We tested the hypothesis that circadian adaptation to night work is best achieved by combining bright light during the night shift and scheduled sleep in darkness. Fifty-four subjects participated in a shift work simulation of 4 day and 3 night shifts followed by a 38-h constant routine (CR). Subjects received 2,500 lux (Bright Light) or 150 lux (Room Light) during night shifts and were scheduled to sleep (at home in darkened bedrooms) from 0800 to 1600 (Fixed Sleep) or ad libitum (Free Sleep). Dim light melatonin onset (DLMO) was measured before and after the night shifts. Both Fixed Sleepand Bright Light conditions significantly phase delayed DLMO. Treatments combined additively, with light leading to larger phase shifts. Free Sleep subjects who spontaneously adopted consistent sleep schedules adapted better than those who did not. Neither properly timed bright light nor fixed sleep schedules were consistently sufficient to shift the melatonin rhythm completely into the sleep episode. Scheduling of sleep/darkness should play a major role in prescriptions for overcoming shift work-related phase misalignment.
- shift work
modern society requires that many people reverse their inherent diurnal activity pattern to ensure 24-h availability of services. There are roughly 8 million workers in the United States who regularly work at night (40). Many are employed in occupations in which peak functioning is critical (e.g., nurses and physicians, airline pilots, and operators of nuclear power plants and heavy machinery). However, night shift work exacts a substantial cost in terms of degraded health and disrupted performance. Night shift workers experience both sleep loss and misalignment of circadian phase. They suffer from greater risk of gastric and duodenal ulcers (51) and cardiovascular disease (7, 12). Night workers are particularly prone to vehicular accidents (2, 11, 32, 36, 43). Their decreased alertness, performance, and vigilance (16) are likely to blame for a substantially higher rate of industrial accidents and quality-control errors on the job (35), injuries (33, 47), and a general decline in work rate (46).
It is now well established that sleep, alertness, and cognitive functioning are determined by the interaction of two processes, the endogenous circadian pacemaker and a sleep homeostat (20,31). The circadian pacemaker, located in the suprachiasmatic nucleus of the hypothalamus, generates an endogenous, near-24-h rhythm (15) that regulates subjective alertness, sleep propensity, and a wide variety of cognitive functions (24,26), as well as core body temperature and melatonin secretion. The pacemaker is known to be highly sensitive to light, which is now considered to be the primary synchronizer of the circadian system (14).
The homeostat mediates a continual decline in performance and corresponding increase in sleepiness with time elapsed since awakening. Extended sleep deprivation experiments show a steady decay in alertness and cognitive functioning superimposed over the daily rhythm produced by the pacemaker (34).
Night workers, who are attempting to invert their normal sleep/wake schedule, suffer because the timing of their sleep/wake and work schedule remains permanently out of phase with the timing of environmental light, which probably accounts for the fact that the endogenous circadian rhythms of most permanent night shift workers fail to adapt completely (45). Ingestion of meals at an inappropriate circadian phase may be an important contributor to the gastrointestinal problems that shift workers suffer (51). Circadian misalignment leads to a substantial loss of sleep efficiency during the (daytime) sleep episode (1, 24), in addition to environmental obstacles to sleep (e.g., noise, light). Finally, night shift workers typically begin their workday 5–10 h after awakening, leaving them with more accumulated homeostatic sleep drive at the beginning of their night work shift compared with day workers (1).
Several strategies could mitigate the debilitating effects of shift work, including improved schedule design (18, 38), pharmacological agents to improve alertness on the job (3), and changes in diet, sleep scheduling, or the work environment itself (46). Appropriately timed bright light is effective in resetting the circadian rhythms of subjects undergoing simulated night work protocols (16, 21, 22, 29, 30, 48). For example, Czeisler et al. (16) demonstrated that physiological maladaptation to night work could be effectively treated by a regimen of exposure to bright light during night work and darkness during day sleep. This was accompanied by a significant improvement in alertness and performance during the night shift hours. However, it is not known how much of this effect was due to the bright light during work and how much was due to the scheduled daytime sleep in darkness (50).
The role of scheduled sleep in darkness as a circadian synchronizer in humans is unclear. Absence of light during the sleep episode may function as a photic synchronizer by changing the timing and distribution of light; the sleep itself may function as a behavioral, nonphotic synchronizer (37, 41); or the darkness may act as a synchronizer in its own right (a “dark pulse”) (10,49). Whatever the underlying mechanism, it is important to know how bright light treatments are affected by sleep schedule when designing a treatment regimen to alleviate circadian maladaptation to night work. It is possible that bright light may be sufficiently powerful to overcome all other synchronizers, and that shift workers' sleep habits are largely irrelevant in determining the effectiveness of bright light intervention. Alternatively, the powerful phase-shifting effect observed by Czeisler et al. (16) might have been due entirely to fixing the treatment subjects' sleep schedules, with the bright light playing only an incidental role. Accordingly, we designed an experiment to independently manipulate these two factors.
Twenty-seven men and 27 women aged 20–40 [mean 26.99 ± 6.22 (SD) yr] were included in the study after extensive clinical evaluation. Medical screening included a complete physical examination, clinical biomedical tests on blood and urine, electrocardiogram, psychological screening tests (Minnesota Multiphasic Personality Inventory and the Beck Depression inventory), and a Sleep Disorders Questionnaire (27).
Subjects were instructed to abstain from caffeine, nicotine, alcohol, and medication use for 3 wk before the study. On admission, and at the start of the constant routine, a comprehensive toxicological urinalysis verified subjects to be drug free. Subjects who reported travel greater than two time zones in the 3 mo before study, or a history of night work during the prior 3 yr or for >2 yr, were excluded.
Subjects wore a wrist activity monitor and called in sleep and wake times for 1 wk before study but were not required to maintain a regular sleep/wake schedule.
On days 1–4, subjects practiced the battery of computerized tests for 4 h per day, 0700–1100. The purposes of this segment were to reduce practice effects during the night shift and Constant Routine (see Constant routine) segments of the study and to ensure that subjects began the night shifts with their endogenous circadian phases similar both to each other and to workers on a typical shift rotation. From day 1 on, activity was monitored continuously with a wrist actigraph equipped with a light sensor (Actiwatch-L, MiniMitter, Sun River, OR). After testing was completed (1100), subjects left the laboratory and assumed normal activity and sleep cycles (see Fig. 1).
On day 5, subjects returned to the laboratory at 1700. Subjects were seated in dim light (<8 lux) from 1700 to 2300 and provided hourly saliva samples (SaliSaver, ALPCO). Compliance with the protocol was ensured by having a technician remain with the subject throughout the constant posture (CP) regimen.
Beginning on day 5, subjects worked three consecutive 8-h night shifts from 2300 to 0700. Subjects performed four iterations of a cognitive performance battery, including a visual analog scale measure of subjective alertness.1There were short breaks between individual tasks in the battery and 30-min breaks between batteries. During long breaks, subjects were instructed not to lie down, nap, or leave the suite, with video monitoring for compliance.
Final endogenous circadian phase was measured during a 38-h constant routine (CR). The CR, described in detail in Refs. 8,9, 16, 17, and 28, is designed to minimize or distribute evenly across the circadian cycle factors known to mask the endogenous component of core body temperature rhythm. Subjects were restricted to semirecumbent wakeful bed rest in dim light conditions (<8 lux). Food was distributed in hourly snacks. A technician was present at all times during the CR to ensure compliance with the protocol and to help the subject maintain wakefulness. Saliva samples were acquired and a 30-min test battery was administered hourly. Core body temperature (CBT) was monitored via a rectal thermistor (Yellow Springs Instruments, Yellow Springs, OH). Subjects were allowed 8 h of recovery sleep in the laboratory after the CR. The CR began at 1700 on day 8 and ended at 0700 on day 10.
During the night shift, there were two experimental manipulations: light, and sleep schedule. Bright Lightsubjects were exposed to ≈2,500 lux (measured in the angle of gaze from the seated position at the work desk) from 2300 to 0500, and ≈150 lux from 0500 to 0700. Room Light subjects were exposed to ≈150 lux for the full 8 h. The light was administered from ceiling-mounted fluorescent lamps.
Subjects' sleep schedule at home was the second experimental variable. Starting on day 6, Fixed Sleep subjects were given opaque material to cover their windows and instructed to be in bed with the lights off, trying to sleep from 0800 to 1600 (±30 min) each day. Subjects thus had a maximum of 1.5 h to travel home and prepare for bed.2 Subjects were not given goggles or sunglasses for the trip home. Compliance was monitored via actigraphy. Free Sleep subjects were told that they could sleep whenever they wanted to.
This design yields four groups of subjects: Bright Light & Fixed Sleep (hereafter Bright Fixed; Fig.1 A; n = 13, 2 women), Room Light & Fixed Sleep (Room Fixed; Fig. 1 B; n = 14, 6 women), Bright Light & Free Sleep (Bright Free; Fig. 1 C;n = 13, 8 women), and Room Light & Free Sleep (Room Free; Fig. 1 D; n= 14, 12 women). Subjects were randomly assigned to groups. All subjects were treated identically until the first night shift. Subjects were informed of their sleep schedules at the end of the first night shift.
Saliva samples were immediately centrifuged and frozen. They were assayed for melatonin concentration by radioimmunoassay (assay sensitivity of 22 pmol/l, intra-assay coefficient of variability of 8%, interassay coefficient of variation of 13%; DiagnosTech, Osceola, WI).
Final melatonin phase was defined as the midpoint of the melatonin secretion episode (MMSE). Mean melatonin concentration during the first 24 h of the CR was calculated, and the MMSE was computed as the midpoint of the upward and downward mean crossings (53,54). The dim light melatonin onset [DLMO (23,39)] was defined as the time that the melatonin levels during the CP reached 20% of the maximum melatonin level observed during the CR. A DLMO was also calculated for the CR.
Final CBT phase was computed by a two-harmonic fit to the temperature data (9). The phase was defined as the average of the nadir of the fundamental and of the composite fit (CBTmin).
Two subjects (one in the Bright Fixed group and one in the Room Fixed group) were excluded from the melatonin analyses because of data collection problems. The mean MMSE is plotted by group in Fig. 2 A. The data clearly show that the combination of bright light and scheduled sleep in darkness produced the greatest adaptation to night work. The mean MMSE of this group, 0806, has moved into the sleep episode. Data were submitted to a 2 (Bright Light vs. Room Light) × 2 (Fixed Sleep vs. Free Sleep) ANOVA. TheBright Light group's phase was significantly delayed with respect to the Room Light group [F(1,48) = 28.72, P < 0.0001], and a fixed sleep schedule significantly delayed MMSE phase compared with a free sleep schedule [F(1,48) = 17.23, P < 0.0005]. There was no interaction [F(1,48) <1].
The CBT phase analysis (shown in Fig. 2 B) was consistent with analysis of the melatonin data. Bright light significantly delayed the CBT nadir [F(1,50) = 25.32,P < 0.0001], as did a fixed sleep schedule [F(1, 50) = 16.97, P < 0.0005]; again the two factors did not interact [F(1,50) <1].
Initial phase was computed from the CP data. Nine subjects either did not exhibit a DLMO during the CP, according to the criterion, or had insufficient data during the CP. Data from the remaining subjects were analyzed via ANOVA. There was no effect of light treatment [F(1,39) <1]. The Fixed Sleep subjects' DLMO was 44.2 min later than that of theFree Sleep group [F(1,39) = 5.76, P < 0.05]. There was no interaction [F(1,39) <1]. We do not know why the initial DLMO, measured before subjects were given a sleep schedule, should have differed as a function of sleep schedule. To ensure that this initial difference in phase was not responsible for producing our observed distribution of final phases, we reanalyzed both the final MMSE phase data and the CBTmin data by ANCOVA by use of the CP DLMO as a covariate.3 The ANCOVAs produced nearly identical results to the original ANOVAs. For the MMSE analysis, the main effect of light was significant [F(1, 38) = 31.25,P < 0.0001], as was the main effect of sleep schedule [F(1,38) = 17.05, P < 0.0005], with no interaction [F(1,38) <1]. Similarly, for the CBTmin analysis, the main effect of light was significant [F(1,38) = 30.36,P < 0.0001], as was the main effect of sleep schedule [F(1,38) = 13.58, P < 0.001], with no interaction [F(1,38) <1].
We determined the phase shift from CP to CR by computing the first DLMO during the CR and subtracting this from the CP DLMO. Figure2 C plots the mean phase shift as a function of group. An ANOVA on the phase shifts showed a main effect of light [F(1,39) = 19.64, P < 0.0001], with larger delays for the Bright Light subjects, and a main effect of sleep [F(1,39) = 6.60, P < 0.05], with Fixed Sleepsubjects' final phase delayed more than that of the Free Sleep subjects. These results confirm that the pattern of final phases observed in the MMSE data was due to the phase-shifting effects of bright light and scheduled sleep in darkness, rather than to preexisting group differences.
We analyzed the subjective alertness data from the part of the CR corresponding to the hours of the night shifts, from 2300 to 700 ondays 8 and 9. We selected these data, instead of data from the actual night work shifts, because all subjects are under exactly the same conditions during the CR, so the data were not contaminated by the acute effects of light (52). Because the study schedule required all subjects, regardless of sleep schedule group, to sleep between the end of the third night shift at 0700 and the start of the CR at 1700 on day 8, differences in homeostatic sleep pressure should also be minimized. The eight hourly measurements per day were averaged and then entered into a 2 (Bright vs. Room Light) × 2 (Fixed vs. Free Sleep schedule) × 2 (day 8 vs. day 9) mixed ANOVA. There were significant main effects of all three factors. Bright Lightsubjects were more alert than Room Light subjects [F(1,50) = 7.72, P < 0.01], and Fixed Sleep subjects were more alert thanFree Sleep subjects [F(1, 50) = 12.19, P < 0.005]. Finally, alertness was substantially lower on the 2nd day of the CR [F(1,50) = 103.25, P < 0.0001], because subjects had been awake for 24 h longer by this time. There were no interactions (all P < 0.10). Data are shown in Fig. 3 as a function of group. Figure 3 A plots the data on the 1st day of the CR, and Fig. 3 B shows data from the 2nd day.
For each subject, we separately analyzed four activity measures: the average sleep start time, the average wake time, and the standard deviations of sleep start time and wake time. These values were computed using the SleepWatch software (MiniMitter). Data were incomplete for the night shift days for 9 subjects (3 each in theFree Sleep groups, 1 in the Bright Fixed group, and 2 in the Room Fixed group). For the remaining subjects, the mean sleep start time (hours of the day ± SD) was 0807 (±07) for the Bright Fixed group, 0935 (±41) for theBright Free group, 0818 (±06) for the Room Fixedgroup, and 0842 (±22) for the Room Free group. Mean wake times (SD) were 1548 (±06) for the Bright Fixed group, 1530 (±29) for the Bright Free group, 1556 (±05) for theRoom Fixed group, and 1533 (±37) for the Room Free group. The SD of sleep start time over the 3 days were 17 (±02) min for the Bright Fixed group, 1 h 23 (±33) min for the Bright Free group, 15 (±02) min for theRoom Fixed group, and 29 (±13) min for the Room Free group. The SD of wake times over the 3 days were 31 (±10) for the Bright Fixed group, 1 h 43 (±21) min for theBright Free group, 31 (±04) min for the Room Fixed group, and 1 h 37 (±21) min for the Room Free group.
ANOVA showed that Free Sleep subjects went to sleep later than Fixed Sleep subjects [F(1,41) = 6.39, P < 0.05]. Light had no effect on sleep start times [F(1, 41) <1], and there was no interaction [F(1,41) = 2.08, P > 0.10]. Wake time did not differ significantly among groups [allF(1,41) <1]. The Free Sleepsubjects were also more variable in their behavior. The standard deviation of sleep start time was higher for the Free Sleepgroup [F(1,41) = 6.28, P< 0.05]. There was a trend toward more variable sleep onset times in the Bright Light condition [F(1,41) = 2.91, P = 0.096], but there was no interaction [F(1,41) = 2.68, P > 0.10]. Variability in the wake time was substantially higher for theFree Sleep group [F(1,41) = 21.84, P < 0.0001]. Again, there was no effect of light or any interaction [both F(1,41) <1].
In the Free Sleep conditions, there was wide variability in the final (MMSE) phase. Figure 4illustrates that this measure is negatively correlated with variability in wake time. The more variable the subjects' sleep patterns, the less likely they were to adapt to the night shift schedule. Across groups, variability in wake times is negatively correlated with final phase [r = −0.546; t(43) = 4.28, P < 0.0001]. Mean wake time itself did not predict final phase [r = −0.136,t(43) <1]. The mean time that subjects went to sleep was also negatively correlated with final phase, although not as strongly as the standard deviation of wake time [r= −0.426, t(43) = 3.09, P< 0.005]. Variability in sleep onset time was not significantly related to final phase [r = −0.235,t(43) = 1.58, P > 0.10]. Of course, all four of these variables are strongly correlated with one another. Sleep onset time naturally determines wake time (r = 0.858, P < 0.0001). More importantly, mean sleep and wake times will determine the variability because of the study schedule. On day 8, subjects have only 10 h between the end of the final night shift at 0700 and the start of the CR at 1700, so all sleep episodes on day 8 take place within this window. Therefore, subjects who go to sleep later ondays 6 and 7 will have high variability, as well as later mean sleep and wake times. We focus on the standard deviation of wake time, because it is the best predictor of final phase in theFree Sleep groups. The four panels in Fig. 4 scatter plot final MMSE phase against the standard deviation of wake time for each group. It is clear from Fig. 4 that the overall correlation between waketime variablity and phase is generated by the Free Sleepschedule subjects (Fig. 4, C and D). There is little variability in the wake times of subjects in the Fixed Sleep groups (Fig. 4, A and B), indicating that subjects followed instructions. For the Bright Fixedsubjects, the correlation with final phase is −0.136 [t(10) <1], and for the Room Fixed subjects, the correlation is close to zero [r = 0.059, t(10) <1]. For the Room Free subjects, however, wake time variability was strongly negatively correlated with final phase [r = −0.634, t(9) = 2.46 < 0.05]. The relationship is strongest for the Bright Free subjects [r = −0.750, t(8) = 3.21, P < 0.05].
The ability of bright light to induce precise phase shifts under controlled laboratory conditions is now well known (14, 14,19). Controlled simulations of shift work schedules have convincingly demonstrated the potential value of applying circadian principles to the problem of night work (13, 30). Our data demonstrate that scheduling of sleep/darkness should play a major role in prescriptions for overcoming shift work-related phase misalignment. Both bright light and a fixed sleep/dark schedule significantly delayed melatonin phase, and the effects of these two factors were additive. Neither factor alone was sufficient to induce consistently adequate phase shifts. In fact, we found that a fixed sleep/wake schedule accounted for 3.17 h of phase delay, compared with 4.10 h for 2,500 lux of light; together, they yielded 7.28 h, sufficient to induce complete physiological adaptation to night shift work. Of course, Fig. 4 indicates that our manipulation was complicated by the fact that seven subjects in the Free Sleep condition voluntarily adopted schedules that were less variable than the most variable Fixed Sleep subject, so we may be underestimating the contribution of a fixed sleep schedule to circadian adaptation to night work.
The Bright Fixed subjects (Fig. 4 A) are all clustered in the upper left of the plot. This is the outcome of a typical controlled laboratory study: subjects are on consistent, experimenter-selected schedules and receive the full benefit of properly timed bright light treatment. The midpoint of the melatonin episode moved into their sleep episodes (0800–1600). Figure4 D depicts subjects who behaved most like night shift workers in the real world, in that they manifested a variety of sleep/wake schedules, generally did not have lightproof bedrooms, and were exposed to room light only at work. This group fares quite poorly.
In Fig. 4 B, the Room Fixed subjects generally did not shift, although some did achieve a phase near 0800 at the start of the sleep episode. It is important to remember that even normal room light can elicit a significant phase-shifting effect in the laboratory (6). Measurements of the human circadian pacemaker's dose-response curve to light suggest that one-half of the maximal type I resetting response is achieved at ∼100 lux (5, 55). The presence of competing synchronizers, to which these subjects were exposed outside of the laboratory, probably accounts for why most of our Room Light subjects failed to adapt. Actual night shift workers face a similar situation, in which the light exposure they receive at work may be insufficient to overcome the effects of the bright light they encounter on the drive home or while running errands.
Figure 4 C shows the subjects who received bright indoor light treatment sufficient to overcome the competing environmental synchronizers, but who were not required to keep a fixed sleep schedule in a darkened bedroom. The outcome for these subjects strongly depended on whether they chose to maintain a consistent sleep/wake schedule. Those who did adapted as well as the Bright Fixed subjects; indeed, the latest midpoint, near 1100, is from this group. But according to the slope of the regression line (not shown), increasing the variability of wake time by one standard deviation entailed the sacrifice of 2 h 36 min of phase delay, so that subjects with highly variable sleep/wake schedules were as poorly adapted to night work as the worst-off subjects in the Room Light conditions; the earliest midpoint, near 2300, is also from this group.
It is important to recognize that we cannot, from these experiments, determine what it is about the scheduled sleep in darkness that promotes adaptation. There are several (not mutually exclusive) possibilities. A fixed period of darkness changes the distribution of light throughout the day, therefore changing the photic effects on the pacemaker. Alternatively, the timing of sleep itself may act as a synchronizer. Finally, a sleep episode in a darkened bedroom may act as a “dark pulse” on the pacemaker (10, 49). Whatever the underlying mechanism (and more than one mechanism may be acting in concert), a consistent sleep/wake schedule should minimize competition from the natural schedule of synchronizers.
Of course, generalizing from laboratory studies to actual shift work situations can be challenging. Our goal was to create a “high-fidelity” laboratory simulation of night shift work. Except for the CR and CP episodes, subjects left the laboratory when not working their shifts. They were thus exposed to typical, uncontrolled patterns of natural light and social interaction experienced by people working the night shift. We are thus more confident about the generalizability of our results than we would have been if our subjects had spent the entire experimental protocol isolated in controlled conditions in the laboratory. However, one caveat is that our subjects were not allowed to use psychopharmacological agents. Use of drugs such as alcohol, nicotine, and caffeine, as well as over-the-counter sleep medications, is quite common among shift workers (42). All of these agents can affect sleep quality, and there is some evidence that alcohol (4) and nicotine (44) can directly alter circadian rhythms. We have no evidence from this study on how light or scheduled sleep may interact with drug use to affect circadian adaptation to, or performance on, the night shift.
Our results suggest that shift workers should be strongly encouraged to adopt a consistent sleep schedule, avoiding the common practice of changing their sleep/wake schedules during the work week. Further research involving multiple light levels will be required, so that appropriately higher light intensities may be recommended in situations in which fixed sleep/dark schedules are impractical. Conversely, given the expense of installing bright light systems, it is important to know the minimal amount of light to recommend when workers can maintain an optimal fixed sleep schedule. Although controlled laboratory studies suggest that the phase-shifting effects of light saturate ∼600 lux of the phase-shifting effects of light (5,55), this finding requires recalibration in a shift work situation in which there are competing synchronizers. Nevertheless, the implications of our findings are clear: use of both bright light exposure and scheduled darkness/sleep is required to achieve a reliable treatment for circadian maladaptation to night work.
We thank the subject volunteers, the recruiters (Conor O'Brien, Naomi Gonzalez, and Serena Ma), and the research technicians.
This work was supported by National Institutes of Health (NIH) Grant HL-52992 and the Air Force Office of Scientific Research Grant F49620-95-1-0388, and an NIH Division of Research Resources General Clinical Research Center Grant GCRC-2-MO1-RR-02635.
Address for reprint requests and other correspondence: C. A. Czeisler, Division of Sleep Medicine, Dept. of Medicine, Brigham and Women's Hospital, 221 Longwood Ave., Boston, MA 02115 (E-mail:).
↵1 Other data from this battery will be published in a separate paper.
↵2 Subjects who lived more than an hour's travel from the laboratory (by car or public transit) were excluded during the screening process.
↵3 We are grateful to an anonymous reviewer for this suggestion.
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
- Copyright © 2001 the American Physiological Society