Reduced Tolerance to Night Shift in Chronic Shift Workers: Insight From Fractal Regulation

Abstract

Study Objectives:

Healthy physiology is characterized by fractal regulation (FR) that generates similar structures in the fluctuations of physiological outputs at different time scales. Perturbed FR is associated with aging and age-related pathological conditions. Shift work, involving repeated and chronic exposure to misaligned environmental and behavioral cycles, disrupts circadian coordination. We tested whether night shifts perturb FR in motor activity and whether night shifts affect FR in chronic shift workers and non-shift workers differently.

Methods:

We studied 13 chronic shift workers and 14 non-shift workers as controls using both field and in-laboratory experiments. In the in-laboratory study, simulated night shifts were used to induce a misalignment between the endogenous circadian pacemaker and the sleep–wake cycles (ie, circadian misalignment) while environmental conditions and food intake were controlled.

Results:

In the field study, we found that FR was robust in controls but broke down in shift workers during night shifts, leading to more random activity fluctuations as observed in patients with dementia. The night shift effect was present even 2 days after ending night shifts. The in-laboratory study confirmed that night shifts perturbed FR in chronic shift workers and showed that FR in controls was more resilience to the circadian misalignment. Moreover, FR during real and simulated night shifts was more perturbed in those who started shift work at older ages.

Conclusions:

Chronic shift work causes night shift intolerance, which is probably linked to the degraded plasticity of the circadian control system.

shift work tolerance, circadian misalignment, adaptation, temporal correlations.

Statement of Significance

Shift work is prevalent in the modern society. Accurate and subjective assessment of tolerance to shift work is important but remains a challenge. Fractal patterns in spontaneous fluctuations of physiological functions are a hallmark of healthy physiology. The perturbed fractal activity regulation in chronic shift workers during night shifts indicates degraded adaptability and reduced night shift tolerance in these individuals. It is warranted for future studies to determine whether the degraded adaptability is caused by altered sleep/circadian control and associated cognitive declines. This study suggests that fractal regulation in motor activity may serve as a promising, unobtrusive tool to assess adverse impacts of shift work on health and behavior.

INTRODUCTION

Shift work is more and more prevalent in the modern society. In the United States, approximately 15% of the workforce undertakes shift work, and about 10% engaged in true night work, including permanent night work, rotating shifts, or irregular schedules.¹ Epidemiological studies provide strong evidence that shift work has significant, negative impacts on task performance^2–5 and cardiovascular health.^6–8 Whether or not chronic shift workers can better adapt to shift schedule when compared to individuals without any experience of shift work is not clear. It is possible that chronic shift workers have increased shift work tolerance due to a training/adaptation effect (ie, the system is better prepared for night shift schedules after repeated exposures) and the selection bias (ie, only those who think/know they can cope with shift schedule start and continue to perform shift work).⁹ Alternatively, the tolerance may be reduced in these individuals due to the long-term adverse effects of night shifts on behavior and physiology.¹⁰

One common pathway through which shift schedules influence alertness, cognition, and physiological function is related to the mistiming of the behavioral cycle relative to the endogenous circadian system.⁸ The circadian system optimally regulates physiological processes and behaviors when it is in synchrony with the daily environmental and behavioral cycles. However, shift work, which is characterized by behaviors and environmental exposures such as light and sound occurring at “wrong” times,^11–14 disturbs the normal synchronization between the endogenous circadian rhythm and the environmental/behavioral cycles. This misalignment is thought to contribute to the increased risk of cardiometabolic diseases in shift workers,^{7,12,13,15–20} and long-term exposure to the misalignment is also believed to cause circadian dysfunction and sleep disorders.²¹ In addition to the rhythms of ~24 hours, our recent studies showed that the circadian system impacts physiological control over a wide range of time scales.^22–24 For instance, motor activity and heart rate display fractal fluctuations (ie, the patterns of fluctuations are very similar at different time scales), and disrupted function of the central circadian pacemaker (suprachiasmatic nucleus) causes loss of the fractal patterns.²⁵ Fractal regulation (FR) is an intrinsic property of many biological systems and are believed to be a hallmark of healthy physiology, indicating system integrity and adaptability.²⁶ Numerous studies have demonstrated that alterations in fractal physiological fluctuations are associated with aging, pathological conditions, and diseases,^26–32 in which systems become less adaptive to perturbations and more vulnerable to catastrophic events. Based on the previous findings, we here tested the hypotheses that FR is disrupted in humans during night shifts and that the night shift-induced disruption is more pronounced in chronic shift workers due to the dysfunction of the circadian system in these individuals when compared to people who do not work during the night.

To test our hypotheses, we conducted a field study in which we collected and examined ambulatory activity recordings of 13 chronic shift workers (24.6–48.8 years old; 8 females and 5 males) who typically worked night shifts for 1–3 days followed by several days of rest or “normal” sleep–wake cycles during which they slept at night (Figure 1A). These participants had undergone night shift work for 1–25 years (Mean ± SD: 5.5 ± 7.2 years). As controls, we also studied ambulatory activity recordings of 14 individuals (20.8–50 years old; 8 males and 6 females) who had no night shifts at least 3 years prior to this study and no consecutive night shifts for more than 6 months across their life (Figure 1B). To determine whether circadian misalignment perturbs FR (while controlling behaviors such as sleep and food intake and environmental conditions such as temperature and light) and whether chronic shift workers and non-shift workers have different response to night shift schedules, we further studied the same participants (except for 2 chronic shift workers; see Methods) in two in-laboratory protocols—the in-laboratory study portion, in which night shift schedules (12-hour inverted sleep–wake cycles) and day shift schedules (normal sleep–wake cycle) were simulated (Figure 1C–F). To evaluate FR, detrended fluctuation analysis (DFA) was performed to examine the temporal correlations in the activity fluctuations at different time scales from ~0.2 to 8 hours.²² Since the mean activity values during sleep episodes are close to the sensitivity threshold of the device (Actiwatch; see details in Methods), the assessment of correlations might be unreliable. We therefore excluded all data during reported/scheduled sleep episodes and when participants underwent test batteries with highly controlled/restricted behavior in laboratory. We observed that the motor activity fluctuations of the control participants in the field study possessed robust fractal patterns as characterized by similar temporal correlations at different time scales. The correlations were degraded in shift workers during night shifts, and the degradation remained even two days after a night shift. The in-laboratory study confirmed this perturbed FR in the same shift workers during simulated night shifts, while the same control participants showed more resilience to the simulated night shift. Besides, we found that the night shift-induced perturbation was worse in those chronic shift workers who started to perform shift work at older ages.

Figure 1

Activity recordings of a representative shift worker and a control participant. (A) A chronic shift worker with 2 night shifts followed by 1 day off and then 1 nightshift followed by 2 days off. (B) A control participant with regular sleep–wake cycles (bed time ~12 AM and wakeup time ~8 AM). (C) The same shift worker as in (A) in a simulated night shift protocol. (D) The same shift worker as in (A) in a simulated day shift protocol. (E) The same control participant as in (B) in a simulated night shift protocol. (F) The same control participant as in (B) in a simulated day shift protocol. Gray-shaded areas indicate self-selected (field study) or scheduled (in-laboratory study) sleep or nap episodes.

METHODS

Participants

Thirteen chronic shift workers (8 females and 5 males; 24.6–48.8 years old; mean [SD] age: 35.9 [7.3] years) and 14 control participants (8 males and 6 females; 20.8–50 years old; mean [SD] age: 28.2 [9.2] years) were enrolled in this study. The non-shift worker group was slightly but significantly younger than the shift worker group (p = .02). All participants were healthy, nonsmoking, and drug- and medication free (except for oral contraceptives), and they all did not have sleep disorders. All shift workers had 1–25 years of consecutive shift work (mean [SD] 5.5 [7.2] years) and a mean cumulative duration of shift work exposure of 5.3 years (SD 7.7 years). All controls reported no shift work in the past 3 years and less than 6 months of cumulative lifetime shift work exposure. In these control participants, seven were unemployed, four were daytime workers (three full-time with regular work schedules and one with varied work time during the daytime, and three were students). All shift workers typically had night shifts for two or three days followed by several days off from work with normal sleep–wake cycles. Control participants maintained normal sleep–wake schedules with sleep episodes between 9PM and 9AM. Sleep schedules of all participants were assessed during the study period based on the sleep–wake diaries, time-stamped voice recordings, and actigraphy data. The study was approved by the local Institutional Review Board. Participants provided informed consent prior to participation.

Protocol

Field Protocol

Each participant undertook an ambulatory protocol of 1–3 weeks in which participants maintained their “normal” daily schedules. Control participansts kept their habitual sleep–wake cycles with sleep during the nighttime. Chronic shift workers might undergo their night work schedules during night shifts or normal sleep–wake cycles during days off (Figure 1A and B). The only requirement for the shift workers was that their sleep episodes should be scheduled at 11PM–7AM the day before they entered the laboratory.

In-Laboratory Protocol

Following the home monitoring, all 14 control participants and 11 of the 13 chronic shift workers continued to participate in the in-laboratory portion of the study. The ages of the 11 shift workers (34.5 ± 7.6 years old; 24.6–48.8 years old) were matched to those of the 14 control participants (28.2 ± 9.2; 20.8–50 years; p > .05). All participants underwent two laboratory protocols according to a crossover design as previously published^15,16: (i) The day shift protocol with normal sleep–wake cycles and (ii) the simulated night shift schedule. For control participants, the day shift protocol included 7 days of normal sleep–wake cycles (ie, 8-hour sleep during the nighttime), and the night shift protocol included 3 days of baseline with regular sleep–wake cycles followed by a 4-hour nap on day 4 and additional 3 days with sleep episodes scheduled during the daytime (ie, 12-hour shift when compared to the baseline days; Figure 1E and F). Light intensity during wakefulness was ~4 lux on day 4 and ~90 lux during the additional 3 days. Light intensity during scheduled sleep/nap episodes was 0 lux. For chronic shift workers, the day shift protocol included only 2 days of normal sleep–wake cycles, and the night shift protocol included 1 day with a 4-hour nap and another day with the sleep episode scheduled during the daytime (ie, no baseline days; Figure 1C and D). Light intensity during wakefulness was ~4 lux on day 1 and ~90 lux on day 2. Light intensity during scheduled sleep/nap episodes was 0 lux. Among all the 14 control participants, 7 undertook the circadian alignment during their first visits and 7 the circadian misalignment first. Among the 11 shift workers, 5 participants undertook the circadian alignment protocol first and 4 took the circadian misalignment first. Throughout the in-laboratory study, the room temperature was adjusted to each individual’s preference and then kept relatively constant (Mean ± SD: 75°F ± 3°F). Diet was also controlled and details have been previously published.^15,16 Briefly, all participants received an ad libitum lunch at ~12 PM on day 1 in both protocols. Thereafter, on each laboratory day, participants received an isocaloric diet calculated according to the Harris–Benedict equation with an activity factor of 1.4. The time of each meal was fixed, with breakfast scheduled at 8 AM (day shift) or 8 PM (night shift), lunch at 11:30 AM (day shift) or 11:30 PM (night shift), dinner at 8 PM (day shift) or 8 AM (night shift), and a snack opportunity at 3 PM (day shift) or 3 AM (night shift).

There were 4 shift workers who did not follow the required sleep schedule the day before entering the laboratory (n = 2) or had unstable timing of their central circadian clock based on the dim light melatonin onset and offset comparisons between both visits (n = 2),¹⁵ who were excluded prior to the analyses.

Data Collection

Motor activity levels were continuously monitored during the whole study period (including both ambulatory and laboratory protocols) using a wrist actigraphy (Actiwatch Spectrum, Philips-Respironics, Murrysville, PA, or Actiwatch-L, Mini Mitter, Bend, OR) worn on the nondominant hand. Acceleration was sampled at 32 Hz and was integrated to a proprietary “count” value every minute.

Assessment of Fractal Regulation

To quantify FR in activity fluctuations, we performed DFA.²² The DFA algorithm includes four steps: (i) integrating the time series of an activity signal after removing its global mean; (ii) dividing the integrated signal into nonoverlapping windows of the same chosen size n (ie, time scale); (iii) detrending the integrated signal in each window using polynomial functions to obtain residuals; and (iv) calculating the root mean square of residuals in all windows as fluctuation amplitude F(n). The same four steps are repeated for different time scale n. To reliably estimate F(n) at a time scale n, at least 4 segments of data each with n points must be available. Otherwise, F(n) will not be obtained at that time scale. To eliminate the effect of possible linear trends in original data, we applied the second-order DFA, that is, the second order of polynomial functions were used to detrend data.³³ A power-law form of F(n) indicates self-similarity (scale invariance) in the fluctuations, yielding F(n)~n^α. The parameter α, called the scaling exponent, quantifies the correlation property in the signal as follows: If α = 0.5, there are no correlations in the fluctuations (“white noise”), and if α > 0.5, there are positive correlations, where large activity values are more likely to be followed by large activity values (and vice versa for small activity values). The exponent α = 1.0 indicates highest complexity in the systems, and similar α values close to 1.0 have been observed in many physiological outputs under normal conditions.^26,34

Previous studies suggested that FR in humans is disrupted with aging and in dementia, leading to distinct correlations over two time scale regions with the boundary at ~1.5–2 hours.^30,32 Thus, for each activity recording in this study, we calculated the scaling exponent in two regions, separately, that is, α₁ at <90 minutes and α₂ at >2 hours (up to 8 hours because the minimal segment length without gaps was ~ 8 hours), omitting the variable transitional region of time scales between 1.5 and 2 hours.

Note that data points with activity levels below the sensitivity of Actiwatch (i.e., 0.01 g, where g is the unit of acceleration, matching the earth’s gravity) are assigned a value of zero.³⁵ Such threshold influence is negligible when activity is relatively high during the active periods. But when the activity levels are very low, the influence can greatly reduce the signal-to-noise ratio that can significantly affect the DFA results.³⁶ Thus, we excluded data during sleep episodes at home and in-laboratory data when participants underwent test batteries with highly restricted behaviors (i.e., confined to bed without movements). As a consequence, we could not obtain α₂ in many cases because F(n) could not be obtained over a time scale region wide enough (ie, from 2 hours to at least 5 hours) for a reliable estimation of α₂: (i) For the analysis of individual days after a night shift at home (eg, Day off 1, Day off 2, and etc.), α₂ could not be obtained in some participants for certain days. Five shift workers had both α₁ and α₂ values for day off 1; 10 shift workers for day off 2; 3 shift workers for day off 3 and for day off 4; and 1 shift work for day 5. Note that there were less participants for day off 1 than for day off 2 because shift workers typically had an additional sleep episode (excluded from the correlation analysis) after a night shift. Thus, we only considered the results of day off 1 and day off 2. (ii) For in-laboratory data, recordings in shift workers were not long enough to quantify F(n) over such a wide region at large time scales. In addition, there were frequently scheduled tests/tasks such as metabolic tests that did not have a huge influence on participants’ activity at small time scales but could impose certain pattern at large time scales (>2 hours) to overall activity fluctuations. Thus, true activity correlations at >2 hours might be masked by these scheduled events.

Statistical Analysis

Mixed models with participant as a random factor for intercept were used to examine the differences in activity correlations between different scale regions (ie, <1.5 hours and >2 hours) for controls during the daytime, shift workers during free days (ie, days off from work), and shift workers during night shifts, separately. Similar mixed models were used to determine (i) the changes in activity correlations and mean activity levels between night shifts and free days in shift workers at home; (ii) the changes in activity correlations between simulated night and day shifts in both control and shift workers in the laboratory; (iii) the differences in correlations between controls and shift workers; and (iv) the changes in activity correlations between home and in-laboratory conditions (ie, simulated day shifts vs. daytime for controls; day shift vs. days off for shift workers; simulated night shifts vs. real night shifts in shift workers). We also performed three additional mixed models to determine the effects of age, cumulative duration of shift work exposure (“duration”), and age at which participants started to perform shift work (“starting age”), respectively. In each model, we combined the data of the field and in-laboratory studies and included one of three variables (age, duration, or starting age), the condition (night/day shift), and their interaction as fixed factors and participant as a random factor. For the above additional mixed models, we explored the potential differences between the field study and the in-laboratory study. Statistical significance was accepted at p < .05. Statistical analyses were performed using JMP Pro 12 (SAS Institute, Cary, NC).

RESULTS

Results of the Field Study

In control participants, activity during daytime at home showed fractal temporal fluctuations as characterized by a power-law form of the DFA-derived fluctuation function, F(n)~n^α, across time scales ~0.1–8 hours (Figure 2). The scaling exponent of the power-law function, α~1.0, indicates strong temporal correlations in activity fluctuations. The correlations were similar at all tested time scales (Figure 3A), for example, α₁ = 1.00 ± 0.02 (mean ± SE; same below) at small time scales (<1.5 hours) and α₂ = 0.97 ± 0.03 at large time scales (>2 hours, p > .2), suggesting robust fractal temporal correlations in activity fluctuations.

Figure 2

Temporal correlations in activity fluctuations in two representative participants. (A) Detrended fluctuation functions were obtained from the signals shown in Figure 1. On the abscissa, n represents the time scale in hours. The fluctuation functions F(n) are vertically shifted for better visualization of the differences between the control and the shift worker and between night shifts (field and in-lab) and day shifts (field and in-lab). A power-law function, F(n)~n^α, (ie, a straight line in the log–log plot) indicates a fractal or scale-invariant structure in the fluctuations of raw data. The exponent α indicates temporal correlations in the fluctuations. For white noise without correlations, α = 0.5 (dashed line). α > 0.5 indicates positive correlations. (B) F(n)/n obtained from the same functions in (A) for better visualization of the differences between controls and shift workers and between night shifts (field and in-lab) and day shifts (field and in-lab).

Figure 3

Group averages of the scaling exponents that quantify correlations in activity fluctuations. (A) Control participants during wakeful period at home and in the in-laboratory day shift and night shift protocols. (B) Shift workers during day and night shifts in ambulatory protocol and simulated day and night shifts in the in-laboratory protocols. (C) Shift workers in individual days after night shifts. The scaling exponents were obtained from power-law fits of fluctuation functions in two regions: 0.1–1.5 hours (α₁) and 2–8 hours (α₂). Data are presented as mean ± SE. The significant difference between the exponents over the two regions is indicated by *(p < .05), **(p < .01), or ***(p < .001) displayed above α₂. The significant difference between night shift (field or in-lab) and day shift (field or in-lab) is indicated by ^#(p < .05), ^##(p < .01), or ^###(p < .001).

Fractal patterns broke down in the activity of chronic shift workers during night shifts (Figure 2), that is, correlations at time scales >2 hours (α₂ = 0.74 ± 0.04) were weaker than those at time scales <1.5 hours (α₁ = 0.89 ± 0.01; p = .0007; Figure 3B). In addition, both α₁ and α₂ in chronic shift workers were significantly smaller than those in control participants (p = .0005 for α₁; p < .0001 for α₂), indicating more random activity fluctuations in shift workers.

Next, we analyzed activity data of the same shift workers on free days when they reverted to a normal sleep schedule (ie, sleeping at night and being awake during the daytime). First, we combined data from all days without work. We found that the activity correlations became stronger as characterized by increases in the scaling exponents, that is, α₁ = 0.97 ± 0.01, p = .0001; α₂ = 0.87 ± 0.04, p = .02. However, the fractal patterns were not fully restored as indicated by the slight but significant difference in correlations between the two time-scale regions (ie, α₁>α₂; p = .02; Figure 3B). When compared to controls, the α₁ of shift workers on free days was similar (p > .1), and α₂ had a tendency to be smaller although the difference did not reach significant level (p = .087).

To better understand how the disrupted fractal activity patterns (partially) recovered after a night shift, we further examined activity data during consecutive individual days following a night shift, separately. For the first day after a night shift (day off 1; within 24 hours after the completion of a night shift), α₂ (0.75 ± 0.06) was significantly smaller than α₁ (0.98 ± 0.03, p = .002), indicating that fractal activity patterns remained disrupted (Figure 3C). Compared to the night shifts, α₁ in day off 1 significantly increased (p = .0003), but α₂ remained similarly low (p > .8). For day off 2 (within 24–48 hours) after the last night shift, the values of α₂ (0.86 ± 0.04) and α₁ (0.99 ± 0.03) were much closer when compared to day off 1, but they were still significantly different (α₂ < α₁; p = .005; Figure 3C). There were not enough data/participants to test whether the perturbed correlations persisted beyond 2 days after the last night shift (see Methods).

Despite the changes in activity correlations from night shifts to days off, mean activity level during the wake time of days off was similar to that of night shifts, for example, mean activity levels in day off 1 (107.4% ± 10.3%) and day off 2 (104.6% ± 8.6%) were similar and not different from those during night shifts (p values > .1).

Results of the In-Laboratory Study

For the in-laboratory data, we did not consider α₂ because required duration of data records was insufficient for the assessment of fractal patterns at these larger time scales (see details in Methods). Thus, we presented only the results of α₁ at small time scales (<1.5 hours).

In control participants, α₁ under normal sleep–wake cycles in laboratory (ie, α₁ = 0.97 ± 0.02 in the day shift protocol; and α₁ = 0.96 ± 0.02 in the 3 days of baseline before the simulated night shifts, see Methods for details) was comparable to those of the same participants at home (both p values >.1; Figure 3A). During the simulated night shifts, α₁ (0.96 ± 0.02) showed no significant change when compared to the values at baseline, in the day-shift protocol, or at home (all p values >.1). There was no significant change in α₁ across the 4 days of simulated night shifts (p > .5; Figure 3A).

In chronic shift workers during the day-shift protocol, α₁ (1.00 ± 0.02) was not significantly different from those of the same participants during days off at home and control participants in the laboratory or at home (all p values >.5; Figure 3A and B). During the simulated night shifts, α₁ of these chronic shift workers was significantly reduced when compared to the value during the simulated day shift (0.93 ± 0.02; p = 0.03; Figure 3B), indicating more random activity fluctuations. Note the mean values of α₁ were slightly larger than that of the same participants during real night shifts (0.89 ± 0.01; p = .05).

Similar to the field study, the mean activity level during the simulated night shifts (111.3% ± 11.8%) was not different from that during the day shifts (p > .4).

Effects of Age of Starting Shift Work

It has been shown that the chronic effect of shift work on health depends on the age at which participants started to perform shift work (“starting age”), that is, increased cancer and cardiovascular diseases risks were observed in participants who started to perform shift work at ages >25 years old.³⁷ Thus, we further explored how starting age affects the response of fractal activity regulation to the night shift schedule. We found that shift workers who started to perform shift work at older ages had larger reductions in α₁ during night shifts (p = .0003), and this association was consistent in the field study and in the in-laboratory study (p > .05). Indeed, the reduction in α₁ was ~3 times larger in participants who started shift work at >25 years old when compared to other shift workers, in both the field and in-laboratory studies (Figure 4).

Figure 4

Dependence of the reduction in the correlations at <1.5 hours during night shifts on the age at which chronic shift workers started to perform shift work (‘starting age’). The reduction in the correlation was estimated by the difference in the scaling exponent at <1.5 hours between day shifts (or days off) and night shifts (Δα = α_1day − α_1night). For demonstration, data were presented as mean ± SE for those who started shift worker younger and older than 25 years, separately. Open bars were the data of the field study only, slashed bars the data of the in-laboratory study only, and black filled bars for the combined data of the field and in-laboratory studies. The p value indicates the significance level for the effect of starting age on the reduction of correlations at <1.5 hours (α₁) during night shifts. Results were based on the mixed model with starting age (a continuous variable), condition (night/day shift), and their interaction as fixed factors and participant as a random factor. Five of 13 shift workers started shift work at an age <25 years old and 4 at an age >25 years old. The remaining four shift workers could not recall the time when they started shift work.

We also performed additional mixed models to explore the effect of age and the effect cumulative duration of exposure to shift work (“duration”) separately. We observed no significant effect of age on the reduction in α₁ during night shifts (p > .1). Longer cumulative duration of exposure was weakly associated with less night shift-induced reduction in α₁ (p = .04) when not considering the effect of starting age, but the association became not significant (p > .1) after accounting for the effect of starting age. On the other hand, the association between age of starting shift work and the response to night shift remained highly significant (p < .002) when age or duration of exposure was included in the models.

DISCUSSION

Motor activity fluctuations in healthy young individuals display robust FR that is degraded with aging and in diseases.³⁸ Here, we showed that FR was disrupted in chronic shift workers during night shifts, leading to more random activity fluctuations over a wide range of time scales from 0.1 to 8 hours. Interestingly, night shifts in the field study had a much stronger impact on FR at >2 hours when compared to that at smaller time scales (Figure 3A). Such scale-dependent effect is reminiscent of previous human studies of aging and dementia^30,31 and animal lesion studies of key neuronal nodes in the circadian control system (eg, suprachiasmatic nucleus and dorsomedial hypothalamic nucleus).^39,40 In all these cases, the circadian control is perturbed and activity correlations are reduced more or mainly at larger time scales (> ~2–4 hours). Thus, the stronger influence of night shifts on activity fluctuations at time scales >2 hours is consistent with perturbed circadian control that is associated with night shifts. This interpretation is also supported by the slow recovery of activity correlations at large time scales in shift workers after night shifts (Figure 3C) because the circadian system adapts slowly to the changes in participants’ daily behavioral cycles.

The reduction in activity correlations at small time scales (<1.5 hours) in the shift workers during night shifts was also enlightening. Altered activity patterns at <2 hours are associated with mood disorders such as major depression, bipolar disorder, and schizophrenia,^41–45 and the reduction in activity correlations at <2 hours relates to worsening of depressive symptoms and cognition in dementia.³¹ In addition, mood and cognition disturbances are common during night shifts.^46,47 Based on these findings, one possible hypothesis is that the observed reduction in short-term activity correlations is linked to acute, adverse influences of circadian misalignment on mood and/or cognition. If this hypothesis is true, temporal activity correlations may serve as a useful, unobtrusive measure for monitoring mood and cognition during night shifts. Alternatively, stress related to night work itself instead of circadian misalignment may cause disrupted FR. But our consistent results in the field and in-laboratory studies appear to refute this possibility, given that the scheduled behaviors in the laboratory were identical and that the average 24-hour circulating cortisol concentration—a measure of stress—was not different between the simulated day and night shift protocols.^15,16 With the revealed impacts of circadian misalignment on FR, it will be important to determine the relationship between the degrees of circadian misalignment and FR disruption. However, this study was not suitable to determine such an association because the in-laboratory night shift protocol was designed to introduce near-maximal circadian misalignment such that the degree of circadian misalignment was in a narrow range (ie, phase shift was ~12 hours; see Figure S1 in the Supplementary Material).

One unexpected finding is that fractal activity correlations at small time scales were reduced in shift workers but not in the controls during the night shifts in the laboratory (Figure 3A). This finding suggests that there are certain compensatory responses in non-shift workers that can help to maintain FR during night shifts. Thus, the most likely interpretation of the finding is that the tolerance to night shifts is impaired by chronic shift work in term of FR. Regarding the underlying mechanisms, we consider chronic exposure to circadian misalignment that can cause dysfunction of circadian control as the main factor contributing to reduced adaptability. In addition to pure circadian misalignment, night shifts are associated with many other altered lifestyle factors including sleep deprivation, changes in food intake, and changes in exercise that can also impact physiology and behavior. A few recent studies explored interventions to improve/maintain FR. For instance, our animal studies showed that maintaining high physical levels can help to counteract the adverse effect of aging on FR,⁴⁸ while increasing daily activity amplitude does not necessarily improved disrupted FR⁴⁹; and our study of humans with dementia showed that increased daily light exposure can decelerate the degradation of FR with aging and dementia.³¹ Further studies with a longitudinal design are required to determine whether reducing circadian misalignment, improving lifestyle factors, and controlling daily environmental influences can help to maintain FR and prevent its degradation in shift workers during night shifts.

Another interesting finding is that the night shift-induced perturbation of fractal activity regulation worsened in those shift workers who started to perform shift work at older ages. This finding is consistent with the previous finding from a large cohort study of 54724 shift workers in which a later age at which night shift work was started was associated with an increased number of chronic disease risk factors.³⁷ The mechanisms underlying the effects of age of starting shift work are still unknown. One possibility is that the neural control system such as the circadian control network has a better “plastic” ability to rewire at younger ages and the ability is reduced with aging. An alternative hypothesis is that people who start shift work at an earlier age are more attracted to shift work because they can cope well with shift work (internal motivation), while those who start at a later age only start shift work because they are forced to start shift work due to external pressures, for example, financial (external motivation). It is warranted for future studies to test these hypotheses.

Our study has limitations that should be considered. First, the study protocols had their limitations. The duration of the study was not long enough to examine how long it will take for the chronic shift workers to fully recover from the effect of night shifts. In the in-laboratory protocols, there were many, frequently scheduled tests with highly restricted behaviors which could mask intrinsic activity patterns, especially at large time scales, although these were identical in the day shift and night shift protocols. Second, the controls were younger than the shift worker group. To ensure that the observed group differences were not simply caused by the age difference, we repeated all the statistical analyses by controlling for age. The results confirmed that the findings were independent of age (see details in Supplementary Materials). Third, the sample size was small. To optimize the statistical power, most analyses were based on within-participant comparisons. The only exceptions are the comparison between controls and shift workers and the influence of age of starting shift work on FR. Future studies with a large sample size are required to confirm these findings. Finally, in addition to motor activity, a wide range of physiological signals, such as brain activity,^50,51 heart rate,^27,52 respiration,²⁸ and gait,^53,54 show fractal patterns. Assessing FR in these variables should be important for future studies to understand the impact of night shifts on different physiological functions.

Nevertheless, this was the first study to show that healthy chronic shift workers have perturbed FR during night shifts and that the perturbation was more severe in those who started to perform shift work at older ages. These results suggest that chronic shift workers have reduced adaptability and lower night shift tolerance that may further exacerbate the adverse impacts on physiology from the environmental and occupational challenges associated with night shift schedules. FR offers a unique opportunity to understand how well the body functions at an integrative system level. The adopted fractal measure may allow noninvasive monitoring of detrimental effects of night shifts.

SUPPLEMENTARY MATERIAL

Supplementary material is available at SLEEP online.

FUNDING

This work was supported by NIH grants R01HL094806 (to F.A.J.L.S.) and R00-HL102241 (to K.H.). P.L. was supported by the International Postdoctoral Exchange Fellowship 20150042 from the China Postdoctoral Council; C.J.M. was supported by the NSBRI through NASA Grant NCC 9–58, R01-HL094806, R01-DK099512, and R01-HL118601; F.A.J.L.S was supported by R01-HL094806, R01-DK099512, and R01-HL118601, and K.H. was supported by R00-HL102241, R01AG048108, and P01AG009975.

DISCLOSURE STATEMENT

FAJLS reports speaker fees from Bayer Healthcare and Sentara Healthcare. All others have nothing to disclose.

ACKNOWLEDGMENTS

We thank Tommy To for preparing the figures. We also thank Dr. Jingyi Qian for the helpful discussion about the quantification of the degree of circadian misalignment.

Author notes

Address correspondence to: Kun Hu, PhD, Division of Sleep and Circadian Disorders, Brigham and Women’s Hospital, Boston, MA 02115, USA. Telephone: [phone]; Fax: [phone]; Email: [email protected]. harvard.edu; or Frank A. J. L. Scheer, PhD, Division of Sleep and Circadian Disorders, Brigham and Women’s Hospital, Boston, MA 02115, USA. Telephone: [phone]; Email: [email protected]; or Peng Li, PhD, Division of Sleep and Circadian Disorders, Brigham and Women’s Hospital, Boston, MA 02115, USA. Telephone: [phone]; Email: [email protected].