Predictors of first-onset major depressive episodes among white-collar workers


  • Madoka Tokuyama , MD, PhD,

    Corresponding author
    1. Division of Psychiatry and Behavioral Proteomics, Department of Post-Genomics and Diseases, Course of Advanced Medicine and
    Search for more papers by this author
  • Kazuhisa Nakao , MD, PhD,

    1. Division of Psychiatry and Behavioral Proteomics, Department of Post-Genomics and Diseases, Course of Advanced Medicine and
    Search for more papers by this author
  • Masako Seto , MD,

    1. Department of Social and Environmental Medicine, Course of Social Medicine, Osaka University Graduate School of Medicine, Osaka, Japan
    Search for more papers by this author
  • Akira Watanabe , MD,

    1. Division of Psychiatry and Behavioral Proteomics, Department of Post-Genomics and Diseases, Course of Advanced Medicine and
    Search for more papers by this author
  • Masatoshi Takeda , MD, PhD

    1. Division of Psychiatry and Behavioral Proteomics, Department of Post-Genomics and Diseases, Course of Advanced Medicine and
    Search for more papers by this author

address: Dr Madoka Tokuyama, Division of Psychiatry and Behavioral Proteomics, Department of Post-Genomics and Diseases, Course of Advanced Medicine, Osaka University Graduate School of Medicine, D3, 2-2 Yamadaoka, Suita, Osaka 565-0871, Japan.


Major depression is a multifactorial disorder. Previous studies have mainly evaluated work stress to determine the risk factors for depression among workers. The present study aimed to determine factors predictive of the first depressive episode 1 year later among white-collar workers, and to examine whether work ‘stress’ is associated with an elevated risk of depression. A 5 year open-cohort study was carried out in a Japanese company. The odds ratios (OR) of the development of depression 1 year later were calculated. Ninety-eight first-onset cases were compared with 1267 never-ill cases. Forward stepwise multiple logistic regression indicated that the first onset of depression was associated with a past history of panic attack (OR: 5.14; 95% confidence interval (CI): 1.64–16.10), neuroticism (OR: 3.59; 95%CI: 2.06–6.26), perceived overprotection (OR: 2.75; 95%CI: 1.66–4.55), poor support (OR: 2.55; 95%CI: 1.58–4.10), and low care (OR: 2.23; 95%CI: 1.23–4.04). First-onset cases were more likely to have had objective work events (OR: 1.50; 95%CI: 1.18–1.90) but they did not differ from never-ill cases in subjective job stress. The development of major depression in white-collar workers is associated with multiple factors, as is depression in the community.


The prevention of mental illness and promotion of mental health has attracted much attention in recent decades.1 In particular, mental health care in the workplace has been gaining more interest. Depressive disorders are one of the most common forms of mental illness encountered in employees. Depression in the workplace has a great impact upon the economy.2–4 The detection of depressed workers at an early stage and encouragement to obtain treatment might be cost-effective for employers.

Risk factors for depression among workers

The relationship between the occurrence of depression and job stress has been confirmed. Kawakami and Haratani conducted a 1 year follow-up prospective study of male workers in a Japanese electrical company.5 They examined the relationship between lifestyle and depressive symptoms. Subsequent depressive symptoms were related to insomnia, irregular meals, infrequent drinking, depressive symptoms at baseline, poor health status, stress at work, no relaxation time, and poor economic status. Kawakami et al. also reported that job stress due to unsuitable jobs and human relations was associated with the occurrence of major depression.6 They suggested that an unsuitable job might decrease a worker's self-esteem and their ability to cope with other psychosocial stressors such as life events, making the worker vulnerable to the disorder. There are other studies also reporting that depressive symptoms are influenced by quality of work (i.e. perceived job stress due to unsuitable jobs) rather than by quantity of work (i.e. job overload).7,8 The results of the Gazel cohort in France point to the possibility that a high level of psychosocial demands, low levels of decision latitude, and low levels of social support at work were predictive of depression.9 A Canadian health survey showed that women who worked long hours had an increased risk for depression.10 Most of these previous studies have mainly evaluated work stress and a few other domains, and there are few studies treating multifarious factors such as community studies.

Risk factors for depression in the community

There are a number of community studies regarding risk factors for major depression.11–13 Depression is thought to be a multifactorial disorder, and prior investigators identified the following putative risk factors: female gender,11,13 young age,11,12 lower level of education,13 separated/widowed/divorced and never married status,11,13 premature parental loss,14,15 exposure to pathogenic parental rearing,16 neurotic personality,17,18 a previous history of major depression,19 low social support,20 recent stressful life events and difficulties,21–23 and predisposing genetic influences.19 A large population-based twin study found that genetic factors affected vulnerability to depression in part by influencing coping behavior and temperament. The coping strategies of ‘turning to others’ and ‘problem solving’ had a buffering effect on the impact of life events on depression, and an individual's tendency to use these coping strategies was significantly influenced by genetic factors.24 Kendler et al. concluded that a set of genetically influenced traits, which reflect a ‘difficult’ or ‘neurotic’ temperament, also predispose exposure to stressful events and to episodes of major depression.23 It has been confirmed that psychosocial stressors play a greater role in the first episode of affective disorder than in subsequent episodes.25–27 Psychosocial stressors may not only play a role in triggering an affective episode but also, because of the neurobiological encoding of memory-like functions related to the stressors, provide a long-term vulnerability to subsequent recurrences and a mechanism for retriggering episodes with lesser degrees of psychosocial stress.26

State effect

In order to assess the personality traits of depressives, it is necessary to consider the possible influence of the current mental status and past episodes. An assessed personality may have changed after an episode of depression (scar hypothesis) or it may be a current depressive symptom (complication hypothesis).12,17,28 The Collaborative Depression Study presented two major reports about the relationship between observed personality trait and depressive disorder. The earlier report by Hirschfeld et al. concluded that the state of depression influenced personality assessments.29 Additionally, they proposed that neuroticism and extraversion did not return to the premorbid level after recovery from depression.12,29 These two reports supported the scar or complication hypothesis, suggesting that the depressive episode was the cause of a lasting personality change. According to later reports, self-reported personality traits did not change after a typical episode of major depression, and the scar hypothesis was not supported.28 Their findings that neuroticism does not increase after episodes of depression was consistent with findings reported by Duggan et al. and Clayton et al.30,31

In order to obviate the state effect on rating personality traits as well as subjective stress and perceived parental rearing style, a prospective and longitudinal study is necessary. Furthermore, it is desirable to evaluate individuals before they develop a first episode of depressive disorder.

Major depression is a multifactorial disorder in which risk factors interact with one another to cause disease. The present study measured variables in four domains: subjective/objective stress at work and in one's private life; past history of anxiety disorders; parental rearing style; and personality traits. The aim was to discover the powerful predictors of the first depressive episode and whether ‘stress’ at work increases the risk of the first episode. We used a multiple logistic regression analysis adjusting possible interrelations among those risk factors. Our report is from a prospective study of a large sample of white-collar workers in a Japanese company, first surveyed in 1997.


Data collection

Original subjects in the present study consisted of approximately 2000 white-collar workers of a marine insurance company and its affiliated companies in Japan. The data used in this analysis were gathered on five occasions at 1 year intervals: 1997, 1998, 1999, 2000 and 2001. Individuals who consented to answer the questionnaire developed by one of the authors (KN) participated in the study. The response rate in 1997, 1998, 1999, 2000 and 2001 was 68.2% (1400/2053), 58.7% (1409/2399), 61.7% (1230/1992), 63.4% (1211/1910), and 49.2% (889/1808), respectively. A self-administered questionnaire at T1 (1997, 1998, 1999 and 2000) assessed the predictive variables: age, gender, subjective stress at work and in one's private life during the past 1 year, a history of depression or anxiety disorders, trauma before the age of 18 years, perceived parental rearing style, neuroticism, and extraversion. The same questionnaire at T2 (12 months later: 1998, 1999, 2000 and 2001, respectively) assessed the outcome variable: the presence or absence of major depressive episodes. Personal and work events within the preceding 6 months, another predictive variable, was assessed at T2.


We assessed the prevalence of major depression five times, so that there were four data sets consisting of two (T1/T2) components: the 1997/1998 (n = 877), 1998/1999 (n = 850), 1999/2000 (n = 809) and 2000/2001 cycles (n = 772). The subjects had, at most, four data sets consisting of the two components. In our open-cohort study, 1605 subjects in total were followed over 5 years. Data sets consisting of 1 year before onset/the year of onset (T1/T2) were used for onset cases. The data sets of the latest 2 years at follow up were used for other cases. To reduce the possible impact of current and past depression on the recollection of neuroticism, we restricted samples to first-onset cases and never-ill cases. People who reported having a current depressive episode at T1 (n = 11) were eliminated from the study. In addition, individuals with a past history of major depression at T1 (n = 161) were excluded. At T2, people without current depression but with a past history of depression (n = 68) were also eliminated. Subjects with a current major depressive episode at T2 without a current episode at T1 nor past history at T1 and T2 were identified as first-onset cases (n = 98). Subjects free from illness at both T1 and T2 were defined as never-ill cases (n = 1267). The final number of subjects was 1365. The sample consisted of 647 men and 718 women with a mean age of 34.2 ± 10.3 years (range: 20–73 years).

Diagnostic evaluation

In order to identify a current depressive episode, the Zung Self-rating Depression Scale32 (SDS) and the Diagnostic and Statistical Manual of Mental Disorders (4th edn; DSM-IV) criteria33 were used for both T1 and T2 investigations. According to DSM-IV criteria, subjects were asked whether they had experienced five (or more) of the following symptoms during the same 2 week period (if they had experienced these symptoms after the loss of a loved one, the symptoms should persist for longer than 2 months): (i) depressed mood; (ii) diminished interest or pleasure in all; (iii) decrease in appetite; (iv) insomnia; (v) fatigue or loss of energy; (vi) feeling of worthlessness; (vii) diminished ability to think or concentrate, or indecisiveness; or (viii) recurrent thoughts of death. Cases reporting the five or more symptoms and scoring >40 in the SDS were defined as having major depression. The SDS cut-off score was determined according to the symptom level of depression defined by Barrett et al.34 For a screening of past history of major depression, the DSM-IV criteria was also used.

Predictor variables

Age was divided into four categories: ≤29; 30–39; 40–49; or ≥50. The presence or absence of the following five major traumas before the age of 18, were investigated at T1: severe injury; severe illness; assault or robbery; divorce of parents; and death of parents. Perceived parental rearing style was assessed at T1. It contained overprotection, low care, and lack of parental warmth. These variables were reported as present or absent. The 1 year test–retest reliability was calculated using the kappa coefficient. Kappa coefficients for overprotection, low care, and lack of parental warmth were 0.59, 0.56, and 0.47, respectively, indicating fair reliability. The T1 questionnaire screened for a past history of panic attack and generalized anxiety disorder by DSM-IV criteria. At T1 we measured neuroticism and extraversion by questions adopted from the Neuroticism, Extraversion, Openness (NEO) Five-Factor Inventory (NEO-FFI).35 Both scores were collapsed to dichotomous variables (higher vs lower than mean score of all subjects). The T1 questionnaire contained nine items assessing the subjective experience of stress: job overload; difficult job; inadequate evaluation of contribution; problems with coworkers; family problems; close friend problems; health problems; poor economic status; and lack of support. Nine items were estimated as present or absent. Subjects were asked which events in the following list had happened to him/her within the preceding 6 months at T2: personal events (divorce/living apart from one's spouse/separation from loved one/severe illness/severe injury/death of family members/severe illness of family members/severe injury of family members/death of close friends/severe illness of close friends/severe injury of close friends/assault or robbery/legal problems/property loss/living with parents/leaving one's family behind for a new post/moving), and work events (transfer/promotion at work/change of superior/change of job description/≤2 days holiday per month/≥40 h overtime per month).

Statistical analyses

First, the odds ratios (OR) of a depressive episode 1 year later, weighted but unadjusted for other covariates, were calculated for all covariates. The first-onset subjects were compared with the never-ill subjects.

Multiple logistic regression was then used to examine the power of independent variables to predict the outcome of the first onset versus the never ill. Personal event variables were totaled, as were work event variables. The stepwise selection method was used with entry testing based on the significance of the score statistic (P < 0.05), and with removal testing based on the probability of a likelihood ratio statistic based on the maximum partial likelihood estimates (P > 0.10).

spss 10.0 J software (SPSS Japan, Tokyo, Japan) was used for all analyses.


The prevalence rate of major depression in 1997, 1998, 1999, 2000 and 2001 was 3.8% (33/877), 3.1% (36/1180), 5.0% (54/1075), 6.0% (63/1052), and 7.0% (54/771), respectively.

Univariate analyses

Unadjusted OR for a first episode of depression are presented in Table 1. Odds ratios were significantly high for those losing property and being separated. A person with a history of panic attacks had an eightfold higher risk of depression at follow up. Neuroticism had a higher risk for subsequent depression. Absence of social support, perceived overprotection, and work events (change of job description, change of superior) also significantly affected the first onset of depression after 1 year. Younger age, interpersonal stress with relatives, job stress (inadequate evaluation of contribution, problems with coworkers), a history of generalized anxiety disorder and health problems appeared to predict subsequent depression 1 year later.

Table 1. Univariate analysis of the estimated OR for depression during 1 year follow up
VariablesOR95% CIP
  1. OR, odds ratio; CI, confidence interval.

  2. The OR and the 95%CI for ‘divorce’, ‘living apart from one's spouse’, ‘severe illness of close friend’, ‘assault or roberry’, and ‘leaving one's family behind for a new post’ were not calculated because no subjects developed a major depressive episode.

Gender (female vs male) 1.400.92–2.12 0.12
Age (years)   0.11
 ≤29 4.061.24–13.23 0.02
 30–39 3.801.15–12.58 0.03
 40–49 4.661.34–16.20 0.02
 ≥50 1.00  
Trauma before age of 18
 Severe injury
  Yes vs no 2.770.93–8.32 0.07
 Severe illness
  Yes vs no 2.450.70–8.56 0.16
 Assault or robbery
  Yes vs no 6.470.58–72.02 0.13
 Divorce of parents
  Yes vs no 0.560.08–41.44 0.57
 Death of parents
  Yes vs no 0.770.27–2.16 0.62
Parental rearing style
  Yes vs no 2.721.77–4.18<0.001
 Low care
  Yes vs no 1.420.85–2.38 0.18
 Lack of parental warmth
  Yes vs no 1.320.46–3.76 0.61
  High vs low 4.422.58–7.55<0.001
  High vs low 0.720.46–1.11 0.13
Past history of any anxiety disorders
 Panic attack
  Yes vs no 8.013.08–20.83<0.001
 Generalized anxiety disorder
  Yes vs no 2.101.32–3.34 0.002
Subjective stress
 Job overload
  Yes vs no 1.160.73–1.84 0.53
 Difficult job
  Yes vs no 1.140.74–1.76 0.56
 Inadequate evaluation of contribution
  Yes vs no 2.101.39–3.18<0.001
 Problems with coworkers
  Yes vs no 2.471.62–3.76<0.001
 Family problems
  Yes vs no 2.471.43–4.28 0.001
 Close friend problems
  Yes vs no 1.700.85–3.40 0.13
 Health problems
  Yes vs no 2.031.31–3.14 0.002
 Poor ecnomoic status
  Yes vs no 1.280.83–1.97 0.26
 Lack of social support
  Yes vs no 3.252.10–5.04<0.001
Life events within 6 months
 Personal events
  Separation from loved one
   Yes vs no 3.931.74–8.88 0.001
  Severe illness
   Yes vs no 3.280.69–15.65 0.14
  Severe injury
   Yes vs no 4.430.45–42.15 0.21
  Death of family members
   Yes vs no 0.920.22–3.93 0.91
  Severe illness of family members
   Yes vs no 0.780.18–3.30 0.74
  Severe injury of family members
   Yes vs no 3.260.36–29.40 0.29
  Death of close friends
   Yes vs no 1.600.71–3.61 0.26
  Severe injury of close friends
   Yes vs no 4.340.45–42.15 0.21
  Legal problems
   Yes vs no 1.300.16–10.23 0.81
  Property loss
   Yes vs no10.742.84–40.66<0.001
  Lliving with parents
   Yes vs no 3.260.36–29.40 0.29
   Yes vs no 0.640.20–2.07 0.45
 Work events
   Yes vs no 1.580.70–3.55 0.27
  Promotion at work
   Yes vs no 1.740.67–4.52 0.26
  Change of superior
   Yes vs no 1.871.09–3.21 0.02
  Change of job description
   Yes vs no 2.931.83–4.69<0.001
  ≤2 days holiday per month
   Yes vs no 0.720.10–5.42 0.75
  ≥40 h overtime per month
   Yes vs no 1.430.90–2.26 0.13

Multiple logistic regression

Results from the multiple logistic regression analysis are summarized in Table 2. The final model included six variables: past history of panic attack, neuroticism, overprotection, lack of social support, low care and work events. The log likelihood of the final model was 579.73 (χ2 = 92.09, d.f. = 6, P < 0.001). The Hosmer and Lemshow goodness-of-fit statistic was 1.62 (d.f. = 6, P = 0.95), indicating an acceptable fit for the data. The strongest predictor was past history of anxiety disorders. For workers who reported an experience of panic attack, the estimated OR of major depression at follow up was 5.14 (95% confidence interval (CI) = 1.64–16.10, Wald statistic = 7.87, d.f. = 1, P = 0.005). A person with a high level of neuroticism had a fourfold higher risk of depression at follow up (95%CI = 2.06–6.26, Wald statistic = 20.34, d.f. = 1, P < 0.001). Lack of social support, perceived overprotection and low care also contributed to subsequent depression. Work events and objective job stress influenced the first onset of depression after 1 year. When the final model was adjusted for the other variables, gender, age, trauma before the age of 18, lack of parental warmth, subjective job stress, health problems, poor economic status and personal events did not lead to increased risk of major depression.

Table 2. Logistic regression analysis of estimated OR for depression during 1 year follow up
VariablesOR95% CIP
  1. OR, odds ratio; CI, confidence interval.

  2. −2 log likelihood = 579.73; χ2 = 92.09 (d.f. = 6, P < 0.001), Negelkerke R2 = 0.169.

Past history of panic attack5.141.64–16.10 0.005
Lack of social support2.551.58–4.10<0.001
Low care2.231.23–4.04 0.009
Work events1.501.18–1.90 0.001


In a large white-collar sample, we found that a past history of panic attack, neuroticism, lack of social support, perceived overprotection, perceived low care, and work events predicted the onset of a new episode of depression in the subsequent year. However, the findings of the present study must be viewed against important limitations. The present study used self-administered questionnaires to assess both the predictive variable and the outcome variable. We relied on self-reported symptoms of depression rather than on a formal diagnosis based on a structured psychiatric interview. Myers and Weissman suggested a modest relationship between self-reported symptoms and clinically diagnosed depression.36 Symptom scales were useful for screening depressed individuals in research studies but were only approximate indicators of clinical depression, therefore we used both SDS and DSM-IV criteria to identify a case of depression. The results of the present study should be replicated by research using a structured interview to collect data. Another limitation is the lack of an important variable regarding genetic influence (i.e. family history). Studies containing genetic variable are needed.

Previous research has observed that patients with primary anxiety disorders often developed secondary depression.37 Results from the USA National Comorbidity Survey reported by Kessler et al. showed that 51.2% of subjects with a 12 month major depressive disorder also experienced an anxiety disorder and 67.9% of cases of secondary depression were associated with a primary anxiety disorder.38 Their results also documented that the prevalence of secondary major depressive disorder has increased over the past 40 years, while the prevalence of pure and primary depression has not changed over this same time period. Most lifetime cases of major depression are secondary, and anxiety disorders are the most common primary disorders associated with secondary depression. Furthermore anxiety disorders are the primary disorders predicting the greatest subsequent risk of secondary depression. Our results also showed that earlier anxiety disorder was the risk factor for secondary depression. Workers with a prior panic attack had a 5.14-fold risk of onset of major depression after 1 year compared with those without. Secondary depression is more severe than pure or primary depression38 so we should pay attention to individuals with a past history of anxiety disorder.

In order to prevent a state of anxiety affecting the rating of personality, subjective stress, or rearing style, workers having a past history of any mental disorder, including anxiety disorder, should be excluded from the subjects. However, we chose to avoid a limitation that would reduce the sample size of this stringently defined first-episode subject group. We therefore defined a broader group, including those who at T1 were negative for a past history of major depression but may have had a past history of other mental disorders before the T1 assessment.

The result of logistic regression analysis showed that the second strongest predictor was the neuroticism variable. This is not a new finding because there are a number of investigations that support the relationship between depression and neuroticism.12,19,39,40 In addition, introversion, dependency17,41 and interpersonal sensitivity39 were suggested as personality risk factors for depression. However, our result is one of the few findings that have been obtained from a prospective design with a large sample size, in first-onset cases who were not influenced by the scar effect of a past depressive episode.

Many previous studies have indicated that less social support leads to an increased risk of major depression.19,42–44 In model by Kendler et al. using a path diagram, low social support directly affected the tendency of depression.19 However, there was an inverse conclusion from a prospective population study in Norway.45 Wade and Kendler examined the association between perceived social support and major depression by longitudinal twin study.46 They used eight variables of perceived social support, including spouse problems, spouse support, relative problems, relative support, friend problems, friend support, confidants, and social integration. Interestingly, cross-time associations were seen only with spouse problem and social integration. An earlier report from Kessler et al. showed that perceived friend and relative support were more powerfully related to major depressive episodes than perceived spouse support.43 The results were also contradictory. We could not examine spouse support because 42.3% of our subjects were under 29 years old and most of them were probably not married. Our analysis indicated that impaired subjective social support significantly predicted a major depressive episode. There are two inconsistent studies, one suggesting that the effect of perceived support is due to an underlying genetic influence, which also affects vulnerability to stress;47 and another failing to replicate it.43 More work on these interactions is needed.

We could confirm the associations of pathogenic parental rearing with high levels of depressive symptoms, as shown by previous investigators. Parker explored the associations between parental child-rearing behavior and depression using three separate groups of bipolar manic-depressive patients, neurotic depressive patients and non-clinical student samples.48 Neurotic depressive patients perceived their parents to be less caring with greater maternal overprotection. Depressive experience in the non-clinical group was negatively associated with low parental care and weakly associated with parental overprotection. Additionally, he found that overprotection from the more important parent was associated with neuroticism, trait anxiety, longer duration of depressive episodes and lower self-esteem. Narita et al. proposed that low parental care was highly associated with a lifetime history of depression among Japanese employees, and provided some evidence that overprotection could also be associated with depression.49 In the initial separate analyses, perceived low parental care appeared not to influence subsequent depression. However, the logistic regression analysis indicated that after the effects of other variables (including neuroticism or history of anxiety disorder variables) had been taken into account, low care contributed to depressive episodes. Our results also showed that perceived overprotection might increase the risk of depression onset (Tables 1,2).

We regarded self-administered job stress as subjective stress, while we regarded work events as objective stress. Onset cases had significantly more work events within the preceding 6 months than never-ill cases. We assessed the unadjusted OR of each perceived job stress item for depressive episodes in the univariate analysis. Job overload had a 1.16-fold risk (95%CI = 0.73–1.84, Wald statistic = 0.40, d.f. = 1, P = 0.53), and difficulty had a 1.14-fold risk (95%CI = 0.74–1.76, Wald statistic = 0.34, d.f. = 1, P = 0.56). These items, which represented the quantity aspect of job stress, were not significant predictors. In contrast, the quality aspect of job stress was an important risk factor for depression. Odds ratios were significantly high for inadequate evaluation of contribution (OR = 2.10, 95%CI = 1.39–3.18, Wald statistic = 12.30, d.f. = 1, P < 0.001), and problems with coworkers (OR = 2.47, 95%CI = 1.62–3.76, Wald statistic = 17.73, d.f. = 1, P < 0.001). Consistent with earlier research, the first episode of depression was influenced by quality of work rather than by quantity of work. However, this influence was not confirmed in the multiple logistic analysis. We used the data at T1 to assess subjective stress to remove the influence of current depression but a 1 year interval might be too long to indicate impact on an onset. In the results we confirmed that the risk for a first episode of major depression was associated with objective work events but we failed to confirm its positive relationship with subjective job stress.

The data indicate that evaluation of an individual's personality, learning how to handle job stress, and maintaining good communication with coworkers may contribute to preventing depression in the workplace.