Retrospectively assessed psychosocial working conditions as predictors of prospectively assessed sickness absence and disability pension among older workers

This prospective follow-up study links data on work environment and health from the Copenhagen Aging and Midlife Biobank (CAMB) with a Danish register containing information on labor market attachment. The design and methods of the study are described elsewhere [21]. To ensure transparent and standardized reporting of the study, the STROBE checklist was followed [22].

CAMB contains data on biological, psychological and social factors for persons between 49 and 63 years of age from the merging of three established cohorts: The Metropolit Cohort [23], The Copenhagen Perinatal Cohort [24] and the Danish Longitudinal Study on Work, Unemployment, and Health [25]. Of the 17,937 individuals invited for the CAMB data collection in 2009–2011, 7190 responded to the questionnaire and 5575 attended a clinical examination. More information on the study population can be found elsewhere [26]. For the present study, employed wage earners at baseline (n = 5076) were included. Because not all participants answered all the survey questions, the exact number of individuals for each analysis varies. Baseline characteristics of the study population are illustrated in Table 1. The present manuscript is part of a larger study setup investigating the influence of physical and psychosocial work environment throughout life and physical and cognitive capacity in midlife on labor market attachment among older workers. From this larger study setup, a study protocol and analyses on the physical work environment have previously been published [21, 27, 28].

Psychosocial working conditions throughout working life were assessed retrospectively by 12 items on specific psychosocial work characteristics derived and modified from the COPSOQ [29]. Each item represents one distinct COPSOQ scale. The original COPSOQ scales were of multiple items (with the exception of work pace that is a single item COPSOQ scale), but due to limited space, it was decided in the CAMB study to only include one item per scale.. The items and their associated response categories can be seen in Table 2. They belong to three COPSOQ domains: Demands at work (quantitative demands, work pace, cognitive demands, emotional demands), Work organization and job contents (influence at work, possibilities for development), and Interpersonal relations and leadership (recognition from management, role clarity, role conflicts, social support from colleagues, social support from supervisors, social community at work) [29]. For the purpose of analysis, the five response categories of each of the 12 psychosocial work characteristics were reduced to three categories (high, medium, low) with the high category consisting of the two response categories in strongest agreement with the item, the medium category corresponding to the middle response category, and the low category corresponding to the two response categories in least agreement with the item [3].

In addition, supplementary post-hoc analyses combining the 12 psychosocial work environment factors into higher-ordered constructs were performed. This included: (i) three COPSOQ domains of demands at work, work organization, and job content, and interpersonal relations and leadership, (ii) a demand - job control ratio, and (iii) a demand - reward ratio. The demand – job control ratio was created by dividing the demand domain by the domain of work organization and job content to approximate the job strain model [30]. The demand-reward ratio was created by dividing the demand domain by score combining the items of “recognition from management”, “social support from co-workers” and “social support from supervisors” to approximate the effort-reward imbalance model [31].

Information on LTSA and disability pension was derived from the Danish Register for Evaluation of Marginalization (DREAM), containing information on all social transfer payments in Denmark [32] and linked to the CAMB cohort via the unique Danish personal identification number. In DREAM, sickness absence is recorded on a weekly basis when the employer is entitled to reimbursement of the sickness pay [33]. In Denmark, the employers have the right to receive sickness compensation benefits from the municipalities for an employee experiencing sickness absence for a given minimum period (up to 2007,> 13 days; 2007–2008,> 14 days; since 2008,> 21 days; since 2012 > 30 days) [32]. Hence, during our follow-up, the period during which the employer received no reimbursement for employee sickness absence changed from 21 days to 30 days of sickness absence (January 2012). To define LTSA consistently throughout this period, it was defined as sickness absence > 30 calendar days, corresponding to ≥6 consecutive weeks in DREAM. In Denmark, a disability pension is a social benefit for people with a significant and permanent loss of work ability with a compensation period lasting until retirement age. Individuals with a permanent loss of work ability and working on special terms, such as “light jobs” (work on special terms with a wage subsidy offered to people on disability pension) and “flexible jobs” (a job offer on special terms for people with permanently reduced work ability), or vacancy benefit for individuals with flexible job, were also classified as receiving disability pension [34].

Physical work environment throughout working life was assessed by the following question: “Looking back on your entire working life: For how many years of your working life have you had…, 1) mostly sedentary work without physical strain?, 2) mostly standing or walking work without major physical strain?, 3) mostly standing or walking work with some lifting and carrying?, 4) mostly heavy, fast or physically demanding work?”. For each response category respondents listed the number of years of working life (cumulative exposure assessment) with the specific effort level [35]. The data on exposure years in each of the 4 categories was transformed to a number between 0 and 100, where 0 indicates that all exposure years belong to category 1 (sedentary work) and 100 indicates that all exposure years belong to category 4 (very hard physical work), and anything in between was linearly scaled. The categories were defined as “low physical work demands” (0–24.99), “moderate physical work demands” (25–49.99), “high physical work demands” (50–74.99) and “very high physical work demands” (75–100) [27].

Physical activity level during leisure time was assessed by the following question: “What would you say best describes your spare time physical activities?” with the following response categories: 1) go for competitive sport regularly and several times a week; 2) go for physical training or heavy house or garden work at least 4 h per week; 3) go for walks, biking or other kinds of light exercise at least 4 h a week (including Sunday excursions, lighter garden work and biking/walking to and from work); 4) read, watch television or have other sedentary activities. For further analyses, a variable was generated with the number ranging from 1 to 4 representing the selected response category.

Chronic diseases were assessed by the following question: “Do you have or have you had any of the following diseases?” with the response options “yes, have now”, “yes, previously” or “no” to the following diseases: back disease, cancer including leukemia, chronic anxiety or depression. Three diseases were included - back disease, cancer including leukemia, and chronic anxiety or depression - because these were the only diseases that predicted LTSA in a previous analysis of Danish employees [36]. For each chronic disease, a binary variable was generated: 1) representing that the participant has or have had the specific disease, and 2) representing that the participant never had the specific disease.

Level of education was categorized into five groups; unskilled, skilled, and short-, medium-, and long education [37]. Skilled labor refers to labor that requires workers who have specialized training or a learned skill-set to perform the work. These workers could be either blue-collar or white-collar workers, with varying levels of training or education. Unskilled labor refers to labor that requires no other education or professional qualifications than primary school education. Short-, medium-, and long education refer to short-, medium-, or long-cycle further education than a high-school education. For further analyses, a variable was generated corresponding to the educational level, with 1 representing the lowest level of education (i.e. unskilled) and 5 representing the highest level of education (i.e. long education).

Smoking was evaluated by a question from the CAMB questionnaire: “Do you smoke?” With the response categories: 1. Yes, daily; 2. Yes, but not daily; 3. No, but I have smoked previously; 4. No, I have never smoked. For the present analyses, response category 1, 2 and 3 were collapsed into “smoking” while response category 4 represented “never smoking”.

Height and weight of participants were measured by the clinical personal and body mass index (BMI) was calculated as BMI (kg/m²) = body weight/(height)².

The Cox proportional hazard model [38, 39], censoring for competing events, was used for modeling the risk of register-based LTSA and disability pension during the 4–6 year follow-up period (i.e. baseline measurements were collected from 2009 to 2011 and register follow-up took place in 2015). A competing risk approach was employed censoring for all events of permanent drop-out from the labor market within the follow-up period. For instance, the analysis with disability pension as outcome was censored for statutory retirement, early retirement, immigration, and death (derived from DREAM). The participant was censored in the sense that nothing was observed or known about the participant after the time of censoring. A censored participant may or may not have an event after the end of observation time. On the contrary, when individuals had an onset of LTSA or disability pension within the follow-up period, the survival times were non-censored and referred to as event times. Analyses were carried out separately for each individual psychosocial work characteristic and adjusted for multiple confounders: age, gender, physical work environment, lifestyle factors (physical activity level, BMI, smoking), chronic diseases, education, and LTSA over the preceding two years prior to baseline (derived from the DREAM register). The estimation method was maximum likelihood and the results are reported as hazard ratios (HR) with 95% confidence intervals (CI).

In the supplementary post-hoc analyses, the 12 single psychosocial work environment factors were replaced with five higher-ordered constructs, including the three COPSOQ domains of demands at work, work organization and job content and interpersonal relations and leadership, a demand-job control ratio and a demand-reward ratio. The standardized Cronbach’s alpha of these new constructs were 0.57 (COPSOQ demands domain), 0.48 (COPSOQ work organization and job content domain, which was identical with job control score), and 0.65 (COPSOQ interpersonal relations and leadership domain), respectively.

During the follow-up period, the following number of outcome events occurred: 970 individuals (19.3%) had at least one episode of LTSA and 85 individuals (1.7%) received disability pension.

Table 3 shows the result for the association of psychosocial work characteristics with risk of LTSA and disability pension. LTSA was predicted by high levels of cognitive demands (HR 1.31 (95% CI 1.10–1.56), p = 0.003), high levels (HR 1.26 (95% CI 1.07–1.48), p = 0.005) and medium levels of emotional demands (HR 1.19 (95% CI 1.02–1.40), p = 0.031), low levels of influence at work (HR 1.30 (95% CI 1.03–1.64), p = 0.028), medium levels of role clarity (HR 1.33 (95% CI 1.11–1.59), p = 0.002) and high levels of role conflicts (HR 1.34 (95% CI 1.09–1.65), p = 0.007).

Disability pension was predicted by low levels of influence at work (HR 2.73 (95% CI 1.49–5.00), p = 0.001) and low levels of recognition from management (HR 2.04 (95% CI 1.14–3.67), p = 0.017).

Table 4 shows the result of the post-hoc analyses on the three COPSOQ domains, and on the demand-job control ratio and the demand-reward ratio. LTSA was predicted by high demands, poor work organization and job content and medium demand-reward ratio. Disability pension was predicted by medium and poor work organization and job content.

It was examined whether numerous psychosocial work environment factors, measured with single items, predicted the risk of LTSA or disability pension. The study was motivated by reviews showing that the vast majority of psychosocial work environment studies have examined job strain as a risk factor for LTSA and disability pension while failing to study other potentially important psychosocial work environment factors [12, 20, 48]. Thus, the study was explorative, i.e. the aim of the study was not to test a specific hypothesis, but to explore a wide range of psychosocial exposures.

Psychosocial working conditions during working life were retrospectively assessed by self-reports at mid-life and could, therefore, be prone to potential bias, in particular, recall bias. Thus, it was not possible to analyze the psychosocial work environment throughout the whole working life, as the psychosocial working conditions were only measured once when he participants were asked to assess retrospectively how these conditions were during their working career. It seems reasonable to assume that psychosocial working conditions are not static during the course of a working life, but instead are prone to changes. Asking participants to combine exposures during their working life in a single number for their average level of influence at work, social relations or role conflicts (among others) is only a first and very limited step towards a life course perspective in occupational psychosocial epidemiology. At the next step, studies are needed examining changing psychosocial working conditions repeatedly over time and whether exposure is more important at specific time points and less important at other time points. In addition, the retrospective assessment of earlier working conditions in our study may be affected by current health- and psychological status [49, 50], although this bias was probably reduced by the adjustment for baseline chronic diseases.

Another limitation to the study is that all psychosocial constructs were measured with single-items from the COPSOQ questionnaire and not with multi-item scales that have undergone a thorough psychometric testing. Not using the entire validated scales hampers the comparison to other studies measuring psychosocial working conditions by all multi-item scales from the COPSOQ. Due to competition in space in the CAMB questionnaire survey, a high number of scales with single items were preferred over a low number of scales with multiple items. The single-item approach allowed for the exploration of a wide range of different exposures, responding to requests in the literature to broaden the perspective on psychosocial risk factors of sickness absence and disability pension [20]. However, the result of the post-hoc analyses on the three COPSOQ domains exhibited a similar pattern of results as the analyses based on single-items, which lends credence to the findings in the main analysis of the present study.

The low response rate of the CAMB survey is a limitation to the study and could have led to selection bias. Previous analyses of the CAMB cohort found that participants and non-participants were comparable in terms of health and education but that a greater proportion of participants than non-participants were employed at the time of the survey suggesting that the study sample may represent a partly socially selected group [26]. In addition, it cannot be ruled out that those with the longest history of hazardous psychosocial exposures could already be outside the labor market when the present study was initiated, suggesting that the healthy worker effect could be a source of confounding bias.

The low number of cases for disability pension is an important limitation to the study. Thus, a possible explanation for the lack of association between most psychosocial factors and risk of disability pension could have been insufficient statistical power. As two predictor variables and two endpoints were included, the number of statistical tests conducted was high and it cannot be ruled out that some of the statistically significant associations might have occurred by chance. Whether or not analyses should be adjusted for multiple testing is controversially discussed in the literature [51, 52]. After weighing the arguments it was decided not to adjust for multiple testing, also because of the exploratory nature of the study.

Caution is warranted regarding the generalization of the results to countries other than Denmark or the Scandinavian countries. Van der Wel et al. [53] previously stated that the welfare systems of Scandinavian countries are better at protecting against non-employment due to illness than other systems. Lunau et al. [54] reported that the association of adverse psychosocial working conditions and risk of depressive disorders was weaker in countries with active labour market policies, high unemployment benefits and low-income inequality (e.g. Denmark) and stronger in countries that lack these labour and social policies. Hence, future studies should examine the generalizability of the present results to other countries with different arrangements of disability pensions and labor market protection.

A strength of the study is the use of data on labor market outcomes derived from the DREAM register that has high reliability, because all transfer payments are systematically recorded [55]. Hence, DREAM makes it possible to assess labor market attachment that is free from potential bias from self-reported LTSA and disability pension. Additionally, by using register data on labor market attachment, common methods variance (i.e. associations between exposure and outcome due to same assessment methods) and recall bias with regard to outcome ascertainment was eliminated [56]. A somewhat high prevalence of LTSA was observed in the present study, where 19.3% of the study population had at least one episode of this outcome event during the follow-up period. Pre-study power calculations, based on extracts from the DREAM database of people aged 50–59 years who were working at baseline (n = 757,226), revealed an incidence rate of 11.7% (new cases of LTSA) over a 3 year period (from 2009 to 2011) [21]. Taken into account the longer follow-up time (3-5 years), this is somewhat in agreement with the proportion that we observed in the present study sample. In addition, the introduction of a new disability pension reform in 2012 have made it more difficult to be granted a disability pension in Denmark and the average age to obtain a disability pension has increased from 45.8 years in 2011 to 48.1 years in 2015. Thus, it can be suggested that a proportion of people between 50 and 60 years with declining health are not anymore applicable for a disability pension, which could cause an increased probability of sickness absence. In line with this, we observed a lower incidence of disability pension in the study population compared with pre-study power calculations performed before the introduction of the new disability pension reform [21].