep depri- of light at all times. During scheduled sleep times, all lights were turned
vation. off (less than 1 lux) in every experimental condition. Subjects did not use
Mathematical models of sleepiness and performance based on the any caffeine, alcohol, tobacco, and/or medications in the 2 weeks before
two-process model of sleep regulation17 predict that compensatory the experiment, as verified by means of blood and urine screens and
homeostatic sleep responses would make it possible to chronically questionnaires, and during the experiment, as per the experimental pro-
reduce sleep duration without cumulative changes in neurobehavioral tocol.
functions.18 This prediction does not appear to be supported by results Volunteers were screened to ensure they had no medical, psychiatric,
from laboratory experiments involving a week of sleep limited to 5 h per or sleep-related disorders and were drug-free. This was determined by
night.10,15 Yet, it has been found that losing one night of sleep (i.e., total history, physical examination and psychological questionnaires, and by

Downloaded from https://academic.oup.com/sleep/article/26/2/117/2709164 by guest on 26 November 2023

sleep deprivation) leads to greater neurobehavioral deficits than when clinical blood and urine laboratory tests and toxicological screening.
the same total amount of sleep is lost across multiple nights of sleep Subjects reported working neither regular night nor rotating shift work
restriction.16 This suggests that either some adaptation to chronic sleep within the past 2 years. They also reported not having traveled across
restriction occurs, or that it is not the amount of cumulative sleep lost time zones in the 3 months before the experiments. The Institutional
that predicts waking neurobehavioral functions. Review Board of the University of Pennsylvania reviewed and approved
The possible occurrence of adaptation to sleep restriction needs to be the studies, and each subject gave written informed consent.
evaluated in chronic paradigms, in which quantitative estimates of the Subjects randomized to the 8 h sleep periods (n = 9; 2 females) were
temporal profile of waking responses across days can be reliably 24.1 ± 2.2 years old (mean ± s.d.); subjects randomized to the 6 h sleep
obtained. Comparison of the neurobehavioral effects of chronic sleep periods (n = 13; 3 females) were 30.1 ± 4.5 years old; and subjects ran-
restriction to those of total sleep deprivation should help to answer the domized to the 4 h sleep periods (n = 13; 1 female) were 27.7 ± 5.4 years
question of whether chronic sleep loss can induce waking neurobehav- old. Subjects in the 0 h sleep condition (n = 13; all males) were 27.3 ±
ioral changes and sleep physiological responses equivalent to those 4.6 years old. The subjects in the 8 h sleep period condition were signif-
found for total sleep loss. We report here the results of dose-response icantly younger than those in the 6 h sleep period condition (one-way
experiments on chronic sleep restriction for 14 consecutive days and ANOVA, F3,44 = 3.16, P = 0.034; Bonferroni post-hoc test, P = 0.022).
total sleep deprivation for 3 days. In these experiments, sleep and wake There were no other significant age differences among sleep restriction
timing and confounding factors were controlled by having subjects conditions.
remain in a laboratory (for a cumulative total of 830 days) with contin- A behavioral estimate of circadian phase position was obtained with a
uous behavioral and physiological monitoring, and random assignment morningness/eveningness questionnaire.19 Subjects in the 8 h sleep peri-
to experimental conditions. The experiments had five goals: (1) Deter- od condition had a morningness/eveningness score of 54.3 ± 8.1 (mean
mine whether sleep chronically limited to 4 h, 6 h or 8 h per night for 14 ± s.d.); those in the 6 h sleep period condition had a score of 52.5 ± 9.6;
consecutive nights results in cumulative changes in waking cognitive those in the 4 h sleep period condition had a score of 55.6 ± 13.4; and
performance functions, subjective sleepiness, and sleep physiology; (2) those in the 0 h sleep condition had a score of 52.8 ± 7.4. There were no
Determine if cumulative changes in cognitive performance, subjective significant differences among conditions in morningness/eveningness
sleepiness, and sleep physiology induced by chronic sleep restriction (one-way ANOVA, F3,44 = 0.28, P = 0.84).
reach levels obtained after 1, 2 and 3 consecutive nights of total sleep Each subject’s average sleep duration when living outside the labora-
deprivation; (3) Obtain quantitative estimates of inter-individual differ- tory in the 5 days prior to the experiment was assessed by means of
ences in response to chronic sleep restriction; (4) Identify factors under- actigraphy combined with complementary diary reports and time-
lying behavioral alertness changes across days for both chronic sleep stamped phone records for time to bed and time awake. Pre-study esti-
restriction and total sleep deprivation; and (5) Estimate nightly sleep mated sleep duration was 7.64 ± 0.65 h (mean ± s.d.) for subjects in the
needed by the subject population to prevent accumulation of waking 8 h sleep period condition; 7.92 ± 0.65 h for subjects in the 6 h sleep
cognitive deficits resulting from insufficient sleep. period condition; 7.84 ± 0.61 h for subjects in the 4 h sleep period con-
dition; and 7.70 ± 0.87 h for subjects in the 0 h sleep condition. There
METHODS were no differences among conditions in pre-study sleep duration (one-
way ANOVA, F3,44 = 0.38, P = 0.77).
Study Design and Participants
Healthy adults (n = 48) participated in a chronic sleep restriction Neurobehavioral Performance
experiment or in a total sleep deprivation experiment. They were physi-
Throughout all experimental conditions, subjects underwent neurobe-
ologically and behaviorally monitored in a laboratory in the General
havioral assessments every 2 h during scheduled wakefulness. Com-
Clinical Research Center (GCRC) of the Hospital of the University of
pared to subjects in the 8 h sleep period condition, subjects in the 6 h and
Pennsylvania, under controlled conditions and with strict schedules for
4 h sleep period conditions had one more neurobehavioral test bout per
time in bed (TIB). The sleep restriction experiment involved one adap-
day, scheduled at 00:50, when the subjects in the 8 h sleep period con-
tation day and two baseline days with 8 h sleep opportunities (TIB
dition were in bed. Subjects in the 0 h sleep condition were tested every
23:30–07:30), followed by randomization to 8 h, 6 h or 4 h periods for
2 h throughout the 88 h of sleep deprivation. Only test bouts in the peri-
nocturnal sleep (TIB ending at 07:30) for 14 days. The total sleep depri-
ods from 07:30 until 23:30 were included in the data analyses.
vation experiment consisted of one adaptation day and two baseline days
The neurobehavioral assessment battery included a psychomotor vig-
with 8 h sleep opportunities (TIB 23:30–07:30), after which subjects
ilance task20 to measure behavioral alertness. The psychomotor vigilance
were kept awake for 88 h. Both experiments concluded with 3 recovery
task (PVT) is a sustained-attention reaction time task with a random
days.
inter-stimulus interval of 2–10 s. Lapses (reaction times greater than 500
At all scheduled wake times, subjects were kept awake in the labora-
ms) were counted per 10 min test bout as a measure of performance
tory under continuous behavioral monitoring, and they underwent neu-

SLEEP, Vol. 26, No. 2, 2003 118 The Cumulative Cost of Additional Wakefulness—Van Dongen et al
impairment indicative of reduced behavioral alertness.21 The neurobe- Data Analyses
havioral assessment battery also included a computerized digit symbol
substitution task22 to measure working memory. This subject-paced task Traditional repeated-measures analysis of variance (ANOVA) is not
involves the matching of digits (0–9) to symbols (circle, triangle, etc.). concerned with the temporal arrangement (i.e., order and intervals) of
The number of correct responses in 1.5 min was counted to measure data points, making this technique poorly suited for distinguishing con-
working memory performance. A serial addition/subtraction task23 was sistent, cumulative changes from error variance in the data. Moreover,
included in the assessment battery to measure cognitive throughput. The repeated-measures ANOVA assumes that the response to the experimen-
serial addition/subtraction task is a subject-paced task requiring the com- tal conditions is homogeneous among subjects. Thus, inter-individual
pletion of 50 mental arithmetic trials. The average number of correct differences in the effects of sleep deprivation8 are ignored. Traditional
responses per min was used as a neurobehavioral assay of cognitive statistical regression techniques overcome the temporal arrangement
throughput performance. Further, the neurobehavioral assessment bat- problem, but do not readily handle inter-individual differences. In order
tery included the Stanford Sleepiness Scale.24 Subjects self-rated their to be able to also quantify inter-individual variability in the responses to
sleepiness on this 7-point scale at the beginning of each test bout. The chronic partial and total sleep loss, we applied mixed-effects regression
battery also included the Karolinska Sleepiness Scale.25 Subjects self- models27,28 for time series analysis. By incorporating random effects,
rated their sleepiness on this 9-point scale near the end of each test bout. these models allow proper separation of between-subjects (i.e., inter-
The results for the Karolinska Sleepiness Scale were very similar to individual) and within-subjects (i.e., temporal) variance in the data. As

Downloaded from https://academic.oup.com/sleep/article/26/2/117/2709164 by guest on 26 November 2023

those for the Stanford Sleepiness Scale. Therefore, results pertaining to an additional advantage, mixed-effects regression models do not require
subjective sleepiness are represented by the Stanford Sleepiness Scale missing data to be filled in. Furthermore, these models provide a frame-
throughout this paper. work for empirical Bayes estimation29 of subject-specific parameters for
For each of the neurobehavioral assays, daily averages over nine test temporal profiles across days of sleep restriction.
evaluations in the period from 07:30 until 23:30 were computed to assess Neurobehavioral variables. The following non-linear mixed-effects
the profiles of sleepiness and performance across days of sleep restric- model was fitted to the neurobehavioral performance data y, expressed
tion. All subjects in each condition had complete data for the psy- as difference from baseline, for each neurobehavioral assay:
chomotor vigilance task, the Stanford Sleepiness Scale, and the digit
symbol substitution task. For the serial addition/subtraction task, com- (1) yt ~ β . t θ
plete data were available for 8 of the 13 subjects in the 0 h sleep condi-
tion and for all subjects in the chronic sleep restriction conditions. There where t denotes days of sleep restriction. The parameter β is a normally
were no significant differences at baseline among the four conditions distributed random effect with condition-specific mean representing rate
(one-way ANOVA) for the psychomotor vigilance task (F3,44 = 2.00, P = of change, used to quantify the build-up of neurobehavioral performance
0.13), the Stanford Sleepiness Scale (F3,44 = 0.44, P = 0.73), and the seri- impairment across days of sleep restriction for each chronic sleep
al addition/subtraction task (F3,36 = 2.12, P = 0.12). Subjects in the 6 h restriction condition. The parameter θ represents curvature in the
response profile, which is necessary to accommodate any non-linearity
sleep period condition performed the digit symbol substitution task
in the metric of the outcome variable y. A single model was fitted for the
slightly less well at baseline than those randomized to the 8 h and 4 h
three sleep restriction conditions, and a separate model was fitted for the
sleep period conditions (F3,44 = 2.83, P = 0.049; Bonferroni post-hoc
total sleep deprivation condition. For statistical testing of differences
tests, P = 0.19 and P = 0.14, respectively).
among the 8 h, 6 h and 4 h sleep restriction conditions in the rate of
change across days, the condition-specific estimates for β were com-
Sleep Architecture
pared using an F test followed by pair-wise contrasts.
Polysomnography. Polysomnographic (PSG) recordings were made Sleep variables. To analyze changes in sleep variables over the 10
during the third baseline sleep period and during 10 of the 14 restricted restricted sleep periods during which polysomnographic recordings were
sleep periods (days 1, 2, 4, 5, 7, 8, 10, 11, 13 and 14). Only subjects with made, the following linear mixed-effects model was fitted to the data y
no more than 4 condition nights of missing data were included in PSG for each sleep variable:
analyses. Thus, we used the PSG data of 7 subjects in the 8 h sleep peri-
od condition, 8 subjects in the 6 h sleep period condition, and 9 subjects (2) yt ~ β . t + α
in the 4 h sleep period condition. For these subjects, data for 1 night of
PSG recording were missing on average. Sleep records were scored where t denotes days of sleep restriction. Parameter α is a normally dis-
using conventional criteria,26 with sleep onset conservatively defined by tributed random effect with condition-specific mean representing the
the occurrence of at least three consecutive 30 s epochs of stage 2-4 or acute change in the outcome variable y from baseline to the first restrict-
REM sleep. There were no significant differences (one-way ANOVA) ed sleep period, as resulting from the acute manipulation of time in bed.
among conditions in key variables describing baseline sleep: total sleep Parameter β is a normally distributed random effect with condition-spe-
time (F2,21 = 1.78, P = 0.19), stage 1 sleep (F2,21 = 0.65, P = 0.53), stage cific mean representing the rate of change across subsequent days; this
2 sleep (F2,21 = 1.05, P = 0.37), slow-wave sleep (F2,21 = 0.39, P = 0.68), parameter quantifies progressive changes in sleep physiology across
REM sleep (F2,21 = 1.74, P = 0.20), sleep latency (F2,21 = 1.95, P = 0.17), days of sleep restriction. A single model was fitted for the 8 h, 6 h and 4
slow-wave sleep (SWS) latency (F2,21 = 1.19, P = 0.33), REM sleep h sleep period conditions. For statistical testing of differences among
latency (F2,21 = 0.58, P = 0.57), and wakefulness after sleep onset (F2,21 these conditions in the acute response to sleep restriction, the condition-
specific estimates for α were compared using an F test. For statistical
= 1.33, P = 0.29).
testing of differences in the rate of change across days, the condition-
Non-REM EEG delta power. The C3 derivation of the EEG was
specific estimates for β were compared using an F test.
sampled at 128 Hz and subjected to spectral analysis in 2 s bins after
removal of artifacts. Power spectra were averaged across 30 s epochs.
For each night, power in the δ band (0.5–4.0 Hz) was totaled over all RESULTS
epochs of non-REM (stages 2–4) sleep, as a marker of sleep homeosta- Cognitive Performance
sis.17 For one subject in the 6 h sleep period condition, EEG signal qual-
ity was insufficient for reliable power spectral analysis. Thus, δ power Chronic sleep restriction conditions. Chronic restriction of the noc-
data were available for a total of 23 subjects in the 8 h, 6 h and 4 h sleep turnal sleep period to either 6 h or 4 h per day for 14 days resulted in sig-
period conditions. nificant cumulative performance deficits relative to the 8 h sleep period
condition (Figure 1). Significant differences among conditions in the

SLEEP, Vol. 26, No. 2, 2003 119 The Cumulative Cost of Additional Wakefulness—Van Dongen et al
rate of change across days (parameter β in equation (1)) were found for tion in the 5 days prior to the experiment and rate of increase in PVT
psychomotor vigilance task performance (F2,30 = 3.67, P = 0.037), digit lapses over the 14 days of sleep restriction (r32 = 0.29, P = 0.048). This
symbol substitution task performance (F2,30 = 5.33, P = 0.010), and seri- suggests that those subjects who habitually slept longest tended to be
al addition/subtraction task performance (F2,30 = 6.19, P = 0.006). more affected by the 14 days of imposed sleep restriction.
Subjects allowed an 8 h sleep period per night displayed only minor, Chronic sleep restriction versus total sleep deprivation. In the 4 h
non-significant increases in lapses of behavioral alertness over the 14 sleep period condition, lapses in behavioral alertness and reductions in
days. The statistically estimated mean of β in equation (1) for the 8 h working memory performance reached levels equivalent to those
sleep period condition was not significantly different from zero (t30 = observed after 2 nights without sleep (Figures 1A, 1C). Cognitive
0.77, P = 0.45) for the psychomotor vigilance task (Figure 1A). Subjects throughput performance after 14 days of sleep restriction was equivalent
in the 8 h sleep period condition demonstrated normal performance to that observed after 1 night without any sleep (Figure 1D). Subjects in
learning curves on the digit symbol substitution task (Figure 1C) and the the 6 h sleep period condition also reached levels of impairment equiv-
serial addition/subtraction task (Figure 1D). In contrast, subjects in the 4 alent to those observed after 1 night of total sleep loss for lapses in
h sleep period condition displayed escalating numbers of lapses in behavioral alertness and working memory performance (Figures 1A,
behavioral alertness and decreasing cognitive accuracy and speed across 1C).
the 14 days. The magnitude of changes in performance over days of For the 8 h, 6 h and 4 h sleep period conditions, curvature (parameter
sleep restriction in the 6 h sleep period condition was between that θ in equation (1)) was statistically estimated to be 0.78 ± 0.04 for psy-

Downloaded from https://academic.oup.com/sleep/article/26/2/117/2709164 by guest on 26 November 2023

observed in the 8 h and 4 h sleep period conditions. chomotor vigilance task performance, 0.59 ± 0.04 for digit symbol sub-
For PVT lapses, the subject-specific values of the rate of change β in stitution task performance, and 0.45 ± 0.04 for serial addition/subtraction
equation (1) were statistically determined with empirical Bayes estima- task performance (estimate ± s.e.). For the 0 h sleep condition, curvature
tion.29 To examine the relationship between pre-study sleep duration and was statistically estimated to be 0.57 ± 0.19 for psychomotor vigilance
the subject-specific rate of change in behavioral alertness over the 14 task performance, 0.74 ± 0.12 for digit symbol substitution task perfor-
days of sleep restriction, these subject-specific β values were correlated mance, and 0.67 ± 0.12 for serial addition/subtraction task performance
with habitual sleep time (estimated by actigraphy combined with diary (estimate ± s.e.). These values indicate that cognitive performance
reports and phone records during the 5 days prior to the experiment). impairment accumulated near-linearly over days for all four experimen-
Partial correlation, controlling for experimental condition, revealed a tal conditions, with slight differences between chronic sleep restriction
modest but significant positive relationship between average sleep dura- and total sleep deprivation and among cognitive performance tasks.

Subjective Sleepiness
Chronic restriction of the nocturnal sleep
period to either 6 h or 4 h per day for 14 days
resulted in a relatively small but significant
build-up of subjective sleepiness, as measured
with the Stanford Sleepiness Scale (SSS) rela-
tive to the 8 h sleep period condition (Figure
1B). Among the three sleep restriction condi-
tions, a significant difference was found in the
rate of change across days (F2,30 = 4.26, P =
0.024). Subjects in the 8 h sleep period condi-
tion displayed only minor, non-significant
increases in self-rated sleepiness: The statisti-
cally estimated condition-specific mean of β in
equation (1) was not significantly different
from zero (t30 = 1.32, P = 0.20). Similarity in
the rate of change across days was observed for
the 6 h and 4 h sleep period conditions (F1,30 =
0.10, P = 0.75).
The average response to 3 days of total sleep
deprivation spanned more than 2 units on the
Stanford Sleepiness Scale, while the response
to 14 days of sleep restricted to 6 h or 4 h per
day was only approximately 1 unit on this
scale. In contrast to cognitive performance
measures, the curvature (parameter θ in equa-
tion (1)) of the response to sleep loss over days
Figure 1—Neurobehavioral responses to varying doses of daily sleep. Four different neurobehavioral assays served to mea- was considerably different for chronic sleep
sure cognitive performance capability and subjective sleepiness. Each panel displays group averages for subjects in the 8 h restriction versus total sleep deprivation. The
( ), 6 h ( ), and 4 h ( ) chronic sleep period conditions across 14 days, and in the 0 h ( ) sleep condition across 3 days. curvature for subjective sleepiness as assessed
Subjects were tested every 2 h each day; data points represent the daily average (07:30–23:30) expressed relative to baseline
(BL). Panel A shows psychomotor vigilance task (PVT) performance lapses; panel B shows Stanford Sleepiness Scale (SSS) by the SSS was statistically estimated to be
self-ratings; panel C shows digit symbol substitution task (DSST) correct responses; and panel D shows serial addition/sub- 0.86 ± 0.14 for the 0 h sleep condition, and 0.24
traction task (SAST) correct responses per min. Upward corresponds to worse performance on the PVT and greater sleepi- ± 0.04 for the 8 h, 6 h and 4 h sleep period con-
ness on the SSS, and to better performance on the DSST and the SAST. The curves through the data points represent statis-
ditions (θ estimate ± s.e.). Thus, the profile of
tical non-linear model-based best-fitting profiles of the response to sleep deprivation (equation (1)) for subjects in each of
the four experimental conditions. The mean ± s.e. ranges of neurobehavioral functions for 1 and 2 days of 0 h sleep (total subjective sleepiness across days was near-lin-
sleep deprivation) are shown as light and dark gray bands, respectively, allowing comparison of the 3-day total sleep depri- ear for the 0 h sleep condition, while it was
vation condition and the 14-day chronic sleep restriction conditions. For the DSST and SAST, these gray bands are curved near-saturating for the 4 h and 6 h sleep period
parallel to the practice effect displayed by the subjects in the 8 h sleep period condition, to compensate for different amounts
of practice on these tasks. conditions (Figure 1B).

SLEEP, Vol. 26, No. 2, 2003 120 The Cumulative Cost of Additional Wakefulness—Van Dongen et al
 Results for the Karolinska Sleepiness Scale (KSS) were nearly identi- period conditions. Total δ power during the first 4 h of nocturnal recov-
cal to those found for the SSS. Among the three sleep restriction condi- ery sleep after 88 h of total sleep deprivation was 172% ± 11% of base-
tions, a significant difference was found for the KSS in the rate of line (mean ± s.e.). This was significantly more than the average for total
change across days (F2,30 = 7.76, P = 0.002). Subjects in the 8 h sleep δ power during sleep chronically restricted to 4 h per night (F1,18 = 5.40,
period condition displayed only minor, non-significant increases in self- P = 0.032). Thus, the δ power response to total sleep deprivation was
rated sleepiness (t30 = 0.56, P = 0.58). Again, similarity in the rate of greater than the δ power response to chronic sleep restriction (Figure
change across days was observed for the 6 h and 4 h sleep period condi- 2F). This observation demonstrates that the absence of δ power accu-
tions (F1,30 = 1.55, P = 0.22). The curvature for KSS responses to sleep mulation over days in the 6 h and 4 h sleep restriction conditions was not
loss was statistically estimated to be 0.81 ± 0.16 for the 0 h sleep condi- merely an artifact related to limited brain capacity for generating δ
tion, and 0.16 ± 0.03 for the 8 h, 6 h and 4 h sleep period conditions (θ power.
estimate ± s.e.). Thus, as with the SSS, the profile of subjective sleepi-
ness rated on the KSS was near-linear across days for the 0 h sleep con- Cumulative Sleep and Cumulative Sleep Loss
dition, while it was near-saturating for the 4 h and 6 h chronic sleep
To understand the nature of the relationship between daily sleep dose
restriction conditions.
and the build-up of neurobehavioral performance impairment, we con-
sidered the cumulative build-up of sleep and wake time over days in the
Sleep Physiology

Downloaded from https://academic.oup.com/sleep/article/26/2/117/2709164 by guest on 26 November 2023

four different experimental conditions (Figure 3). First, the accumula-
The temporal dynamics of polysomnographically recorded sleep peri- tion of polysomnographically-recorded total sleep time across days of
ods across days of sleep restriction are shown in Figure 2. Polysomno- sleep restriction was assessed for each condition. As expected, cumula-
graphic assessments of total sleep time, based on a conservative criteri- tive total sleep time increased near-linearly over days in the 8 h, 6 h and
on to determine sleep onset (at least three con-
secutive 30 s epochs of stage 2–4 or REM
sleep), showed that we were successful in dif-
ferentially reducing sleep in the three chronic
sleep restriction conditions (Figure 2A). There
were condition-specific acute responses
(parameter α in equation (2)) in almost all sleep
variables as a result of the acute change of time
in bed from baseline to the first restricted sleep
period. Significant differences among condi-
tions in response to the first night of sleep
restriction were found for total sleep time (F2,16
= 135.6, P < 0.001), stage 1 sleep (F2,16 = 12.3,
P < 0.001), stage 2 sleep (F2,16 = 29.2, P <
0.001), REM sleep (F2,16 = 12.9, P < 0.001),
and total δ power in the non-REM sleep EEG
(F2,15 = 5.29, P = 0.018), but not for SWS (F2,16
= 0.18, P = 0.84).
Across the 14 days of sleep restriction, only
marginal additional changes were observed in
sleep architecture (parameter β in equation (2)).
In the 8 h sleep period condition, stage 1 sleep
decreased (t16 = –2.49, P = 0.024) while stage 2
sleep (t16 = 2.18, P = 0.045) and REM sleep (t16
= 2.41, P = 0.029) increased, although the mag-
nitudes of these changes were small (Figures
2B, 2C, 2E). REM latency decreased by an
average of 2.2 min per day in the 4 h sleep peri-
od condition (t16 = –7.55, P < 0.001), but did
not change significantly over days in the 6 h
and 8 h sleep period conditions. Slow-wave
sleep (SWS) increased slightly but progressive-
ly over days (t16 = 3.98, P = 0.001) in the 6 h
sleep period condition; no significant systemat-
ic changes in SWS were found across days of
sleep restriction in the 8 h and 4 h sleep period
conditions (Figure 2D).
For total δ power in the non-REM sleep EEG
(Figure 2F), no significant progressive changes
across days of sleep restriction were observed
in any of the three chronic sleep restriction con- Figure 2—Sleep architecture responses to varying doses of daily sleep. The panels display sleep variables assessed from
ditions (|t15| < 1.61, P > 0.13). The level of δ polysomnographic recordings during baseline (BL) sleep and across 10 of the 14 restricted sleep periods. Panel A shows total
sleep time (in h); panels B–E show different stages of sleep (in h); and panel F displays total δ power in the non-REM sleep
power observed in the non-REM sleep EEG
EEG (as % of baseline). Daily means ± s.e. are shown for subjects in the 8 h ( ), 6 h ( ), and 4 h ( ) sleep period condi-
during recovery sleep after the 0 h sleep condi- tions. Panel F also shows the mean ( ) and the mean ± s.e. range (gray band) for total δ power in the first 4 h of recovery
tion was considerably higher than that observed sleep after 88 h of total sleep deprivation, which was 172% ± 11% of baseline (mean ± s.e.). This was significantly more
during sleep across days in the 6 h and 4 h sleep than the average for total δ power during sleep chronically restricted to 4 h per day (F1,18 = 5.40, P = 0.032).

SLEEP, Vol. 26, No. 2, 2003 121 The Cumulative Cost of Additional Wakefulness—Van Dongen et al
4 h sleep period conditions (Figure 3A). This is a direct result of the on performance on the PVT (Figure 1A), because this measure of behav-
nearly flat temporal profiles for polysomnographically assessed total ioral alertness displayed no learning curve and no significant cumulative
sleep time observed in these conditions (Figure 2A). impairment in the 8 h sleep period condition. PVT performance lapses
The accumulation of sleep loss across days of sleep restriction was showed evidence of decreased behavioral alertness as a sleep dose-
also calculated. For each subject, we compared total sleep time to habit- dependent, near-linear function of the number of days of sleep restriction
ual sleep time (estimated by actigraphy combined with diary reports and (Figure 1A). This could indicate that the development of neurobehav-
phone records during the 5 days prior to the experiment), and defined the ioral performance deficits over days of sleep restriction may be account-
difference as sleep loss. Figure 3B shows the accumulation of sleep loss ed for solely by cumulative sleep loss. It would then be predicted, how-
across days of sleep restriction. It is noteworthy that cumulative sleep ever, that the greatest performance impairment should be observed dur-
loss over 14 days in the 4 h sleep period condition was significantly ing days 7–14 in the 4 h sleep period condition, when cumulative sleep
greater than cumulative sleep loss over 3 days in the total sleep depriva- loss was greater in this condition than in the 0 h sleep condition (Figure
tion condition (t20 = 10.58, P < 0.001). 3B). Even after 14 days, however, performance deficits in the 4 h sleep
In order to compare cumulative sleep loss (Figure 3B) to cumulative period condition did not exceed those observed after 3 days in the total
neurobehavioral functions during chronic sleep restriction, we focused sleep deprivation condition (Figure 1A). Thus, cumulative sleep loss
cannot by itself explain the profiles of waking neurobehavioral perfor-
mance impairment for both chronic sleep restriction and total sleep

Downloaded from https://academic.oup.com/sleep/article/26/2/117/2709164 by guest on 26 November 2023

deprivation.
Among experimental conditions, there was a high degree of covaria-
tion between reductions of total sleep time and reductions of stage 1,
stage 2 and REM sleep (cf. Figures 1A, 1B, 1C, 1E), and a relative lack
of variation in SWS and non-REM δ power (Figures 1D, 1F). By the
same reasoning as for cumulative sleep loss, therefore, the cumulative
loss of any of these components of sleep cannot explain the observed
profiles of waking neurobehavioral performance impairment in all four
experimental conditions.

Cumulative Additional Wakefulness

Since the cumulative loss of sleep time did not parsimoniously
explain the rate of change in PVT performance for the chronic sleep
restriction conditions as well as the total sleep deprivation condition, we
focused on calculation of cumulative wake time extension (Figure 3C).
For every subject, the duration of each continuous period of wakefulness
during the experiment was compared to habitual wake time (which was
defined as 24 h minus habitual sleep time). Wake extension was defined
as the difference between the duration of each continuous wake period
and the duration of habitual wake time. Accordingly, cumulative wake
time extension was calculated as the sum of all consecutive hours of
wakefulness extending beyond the habitual duration of wakefulness
each subject was accustomed to at home. In the 8 h, 6 h and 4 h sleep
restriction conditions, this yielded the same results as for cumulative
sleep loss, because the definitions of cumulative wake extension and
cumulative sleep loss were arithmetically equivalent. For the 0 h sleep
condition (i.e., total sleep deprivation), however, each day without sleep
added 24 h to the cumulative wake extension. Thus, over 3 days with 0
h sleep, cumulative wake extension was equal to 72 h for each subject,
while cumulative sleep loss was only 23.1 ± 2.6 h (mean ± s.d.). This
illustrates that cumulative sleep loss and cumulative wake extension are
different constructs that can have different quantitative values, depend-
ing on the manner in which sleep loss occurs (cf. Figures 3B and 3C).
As described above, the two modes of sleep loss yielded similar max-
imum deficits for PVT performance (Figure 1A), but chronic sleep
restriction resulted in much greater cumulative sleep loss than did 3 days
of total sleep deprivation (Figure 3B). By focusing on cumulative wake
extension rather than cumulative sleep loss, we sought to determine if it
was the direct cost of additional wakefulness that could reconcile the
Figure 3—Cumulative build-up of total sleep time, sleep loss and wake time extension PVT performance profiles for these different modes of sleep loss.
across days of sleep restriction and total sleep loss. Panel A shows cumulative total sleep Indeed, the changes in behavioral alertness over days of sleep restriction
time (in h) derived from polysomnographic recordings (i.e., a cumulative representation of (Figure 1A) had a greater similarity with the temporal profiles of cumu-
Figure 2A). Panel B shows cumulative sleep loss relative to habitual sleep duration, that
is, all hours of sleep habitually obtained (as measured at home during the 5 days prior to
lative wakefulness extension (Figure 3C) than with the temporal profiles
the experiment), but not received due to sleep restriction in the experiment. Panel C shows of cumulative sleep loss (Figure 3B). For quantitative investigation of
cumulative wake extension relative to habitual wake duration, that is, all consecutive hours this discovery, a statistical model was developed to describe lapses in
of wakefulness in excess of the habitual duration of a wakefulness period. Daily means are
behavioral alertness (Figure 1A) as a function of cumulative additional
shown for subjects in the 8 h ( ), 6 h ( ), 4 h ( ) and 0 h ( ) sleep period conditions.
Panel B also shows the mean ± s.d. range (gray band) of cumulative sleep loss (relative to wakefulness across all four experimental conditions.
habitual sleep duration) after 3 days in the 0 h sleep condition, which was 23.1 ± 2.6 h We postulated that the build-up of neurobehavioral deficits was not
(mean ± s.d.). This was significantly less than the cumulative sleep loss after 14 days in caused by sleep loss directly, but rather by wakefulness in excess of a
the 4 h sleep period condition (t20 = 10.58, P < 0.001).
(subject-specific) critical wake period, that is, a maximum period during

SLEEP, Vol. 26, No. 2, 2003 122 The Cumulative Cost of Additional Wakefulness—Van Dongen et al
which stable neurobehavioral functioning could be maintained in our standard deviation reflects considerable inter-individual variability in
healthy young adult subjects. Cumulative excess wakefulness was the postulated critical wake duration ξ. The subject-specific values for ξ,
defined as the sum of all hours of wakefulness in excess of this critical statistically determined with empirical Bayes estimation,29 were similar
wake period. This construct was similar to cumulative wake extension, to the subject-specific habitual wake durations (derived from average
but did not rely on assessment of habitual wake duration for establishing sleep durations in the 5 days prior to the experiment). The difference
the critical wake period beyond which neurobehavioral impairment between habitual wake duration and critical wake duration ξ was 0.1 ±
would accumulate. Instead, the postulated critical wake duration was 0.5 h (mean ± s.e.), which was not significantly different from zero (t23
estimated from the available data. = 0.17, P = 0.86).
In the statistical model we developed, parameter ξ was defined as the Taking into account between-subjects variance in ξ and γ, the statisti-
(a priori unknown) critical wake duration (i.e., the postulated maximum cal model in equation (4) explained 83.0% of the variance in the PVT
period of stable waking neurobehavioral functioning). Cumulative data (Figure 1A). The value for curvature θ was 0.67 ± 0.05 (estimate ±
excess wakefulness Σ in the 8 h, 6 h and 4 h sleep period conditions was s.e.). Thus, across days of sleep restriction, the build-up of psychomotor
then obtained by: vigilance performance impairment in all four experimental conditions
was well approximated by a single near-linear function of cumulative
(3a) Σ t = (24 h – ξ) t – CTSTt excess wakefulness. This is illustrated in Figure 4, which shows PVT
performance lapses for all subjects as a function of cumulative sleep debt

Downloaded from https://academic.oup.com/sleep/article/26/2/117/2709164 by guest on 26 November 2023

where t denotes days of sleep restriction. The variable CTSTt represents (Figure 4A), and as a function of cumulative excess wakefulness (Figure
cumulative total sleep time (measured polysomnographically) as a func- 4B). When expressed as a function of cumulative sleep debt—the sum
tion of t. Over days of sleep restriction, cumulative excess wakefulness of all hours of sleep loss relative to the above-estimated subject-specific
was equivalent to cumulative sleep loss relative to a critical daily sleep daily sleep need—the neurobehavioral response to chronic sleep restric-
duration of 24 h – ξ. In the 0 h sleep condition, however, each day with- tion appeared to be fundamentally different than the neurobehavioral
out sleep (beyond the critical wake duration ξ) added 24 h to the cumu- response to total sleep deprivation (Figure 4A). When expressed as a
lative excess wakefulness. Thus, in the 0 h sleep condition, cumulative function of cumulative excess wakefulness, however, the neurobehav-
excess wakefulness Σ was given by: ioral responses to chronic sleep restriction and to total sleep deprivation
were well approximated by a single near-linear model (Figure 4B). This
(3b) Σ t = 24 h . t illustrates the monotonic, near-proportional relationship between cumu-
lative excess wakefulness and neurobehavioral performance impairment
where again t denotes days of sleep restriction. irrespective of daily sleep ration in these experiments.
A non-linear mixed-effects model27,28 was formulated to describe
lapses in behavioral alertness as a function of cumulative excess wake- DISCUSSION
fulness:
This study evaluated the waking neurobehavioral and sleep physio-
logical effects of chronic sleep restriction in healthy young adults using
(4) yt ~ γ (Σ t) θ
random assignment to dosages of sleep maintained over 14 consecutive
days under continuous behavioral, physiological and medical monitor-
where y denoted PVT performance lapses (expressed as difference from ing. Such continuous laboratory control of subjects’ sleep-wake times,
baseline), and θ was a parameter quantifying curvature (as in equation
(1)). The parameter γ was a nor-
mally distributed random effect
representing the rate of increase in
PVT lapses per hour of cumulative
excess wakefulness. The critical
wake duration ξ (i.e., the postulat-
ed maximum period of stable wak-
ing neurobehavioral functioning)
was incorporated in equation (4),
via equation (3a), as a second nor-
mally distributed random effect.
These random effects allowed
inter-individual differences in the
rate of increase in PVT lapses and
in the critical wake duration.8,30
The statistically estimated value
for ξ was 15.84 ± 0.73 h (estimate
± s.e.). For the subject population
in our experiments, limiting daily Figure 4—Behavioral alertness as a function of cumulative sleep debt versus cumulative excess wakefulness. The panels show behavioral
wakefulness to this critical wake alertness, as measured by psychomotor vigilance task performance lapses (relative to baseline), plotted as a function of cumulative sleep debt
(panel A) and as a function of cumulative excess wakefulness (panel B). Cumulative hours of sleep debt (panel A) were determined relative
duration would be expected to pre- to the statistically estimated subject-specific daily sleep need of 8.16 ± 0.73 h (mean ± s.e.). For an average individual, this is the cumula-
vent the build-up of neurobehav- tive time of sleep reduction below 8.16 h per day. Cumulative hours of excess wakefulness (panel B) were calculated as those hours in which
ioral deficits. Accordingly, daily wakefulness exceeded the statistically estimated subject-specific critical wake period of 15.84 ± 0.73 h (mean ± s.e.). For an average indi-
vidual, this is the cumulative time of wakefulness in excess of 15.84 h per wake period. Each point represents the average for a day (across
sleep need to prevent cumulative
14 days of sleep restriction and 3 days of total sleep deprivation) for subjects randomized to the 8 h ( ), 6 h ( ), 4 h ( ) and 0 h ( ) sleep
neurobehavioral deficits in these period conditions. Panel A shows that from the perspective of sleep debt, the response to the 0 h sleep condition (total sleep deprivation) is
subjects would appear to be 24 h – fundamentally different from the response to the 8 h, 6 h and 4 h sleep period conditions (chronic sleep restriction). However, panel B reveals
ξ = 8.16 ± 0.73 h (estimate ± s.e.). that these two experimental paradigms can be described by a single near-linear model when focusing on the cumulative time of excess wake-
fulness. This difference in perspective affects the position of the data points for the 0 h sleep condition: Subjects who received 0 h sleep each
The statistically estimated standard day built up a statistically estimated average sleep debt of only 8.16 h of sleep per day, but extended their consecutive wakefulness by 24 h
deviation over subjects for ξ was per day. Thus, panel B illustrates the monotonic, near-proportional relationship between cumulative excess wakefulness and neurobehavioral
3.58 ± 1.19 h (estimate ± s.e.); this performance impairment irrespective of daily sleep ration.

SLEEP, Vol. 26, No. 2, 2003 123 The Cumulative Cost of Additional Wakefulness—Van Dongen et al
activities and assessments sets this experiment apart from previous pub- development of cumulative neurobehavioral performance deficits when
lished reports on the effects of prolonged chronic sleep restriction (i.e., young adults were chronically restricted to 4 h and 6 h nocturnal sleep.
more than a week). Contrary to earlier, uncontrolled studies of prolonged It is likely that such deficits would also be found in younger subjects
sleep restriction, this experiment yielded convergent findings of sleep (e.g., adolescents32) and older individuals. We do not know if the same
dose-response effects on all three cognitive performance functions. degree of cumulative impairment occurs across the continuum of habit-
Sleep periods chronically limited to 4 h and 6 h per night progressively ually short to habitually long sleepers, among those sleeping at different
eroded the effectiveness of psychomotor vigilance performance, work- circadian times (e.g., night shift workers), and in females compared to
ing memory performance and cognitive throughout performance, pro- males. Additional investigations are underway to address some of these
viding convergent evidence for the adverse effects of chronic sleep issues.
restriction on cognitive functions. These results confirm and substantial- Chronic restriction of sleep to 4 h and 6 h initially elevated subjective
ly extend those obtained in earlier laboratory-controlled studies of sleepiness ratings on both the Stanford Sleepiness Scale and the Karolin-
chronic sleep restriction between 4 h and 6 h per night for up to 7 ska Sleepiness Scale, but as sleep restriction continued, there were only
days.10,15 Claims that humans adapt to chronic sleep restriction within a minor further increases in these ratings (Figure 1B). In fact, unlike PVT
few days, on the other hand, are not supported by the present findings. and DSST performance functions (Figures 1A, 1C), sleepiness ratings
Since chronic restriction of sleep between 4 h and 6 h per night for 14 never reached levels equivalent to those found after 2 nights of total
days produced cognitive performance deficits comparable to those found sleep deprivation. Surprisingly, by the end of the 14 days of sleep restric-

Downloaded from https://academic.oup.com/sleep/article/26/2/117/2709164 by guest on 26 November 2023

under conditions of 1 to 2 days of total sleep deprivation (Figure 1), it tion, when performance was at its worst levels, subjects in the 4 h and 6
appears that even relatively moderate sleep restriction—if sustained h sleep period conditions reported feeling only slightly sleepy. There-
night after night—can seriously impair waking neurobehavioral func- fore, unlike performance measures, sleepiness ratings appeared to show
tions in healthy young adults. This conclusion is reinforced by four addi- adaptation to chronic partial sleep deprivation. In addition, there were no
tional observations: (1) The sensitivity of waking cognitive performance significant differences in sleepiness scores between the 4 h and 6 h sleep
functions to chronic sleep restriction was found both between conditions period conditions. It is unlikely that this was the result of a ceiling effect
(i.e., impact of 4 h versus 6 h versus 8 h nightly sleep periods) and with- or other metric-related artifact, because subjects in the total sleep depri-
in subjects (i.e., impact of increasing days in condition); (2) The effects vation condition reported considerably greater levels of sleepiness, and
of chronic sleep restriction were not limited to a few times of day, but did not show evidence of adaptation.
rather they were evident in performance throughout the waking day These findings for subjective sleepiness suggest that once sleep
(each point in the cognitive performance data shown in Figure 1 reflects restriction is chronic, subjects either cannot reliably introspect with
the average daily performance on each task for assessments taken during regard to their actual sleepiness levels, or as long as they are receiving
the 16 h between 07:30 and 23:30); (3) Mixed-effects regression models at least approximately 4 h of sleep nightly they do not experience a sense
were used to quantify the waking neurobehavioral responses to chronic of sleepiness anywhere near the levels found for total sleep deprivation.
sleep restriction while taking into account inter-individual variability in Regardless of the explanation, the lack of reports of intense feelings of
these responses (ensuring that all statistical analyses and the curves in sleepiness during chronic sleep restriction may explain why sleep
Figure 1 are representative of the subject population’s responses to con- restriction is widely practiced—people have the subjective impression
dition); and (4) The cumulative cognitive deficits are not likely to be due they have adapted to it because they do not feel particularly sleepy. More
to boredom, monotony, non-compliance, or any other non-sleep-related research will be needed to identify the factors that shape subjects’ per-
hypothetical construct, since subjects in the 8 h control condition ceptions of their sleepiness during chronic sleep restriction.
showed no significant progressive deficits (and displayed continued Measures of sleep physiology were less responsive to chronic sleep
learning on the working memory and cognitive throughput tasks) despite restriction than were waking neurobehavioral functions. The primary
being exposed to the same degree of laboratory control, experimental effects on sleep architecture were immediate, overall reductions in the
procedures, and repeated testing. amounts of stages 1, 2 and REM sleep (Figures 2B, 2C, 2E). SWS and δ
We conclude that the effects of sleep chronically limited to 4 h and 6 power in the non-REM sleep EEG were conserved among conditions
h per night on cognitive performance appear to reflect progressive neu- (Figures 2D, 2F). A recent sleep restriction study of within-subjects,
rocognitive dysfunction in systems underlying sustained attention and non-counterbalanced design reported that when analyzing sleep over a
working memory. They implicate an as yet unknown neurobiological standardized period common to two different study conditions, SWS and
process that is sensitive to sleep duration over consecutive days. There δ power (or slow-wave activity; SWA) in restricted sleep (4 h TIB) were
has been little basic research published on possible mechanisms for the greater than SWS and SWA in the first part of extended sleep (12 h
effects of chronic sleep loss on waking cognitive performance, but TIB).33 In the present study, there were no statistically significant differ-
recently it has been proposed that an increase in A1 adenosine receptors ences for SWS and SWA among the three sleep restriction conditions
in the basal forebrain in response to initial sleep loss may sensitize the when analyzing sleep architecture over the first 4 h of each sleep peri-
brain to subsequent sleep loss.31 Whatever the mechanism, it is parsimo- od.34
nious to suggest that the same process underlies the progressive perfor- The conservation of SWS and SWA among all three sleep restriction
mance degradation we observed on all three cognitive tasks (Figures 1A, conditions relative to the marked development of cognitive performance
1C, 1D). Elsewhere, we have suggested that this process may be charac- deficits in the 4 h and 6 h sleep restriction conditions is inconsistent with
terized as wake state instability.21 Those seeking to identify the neurobi- the “core sleep” hypothesis,6,7 which asserts that “core” or “obligatory”
ological basis of cumulative cognitive deficits engendered by chronic sleep occupies the first part of the night and serves to “repair the effects
sleep restriction will have to explain how the brain can be increasingly of waking wear and tear on the cerebrum” (ref. 7, p. 57). In the core
affected for a period of at least 14 consecutive days. sleep hypothesis, all sleep obtained beyond this core sleep duration is
The participants in our experiments were healthy younger adults, considered to be “optional” or “facultative” sleep, which “fills the
21–38 years of age. This is the age range commonly found in occupa- tedious hours of darkness until sunrise” (ref. 7, p. 57). According to this
tions associated with chronic sleep restriction (e.g., shift workers, mili- hypothesis, only core sleep—especially that dominated by SWS and
tary personnel, medical and surgical residents). Many factors—such as SWA—is required for adequate daytime functioning to be maintained.
the added responsibilities of rearing young children, or the desire to Considering that SWS and SWA were conserved among sleep restriction
obtain additional income—contribute to lifestyles that result in chronic conditions in the present study, the finding that cumulative cognitive
sleep restriction. Younger adults are often assumed to be better able to impairment developed in cerebral functions at 4 h and 6 h time for sleep
cope with the demands of prolonged wakefulness and lifestyles that lead per night indicates that the current threshold of 6 h for core sleep dura-
to chronic sleep loss. It is of concern then that our results revealed the tion (see ref. 8) cannot be correct. If 6 h sleep per day were the maxi-

SLEEP, Vol. 26, No. 2, 2003 124 The Cumulative Cost of Additional Wakefulness—Van Dongen et al
mum duration of sleep required to maintain normal cerebral functions, rather than cumulative loss of sleep (i.e., sleep debt), is the primary
cumulative cognitive performance deficits should not have developed in cause of progressively reduced behavioral alertness both across days of
that condition. Thus, the results from the present study do not support a chronic sleep restriction and across days of total sleep deprivation (cf.
functional distinction between “core” and “optional” sleep. Figures 4A and 4B). Subjects in all four experimental conditions
Regardless of which physiological measure of sleep is theorized to appeared to experience the same cumulative “cost” (i.e., increase in laps-
reflect homeostatic sleep drive, there was scant evidence that homeo- es of behavioral alertness) for each consecutive hour they extended their
static sleep drive accumulated across days of sleep restriction. Sleep wake periods (near-linear relationship displayed in Figure 4B).
architecture variables did not show any substantial, consistent cumula- With mixed-effects regression modeling27 of the psychomotor vigi-
tive changes across the 14 days of sleep restriction in the 4 h and 6 h lance performance data, the critical wake period beyond which lapsing
sleep period conditions. There appeared to be no significant cumulative would be expected to increase was statistically estimated to be 15.84 ±
pressure for non-REM sleep, and only minor cumulative changes in 0.73 h (mean ± s.e.). For the average healthy young adult in the experi-
REM pressure, as sleep restriction continued. Power spectral analysis ments, limiting daily wakefulness to this level would be expected to pre-
showed that δ power in the non-REM sleep EEG, which is a putative vent the build-up of neurobehavioral deficits over days. Accordingly, per
marker of homeostatic sleep drive,17 increased only modestly over the 24 h day, the average value for human sleep need to prevent cumulative
first few days of sleep restriction, and thereafter displayed negligible fur- neurobehavioral deficits would appear to be 8.16 h. Although we found
ther increases (Figure 2F). This was not due to limited brain capacity for no evidence that subjects had any significant neurobehavioral impair-

Downloaded from https://academic.oup.com/sleep/article/26/2/117/2709164 by guest on 26 November 2023

generating δ power, since the δ power response to total sleep deprivation ment at the beginning of sleep restriction, it is possible that the 8 h base-
was much greater (Figure 2F). line sleep periods were not sufficiently long to completely prevent the
The modest increase observed in δ power during sleep restriction is build-up of neurobehavioral impairment. Given that all subjects under-
consistent with previous work,35 and with mathematical simulations we went the same baseline procedures regardless of experimental condition,
ran using the two-process model of sleep regulation17 which predicted and since neurobehavioral performance measures were expressed rela-
that the average level of homeostatic sleep drive across days of sleep tive to baseline (Figure 1), any neurobehavioral deficits present at the
restriction would show an acute sleep dose-dependent change followed beginning of sleep deprivation should not have affected the statistical
by stabilization within 3 days. Even though the sleep physiological evaluation of the results reported here. Yet, it is noteworthy that future
responses to both chronic sleep restriction and total sleep deprivation studies of sleep or sleep deprivation could benefit from extending base-
were consistent with contemporary model-based theory for the regula- line sleep periods to more than 8 h time in bed per day.
tion of sleep,17 it is remarkable that the changes in cognitive perfor- We found that the statistically estimated duration of the critical wak-
mance functions over days of sleep restriction (Figures 1A, 1C, 1D) ing period varied greatly among the 48 young adults in our experiments.
were not matched by progressive changes in sleep architecture over days Large inter-individual variability has also been reported for the duration
(Figure 2F). This means that the common implicit assumption that of nocturnal sleep in adult populations.3,37 We observed a positive corre-
momentaneous homeostatic sleep drive is indicative of concurrent wak- lation between average sleep duration in the 5 days prior to the experi-
ing performance capability must be put to rest (see also ref. 36). It ment and the rate of change in behavioral alertness during the 14 days of
appears that the concept of homeostatic sleep drive cannot account for sleep restriction, suggesting that those subjects who habitually slept
the cumulative neurobehavioral performance changes observed across longest tended to be more affected by the 14 days of sleep restriction.
consecutive days of sleep restriction. This correlation was modest, however, suggesting that other factors may
We evaluated whether the neurobehavioral effects observed with also contribute to inter-individual differences in the neurobehavioral
chronic sleep restriction could be explained by cumulative total sleep responses to chronic sleep restriction.8,38
loss, which we refer to here as the “sleep debt” hypothesis.5 Psychomo- The physiologic expression of sleep in humans appears to have mul-
tor vigilance lapses were used as the primary neurobehavioral metric for tiple functions, ranging from metabolic39 to neurocognitive.40 Chronic
this evaluation because PVT lapses displayed no learning curve and no loss of physiological sleep has been documented to adversely affect
significant cumulative impairment in the 8 h sleep period condition. endocrine function,39 cardiovascular events,41 and other health-related
PVT lapses showed evidence of decreased behavioral alertness as a sleep outcomes.37 Yet, it has remained unclear why homo sapiens should
dose-dependent, near-linear function of the number of days in the 6 h, 4 invest so much time in sleep—roughly one-third of every day in adults—
h and 0 h sleep period conditions (Figure 1A). This would seem to sug- to fulfill the various physiological, cognitive and health-related func-
gest that the development of neurobehavioral performance deficits over tions that sleep may have. It is well established that the temporal regula-
days of sleep restriction could be accounted for solely by sleep debt (i.e., tion of sleep is governed by an interplay of homeostatic and circadian
the cumulative loss of time for sleep) regardless of homeostatic sleep processes.17,42–44 The present data suggest that this temporal regulation
drive. On the other hand, the sleep debt hypothesis would predict that the of sleep serves to protect human neurobehavioral functions from degra-
highest level of psychomotor vigilance performance impairment should dation due to excessive wakefulness within and between circadian
be observed during days 7–14 in the 4 h sleep period condition, when the cycles.
cumulative reduction of sleep time was greater in this condition than in
any of the other experimental conditions (Figure 3B). Even after 14 ACKNOWLEDGMENTS
days, however, the performance deficits in the 4 h sleep period condition
did not exceed those observed after 3 days in the 0 h sleep period con- We thank the subject volunteers; the research technicians and student
dition (Figure 1A). Thus, the cumulative reduction of time for sleep can- behavioral monitors; physician of record Martin Szuba, MD; senior
not by itself explain the profiles of waking performance impairment in research staff Michele Carlin, John Powell IV, Christine Dinges, and
all four experimental conditions (Figure 4A). Claire Fox; the staff of the General Clinical Research Center (GCRC) of
In introducing a new hypothesis, we postulated that the build-up of the Hospital of the University of Pennsylvania; Gregory Belenky, MD
neurobehavioral deficits was not caused by reduction of sleep time per and David Thorne, PhD for the serial addition/subtraction task; and
se, but rather by excessive wakefulness beyond a maximum period dur- Willard Larkin, PhD, Naomi Rogers, PhD, Alicia Levin and two anony-
ing which stable neurobehavioral functioning could be maintained. mous reviewers for helpful suggestions. This work was supported by
When we defined “excess wakefulness” as all waking time beyond this NIH awards R01-NR04281, R01-HL70154 and M01-RR00040, Air
hypothetical critical period, behavioral alertness measured by perfor- Force Office of Scientific Research grants F49620-95-1-0388 and
mance lapses was a near-linear function of excess wakefulness across F49620-00-1-0266, and NASA Cooperative Agreement NCC 9-58 with
days of sleep restriction for all four experimental conditions (Figure 4B). the National Space Biomedical Research Institute.
This suggests that cumulative wake extension (i.e., excess wakefulness),

SLEEP, Vol. 26, No. 2, 2003 125 The Cumulative Cost of Additional Wakefulness—Van Dongen et al
REFERENCES 39. Spiegel K, Leproult R, Van Cauter E. Impact of sleep debt on metabolic and endocrine
function. Lancet 1999;354:1435–1439.
1. Dinges DF, Kribbs NB. Performing while sleepy: Effects of experimentally-induced 40. Stickgold R, Hobson JA, Fosse R, Fosse M. Sleep, learning, and dreams: Off-line mem-
sleepiness. In: Monk TH, ed. Sleep, sleepiness and performance. Chichester: John Wiley ory reprocessing. Science 2001;294:1052–1057.
& Sons, 1991:97–128. 41. Ayas NT, White DP, Manson JE, Stampfer MJ, Speizer FE, Malhotra A, Hu FB. A
2. Bonnet MH. Sleep deprivation. In: Kryger MH, Roth T, Dement WC, eds. Principles and prospective study of sleep duration and coronary heart disease in women. Arch Intern
practice of sleep medicine (3rd edn). Philadelphia: WB Saunders, 2000:53–71. Med 2003;163:205–209.
3. National Sleep Foundation. 2002 Sleep in America Poll. Washington DC: National Sleep 42. Czeisler CA, Duffy JF, Shanahan TL, et al. Stability, precision, and near-24-hour period
Foundation, 2002. of the human circadian pacemaker. Science 1999;284:2177–2181.
4. Kleitman N. Sleep and wakefulness (2nd edn). Chicago: University of Chicago Press, 43. Dijk DJ, Edgar DM. Circadian and homeostatic control of wakefulness and sleep. In:
1963. Turek FW, Zee PC, eds. Regulation of sleep and circadian rhythms. New York: Dekker,
5. Dement WC, Vaughan C. The promise of sleep. New York: Dell Publishing, 1999. 1999:114–147.
6. Horne JA. Sleep function, with particular reference to sleep deprivation. Ann Clin Res 44. Van Dongen HPA, Dinges DF. Circadian rhythms in fatigue, alertness, and performance.
1985;17:199–208. In: Kryger MH, Roth T, Dement WC, eds. Principles and practice of sleep medicine (3rd
7. Horne JA. Why we sleep. Oxford: Oxford University Press, 1988. edn). Philadelphia: WB Saunders, 2000:391-399.
8. Van Dongen HPA, Rogers NL, Dinges DF. Sleep debt: Theoretical and empirical issues.
Sleep Biol Rhythms 2003;1:4–12.
9. Carskadon MA, Roth T. Sleep restriction. In: Monk TH, ed. Sleep, sleepiness and per-
formance. Chichester: John Wiley & Sons, 1991:155–167.
10. Dinges DF, Pack F, Williams K, et al. Cumulative sleepiness, mood disturbance, and

Downloaded from https://academic.oup.com/sleep/article/26/2/117/2709164 by guest on 26 November 2023

psychomotor vigilance performance decrements during a week of sleep restricted to 4–5
hours per night. Sleep 1997;20:267–277.
11. Webb WB, Agnew Jr HW. The effects of a chronic limitation of sleep length. Psy-
chophysiology 1974;11:265–274.
12. Friedmann J, Globus G, Huntley A, Mullaney D, Naitoh P, Johnson L. Performance and
mood during and after gradual sleep reduction. Psychophysiology 1977;14:245–250.
13. Horne JA, Wilkinson S. Chronic sleep reduction: Daytime vigilance performance and
EEG measures of sleepiness, with particular reference to “practice” effects. Psy-
chophysiology 1985;22:69–78.
14. Blagrove M, Alexander C, Horne JA. The effects of chronic sleep reduction on the per-
formance of cognitive tasks sensitive to sleep deprivation. Appl Cogn Psychol
1995;9:21–40.
15. Carskadon MA, Dement WC. Cumulative effects of sleep restriction on daytime sleepi-
ness. Psychophysiology 1981;18:107–113.
16. Drake CL, Roehrs TA, Burduvali E, Bonahoom A, Rosekind M, Roth T. Effects of rapid
versus slow accumulation of eight hours of sleep loss. Psychophysiology
2001;38:979–987.
17. Borbély AA, Achermann P. Sleep homeostasis and models of sleep regulation. J Biol
Rhythms 1999;14:557–568.
18. Van Dongen HPA. Comparison of model predictions to experimental data. Aviat Space
Environ Med; submitted.
19. Horne JA, Östberg O. A self-assessment questionnaire to determine morningness-
eveningness in human circadian rhythms. Int J Chronobiol 1976;4:97–110.
20. Dinges DF, Powell JW. Microcomputer analyses of performance on a portable, simple
visual RT task during sustained operations. Behav Res Meth Instr Comp
1985;17:652–655.
21. Doran SM, Van Dongen HPA, Dinges DF. Sustained attention performance during sleep
deprivation: Evidence of state instability. Arch Ital Biol 2001;139:253–267.
22. Wechsler D. Wechsler Adult Intelligence Scale–Revised. New York: Psychological
Corp., 1981.
23. Thorne DR, Genser SG, Sing HC, Hegge FW. The Walter Reed performance assessment
battery. Neurobehav Toxicol Teratol 1985;7:415–418.
24. Hoddes E, Zarcone V, Smythe H, Phillips R, Dement WC. Quantification of sleepiness:
A new approach. Psychophysiology 1973;10:431–436.
25. Åkerstedt T, Gillberg M. Subjective and objective sleepiness in the active individual.
Intern J Neurosci 1990;52:29–37.
26. Rechtschaffen A, Kales AA. Manual of standardized terminology, techniques and scor-
ing system for sleep stages of human subjects. Los Angeles: UCLA Brain Information
Service, 1968.
27. Vonesh EF, Chinchilli VM. Linear and nonlinear models for the analysis of repeated
measurements. New York: Marcel Dekker, 1996.
28. Wolfinger RD. Fitting nonlinear mixed models with the new NLMIXED procedure.
Cary: SAS Institute Inc., 2000.
29. Efron B. Empirical Bayes methods for combining likelihoods. J Am Stat Assoc
1996;91:538–550.
30. Van Dongen HPA, Dinges DF. Modeling the effects of sleep debt: On the relevance of
inter-individual differences. Sleep Res Soc Bull 2001;7:69–72.
31. Basheer R, Porkka-Heiskanen T, Strecker RE, Thakkar MM, McCarley RW. Adenosine
as a biological signal mediating sleepiness following prolonged wakefulness. Biol Sig-
nals Recept 2000;9:319–327.
32. Fallone GP, Seifer R, Acebo C, Carskadon MA. Prolonged sleep restriction in 11- and
12-year-old children: Effects on behavior, sleepiness, and mood. Sleep 2000;23S2:A28.
33. Spiegel K, Leproult R, Colecchia EF, L’Hermite-Balériaux M, Nie Z, Copinschi G, Van
Cauter E. Adaptation of the 24-h growth hormone profile to a state of sleep debt. Am J
Physiol 2000;279:R874–R883.
34. Van Dongen HPA, Maislin G, Dinges DF. Chronic partial sleep deprivation: Neurobe-
havioral response and sleep architecture. Sleep Res Online 1999;2S:735.
35. Brunner DP, Dijk DJ, Borbély AA. Repeated partial sleep deprivation progressively
changes the EEG during sleep and wakefulness. Sleep 1993;16:100–113.
36. De Brunier AB, Dinges DF, Van Dongen HPA. Dynamics of the waking EEG during
chronic partial sleep deprivation. Sleep-Wake Res Netherlands 2002;13:27–29.
37. Kripke DF, Garfinkel L, Wingard DL, Klauber MR, Marler MR. Mortality associated
with sleep duration and insomnia. Arch Gen Psychiatry 2002;59:131–136.
38. Aeschbach D, Cajochen C, Landolt H, Borbély AA. Homeostatic sleep regulation in
habitual short sleepers and long sleepers. Am J Physiol 1996;270:R41–R53.

SLEEP, Vol. 26, No. 2, 2003 126 The Cumulative Cost of Additional Wakefulness—Van Dongen et al