Introduction

Over the recent decades work stress has emerged as a major psychosocial influence on physical and mental health in the developed countries resulting in raised social costs and weighting heavily in work organization: in a large investigation analysing the relationship between health risks and medical expenditure on 46,026 American employees, stressed individuals were 46% more costly than those lacking this risk (Goetzel et al. 1998); in the European Union work-related stress is the second most common work-related health problem affecting nearly one out of every three workers (Cox et al. 2000). The increasing interest on the relationship between work stress and physical and mental health has resulted on psycho-social models as well as patho-physiological models. One of the leading theoretical model in the job-stress literature is the “Demand/Control” model developed by Karasek and Theorell (1990) involving two major dimensions of job stress: psychological demand and decision latitude at work. The Karasek’s job strain model proposes that people working in highly demanding jobs also have low control and limited opportunities to use skills will experience high job strain. A high level of job strain has been found to be associated with physiological coronary heart diseases risk factors, such as high blood pressure (Tsutsumi et al. 2001; Niedhammer et al. 1998; Alfredsson et al. 2002), high serum total cholesterol (Jonsson et al. 1995), and high triglycerides (Netterstrom et al. 1991).

In the investigation of psychosocial influences on health, knowledge on biochemical pathways has been greatly expanded, as well as the availability of physiological measures. The results of the psychobiological research suggest that two neuroendocrine systems are activated as a result of the stress stimuli: the sympathetic-adrenomedullary system (SAM axis), with the secretion of epinephrine and norepinephrine, and the hypothalamic-pituitary-adrenocortical system (HPA axis) with the secretion of cortisol. The measurements of these two systems have formed the basis for psychoneuroendocrine stress research starting with Cannon (Lundberg 1984) and Seyle (Goldstein 1995): the former focusing on the emergency function of the SAM axis, whereas the HPA axis was the central part of theories on stress and the general adaptation syndrome of the latter.

Salivary cortisol levels accurately reflect serum-free cortisol, the physiologically active component (Kirschbaum and Hellhammer 1994). Measurement of cortisol in saliva has many advantages over determination from blood samples: sampling is non-invasive and therefore not painful (avoiding the risk of stress-induced modulation of levels), it does not require trained medical personnel and can be repeated at frequent intervals. In addition the collected samples do not need any special treatment, being stable at room temperature for up to 7 days. These advantages mean that samples can be reliably taken in a normal ambulatory setting or in the subjects’ own home during their routine activities. Unlike urinary measurement, detail of dynamic change can be explored using saliva samples as the medium for assay. A moderate to high within-subject stability in the extent of the awakening cortisol response (measured at intervals for up to an hour following awakening) across days and weeks was reported (Pruessner et al. 1997). The response is unrelated to quality of sleep, smoking status, alcohol consumption, time of awakening or post-awakening ambulatory activities (Clow et al. 2004) and phase of the menstrual cycle (Kudielka and Kirschbaum 2003). Increased knowledge has been achieved in the recent years relating both to the well-documented circadian rhytmicity of cortisol output and to the cortisol awakening response (CAR). The CAR, the change in cortisol concentration that typically takes place in the 20–60 min after awakening in the morning, has been recognized over recent years as a distinctive phenomenon in the diurnal profile of cortisol output that is of considerable psychoneuroendocrinological significance (Wust et al. 2000a).

No general association between self-reported mental stress at the workplace and the salivary cortisol response has been observed in studies carried out in this topic. A positive association between magnitude of the CAR and general chronic stress has been observed (Schultz et al. 1998; Pruessner et al. 1999, 2003b), as well as between CAR and work stress (Ockenfels et al. 1995; Kunz-Ebrecht et al. 2004; Schlotz et al. 2004). On the contrary, negative associations have been reported between self-reported mental job stress and cortisol concentration (Steptoe et al. 1998; Yang et al. 2002), as well as no associations between self-reported mental stress and the salivary cortisol response (Evans and Steptoe 2001; Fischer et al. 2000). A plausible explanation for the observed inconsistency may be due to large differences in cortisol sampling procedures, types of mental stressors and the self-report scales used to evaluate them, differences in study design and the absence of control for factors, which are known to or potentially interfere with the cortisol response. Nevertheless, the putative role of the CAR in the regulation of physiological function across the day and its sensitivity to psychosocial variables make it a prime candidate as an intermediary linking mind and health (Clow et al. 2004). Although it is well documented that other steroid hormones are also affected by stress, cortisol is still considered a major indicator of altered physiological states in response to stressful stimulation.

This study was carried out to assess the correlation between self-reported stress assessed by Karasek’s questionnaire and salivary cortisol level in working and weekend days in volunteer subjects employed in a call-centre. Recent studies suggest that most jobs in call centres can be characterized as unskilled work, which some authors called an advanced form of Taylorism (Knights and McCabe 1998). Several studies (Isic et al. 1999; Hutchinson et al. 2000) showed that call-centres operators have low levels of job control, and this condition predicted depression among inbound call agents of a national UK bank in cross sectional as well as in longitudinal data (Holman 2002). Furthermore, in a study among US teleservice centre, lack of job control was associated with musculoskeletal disorders (Hoekstra et al. 1995).

Aims of our investigation were to test the following hypothesis: (1) the cortisol excretion on work day is greater than on weekend day; (2) the cortisol output would be more pronounced in the group of working people defined by high job demand and low control; (3) the cortisol awakening response is the most sensitive neuroendocrine measure of psychosocial job stressors compared with measures of the cortisol excretion during the day.

Materials and methods

Subjects

This cross-sectional study has been conducted in a call-centre setting, where subjects have been recruited through e-mail advertisements and on work side meetings asking volunteers to participate in an investigation on daily activities and cortisol. A participation rate of 25.7% was reached, and among 120 adherent subjects, 15 individuals were excluded on the basis of the criteria of selection (hormone replacement therapy, blood glucose levels higher than 120 mg%, medication for psychiatric disorders). 105 subjects (89 women and 16 males) constituted the final sample.

Study protocol

Saliva samples were collected using cotton dental rolls held in the mouth until satured and then stored in Salivette tubes (Sarstedt, Leicester, UK). Participants were instructed to take seven samples on two working days (one judged as pleasant, the other as unpleasant by the subject itself) and a weekend day, with measures at awakening, 30, 60 min thereafter, at the start of work-shift and then every 3 h (Fig. 1). Participants filled in a diary the actual time of day they had taken each sample. Participants were asked to take the first sample while lying in bed and the second after 30 min without brushing their teeth, eating , drinking or smoking. Tubes were returned to the investigators personally, and cortisol concentration was determined by solid phase radioimmunoassay (LOD = 1.5 nmol/l; intra- and interassay variability were 10 and 14%, respectively). Cortisol values higher than 75 nmol/l were excluded (one participant), because these may be due to altered pH-values or blood contamination of the sample (Steptoe et al. 2004). Separate analyses were carried out for cortisol responses to waking and for cortisol levels along the day. Cortisol responses to waking can be erroneously measured if people fail to take the first sample immediately after waking. Because the increase is rapid, a delay in the first sample means that the “waking” sample is taken on this upward curve, and that the overall increase may be correspondingly reduced.

Fig. 1
figure 1

Sampling schedule

Full size image

In this study compliance with sampling regime has not objectively been monitored: adherence to protocol was judged by calculating the difference between the time participants stated they had taken each saliva sample and the time demanded by the protocol. Individuals with time differences of more than 10 min as well as individuals with any missing value were excluded (36 subjects): excluded subjects are similar to those studied considering sex, age, educational level, marital status, job contract, hours worked per day and work stress measures. Likewise, no differences were showed between adherent and non-adherent subjects on sociodemographic characteristics. We studied 68 workers, 56 females and 12 males and collected 1,428 samples of salivary cortisol. All subjects gave written, informed consent, and ethical approval for the study was obtained from the local research ethics committee.

Work stress measures

Background information was provided by questionnaire, including measures of gender, age, marital status, educational level, hours worked per day, job contract. Psychosocial job stressors were measured according to the job strain model, using the Italian version of the Job Content questionnaire (JCQ) which contains 11 Likert-scaled items (Cesana et al. 2003). This model predicts that the combination of job demand and job control results in different degrees of perceived strain, stress-related risk, and active–passive behavioural correlates of jobs (Karasek and Theorell 1990). There were five items measuring “job demand” (i.e. “Do you have to work very intensively?”), and six items composing the “job control” scale (i.e. “Do you have a choice in deciding how you do your work?”), each of which was rated on a 4-point scale ranging from 1 (“strongly disagree”) to 4 (“strongly agree”). Job demand and job control scales have been obtained by mean of the sum of the items that represent the respective dimension. The Cronbach α for the scales were 0.71 and 0.74 for demand and control, respectively. The scores of each dimension were transformed to range from 25 to 100, where 100 indicated maximum demand and maximum control. The scores were then divided into two categories (high and low) on the basis of median split of each dimension, and four job strain levels (high, active, passive and low) were obtained combining the categories of job demand and job control (see Appendix for details). Table 1 shows the characteristics of the study population.

Table 1 Characteristics of 68 call-centre operators
Full size table

Statistical analysis

The CAR has been quantified by calculation of: the area under the curve (AUC) relative to zero (or ground: AUCG), the area under the curve with respect to increase (AUCI) (Pruessner et al. 2003a) and the mean increase (MnInc) (Wust et al. 2000b). The cortisol excretion during the day has been calculated by mean of the area under the curve with respect the ground (AUCG) using the samples from 4 to 7(AUCG-work), and the area under the curve with respect to the ground using the samples from 1 to 7 (Diurnal cycle) (Pruessner et al. 2003a). Given the skewed distribution of the salivary cortisol measures, logarithmic transformations were performed: the transformation did not generate distribution suitable for parametric analysis, and then all statistic analysis were carried out using non-parametric tests. The cortisol values have been analysed in various ways: firstly, a two-way (time by day) within-subject non-parametric test (Friedman test) was computed to examine differences in each of the seven cortisol samples as well as in CAR, AUCG-work, Diurnal cycle values obtained according to the above-reported calculations. The difference between the two first cortisol samples has been analysed by mean of the Wilcoxon test. The same test has been applied to explore differences between the high and low scores of the demand and control dimensions; the differences among job strain scores have been tested by mean of the Kruskal–Wallis test. Differences on sociodemographic characteristics and work stress measures between included and excluded subjects as well as between adherent and non-adherents were explored by chi-square test. The Spearman rank correlations were used to explore associations between job demand, job control, job strain and cortisol levels. Variables found to be associated with cortisol level were entered into a general linear model (GLM). The categorical predictors were job demand (two levels), job control (two levels), and job strain (four levels). Age, gender, educational level (two levels determined by years of school completed), marital status (two levels: married, not married), working hours per day (two levels: full-time, and part-time), job contract (two levels: confirmed unlimited, temporary contract) were introduced as covariate in the model. The dependent variables were CAR, AUCG-work, and diurnal cycle values. In the first analysis all the factor terms were entered simultaneously into the model, including the covariates, to explore the main effects for all the variables; in the second step interactions between covariates and factors as well as any covariate by covariate interactions were analysed to test the homogeneity of regression slopes assumption. The selection of covariate factors and interactions to be included in the final model was based on a stepwise procedure deletion model: the limit of the significance of covariates to be included in the final model was set at 0.30, that is, variables or interactions not reaching a P-value below 0.30 were eliminated from the model in the next step. This higher limit prevented much of the bias that otherwise may have arisen from this selection method. Significance was set at P < 0.05, two-sides for all the analysis. All data analysis was carried out using SPSS 14.0.

Results

Cortisol values showed the expected diurnal rhythm with high morning and low evening values. The cortisol excretion in weekend was lower than in working days in both sex, but only in females there was a significant effect of day in samples no. 5 (χ2 (2) = 1433, P < 0.001; the median scores on pleasant, unpleasant and weekend day were 13.24, 11.06 and 9.55 nmol/l, respectively), no. 6 (χ2 (2) = 22.30, P < 0.0001; the median scores on pleasant, unpleasant and weekend day were 11.06, 8.87, 4.37 nmol/l, respectively), and no. 7 (χ2 (2) = 29.45, P < 0.0001; the median scores on pleasant, unpleasant, and weekend day were 10.51, 6.41, 1.36 nmol/l, respectively) (Fig. 2). In both sex the CAR values, computed as AUCG, AUCI and MnInc, were higher in work days than in weekend days, but this difference did not achieve the statistical significance. On the other side, the cortisol values showed the typical increase from the awakening to 30 min post-awakening samples in males (6.0 and 23.5% in unpleasant and in pleasant shift, respectively) and in females (17.5 and 21.35% in pleasant and unpleasant shift, respectively): the difference was statistically significant (males: Z = 2.17; P < 0.05; females: Z = 4.88; P < 0.001) (Fig. 3). The cortisol excretion during the day showed higher values in working days, but only in women there was a significant effect of day: in AUCG-work (χ2 (2) = 8.10, P < 0.05; the median scores on pleasant, unpleasant, and weekend day were 6416.77, 4760.70, 4371.72 nmol/l, respectively) (Fig. 4), and in diurnal cycle (χ2 (2) = 15.75; the median scores on pleasant, unpleasant, and weekend day were 11684.07, 13988.17, 8627.05 nmol/l, respectively) (Fig. 5). Although the two categories (high and low) of the job demand and job control dimensions as well as the four levels of job strain exhibited significant differences in score values (Table 1), the work stress measures and cortisol values showed no widespread associations (Table 2). Indeed, job control was inversely related to the sample no. 3 and the CAR (area under the curve with respect to ground) in males only, and job demand was statistically positively associated with samples nos. 2, 3, 5 and the CAR indices in females, but with the sample no. 5 only in males. In both sex no significant association was found between job strain and salivary cortisol values. No significant linear relation between work stress measures and salivary cortisol values could be seen in any of the final general linear models. Salivary cortisol values in the day, measured as AUCG-work and diurnal cycle, and sociodemographic characteristics showed a statistically significant linear relation. In all the final build model, the AUCG-work and the diurnal cycle dependent variables increased as the age increases: in the job control model (Table 3): F = 16.73, P < 0.001 and F = 13.26, P = 0.001, respectively; in the job demand model (Table 4): F = 18.70, P < 0.001 and F = 13.88, P < 0.001, respectively; and in the job strain model (Table 5): F = 14.20, P = 0.001 and F = 8.70, P = 0.005, respectively. Furthermore the covariate gender showed the same linear relation with the diurnal cortisol excretion: the difference between gender group explains a significant size of the variance in both AUCG-work and diurnal cycle cortisol excretion: in the job control model: F = 5.33, P = 0.025 and F = 8.63, P = 0.005, respectively; in the job demand model: F = 5.08, P < 0.028 and F = 12.29, P < 0.001, respectively; and in the job strain model: F = 5.34, P = 0.025 and F = 10.66, P = 0.002, respectively.

Fig. 2
figure 2

General profile of cortisol levels. PS Pleasant shift, US Unpleasant shift, Wd Weekend

Full size image
Fig. 3
figure 3

Values at awakening and 30 min post-awakening (medians and first and third percentile of salivary cortisol concentration)

Full size image
Fig. 4
figure 4

Salivary cortisol values calculated as area under the curve with respect to ground (AUCG) during the day using the samples from no. 4 to no. 7 (AUCD). PS Pleasant shift, US Unpleasant shift, Wd Weekend

Full size image
Fig. 5
figure 5

Cortisol values calculated as area under the curve with respect to ground (AUCG) during the day using the sample no. 1, and samples from no. 4 to no. 7 (diurnal cycle). PS Pleasant shift, US Unpleasant shift, Wd Weekend

Full size image
Table 2 Associations between job strain model and salivary cortisol values using Spearman’s Rho correlations
Full size table
Table 3 Relationship between job control, covariates (age, gender, educational level, work hours per day, job contract, marital status) and salivary cortisol output during the day using general linear model (GLM)
Full size table
Table 4 Relationship between job demand, covariates (age, gender, educational level) and salivary cortisol output during the day using general linear model (GLM)
Full size table
Table 5 Relationship between job strain, covariates (age, gender, educational level, work hours per day, job contract, marital status) and salivary cortisol output during the day using general linear model (GLM)
Full size table

Educational level, marital status, working hours per day and job contract covariates showed a positive linear relation, although without statistical relevance, with the cortisol excretion during the day computed as diurnal cycle in the job control final model (Table 3), indicating that these sources of variance do not play a significant role in the total variance of the cortisol excreted over the day. Examining the job demand final model (Table 4), F statistic (7.42) and its associated significance level (P = 0.009) of the covariate educational level indicated that it has a significant linear relationship with the diurnal cycle (F = 7.42, P = 0.009), suggesting that the values of the dependent variable (cortisol excreted over the day) increases as the educational level increases.

Discussion

The first major prediction directing this research was confirmed: in healthy call-centre operators, the salivary cortisol excretion in work day was greater than in weekend day. On the other side, the results for cortisol output over the day present a conflicting picture, as the cortisol values in the post-awakening phase only were higher in the workday, and in females only this difference reached the statistical significance. These findings are only partially consistent with those of the other two studies exploring this topic (Schultz et al. 1998; Kunz-Ebrecht et al. 2004) that reported higher cortisol awakening response in workday in both sex. Several explanations of the reported differences can be proposed. Firstly, both of the two mentioned studies explored the cortisol awakening response only, failing to analyse the cortisol excretion over the day. Secondly, the size of the male group is too small in our study, as it’s not possible to expect any significant effect. Thirdly, non-compliance with the sampling schedule is a potential confounding factor, as a delay in taking the first sample may mean that cortisol has already begun to rise prior to the first assessment and then the shape of the cortisol awakening response can be altered. The moderate intra-individual stability of the CAR between days observed in this study restricts this possible confusion, as it’s unlikely that a subject did a systematic mistake collecting the first sample.

According to the literature data (Pruessner et al. 1997; Wust et al. 2000b; Edwards et al. 2001) the cortisol awakening response showed to be unrelated to age and gender. On the contrary, the cortisol values in the post-awakening period increased as the age increased, and the cortisol values of females were higher than in males. The latter result supports similar findings (Pruessner et al. 1997; Wust et al. 2000b) suggestive of a rather consistent sex difference on the HPA axis activation. Interestingly, the cortisol awakening response values in this research showed values higher than in the other studies: from 26.4 to 33.5 nmol/l, and from 4.7 to 18.5 nmol/l (Clow et al. 2004), respectively. A plausible reason could be related to the different methods used to analyse the samples: in our research the salivary cortisol samples were analysed by mean of the sensitive radioimmunoassays (RIA) method, while a range of various analytical techniques (enzyme-linked immunoassay, time-resolved fluorescence, luminescence, or enzyme-linked read-out systems) have been employed in the studies, that bounding the comparisons among studies. The alternative explanation could be related to a positive association between job stress and cortisol excretion.

The high and low job demand and job control categories as well as the job strain levels significantly differed on score values, but any significant linear relation has been observed between work stress measures and cortisol output in the various general linear models calculated. These findings do not confirm the results pointed out by the studies employing the job strain model as a measure of work stress (Steptoe et al. 1998, 2000).

When the models took into account the variance accounted for by other factors (gender, age, marital status, educational level, hours worked per day, job contract), the variance that could be attributed to job strain or to the two dimensions (demand–control) of the model was not significantly different from the models in which job strain explains no variance. This lack of association between the measures used in this research to assess the environmental aspects of work stress and the diurnal pattern of cortisol salivary activity is not quite surprising considering that the HPA axis activation is thought to be associated with inability to cope, helplessness, affective distress and perceived uncontrollability (Peters et al. 1998), whereas, the constructs of the job strain model are restricted to the situational component (the assessment of an adverse work environment), and do not take into account the psychological aspect of the stress process (i.e. the cognitive and/or affective characteristics of the working person). Indeed surveys that used psychometric scales assessing both the coping characteristics of working person and the characteristics of the perceived work environment (Steptoe et al. 2000) or that associated with the measures of the psychological aspects of the stress process showed better correlation with biomarkers of stress (i.e. The Perceived Stress Scale; Cohen et al. 1983); Impact of Event Scale (Horowitz et al. 1979); Stress Appraisal (Peacock and Wong 1990); State-Trait Anxiety Inventory (King et al. 1983); Multiple Affect Adjective Checklist (Spielberger 1966).

The results of this study partially confirm our third hypothesis. Indeed, strong associations between job control and the cortisol awakening response indices appeared in females only, whereas males showed a good association between job demand and the total cortisol excretion during the awakening period (computed as area under the curve with respect to ground). No significant associations have been observed between cortisol excretion during the day (calculated both as AUCG-work, and diurnal cycle) and work stress measures. These results agree with the conclusions of the psychoneuroendocrinological literature (Clow et al. 2004) suggesting that the measure of the awakening cortisol response constitutes today the most sensitive tool in the assessment of the physiological response to psychosocial job variables.

Limitations of this study should be recognized. First, the survey design was a typical cross-sectional study performed on a relative limited number of volunteers, so it is not possible to deliver “cause or effects analysis” to describe the complex relationships between the different variables investigated. Secondly, the enrolment in the study of volunteers only could have caused a selection bias: on the other side no significant differences have been observed between adherent and no-adherent subjects about sex, age, sociodemographic characteristics. Third, the investigation was performed with subjects from a single occupation rather than comparing job strain and other work characteristics across occupational groups; therefore the results cannot be generalized to other populations. The reasons for taking this approach is that professions or occupations differing in job strain may also vary in other factors (i.e. social status, physical activity) that might influence neuroendocrinal function independently of work stress. Finally, there was no monitoring of participants’ compliance with the study protocol. Participant adherence to the study protocol is particularly important for cortisol awakening response interpretation, because this aspect of cortisol secretory activity is especially sensitive to deviation from instructions, leading to timing error, so even small deviations from the protocol would have resulted in substantial consequence for the values obtained.

In conclusion, the results of this study, although no apparent support for a close association could be found between work stress measures and salivary cortisol excretion, suggest that a lot of efforts must be reserved to elaborate theoretical models suitable for the assessment of job stress. In particular, more specific measures of mental stress, i.e. those involving uncertainty, novelty, lack of control, distress, anxiety, ego-involvement and helplessness, need to be tested in the assessment of physiological response that employs a promising biomarker of mental stress as the salivary cortisol.