The Impact of Changes in Job Strain and Its Components on the Risk of Depression

Abstract

Objectives. We assessed the impact of changes in dimensions of the psychosocial work environment on risk of depression in a longitudinal cohort of Canadian workers who were free of depression when work conditions were initially reported.

Methods. Using a sample (n = 3735) from the Canadian National Population Health Survey, we examined the effects of changes in job control, psychological demands, and social support over a 2-year period on subsequent depression. We adjusted models for a number of covariates, including personal history of depression.

Results. Respondents with increased psychological demands were more likely to have depression over the following 2 years (odds ratio = 2.36; 95% confidence interval = 1.14, 4.88). This risk remained statistically significant after adjustment for age, gender, marital status, presence of children, level of education, chronic health conditions, subclinical depression when work conditions were initially assessed, family history of depression, and personal history of depression.

Conclusions. These results demonstrate that changes in psychological demands have a stronger influence than changes in job control on the onset of depression, highlighting the importance of not assuming an interaction between these 2 components of job strain when assessing health outcomes.

Major depression is 1 of the top 3 causes of disability burden in high-income countries.¹ The burden of depression could be reduced by identifying predictors of the disease that are amenable to change and then intervening accordingly. Job strain has been identified as one such predictor.² Job strain refers to a situation where job control (people's ability to make decisions and use their skills at work) is low and the job's psychological demands (the pace and mental intensity of work) are high.³

Although cross-sectional and longitudinal research has demonstrated that job strain and its components are related to an increased risk of depression,^2,4,5 this does not constitute evidence that changing these conditions would result in changes in the risk of depression (for better or worse). Yet, from both an organizational and public policy perspective, evidence that changes in psychosocial working conditions are associated with subsequent increased (or decreased) risk of depression is important if the potential mental health effects are to be accounted for when making decisions that will affect the psychosocial work environment. Ideally, this evidence would be generated from trials (preferably randomized). In lieu of trial data, longitudinal survey data where job strain is measured at 2 or more time points can be used to explore the impact that naturally occurring changes in job strain have on the risk of depression.

Four longitudinal studies have examined whether naturally occurring changes in job strain were associated with changes in the risk of both depression and psychological distress.^6–9 The findings of all 4 studies suggested that beneficial changes (reductions) in job strain resulted in a lower risk of depression; however, the effect sizes were small in 3 of the 4 studies, and statistical significance was achieved in only 1 study.⁸ In addition, each of these studies suffered from important methodological limitations in their design, which may have biased the findings reported.

None of the previous studies took into account the day-to-day variability in job strain scores when assigning respondents to exposure groups. In each study, a respondent was classified as “exposed” to a change in job strain if his or her job strain score crossed a particular threshold between time points, regardless of the actual size of the change in the score between baseline and follow-up. As a result, some respondents were classified as “exposed” when the change was no greater than the day-to-day variability observed for the score, whereas other respondents were classified as “unexposed” when the change was greater than this variability. These classifications could dilute the “exposed” group, making it more similar to the “unexposed” group with respect to risk of depression (and vice versa). In other words, the use of thresholds in previous studies may have resulted in nondifferential misclassification across exposure groups that, in turn, may have produced an underestimate of the true effect of changing job strain on the risk of depression.^10–12

Most previous studies focusing on change in job strain have not considered the potentially different impacts of changes in the underlying components of job strain (job control and psychological demands). In addition to job control and psychological demands, it is also important to examine changes in social support because previous work has linked this psychosocial characteristic of work to an increased risk of depression.^7,9,13,14 Studies that focused on job strain (but not changes in job strain) have found that these components pose different risks for depression, with high psychological demands more strongly associated with the risk for depression than low job control.^15–17 Accordingly, a change in job demands may have a larger effect on depression than a change in job control. If this is the case, investigations focused only on changes in job strain (without consideration of beneficial or adverse changes in the underlying components) would result in an underestimate of the true potential impact of efforts to reduce job strain as a strategy for preventing depression. Only 1 of the 4 studies explored this possibility, reporting that changes in psychological demands had a stronger effect on the risk of depression than changes in job control, although neither change produced a statistically significant effect.⁹

Because of data availability, each of the previous change studies had different time lags between when the change in job strain occurred and the measurement of subsequent depression (or depressive symptoms). Stansfeld et al.⁹ allowed 2 to 7 years to elapse between the change in job strain and the measurement of depression, whereas Wang et al.⁸ allowed 1 to 10 years. Research examining the effect of work conditions on mental health has demonstrated that different time lags can result in different study findings, with the strongest relationships between work and mental health found over 1- to 2-year periods.^18–20 As a result, 4 previous studies in this area of research have likely underestimated the effect of the change by collapsing short and long lag periods. The study design by de Lange et al. allowed for the risk of depression at year 1 to be attributed to changes in job strain that took place after depression onset.⁶ Bourbonnais et al.⁷ took their follow-up measurement of job strain at the same time that they assessed depression, which may have resulted in a spurious, or elevated, correlation between the 2 measures, because of the likely impact that depression has on self-reported job strain assessments.⁹

Finally, each of the previous change studies failed to adjust for personal history of depression, which is a potentially important confounder. Personal history of depression is the strongest predictor of a future episode of depression.²¹ Because it is plausible that a personal history of depression could also increase the risk for negative changes (increases) in job strain, this history may confound the relationship between changes in job strain and the risk of depression. Some authors have suggested that respondents with a previous history of depression should be removed from analyses focused on the impact of change in psychosocial working conditions on the risk of depression.¹³ Although removal of respondents who have a potential confounder is one method to deal with confounding, this approach has the disadvantage of creating a hypothetical sample of the working population (i.e., only those workers without any previous history of depression).^22,23 The data used in our analysis (from the National Population Health Survey) demonstrate that up to 20% of the working population who are not currently depressed have a personal history of depression. Therefore, excluding these respondents from the analysis creates a sample of workers that is no longer representative of the working population. From a population health perspective, a more reasonable approach to dealing with confounding is to include personal history of depression as a covariate in multivariate analyses.

We have designed our study to overcome the limitations of previous work by (1) focusing on psychosocial work environment changes that are greater than those expected as a result of day-to-day variability; (2) examining the separate impacts of job control, psychological demands, and social support in addition to job strain; (3) assessing change in work environment only among respondents not currently depressed; and (4) adjusting for the potential confounding effects of personal history and family history of depression. By overcoming these limitations, we aimed to generate more accurate estimates of the effect of changes in the psychosocial work environment on the risk of depression.

METHODS

We conducted a secondary analysis of data from the Canadian National Population Health Survey (NPHS). Starting in 1994, the NPHS has collected information every 2 years on health conditions, health behaviors, and labor market participation from a representative cohort of Canadians.^24,25 Respondents aged 15 to 74 years are asked questions on labor market participation. To date, 7 cycles of the NPHS are available for analysis (1994–1995, 1996–1997, 1998–1999, 2000–2001, 2002–2003, 2004–2005, and 2006–2007). We focused on 3 specific cycles (2000–2001, 2002–2003, and 2004–2005) for 2 reasons: the 2000–2001 and 2002–2003 cycles were the first consecutive cycles in the NPHS to measure job strain, and the 2004–2005 cycle of the NPHS contains information on personal history of depression, which is an important covariate in our analysis.

Depression

We defined the occurrence of depression using 2 variables available in the 2004–2005 cycle of the NPHS. The first variable was based on the Composite International Diagnostic Interview-Short Form for Major Depression (CIDI-SFMD), a structured diagnostic interview tool designed to assess major depression in population-based surveys.^26,27 Respondents who endorse 5 or more depressive symptoms included in the CIDI-SFMD have a 90% probability of having experienced a major depressive episode in the 12 months prior to the survey. We classified respondents who met this criterion as depressed.

Because NPHS cycles are 2 years apart but the CIDI-SFMD only assesses depression in the 12 months prior to the survey, respondents who experience a major depressive episode 13 to 24 months prior to completing a cycle of the NPHS will not be identified as depressed by the CIDI-SFMD. However, the NPHS does capture the date a respondent is first diagnosed with depression (if applicable). Respondents who reported first being medically diagnosed with major depression 13 to 24 months before the 2004–2005 survey were also classified as depressed, partially overcoming the limitation imposed by the 2-year interval in the NPHS design.

Changes in Job Strain and Its Components

The 2000–2001 and 2002–2003 cycles of the NPHS contain an abbreviated version of the Job-Content Questionnaire.³ For our study, we derived 4 measures of the psychosocial work environment from the responses to this questionnaire: job control (5 questions), psychological demands (2 questions), social support (3 questions), and job strain (the ratio of the average score of the psychological demand questions to the average score of the job control questions). We have listed the questions used to assess the psychosocial work environment in the NPHS in the box on page e3.

Work Stress Questions in the National Population Health Survey

Response scores for all questions are (1 = strongly disagree, 2 = disagree, 3 = neither agree or disagree, 4 = agree, 5 = strongly agree)

Job control

Your job requires you learn new things

Your job requires a high level of skill

Your job requires that you do things over and over (reverse scored)

Your job allows you freedom to decide how you do your job

You have a lot to say about what happens in your job

Psychological demands

Your job is very hectic

You are free from conflicting demands that others make (reverse scored)

Social support

You are exposed to hostility or conflict from the people you work with

Your supervisor is helpful in getting the job done (reverse scored)

The people you work with are helpful in getting the job done (reverse scored)

We assessed changes in each psychosocial work dimension by classifying changes greater than the minimal detectable change, as previously described by Jacobson et al.^10,11 This method incorporates both the observed variance in scores at time 1 and an estimate of day-to-day variability in the measures (i.e., the variability in scores when no change has actually occurred). We generated the observed variances in each score (job control, job demands, and social support) at time 1 using the NPHS data. We drew the day-to-day variability estimates from another data source because these were not available in the NPHS. Using test-retest data over a 2-week period from 48 participants in the Ontario Child Health Survey Follow-Up Study (the participants in this follow-up study were aged 21–35 years),²⁸ we generated estimates of the day-to-day variability for each of the psychosocial work dimensions. The estimates generated for each of the 3 dimensions (job control, psychological demands, and social support) were the same: ±3. In other words, 95% of the time, respondents who have no real change in job control, psychological demand, or social support would have score changes of ±3. The fact that the estimates of day-to-day variability for all 3 work stress dimensions were the same (±3) is coincidental, given that they were derived separately. Interpretations of these estimates of day-to-day variability should take into account the range of scores possible on each scale. There is thus relatively greater day-to-day variability observed in the psychological demands scale (range = 0–8) than in the job control and social support scales (range = 0–12), reflecting the poorer test–retest reliability of the psychological demands scale.

Covariates of Interest

Using respondents’ data from the 2000–2001 survey cycle, we adjusted our regression models for gender, age (grouped), marital status, the presence of children younger than 12 years in the household, highest level of education, presence of chronic health conditions (excluding allergies), and psychosocial dimensions of work (job control, psychological demands, and social support). Models also included separate binary variables indicating whether the respondent (1) had changed occupations between the 2000–2001 and 2002–2003 survey cycles, (2) had a family history or depression as of the 2004–2005 survey cycle, (3) had been diagnosed with depression by a health professional prior to the 2000–2001 survey cycle (as self-reported in the 2004–2005 cycle), or (4) had subclinical depression in the 2000–2001 or 2002–2003 survey cycles (defined as a probability of depression above 0 but below 90%, based on the CIDI-SFMD).

Analysis

A total of 10 645 respondents participated in all 3 cycles used for this analysis (2000–2001, 2002–2003, and 2004–2005), which represents 69% of the subset of the original sample who had not died or been institutionalized. Respondents lost to follow-up were more likely to be of lower education (as measured in 1994), male, and younger. For the purpose of this analysis, we focused on respondents aged 25 to 60 years in 2000–2001 who were working at some point in the 12 months preceding both the 2000–2001 and 2002–2003 surveys (to ensure that change in work conditions could be assessed for each respondent). This sample comprised 4608 respondents (74% of the sample aged 25 to 60 years).

To prevent cognitive distortion bias (a bias caused by depressed respondents’ tendency to perceive situations more negatively than nondepressed respondents) from affecting reports of psychosocial work conditions, we removed respondents who were depressed during the period when these conditions were assessed (n = 388; 9% of working sample), who did not respond to questions on depression during these cycles (n = 100), or who reported being first diagnosed with depression (based on questions on personal history in the 2004 survey) between the 2000–2001 survey cycle and the 2002–2003 cycle (n = 68). The remaining sample of respondents—who were working in both 2000 and 2002, responded to the 2004 survey, and were free of depression when working conditions were assessed (in 2000–2001 and 2002–2003)—totaled 4052. Of this sample, 66 respondents (2% of sample) were missing information on depression in 2004, with another 251 respondents (6%) missing information on job strain or covariates, leaving a final sample of 3735. Respondents with missing information had lower levels of education and were younger than those with complete information.

We used logistic regression to examine the risk of depression associated with a change in each dimension of the psychosocial work environment, using no change as the reference group (allowing us to examine both positive and negative changes in each component). We explored the possibility of interactions among changes in each dimension of work and gender,⁹ education, and occupation groups.^14,29 We did this by including a multiplicative interaction term (separately for each interaction) in our regression models (assessing whether the association between change in work dimensions and depression differed across these groups). To account for the complex sample design of the NPHS, and in line with guidelines from Statistics Canada, we used a bootstrap technique to adjust the confidence intervals around each point estimate.³⁰ In addition, we weighted all analyses to account for the probability of selection into the original sample and nonresponse.

RESULTS

Table 1 presents the distribution of our main study variables. It also shows the percentage of depressed respondents in each subgroup of each study variable, along with the unadjusted odds ratios testing the difference in the risk of depression across the subgroups of each variable. A total of 4% of our sample had experienced depression between the 2002–2003 and 2004–2005 surveys. Focusing on the components of job strain, we found an elevated risk of depression only for negative changes (increases) in psychological demands between the 2000–2001 and 2002–2003 surveys. Other factors associated with an increased risk of depression were the presence of chronic health conditions at baseline, being female, having a family history of depression, and having a personal history of depression.

TABLE 1—

Distribution of Depression Across Study Variables: Canadian National Population Health Survey, 2004

Variable	No.	Respondents Depressed, %	Unadjusted OR (95% CI)
Full sample	3735	4.0
Change in job control (2000–2002)
Decrease	599	4.0	1.02 (0.58, 1.80)
No change (Ref)	2511	3.9	1.00
Increase	625	4.2	1.07 (0.66, 1.73)
Change in psychological demands (2000–2002)
Increase	295	8.2	2.37* (1.26, 4.46)
No change (Ref)	3162	3.6	1.00
Decrease	278	3.6	0.99 (0.12, 8.09)
Change in job strain (2000–2002)
Increase	314	5.2	1.38 (0.65, 2.94)
No change (Ref)	3118	3.8	1.00
Decrease	302	4.8	1.28 (0.62, 2.62)
Change in social support (2000–2002)
Decrease	445	5.4	1.34 (0.78, 2.31)
No change (Ref)	2777	4.1	1.00
Increase	454	2.2	0.54 (0.22, 1.33)
Gender
Men (Ref)	2030	2.9	1.00
Women	1705	5.3	1.84* (1.21, 2.80)
Age group, y
25–34	805	4.7	1.33 (0.80, 2.23)
35–44 (Ref)	1455	3.5	1.00
45–60	1475	4.0	1.15 (0.73, 1.82)
Marital and household status
Single with children aged < 12 y	111	6.1	1.46 (0.56, 3.79)
Single without children aged < 12 y	818	4.1	0.97 (0.59, 1.58)
Married with children aged < 12 y	1223	3.4	0.80 (0.47, 1.34)
Married without children aged < 12 y (Ref)	1583	4.2	1.00
Chronic health conditions in 2000
No (Ref)	2087	2.9	1.00
Yes	1648	5.3	1.88* (1.29, 2.75)
Level of education
Less than secondary	340	4.0	1.14 (0.51, 2.51)
Secondary graduation	1564	3.9	1.09 (0.61, 1.94)
Postsecondary graduation	906	4.6	1.32 (0.78, 2.23)
Bachelor's and higher (Ref)	925	3.6	1.00
Same occupation at both time points
Yes	3086	3.7	0.67 (0.40, 1.12)
No (Ref)	649	5.4	1.00
Family history of depression
No (Ref)	2654	2.3	1.00
Yes	1081	8.0	3.66* (2.48, 5.41)
Personal history of depression
No (Ref)	3494	3.7	1.00
Yes	241	8.6	2.47* (1.35, 4.50)
Probability of depression below 90% in 2000 or 2002
No (Ref)	3576	3.9	1.00
Yes	159	6.6	1.76 (0.73, 4.21)

Note. CI = confidence interval; OR = odds ratio.

*P < .05.

Table 2 presents the odds ratios and 95% confidence intervals for depression across changes in job strain and social support in our fully adjusted model. Neither change in job strain nor change in social support was associated with an increased risk of depression. There was a trend toward a protective effect of positive changes (increases) in social support; however, the effect was not statistically significant (odds ratio [OR] = 0.45; 95% confidence interval [CI] = 0.17, 1.21).

TABLE 2—

Adjusted Odds Ratios of Depression in 2004 Associated With Changes in Job Strain and Social Support Between 2000 and 2002: Canadian National Population Health Survey

Variable	AOR (95% CI)
Change in job strain
Increase	1.24 (0.57, 2.68)
No change (Ref)	1.00
Decrease	1.17 (0.50, 2.74)
Change in social support
Decrease	1.33 (0.76, 2.33)
No change (Ref)	1.00
Increase	0.45 (0.17, 1.21)

Note. AOR = odds ratio. CI = confidence interval; Odds ratios are adjusted for gender, age group, marital status and presence of children, occupation change, level of education, baseline levels of job strain and social support, presence of chronic health conditions, having subclinical depression in 2000 or 2002, family history of depression, and personal history of depression. Job strain model did not include separate measures of job control or psychological demands. The total sample was 3735.

Table 3 presents the odds ratios and 95% confidence intervals for depression across change in the components of job strain (job control and psychological demands). Respondents reporting increases in psychological demands between the 2000–2001 and 2002–2003 surveys had more than twice the risk of depression compared with respondents with change scores for psychological demands that were within the limits of day-to-day variability (OR = 2.36; 95% CI = 1.14, 4.88). We observed no appreciable differences in the risk of depression across job control change groups. The risk of depression across psychosocial work condition change groups did not differ across gender, occupation, or education groups (results not shown, available upon request).

TABLE 3—

Adjusted Odds Ratios of Depression in 2004 Associated With Changes in Components of Job Strain Between 2000 and 2002: Canadian National Population Health Survey

Variable	AOR (95% CI)
Change in job control
Decrease	1.11 (0.60, 2.06)
No change (Ref)	1.00
Increase	0.93 (0.52, 1.66)
Change in psychological demands
Increase	2.36* (1.14, 4.88)
No change (Ref)	1.00
Decrease	1.04 (0.12, 8.66)

Note. AOR = odds ratio. CI = confidence interval; Odds ratios are adjusted for gender, age group, marital status and presence of children, occupation change, level of education, baseline levels of job control, psychological demands and social support, change in social support between 2000 and 2002, presence of chronic health conditions, having subclinical depression in 2000 or 2002, family history of depression, and personal history of depression. The total sample was 3735.

*P < .05.

DISCUSSION

We examined the impact of changes over a 2-year period in job strain and its components (job control and psychological demands), and of changes in social support, on the risk for depression 2 years after the changes. We found that over a 2-year period following the change, an increase in psychological demands increased the risk of depression. Other changes in the psychosocial work environment were not associated with risk of depression by traditional measures of statistical significance. The risk conferred by increases in psychological demands remained statistically significant after adjustment for age, gender, marital status, presence of children, level of education, chronic health conditions at baseline, subclinical depression in 2000 or 2002, personal and family history of depression, and baseline psychosocial work characteristics. Furthermore, the size of this risk was similar to the size of the risks associated with family and personal histories of depression, 2 major but nonmodifiable risk factors.²¹ The risk associated with psychological demands was similar for males and females and across respondents from different socioeconomic groups (defined by education or occupation). This finding suggests that organizational changes that cause increased psychological demands at work may have a deleterious impact on the mental health of workers over a relatively short period.

The differential impact of psychological demands and job control highlights the importance of not assuming an interaction between these 2 components a priori when assessing health outcomes.^31,32 Like other researchers,^4,13,33 we found that changes in psychological demands were more important than changes in job control in the onset of depression. Unlike previous investigators,^8,9 we did not find that positive changes in the psychosocial work environment reduced the risk for depression. This divergence may be explained by the fact that we focused on depression occurring within 2 years of the change in work, whereas the previous studies examined depression up to 7 to 10 years following the change. Therefore, our failure to find this particular effect may suggest that negative psychosocial work experiences have a residual effect on the mental health of workers. As a result, the benefits of positively changing psychosocial work conditions may require more than 2 years to be realized.

The following limitations should be considered when interpreting our results. Because all measures included in this study were based on self-report, it is possible the associations presented here were inflated by common method variance bias in survey responses (e.g., because of negative affectivity among respondents).^34–36 However, we believe that by restricting the analysis to respondents who were free from depression while the work conditions were assessed, and measuring work conditions and depression at different times, we have limited the likelihood of this bias.³⁷ A second limitation pertains to the possibility that we missed some cases of depression in the 13 to 24 months between the 2002–2003 and 2004–2005 survey cycles. A limitation in the design of the NPHS is that it is conducted every 2 years, but the CIDI-SFMD assesses depression only in the 12 months preceding the survey. We partially overcame this design limitation by classifying respondents as depressed if they reported being first medically diagnosed with depression during that period. It is unlikely, however, that we identified all respondents who experienced depressive episodes during that period, as we would have missed those respondents experiencing a recurrence of depression that was first diagnosed before 2000, as well as respondents who were depressed but either did not seek medical attention or sought it but were not diagnosed with depression.

The measure of psychological demands used in our analysis is based only on 2 broad questions that do not distinguish between working pace and working hours.^38,39 However, this limitation in our questions assessing psychological demands would likely lead to bias toward the null. Therefore, if we had used a more precise measure of psychological demands, we would likely have reported even stronger effects. Finally, given the self-reported nature of our psychosocial work measures, we cannot be sure whether the changes in psychological demands reported were a result of actual changes in the work environment or changes in perceptions (within-person shifts). Therefore, our results cannot be interpreted as evidence that intentionally changing psychological demands will result in differences in depression. Such an interpretation must be based on the results of well-designed intervention studies with reliable and valid objective measures of job demands.

Despite these limitations, our study has moved this area of research forward by examining changes in work conditions among respondents free of depression at each point where working conditions were assessed (thereby minimizing the likelihood of common method variance bias), by adjusting our models for personal history of depression (thereby reducing the potential for confounding to bias the findings), and by using a relatively short period (2 years) between the measurement of change in working conditions and measurement of depression.

A final strength of our study, which warrants further discussion, is that our measure of change in each psychosocial work dimension accounted for changes attributable only to day-to-day variability. Although complexities in measuring change between time points have been discussed for many decades,^40–42 methods to accurately assess change have not been incorporated readily into work stress research. As outlined in the introduction, not taking day-to-day variability into account in a measure may lead to inferring that change has taken place when it is no more than error, or that no change has taken place when it has (just not across a specific threshold). We also ran models classifying respondents as “changed” if their job strain ratio score changed from above 1 to below 1 or vice versa (as previously done by Wang et al.⁸). Despite using the same data source as Wang et al., we did not find a difference in the risk of depression between respondents who “changed” compared with those who did not, according to this definition (results not presented, available on request). The lack of agreement between our findings and those of Wang et al. could be caused by differences in either the timing of the assessment of depression (Wang et al. examined the occurrence of depression up to 4 years after the change in work conditions were assessed) or the selection of the sample (e.g., we used a sample free of depression when working conditions were assessed).

Available data from the European Union and the United States suggest that over the past 20 years, employees feel they are working harder and have less time to get things done (both aspects of psychological demands).^43,44 Unfortunately, in the case of the United States, monitoring of the psychosocial work environment is becoming more limited.⁴⁵ Given the potential importance of increasing psychological demands in the etiology of depression, more work is required to better understand this relationship, to monitor psychological demands and other work stressors routinely at the population level, and to assess how the changes in nature of work affect the mental health of workers.