Article Text
Abstract
Objectives The main objective was to determine whether sleep duration on work shifts mediates the relationship between a current alternating day and night shift work schedule and metabolic syndrome among female hospital employees. The secondary objective was to assess whether cumulative lifetime shift work exposure was associated with metabolic syndrome.
Methods In this cross-sectional study of 294 female hospital employees, sleep duration was measured with the ActiGraph GT3X+. Shift work status was determined through self-report. Investigation of the total, direct and indirect effects between shift work, sleep duration on work shifts and metabolic syndrome was conducted using regression path analysis. Logistic regression was used to determine the association between cumulative shift work exposure and metabolic syndrome.
Results Shift work is strongly associated with metabolic syndrome (ORTotal=2.72, 95% CI 1.38 to 5.36), and the relationship is attenuated when work shift sleep duration is added to the model (ORDirect=1.18, 95% CI 0.49 to 2.89). Sleep duration is an important intermediate between shift work and metabolic syndrome (ORIndirect=2.25, 95% CI 1.27 to 4.26). Cumulative shift work exposure is not associated with metabolic syndrome in this population.
Conclusions Sleep duration mediates the association between a current alternating day–night shift work pattern and metabolic syndrome.
- Shift Work
- Sleep
- Epidemiology
- Cardiovascular
Statistics from Altmetric.com
What this paper adds
Research suggests that shift work may be associated with cardiovascular and metabolic diseases, but few studies have explored the mechanisms of this relationship.
Sleep disturbance is a common consequence of shift work and has been hypothesised as an intermediate in the shift work–metabolic syndrome relationship.
In a group of female hospital employees, women who work on alternating day and night shifts sleep less on work shifts and are more likely to have metabolic syndrome compared with those who work only on day shifts.
Work-related sleep duration mediates the association between shift work and metabolic syndrome.
Healthy workplace policies focused on increasing sleep duration among shift workers may help alleviate the negative cardiometabolic effects of shift work.
Introduction
Current research suggests shift work is a risk factor for numerous negative health outcomes, including cardiovascular and metabolic diseases.1–5 For example, recent data from the Nurses’ Health Study found that rotating shift work for 6–14 and ≥15 years was associated with a 19% and 23% increased risk of cardiovascular disease mortality, respectively,4 while a meta-analysis concluded that any exposure to night shift work increased the risk of metabolic syndrome by almost 60%.2
A number of potential pathways by which shift work affects cardiovascular health have been postulated, including social problems, behavioural changes and circadian disruption.6 It is likely that these hypothesised pathways linking shift work and cardiovascular disease should similarly affect metabolic syndrome occurrence because metabolic syndrome is a strong predictor of cardiovascular disease.7
Sleep disturbance is a common consequence of shift work8–10 and is on the circadian disruption pathway.6 Sleep duration is particularly affected by shift work, where two reviews concluded that early morning and night shifts are associated with short sleep duration.8 9 Reduced sleep duration may have negative cardiometabolic effects, as demonstrated in three recent meta-analyses that have identified short sleep duration as a risk factor for metabolic syndrome.11–13
In a systematic review that looked at the relationship between shift work and metabolic syndrome, it was suggested that sleep duration should be assessed as a potential mediating variable.3 To our knowledge, a previous study by our research group is the only study that has investigated whether sleep disturbance is an intermediate in the shift work–metabolic syndrome relationship.14 Our previous research did not investigate sleep duration specifically, but instead focused on other indices of sleep disturbance measured by the Pittsburgh Sleep Quality Index (PSQI), a self-report questionnaire in which average sleep patterns are recalled over the past 30 days.15 In that study, the potential mediators (sleep efficiency, sleep latency and the total PSQI score) were not identified as intermediates,14 although these results may be attributed to limitations surrounding the use of self-report measurements of sleep. Measuring sleep with objective methods such as actigraphy can provide more accurate and precise estimates of sleep–wake behaviours to better characterise shift work-related sleep disturbance.
The primary objective of this project was to assess whether sleep duration mediates the relationship between a current alternating day–night shift work schedule and metabolic syndrome among female hospital employees. We chose to focus on sleep duration on work shifts to best capture the effect of a work schedule on sleep. An additional objective was to assess whether cumulative lifetime shift work exposure is associated with metabolic syndrome. Mediation analysis with this additional shift work parameter is not warranted because work by our group in this study population did not find that cumulative shift work exposure was associated with current sleep duration.16
Methods
A cross-sectional study was conducted in an acute care teaching hospital in Kingston, Ontario, between 2011 and 2014. The target population included female shift workers and day workers (n=294), where shift work was defined as a rotating day and night schedule. Participants were volunteers recruited through advertisements in the hospital. Women self-excluded if they were pregnant, had given birth in the previous year or had less than 1 year of work history. Most participants were nurses, although women in research, laboratory, diagnostic and support services were also included. All participants provided informed consent, and the study was approved by the Health Sciences Research Ethics Board at Queen’s University.
Shift work exposure was determined through self-report. The routine schedule among current shift workers was two 12-hour day shifts, two 12-hour night shifts, and 4–5 off days, although women often traded shifts so this schedule varied. Current day workers typically worked five 8-hour days followed by 2 days off. Cumulative lifetime shift work exposure was estimated as full-time years+0.5*part-time years. For comparative purposes, women were classified as working ≥10 and <10 years of shift work, because a cut-off of 10 years has been linked to increased metabolic syndrome risk.2
Participants completed a physical exam and a detailed questionnaire to collect information on sociodemographic factors, lifestyle characteristics and working patterns. Physical activity was measured with the Global Physical Activity Questionnaire,17 life stress was measured with the Derogatis Stress Profile,18 emotional health was measured with the Center for Epidemiologic Studies Depression Scale Revised,19 and occupational stress was measured with the Job Content Questionnaire.20
Sleep was objectively monitored for 1 week with the ActiGraph GT3X+, a small (4.6 cm × 3.3 cm × 1.5 cm), lightweight (19 g) actigraph that contains a triaxial accelerometer. Accelerometers continuously measure movement, and sleep can be identified by lack of movement. During sleeping hours, the ActiGraph GT3X+ was worn on the non-dominant wrist attached with an elastic band. Data were collected at a sampling rate of 30 Hz, downloaded in 10 s epochs and were aggregated to 1 min epochs representing counts/min. Participants were also required to track their sleep timing using sleep logs, which were then used to aid the collation of actigraph data. Compared with polysomnography, the gold standard in sleep measurement, the accuracy and sensitivity of the ActiGraph GT3X+ (worn on the wrist) are high (>80%), while specificity is lower (approximately 45%–60%).21 22
ActiLife (V.6.5.3) software and the embedded Cole-Kripke algorithm23 were used to determine main sleep duration on work shifts, where main sleep refers to the single longest sleep episode in a 24-hour period. The Cole-Kripke algorithm scores each 1 min epoch as sleep or wake time, and sleep duration was defined as the total number of 1 min epochs scored as sleep during a main sleep episode (ie, from ‘lights on’ to ‘lights off’). ‘Lights off’ and ‘lights on’ times were manually entered into the software by referencing the sleep logs and through a visual inspection of the data.
Sleep was attributed to a day shift or night shift depending on the impact of the shift on sleep. A day shift influences the night-time sleep prior to the workday, so this was coded as a day shift sleep. Sleep after a night shift is influenced by the previous night of work so was coded as a night shift sleep. Mean work-related sleep duration was calculated using all available work shift main sleep periods. On average, this included 5 days for day workers and 4 days for shift workers.
Of the total 330 women recruited for the original study, 36 were excluded due to missing sleep data. A minimum of three work-related sleep periods were required to be included in this analysis, and among shift workers this must include at least one day and one night shift.
The metabolic syndrome was defined according to the 2009 Joint Interim Studies Consensus Statement.24 To be positive for the metabolic syndrome, at least three of the following five criteria must be fulfilled: high waist circumference (≥80 cm), elevated triglycerides (≥1.7 mmol/L), reduced high-density lipoprotein (HDL) cholesterol (<1.3 mmol/L), elevated blood pressure (systolic ≥130 and/or diastolic ≥85 mm Hg) and elevated fasting blood glucose (≥100 mg/dL).24 Participants reported whether they were currently taking medication for high blood pressure or cholesterol, and women responding positively were categorised as positive for these respective risk factors.
On enrolment, a trained research coordinator measured blood pressure while the participant was seated using the BpTRU blood pressure monitor (VSM MedTech, Coquitlam, Canada), and the mean of three consecutive readings (with 1 min intervals between readings) was used. A study nurse measured waist circumference at the midway point between the iliac crest and the lower rib. To measure triglycerides, HDL cholesterol and blood glucose, blood samples (after a nightly fast) were drawn by the study nurse and kept in a freezer until analysis at the Pathology and Molecular Medicine Laboratory at Kingston General Hospital using standardised laboratory techniques. Blood was drawn from day workers before or during a day shift, while blood was drawn from shift workers typically in the morning of a day off work.
Current shift workers and day workers were compared by sociodemographic factors, work patterns and lifestyle characteristics using Wilcoxon rank-sum tests for continuous variables and χ2 tests for categorical variables. The presence of metabolic syndrome (as well as its individual components) was compared between current shift workers and day workers, as well as between women who worked ≥10 and <10 years of shift work, using χ2 tests.
Mediation analysis was conducted using the PROCESS procedure for SAS V.9.4.25 A conceptual model to illustrate the proposed association between shift work (predictor variable X), sleep duration (mediator M) and the metabolic syndrome (dependent variable Y) is shown in figure 1. Using a standard regression analysis approach, we estimated (1) the total effect of shift work (X) on the metabolic syndrome (Y), unadjusted for group differences in sleep duration (M) — path c; (2) the direct effect of shift work (X) on the metabolic syndrome (Y), holding the effect of sleep duration (M) constant — path c’; (3) the relationship between shift work (X) and the proposed mediator, sleep duration (M) — path a; and (4) the association between sleep duration (M) and the metabolic syndrome (Y), holding the independent variable shift work (X) constant — path b. Three statistical equations can describe the regression models used to estimate all pathways:
Y= B0+cX + confounders
M=B0+aX + confounders
Y=B0+c’X+bM + confounders
Logistic regression was used for models 1 and 3 because metabolic syndrome is a dichotomous outcome, while ordinary least squares regression was used for model 2 because sleep duration is continuous. The PROCESS command is a flexible procedure that allows for a combination of linear and logistic regression.25 The indirect (mediating) effect was calculated as the product of coefficients ab and was estimated from 10 000 bootstrap samples. To test the significance of the indirect effect (ie, whether sleep duration is a significant mediator), bias-corrected 95% CIs around the indirect effect were constructed. The indirect effect is considered statistically significant if the CI does not include the null value. Bootstrapping is a recommended approach to testing the significance of the indirect effect because it does not assume normality in the sampling distribution.26 27
The primary confounders of interest in the mediation analysis were age, menopausal status, number of children, marital status and education because these variables may be related to shift work, sleep duration and metabolic syndrome, but are unlikely to be on the causal pathway in any of these relationships. Before conducting the PROCESS procedure, each individual regression analysis was run to select empirical confounders using the 10% change-in-estimate approach.28 Additional variables including smoking, caffeine use, alcohol consumption, physical activity, stress and emotional health are potential intermediates between shift work and metabolic syndrome and would dilute the association if erroneously adjusted for. However, due to some uncertainty as to whether these variables are true intermediates and to ensure we did not exclude important confounders, the relationship between these variables and metabolic syndrome was determined using Wilcoxon rank-sum tests for continuous variables and χ2 or Fisher’s tests for categorical variables. If any variable was found to be strongly related to metabolic syndrome, a sensitivity analysis was to be conducted to adjust for the additional variable(s). Obesity and/or being overweight were not considered as confounders due to the high degree of correlation to waist circumference, a component of the outcome metabolic syndrome.
A similar change-in-estimate approach was used to select confounders to include in the multivariable model assessing the relationships between cumulative shift work exposure and the metabolic syndrome. The same subset of potential confounders was considered, but current shift work status was also included as a possible confounder. All statistical analyses were conducted using SAS V.9.4.
Results
Comparisons between current shift workers and day workers by sociodemographic, lifestyle and work characteristics are seen in table 1. Shift workers are younger, have fewer children, are more likely to have an undergraduate or graduate education, are less likely to smoke, have less life stress, greater psychological job demands, less social support from coworkers and supervisors, and more job security. Shift workers sleep less than day workers on work shifts: the main sleep duration of shift workers is 332 min (5 hours 32 min), while the main sleep duration of day workers is 404 min (6 hours 44 min). Shift workers are also more likely to be nurses and have greater cumulative shift work exposure.
Comparisons between shift work exposure parameters for the presence of the metabolic syndrome and its individual components are shown in table 2. In terms of the components of the metabolic syndrome, there is a trend of higher blood glucose and blood pressure among current shift workers compared with day workers, but these differences are not statistically significant. Waist circumference of current shift workers and day workers is very similar, while shift workers are more likely to have elevated triglyceride levels and low HDL cholesterol. Current shift workers are also more likely to have metabolic syndrome than day workers (20% vs 12%), corresponding to a crude OR of 1.91 (95% CI 1.00 to 3.62). After adjusting for age (the only empirical confounder), the relationship is stronger (OR=2.72, 95% CI 1.38 to 5.36). This adjusted OR reflects the total effect of shift work on metabolic syndrome.
Women with ≥10 years of shift work history are more likely to have elevated blood pressure, high blood glucose and metabolic syndrome than those with <10 years of shift work history. Although the crude OR demonstrates an almost threefold higher risk (OR=2.97, 95% CI 1.55 to 5.69), adjustment for age, menopausal status and current shift work status (confounders which changed the coefficient by ≥10%) attenuates the association between cumulative shift work and metabolic syndrome (OR=1.60, 95% CI 0.69 to 3.74).
Table 3 presents the results of the mediating role of work-related main sleep duration in the shift work–metabolic syndrome relationship. After controlling for age, shift work is related to work shift sleep duration (path a), where current shift workers sleep 74 min (1 hour 14 min) fewer than day workers. After controlling for current shift work status and age, each additional minute of sleep duration decreases the odds of the metabolic syndrome (path b; OR=0.99, 95% CI 0.98 to 0.99). This corresponds to an OR of 0.52 (95% CI 0.33 to 0.82) per each additional hour of sleep. Controlling for sleep duration (in addition to age) attenuates the relationship between shift work and the metabolic syndrome (ORDirect=1.18, 95% CI 0.49 to 2.89). The CI for the indirect effect of work shift sleep duration on the relationship between shift work and the metabolic syndrome excludes the null value of 1 (path ab; OR=2.25, 95% CI 1.27 to 4.26), indicating that sleep duration on work shifts is an important mediator. Because certain variables we hypothesised to be intermediates but could arguably instead be confounders (eg, physical activity, smoking status and so on) were not associated with metabolic syndrome (online supplementary table 1) and therefore do not meet the basic criteria of confounding, a sensitivity analysis adjusting for these variables was not warranted.
Supplementary file 1
Discussion
We demonstrated that sleep duration on rotating (day and night) work shifts mediated the relationship between a current shift work schedule and metabolic syndrome in female hospital employees. Current shift workers slept less than day workers and were more likely to have metabolic syndrome. Although we found that some individual components and the overall presence of the metabolic syndrome were more common in women with ≥10 years compared with those with <10 years of shift work history, cumulative shift work exposure was no longer associated with metabolic syndrome after confounder adjustment.
A strong association between shift work and metabolic syndrome was found. This effect was larger than that reported in a recent meta-analysis where shift work elevated the risk of metabolic syndrome by approximately 60%.2 Differences may be attributed to inconsistent definitions of shift work among studies included in the meta-analysis (eg, ‘ever exposed’ to night shift work, night work ≥1 night/week and so on), different diagnostic criteria for the metabolic syndrome, different confounders and the diversity of study populations.
Although the risk of metabolic syndrome was elevated among women who worked ≥10 years of shift work, the CI was wide and crossed the null value of 1 after confounder adjustment. This finding is in contrast to results from a meta-analysis that reported 10 or more years of shift work increased the risk of metabolic syndrome by 77%.2 Furthermore, a dose–response relationship between increasing years of shift work history and metabolic syndrome was found among women in a recent large study of Chinese workers, where every 10 years of shift work increased the OR for metabolic syndrome by 10%.29 It is possible that our study was underpowered, did not include enough variability in cumulative shift work exposure to detect a difference, or our comparator group included employees with other unmeasured personal or work characteristics that contributed to cardiometabolic risk.
We chose to investigate work-related sleep duration as a mediator to best capture the effect of a rotational shift work schedule on sleep. In another work by our group, we found that this sample of shift workers slept approximately 20 min fewer than day workers when sleep duration was averaged across the entire week (including work shifts and free days).16 The reason for this relatively small difference is because shift workers slept considerably longer than day workers on free days, a finding that is consistent with previous research.30 Although shift workers ‘catch up’ on sleep on free days, it is conceivable that a repeated pattern of sleep restriction on work shifts increases cardiovascular and metabolic disease risk. A number of biological mechanisms may explain why sleep restriction increases the risk of metabolic syndrome. For example, sleep restriction for one to three nights has been associated with altered cortisol levels,31 insulin resistance,32 glucose homeostasis,33 disrupted appetite-regulating hormones (eg, leptin, ghrelin),34 increased subjective hunger34 and higher caloric consumption.35
This study has a number of key strengths. It is novel in that it is one of the few studies that have assessed the mediating role of sleep in the relationship between shift work and metabolic syndrome. This is a large sample size in relation to other occupational health studies with such detailed data collection, and the use of actigraphy to objectively measure sleep is a more accurate method than estimates from self-report approaches.36 In addition, the use of bootstrapping to construct 95% CIs around the indirect effect provides a more accurate estimate of the significance and variability of the mediating effect compared with traditional methods that assume normality in the sampling distribution.27
There are some limitations in this study. First, mediation analysis should ideally be conducted using longitudinal data. To provide evidence of causality, this analysis should be replicated in cohort studies. Second, some non-differential misclassification of sleep duration is likely to have occurred. Actigraphy is a less accurate approach to monitor sleep–wake cycles than laboratory-based methods such as polysomnography, and the relatively short data collection period in this study reduces the reliability of actigraph measures of sleep duration. Although the recommended minimum length for actigraph monitoring is 3 days,37 as many as 3 weeks may be needed to obtain valid measures of sleep–wake cycles.38 Therefore, to best capture general sleep patterns, this study would have benefited from longer term actigraph monitoring. Although we cannot quantify the degree of sleep misclassification in this study, any misclassification was likely non-differential and would attenuate the measure of association towards the null value. Despite some potential misclassification, actigraphy is able to measure sleep in the natural environment so is the most practical method to objectively measure sleep in large observational studies. In a similar manner, there may be some misclassification of metabolic syndrome. Blood pressure and biological samples used to diagnose metabolic syndrome are modified depending on the time and date when measurements are conducted. Blood pressure was measured on enrolment, and the date and time of enrolment varied by participant. Although blood was drawn for biological samples typically in the morning for both shift workers and day workers, differences in shift type at time of collection (day shifts for day workers, free days for shift workers) might have impacted lipid and glucose profiles. Future studies should attempt to standardise timing of blood pressure measurements and biological sample collections.
Another limitation of this study is that we investigated only one potential mechanism in which shift work affects the metabolic syndrome, whereas this relationship is undoubtedly multifactorial. For example, other potential mediators may include changes in cortisol, melatonin, other disturbances to the sleep–wake cycle (eg, sleep quality, timing of sleep and so on) and diet. The strong mediating effect found for sleep duration is likely related to some of these other variables, and when assessing sleep duration as a mediator, we also captured the effect of other mediators related to sleep duration. Future studies should build on this model and better disentangle individual effects by incorporating other possible mediators. In addition, uncontrolled confounding may be possible in this study. We did not have good information on current comorbidities or medication use that may influence sleep duration or the metabolic syndrome, so we were unable to account for these variables. Future studies should attempt to incorporate these variables into their data collection and analyses. Lastly, the results may not be generalisable to different populations (eg, different occupational settings, shift work schedules and so on).
In conclusion, these results provide evidence that reduced sleep duration is a pathway by which female alternating day–night shift workers working in the healthcare sector are at increased risk for metabolic syndrome. Future work that builds on this model will allow for the development of healthy workplace policies to help reduce the adverse health effects of shift work.
Acknowledgments
We thank the participants, the investigative team (L McGillis-Hall, I Janssen, A Day, C Collier) involved in the design, staff (including R Corbin and C Kelly), and the students (including Michael Leung and Eleanor Hung) involved in data cleaning.
References
Footnotes
Contributors JK was responsible for the conceptualisation of this project. She conducted most data cleaning and all statistical analyses, interpreted the results, and prepared and wrote the manuscript. JT is the primary investigator of the overall study from which this project is based, and a cosupervisor of this specific project. She conceptualised and designed the overall study along with coinvestigators, contributed to the interpretation of results and made revisions to this paper. AD contributed to the analytical plan and provided revisions of this paper. KJA is a coinvestigator of the overall study and was the primary supervisor of this specific project. She was involved in study design, provided statistical guidance, assisted in the interpretation of results and made revisions to this manuscript.
Funding This study was funded by the Canadian Institute for Health Research and the Workplace Safety and Insurance Board (Ontario).
Competing interests None declared.
Patient consent Obtained.
Ethics approval Queen’s University Health Sciences Research Ethics Board.
Provenance and peer review Not commissioned; externally peer reviewed.