Job stress is a risk factor for a range of adverse effects on health, including major contributors to the overall burden of disease in developed countries, such as cardiovascular disease (CVD) and depression [1–4]. Previous estimates of the proportion of various diseases attributable to occupational exposures have been made by combining population-based exposure prevalence data with estimates of exposure-related increases in specific disease risks taken from epidemiologic studies, yielding exposure-specific 'population attributable risks' (PAR) [5]. PAR is the proportion of disease cases attributable to the exposure in question, or the fraction of disease cases that would not have been observed if the exposure was non-existent [6].

Previous PAR estimates for job stress have utilized Karasek and Theorell's demand-control measures of job stress [7] in relation to cardiovascular disease (CVD) mortality [5, 8, 9]. We found only one estimate of job stress-related PAR for depression, from a Finnish study [9]. Using the prevalence estimates for job strain of 19% for men and 23% for women, this study estimated that 14.6% of deaths related to 'depressive episodes' among men and 9.8% among women were attributable to job strain [9]. While a focus on mortality in work-related PAR estimation may be suitable for CVD and cancer, morbidity estimates would be more useful for mental illness. A recent WHO global quantitative risk assessment initiative generated integrated morbidity and mortality disease burden estimates (disability-adjusted life years) for selected occupational risk factors and associated disease outcomes, but these did not include psychosocial working conditions or mental health outcomes [10].

In the present study, we seek to extend previous work by estimating PAR for depression, and examining variation in job strain exposure and associated attributable disease risk across occupational skill levels as a measure of social class [11]. We have focused on depression because it is a growing contributor to the general burden of disease, and is the most prevalent form of mental illness in developed countries [12]. Examination of exposure and related disease patterns by occupational level contributes to the growing awareness and investigation of psychosocial and other working conditions as important contributors to health inequalities [1, 11, 13].

Effect size estimates for job strain in relation to depression were extracted from the published literature, and combined with exposure prevalence data from the Victorian Job Stress Survey to estimate the proportions of depression attributable to job strain. These proportions were then applied to prevalent cases of depression to yield numbers of job strain-attributable cases of depression. Further, to determine the extent to which regularly collected claims data captures work-related depression or other mental health conditions, we compared our prevalence estimates with the number of compensated "mental stress"-related workers compensation claims. Data sources for each are described below.

Estimated proportions of depression attributable to job strain among working Victorians indicate that job stress is a substantial public health problem. Findings also show that job strain and associated depression risks are inequitably distributed, with workers in lower skill level jobs most likely to be adversely affected, particularly among males. Both in Victoria and in industrialised democracies internationally, poor mental health is disproportionately prevalent among those in lower status occupations, and with lower educational attainment and lower incomes [23–26]. Our findings suggest that job strain may be an important contributor to these mental health inequities.

Our job strain–attributable depression estimates are in the same range to those previously published for job strain in relation to cardiovascular disease (7–16% among men for job strain assessed at a single point [8]) and coronary heart disease (16% in men and 19% in women [9]). Taken together, job strain-attributable risks for these two major chronic diseases represent a substantial preventable disease burden. With respect to the burden of occupational disease more broadly, this study complements and extends previously published Australian mortality estimates for occupational exposure to hazardous substances [27] and morbidity estimates for occupational cancers [28].

The PAR estimates presented are conservative in several ways. The analyses are based on a conservative effect size estimate that is lower than those published for most individual studies. Some cross-sectional studies have found strong associations between job strain and prevalent cases of depression, such as a US study that presented adjusted ORs for job strain and major depressive episode (OR = 7.0), depressive episode (OR = 4.1), and dysphoria (OR = 2.9) among women, with no significant associations among men [29]. Longitudinal studies, by contrast, tend to find smaller effect sizes for incident cases. In a 4-year longitudinal study of depression outcomes in Swedish workers that also accounted for the role of non-occupational factors such as coping ability and stressful life events, job strain remained significantly associated with sub-clinical depression (RR = 2.8) for women, but not for men [30]. The French longitudinal GAZEL study found that the demand/control model measures of high psychological demands (OR = 1.77 men, 1.37 women), low job control (OR = 1.38 men, 1.41 women), and low social support (OR = 1.58 men, 1.29 women) predicted subsequent depressive symptoms at 1-year follow-up [31]. All effects were statistically significant and were unchanged after adjustment for potential confounders. Similar results were found on 3-year follow-up [32]. These investigators did not assess job strain as a predictor variable, but found similar effects of demand/control variables for men and women. Another French (cross-sectional) study of a different population found that job strain was a significant risk factor for depressive symptoms for men (OR = 1.94), but not for women [33]. This finding was largely attributed to job control, which was a significant risk factor for depressive symptoms (OR = 2.84) as well as psychiatric disorders (OR = 2.02) in men [33]. The literature includes a limited number of Australian studies. In a sample of 40–44 year old workers (n = 2249 from low, middle and high status jobs), a statistically significant cross-sectional association was observed between job strain and depression (OR= 2.46) for men and women [34]. Another analysis of a subset of 1,188 employed professionals from the same sample found an OR of 2.54 for depression, again for men and women combined [35]. These cross-sectional Australian estimates are similar to those obtained internationally from longitudinal studies. Finally, we would note that for the purposes of our Victorian working population PAR estimates, effect size estimates would most appropriately come from population-based studies and not studies of specific occupational groups with more restricted generalisability. Of the two studies on which the meta-analysis was based, only one was population-based: the Canadian study in which job strain showed different associations with depression for men (aOR 3.30) versus women (aOR 2.10) (further details in Methods section above). The second was from a longitudinal Finnish cohort of hospital workers, which found no association between job strain and depression. The less relevant study (for the purposes of the analyses presented in this paper) brought the meta-analysis effect size estimate down to 1.8. The net effect of using the meta-analysis effect size estimate in our calculation is that we are more likely to be under-estimating than over-estimating job strain–attributable depression in the Victorian working population.

The PAR estimates presented are also conservative in that the steepness of the gradient observed may be underestimated. It is postulated that job stress has a greater impact on mental health among people in lower socioeconomic positions [11]; as has been empirically demonstrated with respect to hypertension and CVD [11, 36]. In addition, to produce a comprehensive estimate of the effects of job strain on mental health, other associated mental health outcomes would need to be included, such as anxiety, work-related suicide, and behavioural disorders (for example, alcoholism and nicotine addiction) [4, 37, 38]. Further, job strain represents only one of several work-related psychosocial hazards. Others that have been linked to depression include effort-reward imbalance, injustice at work, job insecurity, and bullying [4, 39, 40]. All such hazards would need to be included to estimate the full effect of psychosocial work hazards on depression in particular, and on mental health disorders in general. Thus, we would argue that the impact of all psychosocial working conditions on depression would be higher than the estimates we have presented, and corresponding estimates for all affected mental health outcomes would be higher still.

We also acknowledge certain limitations of this study. Although the survey was designed to be representative of the working population, the sample was taken from telephone listings. This may disproportionately exclude those workers who are in less secure employment, in lower status groups, shift workers and those working longer hours. This bias would lead to underestimation of differentials in exposure and disease distributions across occupational levels. Further, job strain has been studied in relation to various measures of depression, including the CES-D scale [32, 33], Goldberg's 9-item scale [34], major depressive episode as determined by CIDI method [14], and doctor-diagnosed clinical depression [15]. The effect size estimate and the ABS depression prevalence estimates used in this analysis, however, were both based on 12-month depression and included the use of CIDI measurement method. In addition, there are underlying assumptions for the PAR methods used [6]. We combined local exposure prevalence data with an effect size estimate from international studies. This approach assumes that disease aetiology and co-exposures internationally will be similar to the Australian context. This assumption is reasonable because: 1) studies relied upon for effect size estimation came from developed countries, 2) adjusted effect sizes across the various developed countries studied were similar, 3) adjusted (cross-sectional) Australian estimates are consistent with and higher than the effect size used.

We also acknowledge the need for further study of job stress in general and job strain in particular in relation to mental health outcomes. Though the majority of studies of job strain or its component scales of job control and job demands show positive associations, studies of job strain and depression show inconsistent effect sizes, with some studies finding a null association. Substantial differences in the effect size estimates for the contributing studies to the meta-analysis used in the PAR calculations led to its wide confidence intervals (1.815 [95% CI 1.062, 3.101]), and in turn to wide confidence intervals in our PAR estimates. But we also argue from a precautionary principle perspective that the evidence presented is adequate to justify expanded policy and practice responses, particularly given the scale of the problem and the availability of intervention strategies to address it [41]. Several reviews of the intervention evaluation literature have demonstrated that job stress can be effectively addressed by comprehensive or systems approaches–that is, combining individual- and organisation-directed, or work- and worker-directed intervention strategies [42–46].

Finally, comparison of job strain-attributable depression estimates to compensated "mental stress" claims shows that the scale of the problem is grossly underestimated by WC statistics. There are a number of contributing factors for this discrepancy. WC statistics are based on accepted claims. Information regarding the numbers of claims submitted is unavailable. However, given the adversarial nature of the WC system, it is likely that many workers with stress-related illnesses, including depression, have their claims rejected, or may be deterred from filing a claim. Further, when a worker presents to a medical practitioner with depression, the practitioner may or may not identify an underlying occupational causation or contribution. Even if job stress is medically recognized as a contributory factor, there is a documented reluctance amongst Australian general practitioners to initiate WC claims for patients presenting with job stress-related conditions [47, 48]. Further reinforcing this point, a study of Australian human services workers found that psychological claims were four times more likely to be rejected than other kinds of claims, and that the investigation process was reported by claimants to be particularly stressful [49].

This discrepancy between epidemiologic versus insurance-based estimates of job strain–attributable depression demonstrates an urgent need for population-based exposure and health data to complement claims statistics as a basis for guiding policy and practice strategies and priorities in this area. Further, the social patterning of job strain-attributable depression indicates that concerted action by occupational health & safety regulatory agencies and other workplace stakeholders, particularly in lower-skilled work contexts, has the potential to reduce mental health inequities on a population level. Applying a health inequalities intervention framework consistent with the systems approach described above, Whitehead recently described how stressful psychosocial working conditions can be addressed by a combination of strengthening individuals, strengthening communities, improving living and working conditions, and promoting healthy macro-policies [50]. There is an urgent need and a significant opportunity to better link work and health concerns and intervention strategies with the more broadly embraced efforts to understand and address health disparities.

Given the availability of feasible and effective intervention strategies, our findings add to the growing evidence base in support of expanded public health policy and practice interventions to address the adverse effects of job stress. A substantial and inequitable disease burden could be addressed by applying a systems approach to job stress in Victoria and elsewhere. In addition to being a concern for workers, unions, employers, occupational health and safety, and workers' compensation systems, job stress should be a concern for physical and mental health promotion agencies, government public health authorities (and health departments), medical practitioners, community advocacy groups, and others. An optimal public health response to job stress would encompass participation by the full range of stakeholders.