Article Text
Abstract
Objectives: To examine the relationships between exposure to workplace factors (night work, extended working hours, psychosocial work stress) and cortisol secretion, and to test whether workplace factors interact, resulting in combined effects.
Methods: Multiple linear and logistic regression was used to test relationships between workplace factors and cortisol secretion in the 1958 British birth cohort at 45 years. Salivary cortisol was measured twice on the same day to capture the post-waking decline, facilitating the analysis of different cortisol patterns: (1) time 1 (T1, 45 minutes post-waking); (2) time 2 (T2, 3 h after T1); (3) average 3 h exposure from T1 to T2 cortisol; and (4) T1 to T2 change. To identify altered diurnal cortisol patterns we calculated: (1) flat T1–T2 change in cortisol; (2) top 5% T1; (3) bottom 5% T1; and (4) T1 hypo-secretion or hyper-secretion. Models were adjusted for socioeconomic position at birth and in adulthood, qualifications, marital status, dependent children, and smoking status.
Results: 25% of men and 8% of women were exposed to >1 workplace factor (night work, extended work hours, job strain). Night work was associated with a 4.28% (95% CI 1.21 to 7.45) increase in average 3 h cortisol secretion independently of job strain or work hours. Night workers not exposed to job strain had elevated T1 cortisol (5.81%, 95% CI 1.61 to 10.19), although for T2 cortisol it was night workers exposed to low job control who had elevated levels (11.72%, 95% CI 4.40 to 19.55). Men (but not women) working >48 h/week had lower average 3 h cortisol secretion (4.55%, 95% CI −8.43 to −0.50). There were no main effects for psychosocial work stress. All associations for T2 and average 3 h cortisol secretion weakened slightly after adjustment for confounding factors, but associations for T1 cortisol were unaffected by adjustment.
Conclusions: Our study suggests that night work in particular is associated with elevated cortisol secretion and that cortisol dysregulation may exist in subgroups with specific combinations of stressors.
Statistics from Altmetric.com
Policy makers recognise that being in work and experiencing good working conditions are vital for health and wellbeing.1 2 Poor-quality jobs, such as those characterised by poor working conditions, may be detrimental for health. Some argue that psychosocial work stress, such as low job control or effort–reward imbalance, increases the risk of coronary heart disease, obesity and metabolic changes.3 4 5 6 Working at night or long hours is also associated with several outcomes including cardiovascular disease (CVD), type 2 diabetes and elevated body mass index.7 8 Explanations for such findings are unclear. Workplace factors could increase the risk of CVD through direct metabolic consequences, for example, circadian disruption by night work; direct effects on behaviour arising from particular work patterns, for example, long hours may prevent access to healthy food; or indirectly through stress-induced behaviour changes including insufficient sleep.8 9 10 Furthermore, workplace stressors could influence CVD through neuroendocrine responses to stress if abnormal cortisol secretion patterns are involved in the development of the metabolic syndrome and CVD as suggested.11
Several studies have reported associations between workplace factors and cortisol secretion. A Swedish population-based study found that the cortisol response to waking was enhanced for women, but not men, in association with psychosocial work stress.12 Cortisol dysregulation has also been reported from the Whitehall II study in relation to work stress.13 Because the Whitehall study is a study of civil servants in London, England the findings may not apply to the general population. Also, civil service jobs will not usually involve other working practices such as regular night work. Disruption of the diurnal cycle of cortisol secretion in night shift workers has been demonstrated,14 15 16 while studies examining long working hours, find little evidence to suggest that cortisol is affected.17 18 19 However, most are studies of small samples selected from specific occupational settings.
What this paper adds
Workplace factors such as night work, long working hours and psychosocial work stress are associated with obesity, cardiovascular disease and type 2 diabetes. Workplace stressors could influence cardiovascular risk through neuroendocrine responses to stress.
Night work was associated with cortisol dysregulation independently of work duration (h/week) or psychosocial work stress.
Psychosocial work stress alone was not associated with cortisol secretion, but women and night workers exposed to job strain may be vulnerable to cortisol dysregulation.
Men working more than 48 h/week had lower average 3 h cortisol secretion.
Further investigation of workplace factors over the life-course that take into account pre-employment selection processes is needed in order to understand the causal relationships between work and cortisol secretion.
The effects of combinations of different workplace factors in relation to cortisol secretion have not been previously examined. Since work hours, night work and job strain are related, the effects of one factor may be modified by the other.20 The 1958 British birth cohort provides an opportunity to address this issue in a large, unselected population, that is, spanning the whole working population and not just particular occupations. The aim here is to investigate the inter-relationships of night work, long working hours and psychosocial work stress (job control and demands) with cortisol secretion at 45 years, to establish whether there are differential effects on cortisol by exposure to combinations of work factors. We examine cortisol patterns using a range of indicators to reflect the diurnal rhythm of cortisol secretion which is characterised by a peak soon after waking in the morning and a gradual decline throughout the day.21 There is evidence of a U-shaped relationship such that health risks have been found for both hyper- and hypo-cortisolism. Hence, we examine indicators of cortisol patterns including a post-waking cortisol measure that indicates either hyper-secretion or hypo-secretion.
Study population and methods
Study sample
The 1958 British birth cohort consists of 18 558 individuals: 17 638 participants enrolled in the Perinatal Mortality Survey all born 1 week in March 1958 in England, Scotland and Wales, and 920 immigrants with the same birth dates included to age 16 years. The cohort has been interviewed in childhood (ages 7, 11 and 16 years) and adulthood (23, 33, 42 and 45 years).22 At 45 years, 11 971 cohort members who had not died or emigrated and were still in contact with the survey, were invited to a clinical examination undertaken in their home by a trained nurse; 9377 (78%) participated. Ethical approval for the clinical study was given by the South East Multi-Centre Research Ethics Committee. The analyses presented in this paper are based on 7916 cohort members who were in paid employment at 45 years.
Measures
Salivary cortisol
The collection of saliva and measurement of cortisol has been described in detail elsewhere.23 Briefly, participants were asked to collect two saliva samples on the next convenient day after the 45 years clinical examination, the first 45 minutes after awakening (T1) and the second 3 h later on the same day (T2). Participants were instructed to avoid brushing or flossing their teeth, or eating or drinking for 15 minutes before taking each sample. They were asked to chew on a salivette until it was soaked, record the date and time of collection, and store the sample at room temperature until mailed to the laboratory. Cortisol levels were measured at the University of Dresden with a commercial immunoassay kit with chemiluminescence detection (CLIA, IBL-Hamburg, Germany).
Workplace factors
Hours worked: at 45 years, participants reported the total hours worked per week in their main job. Hours/week was categorised to identify part-time (<35), full-time (35–40), lower overtime (41–48) and extended working defined as >48 h/week in accordance with the European Directive on work hours.24 25
Night work: participants reported the frequency of working between the hours of 22:00 and 4:00 and 4:00 and 7:00 as “never”, “less often than once a month”, “at least once a month”, and “at least once a week” at 42 years (45 years data were not available). Regular night work was defined as working from 22:00 to 7:00 at least once a month.
Psychosocial work stress: job control and demand was measured at 45 years based on Karasek’s Job Content Instrument,25 26 as previously used by the Whitehall II study.27 Job control was measured by six questions (learning new things; monotony; level of skill; choice in how to do work; choice in what work to do; say in decisions) and job demands by four questions (work fast; work intensively; time to do everything; conflicting demands). Job strain was defined as low control (score above the median) and high demands (score below the median).
Confounding factors
An important potential confounder is socioeconomic position (SEP), in childhood and in adulthood, as social disadvantage over the life-course is associated with a range of health outcomes in adulthood, and also associated with cortisol dysregulation.23 In addition, relationships between marital status, dependent children and cortisol levels, particularly for women, have been reported.28 Smoking is related to cortisol hyper-secretion29 and has therefore been treated as a potential confounding factor in accordance with our previous studies,23 30 although its associations with shift work and work stress suggest that it may mediate work/cortisol associations.9 31 An examination of mediating effects was not an aim of this paper.
SEP at birth was based on father’s occupation using the Registrar General’s Social Class (RGSC), grouped as I&II (professional/managerial), IIINM (skilled non-manual), IIIM (skilled manual), and IV&V (semi-unskilled manual, including single-mother households). The RGSC is a standard method of classification of occupations in the UK.32 SEP at 42 years, was based on the participant’s current or most recent occupation and categorised as above. Highest qualification attained by 42 years (none, some qualifications, O-level or equivalent, A-level or equivalent, degree) was also used to indicate adult social position. Smoking habits were reported at 42 years, ranging from “never” to “current smoker of ⩾20 cigarettes/day” (seven categories). Participants reported their marital status at 45 years (single never married, first marriage, second or later marriage, separated, divorced, widowed) and dependent children (⩽18 years).
Data analysis
T1 and T2 values were truncated at 2 nmol/l (for <2 nmol/l), the lower limit of the detection range, and 100 nmol/l (for >100 nmol/l) in order that potentially implausible values did not disproportionately influence the analyses. T1 and T2 values were log10 transformed to reduce skewness in the distribution. Since cortisol level was influenced by both time of awaking and time since awaking, the transformed values for each individual were centred 08:08 (45 minutes after mean awaking time of 7:23) for T1, and at 11:08 (3 h 45 minutes after mean awaking time) for T2, by back transforming the predictions from linear regression models at specified times, to the original scale (nmol/l). Therefore, cortisol measures used in analyses are adjusted for both time of awaking and time since awaking. We analysed outcomes that differentiate cortisol patterns: (1) T1 cortisol level; (2) T2 cortisol level; (3) average 3 h cortisol exposure from T1 to T2: calculated by summing the back-transformed T1 and T2 values, multiplying the results by 3 and dividing by 2; and (4) T1–T2 slope calculated as the difference between T1 and T2 divided by the time between the measures. To identify altered diurnal cortisol patterns we calculated groups with (1) flat T1 to T2 defined as T2 cortisol within 20% of the T1 value, or, when the T1 value was <7.5 nmol/l, flat was defined as a T2 value <120% of T1; (2) top 5% T1; (3) bottom 5% T1; and (4) extreme T1 (top or bottom 5%). Cut-offs for the highest 5% were 40.73 nmol/l for males and 41.69 nmol/l for females; cut-offs for the lowest 5% were 7.24 nmol/l and 7.94 nmol/l, respectively. The outcomes described here have been used in our previous studies of cortisol.23 30 33
Multiple linear regression was used to analyse the relationship between workplace factors and continuous cortisol measures. As T1, T2 and 3 h cortisol were log10 transformed, relative change (in per cent) in these measures was calculated from the regression coefficient (β) as 100×(10β−1). For example, T1 cortisol changes from value X1 to 10β X1 when the workplace exposure changes by one level, for example, from none to night work. Logistic regression was used for binary cortisol measures (flat T1–T2, top or bottom 5% T1, bottom 5% T1, top 5%) from which odds ratios (ORs) and 95% confidence intervals were obtained.
Our modelling strategy consisted of testing: (1) main effects of the work variables on each cortisol outcome; (2) gender interactions with the work variables; (3) interactions between work variables; and (4) adjustment for potential confounding factors in final models. The Wald test was used to assess interactions. We chose p⩽0.10 to indicate interaction rather than the conventional p value of 0.05 because statistical tests for interaction are known to have low power and could miss important interactions.34 35 Where interactions were found, stratum specific estimates were calculated. Job control and demands were modelled simultaneously and compared with a second model using job strain, which combines both demand and control in one variable. Job strain was used in subsequent models unless the models of job control and demands indicated either of these components to be significant.
The greatest source of missing data was sample attrition to 45 years36 and lack of a cortisol measure. Most participants at 45 years had consented to saliva collection for cortisol measurement (n = 9165), only 6467 had a T1 cortisol value, 6506 had T2 cortisol, and 6449 had both. Analyses were undertaken on those in paid employment (84% of participants at 45 years) and had cortisol measured (n = 5546 T1 cortisol; n = 5576 T2 cortisol). Of these, approximately 8% had missing data on workplace factors while a further 11% were missing information on confounders. Missing data were handled, first, using multiple imputation of exposure and confounding factors because cortisol levels varied for those with missing exposure data. The imputation by chained equations (ICE) method in STATA was used, creating 10 copies of the dataset. All exposures and covariates reported in this paper plus the particular cortisol outcome of interest were included in the imputation equations. Second, as we did not impute on the dependent variables (ie, the CVD risk factors)37 multiple imputation alone would not account for attrition bias. We therefore weighted our analyses to take account of previous observations of moderate under-representation of more extreme groups, including those from families with no male head and with childhood behavioural or cognitive problems.36 Therefore, the combined analyses from the imputed datasets were weighted on key factors related to attrition, namely, maths score at 7 years, behavioural problems at 7 years, and social class at birth. Regression analyses presented are based on imputed and weighted data. Final sample sizes in regression models were determined by having the relevant cortisol outcome and data to construct sample weights: T1 = 4893, T2 = 4923.
Results
Men had lower T1 cortisol (median 18.8 nmol/l), but higher T2 cortisol (7.10 nmol/l) than women (19.6 and 6.6 nmol/l, respectively) (table 1). On average, cortisol levels declined from T1 to T2 (slope), with women having a greater decline than men.
Table 2 presents descriptive information for cortisol levels for employed men and women according to exposure to extended work hours, night work, and psychosocial work stress. Exposure patterns for workplace factors were considerably different for men and women: men were more likely to work nights and long hours but women were more likely to have low control and job strain, while similar proportions reported high demands. Approximately 25% of men compared to 8% of women were exposed to >1 work stressor, the most common combination being long hours and night work (15% and 3% of men and women, respectively) although 3% of women experienced night work and job strain (data not shown). In simple analyses of each workplace factor separately there was no statistically significant variation in T1 cortisol levels (table 2). Higher levels of T2 cortisol were seen for night work while lower T2 levels were associated with longer work hours per week for men only.
Associations between workplace factors and cortisol measures are presented for men and women combined (adjusted for sex) unless a gender interaction at p<0.1 was found (tables 3 and 4). For work hours, the only effect seen was among men working extended hours, who had 4% lower 3 h cortisol that was not explained by confounding factors (table 3). The gender interaction seen for work hours in univariate analyses of T2 cortisol was not apparent in the multivariate, imputed and weighted models (p = 0.206).
For night work, T1 cortisol was higher but only for those without job strain and was not attenuated following adjustment (table 3). Correspondingly, night work was associated with a 73% increased risk of top 5% of T1 cortisol (table 4) in those who were not exposed to psychosocial work stress, in this case, high demands. There was only weak evidence to suggest that T1 levels may in comparison be lower for night workers exposed to psychosocial work stress, who had a 30% increased risk of being in the bottom 5% T1, although the effect was not significant (table 4). Night work was independently associated with higher T2 and 3 h levels (table 3), with some weak indication that the effect for T2 existed in the subgroup exposed to low job control. Associations for T2 and 3 h cortisol attenuated slightly after adjustment for potential confounding factors (table 3). Accordingly, weak associations were found for cortisol change between T1 and T2 (data not shown) either as (1) a slower decline for night workers with job strain (0.19 nmol/l/h, 95% CI −0.37 to 0.76) (interaction p = 0.06), or (2) a flattened T1–T2 response (OR = 1.06, 95% CI 0.78 to 1.44) for night work and low control (interaction p = 0.09).
There were no independent findings for psychosocial work stress. There was an increased, but statistically insignificant, risk of top 5% T1 cortisol for women with job strain (OR = 1.36, 95% CI 0.91 to 2.03), but not men (OR = 0.82, 95% CI 0.49 to 1.36) (gender interaction p = 0.03). Finally, job strain was associated with a 35% increased risk of extreme (high or low) T1 cortisol (OR = 1.35, 95% CI 1.05 to 1.73) but only among those who did not work extended hours (interaction p = 0.08, strengthening to p = 0.065 after adjustment).
Discussion
Our main finding from this large, general population sample was that night work, a common characteristic in this population, was associated with altered cortisol patterns. Night workers had elevated cortisol soon after waking, an association that was most evident for those not exposed to work stress. For night work with psychosocial work stress there was a weak suggestion of a lower post-waking response and elevated cortisol later in the day. Work hours tended to be unrelated to cortisol, except for lower 3 h cortisol secretion in men working extended hours. Job strain alone was unrelated to cortisol patterns, although there was an increased risk of extreme T1 levels (high or low) in those who did not work extended hours and women with job strain were more likely to have high cortisol soon after waking than men. Associations were not explained by adjustment for smoking, SEP or family factors.
Strengths and limitations
The major strength of this study is that it uses data from a large, population-based sample from which estimates for exposures are generalisable to the whole working population, rather than a narrowly defined occupational group. Information on multiple workplace factors, including shift work, working hours and psychosocial work characteristics, was collected and the study has greater power to detect joint effects of exposures on cortisol patterns that cannot be studied in small samples.
The study has several limitations. First, residual confounding remains a possibility as our confounding factors reflecting SEP and personal circumstances may not fully capture non-work influences. Other than smoking, we have not examined factors that may mediate work/cortisol associations, such as health behaviours. Thus, it is beyond the scope of this paper to draw conclusions regarding causal mechanisms, especially as the directions may be complicated, for example, whether health behaviours alter cortisol levels or vice versa. Smoking strongly influences cortisol secretion and was treated as a measurement factor, requiring adjustment, rather than a true confounder.
Second, cortisol levels are lower in saliva than in blood, however, salivary cortisol is strongly correlated with serum cortisol.38 Because saliva samples are easy and inexpensive to collect they are well suited to population studies. Ideally, multiple saliva collections throughout a day, repeated over several days are required to measure the precise diurnal rhythm of cortisol secretion within individuals. The usual pattern is for cortisol levels to decline within a few hours from a post-waking peak. Thus, the timing of our samples should capture change over this crucial interval.21 The purpose of our study was to assess associations in a large population rather than for selected individuals; a maximum of two samples on one day was both affordable and feasible. In a study of this size, precision in the estimates of effect is gained at the group level even though estimates for individuals may be less reliable.
Third, studies of stress and illness are vulnerable to reporting bias as stressed individuals may report more symptoms.39 The associations here are unlikely to be due to such bias as we used an objective outcome measure, and work information was reported by participants without the knowledge of the hypotheses to be investigated. However, the reporting of work stress could be influenced by cortisol levels, as people with high cortisol may be highly stressed and more likely to report work stress. Information on shift work was collected at 42 years, therefore some misclassification of 45 years night work may have occurred.
Finally, due to attrition, the sample with complete data was less than half of the original birth cohort, leading to an under-representation of participants from disadvantaged backgrounds in terms of SEP, cognitive and behavioural problems.36 Such bias could lead to underestimation of associations between work and health. Analyses were therefore weighted to address this problem.
Interpretation of the findings
We are not aware of any study to date that has examined the joint effects of work-related factors on cortisol secretion. Several studies have reported dysregulation of the diurnal cortisol profile associated with exposures such as job strain, night work and to a lesser extent long work hours, usually for a single occupational group and exposure.12 13 14 15 16 17 18 19 We were able to separate the effects of the workplace factors studied, and also examine their joint effects. Our findings for cortisol dysregulation, in a large sample representative of middle-aged, working adults, implied that night work, in the absence of work stress and independent of the number of hours worked, is associated with post-waking cortisol hyper-secretion which could be due to circadian disruption.8 This is an important finding because of the substantial size of the employed population who work at night (⩾1/month): more than one-third of men and 16% of women, of which 79% did not experience job strain. There was some suggestion that night work in combination with work stress may be associated with lower post-waking cortisol and higher cortisol later in the day. This is a tentative finding requiring confirmation in other studies but if true, would have important policy relevance as organisations can influence workplace factors, for example, through the introduction of shift design practices that incorporate aspects of job satisfaction, for example, varied work load and a work environment comparable to that provided for day workers.40
We found no strong evidence in support of an association between cortisol dysregulation and psychosocial work stress, when other work exposures were taken into consideration. This contrasts with findings from Whitehall II whereby psychosocial stress at work was associated with an enhanced post-waking response and elevated cortisol at later times in the day.13 It is difficult to directly compare results from the 1958 cohort to those from Whitehall II because of differences in study design, such as the more diverse range of working environments in the 1958 cohort and differences in the timing of cortisol collection. There was some suggestion that women in the 1958 cohort exposed to job strain had an increased risk of being in the top 5% of post-waking cortisol levels, although no effect was seen on cortisol measured later in the day which is consistent with findings from Sweden.12 Our finding of an association for job strain with extreme post-waking values, that is, either high or low T1, in those who do not work extended hours is interesting because it may reflect underlying stressors, such as poor physical or mental health, that may be associated with hypo-secretion or hyper-secretion, but have resulted in shorter hours of work and enhanced an individual’s susceptibility to job strain. More detailed examination of the relationship between working shorter hours on cortisol secretion is warranted.
Working extended hours tended to be associated with hypo-cortisolism in men. Small, work-place-based studies that have experimentally increased work hours have found no effects on cortisol secretion.17 18 More generally, evidence is growing that long work hours adversely affect a range of health outcomes including CVD, but findings are inconsistent possibly due to methodological limitations. First, since shift duration can be long, effects of long hours need to be disentangled from shift-work patterns. In addition, control over work hours may alleviate adverse health effects of long work hours, although evidence is scant.25 Second, definitions of extended hours of work vary, although >48 h/week has been adopted by the European Directive on Working Hours and therefore used in our study because of its policy relevance,25 but more focus on the relative health effects of moderate and extensive overtime is needed. Finally, others have shown that the total hours worked arising from domestic work in addition to paid hours is an important consideration.41
Our study findings were not explained by other factors such as family circumstances, education or SEP. This is surprising given that night work and work stress are associated with less advantaged SEP, which, in turn is associated with extreme post-waking cortisol levels in the 1958 cohort.23 We did not seek to examine potential mediators, such as insufficient sleep, which is associated with increases in blood pressure, inflammatory and metabolic cardiovascular risk factors and obesity. For cortisol, and the hypothalamic-pituitary-adrenal (HPA)-axis more generally, the relationship with sleep appears to be bidirectional10 42: circadian disruption of the HPA-axis in shift workers can lead to sleep disruption which in turn can further disrupt the HPA-axis. Further investigation into causal pathways linking night work to cortisol dysregulation is needed. More generally, studies to date have tended to ignore the possibility that individuals may have been selected into their work roles due to other, underlying factors, which are in fact the proximate cause of cortisol dysregulation. Meanwhile, our results highlight the need for a better understanding of the differential effects of workplace exposures given the substantial proportion of the population experiencing these exposures.
Acknowledgments
Cortisol levels were measured under the direction of Professor Kirschbaum (Biological Psychology, Department of Psychology, University of Dresden, Germany) and we are grateful to Leah Li (Centre for Paediatric Epidemiology, UCL Institute of Child Health) who helped develop the cortisol measures used in the analyses.
REFERENCES
Footnotes
Funding Data collection at 45 years was funded by the Medical Research Council (grant G0000934). CT is funded by a fellowship from the Economic Social Research Council under the “Understanding Population Trends and Processes” initiative (grant RES-163-27-1011). The analyses also benefit from funding from the Human Early Learning Partnership (HELP), Vancouver, Canada. This work was undertaken at GOSH/UCL Institute of Child Health which received a proportion of funding from the Department of Health’s NIHR Biomedical Research Centres funding scheme. The Centre for Paediatric Epidemiology and Biostatistics also benefits from funding support from the Medical Research Council in its capacity as the MRC Centre of Epidemiology for Child Health.
Competing interests None.
Ethics approval Ethical approval for the clinical study was given by the South East Multi-Centre Research Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed.