A multi-wave study of organizational justice at work and long-term sickness absence among employees with depressive symptoms

Sickness absence due to mental ill-health increases, stressing the importance of focusing on the many employees at work with mental health issues. In this study, organizational justice had a protective effect on the risk of long-term sickness absence among men, but not women, with depressive symptoms. A protective effect of favorable changes in organizational justice was not found. A multi-wave study of organizational justice at work and long-term sickness absence among employees with depressive symptoms. Scand J Work Environ Health . 2014;40(2):176–185. doi:10.5271/sjweh.3401 Objectives Mental health problems are strong predictors of long-term sickness absence (LTSA). In this study, we investigated whether organizational justice at work – fairness in resolving conflicts and distributing work – prevents risk of LTSA among employees with depressive symptoms. Methods In a longitudinal study with five waves of data collection, we examined a cohort of 1034 employees with depressive symptoms. Depressive symptoms and organizational justice were assessed by self-administered questionnaires and information on LTSA was derived from a national register. Using Poisson regression analyses, we calculated rate ratios (RR) for the prospective association of organizational justice and change in organizational justice with time to onset of LTSA. All analyses were sex stratified. Results Among men, intermediate levels of organizational justice were statistically significantly associated with a decreased risk of subsequent LTSA after adjustment for covariates [RR 0.49, 95% confidence interval (95% CI) 0.26–0.91]. There was also a decreased risk for men with high levels of organizational justice although these estimates did not reach statistical significance after adjustment (RR 0.47, 95% CI 0.20–1.10). We found no such results for women. In both sexes, neither favorable nor adverse changes in organizational justice were statistically significantly associated with the risk of LTSA. Conclusions This study shows that organizational justice may have a protective effect on the risk of LTSA among men with depressive symptoms. A protective effect of favorable changes in organizational justice was not found.

Long-term sickness absence (LTSA) involves great individual and societal costs (1)(2)(3). Major depression (4,5) but also less severe mental health problems, such as depressive symptoms (6)(7)(8)(9), are strongly related to risk of LTSA. Thus, employees with mental health problems are at increased risk of LTSA and subsequent permanent exclusion from the labor market (10,11). This is a growing public health concern as disability benefit claims and sickness absence due to mental health problems have increased in OECD countries (10). A newly published OECD report on mental health and labor market participation in Denmark concluded there is a need to address the large number of employees with mental health problems who are able to work but often struggling in their jobs (12).
In accordance with the International Classification of Functioning, Disability and Health of the World Health Organization (13), we regard the etiology of sickness absence as multifactorial (14) and characterized by a complex interplay between health conditions and contextual factors (13), including psychosocial working conditions. As working conditions are modifiable, identifying protective working conditions would be an important step in creating work environments that contribute to lower sickness absence rates and prevent labor market exclusion among employees with mental health problems. Hjarsbech et al In recent years, organizational justice has been increasingly discussed as a psychosocial working condition that might protect employees' health (15,16). Organizational justice refers to equity in decision-making procedures and the extent to which employees are fairly treated at their workplace (15,16). A recent review of 11 prospective studies showed that high organizational justice predicts a lower risk of poor mental health. Moreover, 4 of the studies reported an association of high organizational justice with decreased risk of sickness absence (17)(18)(19)(20). The review concluded there is a need for further studies that examine whether changes in organizational justice may affect health (16).
To the best of our knowledge, no study has yet investigated whether organizational justice has beneficial effects on LTSA among employees with mental health problems. In this study, we examine whether organizational justice prevents risk of LTSA among employees with depressive symptoms. We applied a longitudinal design, measuring organizational justice and depressive symptoms during five waves of data collection while events of LTSA were continuously recorded in a national register. We hypothesized that high levels of organizational justice, and improvement in organizational justice from one wave to the next, decrease the risk of LTSA in a population at high risk of LTSA.

Study design and population
Psychosocial Work Environment and Workability among Employees with Depressive Symptoms (PEWED) is a follow-up study of participants recruited from a survey of the Danish population: the Danish National Working Environment Study (DANES).
During September-December 2008, 12 584 Danish residents responded to the DANES survey (response rate 69%) (21,22). We invited these respondents to participate in PEWED if they (i) were between 18-65 years, (ii) were employed, (iii) responded by mailed or internet questionnaire, but not by telephone interview, as mode of administration can affect the answers (23), and (iv) had depressive symptoms, defined as a score of ≥10 points on the Major Depression Inventory (MDI) (24,25). Figure 1 gives an overview of the DANES and PEWED surveys. Of the 12 584 DANES respondents (T0: baseline), 2271 fulfilled the inclusion criteria for PEWED and received a questionnaire measuring organizational justice and depressive symptoms in May 2009 (T1: first follow-up). Of those, 1478 (65%) responded at T1 and constituted the PEWED study population.
We excluded participants who had stopped working (N=191) or changed job (N=103) between T0 and T1. After list-wise deletion of participants with missing values on key variables (N=150), the final study sample comprised 1034 participants.

Measures
Measurement of organizational justice. Organizational justice was measured with two items from the justice scale of the Copenhagen Psychosocial Questionnaire version II (26). After the introductory sentence "These questions are not about your own work but about the whole organization where you are employed", participants were asked: "Are conflicts resolved in a fair way?" and "Is work distributed in a fair way?" Each item had five response categories: 0 (to a very low degree), 2.5 (to a low degree), 5.0 (partly), 7.5 (to a high degree), and 10 (to a very high degree). The Cronbach's alpha of the scale was 0.73.
Responses were summed and the sum transformed to a scale ranging from 0-10, with higher scores indicating higher levels of organizational justice. We calculated the average organizational justice score from the measurements of two successive waves (ie, the average of T0 and T1; T1 and T2; T2 and T3; and T3 and T4). The average scores were then divided into groups of high (>5 points), intermediate (>3.75-5 points) and low (≤3.75) levels of organizational justice based on tertile split. Furthermore, we divided the organizational justice score from each single wave into groups of high, intermediate, and low levels of organizational justice based on tertile split. Then, we calculated the change in the level of organizational justice, as the change from one group to another at the subsequent wave (ie, the change from T0 to T1; T1 to T2; T2 to T3; and from T3 to T4). The changes were categorized depending on their direction as favorable (eg, from low to intermediate), adverse (eg, from high to intermediate), and as "no change".
Measurement of long-term sickness absence. We defined incident LTSA as >3 consecutive weeks of sickness absence during follow-up. We had four periods of approx-imately four months of follow-up from T1-T2, T2-T3, T3-T4 and T4-T4+four months, respectively. We retrieved data on LTSA by linking the unique social security number of each participant with the Danish Register for Evaluation of Marginalization (DREAM). DREAM contains all social transfer payments in Denmark for each week, including sickness absence benefits (27).
Measurement of depressive symptoms. We measured depressive symptoms with the MDI (24,25), a rating scale of 10 items assessing the presence of depressive symptoms (24). The response categories measure symptom frequency during the last two weeks as: 0 (at no time), 1 (some of the time), 2 (slightly less than half the time), 3 (slightly more than half the time), 4 (most of the time), and 5 (all of the time). Responses are summed, yielding a scale from 0-50, with higher scores indicating more severe depressive symptoms. We defined prevalence of depressive symptoms as an MDI score of ≥10. We chose this cut-off point because earlier Danish cohort studies showed that participants scoring from 9-10 points had a statistically significant increased risk of LTSA (9,28). The MDI score was assessed at T0 and in all follow-up surveys. The MDI score at T0 was used to define the study population while scores from T0-T3 were used as time-varying covariates in the analyses.
Measurement of covariates. We measured age (continuous variable), social class [categorized into (i) the salariat, (ii) intermediate occupations, and (iii) working class, in accordance with the three class model of the European Socioeconomic Classification (29)], cohabitation with partner or spouse (yes versus no), having children living at home (yes versus no), body mass index (BMI) [normal versus non-normal weight (underweight, overweight, obese)], smoking habits (never smoked versus current or previous smoker), units of weekly alcohol consumption, hours of weekly physical activity, pre-existing physician-diagnosed mental health problem (yes versus no), number of pre-existing physician-diagnosed diseases (hay fever/allergy, asthma, diabetes, cardiovascular disease, back disease, reduced hearing, skin disease, other diseases). See table 1 for a detailed description of these covariates. We selected these covariates because they have been associated with sickness absence (30)(31)(32)(33). All information on covariates came from the baseline questionnaire (T0) except age, which was obtained from registers. The covariates were only assessed at baseline and not during follow-up and were therefore treated as time-invariant covariates.
In addition to the time-invariant covariates, we measured the continuous MDI score, psychological treatment (both non-medicinal and psychopharmacological), and stressful life events as time-varying covariates during follow-up (T1-T4). Stressful life events included (i) severe illness, accident or death of close family member, (ii) own severe illness or accident, and (iii) serious problems in relationship, divorce, or end of permanent relationship. Both psychological treatment and stressful life events were used as binary variables (yes, ≥1 versus no) and selected because of their associations with mental health problems and sickness absence (33,34).

Statistical analysis
Using Chi-squared and two sample t-tests we examined differences between PEWED and employed DANES participants without depressive symptoms (MDI score <10).
We calculated time at risk of LTSA as the time until onset of LTSA or time of censoring. Participants were censored if they were no longer at risk of LTSA (eg, if they were unemployed, retired, had emigrated or died indicated by entries in the DREAM registry), when they changed their job (based on reported date of job change in the questionnaires), or at the end of follow-up, whichever came first. Follow-up time was from T1-T2, T2-T3, T3-T4, and T4-T4+four months, respectively.

Hjarsbech et al
Using Poisson regression analyses, we calculated rate ratios (RR) and 95% confidence intervals (95% CI) for the prospective association of level of organizational justice and change in organizational justice with time to onset of LTSA. The analyses included exposure to organizational justice in the four intervals between the five waves of data collection, ie, T0-T1, T1-T2, T2-T3 and T3-T4. For each of these four intervals, the prospective association between the exposure and onset of LTSA in the subsequent follow-up period was studied. For example, for the exposure period T0-T1, we examined the risk of LTSA in the four months after T1, ie, the time period T1-T2. This design is depicted in figure 2. We combined information from the four intervals in an overall analysis and used the logarithm to time at risk as an offset variable. The assumption of independence between observations is violated when using repeated measurements as measurements from the same individual are correlated. Therefore, we used Generalized Estimating Equations (GEE) to account for the dependence between the repeated observations within each participant. Analyses were conducted in PROC GENMOD, SAS 9.2 (SAS Institute, Cary, NC, USA).
Previous research on psychosocial working conditions and LTSA showed sex-specific effects (35,36). Therefore, we tested if associations between organizational justice and LTSA were sex-specific. The test of interaction was statistically significant (P=0.02) and consequently sex-stratified analyses were performed. There was no interaction between organizational justice and depressive symptoms (P=0.98).
Except for age and social class, covariates were only included in the analyses if they were associated with LTSA (P<0.05) in separate Poisson regression analyses. Accordingly, we included: age, social class, BMI, smoking habits, pre-existing physician-diagnosed mental health problem, pre-existing physician-diagnosed diseases, psychological treatment, stressful life events (model I) and MDI score (model II). The analyses of changes in organizational justice were further adjusted for the initial level of organizational justice (eg, change between T0-T1 was adjusted for level of organizational justice at T0) (model III).

Sensitivity analysis
We conducted three sensitivity analyses. First, we repeated the analyses excluding participants with LTSA one year before baseline (T0). Second, we restricted the analyses to first occurrence of LTSA only. Third, we included weekly working hours as a covariate to examine if differences between men and women in the association of organizational justice and LTSA might be explained by differences in working hours.

Results
Characteristics of the study population Table 1 compares characteristics of PEWED with DANES participants without depressive symptoms. Men and women from PEWED were more often diagnosed with a mental health problem by a physician, had more diseases, were more likely to be current or previous smokers, and had lower mean organizational justice scores than DANES participants. Men from PEWED consumed more alcohol than men from DANES, a difference that approached statistical significance. Women from PEWED were less likely to be (i) cohabiting, (ii) of normal weight, and (iii) physically active than DANES women. There were no statistically significant differences between the two samples regarding age, social class, and parental status.

Organizational justice and long-term sickness absence
The 438 men and 596 women in PEWED yielded 1440 and 2007 observations, respectively. Of the 438 men, 47 men experienced ≥1 LTSA event. A total of 82 events were observed, 23 men experienced 1 event and 24 experienced ≥1 event. Of the 596 women, 109 experienced ≥1 LTSA event. A total of 201 events were observed, 49 women experienced 1 event and 60 women experienced ≥1 event. The mean time at risk was 66.7 weeks for men and 66.4 for women. Table 2 shows the prospective association of levels of organizational justice with risk of LTSA stratified by sex. Compared to men with low levels of organizational justice, men with intermediate and high level of orga-  T0   T0   T0   T1   T1   T1   T2   T2   T2   T3   T3   T3  T4   T4   T4 T4+4 months  nizational justice had less than half the risk of LTSA. Results were statistically significant for intermediate levels of organizational justice after adjusting for all covariates, including MDI score. For high levels of organizational justice, the RR was statistically significant only in the unadjusted analysis. Among women, neither intermediate nor high levels of organizational justice were statistically significantly associated with risk of LTSA. Table 3 shows the prospective associations of changes in organizational justice and risk of LTSA. For both men and women, there was no statistically significant effect of either favorable or adverse change in organizational justice on LTSA.

Sensitivity analysis
When we excluded participants with LTSA one year before baseline from the analyses, effects of levels of organizational justice and risk of subsequent LTSA were similar to the main analyses. For the analyses of changes in organizational justice, the effect of adverse changes for men became stronger and statistically significant, however with wide confidence limits. Analyses restricted to first occurrence of LTSA attenuated the risk estimates, and the risk estimate for intermediate levels organizational justice for men was no longer statistically significant. Finally, results remained unchanged when working hours were included as a covariate. Detailed results of these analyses are available on request from the first author.

Discussion
Our hypotheses that high levels of organizational justice and the improvement thereof would be associated with a decreased risk of LTSA among employees with depressive symptoms were only partly confirmed. Among men, we found that intermediate levels of organizational justice statistically significantly predicted decreased risk of subsequent LTSA. There was also a tendency towards a decreased risk for men with high organizational justice, although not statistically significant in the adjusted models. Among women, organizational justice was not related to LTSA. In both sexes, changes in organizational justice did not predict risk of LTSA.
Findings on sex differences from previous studies of organizational justice and morbidity are inconsistent.  Two previous studies have reported that favorable changes in relational justice predicted lower risk of poor self-rated health (39) and minor psychiatric morbidity (40). To the best of our knowledge no previous studies have examined the effect of changes in organizational justice on sickness absence. In our study, changes in organizational justice were not statistically significantly associated with LTSA. Several explanations for the lack of association are possible. First, the majority of employees in our study did not experience changes in organizational justice, and the changes that occurred might have been too small to create sufficient exposure contrast to generate an effect. Second, sometimes positive changes in organizational justice might be accompanied with negative changes in other areas at work, and vice versa, which may offset any effects on LTSA. For example, the replacement of a problematic supervisor might lead to an increase in organizational justice, however, the effect on health and subsequent LTSA might be offset by heightened insecurity, lower role clarity, and lower predictability during the replacement process. Third, the potential effects of change in organizational justice might be delayed and therefore not captured within four months of follow-up.
As delineated in the introduction, we are not aware of other studies that have examined the association of organizational justice on LTSA among employees with mental health problems. A few studies have taken related approaches to examine the interplay between work environment and vulnerability on risk of sickness absence. Väänänen et al (45) found that high social support at work decreased the negative effect of psychological symptoms on sickness absence among male but not female employees. Virtanen et al (8) found no statistically significant interaction between job strain and psychological distress in predicting LTSA in a predominantly female population. Munir et al (46) reported that quality of leadership was more strongly associated with risk of LTSA among employees with depressive symptoms, however, the interaction was not statistically significant. Finally, Elovainio et al (47) found that the association of vulnerability, defined as exposure to a stressful life event, with risk of LTSA was weaker in work units with high levels of organizational justice. In sum, although the findings from these studies are not consistent, they generally appear to be in line with our finding that good psychosocial working conditions might buffer the effect of mental health problems on LTSA, and that effects might be different for men and women.

Strengths and weaknesses of the study
The repeated exposure measurements are a strong feature of this study as averages of repeated measurements are more stable and reliable than single measurements Some studies have reported no difference for men and women (15,19,20), whereas other studies have found distinct differences and suggested that possible explanations for the differences might include differences in occupational positions, social roles, expectations of supervisors (37), and relational orientation (38). Some studies (17,39,40) have reported no statistically significant sex interaction with organizational justice, but still there are indications of differences. Head et al (17) found that relational justice, a component of organizational justice pertaining to a considerate, fair and respectful treatment from supervisors (15,16), predicted sickness absence and that this association appeared to be stronger among women than men. In contrast, two other studies found relational justice to predict poor self-rated health (39) and psychiatric morbidity (40) more strongly among men than women (40). These earlier studies were all conducted in samples that did not examine the level of depressive symptoms among the employees, whereas the present study examined a sample of employees with depressive symptoms.
Mental health problems may affect men and women differently. For example, a study by Angst et al (41) found that men more frequently attributed the onset of their depression to problems at work, whereas women attributed onset of depression to relationship problems and illness or death in the family. Furthermore, depressed men experienced impairment in workability more adversely than depressed women. In another study, Stansfeld et al (42) found greater effect estimates of mental health problems with regard to sickness absence among men than women. When the analyses were adjusted for work and non-work social support and other non-work factors, the effects were no longer statistically significant for women.
Women with depressive symptoms have been found to seek treatment more often than men with depressive symptoms (41). In our study, 19.4% of men and 28.5% of women used some kind of psychological treatment. That men seek help less often than women when experiencing mental health problems might render them more vulnerable to influences from their environment, including the work environment. Thus, men may benefit more from protective working conditions than women. Furthermore, the importance of organizational justice for men may reflect the importance of the work role for men's mental health, as suggested by Plaisier et al (43) who found that paid employment was strongly associated with a reduced prevalence of depression and anxiety among men but not women.
Finally, in Denmark, the labor market is strongly sex-segregated, with women more likely to work in the public sector and men more likely to work in the private sector (44). Work values might differ for employees in the public versus the private sector and thus affect the importance of organizational justice differently.
Hjarsbech et al of working conditions (48). We used a national register to assess LTSA as register data on sickness absence are considered more reliable than self-reported data (27). The register-based outcome measure also ensured that results were not biased due to common method variance (49) and almost all participants could be followed-up. Furthermore, we adjusted for time-invariant and timevarying covariates known for their association with both sickness absence and mental health problems.
Of the 2271 invited DANES respondents, 65% agreed to participate and response rates during followup ranged from 86-91%. Given the vulnerable study population, we consider these response rates as very satisfactory. The non-response analysis revealed no selection in relation to exposure and outcome.
Several limitations of the study must be considered. First, we were restricted to measure organizational justice with only two items, derived from the four-item justice scale of the COPSOQ II (26). Other prospective studies have used more comprehensive measures of organizational justice, often the scales originally developed by Moorman (50) and modified by Elovainio et al (15) and Kivimäki et al (19). These scales distinguish between distributive, procedural, and relational justice. We cannot rule out that a more comprehensive measure of organizational justice, and a distinction between different organizational justice components, might have yielded different results.
Second, we measured both organizational justice and depressive symptoms self-reportedly, which involves the risk of reporting bias (51) that may artificially inflate associations between exposure and outcome. Although reporting bias is an important concern, we consider this issue less problematic in the present study as all of our participants had some degree of depressive symptoms. Consequently, the selection of participants likely reduced the magnitude of any reporting bias. Moreover, we assessed depressive symptoms in all waves of data collection, and our results were robust to adjustment for the time-varying MDI score.
Third, some covariates (eg, BMI and smoking habits) were only assessed at baseline because we wanted to keep the follow-up questionnaire brief to lower the burden on respondents and ensure a high response rate. Consequently, these covariates could only be analyzed as time-invariant covariates, even though some of them might have changed during follow-up.
Fourth, the study sample and number of events were relatively small for an epidemiological study on LTSA. This did not allow us to conduct further sensitivity analyses. For example, we could not conduct stratified analyses to examine whether changes from low to intermediate organizational justice scores had different effects than changes from intermediate to high organizational justice scores.
Fifth, the relation of depressive symptoms, organizational justice and LTSA is complex. We used organizational justice as a predictor variable and LTSA as the outcome in a cohort of employees who at baseline had depressive symptoms. However, other pathways are also possible. Findings from a recent review indicate that exposure to low organizational justice may be an important risk factor in the etiology of mental health problems (16). We do not know the causes of depressive symptoms at baseline and thus we cannot preclude that organizational justice before baseline (T0) could have played a role in the etiology of the depressive symptoms at baseline.
Finally, we intentionally sampled the PEWED study population from a sample of the Danish population (DANES) to represent a subgroup of employees at high risk of LTSA. Consequently, the findings cannot be generalized to general working populations, however, the findings may be generalizable to employees with depressive symptoms.

Concluding remarks
In summary, we found that organizational justice is associated with a decreased risk of LTSA among men, but not women, with depressive symptoms. Thus, improving organizational justice in organizations might be a venue for reducing LTSA among employees with mental health problems at least for male employees. Further studies with more comprehensive measurements of organizational justice are needed to corroborate the findings. These studies should also further examine sex differences in association between organizational justice and LTSA among employees with depressive symptoms and other mental health problems.