Of the working-age population in Finland, 7.4% received disability pension benefits in 2006,1 which is parallel with data from other European countries.2 This causes a substantial economic burden for society.3 Early exit from the work force due to permanent disability for work is usually a long process, often with increasing duration of sickness absences and increasing difficulties in returning to work after illness.4 5 However, research on the contribution of stressful work characteristics, such as job strain, high job demands and low job control, in this process has until recently been scarce.

The demand–control (or job strain) model is the most frequently used concept to study job characteristics provoking stress at work.6 According to this model high job strain is a combination of high psychosocial demands of the work and low control over one’s work situation.6 There is evidence, although not entirely consistent, that job strain or its components are associated with elevated sickness absence rates and increased risk of diseases that may contribute to disability for work, such as mental health problems, musculoskeletal disorders and cardiovascular diseases.5^–16 It has also been found that large organisational changes such as personnel downsizing may predict increased sickness absence rates, health problems and disability pensions,17 and that this effect may be related to increased job demands combined with reduced job control.18

To date few studies have directly examined the association between job strain (a combination of high job demands and low job control) or its components and disability pension. Self-reported job control (but not the full job strain model) was found to predict disability pension with a 1.3-fold relative risk in the low job control group compared with those having high control over their work.19 In a study of 1038 Finnish men aged 42–60 years, those reporting job strain from mentally strenuous work were 2.5 times more likely to retire due to disability than the non-strain group.20 In addition, some studies have found associations between job demands, job control and intentions to retire early.21^–23

Two limitations may compromise the value of this evidence. First, a potential source of bias is reliance on self-reports, as dispositional differences between individuals are known to affect their response style. In all the aforementioned studies, the measurement of job strain was based on employees’ own reports. One way to reduce potential self-reporting bias is to use work unit-aggregated scores by calculating the mean levels of job demands and job control for each work unit and link these work unit-aggregated scores to each participant. The aggregated scores method has been used in studying the associations of job control, job demands and sickness absence,24 which is a risk factor for early retirement,24 but as far as we know no such study with respect to disability pension has been conducted.

Second, no earlier study has taken into account the fact that employees belong to work units. Thus, the statistical dependence among the scores of employees in the same group has been discounted in previous analyses treating each subject as an independent observation. Among people working in the same unit, intentions to leave work or actual exit from work may have their origin in the shared attitudes between co-workers. Ignoring variation between work units and groups may therefore inflate the statistical strength of the association between the variables. Multilevel modelling provides an adequate tool to take into account this variation. There is prospective evidence from multilevel studies of an association between job strain and sickness absence,13 14 but evidence from multilevel studies with respect to disability pension is largely lacking.

To take into account the limitations above in studying the effect of job strain on disability pension, the present study used two indicators of job strain, self-assessed and work unit-aggregated, and took into account in all analyses the fact that individuals were nested in their work units.

METHODS

Study sample

The Finnish public sector study is an ongoing prospective study exploring the relationships between behavioural and psychosocial factors and health among public sector employees working in 10 towns and 21 hospitals.25 The sample covers almost 20% of the full-time employees working in the Finnish municipal sector, the most common occupations of the participants being nurses, teachers and cleaners.

Sample selection is described in fig 1. The baseline survey was conducted in 2000–2, 48 598 responded (response rate 68%). In the Finnish public sector, the granting of pension for people less than 55 years of age is exclusively determined by medical reasons and disability, whereas among older age groups other reasons for early retirement also exist. In order to ensure that the retirement was due to disability rather than any other cause (eg, old age pension, voluntary pension), we excluded those over 50 years at baseline, thus all participants were younger than 55 years at follow-up in 2004–5 (31 481 participants were left in the sample). Furthermore, we excluded all those with missing data on retirement status at follow-up, that is baseline respondents who were not in the service of the target organisations in 2004 and did not respond to the follow-up survey in 2005 (n = 3068). Finally, respondents with missing data in any of the study variables were excluded (n = 3283). Thus, the final sample consisted of 25 150 participants (4764 men and 20 386 women) aged 50 years or younger at baseline and either at work or retired because of disability at follow-up.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Sample selection.

Job strain

Job strain was assessed at baseline by using measures derived from the Job Content Questionnaire.26 The job demands scale consists of three items enquiring about workload and pace of work (Cronbach α = 0.77):

1. I have to work very hard.

2. My job involves an excessive amount of work.

3. I don’t have enough time to get my work done.

Job control was measured by a nine-item indicator (Cronbach α = 0.82, items 4 and 9 reversed when computed):

1. My job allows me to make a lot of decisions on my own.

2. My job requires me to be creative.

3. My job requires that I learn new things

4. My job involves a lot of repetitive work.

5. I have a lot of say what happens on my job.

6. My job requires a high level of skill.

7. I get to do a variety of different things on my job.

8. I have an opportunity to develop my own special abilities.

9. On my job, I have very little freedom to decide how I do my work.

All items used a 5-point Likert-type response format ranging from 1 (strongly disagree) to 5 (strongly agree). A total score for both constructs was computed and the scores were further divided into quartiles to indicate different exposure levels. To create a job strain indicator, demands and control were split on the median and combined to four categories: low strain jobs (low demands combined with high control), active jobs (high demands combined with high control), passive jobs (low demands combined with low control) and high strain jobs (high demands combined with low control).6

In addition to the self-assessed individual measures, we computed a work unit-aggregated score by assigning to each employee the work unit average of co-workers’ job strain scores (ie, the average of scores of all other respondents in the work unit except the individual respondent’s own score).24 27 28 Work unit-aggregated scores were computed only for units with at least five persons as work unit level job strain, job demands and job control are only meaningful in groups. The total number of work units (eg, schools or hospital wards) was 2640, mean size 15 people and range from 5 to 366 people.

Other baseline characteristics

Demographic characteristics were obtained from employers’ registers and included participants’ gender, age and socioeconomic position (higher grade non-manual, lower grade non-manual, manual; based on Statistics Finland classification).29

The health risk behaviour assessed with standard survey measures included current smoking status (smoker vs non-smoker), high alcohol consumption (average weekly consumption ⩾190 g of absolute alcohol for the women and >275 g for the men),30 obesity (body mass index, BMI, from self-reports of height and weight ⩾30 kg/m²), and leisure time physical activity (<2.0 metabolic equivalent task (MET) hours per week, corresponding approximately to 30 min of walking per day).31

Health indicators

Participants were asked whether they have or have had any of the following diseases diagnosed by a physician:32 cardiovascular disease (including myocardial infarction, stroke, diabetes, chronic hypertension), musculoskeletal disorders, respiratory diseases (asthma and chronic bronchitis) and psychiatric disorders (depression or other). Other indicators of health were self-rated health status (average or worse vs good or very good health) and psychological distress (GHQ caseness, score ⩾4, on the 12-item version of the General Health Questionnaire (GHQ).33

Disability pension

The employment status of the participants at follow up was assessed by asking them to choose the most suitable of alternative response options asking them what they mainly do (for a living) at the moment. From the list of 15 different options, the alternative of “being retired because of work disability, either permanently or temporarily” was taken as an indicator of disability pension (disability pension cases n = 93). In the Finnish pension system a disability pension is granted either for a fixed term (temporary) or permanently and mainly after being on sickness leave for 1 year. Unfortunately our data did not enable us to make a distinction between a temporary and a permanent disability pension or with part-time and full-time disability pension (granted due to 40% or 60% loss of work capability, respectively).

Statistical analysis

We applied a multi-level logistic regression analysis that used the generalised estimating equations method (GEE) to study the associations between job strain and its components, demands and control and the risk of disability pension. This method takes into account the non-independence of observations within groups (ie, participants within work units). We used intraclass correlation (ICC) to study the similarity of individual responses within work units. In a multilevel setting, the ICC is the proportion of the variance in each job strain indicator that occurs at the work unit level.34 In our sample, the ICC was 21% for job control and 15% for job demands, which indicate significant variance between work units and justify the operationalisation of the concept of job strain, job control and job demands as a work unit level phenomena.

In analysis, women and men were pooled because there was no strong evidence for gender-specific associations (for all interactions between gender and job strain or its components p>0.15). ORs and their 95% CIs were sequentially adjusted for age and gender, socioeconomic position, health risk behaviours and health indicators. The analyses were conducted separately for self-assessed and work unit-aggregated measures of job strain and its components. The hypothetically least stressful work condition (ie, low demands, high control or low strain) was selected as the reference category in each indicator. Statistical analyses were performed using SAS 9.1 software.

RESULTS

Table 1 shows the baseline characteristics of the 25 150 participants and age-adjusted and sex-adjusted ORs for subsequent disability pensioning by levels of baseline characteristics. Of the participants, 81% were women, 84% were non-manual employees and the mean age was 40.2 years. Ninety-three individuals were granted a disability pension during the follow-up. An increased risk of becoming the recipient of a disability pension was found for older employees, manual workers, current smokers, those physically inactive and obese employees. An increased risk for disability pension was also found for people with musculoskeletal disease, respiratory disease, psychiatric disorder, psychological distress or non-optimal self-rated health. Gender, alcohol consumption and the status of cardiovascular disease did not predict the disability pension.

Table 2 shows the results from logistic regression analyses on the associations between self-assessed job demands, job control, job strain and subsequent disability pensioning. Self-assessed job demands were not associated with disability pensioning. Self-assessed job control predicted disability pension in models adjusted for age, gender and socioeconomic position, but this association attenuated after further adjustments for health risk behaviour and health indicators. For employees having a high strain job compared with those with a low strain job the risk for disability pensioning was 2.6 times greater after adjustments for age, sex, socioeconomic position and health risk behaviour. After further adjustment for health indicators this OR attenuated to 1.7 and did not remain statistically significant. However, there was an association between passive jobs and increased risk for disability pension irrespective of adjustments.

Table 3 shows the results on the associations between work unit-aggregated job demands, job control, job strain and subsequent disability pensioning. The association between job demands and disability pensioning was not statistically significant and job control predicted disability pension only in the age-adjusted and sex-adjusted model. Employees with high work unit job strain had a 2.3-fold increased risk of disability pension compared with those with low work unit-aggregated job strain after adjustment for all baseline covariates.

DISCUSSION

This prospective study of 25 150 public sector employees showed high job strain (ie, high job demands combined with low job control) to be predictive of subsequent disability pension. The odds for becoming the recipient of a disability pension was 2.3-fold for the work unit high strain group compared to the low strain group after taking into account for demographic characteristics, health risk behaviour and various health indicators. To our knowledge, this is the first study testing the job strain model in regard to disability pension taking into account the potential bias arising from the fact that employees are nested within work units. Furthermore, in order to reduce subjectivity bias we measured job strain and its components demand and control by using work unit-aggregated scores in addition to individual’s self-reports. Since job strain measurement in previous studies has been based on employees’ own reports, it may have been systematically biased by an individual’s psychological disposition or other characteristics. Given that such dispositional characteristics may also affect intentions to retire early, the observed associations between self-reported job strain and disability pension may have been overestimates or spurious.

Although our findings of self-assessed and work unit job strain indicators were largely converging, there were some differences in regard to health adjusted models. The association between work unit job strain and disability pension was robust to adjustments for health indicators, but when self-assessed strain measure was used the effect attenuated. Because disability pension is granted for medical reasons only and this study excluded other reasons for early retirement than disability, it is apparent that the effect of job strain is mediated by poor health. In fact, disability pension could be regarded as an endpoint of occupational health deterioration. The attenuated association between self-assessed job strain and disability pension after adjustment for health indicators may thus reflect the effect of health as a mediator. On the other hand, alternative explanations cannot be excluded. Thus, it is also possible that the association between job strain and disability pension is confounded by the current health status of the respondent, that is that poor health both leads to high strain at work and results in disability. However, this is not a plausible explanation for the association between work unit job strain and disability pension, because the work unit-aggregated measurement was independent of the respondents own characteristics, including his/her health at baseline.

Our finding that self-assessed passive jobs predict disability pension was not initially hypothesised, although there is previous evidence of passive work conditions being hazardous to health.35 It may be speculated that passive jobs (ie, low demand combined with low control) are experienced as unimportant or worthless jobs not giving any occupational satisfaction and in this sense they may increase intentions to retire. Thus, the association between passive work and disability pension may relate to a motivational rather than the disease-mediated pathway discussed above. We may also have found this association by chance because it was not replicated in the analyses based on work unit-aggregated assessment of passive jobs.

The findings of this study show that low socioeconomic position, health risk behaviour, doctor-diagnosed diseases, psychological distress and poor self-rated health increase the risk of disability retirement and are in accordance with the results of previous studies on the risk factors of disability pensions.5

Study strengths and weaknesses

Prospective design, large employee data, acceptable response rate, control for a number of potential and already known confounding factors, and accounting for clustering by work units by the use of multilevel statistical procedures were the specific strengths of this study. We assessed job strain by using individual scores and work unit-aggregated scores. The two methods have different strengths and different potential limitations. Self-assessment is sensitive to individual differences in job strain, but it is also open to subjectivity bias. In contrast, work unit assessment does not capture differences in job strain between members of the same work unit but as an inferred measure it is less open to subjectivity bias. Having consistent findings from both of these methods strengthens the evidence that job strain is a robust predictor of disability pension. In addition to the methods used in our study, some studies assessing job strain by external evaluators have also been conducted.36 37 In the future, this method should also be tested with regard to the disability pension.

The following limitations need to be considered when interpreting the findings. First, job strain was assessed at one point in time only. As such a measurement may imperfectly reflect long-term job strain, our findings may underestimate the association between job strain and disability pension. Second, measures of health risk behaviour, health indicators and disability pensioning were based on self-reporting. Imprecision in the measurement of covariates and outcome could either artificially inflate or attenuate the association between job strain and disability pension. Third, the follow-up period for disability pension was relatively short (max 5 years); only 93 employees younger than 55 retired due to disability during this period. Furthermore, it was not possible to make a distinction between temporary and permanent disability pension. Although temporary disability typically precedes permanent disability pension, it may be hypothesised that the effect of job strain on temporary disability pensions may differ from that on permanent disability. Thus, further research is needed to examine whether job strain is a long-term predictor of various forms of disability pension. Four, there was a time lag of 3 years in baseline measurements (2000–2); thus, the follow-up period to 2005 varied between 2 years and 5 years. This may have affected the association between job strain and disability. However, the interaction term “job strain × length of follow-up period” was non-significant (p>0.07 for all individual and work unit-aggregated scores), suggesting that a major effect is unlikely. Fifth, we did not have diagnosis-specific information on the disability pensions, which could have given us some hint of possible pathways through which job strain has influence on disability pensions. Finally, our findings are based on data from a cohort of public sector employees and thus are not necessarily generalisable to other sectors or more male-dominated working populations.

Policy implications

Prevention of job strain among employees is an important target for organisations. If the predictive associations found in this study were causal, our results suggest that reducing work stress may also reduce risk for work disability among employees younger than 55. Several processes might contribute to such benefits: alleviating job strain might directly reduce stress-related morbidity or it could help employees otherwise in a serious risk for work disability to manage with their job despite their disease or disability. Reduced levels of job strain might also keep in the labour force those individuals whose health and stress-coping capacity may not be excellent, but good enough for lower strain jobs. Importantly, there is some evidence that job strain and its components may have an effect on the return to work after disability/sick leave.38^–40 Keeping these people at work will be an issue of great importance in the future labour market. However, intervention studies will be needed to determine effective ways to reduce job strain,41 42 and to assess whether the reduction of job strain has effect on early exit and, if so, what kind of processes underlie these positive outcomes.

Conclusions

Our evidence suggests that high job strain is an independent risk factor for a disability pension. If causal, the strong association also found in work unit-aggregated measurements brings into focus the potential relevance of designing organisational (vs individual) interventions when aiming to reduce disability for work associated with work stress.