Aligning Work and Circadian Time in Shift Workers Improves Sleep and Reduces Circadian Disruption

Transcription

1 Report Aligning Work and Circadian Time in Shift Workers Improves Sleep and Reduces Circadian Disruption Highlights d Extreme chronotypes slept up to 0.5 hr longer on workdays d d Self-reported sleep quality in extreme chronotypes improved on workdays Overall social jetlag was reduced by z1 hr Authors Céline Vetter, Dorothee Fischer, Joana L. Matera, Till Roenneberg Correspondence till.roenneberg@med.uni-muenchen.de d Frequent night shifts are better tolerated by very late chronotypes In Brief The health deficits associated with shift work have been linked to circadian stress and disrupted sleep. Vetter et al. performed the first intervention study in a real-life industrial setting that adjusted shifts to individual chronotype and show that workers sleep longer and better and suffer from less social jetlag. Vetter et al., 2015, Current Biology 25, March 30, 2015 ª2015 Elsevier Ltd All rights reserved

2 Current Biology Report Aligning Work and Circadian Time inshiftworkersimprovessleep and Reduces Circadian Disruption Céline Vetter, 1,2 Dorothee Fischer, 1 Joana L. Matera, 1 and Till Roenneberg 1, * 1 Institute of Medical Psychology, Ludwig-Maximilian-University, Munich, Germany 2 Present address: Channing Division of Network Medicine, Brigham and Women s Hospital, Harvard Medical School, 181 Longwood Avenue, Boston, MA 02115, USA *Correspondence: till.roenneberg@med.uni-muenchen.de SUMMARY Sleep loss and circadian disruption a state of misalignment between physiological functions and imposed sleep/wake behavior supposedly play central roles in the etiology of shift work-related pathologies [1 4]. Circadian entrainment is, however, highly individual [5], resulting in different chronotypes [6, 7]. Chronotype in turn modulates the effects of working times: compared to late chronotypes, earlier ones sleep worse and shorter and show higher levels of circadian misalignment during night shifts, while late types experience more sleep and circadian disruption than early types when working morning shifts [8]. To promote sleep and reduce the mismatch between circadian and working time, we implemented a chronotype-adjusted (CTA) shift schedule in a factory. We abolished the most strenuous shifts for extreme chronotypes (i.e., mornings for late chronotypes, nights for early ones) and examined whether sleep duration and quality, social jetlag [9, 10], wellbeing, subjective stress perception, and satisfaction with leisure time improved in this schedule. Intermediate chronotypes (quartiles 2 and 3) served as a control group, still working morning (6:00 14:00), evening (14:00 22:00), and night (22:00 6:00) shifts, with two strenuous shifts (out of twelve per month) replaced by evening ones. We observed a significant increase of self-reported sleep duration and quality, along with increased wellbeing ratings on workdays among extreme chronotypes. The CTA schedule reduced overall social jetlag by 1 hr, did not alter stress levels, and increased satisfaction with leisure time (early types only). Chronotype-based schedules thus can reduce circadian disruption and improve sleep; potential long-term effects on health and economic indicators need to be elucidated in future studies. RESULTS AND DISCUSSION We designed the chronotype-adjusted (CTA) schedule in view of employees chronotypes (n = 114, Figure 1A), as assessed by mid-sleep time on free days after evening shifts, corrected for sleep debt during the work week (MSF E sc) with the Munich chronotype questionnaire for shift workers (MCTQ Shift ) [11], and production requirements (e.g., equally staffed shift groups). Employees were ranked and assigned by chronotype quartiles to the CTA schedule (Early1 = 2:26 3:34, Early2 = 3:36 4:08, Late1 = 4:11 4:52, and Late2 = 4:55 7:34). The local ethics committee approved the study, and participants gave written informed consent. In general, the chronotype distribution resembled the one of the general population (Figure 1A; Kolmogorov-Smirnov test: D = 0.45, p = 0.21) but lacked extreme mid-sleep times, i.e., measures of %1:00 or R9:00 [5]. We examined the match between sleep log (n = 47) and actimetry-derived (n = 19) homologs with the questionnaire-based chronotype proxy MSF E sc, and both corresponded well to the MCTQ Shift chronotype measure (range, intraclass correlation coefficient [ICC] = , p < 0.001; Figure 1B; for further information, see Supplemental Experimental Procedures, 3.2.). Participants filled out daily sleep logs, reporting sleep onset and offset, quality, and wellbeing (minimum: 0, maximum: 10, respectively) within 4 weeks of baseline measures in a fast-forward rotating schedule (Figure 2A, ), directly after transition to the CTA schedule (CTA1), and within the last 4 weeks (CTA2) of the 5-month intervention period. Similarly, questionnaire-based perceived stress levels [12] and satisfaction with the amount of free time [13] were assessed during baseline and twice within the CTA schedule. Overall, 58 employees (51%) filled out a sleep log; dropout rates were low (3.5%). In the analyses, we included all employees (1) whose chronotype was congruent with their group assignment (group-specific MSF E sc range ± 0:15) and (2) who completed sleep logs at baseline, CTA1, and CTA2 (for information on recruitment and exclusion procedures, see Supplemental Experimental Procedures, 3.1.). Late1 was not eligible for analysis, as two participants only fulfilled those criteria. The final sleep log sample (n = 28) included individuals that were mostly male (96%), 40.4 ± 10.6 years old (mean ± SD), overweight Current Biology 25, , March 30, 2015 ª2015 Elsevier Ltd All rights reserved 907

3 A B Figure 1. Chronotype Distributions and the in % :00 2:00 4:00 6:00 8:00 10:00 Chronotype (MSF E sc, hh:min) German Sample Shift Sample Chronotype ( hh:min, MSF E sc derived from MCTQShift ) Sleep Logs Actimetry Chronotype (av., hh:min, MSF E sc derived from sleep logs or actimetery) Validity of the Chronotype Measures (A) Chronotype distribution in Germany (2012, MCTQ database, n = 72,469) and among the shift workers at the steel factory (n = 114). (B) Correlation between the MCTQ Shift -derived chronotype variable (mid-sleep on free days, corrected for sleep debt accumulated during the work week) with averaged (av.) homologs extracted from sleep logs (open circles and dotted line, n = 47) and actimetry (black circles and solid line, n = 19). (BMI of 28.2 ± 5.6 kg/m 2 ), in a relationship (68%), had at least one child (54%), and had a chronotype of 03:58 ± 01:01. Those characteristics were similar across subgroups and instrument-specific samples (Table S1), with the exception of younger Late2 participants (29.3 ± 7.4 years). We used repeated-measures ANOVAs to assess the effects of the CTA schedule (see also Supplemental Experimental Procedures, 3.2.). The CTA Schedule Increases Sleep Duration and Quality on Workdays We initially aimed at using actimetry for objective sleep timing and duration assessments. However, recruitment rate (n = 26, 22.4%) was low, and after exclusions (see Supplemental Experimental Procedures, 3.1.), only 19 participants were eligible for analysis. We therefore used all data available at baseline and compared sleep log (n = 47) and actimetry-derived (n = 19) sleep duration and timing (see Supplemental Experimental Procedures, 3.3. and 3.4.; overlap between samples: n = 17). Indeed, self-reported and objective sleep duration and mid-sleep parameters (sleep onset sleep duration) corresponded well to one another (range, ICC = , p values < 0.007, Bonferroni corrected), suggesting that self-reports were indicative of actual behavior. Workday sleep duration differed significantly between groups (F(2,25) = 7.40, p = 0.003, partial h 2 = 0.37) with Early1 sleeping less (6 hr 23 min ± 29 min) than Late2 (7 hr 11 min ± 56 min, Bonferroni-corrected post hoc test: p = 0.003; Table S2 summarizes group-specific means ± SD). Both slept longer at CTA2 than at baseline (Early1: +28 min; Late2: +24 min; F(2,50) = 3.84, p = 0.03, partial h 2 = 0.13), while Early2 slept 14 min less (Figure 3A). Free day sleep duration showed a non-significant decrease from baseline to CTA2 in Early1 ( 24 min), Early2 ( 13min), and Late2 ( 58 min) (Figure 3B). Sleep log-derived sleep quality (scale from 0 to 10: very well ) improved on workdays in Early1 (+1.17) and Late2 (+0.60) but remained similar in Early2 (+0.01; F(2,50) = 3.48, p = 0.04, partial h 2 = 0.12; Figure 3C). Free day sleep quality was not affected by the shift schedule change (p = 0.65). Our results suggest that reducing exposure to the most strenuous shifts in extreme chronotypes could ameliorate chronic sleep deprivation, a major health and security hazard in shift workers [1, 14], and improve sleep quality on workdays. Early2, who experienced mainly social and ergonomic changes, but no major reduction of strenuous shifts, showed relatively stable measures of sleep duration and quality throughout the study. Despite the lack of a true control condition, this is suggestive of the potential benefit of CTA schedules on self-reported sleep duration and quality. The CTA Schedule Can Reduce Circadian Misalignment Social jetlag [9, 10], our central measure of circadian misalignment, was derived from sleep logs and computed by the absolute difference between the mid-point of sleep on workdays and free days. In shift workers, total social jetlag is the weighted average of shift-specific social jetlag, thereby accounting for the Figure 2. Shift Schedules (A) The initial schedule. Employees work the identical sequence of shifts. Morning shift (MS) is indicated by the striped box; evening shift (ES) is indicated by the gray box; night shift (NS) is indicated by the black box; and free day is indicated by the white, black-rimmed box. (B) The chronotype-adjusted (CTA) schedule. We abolished the theoretically most strenuous shifts for extreme chronotype groups (NS for Early1, MS for Late2). The intermediate groups, Early2 and Late1, experienced less dramatic changes with regards to the shift sequence (Early2: two NS less, two ES more; Late1: two MS less, two ES more). 908 Current Biology 25, , March 30, 2015 ª2015 Elsevier Ltd All rights reserved

4 A Mean (h ± SEM) C Mean (Scale 0-10, a.u. ± SEM) Sleep Duration on Workdays CTA CTA 1 CTA 2 Sleep Quality on Workdays CTA 2 Early 1 Early 2 Late 2 Early 1 Early 2 Late 2 B Mean (h ± SEM) D Mean (h ± SEM) Sleep Duration on Free Days CTA 1 Social Jetlag CTA CTA 1 CTA 2 Figure 3. Effects of the CTA Schedule on Sleep and Circadian Misalignment (A) Self-reported sleep duration on workdays increases significantly for Early1 and Late2 in the CTA schedule, as compared to baseline assessments. (B) Free day sleep duration decreased in the CTA schedule, especially in extreme chronotypes, but this change was not significant. Sleep duration in Early2 was stable across time points. (C) Sleep quality improved significantly on workdays, again only in extreme chronotypes, while the control group Early2 did not show significant changes across schedules (higher values represent better sleep quality). (D) Social jetlag decreased most in the earlier chronotypes (Early1, D); Early2 showed a slight decrease of social jetlag when comparing baseline to CTA2. Late2, however, on average did not show changes in circadian misalignment. An intentionto-treat analysis comprising all participants irrespective of whether their chronotype was an ideal match with the intervention group, i.e., irrespective of congruence showed that the CTA schedule could reduce circadian misalignment by 1 hr (data not shown), which represents a meaningful change, as we have previously reported that 1 hr of social jetlag can increase the odds of being overweight [10]. frequency of given shifts within a schedule ([11]; for computation details, see Supplemental Experimental Procedures). The CTA schedule reduced participants social jetlag (2-2-2 schedule: 3 hr 17 min ± 20 min versus CTA2: 1 hr 52 min ± 1 hr 10 min; F(2,50) = 8.86, p < 0.001, partial h 2 = 0.29; Figure 3D). Compared to baseline, Early1 and Early2, but not Late2, showed less social jetlag at CTA2 ( 1 hr 55 min and 1 hr 28 min, respectively, F(4,50) = 2.85, p = 0.035, partial h 2 = 0.21). At CTA2, Early1 showed comparable levels to the day-working population [10]. Interestingly, SDs in Late2 increased from 15 min at baseline to 1 hr 55 min at CTA2, suggesting that social jetlag increased for some participants but decreased for others. Further analyses in this group independent of congruence (i.e., whether their chronotype was within the ideal chronotype range for Late2) showed that earlier chronotypes indeed had significantly higher levels of social jetlag at CTA1 (r = 0.64, p < 0.05, n = 11). This association was attenuated at CTA2, potentially due to a reduced sample size (r = 0.44, p = 0.19, n = 6). Sleep duration and quality were also associated with chronotype in Late2 (r values = , p < 0.05, n = 10 12), whereas wellbeing showed a significant correlation for CTA2 only (r CTA1 = 0.54, p = 0.08, n = 11; r CTA2 = 0.86, p = 0.003, n = 9; see Figures S1A S1D). Altogether, these findings suggest that working up to 14 night shifts per month can be beneficial, but only for very late chronotypes. Last, we used an intention-to-treat analysis approach [15] including all participants, irrespective of congruence between CTA group and individual chronotype and observed a total decrease of social jetlag by 1 hr 2 min (F(1.37,47.32) = 8.86, p = 0.002, partial h 2 = 0.21, Greenhouse- Geisser [GHG] corrected). Effects on Wellbeing, Satisfaction with Leisure Time, and Perceived Stress Compared to baseline, sleep log-derived wellbeing ratings on workdays increased in Early1 (+1.3 points, 0 10 scale) and Late2 (+0.76) but decreased slightly in Early2 ( 0.36) at CTA2 (Figure 4A; F(4,50) = 3.52, p = 0.013, partial h 2 = 0.22). Free day wellbeing ratings did not change significantly (p > 0.1). Forty-five participants answered the sub-section Your social and domestic situation of the standard shiftwork index (SSI; German translation, [13]) assessing satisfaction with the amount of leisure time (scale: 1, not at all, to 5, very much ). In general, satisfaction ratings decreased in the CTA schedule (CTA2 versus baseline, 0.24, Figure 4B; F(3,41) = 8.24, p < 0.001, partial h 2 = 0.38); however, patterns differed significantly across groups: ratings increased in Early1 (+0.48, scale from 0 to 5, versus CTA2) but decreased in Early2 ( 0.41), Late1 ( 0.44), and Late2 ( 0.59) (F(5.05,68.95) = 2.95, p = 0.012, partial h 2 = 0.17, GHG corrected). The increase observed in Early1 potentially reflects a gain in socially valuable time. Despite only small changes in work timing, both control groups were less satisfied, suggesting that other factors, such as the change in shift sequence (e.g., four consecutive evening shifts), may have interfered with leisure time. In Late2, satisfaction ratings declined most; it appears plausible that especially for younger employees, working at night on three out of four weekends per month might have significantly interfered with socially valuable time. Additionally, an increase in production demands during, but unrelated to, the CTA schedule led to an augmentation of weekly working hours on some weekends and might have further affected ratings [16]. Last, employees might not have fully Current Biology 25, , March 30, 2015 ª2015 Elsevier Ltd All rights reserved 909

5 A B Figure 4. Changes in Self-Reported Wellbeing and Satisfaction with Leisure Time across Shift Schedules (A and B) Wellbeing on workdays (A) changed significantly, dependent on shift group: extreme chronotypes (Early1 and Late2) felt better in the CTA schedule, as compared to baseline, while intermediate ones showed a slight decrease in wellbeing ratings. Satisfaction with time for social activities left by the schedule only increased in Early1, while it decreased in all three other groups (B). Better wellbeing and satisfaction ratings are represented by higher y axis values. adapted family and social life to the CTA schedule, as it was introduced for a 5-month period only, potentially further entailing discontent. The perceived stress scale (German translation, [12]) measures subjective stress appraisal in example situations (scale: 0, never, to 4, very often ), with higher sum scores indicating higher stress levels (maximum = 35). Compared to baseline, we observed a non-significant decrease in stress scores in all groups at CTA2 (total n = 33; Early1 = 4.58, Early2 = 0.22, Late1 = 3.0, Late2 = 4.59; all p values > 0.14). Limitations, Implications, and Conclusions Several limitations of our study are noteworthy: our sample is small and almost entirely male. Along with low statistical power and residual confounding inherent to field studies, this threatens the generalizability of the findings. However, in view of the consistency between self-reports and objective measures, the longterm recordings, and the concordance with predictions derived from cross-sectional studies, we believe our results constitute a proof of principle. Studies with larger samples are needed to understand to which degree our results can be extrapolated to the general population. Additionally, this study benefitted from a unique financial framework, as employees did not experience any financial losses (e.g., due to fewer night shift bonuses), which may have biased their judgment. Also, we could not assess direct healthrelated, physiological measures in this study; more readily accessible variables, such as absenteeism or sick days, were not useful in the context of a 6-month-long intervention study. Last, it is difficult to attribute the positive effects of the CTA schedule to the absence of the most strenuous shifts, as the shift schedule change concurrently altered other features of the schedule, such as rotation speed and shift sequence; Early1 and Late2, for example, experienced fewer shift changes as compared to their original schedule, which is thought to be advantageous for sleep and safety [17, 18]. Computational approaches may be a useful approach to disentangle the contribution of those highly nested factors. The positive effects of the CTA schedule on sleep measures further suggest that extremely late chronotypes better tolerate frequent night shift work than earlier ones; however, we propose that weekly hours of frequent or permanent night shift workers should be minimized to avoid social disruption. We also recommend to staff night shifts with as little personnel as possible while assuring adequate performance. In our study, production processes required four equally staffed groups, yet we observe a bell-shaped chronotype distribution in the population [5] and in this sample. If night shifts are mandatory, as in many industrial and medical settings, working time arrangements should critically examine the manpower needed. Shift schedules acknowledging circadian principles have been shown to improve satisfaction and subjective health more than three decades ago [19]. We took this a step further by personalizing working times according to chronotype. Our results indicate that such interventions can reduce circadian disruption and improve health-related outcomes, such as sleep [20]. Further studies are needed to extrapolate potential long-term effects of CTA schedules on health, social life, and economic indicators. Our results underline that shift-associated circadian disruption depends on individual internal time. SUPPLEMENTAL INFORMATION Supplemental Information includes Supplemental Experimental Procedures, one figure, and three tables and can be found with this article online at AUTHOR CONTRIBUTIONS C.V. and T.R came up with the study concept and design. C.V., D.F., and J.L.M. collected data. C.V., D.F., J.L.M., and T.R. analyzed and interpreted the data. C.V. drafted the manuscript. D.F., J.L.M., and T.R. were responsible for critical revisions and intellectual contributions. ACKNOWLEDGMENTS This research was funded by ThyssenKrupp Electrical Steel Europe and a scholarship of the Hanns-Seidel-Stiftung (D.F.). First of all, we would like to thank the participants of this study. Numerous people were involved in organizing and implementing this study, in particular Dr. Jörg Augustin, Jens Roßkothen, Frank Patschkowski, Dirk Sievers, Simone Smaga, Thomas Schulte-Strathaus (ThyssenKrupp Electrical Steel), and Klaus Rüttiger (ThyssenKrupp Steel). Special thanks go to Gabriele Konetzka and Dirk Stahlschmidt, ThyssenKrupp Electrical Steel. Received: October 8, 2014 Revised: December 19, 2014 Accepted: January 27, 2015 Published: March 12, 2015 REFERENCES 1. Wright, K.P., Jr., Bogan, R.K., and Wyatt, J.K. (2013). Shift work and the assessment and management of shift work disorder (SWD). Sleep Med. Rev. 17, Current Biology 25, , March 30, 2015 ª2015 Elsevier Ltd All rights reserved

6 2. Stevens, R.G., Hansen, J., Costa, G., Haus, E., Kauppinen, T., Aronson, K.J., Castaño-Vinyals, G., Davis, S., Frings-Dresen, M.H.W., Fritschi, L., et al. (2011). Considerations of circadian impact for defining shift work in cancer studies: IARC Working Group Report. Occup. Environ. Med. 68, Rüger, M., and Scheer, F.A.J.L. (2009). Effects of circadian disruption on the cardiometabolic system. Rev. Endocr. Metab. Disord. 10, Puttonen, S., Härmä, M., and Hublin, C. (2010). Shift work and cardiovascular disease - pathways from circadian stress to morbidity. Scand. J. Work Environ. Health 36, Roenneberg, T., Kuehnle, T., Juda, M., Kantermann, T., Allebrandt, K., Gordijn, M., and Merrow, M. (2007). Epidemiology of the human circadian clock. Sleep Med. Rev. 11, Roenneberg, T. (2012). What is chronotype? Sleep Biol. Rhythms 10, Roenneberg, T., Wirz-Justice, A., and Merrow, M. (2003). Life between clocks: daily temporal patterns of human chronotypes. J. Biol. Rhythms 18, Juda, M., Vetter, C., and Roenneberg, T. (2013). Chronotype modulates sleep duration, sleep quality, and social jetlag in shift-workers. J. Biol. Rhythms 28, Wittmann, M., Dinich, J., Merrow, M., and Roenneberg, T. (2006). Social jetlag: misalignment of biological and social time. Chronobiol. Int. 23, Roenneberg, T., Allebrandt, K.V., Merrow, M., and Vetter, C. (2012). Social jetlag and obesity. Curr. Biol. 22, Juda, M., Vetter, C., and Roenneberg, T. (2013). The Munich chronotype questionnaire for shift-workers (MCTQShift). J. Biol. Rhythms 28, Cohen, S., Kamarck, T., and Mermelstein, R. (1983). A global measure of perceived stress. J. Health Soc. Behav. 24, Barton, J., Spelten, E., Totterdell, P., Smith, L., Folkard, S., and Costa, G. (1995). The standard shiftwork index: a battery of questionnaires for assessing shiftwork-related problems. Work Stress 9, Barger, L.K., Lockley, S.W., Rajaratnam, S.M., and Landrigan, C.P. (2009). Neurobehavioral, health, and safety consequences associated with shift work in safety-sensitive professions. Curr. Neurol. Neurosci. Rep. 9, Hollis, S., and Campbell, F. (1999). What is meant by intention to treat analysis? Survey of published randomised controlled trials. BMJ 319, International Labour Organization (2011). Working time in the twentyfirst century. Report for Discussion at the Tripartite Meeting of Experts on Working-Time Arrangements, October 17 21, org/wcmsp5/groups/public/ ed_protect/ protrav/ travail/documents/ publication/wcms_ pdf. 17. Pallesen, S., Bjorvatn, B., Magerøy, N., Saksvik, I.B., Waage, S., and Moen, B.E. (2010). Measures to counteract the negative effects of night work. Scand. J. Work Environ. Health 36, Driscoll, T.R., Grunstein, R.R., and Rogers, N.L. (2007). A systematic review of the neurobehavioural and physiological effects of shiftwork systems. Sleep Med. Rev. 11, Czeisler, C.A., Moore-Ede, M.C., and Coleman, R.H. (1982). Rotating shift work schedules that disrupt sleep are improved by applying circadian principles. Science 217, Luyster, F.S., Strollo, P.J., Jr., Zee, P.C., and Walsh, J.K.; Boards of Directors of the American Academy of Sleep Medicine and the Sleep Research Society (2012). Sleep: a health imperative. Sleep 35, Current Biology 25, , March 30, 2015 ª2015 Elsevier Ltd All rights reserved 911

7 Current Biology Supplemental Information Aligning Work and Circadian Time in Shift Workers Improves Sleep and Reduces Circadian Disruption Céline Vetter, Dorothee Fischer, Joana L. Matera, and Till Roenneberg

8 SUPPLEMENTAL INFORMATION 1. Supplemental Figures 1.1. Correlation analyses in Late2 A Social Jetlag (NS, h) CTA 1 CTA 2 B Sleep Duration (NS, h) Chronotype (MSF E sc, hh:min) Chronotype (MSF E sc, hh:min) C Sleep Quality (NS, scale 0-10, a.u.) Chronotype (MSF E sc, hh:min) CTA 1 CTA 2 D 10 Wellbeing (NS, scale 0-10, a.u.) Chronotype (MSF E sc, hh:min) Figure S1 Examination of chronotype-dependent effects in Late2 (relates to Fig. 3A- C and 4A). Social jetlag (A), sleep duration (B), sleep quality (C) and wellbeing (D) on night shifts (NS) within Late2 correlate with chronotype. a.u. arbitrary unit. Continuous lines represent regression lines for CTA1, discontinuous ones CTA2. 2. Supplemental Tables 2.1. Subsamples and instrument-specific group characteristics Table S1 provides an overview of instrument-specific sample characteristics by group of the CTA schedule. Participation in all study instruments was self-selected, i.e., employees could chose which parts of the study they wanted to participate in, resulting in instrument-specific sub-samples. The MCTQ Shift [S1] was used to assess

9 chronotype in employees, while Standard Shiftwork Index [SSI, S2] was used to assess satisfaction with the amount of time left by the shift schedule, and the Perceived Stress Scale [PSS, S3] provided a global measure for stress. Table S1. Age, Body Mass Index and Chronotype (MSF E sc, assessed by the MCTQ Shift ) are given as mean±standard deviation; the range is indicated in brackets. Two sub-samples relied on questionnaire data (Standard Shiftwork Index, SSI, and Perceived Stress Scale, PSS). Please note that the qualitative questionnaire did not collect of personal information, thus no demographic data are available. M=male, n=sample size, BMI=Body Mass Index. Children = children living in the same household. Validating self reports samples= validating 1) the chronotype parameter with its analogues derived from actimetry and sleep-logs, and 2) validating self-reported sleep-log data with objective actimetry recordings. Evaluation samples = samples used to probe the effects of the CTA schedule across time points (Baseline vs. CTA1 vs. CTA2). * % with partner. Sample n Age (in yrs) Sex (% m) BMI Chronotype Family Status* Children (%, 1) MCTQ Shift ±9.8 (21-57) ±4.9 ( ) 4:18±1:02 (2:26-7:34) Validating self-reports Sleep-logs ± ±5.0 4:11±0: (21-57) ( ) (2:35-6:37) Actimetry ±8.3 (21-49) ±5.6 ( ) 4:04±0:52 (2:58-5:58) Evaluating the effects of the CTA schedule Sleep-logs ±10.6 (22-57) SSI ±11.2 (21-57) PSS ±10.2 (23-57) ±5.6 ( ) ±5.3 ( ) ±5.7 ( ) 3:58±1:01 (2:35-6:37) 4:10±0:59 (2:35-6:37) 4:08±1:04 (2:35-6:37) Descriptives of the variables Table S2 summarizes descriptive information of the variables used to probe the effects of the CTA schedule, namely sleep duration, social jetlag, sleep quality, wellbeing, satisfaction with the amount of free time in the schedule (SSI) and perceived stress (PSS), as a function of group affiliation, instrument, and work and free days if appropriate. We report values for the sleep log sample (n=28) that was used for repeated measures ANOVAs.

10 Table S2. Means (M) and standard deviations (SD) of all dependent variables CTA 1 CTA 2 Sleep duration Work Days Early ± ± ±0.45 Early ± ± ±0.43 Late2 6.87± ± ±0.79 Free Days Early1 8.26± ± ±0.84 Early2 8.51± ± ±0.81 Late2 9.13± ± ±0.98 Social jetlag Early1 3.41± ± ±0.34 Early2 3.40± ± ±1.38 Late2 2.65± ± ±1.91 Total congruent 3.28± ± ±1.16 Total all 3.59± ± ±1.89 Sleep quality Work Days Early1 5.92± ± ±1.71 Early2 6.76± ± ±1.77 Late2 6.40± ± ±1.72 Free Days Early1 6.72± ± ±1.57 Early2 6.84± ± ±2.10 Late2 6.66± ± ±2.12 Wellbeing Work Days Early1 5.82± ± ±1.66 Early2 6.72± ± ±1.53 Late2 6.44± ± ±1.69 Free Days Early1 6.64± ± ±1.57 Early2 6.93± ± ±1.68 Late2 6.85± ± ±1.60 SSI Early1 3.29± ± ±0.95 Early2 3.59± ± ±0.54 Late1 3.23± ± ±0.74 Late2 3.33± ± ±0.83

11 2-2-2 CTA 1 CTA 2 PSS Early ± ± ±9.79 Early ± ± ±8.66 Late ± ± ±5.26 Late ± ± ± Statistical indices of the omnibus tests We evaluated the effects of the CTA schedule by analyzing sleep duration and quality, social jetlag, as a measure of circadian misalignment, as well as wellbeing, satisfaction with social life, based on answers to a sub-scale of the Standard Shiftwork Index, as well as subjective stress perception with the Perceived Stress Scale. Table S3 gives an overview of statistical indices and effect sizes. Table S3. Overview of statistical indices for all dependent variable. Repeated measures ANOVAs were conducted with shift schedule and shift group as independent factors. Statistical significance (P<0.05) is indicated as follows: * <0.05, ** <0.01, *** < All P-values are indicated two-sided. ME=main effect, INT=interaction effect, n=sample size, GHG=Greenhouse-Geisser correction, SSI=Standard Shiftwork Index; PSS=Perceived Stress Scale. MS: Morning Shift, ES: Evening Shift, NS: Night Shift. # Dependent Variable Effect F-Value (df) P-value (2-sided) Effect size (partial η 2 ) n GHG 1 Sleep Duration A. Work Days ME shift schedule 3.84 (2,50) 0.028* ME shift group 7.40 (2,25) 0.003** 0.37 INT 1.65 (4,50) B. MS Early1: ME shift 1.87 (2,24) schedule C. ES Early1: ME shift 1.44 (2,24) schedule Late2: ME shift 0.96 (2,10) schedule D. NS Late2: ME shift schedule 0.43 (2,10) E. Free Days ME shift schedule 2.94 (2,50) ME shift group 3.10 (2,25)

12 # Dependent Variable Effect F-Value (df) P-value (2-sided) Effect size (partial η 2 ) INT 0.70 (4,50) n GHG 2 Alarm clocks A. Free days 3 Social Jetlag A. Congruent participants Early1: ME shift 0.69 (2,24) schedule Late2: ME shift schedule 0.16 (2,10) ME shift schedule 8.86 (2,50) 0.001** ME shift group 1.67 (2,25) INT 2.85 (4,50) 0.035* 0.21 B. All participants ME shift schedule 8.68 (1.37,47.32) 0.002** yes 4 Sleep Quality A. Work Days ME shift schedule 3.48 (2,50) 0.038* ME shift group 0.02 (2,25) INT 1.84 (4,50) B. Free Days ME shift schedule yes (1.44,27.32) ME shift group 0.17 (2,19) Wellbeing A. Work Days INT 0.63 (2.87,27.32) ME shift schedule 2.99 (2,50) ME shift group 0.06 (2,25) INT 3.52 (4,50) 0.013* 0.22 B. Free Days ME shift schedule yes (1.53,33.61) ME shift group 0.23 (2,22) INT 2.05 (3.06,33.61) 6 SSI ME shift schedule 2.08 (1.68,68.95) yes ME shift group 8.24 (3,41) <0.0001*** 0.38 INT 2.95 (5.05, 68.95) 0.012* PSS ME shift schedule 2.02 (2,58) ME shift group 1.47 (3,31) INT 0.45 (6,58)

13 3. Supplemental Experimental Procedures Recruitment and exclusion criteria All potential participants (N=116) were informed about the study in groups by the lead author of the study via presentations and question and answer sessions in morning, evening and night shifts. Participants could then indicate their willingness to participate and CV, DF and JLM then finalized recruitment in the week before baseline assessments started. Participants could choose the instruments they were willing to answer and end their participation any time. To assess the effects of the CTA schedule, we post-hoc excluded participants for the following reasons: Munich ChronoType Questionnaire for shift workers (MCTQ Shift [S1]). Out of the 116 employees, 114 filled out the MCTQ Shift prior to the study period. Two employees did not answer the MCTQ Shift because of permanent sick leave or lack of interest. Sleep-logs. Out of all employees, we recruited 58 employees to fill out sleep-logs. Six participants (3.5%) dropped out during the study. We also excluded participants from data analyses 1) who couldn t fill out the sleep-log at all three time points due to illness or vacation (n=10; given that repeated-measures ANOVA require a compound symmetry of the covariance structure); 2) who showed deviating sleeping behavior (e.g., frequent insomnia independent from type of shift, assessed from the commentary field in the sleep-log, n=2); 3) incongruent participants (i.e. chronotype measure more than 15 min away from the group-specific cut-off, n=5) ; and finally 4) the shift group Late1, as only two participants were left for analyses after exclusion process; such small sample sizes do not allow for reliable statistical testing. Actimetry. Twenty-six participants agreed to wear actimeters. We then excluded 1) participants exhibiting abnormal sleeping behavior (n=1), and 2) incongruent

14 participants (n=4), and 3) participants who did not participate at all three time points (n=2). Given the extreme small final sample size (n=19) across all four shift groups, we used actimetry data for validation purposes only; we compared key parameters of the MCTQ Shift (n=19) and sleep-logs at baseline with actimetry recordings (see Fig. 1B and main manuscript). Standard Shiftwork Index [S2] and Perceived Stress Scale [S3]. Out of 116, 58 employees (59%) filled out the SSI and PSS at any of the three time points. The final samples comprised all congruent participants who filled out the questionnaires at all three points (baseline, CTA1 & CTA2; SSI: N=45; PSS: N=33) Statistical procedures All analyses were conducted with SPSS Statistics 21.0 (SPSS Inc.) and ChronOSX [S4]. We used repeated-measures ANOVAs to analyze the 3x3-mixed design. Note that questionnaire samples comprise all four shift groups (i.e. included Late1, excluded from sleep-log analyses, see above), resulting in a 3x4-mixed design. Greenhouse-Geisser correction was applied if the sphericity assumption was violated. Significant main effects were examined with Bonferroni-corrected post-hoc comparisons. Late2 data were analyzed with bivariate correlations. We used intraclass-correlations for validation of self-reports with sleep log and actimetry data and to ensure interrater-reliability of actimetry-based sleep-parameters (SOM, section II.3.). Given the small sample sizes, the observed power never exceeded the commonly desired value of 1-β=0.80. Before-after comparisons were mostly conducted with congruent data sets only.

15 3. 3. Actimetry recordings and interrater-reliability of actimetry-derived sleep parameters Loco-motor activity was continuously recorded with wrist-worn devices (Daqtix GmbH, Germany), sampled every second and stored in 30-sec intervals. Data were binned in 10-min intervals for analyses. If participants took off the device, they entered time and date into a protocol, for future data exclusion. We extracted the parameters mid-sleep on shift-specific (morning, evening and night shift, respectively) workdays (MSW) and free days (MSF), as well as sleep duration from actimetry data by applying a threshold analysis [1] to define sleep onset as well as sleep offset. The interrater-reliability of determining sleep onset and sleep offset was estimated by three raters for five randomly selected participants over a period of both, four weeks in the schedule and four weeks in the chronotype-adjusted (CTA) schedule. To calculate intra-class correlations (ICCs), we applied the characteristics single measure (comparisons of single values instead of average values of the raters) and absolute agreement (general differences between raters are still part of the residual variance in contrast to the out-partialization of this variance in the consistency-model), following the most conservative way of calculating interraterreliability. The ICC coefficients showed very high agreement (sleep onset: ICC=0.98, P<0.001; sleep offset: ICC=0.99, P<0.001) Data processing Sleep-logs. We excluded sick days as well as free days with unusual early awakening times using alarm clocks due to e.g. medical appointments, based on the participants indications in the sleep-log commentary field. For free-day analysis, only the first two free days after a workday were included in order to avoid a wash-out of shift-specific

16 effects. Based on sleep on- and offset, we computed the parameters mid-sleep on shift-specific work (MSW MS/ES/NS ) and free days (MSF MS/ES/NS ) as well as sleep duration for all types of free days and workdays. We computed shift specific and weighted average values regarding sleep duration, social jetlag, sleep quality and wellbeing for each type of shift (MS, ES, NS) and each type of following free day (free day after MS, ES, NS). Actimetry. Actimetry data was analyzed using ChronOSX [S4] extracting daily sleep on- and offset of the main sleep episode by using a threshold approach ([S1], for details on interrater-reliability of parameter extraction, see II.3.). Calculation of sleep parameters from actimetry-based sleep on- and offset was analogous to sleep-log data processing. Standard Shiftwork Index (SSI) [S2]. In accordance to the manual guidelines, we calculated average values for each participant for the subsection Your Social and Domestic Situation. Perceived Stress Scale (PSS) [S3]. Sum values were computed across all items for each participant, resulting in a score ranging from 0 to 56 (with higher values indicating higher levels of perceived stress).

17 4. Supplemental References S1. Juda, M., Vetter, C., and Roenneberg, T. (2013). The Munich ChronoType Questionnaire for Shift-workers (MCTQShift). J. Biol. Rhythms 28, S2. Barton, J., Spelten, E., Totterdell, P., Smith, L., Folkard, S., and Costa, G. (1995). The Standard Shiftwork Index: a battery of questionnaires for assessing shiftwork-related problems. Work Stress 9, S3. Cohen, S., Karmarck, T., and Mermelstein, R. (1983). A global measure of perceived stress. J. Health Soc. Behav. 24, S4. Roenneberg, T., and Taylor, W. (2000). Automated recordings of bioluminescence with special reference to the analysis of circadian ryhthms. Methods Enzymol. 305,