Study objective: To study systematically the separate and combined effects of work stress and socioeconomic position on three measures of health in an unselected working population.
Design: Two exposures (high demand/low control (“job strain”); effort–reward imbalance at work) are related to angina pectoris, depression, and poor self-rated health in a cross-sectional study design in which socioeconomic position was measured by occupational position and educational level.
Setting: Baseline data of a prospective population-based cohort study in Germany, collected between 2000 and 2003.
Participants: 1749 employed or self-employed men and women (36.3% of total sample) aged 45–65 years.
Main results: Effort–reward imbalance and job strain were associated with elevated odds ratios of all three health measures, using logistic regression analysis. The prevalence of poorer health was always highest in subgroups defined by high work stress and low socioeconomic position, with respective odds ratios ranging from 2.30 to 2.98 (95% CI 1.38 to 4.52) for self-rated health, 1.70 to 2.24 (95% CI 1.04 to 3.88) for angina and 2.61 to 8.20 (95% CI 1.53 to 14.15) for depression.
Conclusion: Although stress at work was related to poorer health in the total study group, the strongest associations were consistently observed in men and women with low educational level or low occupational position. Worksite health promotion should be directed primarily towards these target groups.
Statistics from Altmetric.com
A social gradient of health has been documented in working populations of many advanced societies leaving those at the lower end of the social hierarchy in poorer health. Such studies show that with each step one moves up on the social ladder, the better is one’s health.1 2 As differences in medical care and health-related behaviours do not sufficiently account for the observed social gradient, psychosocial stress has emerged as a potential explanatory factor.3–5 More specifically, psychosocial stress at work was found to be associated with elevated risks of incident coronary heart disease,6 7 depression,8–10 and poor self-rated health,11 12 among others. These conditions of poorer health are even more prevalent in lower socioeconomic position groups.1 13 14 It can be hypothesised that stress at work either mediates the association of low socioeconomic position with poorer health or interacts with low socioeconomic position to produce a high risk of poorer health.15
Several studies have tested these two hypotheses in a systematic way, and their evidence is mixed.15–18 This may partly be due to the fact that the prevalence of psychosocial stress at work does not always follow a social gradient (a precondition of testing the mediation hypothesis). Moreover, concerning the interaction hypothesis, adequate statistical analyses have rarely been reported.
In this study, we analyse the hypothesis of an interaction of stress at work with socioeconomic position with respect to three health conditions that are more prevalent among lower socioeconomic position working people: angina pectoris, depression, and poor self-rated health. To measure psychosocial stress at work, two established theoretical models are simultaneously tested, the demand–control model and the effort–reward imbalance model. The focus of the former model was placed on a specific combination of job task characteristics, termed job strain. Job strain is defined by an interaction of high quantitative demands with low decision latitude or low degree of skill discretion.19 The latter model is concerned with the reciprocity of contractual exchange at work in which efforts are compensated by rewards in terms of money, career opportunities including job security, and esteem. Lack of reciprocity (high effort and low reward) elicits recurrent stress reactions because a sense of justice of exchange (reciprocity) is violated.20 The two models that complement each other were shown to explain elevated risks of a variety of diseases.21 22
The following questions were answered using baseline data of a prospective population-based cohort study: (1) is there a social gradient of the three health indicators in this population? (2) what is the social distribution of stress at work? (3) is stress at work associated with poorer health? (4) does stress at work interact with socioeconomic position to produce poorer health?
Data were collected during the baseline examination of the Heinz Nixdorf Recall (HNR) Study, an ongoing prospective population-based cohort study in Germany.23 The study base was the general German population aged 45 to 74 years, living in three cities of an industrialised urban region (Ruhr area). Participants were recruited from a random sample derived from mandatory citizen registries. A total of 4814 men and women agreed to participate (55.8% response rate). The proportion was calculated excluding participants reporting a history of coronary heart disease (previous myocardial infarction or coronary revascularisation, n = 327) who were not eligible for the main analyses of the HNR Study and were also excluded from this analysis.24 According to the study questions, the following analyses were restricted to working men and women. Selection criteria were regular working time of at least 15 hours a week and age below the official retirement age (65 years). A total of 1749 individuals met these criteria (1019 men and 730 women; 36.3% of the total sample).
Information was obtained through standardised interviews and self-completed questionnaires. Computer-assisted personal interviews were conducted and supplemented by paper and pencil questionnaires. Special efforts were invested in the quality control of data collection and data handling, as documented in an external certification (DIN EN ISO 9001; 2000).23 24
As mentioned, this report was restricted to three indicators of health that were shown to be more prevalent among lower socioeconomic position individuals: self-rated health, angina pectoris and depression. Self-rated health was measured by a standard five-point Likert scale item (“very good”, “good”, “fair”, “poor” and “very poor”, with less than good indicating poor health.25 Angina pectoris was assessed using the Rose questionnaire.26 A diagnosis was made if chest pain or discomfort was present when walking uphill or hurrying and/or walking on level ground at normal speed, and disappeared within 10 minutes or less. For the assessment of depressive symptoms we applied a short version (15 items) of the Centre for Epidemiological Study–Depression Scale (CES-D).27 28 The internal consistency of this scale was 0.83 for men and 0.88 for women. We calculated a sum score of all items with the range of 0 to 45. As the score was not normally distributed, however, we decided to calculate logistic instead of linear regression models. For this purpose the score was dichotomised at the cut point of a score greater than 17, indicating a depressive syndrome.
Additional information was collected on body weight (obesity; body mass index >30), cigarette smoking (non-smokers/ex-smokers/current smokers) and physical activity (regular sport/no sport) and negative affectivity (high/low).
Socioeconomic position was defined by two alternative indicators, level of education and occupational grade. Education was classified according to the International Standard Classification of Education29 as total years of formal education, combining school and vocational training. The continuous variable was grouped into four categories, with the highest category of 18 and more years of education (equivalent to a university degree) and the lowest category of 10 and fewer years (equivalent to a basic school education and no vocational training). Occupational grade was classified into the following categories: (1) high managerial/director; (2) manager/supervisor; (3) self-employed; and (4) employee.
The demand–control and the effort–reward imbalance models were measured by validated scales. Effort–reward imbalance was measured by 17 items.30 A ratio of the two scales “effort” and “reward” was composed according to the formula: e/(r × c), where “e” is the sum score of the effort scale, “r” is the sum score of the reward scale and “c” represents a correction factor for different numbers of items in the nominator and denominator. The higher the ratio, the more work stress is experienced. This ratio was divided into quintiles, and a dichotomised measure was constructed (scores of the upper quintile of the effort–reward ratio indicating the “high stress” group versus the remaining four groups). The scale measuring the model’s intrinsic component “overcommitment” was not included in this comparative analysis of two work stress models because no intrinsic information was available from the demand–control model.
Concerning the measurement of the second model,31 the control scale (decision latitude) was combined with the demand scale to define the following four categories: (1) “high job strain” with values on the “demand” scale above the median and with values on the “job decision latitude” scale at or below the median; (2) low strain, based on simultaneously low job demands and high job decision latitude; (3) passive, based on simultaneously low job demands and low decision latitude; and (4) active, based on simultaneously high job demands and high job decision latitude. For the analysis (logistic regression) we again composed a dichotomised variable in which the first category (“job strain”) was evaluated against the remaining three groups.
Data were first cross-tabulated and descriptive measures were calculated. Calculations refer to the association of the two work stress models with socioeconomic position and of the three health measures with socioeconomic position. Next, logistic regression models were calculated to estimate odds ratios separately for the two work stress models for different levels of covariate adjustment. Model I displayed crude estimators, whereas model II was adjusted for age and gender. In model III socioeconomic position and lifestyle risk factors were included. In the case of self-rated health and angina pectoris odds ratios were also adjusted for negative affectivity in order to control for a possible bias of self-reported data.32 33 Negative affectivity was not included in the third model in the case of depression because these two constructs were highly interrelated at the measurement level.
To test the interaction of two exposures, socioeconomic position and work stress, composite variables were constructed. To this end, level of education was dichotomised with primary and secondary degrees as the lower category. Occupational grade was dichotomised on the basis of the occupational hierarchy into high (high managerial/director, manager/supervisor and self-employed) and low (employee) grade. The dichotomised exposure variables, work-related stress and socioeconomic status, were combined in the following way: (1) no work stress present, high socioeconomic position; (2) work stress absent, but low socioeconomic position; (3) work stress present, but high socioeconomic position; (4) both work stress and low socioeconomic position present.
In brief, synergetic interaction refers to the extent to which the joint effect of two risk factors of disease exceeds the independent effects of each of the factors. As a test of statistical significance for additive interaction, the synergy index was calculated for each model.34 Significance of trend was tested by calculating the 95% confidence interval (CI) for the synergy index, based on the normal approximation and the variance and covariance estimates of the fitted coefficients for the indicator variables from the logistic regression models, according to Hosmer and Lemeshow.35 When the lower limit of the corresponding 95% CI for the synergy index was greater than 1.0, the joint effect of the two exposures exceeded their additive effects. Although the prevalence of health indicators differed in men and women, logistic regression analyses revealed no consistent differences of results according to gender. Therefore numbers are given for the total group of working men and women. Analyses were calculated using the SPSS statistical package 12.01 R-version 2.3.1. (SPSS Inc, Chicago, Illinois, USA).36
Table 1 gives the main descriptive characteristics of the study population. The proportion of participants with high education or occupational status was greater among men than women. Conversely, depression, angina and poor self-rated health were more prevalent among women.
Results on the first two questions are displayed in table 2. A social gradient of health, as measured by the three indicators, was observed in this sample with poorer health in the lowest compared with the highest socioeconomic position group, especially so in the case of educational level. The second question concerned the distribution of the two work stress models according to socioeconomic position. A social gradient of job strain was evident with regard to education. Working men and women in the lowest compared with the highest educational level were almost three times as likely to exhibit job strain. Effort–reward imbalance was not consistently associated with socioeconomic position. Middle managers exhibited a higher prevalence than lower status employees, and the same held true for secondary versus primary education.
Third, we were interested in associations of the two models of stressful work with health. Although logistic regression models with all quintiles of effort–reward imbalance and with the four demand–control categories were calculated we present results, for the sake of brevity, for the abbreviated (ie two-categorical) work stress variables only. As can be seen from table 3, the odds ratios of either work stress model are significantly elevated with respect to the three health measures in the unadjusted and in the fully adjusted models (with the exception of job strain and angina pectoris in model III). The strongest association is observed between stressful work conditions measured through effort–reward imbalance and depression.
The fourth question concerns the test of an interaction of work stress with socioeconomic position, assuming joint effects of stress at work and low socioeconomic position on poorer health. To this end, multivariate logistic regression analyses based on a combined variable of socioeconomic position and work stress were performed for each health measure. As indicated in table 4, odds ratios of poorer health were found to be consistently highest in the low socioeconomic position/high work stress groups with odds ratios varying between 1.7 and 8.0. In all cases, confidence intervals were greater than 1.0. Again, the strongest association was observed with regard to effort–reward imbalance and depression. This is the only case in which the synergy index indicated that the combined effects of being in a low occupational position and experiencing stressful work exceeded their additive effects on poorer health. Comparable findings were observed for the interaction between education and work stress, although the odds ratios were lower in the fully adjusted models for the low socioeconomic position/high work stress combination (results not shown).
What is already known on this subject
Socioeconomic position is associated with a variety of health outcomes, including those of the present study.
Stress at work, as measured by the demand–control and the effort–reward imbalance models, explains elevated risks of diseases, including those of the present study.
Few studies have tested the interaction of work stress and socioeconomic position in explaining health.
What this study adds
Findings demonstrated poorest health in all low socioeconomic position/high work stress subgroups, with a significant synergy effect of effort–reward imbalance and socioeconomic position on depression.
Results support theory-based worksite health promotion measures, with special emphasis on less socially privileged employment groups.
As a combined exposure of work stress and low socieconomic position is associated with increased vulnerability, worksite health promotion should be directed towards these groups.
Concerning our research questions, four main findings are summarised below. First, the results show a social gradient of all three health indicators, with individuals in a lower position reporting poorer health. Second, concerning the distribution of the two models of an adverse psychosocial work environment according to socioeconomic position, stress at work is not always more prevalent among those in lower positions. For example, middle managers exhibited a higher prevalence of stress in terms of effort–reward imbalance compared with less qualified employees. Third, two complementary work stress models (demand–control and effort–reward imbalance) were each associated with the three indicators of health under study (angina pectoris, poor self-rated health and depression). Finally, the findings consistently demonstrated poorest health in the low position/high work stress subgroups, with a significant synergistic effect of effort–reward imbalance and low socioeconomic position on depression. Although stress at work is not always more prevalent among lower position individuals, its adverse effects on health are most obvious among those with the lowest standing in the social hierarchy, thus lending support to the susceptibility hypothesis of social inequalities in health.
This hypothesis maintains that individuals of low socioeconomic position possess fewer resources for coping with the burden of stressful work, partly because multiple exposures overtax their efforts, and partly because their acquired coping abilities are less efficacious.37 38 For example, low socioeconomic position individuals were found to have low control beliefs and poor skills of mastery more often than individuals of higher social position.39 As a result, they may more often be deprived of experiences of success, reward and control. These experiences in turn are associated with prolonged or more intense stress reactions and delayed recovery, ie with psychobiological reactions that contribute to increased illness susceptibility. These psychobiological reactions were identified with particularly high frequency in stressed men and women of low occupational standing.40–42
The fact that we were not able to provide a direct test of the susceptibility hypothesis by including respective psychological and biological data in a longitudinal design must be considered a major limitation of the study. Data were drawn from baseline measurements of a large observational cohort study, in which future measurement waves will provide opportunities of testing the hypotheses in a longitudinal design. Another limitation was the relatively low response rate. Despite the complex health examinations including physical examinations by physicians and X-ray, the overall response in the HNR Study was similar to response proportions of other German population-based studies.24 Stang et al24 also conducted non-responder analyses and found no support for a substantial response bias. A further limitation concerns the lack of externally validated morbidity measures. Moreover, reported results may be biased to some extent as a result of common method variance. In particular, one can argue that lower socioeconomic position in contrast to higher socioeconomic position individuals are more likely to report poor health and, simultaneously, to report a high level of work stress. We did not, however, find an association of work stress in terms of effort–reward imbalance with socioeconomic position, nor did we observe a similar size of effects on health produced by either work stress model. In addition, by including negative affectivity into multivariate analysis, we adjusted for an established indicator of potential reporting bias, as was done in previous analyses.43 Concerning the distribution of educational levels in the study sample, it should be noted that the low proportion of primary education and the high proportion of participants with higher education are comparable to the German working population of this age group. The reasons for the observed distribution are the “dual” education system in Germany (a mix of mandatory school and voluntary vocational training) and an “educated worker effect” as a result of a high proportion of early retirement among less qualified workers.
These limitations are balanced by several strengths. First, we selected three health measures that were previously shown to be more prevalent among lower socioeconomic position individuals, and we confirmed a social gradient of these conditions in a large working population, applying two alternative indicators of socioeconomic position. Second, we analysed associations of two established complementary models of psychosocial stress at work with these health indicators, using psychometrically validated scales for their assessment. The fact that independent effects on health of the two measurements of work stress were replicated adds to the robustness of the available evidence. Moreover, with one exception, associations remained statistically significant in the final model adjusting for a number of relevant confounders. Third, to our knowledge, this is the first study to test the hypothesis of an interaction of work stress in terms of the two models with socioeconomic position in explaining elevated health risks. In all instances we found relatively high odds ratios of poorer health in the low socioeconomic position/high work stress subgroups, although a significant synergy effect exceeding the additive effects was observed in the case of effort–reward imbalance and depression only.
In conclusion, this study found evidence of an association of psychosocial stress at work with poor self-reported health, depression and angina pectoris. These associations were particularly strong in the case of low socioeconomic position working men and women. Results support theory-based worksite health promotion measures with special emphasis on less socially privileged employment groups.
The authors are grateful to Pablo Verde, MSc, and Andreas Rodel, MA, for their statistical advice. They also thank the Heinz Nixdorf Foundation (Chairman Dr jur G Schmidt, Essen, Germany) for sponsoring the Heinz Nixdorf Recall study and the German Research Council for supporting the additional assessment of psychosocial factors (DFG SI 236/8-1; 9-1). The authors are indebted to the investigative group and the study personnel of the Heinz Nixdorf Recall Study.
Funding: The baseline screening of the Heinz Nixdorf Recall study was funded by the Heinz Nixdorf Foundation (Chairman Dr jur G Schmidt, Essen, Germany). The present project was additionally funded by the German Research Foundation (DFG SI 236/8-1; 9-1).
Competing interests: none.
Ethics approval: The research protocol of the Heinz Nixdorf Recall study has been approved by the local ethics committee (University Duisburg-Essen) and informed consent has been obtained from each participant.
Contributions: NW conceived and designed the research, analysed and interpreted the data, and contributed to writing the paper. ND acquired data, performed the statistical analyses and made critical revisions of the manuscript. SM and AS acquired data and made critical revisions of the manuscript. RE and K-HJ handled funding and supervision, and made critical revisions of the manuscript. JS drafted the manuscript, and handled funding and supervision.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.