The present study indicates that work demands, such as awkward body postures, lifting heavy loads and work pace, are associated with an increased risk of early exit from work, whereas resources at work, such as influence and quality of leadership, might be associated, although with a lower strength, with a decreased risk of early exit. In fact, our analyses suggest that a quarter of early exits are attributable to awkward body postures, lifting heavy loads and work pace. The study also indicates some non-linear associations; the risk of early exit for exposure to awkward body postures was only elevated at very high levels, the risk associated with high work pace was elevated already at relatively low levels (and did not increase with further increasing levels), while the risk for exposure to low quality of leadership was elevated at only very low levels.
An increased risk of early exit associated with physical work demands clearly emerges, possibly attributable to exposure to awkward postures or heavy lifting, although, because of their high intercorrelation, it was not possible to enter in the model simultaneously these two variables to determine their individual effect. Another important risk factor for early exit appears to be that of psychological demands, captured by the variables work pace and amount of work, but also for these dimensions their strong correlation limited the possibility to estimate reliably their independent effect in a multiple regression model. Although work pace may be an indicator of physical demand, it showed only a mild correlation with physical factors (Pearson correlation 0.06–0.17), allowing to assess the effect of exposure to high psychological demand controlling for physical demand and other work factors (adjustment for two physical factors and two psychosocial factors decreased the risk by one fifth). Another possible predictor for early exit might be job control, although influence at work was only marginally associated to early exit in the mutually adjusted model; maybe its effect is too small to be detected in this population (143 premature exits out of 2351 employees).
This study also showed that some physical and psychosocial work dimensions proposed in the literature partly overlap. High correlations were observed especially within work domains, for example, between work pace and amount of work (domain of quantitative demand), between control over working time and influence at work (domain of control), or among different factors in the domain of physical demand (awkward body postures, heavy lifting and walking/standing) (Table 3). From a theoretical point of view, there is a lack of understanding of the interdependence—and uniqueness of—specific working condition dimensions. Statistically this problem shows itself through two insufficient solutions: in multiple regressions, it poses problems to do mutual adjusted analyses when entering all working conditions into the same model, as also intercorrelations as low as 0.25 would lead to multicollinearity, making interpretations of risks impossible (Vatcheva et al. 2016); to solve this problem by constructing metascales, such as demands and resources, poses other problems, because of possible differential subscale effects (Burr and d’Errico 2018). In the present paper, we therefore did mutually adjusted regression models where we only controlled for a limited number of working conditions (Tables 4, 5).
Also, the study seems to indicate an association between the job strain construct and early exit. We found that job strain ceased to predict early exit when controlled for physical demands and quality of leadership. One could argue that these covariates in part contribute to high demands and low control; thus, adjustment for these factors might represent an overadjustment. Further, the study indicated that demands and control interact, i.e., job strain poses a risk for early exit over and above what one could expect when considering the respective risks of demands and control. More well-powered studies should look more at possible interactions among working conditions on early exit.
The results on specific early exit routes seem to indicate that the working environment has somewhat stronger associations to the sickness absence route than to those of unemployment and pensioning. Only work pace and role clarity were associated with the unemployment route, and only control over working time decreased the risk of taking the pension route. However, due to the low number of events, the results on single exit routes should be interpreted with caution. The work environment risk factors found—lifting heavy loads, repetitive movements, work pace and amount of work—have been found to be associated to subsequent poor mental health and/or musculoskeletal complaints leading to sickness absence (da Costa and Vieira 2010; Theorell et al. 2015).
Strengths
This study of employees with a broad age range and examining several work environment dimensions is the first of its kind in Germany. A major strength of this study is that it used validated instruments, such as COPSOQ (Nübling et al. 2006; Pejtersen et al. 2010), and the employment history tool (Borsch-Supan et al. 2013). Second, the study is relatively large, including 2351 subjects. Third, the adjustment for several societal and household covariates, in particular for SEP and income, is expected to have reduced the possibility that the observed associations have been confounded by other subjects’ characteristics. We did not consider a control for SEP as an overadjustment, as SEP is expected to have a major independent impact on early exit from work (Schuring et al. 2013; Visser et al. 2016). Also, the control for other physical and psychosocial exposures in the fully adjusted model on early exit from work allows excluding relevant distortions in the risk estimates due to confounding by other work factors, in contrast to most other studies on the subject. Fourth, this analysis has dealt with more exit routes out of work apart from pensioning, namely sickness absence and unemployment, which are more difficult to operationalize as these states could be recurrent. People taking these routes would in many cases only in retrospect see them as exit routes. We have used two quite stringent criteria to define the exit routes sickness absence and unemployment. Either sickness absence or unemployment disregarding duration had to be followed by pension, or sickness absence or unemployment had to last at least 18 months. This length was chosen, because it was associated to a high risk (< 75%) of later early exit (it is a pure coincidence that we found the same duration cut point for both these routes). Future studies with access to labour market data covering a longer period may enable a definition of early exit routes in a more refined way.
Limitations
The strengths of this study need to be balanced against its weaknesses.
First, this study is observational, where selection bias has to be considered, also in the light of the low participation in the cohort. Based on comparisons with the study’s sampling frame, differences in participation at baseline by gender and age were limited (Table 1), whereas they were greater between SEP strata, with a response fraction almost 10% lower among unskilled workers, compared to professionals, managers and semi-professionals. Similar differences by SEP were also observed for participation at follow-up. Several other researchers have reported lower participation in surveys and epidemiological studies among subjects in more disadvantaged social positions (Cifuentes et al. 2008; Demarest et al. 2013; Goldberg et al. 2001; Goyder et al. 2002; Lissner et al. 2003), for reasons which are still not well understood. Differences in attrition by level of exposure to the different work factors were of magnitude similar to those observed by socioeconomic position (Appendix Table 8) and appeared, at least in part, explained by their association with socioeconomic position, as high levels of walking/standing and awkward body postures were found in lower social class, whereas high levels of amount of work, control over working time and possibilities of development were found in higher social class (Table 3). SEP differences in attrition are not expected to have caused a substantial distortion of the associations away from true effects, considering that differences were relatively small (maximum difference in response rate at follow-up: 13% points between subjects exposed to low or high possibilities for development), and that all analyses were adjusted for socioeconomic position. An analysis with design weights so as to control for attrition did not change the results (Appendix Table 9).
Second, all physical demand dimensions were only measured with one to two items. This might lead to some measurement error regarding these variables and to a consequent non-differential misclassification of the exposures, which in turn would produce an underestimation of their associations with early exit.
Third, the study did not consider social support, as we had concerns regarding the validity of the available COPSOQ 1 question in the S-MGA (Burr et al. 2019). Social support from supervisors has been shown to be strongly correlated to quality of leadership (Burr et al. 2019), which we included as possible risk factor in the present study. In contrast, we could not assess possible effects of social support from colleagues.
Fourth, as the study was based on self-reports, it cannot be ruled out that people with poor health have exaggerated physical or psychosocial demands and underestimated psychosocial resources, which could have led to an overestimation of the associations, if subjects with poorer health, as expected, were more likely to exit from work.
Fifth, we treated work environment variables as continuous variables assuming a linear association. In a sensitivity analysis we—as mentioned above—did only find few signs of non-linearity by treating working conditions as cubic terms.
Sixth, we assessed SEP through occupational social class. This approach overlooks important aspects of the complexity of SEP related to the household and to lifetime biographies. Unfortunately, the study did not entail such data.
Last, exposure to workplace factors was assessed only at baseline, but it could have changed during follow-up, possibly causing a non-differential misclassification of the exposure and an attenuation of the associated risk estimates.
Comparison with earlier studies
The comparability of our results with other studies on early exit as a global measure is limited by differences regarding work dimensions and data analyses (Boot et al. 2014; de Boer et al. 2018; Lund and Borg 1999; Robroek et al. 2013a). One study on Danish employees found that possibilities for development lowered risk of early exit in both genders, and—among women—also decision authority and medium level of social support. A lack of control for exposure to physical or psychosocial demands might overestimate the role of psychosocial resources (Lund and Borg 1999). Two Dutch studies also include physical demands, but their results were stratified by chronic disease status (Boot et al. 2014; de Boer et al. 2018). In the first one, the risk of early exit increased with physical demands and decreased with psychosocial resources, but only among subjects affected by chronic diseases, and increased with high psychosocial demands in the overall sample (Boot et al. 2014). In the other study, only time pressure increased the risk of early exit at 1-year follow-up, this was also the case with emotional demands at 2-year follow-up, in both cases only among workers with chronic diseases, whereas physical demands was not associated (de Boer et al. 2018). A European study based on SHARE—examining the association of early exit with physical and psychosocial demands, job control and rewards—found an increased risk of early exit for exposure to low job control and low rewards, but no association with physical demand and high time pressure (Robroek et al. 2013a); however, adjustment for health status may have led to an underestimation of the effect of work factors in this study, due to the possible mediating role of health.
Also, the finding of a positive association between physical demands and early exit in our study appears consistent with the results of a recent Danish study, which did not consider a single global exit route but four routes separately, and found significant associations between exposure to high physical demands and exit from work through disability pensions, early retirement, and LTSA, whereas the increase in risk was only marginally significant for unemployment (Sundstrup et al. 2018a).
Among studies investigating only one or two exit routes from paid employment, exposure to physical factors at work has been quite consistently associated with an increased risk of exit through disability retirement (Albertsen et al. 2007; Bödeker et al. 2008; Emberland et al. 2017; Karpansalo et al. 2002; Krause et al. 1997; Krokstad et al. 2002; Labriola et al. 2009; Lahelma et al. 2012; Lund and Csonka 2003; Pohrt and Hasselhorn 2015; Stattin and Jarvholm 2005) and early retirement (Friis et al. 2007; Lund et al. 2001, 2005).
Exposure to physical workload has also been found to increase the risk of unemployment (Borg and Burr 1997; Lund et al. 2001; Robroek et al. 2013a), although these studies mainly examined shorter periods of unemployment, which could not be considered a definitive exit from paid employment.
Physical demand has been found associated also with LTSA in several studies (Andersen et al. 2016; Burdorf and Jansen 2006; Christensen et al. 2007; Lund et al. 2006; Lund and Labriola 2006; Melchior et al. 2005; Sterud 2014) with higher risks generally found among blue-collar workers. However, for this outcome the comparability with our results is expected to be limited, because the LTSA definition used in our study was of much longer duration than that employed in the referenced studies, which mainly adopted a cut-off of few weeks. In these studies, physical risk factors for early exit included mainly physical demand or similar indicators of physical workload, with only few reporting associations with specific exposures, such as repetitive movements (Labriola et al. 2009), bending of the back or neck (Albertsen et al. 2007; Lund et al. 2001), or working in awkward postures (Albertsen et al. 2007; Karpansalo et al. 2002; Krause et al. 1997; Labriola et al. 2009; Lund and Csonka 2003).
Psychosocial factors at work have also been found associated in several studies with an increased risk of disability pensions, although a recent review on 39 studies concluded that there is only moderate evidence of an association for low job control and job strain, and limited evidence for other psychosocial dimensions, such as job demands, effort-reward imbalance, low social support and repetitive work (Knardahl et al. 2017).
Non-disability retirement also appears to increase mainly by low job control or its sub-dimensions (Blekesaune and Solem 2016; de Wind and van der Beek 2014; Lund and Villadsen 2005; Robroek et al. 2013a; Thorsen et al. 2016), but other factors, such as low role clarity, low reward, low organizational justice and low leadership or management quality, have been reported among risk factors (Breinegaard et al. 2017; Thorsen et al. 2016). A recent review on the relationship between exposure to psychosocial factors at work and early retirement concluded that there is sufficient evidence that high job control and high social support are associated with later retirement, but not for job demands, organizational justice, effort-reward imbalance or other psychosocial work factors (Browne et al. 2018).
Unemployment was consistently associated with low control and its sub-dimensions in the few available studies (Lund et al. 2001; Lund and Labriola 2006; Robroek et al. 2013a), whereas various psychosocial factors have been found to increase the risk of LTSA, but results appear inconsistent among studies; nonetheless, low control or its sub-dimensions have been repeatedly associated with LTSA (Henderson et al. 2012; Lund et al. 2005; Melchior et al. 2005). Also, one study reported an increased risk of LTSA for exposure to high strain (Wang et al. 2004), while another one also for exposure to conflict, rewards, quality of leadership, emotional demands and demands for hiding emotions (Lund et al. 2005).
It is worth underlining that most studies in the literature did not adjust the results for exposure to other work factors, which may explain the higher number of work factors found significantly associated with early exit in these studies, as well as the stronger risk estimates reported for most work exposures.