Shift Work, Chronotype, and Melatonin Rhythm in Nurses

This article is featured in Highlights of This Issue, p. 1093

Shift work is a common aspect of modern life, with nearly 15 million Americans regularly working non-standard shifts (1). Historically, key epidemiologic studies have identified an association between night-shift work and an increased risk of breast cancer, and the World Health Organization classified circadian desynchrony associated with shift work as a probable carcinogen in 2007; however, some epidemiologic studies have indicated no such association (2–4). Night-shift work also has been linked to greater long-term risks of obesity, diabetes, and cardiovascular disease (5–7). Recent studies have corroborated these shift work-chronic disease associations in the Nurses' Health Studies (8–10).

Melatonin is a pineal hormone with oncostatic, antioxidant, and anti-inflammatory properties; it is entrained to the 24-hour environmental light-dark cycle, and acts as a reliable marker of the circadian system (11). The natural rhythm of melatonin involves peak levels at night, as melatonin is acutely suppressed by light; thus, light at night causing melatonin suppression is one of the hypotheses put forth to explain associations between night-shift work and chronic disease risk (12). A growing number of epidemiologic studies have linked lower levels of melatonin with higher risks of breast cancer, diabetes, and cardiovascular disease (13–16). Other possible mechanisms include sleep, activity, and/or circadian disruption either in conjunction with, or independent of, melatonin suppression (17, 18).

Most (19–27), but not all (28, 29), epidemiologic studies have suggested that night-shift work is associated with decreased melatonin secretion. However, few studies have explored the potential modulating effects of chronotype, which refers to an individual's diurnal preference, and these studies have produced mixed results (25–27). Thus, we evaluated the effects of shift work type, chronotype, and their joint effects on 24-hour light exposure and urinary melatonin rhythms in a sub-study of active nurses participating in the Nurses' Health Study II.

The Nurses' Health Study (NHS) II began in 1989, when 116,434 female nurses (ages 25 to 42 years) returned a mailed questionnaire about their health, lifestyle, and medical history (30); thereafter, similar questionnaires were sent to participants every two years. Rotating night-shift work was assessed on the baseline questionnaire and updated on each subsequent questionnaire, where it was defined as at least three night shifts per month in addition to day and evening shifts in that month. Chronotype was assessed using question 19 on the Morningness-Eveningness Questionnaire (MEQ; ref. 31): “One hears of morning and evening types of people. Which one of these types do you consider yourself to be?” Five responses were possible: definitely a morning type, rather more a morning than evening type, rather more an evening than a morning type, definitely an evening type, and neither. The present study focused on 148 NHS II participants who were involved in a validation study of the Harvard Light Exposure Assessment (H-LEA) Questionnaire, which included real-life light exposure measurements using a head-mounted light meter device called the Daysimeter. Detailed information regarding inclusion/exclusion criteria and subject recruitment have been published elsewhere (32).

Briefly, 18,044 NHS II participants met inclusion criteria for the H-LEA validation study in 2005 (20% of women who were still alive and participating in the cohort; ref. 32). These women were eligible if they: (i) had answered all shift work questions on the cohort questionnaires; (ii) were in the workforce; (iii) had no history of cancer (except nonmelanoma skin cancer) or cardiovascular disease; (iv) were not taking hormonal preparations; and (v) were not pregnant or lactating. On the basis of random sampling, two waves of participants were invited to participate in the validation study between September 2006 and April 2008 (both waves were recruited at similar times of the year). This included an initial group of 720 women (equal numbers of women with no rotating-shift work history, women in current rotating-shift work with <15 years of prior rotating-shift work history, and women in current rotating-shift work with ≥15 years of prior rotating-shift work) and an additional group of 242 current rotating-shift workers invited because most initial enrollees were day-shift workers. The second wave of participants were current rotating-shift workers with ≥15 months of night work between 2005 and 2007; they met the original inclusion criteria in 2005, and continued to meet these criteria in 2007.

Recruitment and enrollment were stopped once the target enrollment of approximately 150 participants was achieved, which resulted in a total enrollment of 148 women (32).

The degree of melatonin suppression induced by light exposure is wavelength dependent, with shorter wavelengths (visible blue light of ∼460–480 nm) inducing maximal suppression (33). The Daysimeter—a small, self-contained, battery-operated headset—was designed specifically for measuring short-wavelength light. This device contained two optical sensors, which were placed near the plane of a participant's cornea while being worn. The first sensor matched closely with the standard photopic luminous efficiency function (λ_max = 555 nm) and the second sensor responded only to light wavelengths shorter than 570 nm, with a λ_max = 470 nm. Light was measured together at regular, 30-second time intervals and stored electronically over a seven-day period. Mean photopic illuminance values (in lux) were derived from the first sensor, and mean spectrally weighted illuminance values (as a measure of circadian light) were derived from the second sensor. Technical details related to the device and circadian light model have been published previously (33, 34).

The device also had an accelerometer and a thermometer, and data from these sensors were used to assess participants' compliance with wearing the device.

Nurses were instructed to collect urine samples for three sequential days following at least three days without night-shift work; this approach was designed to assess each participant's baseline urinary melatonin rhythm. If feasible, rotating-shift workers were asked to start the collection on a day shift followed by a night shift, and day workers were asked to collect their samples on three day shifts (or days off). Women collected each spontaneous urine void in plastic containers for 72 hours, starting after their first morning void on the first day of collection. Women merged all voids within five predetermined time periods (7:00–10:59, 11:00–14:59, 15:00–18:59, 19:00–22:59, and 23:00–6:59), and reported the total urine volume for each period. The timing of the sample collections were not recorded and therefore all samples provided within each fixed time window were assigned that arbitrary time (the midpoint of the collection time period) and duration in the analysis. A 4.5-ml sample from each period was either refrigerated or frozen (the rest was discarded), resulting in 15 samples per participant. At the end of the 72-hour collection period, nurses sent all samples under cool conditions via overnight courier to Channing Laboratory (Boston, MA), where samples were immediately aliquoted into 1.5-mL vials and frozen at −80°F until shipment to the laboratory for analysis.

Urinary 6-sulfatoxymelatonin (aMT6s) concentration was assessed in urine samples using commercially available ELISA kits (Immuno-Biological Laboratories International, Hamburg, Germany). Urinary creatinine was assessed by buffered kinetic Jaffe reaction using the COBAS Integra 400 (Roche Diagnostics, Indianapolis, Indiana). Urine samples were analyzed in two batches, with intra-batch coefficients of variation of 12.6% and 14.3% for urinary aMT6s and 5.2% and 7.0% for urinary creatinine, respectively.

Of 148 nurse-participants in the validation study, 16 were excluded from analysis due to technical issues or noncompliance with light meter recordings (defined as extended episodes of inactivity based on temperature sensor and accelerometer data). Two additional participants were excluded because they did not provide urine samples, leaving 130 women (46 day-shift workers and 84 rotating-shift workers) with complete data for these analyses.

Urinary aMT6s levels below the detection limit of the aMT6s assay were set to this limit (0.4 ng/mL). We calculated mean urinary aMT6s concentrations for each of the five predetermined urine collection time periods. We standardized aMT6s measurements (ng/mL) by creatinine level (mg/mL) of each urine sample to account for differences arising from variations in urine concentrations. We calculated total 24-hour aMT6s production by summing the total aMT6s production for each void (ng/mL × total mL) and expressing it in ng/24 hours.

This study evaluated day-shift workers and current rotating-shift workers over three days. Day-shift workers might be day shift or off shift during this period, and current rotating-shift workers might be night shift, day shift, or off shift. Given that light exposure and melatonin patterns were similar for day versus off shifts, these data were analyzed together within each shift work type. Therefore, we had three main groups for comparison: day-shift workers, rotating-shift workers on day/off shifts, and rotating-shift workers on night shifts.

In statistical models, we evaluated shift work categories in relation to light exposure and urinary aMT6s levels using mixed models for repeated measurements with time as a categorical variable. This method takes into account intra-individual and inter-individual variation of the repeated measures of the outcome variable. Covariance structures were selected on the basis of model fit statistics and visualization of changes in covariance and correlation among residuals on the same individual over lag times.

To evaluate 24-hour light exposure and urinary melatonin rhythms, we used harmonic mixed models and cosinor analysis (35), and estimated circadian mean (mesor), difference between the peak and circadian mean (amplitude), and time of peak onset (acrophase) for urinary melatonin across categories of shift work and chronotype, and across categories of chronotype stratifying by shift work. (As described in the Introduction, melatonin is acutely suppressed by light, and therefore its natural rhythm involves peak levels at night). A multiplicative interaction term was added to the model for shift work and chronotype. Initial models were adjusted for age, and additional models were also adjusted for body-mass index, melatonin batch, region (time zone), and month of melatonin collection. Mutual adjustment for shift work and chronotype was considered, but did not change model estimates materially and was not included in final models.

In all analyses, we log-transformed light exposure and urinary aMT6s levels and reported their geometric means and 95% confidence intervals (CI). Statistical tests were two sided and Bonferroni correction was applied as indicated to adjust for multiple testing and to determine the threshold for statistical significance. All analyses were conducted using SAS version 9.4 (SAS Institute).

In this randomly selected sample of NHS II participants, women had characteristics that were largely similar to the overall NHS II cohort (Table 1). Within this sample, rotating-shift workers were slightly younger and heavier, on average, compared with day-shift workers. In addition, a higher percentage of rotating-shift workers exercised multiple times weekly, and reported use of sleeping pills and NSAIDs. As expected, rotating-shift workers were more likely to have a short-term and long-term history of rotating-shift work compared with day-shift workers.

The spectral responses of the Daysimeter's first and second sensors were highly correlated, and results were similar when we evaluated measurements from the first versus second sensor. Therefore, we present only the results for the photopic sensor in lux.

Overall, day-shift workers recorded significantly greater light exposure over 24 hours compared with rotating-shift workers, and rotating-shift workers on nights shifts recorded significantly greater 24-hour average light exposure compared with rotating-shift workers on off shifts (Table 2). When light exposure was examined during five pre-determined time periods across the 24-hour day, day-shift workers and rotating-shift workers on day/off shifts generally recorded greater light exposure during the day compared with the night; however, light exposure appeared to be more similar during the day and night for rotating-shift workers on night shifts. Across shift work categories, day-shift workers had significantly higher recorded light exposure during the day (7:00 to 18:59) compared with rotating-shift workers on night shifts, with intermediate light exposure recorded by rotating-shift workers on day/off shifts. During the night (23:00–6:59), rotating-shift workers on night shifts had significantly greater light exposure compared with day-shift workers, and rotating-shift workers on day/off shifts had significantly lower light exposure compared with the other groups (Table 2). When the number of hours workers were exposed to >20 lux light was compared across shift worker categories, day-shift workers and rotating-shift workers on off shifts generally recorded significantly longer exposure during the day (7:00 to 18:59), whereas rotating-shift workers on night shifts had significantly longer exposure at night, particularly from 23:00 to 6:59 (Table 2).

Total urinary melatonin concentrations calculated across 24 hours were similar across shift work categories (Table 2). However, these concentrations varied over the course of the day: they were highest during the night (23:00 to 6:59), decreasing in the morning hours (7:00 to 10:59), and were lowest during the day (11:00 to 22:59) across all shift work categories. Notably, melatonin concentrations were significantly lower at night (23:00 to 6:59) in rotating-shift workers on night shifts compared with both day-shift workers and rotating-shift workers on day/off shifts.

When urinary melatonin circadian rhythm parameters were evaluated (Table 3), rotating-shift workers on night shifts had a significantly later peak onset [multivariable-adjusted geometric means: 5.71 hours (95% CI, 4.76–6.85) vs. 4.10 hours (95% CI, 3.37–4.99)] and smaller peaks (11.81; 95% CI, 9.49–14.71 vs.14.83; 95% CI, 11.72–18.75) compared with day-shift workers, and significantly smaller mean levels (multivariable-adjusted geometric means: 16.51; 95% CI, 13.97–19.50 vs. 20.60; 95% CI, 17.33–24.48) and smaller peaks (11.81; 95% CI, 9.49–14.71 vs. 16.26; 95% CI, 12.95–20.40) compared with rotating-shift workers on day/off shifts. Across chronotypes, analyses of urinary melatonin rhythms identified significant differences in time of peak onset only, such that intermediate types and evening types had later peak onsets compared with morning types [multivariable-adjusted geometric means: 5.24 hours (95% CI, 4.48–6.11) and 4.90 hours (95% CI, 3.94–6.09) vs. 3.64 hours (95% CI, 2.99–4.43), respectively]. However, there was a significant interaction between shift work category and chronotype for all circadian parameters related to urinary melatonin (all P_interaction < 0.05). Curves for light exposure and urinary melatonin across shift work and chronotype categories are depicted in Fig. 1 (Fig. 1A, light exposure by shift work; Fig. 1B, urinary melatonin by shift work; Fig. 1C, light exposure by chronotype; Fig. 1D, urinary melatonin by chronotype).

Stratified analyses of chronotype and urinary melatonin parameters by shift work category showed notable differences (Table 4). Across all shift-work categories, there was an earlier peak for morning types compared with intermediate and evening types, with differences reaching statistical significance for the morning type versus intermediate type comparison among rotating-shift workers on day/off shifts and rotating-shift workers on night shifts [multivariable-adjusted geometric means: 3.48 hours (95% CI, 2.62–4.62) vs. 6.01 hours (95% CI, 4.88–7.42) for rotating-shift workers on day/off shifts; 3.56 hours (95% CI, 2.39–5.30) vs. 6.23 hours (95% CI, 4.69–8.26) for rotating-shift workers on night shifts]. For day-shift workers and rotating-shift workers on day/off shifts, there were qualitative trends of decreasing mean levels and peak sizes across morning, intermediate, and evening types; however, the only comparisons reaching statistical significance were for evening types compared with morning and intermediate types among day-shift workers [multivariable-adjusted geometric means: 9.76 (95% CI, 6.58–14.49) vs. 18.53 (95% CI, 13.32–25.78) and 14.62 (95% CI, 11.06–19.33), respectively]. Among rotating-shift workers on night shifts, the opposite qualitative trends were observed: Mean levels and peak sizes increased across categories of morning, intermediate, and evening types, although most comparisons did not reach statistical significance. Curves for urinary melatonin levels and light exposure across chronotype categories, stratified by shift work, are depicted in Fig. 2 (Fig. 2A–C, light exposure in day-shift workers, rotating-shift workers on day/off shifts, and rotating-shift workers on night shifts; Fig. 2D–F, urinary melatonin in day-shift workers, rotating shift-workers on day/off shifts, and rotating-shift workers on night shifts).

This study has two main findings. First, rotating-shift workers on night shifts had higher levels of light exposure and lower levels of urinary melatonin production during the night compared with day-shift workers. These results were generally consistent with previous studies that have measured light exposure and melatonin levels in night-shift workers, and may support the proposition that night-shift workers experience more nocturnal melatonin suppression due to higher levels of light exposure at night compared with day workers—one of the hypotheses underlying the potential association of night-shift work and an increased risk of breast cancer. Other known risk factors for breast cancer (e.g., age, family history of breast cancer, and multiple reproductive factors) are less likely to mediate the association between night-shift work and breast cancer because these factors do not appear to differ according to night-shift work status (10, 36).

Second, analyses that modeled 24-hour urinary melatonin rhythms indicated certain differences in circadian parameters across shift work and chronotype, which were largely consistent with prior literature describing these associations. Furthermore, a significant interaction of shift work and chronotype was identified for all parameters of the melatonin rhythm, suggesting that interpretation of these results should focus on the joint effects of these exposures. In particular, a distinct pattern of greater mean levels, larger peaks, and earlier peaks was observed for morning types compared with all other chronotypes among day-shift workers, and a similar pattern was apparent for evening types compared with the other chronotypes among rotating-shift workers on night shifts. Together, these results suggest that better alignment of shift-work type and chronotype may produce melatonin rhythms more appropriately synchronized with the imposed sleep-wake and dark-light schedules.

Three prior epidemiologic studies have examined joint effects of shift work and chronotype in relation to urinary melatonin (25–27). In Spain, a study of fixed-shift workers (75 night-shift workers, 42 day-shift workers) collected urine over 24 hours on a working day, and found that the circadian mean for urinary aMT6s was higher for evening compared with morning types among night-shift workers (27). The authors reported that the mean levels were highly correlated with peak size study (and therefore results were not presented separately for peak size), suggesting that these results were similar to ours, which identified a larger peak among evening compared with morning types among rotating-shift workers on night shifts. However, this study did not find differential effects of chronotype on circadian phase among night-shift workers. In the United States, another study conducted among fixed-shift workers (354 night-shift workers and 310 day-shift workers) evaluated urinary aMT6s levels during relevant time periods (i.e., during night sleep for both groups, and during day sleep and night work for night-shift workers), and identified that evening chronotypes had higher melatonin levels compared with morning chronotypes during the first day sleep following a night shift (25). The authors interpreted these results to mean that morning chronotypes were better able to maintain “normal” melatonin rhythms because their melatonin levels were closer to those of day-shift workers; however, such an interpretation does not acknowledge that, in their comparison, the early chronotypes were sleeping during their biological day, whereas the evening chronotypes were sleeping during their biological night (a point that we have previously addressed; ref. 37). If this point were acknowledged, the observed difference may indicate that evening types have more adaptability to night-shift work (i.e., higher levels of melatonin during their biological night), which is consistent with the notion that alignment of shift work type and chronotype is important for optimal melatonin rhythms (38).

Finally, in Canada, a study assessed urinary aMT6s rhythms based on 48 hours of urine collection, and compared data from 114 rotating-shift workers (including data on one night shift and one day shift) versus 147 fixed-day workers (26). The authors reported that, among rotating-shift workers, aMT6s curves of evening chronotypes had lower mean levels and earlier peak onset compared with morning chronotypes. These findings seem to contradict the results of our study, although it is difficult to compare results in detail because the definitions of rotating-shift work, as well as recent and long-term history of such work, are not entirely clear in the Canadian study. For example, participants in our study were instructed to begin urine collection following ≥3 days without night-shift work (to assess participants' baseline urinary melatonin rhythms), which was not a requirement in the Canadian study and could lead to less straightforward interpretation of results because their participants' urinary melatonin rhythms could have been influenced by carryover effects of night-shift work performed in the days preceding urine collection. More research is clearly needed to understand the joint effects of different night-shift work schedules and chronotype on 24-hour melatonin rhythms.

Limitations of note include that women who volunteered to participate in our study appeared more health conscious, as reflected in the lower rate of smoking and alcohol consumption among rotating-shift workers in the current study compared with the overall NHS II cohort participants (32). Further adjustment for these covariates did not change our results, however. Also, there was potential for noncompliance with wearing the light meter in our study; however, a thermometer and accelerometer built into the device allowed us to detect noncompliance patterns, and participants with such patterns were excluded from our analyses. Another limitation of note is that even though plasma aMT6s levels are suppressed within 15 minutes of plasma melatonin suppression, it is less clear how quickly these changes can also be detected in urine (39–43). Even though frequency and timing of urine samples will affect the time of measurable urinary effect of light, any delay is likely only in the order of one hour (44, 45); hence, we expect it to have had minimal and only non-differential impact on our data analyses. Finally, in this study, the rotating night-shift workers on off shifts were probably a mixed group of individuals with variable sleep/activity patterns on their days off; thus, it is difficult to interpret the data from this group and compare it with other groups of interest.

In summary, we found that rotating-shift workers on night shifts had greater light exposure and lower urinary melatonin levels during the night compared with day-shift workers, as expected. Moreover, our results provide strong evidence, based on in-field measurements, that circadian rhythm parameters for the major urinary melatonin metabolite are dependent on both chronotype and shift work type. Future studies of shift workers should focus on the joint effects of these parameters, as better alignment of shift-work type and chronotype may produce less disrupted melatonin rhythms.

P. Razavi is a consultant/advisory board member for Novartis and reports institutional research funding from Grail Inc. E.E. Devore is a consultant/advisory board member for Epi Excellence and Bohn Epidemiology. S.W. Lockley is an adjunct professor at Monash University, reports receiving a commercial research grant from F.Lux, has received speakers bureau honoraria from Ineos, Teague, BHP Billiton, has ownership interest (including stock, patents, etc.) in Brigham and Women's Hospital, and is a consultant/advisory board member for Apex2100, Consumer Sleep Solutions, Six Senses, Lighting Science Group, Mental Workout, Light Cognitive, Akili Interactive, Hintsa Performance, Headwaters, and Midwest Lighting Institute. E. Schernhammer is a consultant/advisory board member for Scientific Advisory Board for Houska Award. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S.W. Lockley, M.G. Figueiro, V. Ricchiuti, W.C. Willett, E.S. Schernhammer

Development of methodology: S.W. Lockley, M.G. Figueiro, V. Ricchiuti, E.S. Schernhammer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Razavi, A. Bajaj, E.S. Schernhammer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Razavi, E.E. Devore, S.W. Lockley, M.G. Figueiro, S.E. Hankinson, E.S. Schernhammer

Writing, review, and/or revision of the manuscript: P. Razavi, E.E. Devore, S.W. Lockley, M.G. Figueiro, W.J. Gauderman, S.E. Hankinson, W.C. Willett, E.S. Schernhammer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Razavi, A. Bajaj

Study supervision: E.S. Schernhammer

Other (run tests for the study): V. Ricchiuti

We would like to thank the Nurses' Health Study II participants for their dedication to this research, particularly those who provided additional information for this validation study. We would also like to thank Mark Rea, PhD (Lighting Research Center, Rensselaer Polytechnic Institute) for his insightful comments. This research was supported by the Cooperative Agreement Number R01 OH04192 from Centers for Disease Control and Prevention (CDC) and the National Institutes of Health (R01 OH008171, to E. Schernhammer). The Nurses' Health Study II infrastructure is supported by the National Cancer Institute (UM1 CA176726).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.