Effects of psychosocial work characteristics on hair cortisol – findings from a post-trial study

Abstract

Prolonged work stress, as indicated by the effort-reward imbalance (ERI) model, jeopardizes health. Cortisol represents a candidate mechanism connecting stress to ill health. However, previous findings appear inconclusive, and recommendations were made to assess work stress at multiple time points and also to investigate ERI (sub-)components. This study therefore examines the effects of two single time points, as well as the mean and change scores between time points of ERI and its components on hair cortisol concentration (HCC), a long-term cortisol measurement. Participants were 66 male factory workers (age: 40.68 ± 6.74 years; HCC: 9.00 ± 7.11 pg/mg), who were followed up after a stress management intervention (2006–2008). In 2008 (T1) and 2015 (T2), participants completed a 23-item ERI questionnaire, assessing effort, the three reward components (esteem, job security, job promotion) and over-commitment. In 2015, participants also provided a 3-cm hair segment close to the scalp for HCC analysis, as well as information on relevant confounders (i.e. medication intake, age, work characteristics, socioeconomic and lifestyle factors, number of stressful life events). Linear regressions revealed hardly any cross-sectional or longitudinal effect of ERI and its components on HCC. Only the change scores between T1 and T2 of job security were negatively associated with lower HCC in unadjusted (β = −.320; p = .009) and adjusted (β = −.288; p = .044) models. In this study, only a decrease of perceived job security over time was significantly associated with higher HCC, and other predictors were not related to this outcome. Especially after correction for multiple testing, this study revealed just a weak association of different psychosocial work measurements with HCC.

Lay summary

This study showed that an increase in perceived job insecurity is correlated with higher levels of the stress hormone cortisol. The higher levels of cortisol might represent a biological explanation for the negative health effects of job insecurity. The association was, however, relatively low, and more and more voices are questioning whether cortisol in hair is a reliable marker for perceived work stress.

Keywords:

• Work stress
• effort-reward imbalance
• change
• longitudinal
• hair cortisol
• job security

Introduction

Chronic stress from the environment can elicit a physiological stress response, which is mainly triggered through the sympatho-adrenal medullary (SAM) system – with epinephrine and norepinephrine as end products – and the hypothalamic–pituitary–adrenal (HPA) axis resulting in the release of glucocorticoids, specifically cortisol (McEwen, 1998). If the stress response is repeatedly activated due to chronic demanding conditions, excessive levels of cortisol secretion can be maladaptive for health. Persistently high levels of cortisol have been associated with mental health and coronary heart diseases (Piazza, Almeida, Dmitrieva, & Klein, 2010). Thus cortisol might be a candidate mechanism to explain the close links observed between adverse psychosocial work conditions and indicators of reduced physical and mental ill health (Schnall, Dobson, & Landsbergis, 2016; Theorell et al., 2015).

The effort–reward imbalance (ERI) model is a well-established concept for measuring adverse work situations and has consistently been linked to ill health (Loerbroks, Shang, Angerer, & Li, 2015; Nieuwenhuijsen, Bruinvels, & Frings-Dresen, 2010; Schmidt et al., 2015; Siegrist, 2008, 2010). This model is based on the norm of social reciprocity and defines burdensome work situations in which efforts are insufficiently reciprocated by money, esteem, career opportunities and job security (Siegrist, 1996; Siegrist et al., 2004). Another key element of the ERI model is over-commitment, referring to an excessive degree of commitment at work and a high need of approval (Siegrist et al., 2004). While over-commitment is suggested to increase the strain resulting from ERI (i.e. moderating the negative effects of ERI on health), it also exerts a direct effect as a risky pattern of personal coping with demanding work (Siegrist, 1996; Siegrist & Li, 2016).

ERI might affect health via cortisol, but empirical evidence for the association of the ERI model with salivary cortisol is inconclusive so far and mostly limited to cross-sectional studies (Almadi, Cathers, & Chow, 2013; Bellingrath, Weigl, & Kudielka, 2008; Eller, Netterstrom, & Hansen, 2006; Eller et al., 2012; Hanson, Maas, Meijman, & Godaert, 2000; Harris, Ursin, Murison, & Eriksen, 2007; Irie, Tsutsumi, Shioji, & Kobayashi, 2004; Izawa, Tsutsumi, & Ogawa, 2016; Maina, Bovenzi, Palmas, & Larese Filon, 2009; Ota et al., 2014; Siegrist & Li, 2016; Steptoe, Siegrist, Kirschbaum, & Marmot, 2004; Wright, 2011).

In addition and complementary to measuring cortisol in saliva, measuring cortisol in hair for cumulative long-term stress exposure has only recently started. It yields tangible benefits compared to saliva or blood measurements representing mainly acute stress reactions (Stalder & Kirschbaum, 2012; Weckesser et al., 2014; Weitzman et al., 1971). Hair cortisol concentration (HCC) has increasingly become a biomarker for chronic stress exposure, also with regard to chronic work stress (Russell, Koren, Rieder, & Van Uum, 2012; Stalder & Kirschbaum, 2012; Staufenbiel, Penninx, Spijker, Elzinga, & van Rossum, 2013; Wells et al., 2014; Wosu, Valdimarsdottir, Shields, Williams, & Williams, 2013). A recent meta-analytic integration found, however, no consistent association of HCC with self-reported perceived stress (Stalder et al., 2017), and a recently published study found that HCC had only limited applicability as a biomarker for work stress (Janssens et al., 2016).

The association of long-term cortisol assessment in hair with the ERI model has been investigated in some studies, yielding inconclusive findings. One study reports a significant positive correlation with the ERI ratio (Qi et al., 2014), while another study reports significant associations only on the level of single ERI items (Gidlow, Randall, Gillman, Silk, & Jones, 2015). A study among workers of a ready-made garment factory in Bangladesh found the surprising result that the ERI item on promotion prospects was positively associated with higher hair cortisol levels (Steinisch et al., 2014).

While many studies measure work stress at a single time point, it is recommended to measure work stress over time by multiple measurements in order to improve estimation of long-term exposure and to reduce potential misclassification (i.e. type II error) (Kivimaki et al., 2006; Kivima¨ki et al., 2008). A recent study showed that change scores of ERI and its respective components between two time points had better statistical power to predict depressive symptoms than single ERI scores or mean scores between the waves (Li et al., 2013). Furthermore, there is empirical evidence revealing changes in ERI, in particular increase over time, to be related to mental (Buddeberg-Fischer, Klaghofer, Stamm, Siegrist, & Buddeberg, 2008; Godin, Kittel, Coppieters, & Siegrist, 2005; Li et al., 2013) and cardiovascular health (Chandola, Siegrist, & Marmot, 2005; Trudel, Brisson, Milot, Masse, & Vezina, 2015, 2016). Examining changes in ERI scores over time also appears to be a promising innovation for another reason: it may be instructive to prioritize intervention measures according to the strength of links between distinct aspects of the work stress model and health (Smith & Beaton, 2008).

In light of the previous discussion, this study aimed to examine effects of ERI measurements at two single time points (2008 and 2015), as well as their means and their change scores between time points on hair cortisol levels in 2015. Furthermore, the potential differential effects of the ERI ratio, the single scales “effort”, “reward”, and “over-commitment”, and the sub-components of “reward” (i.e. financial, esteem, career reward, job security) on HCC were explored. It is assumed that more adverse psychosocial work characteristics are associated with higher HCC (i.e. positive association with ERI ratio, effort, over-commitment; negative association with reward and its sub-dimensions [esteem, job security, job promotion]). Regarding the change effects, a deterioration of rewards is suggested to be associated with higher HCC (negative association), while an improvement in adverse aspect should be related to lower HCC (positive association). In addition, multiplicative interaction terms tested the potential effect modification by over-commitment.

Methods

Participants

The data used in this study were post-trial measurements of a randomized control trial with waiting list design among employees of a manufacturing plant in Southern Germany (Limm et al., 2011). Employees were invited to partake in a 2-day stress management intervention with half-day booster sessions after 4 and 6 months, one half in the year 2006 and one half in 2007 (n = 189). The stress management intervention was designed to improve an individual’s ability to identify and cope with stressors in the work environment. In particular, potential efforts and rewards, as well as over-committed work behaviors, were identified, and means for improvement were discussed. Participants were followed up in 2008 (n = 139) and 2015 (n = 119). In 2015 a 3-cm hair segment close to scalp was collected from each participant. The present analyzes comprised men with no missing values on relevant variables in these assessments (n = 71). One person was excluded due to an extremely high hair cortisol value (222.97 pg/mg), identified as an outlier exceeding ±3.5 standard deviation around the mean. In addition, participants using cortisol medication (n = 3) or with dyed hair (n = 1) were excluded. Participants provided written informed consent and information was provided according to the Declaration of Helsinki. The Ethical Committee of the University of Ulm approved the study.

Measures

Adverse psychosocial work conditions

ERI scales were assessed by a validated questionnaire covering effort with six items, reward with 11 items (consisting of four items esteem, two items job security and four items job promotion) and over-commitment with six items (Siegrist et al., 2004). The items of effort and reward scales were rated on a 5-point scale, where a value of 1 indicates no stressful experience and a value of 5 indicates very high stressful experience. The over-commitment items were rated on a 4-point Likert scale from 1 (fully disagree) to 4 (fully agree). Consequently, with such a scoring, the range for the scale “effort” is 6–30, for the scale “reward” 11–55, and for the scale “over-commitment” 6–24, with higher scores reflecting higher effort, reward, and over-commitment. Furthermore, a ratio between the two scales “effort” and “reward” (weighted by item numbers) was calculated to quantify the degree of mismatch between high “cost” and low “gain” at work at individual levels (Siegrist et al., 2004).

Hair cortisol analysis

Hair cortisol was measured by an online solid phase extraction liquid chromatography-tandem mass spectrometry method based on fragments of second order MS³ (SPE LC-MS/MS/MS) following the procedure described by Quinete, Bertram, Reska, Lang, and Kraus (2015). Around 50 mg (when available) of minced hair were weighed, washed with 2.5 mL of isopropanol and dried overnight at room temperature. Then, hair samples were incubated in 2 mL methanol with internal standards (cortisol-d4 and cortisone-d7) at room temperature for 24 h. Finally, samples were centrifuged at 4500 rpm for 10 min, supernatants were transferred to an LC vial and analyzed by online SPE-LC-MS³. The limit of quantitation (LOQ) was 0.05 ng/mL⁻¹, which corresponds to 2 pg/mg⁻¹ hair (for 50 mg of hair). Hair weight varied from 7.3 to 69 mg. The method was adapted using 1 mL methanol when the amount of hair was ≤25 mg. For statistical purposes, cortisol levels were set to 1/2 LOQ in cases where levels were below LOQ (n = 1). Intra- and inter-assay coefficients of variation of the method ranged between 1.4 and 14%.

Potential covariates

Potential covariates that might confound the association between ERI and hair cortisol included: age (years), net household income per month (≤€3000, €3001–4000, > €4000, not specified), shiftwork (yes or no), lifestyle factors, as well as medication intake and number of stressful life events. Lifestyle factors were assessed by questions on physical activity (h/week), and smoking (never, ex, smoker). Participants were asked whether they were taking medication at time of hair cortisol sampling and if so, which. In case of cortisol medication, participants were excluded (cf. section on participants above); otherwise they were categorized into medication intake (mostly blood pressure medication) versus no medication intake. Ten items (without two work-related items) from the list of 12 threatening experiences (LTE) (Rosmalen, Bos, & de Jonge, 2012) assessed number of stressful life events (potential range 0–10).

Statistical analysis

Change scores for the ERI ratio as well as for the underlying components and over-commitment were calculated by subtracting the baseline value (T1) from the follow-up value (T2). Linear regression models estimated the association of ERI scores at T1, T2, mean, and change score with HCC. Two sets of models were calculated. The first set was unadjusted, while the second one was adjusted for previously defined covariates. Based on their established association with cortisol or ERI, these models were adjusted for age, shiftwork, income, physical activity, smoking, medication intake, and number of stressful life events at time of hair cortisol sampling (Manenschijn, van Kruysbergen, de Jong, Koper, & van Rossum, 2011; Siegrist, 2010; Stalder et al., 2017; Staufenbiel et al., 2013). Change score analyzes were additionally adjusted for baseline values (T1) of the corresponding scale to account for regression to the mean (Smith & Beaton, 2008). In order to test effect modification by over-commitment, analyzes were repeated by including a multiplicative interaction term between over-commitment and the ERI scales while controlling for both main effects. Hair cortisol was transformed to approach normal distribution (Log₁₀). These analyzes were performed using IBM SPSS Statistics Version 22. Post-hoc power analyzes were conducted with an online power calculator (Soper, 2017). The observed statistical power (SP) for multiple regressions was based on the observed probability level, the number of predictors, the observed R², and the sample size.

Results

The sample characteristics are presented in Table 1. The mean HCC was 9.00 (± 7.11) pg/mg. Only one sample showed cortisol concentration lower than LOQ. The average age was 40.68 years (± 6.74), and most of the participants did no shiftwork (81.8%) and never smoked (43.9%). The mean duration of physical activity per week was 2.46 h (± 2.98) and 31.8% of the participants took medication (mostly blood pressure medication). Nearly one-third provided no information about income. The analytic sample did not significantly differ from dropouts in terms of sociodemographic variables, lifestyle factors or ERI scales in T1 or T2 (data not shown).

Table 1. Sample characteristics at time of hair cortisol sampling.

	Mean or %	SD or n
Cortisol hair (pg/mg)	9.00	7.11
Age (years)	40.68	6.74
Shiftwork (yes)	18.2%	12
Smoking
Never smoker	43.9%	29
Ex-smoker	33.0%	22
Smoker	22.7%	15
Physical activity (h/week)	2.46	2.98
Medication (yes)	31.8%	21
Income
≤3000	18.2%	12
3001–4000 Euro	34.8%	23
>4000 Euro	16.7%	11
n/s	30.3%	20
Number of stressful life events	2.18	1.60

SD: standard deviation; n: number.

Table 2 represents the mean values and standard deviation for the scores in 2008 (T1), 2015 (T2), the mean of the two time points and the change between them. All scores appeared rather stable over time. The biggest changes were for esteem, which decreased from 22.03 to 20.83 (p = .017), and over-commitment, which increased from 12.71 to 13.38 (p = .125). The effort and reward scale showed good reliability at both time points (Cronbach’s αs ≥ 0.80), which is suggested to be an important consideration to define change between points in time (Smith & Beaton, 2008). For over-commitment the reliability was satisfactory (Cronbach’s αs T1 = 0.74; T2 = 0.79).

Table 2. Scores of ERI scales at T1, T2, as well as mean and change scores.

	T1		T2		Mean (T1 T2)		Change (T1 T2)
	Mean	SD	Mean	SD	Mean	SD	Mean	SD	p
Effort-reward ratio	0.59	0.24	0.59	0.22	0.59	0.20	−0.01	0.22	0.834
Effort	15.00	3.83	14.59	4.07	14.82	3.37	−0.41	3.96	0.415
Reward	48.50	6.60	47.35	6.55	47.92	5.67	−1.15	6.65	0.164
Esteem	22.03	3.57	20.83	3.90	21.43	3.17	−1.20	3.98	0.017
Job security	8.89	1.76	8.95	1.59	8.92	1.38	0.06	1.90	0.796
Job promotion/salary	17.58	2.35	17.56	2.82	17.57	2.09	−0.02	3.07	0.968
Over-commitment	12.71	3.56	13.38	3.92	13.05	3.31	0.67	3.49	0.125

SD: standard deviation; p: p value (paired samples test).

Table 3 represents the unadjusted effects and Table 4 represents the adjusted effects of the ERI scales at two time points, the mean score and the change score on levels of hair cortisol. In unadjusted models, job promotion at 2008 showed a significant association with HCC (β = .253, p = .040); the effects of reward and job security were not statistically significant (β = .225, p = .069; β = .232, p = .061; respectively). The change in the reward scale was negatively associated with hair cortisol levels, however not reaching statistical significance (β = −.212, p = .087), and change in job security showed a strong and significant effect on hair cortisol (β = −.320, p = .009; Figure 1). The change effects were independent of several confounding factors, as in models adjusted for medication intake, number of stressful life events and initial scores, the pattern of effects remained similar (reward: β = −.226, p = .090; job security β = −.288, p = .044). No other associations reached level of significance.

Figure 1. Association of change in the job security scale between T1 and T2 and transformed (Log 10) hair cortisol concentration (HCC) (see file attached).

Table 3. Unadjusted effects of single time points (T1, T2) ERI measures, as well as mean and change scores of ERI scales on hair cortisol concentration (Log₁₀).

	T1				T2				Mean (T1 T2)				Change effect
	B	β	R^2	SP	B	β	R^2	SP	B	β	R^2	SP	B	β	R^2	SP
Effort-reward ratio	−0.227	−0.172	0.030	0.294	−0.106	−0.074	0.006	0.086	−0.226	−0.144	0.021	0.215	0.157	0.111	0.012	0.154
Effort	−0.010	−0.116	0.013	0.145	−0.008	−0.107	0.011	0.128	−0.012	−0.129	0.017	0.180	0.000	0.001	0.000	0.000
Reward	0.011^a	0.225	0.051	0.468	0.001	0.012	0.000	0.040	0.008	0.138	0.019	0.198	−0.010^a	−0.212	0.045	0.420
Esteem	0.012	0.135	0.018	0.189	0.006	0.073	0.005	0.078	0.012	0.121	0.015	0.163	0.045	−0.050	0.002	0.054
Job security	0.041^a	0.232	0.054	0.491	−0.025	−0.126	0.016	0.171	0.017	0.075	0.006	0.086	−0.053^b	−0.320	0.103	0.784
Job promotion	0.034^c	0.253	0.064	0.564	0.000	−0.003	0.000	0.039	0.021	0.140	0.020	0.206	−0.020	−0.197	0.039	0.371
Over-commitment	−0.014	−0.156	0.024	0.241	−0.005	−0.057	0.003	0.062	−0.011	−0.117	0.014	0.154	0.009	0.095	0.009	0.111

^a p < .10.

^b p < .01.

^c p < .05.

SP: statistical power.

Table 4. Adjusted effects of single time points (T1, T2) ERI measures, as well as mean and change scores of ERI scales on hair cortisol concentration (Log₁₀).

	T1				T2				Mean (T1 T2)				Change effect
	B	β	R^2	SP	B	β	R^2	SP	B	β	R^2	SP	B	β	R^2	SP
Effort-reward ratio	−0.088	−0.067	0.276	0.942	0.143	0.100	0.303	0.969	0.000	0.000	0.273	0.939	0.259	0.183	0.317	0.971
Effort	−0.003	−0.037	0.274	0.940	0.002	0.021	0.295	0.962	−0.002	−0.022	0.273	0.939	0.005	0.057	0.298	0.955
Reward	0.004	0.091	0.280	0.947	−0.007	−0.137	0.290	0.958	−0.002	−0.031	0.273	0.939	−0.011^a	−0.226	0.316	0.970
Esteem	0.003	0.036	0.274	0.939	−0.005	−0.060	0.276	0.942	−0.002	−0.018	0.273	0.939	−0.007	−0.093	0.280	0.934
Job security	0.029	0.164	0.298	0.965	−0.030	−0.152	0.295	0.962	0.003	0.015	0.273	0.939	−0.047^b	−0.288	0.347	0.987
Job promotion	0.010	0.072	0.277	0.943	−0.016	−0.142	0.291	0.958	−0.011	−0.071	0.277	0.943	−0.018	−0.172	0.300	0.957
Over − commitment	−0.009	−0.105	0.282	0.949	−0.007	−0.086	0.280	0.947	−0.010	−0.103	0.277	0.943	−0.003	−0.037	0.283	0.938

^a p < .10.

^b p < .05.

SP: statistical power.

Analyses were adjusted for age, shiftwork, income, physical activity, smoking, number of stressful life events, medication intake, and baseline scale values in change analyzes.

Effects of the ERI scales on hair cortisol were not modified by over-commitment, as all interaction terms were not significant (all p values > .1).

The observed SP with an alpha level of 0.05 ranged between 0.145 and 0.564 in the unadjusted analyzes, and between 0.934 and 0.987 in the adjusted regressions (Tables 3 and 4).

Since multiple regression analyzes were run, effects might be due to Type I error. Correcting the p-value for multiple testing according to the Bonferroni correction (α/number of tests) (Shaffer, 1995) results in a p value of .0018 for an α of 0.05 and 28 tests. According to this correction, no significant associations between psychosocial work conditions and HCC were found.

Discussion

The present study examined single time, mean, and change effects of ERI scales on HCC. Results revealed that only changes in ERI components were related to hair cortisol levels. The change in the reward scale was associated with hair cortisol levels, but not reaching level of significance. However, a change in its subscale “job security” was significantly associated with hair cortisol levels, such that a decrease in job security over time was associated with higher cortisol levels in hair. This effect was independent of sociodemographic and lifestyle factors, as well as of baseline scale values, medication intake and number of stressful life events. However, after correction for multiple testing associations of psychosocial work, measurements with HCC did not reach a level of significance. That an established model to measure adverse psychosocial work conditions, such as the ERI model, found hardly any cross-sectional or longitudinal association with hair cortisol, especially after correction for multiple comparisons, appears in line with a growing amount of literature challenging the use of cortisol as a reliable biomarker for perceived work stress. For instance, a systematic review on the relation between salivary cortisol and psychosocial work stress revealed that a large proportion of studies showed non-significant findings (Karlson et al., 2012). A recent meta-analysis reported no consistent association of HCC with self-reports of perceived stress (Stalder et al., 2017), and a study among Belgian workers concluded that HCC has limited applicability as a biomarker for work stress (Janssens et al., 2016). One reason might be that significant HCC elevation requires chronic stress exposure above a certain intensity level to exhaust regulatory capacities of the endocrine system (Stalder et al., 2017). While one study found that levels of hair cortisol were positively related to one item on the job security scale (I have experienced or I expect to experience an undesirable change in my work situation) (Gidlow et al., 2015), other empirical studies stand in contrast to this finding. One study found a significant positive association of the ERI ratio in 39 Chinese female kindergarten teachers in unadjusted analysis (Qi et al., 2014). Similarly, in workers from a garment factory in Bangladesh the ERI item on promotion prospects was positively associated with HCC (Steinisch et al., 2014). The variety of these findings might be explained by both the different cultural backgrounds and the questionable applicability of HCC as a biomarker for self-reported work stress.

An alternative, more statistical, explanation for the weak associations between psychosocial work conditions and HCC might be the lack of statistical power due to the restricted sample size of 66 persons in this study. Especially in the adjusted regression models, the statistical power appears high enough (≥ .934) to detect possible effects.

Although the applicability of HCC as an indicator for work stress is currently debated, change score analyzes in this study were superior to single time and mean score analyzes regarding the association of ERI and HCC. The proportion of explained variance was highest in all adjusted change score analyzes and lowest in mean score models. Thus, multiple measurements of work stress appear more suitable than single time assessments only when applying the change between time points. The observation that the change score of ERI is the best predictor for health outcomes corresponds to the findings of a previous study (Li et al., 2013).

Even though it is questionable whether hair cortisol is an appropriate biomarker for self-reported work stress, there is extensive epidemiological evidence about the negative health effects of perceived low job security. For example, recent systematic reviews and meta-analyzes revealed perceived lack of job security to be a significant risk factor for incident coronary heart disease (Virtanen et al., 2013) and depressive symptoms (Kim & von dem Knesebeck, 2015). High stress hormone levels are supposed to transmit such negative health effects (Pereg et al., 2011; Staufenbiel et al., 2013). Cortisol affects lipid and carbohydrate metabolism, and hair cortisol has been linked to metabolic syndrome (Stalder et al., 2013). While previous studies found job security to be cross-sectionally related to cortisol levels (Yan et al., 2016), our study could only provide conditional evidence of an association between decreased job security and higher long-term cortisol levels.

This study has several limitations. First, the sample comprised male employees, and further studies are thus needed to examine whether our findings could be generalized to women. Second, hair cortisol was only assessed at follow up (T2). Third, a healthy worker effect cannot be ruled out as employees with the worst health or lowest job security might have left the company. Finally, due to lack of data, a potentially relevant covariate associated with HCC, the body mass index (Stalder et al., 2012), was not included in our analysis. In conclusion, the current study provides conditional evidence of an association between a decrease in job security and higher HCC, a biomarker of long-term stress release. This finding might point to a potential pathway by which a core reward frustration in the workplace, the perception of deteriorating job security, might affect health. In general, however, HCC revealed a weak association with different measurements of adverse psychosocial work conditions.

Acknowledgements

The company's medical services supported the research team in conducting the study. The authors are indebted to all participants of the study and to the support of the hosting company.

Disclosure statement

The authors report no conflicts of interest.

Additional information

Funding

This work was supported by the German Federal Ministry of Education and Research (BMBF) under Grant No. 01EL1409B.