Sanne, Bjarte MD; Mykletun, Arnstein MA; Dahl, Alv A. MD, PhD; Moen, Bente E. MD, PhD; Tell, Grethe S. PhD, MPH
Anxiety and depressive disorders are major public health problems, and there is a growing awareness of the economic burden imposed by these disorders. 1,2 Absenteeism and reduced capacity while at work account for a considerable part of this burden. Anxiety and depressive disorders are underdiagnosed and undertreated, although effective treatment exists. 3,4 To reduce the suffering and costs of these disorders, both primary and secondary prevention are needed.
Job conditions are a major source of environmental influence and may affect the development of anxiety and depressive symptoms. 5 The working place may therefore be a strategic arena for interventions against these common health problems. However, knowledge on the relationship between work life and anxiety and depressive disorders and thus about populations at risk is too scarce to develop targeted interventions. 6
One way of identifying groups at risk is to study possible differences in occupational distribution of anxiety and depression symptom load. Differences between occupational groups could result from: 1) a selection into specific occupations (certain personal characteristics may explain both occupational choice and levels of anxiety/depression); 2) a selection out of certain jobs (work conditions may cause exclusion of anxiety/depression prone individuals); and/or 3) a consequence of wear and tear (ie, unfavorable conditions) in the job. The Job Demand–Control (-Support) Model, which has dominated research on occupational stress for more than 20 years, is a wear and tear model. 5 Other examples are Siegrist’s Effort-Reward Imbalance Model 7 and the burnout theories of Cordes & Dougherty 8 and Maslach et al. 9 To answer which of these three factors (selection into, selection out of, and wear and tear) are involved and how, a longitudinal design is required. As far as we are aware, no such study has been published.
To our knowledge, only five research groups have published results from large epidemiological studies on the relationship between depressive disorders/depression symptom load and occupational groups. 10–14 All were conducted in the United States or Canada, and all found differences in prevalence/symptom load among occupational groups. However, the findings differ considerably regarding which groups have the highest prevalences/symptom loads. Only two of the five studies included anxiety. 12,14 Because anxiety and depressive disorders are both frequent and often comorbid, it is important to study both conditions at the same time. 15
Three of the five studies included only major occupational groupings. In addition, different grouping procedures were used, making comparisons of the findings cumbersome. None of the studies used the International Standard Classification of Occupations (ISCO-88). 16 Further shortcomings of the previous studies include lack of industrial classification (which could be useful in specifying groups at risk) and lack of information on other aspects of work life, such as income and work hours. Finally, none of the studies stratified on gender, which is highly recommended because of gender differences in the prevalence of anxiety and depressive disorders as well as in the distribution of occupations. 5,17
The aim of this study was to examine levels of anxiety and depression symptom loads in relation to standard occupational classification, in a large Norwegian population-based sample of men and women living and working in both urban and rural settings. The following research questions were posed:
* Do the levels of anxiety and depression differ between occupations?
* Which occupational groups differ significantly from the average levels of anxiety and depression?
* How strongly is occupation associated with levels of anxiety and depression?
* To what extent may the association between occupation and levels of anxiety and depression be explained by workplace characteristics, demographics, individual lifestyle and somatic health problems?
Materials and Methods
The Hordaland Health study 1997–1999 (HUSK) was conducted as a collaboration among the National Health Screening Service, the University of Bergen, and local health services. The study population included the 29,400 individuals living in Hordaland county of Western Norway born between 1953 and 1957. A total of 8598 men and 9983 women participated, yielding a participation rate of 57% for men and 70% for women. The study also included 2291 men and 2558 women born 1950–1951, with participation rates of 73% and 81%, respectively. The present study encompassed only workers (defined as having worked at least 100 income-giving hours the preceding year) with valid Hospital Anxiety and Depression Scale (HADS) ratings, a total of 17,384 individuals, which constituted 85.7% of all working participants.
Data collection in HUSK was performed in two steps. The first step, which was identical for all participants, included a self-administered questionnaire and a health examination. In the second step, the two age cohorts 1950–1951 and 1953–1957 were given different questionnaires. The analyses of research question 4 were, therefore, carried out without those born in 1950 or 1951. Thus, the number with valid information on all variables was 11,910 (5821 men and 6089 women), constituting 85.5% of the men and 79.5% of the women in the main sample who were born from 1953–1957. This subsample did not differ significantly from the main sample as to the HADS scores (Table 1) and showed a similar, although slightly weaker, association with skill level.
Anxiety and Depression.
Levels of anxiety and depression were assessed by the HADS, which has been found to perform well in assessing symptom load and caseness of anxiety and depressive disorders in both somatic, psychiatric and primary care patients as well as in the general population. 18
Valid HADS scores were defined as having answered at least five of seven items on both the anxiety (HADS-A) and the depression (HADS-D) subscales. Each item was scored on a four-point scale from zero to three, and the item scores were added, giving subscale scores from 0 (minimum symptom level) to 21 (maximum symptom level). 19 Extrapolation of the scores was done when five (applying to 84 individuals for HADS-A and 77 for HADS-D) or six (applying to 703 individuals for HADS-A and 1952 for HADS-D) questions on a subscale had been scored.
Caseness (ie, ‘possible cases’ of HADS anxiety and/or depressive disorders) was defined as a score of eight or above on HADS-A and/or HADS-D, as this cut-off level has been shown to give an optimal balance between sensitivity and specificity on receiver operating curves. 18
The self-administered questionnaires included an open-ended question of main occupation and industry, manually classified according to Standard Classification of Occupations, ISCO-88(COM), 20 and Standard Industrial Classification, SIC94. 21 ISCO-88(COM) has a four-level hierarchical structure and is divided into 10 major groups (eg, professionals), 31 submajor groups (eg, life science and health professionals), 108 minor groups (eg, health professionals), and 353 unit groups (eg, nutritionists). In the ISCO-88, classification is performed according to two principles: skill level and skill specialization. There are four skill levels, based on the International Standard for Classification of Education. Occupations are classified according to which technical and formal skills that are normally required. The skills do not have to be achieved through formal education but can be a result of informal training and experience. The four levels of skills are occupations that normally require: 1) no more than 9 years of primary education; 2) 1 to 3 years of secondary education; 3) 1 to 3 years at university or college level, and 4) education equivalent to a first or postgraduate university degree, or college examinations based on a similar length of study. Thus, the skill level normally required for major occupational group 2 (professionals) is level 4, for group 3 (technicians and associate professionals) level 3, for most of the occupations in the groups 6 (agricultural/forestry/fishery workers), 7 (craft and related trades workers) and 8 (‘plant and machine operators and assemblers’) level 2, and for group 9 (elementary occupations) level 1. There are no requirements regarding formal education for group 1 (legislators/senior officials/managers). Skill specialization is defined by 1) the field of knowledge required, 2) the tools/machinery used, 3) the materials worked on or with, and 4) the types of goods and services produced.
Industrial classification was included in the study as a complement to occupational classification, to increase the specification of groups at risk and enable tracking of patterns and tracing of specific groups. Industrial classification entails grouping homogenous activities as much as possible, ie, classifying production units according to their economic activity. SIC94, being independent of the ISCO-88, has a six-level hierarchical structure, and is divided into 17 sections, 31 subsections, 60 divisions, 222 groups, 503 classes, and 658 subclasses.
The following work related variables were included because of their possible confounding effects on the association between major occupational grouping and HADS scores: Number of paid work hours per week, shift work, night work or duties, the opportunity to use one’s abilities at work and level of physical activities at work (mainly sedentary/work demanding much walking with or without much lifting/heavy manual labor).
Demographics, Individual Lifestyle, and Somatic Health Problems.
The following possible confounders were also examined: Level of education, the household’s total income, marital status, parity, daily smoking (yes/no), alcohol consumption (alcohol units per fortnight, categorized into total abstinence/low-risk consumption/high-risk consumption, the latter defined as consumption above 21 units per week for men and 14 units per week for women), leisure time physical activity (categorized into three groups: 1–2 points/3–5 points/6–8 points, using a scale from 1 point [no exercise] to 8 points [3 or more hours per week of both heavy and light exercise]), perception of having “enough good friends” (yes/no), musculoskeletal problems (pain and/or stiffness the last 12 months, at least 3 months continuously, as well as resulting in reduced work capacity or sick leaves), chronic somatic diseases (having or having had one or more of the following: myocardial infarction, angina pectoris, hypertension, stroke, asthma, chronic bronchitis, diabetes mellitus, or multiple sclerosis), and the physical summary score of the quality-of-life scale SF-12 Health Survey (the higher score, the better is the reported physical health). 22 Body mass index (weight in kg/height in m2) was calculated from measured height and weight.
Table 1 gives an overview of the most important statistical procedures used. All analyses were stratified by gender because HADS-A and HADS-D scores as well as the distribution of occupations differed between men and women. When heteroscedasticity occurred, the univariate analyses of variance (ANOVA) were repeated using the nonparametric Kruskal–Wallis test. This also applied for the testing of possible differences in HADS scores between submajor occupational groups. Only one-way ANOVA results are shown in Tables 2 and 5, but unless it is otherwise is stated, the two methods gave equivalent levels of significance.
Occupational (and industrial) groups that statistically differed significantly from the average HADS scores were focused upon. Groups with ‘higher ’ HADS-A and HADS-D scores filled the following criterion: The lower limit of the 95% confidence interval of the mean HADS subscore was higher than the mean HADS subscore of the corresponding total sample. Correspondingly, the ‘higher’ limit of the 95% confidence interval of ‘lower ’ score groups were lower than the mean HADS subscore of the corresponding total sample. To prevent type 1 errors from multiple comparisons, post hoc tests were performed: When homoscedasticity occurred, Scheffé’s test was used, whereas Tamhane’s T2 test was conducted when heteroscedasticity occurred. All HADS-A and HADS-D scores throughout the article refer to mean scores for the current groups.
ANOVA was used to adjust the HADS-A and HADS-D scores of the major occupational groups for possible confounders, which were tested in bivariate analyses. The variables whose categories differed significantly both across major occupational groups (cross-tables) and in HADS-A and/or HADS-D scores (one-way ANOVA) were further included in two-way ANOVA. Then, for the differences in HADS scores that could not be explained by a single factor, two models were made for the simultaneous adjustment of several explanatory factors. The models were based on the size of these variables’ explained variance in one-way ANOVA analyses with the corresponding HADS score as the dependent variable. Significance level was set to P = 0.05 with two-sided tests. The analyses were performed by means of SPSS for Windows, version 11.0.
The study protocol was cleared by the Regional Committee for Medical Research Ethics of Western Norway and approved by the Norwegian Data Inspectorate.
Before going more thoroughly into the main results shown in Table 1, some general findings will be mentioned. First, women had a significantly higher HADS-A score than men, whereas the reverse was seen for HADS-D score. Second, men were more evenly distributed throughout the occupational (and industrial) classification systems than women. For instance, the five largest of the 31 submajor occupational groups constituted 49% of all the male workers and the eight largest 66% whereas the corresponding numbers for women were 64% and 81%, respectively. Both genders showed a clear tendency toward a traditional gender pattern in the distribution of occupations. Finally, for both occupational and industrial groups, more male than female groups had ‘higher’ or ‘lower’ HADS scores, and more groups had ‘higher’ or ‘lower’ HADS-D scores compared to HADS-A scores (Tables 2 and 3).
Association Between HADS Scores and Skill Level
HADS levels were clearly associated with skill level demands in major and submajor occupational groups: ‘lower’ scores were merely found in groups characterized by high skill levels, whereas ‘higher’ scores only were found in groups with low skill levels, except group 3.3 (teaching associate professionals) for HADS-D scores in women. This pattern was strongest for HADS-D and weakest for HADS-A scores and stronger in men than in women. For HADS-D scores in men in major occupational groups, groups 6 through 9 had ‘higher’ scores whereas groups 1 through 3 had ‘lower’ scores. The examination of industrial sections and divisions showed a similar association between HADS scores and skill levels, although somewhat weaker than for occupational groups. To test the validity of the association between HADS scores and skill level in major occupational groups, the following procedures were undertaken:
For the HADS-D scores, the rankings of means of Table 2 and the mean rank calculated by the Kruskal–Wallis test (which tests if the groups come from populations with the same median) were compared. The rankings were identical, except for group 0 in men (the smallest male group), and the exchange of order of groups 6 (the smallest female group) and 9 for women.
Post hoc tests were performed to investigate which groups differed from which and to account for multiple comparisons. For HADS-A scores, none of the differences between the groups were significant, for neither gender. For HADS-D levels in men, all groups differed from at least one other group (Table 4). The findings showed a strong association between HADS-D scores and skill levels. In women, group 9 differed from groups 1–5.
Thus, HADS-D scores in men showed the strongest association with skill levels, followed by HADS-D scores in women, HADS-A scores in men, and finally HADS-A scores in women, where the association was weak. To investigate if the association between HADS scores and skill levels may be of clinical significance, the following analyses were performed:
Logistic regression was used to examine possible differences between major occupational groups regarding caseness (Table 5). The groups did not differ significantly as regards HADS anxiety caseness in men but differed as to anxiety caseness in women and depression caseness in both genders. The association between depression caseness and skill level was strong, especially for men, while it was rather weak as for anxiety caseness in women. The odds ratios (ORs) for depression caseness in men was three times as high for groups 9 and 6 compared with group 1, and in women 2.6 times as high in group 8 as in group 1.
Explained variance (R2) was used to examine the strength of the association between HADS levels and major occupational grouping. The R2s were 0.4% and 0.3% for HADS-A and 1.8% and 0.5% for HADS-D in men and women, respectively. This effect was within the range of the effect of level of education (0.3–0.9%), except for HADS-D in men, where the R2 of ‘occupation’ was twice that of level of education (0.9 versus 1.8%). The R2 values of marital status and chronic somatic diseases for HADS-D in men were 0.7% and 0.2%, respectively.
Specific Occupational and Industrial Groups
Major Occupational Groups.
Major occupational group 9 (elementary occupations) was the only group which consistently showed ‘higher’ HADS scores (Table 2). Removing this group from the ANOVA analysis rendered the differences in HADS-A scores in women not significant. Group 1 (legislators/senior officials/managers) had ‘lower’ HADS-D scores for both men and women, whereas group 3 showed ‘lower’ scores for both HADS-A and HADS-D in men.
Submajor Occupational Groups.
Submajor occupational groups also differed significantly regarding both HADS-A and HADS-D scores for both genders (Table 3). For both men and women, several of the groups with ‘higher’/‘lower’ HADS-A scores also had ‘higher’/‘lower’ HADS-D levels. Groups 1.2 (corporate managers of large/medium sized enterprises) and 3.2 (life science and health associate professionals, also including nurses) consistently showed ‘lower’ HADS scores.
Three of the six male groups with ‘higher’ HADS-D levels belonged to major group 7 (craft and related trades workers). However, agricultural workers (group 6.1) had the highest HADS-D level, significantly ‘higher’ than the group with the second highest score. All the three sub-major groups belonging to major group 1 (legislators, senior officials and managers) were among those with ‘lower’ scores. In women, group 9.1 (services elementary occupations), which also had ‘higher’ HADS-A scores, had significantly ‘higher’ HADS-D score than the other group on the list, group 3.3. Although means and 95% confidence intervals for male and female teaching associate professionals were very similar, group 3.3 was one of the six male groups with ‘lower’ HADS-D scores.
Industrial Sections and Divisions.
In men, blue-collar workers groups, such as manufacturing and construction workers, had ‘higher’ HADS-D scores. Agricultural workers had the highest HADS-D score on a sectional level, significantly ‘higher’ than the other sections. Women working in education had ‘higher’ HADS-A and HADS-D scores. Also, female manufacturing workers had ‘higher’ HADS-D scores. Manufacturers of food products and beverages had ‘higher’ and the highest HADS-D score on a divisional level. Among the five sections and seven divisions with ‘lower’ HADS-D scores in men, were educational workers and those employed in health and social work. Women employed in health and social work had ‘lower’ HADS-A and HADS-D scores.
To What Extent Could the Association Between Occupation and HADS-A and HADS-D Scores Be Explained by Other Variables?
When examining the potential confounders as described in the Statistics section, the variable chronic somatic diseases did not differ across major occupational groups, and HADS-scores did not differ for those with and without shift work, night work or duties. These variables were therefore not included in the adjustment analyses.
Two-way ANOVA showed that the differences in both HADS-A and HADS-D scores in women could be explained by the question “how often are you able to use your abilities in your work?” The differences in HADS-A scores in men could be partially explained by the category household income.
The differences in HADS-D scores in men could not be explained by a single factor. Therefore, two models were used that included the three and five variables with the highest R2 values, respectively (in descending order): reporting to have “enough good friends,” “how often are you able to use your abilities in your work,” “PCS” (the physical summary score of the SF-12), “household income,” and “leisure time physical exercise.” In both models, the effect of major occupational groups on HADS-D remained significant, and groups 6 (agricultural/forestry/fishery workers) and 9 (elementary occupations) had the highest HADS-D scores.
The study showed significant differences in HADS-A and HADS-D scores between major and between submajor occupational groups. HADS levels showed a distinct and inverse relationship with skill levels, most strongly observed for HADS-D scores in men. The equally strong relationship between skill level and depression caseness indicates that the findings are of clinical significance, especially in men. Elementary occupations consistently showed ‘higher’ anxiety and depression scores, whereas male agricultural workers had the highest depression scores. The differences in HADS-A and HADS-D scores in women could be explained by differences in possibilities to use one’s abilities at work, whereas the differences in HADS-A scores in men could be partially explained by different income levels. However, the differences in HADS-D scores in men could not be explained by factors measured in the study.
Levels of anxiety and depression were investigated simultaneously, which is important because of the high correlation between anxiety and depressive symptoms. 18 The large sample size allowed stratification on gender and the investigation of subgroups. Both the considerably different distribution pattern of occupations and the different HADS scores between genders underscore the importance of gender specific analyses. The inclusion of industrial in addition to occupational classification further facilitated the specification of groups at risk. In addition, information on various work related factors enhanced the understanding of the relationship between HADS scores and occupation.
The most important limitation of the study is its cross-sectional design. However, the study served to identify populations at risk, and may thus form the basis for secondary prevention. The narrow age range reduce the generalizability of the findings. However, because of the large sample size and the age homogeneity, a more thorough investigation of subgroups was possible. Some of the occupational groups, particularly some on the submajor level, are small, thus increasing the risk of type II errors. However, we were nevertheless able to reject the hypothesis of no differences in HADS scores between occupational groups.
The moderate participation rate warrants some remarks: Nonresponders to surveys have been found to have higher prevalences of mental disorders. 17,23. Furthermore, the ECA study showed that “uneducated” men had increased risk of being lost to follow-up, and that “unskilled” male workers had higher lifetime prevalences of psychiatric disorders than “skilled” workers or workers with higher occupational status. 24 The healthy worker effect is well-known, 25 and also in our material unemployed had considerably higher anxiety and depression levels than workers. Thus in the present study it is probable that: 1) the proportion of working individuals was higher among those participating than in the non-participating group; and 2) our findings of higher HADS scores in low skill occupations, particularly in men, would have been strengthened by a higher participation rate.
When adjusting for possible explanatory factors, the subsample used was considerably smaller than the main group of participants. However, the subsample did not differ significantly from the main sample with regard to HADS scores, supporting the generalizability of the findings in the subsample.
The HADS-D scores showed moderate skewness, and the major occupational groups showed heteroscedasticity for HADS-D scores. However, because of the almost identical results from parametric and nonparametric tests (significance levels and ranking of groups), it was considered acceptable to use the mean as a measure of central tendency, and to proceed with post hoc tests and ANOVA analyses for the adjustment of possible confounders. Comparing the groups, it could be argued that instead of using the defined ‘higher’ and ‘lower’ scores, it would be more consistent to compare the mean and 95% confidence interval of a particular subgroup with those of the remainder of the whole group. However, applying this method gave almost identical results, if anything, our findings were strengthened.
The HADS does not provide defined diagnoses of anxiety and depressive disorders. However, because of the healthy worker effect, it is to be expected that the main part of the variation in HADS scores in our sample was found in the sub-clinical area. This strengthens the argument for comparing levels of symptom load in addition to comparing prevalences of cases.
Information acquired through self-administered questionnaires may be biased toward the negative in depressed individuals. 26 Thus subjective responses, such as self-reports on having the opportunity to use one’s abilities at work, could be skewed by a depressed mood.
North American studies have shown high levels of depressive symptoms/high prevalences of depressive disorders in some blue-collar occupations 12,13 and low prevalences of anxiety and depressive disorders in professional and managerial–administrative workgroups. 12–14 However, none of the studies have shown as strong a pattern between anxiety and depressive symptoms/disorders and low skill level jobs as in the present study. This may be due to the American studies using other occupational classification systems than ISCO-88, whose structure is based on differences in skill level. It is of note that the ECA study found the lifetime prevalence rate of any psychiatric disorder in men to be higher among unskilled than among those skilled or with higher occupational status. 24
Although occupational classification explained only a modest part of the variation in HADS scores, the R2 of occupation was two times that of level of education, 2.6 times that of marital status and nine times that of chronic somatic diseases for HADS-D scores in men, factors which are all known to influence the prevalence of depression. 17,27
The higher level and prevalence of depression in men than in women are congruent with the findings from the large Norwegian population-based HUNT study: Using the HADS, the odds ratio for depression caseness (HADS-D score > = 8) was found to be significantly higher in men compared with women, also in the particular age group 40–49 years. 28 Three main explanations for the discrepant findings compared to most other prevalence studies could be: different concepts of depression, different data collection procedures and the difference between population (census) studies and studies based on sampling.
As regards the high depression level in farmers, Roberts and Lee found the field of farming, fishing, forestry to have the highest lifetime risk for major depression, 11 whereas others have shown that farmers have increased suicide rates. 29,30 The findings could be related to the considerable mechanization, rationalization, financial strain and social isolation that have taken place in agriculture over the last years. 31
Because of the considerable similarities between depression and burnout symptomatology, according to Cordes and Dougherty’s burnout theory it might be expected that both teachers and nurses had high depression levels (the theory predicts high emotional exhaustion in workers exposed to high frequency and intensity of interpersonal contact). 8 However, female teachers had ‘higher’ whereas nurses of both genders had ‘lower’ depression levels. Maslach et al.’s more comprehensive person within context burnout theory may be more useful in understanding differences in HADS-D scores between occupational groups. 9
One of the few Norwegian industrial groups that still work on the assembly line is manufacturers of food products and beverages. Women in this industry had a ‘higher’ HADS-D score. Most assembly-line jobs are high strain jobs, with low decision latitude and high psychological demands, which according to Karasek and Theorell’s demand-control theory, could explain the finding. 5 However, such jobs are also low skill-level jobs.
In conclusion, the strong and inverse relation of skill levels to depression score and caseness indicates that the findings have clinical relevance, particularly in men. Programs for early detection and subsequent treatment of depressive disorders in low-skill occupations should be considered. The workplace may be an appropriate arena for intervention. In addition, primary health care workers, especially general practitioners, should be made aware of the increased risk of depressive symptoms and disorders in low-skill occupations. Although anxiety levels show a similar association with skill levels, the patterns are weaker, and results from the study as a whole indicate that the differences in anxiety levels between occupational groups may not be clinically relevant. Longitudinal studies on occupational differences in levels of anxiety and depression are needed.
The data collection was conducted as part of HUSK (the Hordaland Health Study 1997–1999) in collaboration With the Norwegian National Health Screening Service.
This project has been financed with the aid of EXTRA funds from the Norwegian Foundation for Health and Rehabilitation and the National Council of Mental Health and by funds from the Norwegian Ministry of Labor and Government Administration.
1. Greenberg PE, Stiglin LE, Finkelstein SN, Berndt ER. The economic burden of depression in 1990. J Clin Psychiatry. 1990; 54: 405–418.
2. Greenberg PE, Sisitsky T, Kessler RC, et al. The economic burden of anxiety disorders in the 1990s. J Clin Psychiatry. 1999; 60: 427–435.
3. Hirschfeld RMA, Keller MB, Panico S, et al. The National Depressive and Manic-Depressive Association Consensus Statement on the Undertreatment of Depression. J Am Med Assoc. 1997; 277: 333–340.
4. Leon AC, Portera L, Weissman MM. The social costs of anxiety disorders. Br J Psychiatry 1995; 166 (Suppl. 27): 19–22.
5. Karasek R, Theorell T. Healthy Work: Stress, Productivity, and the Reconstruction of Working Life. New York: Basic Books, Inc.; 1990.
6. Baba VV, Jamal M, Tourigny L. Work and mental health: a decade in Canadian research. Can. Psychol. 1998; 39: 94–107.
7. Siegrist J. Adverse health effects of high-effort/low-reward conditions. J Occup Health Psychol. 1996; 1: 27–41.
8. Cordes CL, Dougherty TW. A review and an integration of research on job burnout. Acad Management Rev. 1993; 18: 621–656.
9. Maslach C, Schaufeli W, Leiter M. Job burnout. Annu Rev Psychol. 2001; 52: 397–422.
10. Eaton WW, Anthony JC, Mandel W, Garrison R. Occupations and the prevalence of major depressive disorder. J Occup Med. 1990; 32: 1079–1087.
11. Roberts RE, Lee ES. Occupation and the prevalence of major depression, alcohol, and drug abuse in the United States. Environ Res. 1993; 61: 266–278.
12. Kessler RN, Frank RG. The impact of psychiatric disorders on work loss days. Psychol Med. 1997; 27: 861–873.
13. Grosch JW, Murphy LR. Occupational differences in depression and global health: results from a national sample of US workers. J Occup Environ Med. 1998; 40: 153–164.
14. Dewa CS, Lin E. Chronic physical illness, psychiatric disorder and disability in the workplace. Social Sci Med. 2000; 51: 41–50.
15. Kessler RC, Nelson CB, McGonagle KA, Liu J, Swartz M, Blazer DG. Comorbidity of DSM-III-R major depressive disorder in the general population: Results from the US National Comorbidity Survey. Br J Psychiatry. 1996; 168 (Suppl. 30): 17–30.
16. ILO. International Standard Classification of Occupations: ISCO-88. Geneva: International Labour Office; 1990.
17. Kessler RC, McGonagle KA, Zhao S, et al. Lifetime and 12-month prevalence of DSM-III-R psychiatric disorders in the United States. Arch Gen Psychiatry. 1994; 51: 8–19.
18. Bjelland I, Dahl AA, Tangen Haug T, Neckelmann D. The validity of the Hospital anxiety and depression scale. An updated literature review. J Psychosom Res. 2002; 52: 69–77.
19. Herrmann C. International experiences with the hospital anxiety and depression scale - a review of validation data and clinical results. J Psychosom Res 1997; 42: 17–41.
20. Statistics Norway. Standard Classification of Occupations. Oslo/Kongsvinger: Statistics Norway; 1999.
21. Statistics Norway. Standard Industrial Classification. Oslo/Kongsvinger: Statistics Norway; 1994.
22. Ware J, Kosinski M, Keller S. A 12-item, short-form health survey: Construction of scales and preliminary tests of reliability and validity. Med Care. 1996; 3: 220–223.
23. Allugander C. Psychoactive drug use in a general population sample, Sweden: correlates with perceived health, psychiatric diagnoses, and mortality in an automated record-linkage study. Am J Public Health. 1989; 79: 1006–1010.
24. Tohen M, Bromet E, Murphy J, Tsuang M. Psychiatric epidemiology. Harvard Rev Psychiatry. 2000; 8: 111–125.
25. Choi BCK. Definition, sources, magnitude, effect modifiers, and strategies of reduction of the healthy worker effect. J Occup Environ Med. 1992; 34: 979–988.
26. Morgado A, Smith M, Lecubrier Y, Widlöcher D. Depressed subjects unwittingly overreport poor social adjustment which they reappraise when recovered. J Nerv Ment Dis. 1991; 179: 614–619.
27. Kaelber CT, Moul DE, Farmer ME. Epidemiology of depression. In: Beckham EE, Leber WR, eds. Handbook of Depression, 2nd ed. New York: Guilford Press; 1995: 3–35.
28. Stordal E, Krüger MB, Dahl NH, Krüger Ø, Mykletun A, Dahl AA. Depression in relation to age and gender in the general population: the Nord-Trøndelag Health Study (HUNT). Acta Psychiatr Scand. 2001; 104: 210–216.
29. Boxer PA, Burnett C, Swanson N. Suicide and occupation: a review of the literature. J Occup Med. 1995; 37: 442–452.
30. Malmberg A, Simkin S, Hawton K. Suicide in farmers. Br J Psychiatry. 1999; 175: 103–105.
31. Thelin A. Psychosocial factors in farming. Ann Agric Environ Med. 1995; 2: 21–26.