Sick leave is considered a measure of overall health (1) and is a predictor of mortality (2) among the working population. A persistent finding is that people with lower adult social position have higher levels of sick leave (3), and several longitudinal studies have suggested that working conditions could in part explain this link (4–10). Aspects of the working environment that have been identified as potentially harmful include psychosocial factors (4, 5, 7, 9, 10) and physical workload (5–10), with the latter appearing to be the most important factor explaining the social gradient (7, 9, 10). Despite the consistent pattern of these findings, the pathway linking adult social position to sick leave through physical working conditions has not been clearly elucidated from a causal standpoint (11, 12).

Figure 1 presents the underlying causal diagram representing the relation between adult social position, physical workload, and sick leave. A causal pathway from adult social position to sick leave, via work, is compatible with the aforementioned findings. However, according to the indirect selection hypothesis (13), selection into adult social position is not random; the same early-life-course factors that are driving people into lower adult social positions could also be driving them into jobs with higher physical workload, placing them at increased risk for sick leave. The link between adult social position and sick leave, and the mediating pathway of physical working conditions, could therefore in part be confounded by the contribution of early-life-course factors, and studies that do not control for these will be biased (11, 14). With the exception of the study by North et al (4), who considered the father’s social class when examining the mediating effects of psychosocial work characteristics, none of the previous studies considered the role of early-life-course factors in the analyses.

Figure 1

A causal diagram showing that the direct pathway between adult social position and sick leave, and the indirect pathway via working environment, is confounded by early life course factors.

We consider two early-life-course factors: childhood and adolescent social position and the personality trait “neuroticism”. The rationale for considering these factors is drawn from two bodies of literature. The first is evidence from epidemiology showing that childhood and adolescent social position directly influences educational attainment (15) while also directly and independently influencing health outcomes, including morbidity (16) and work disability (17, 18). Childhood and adolescent social position could therefore be driving individuals into both lower adult social position and poor health, thereby confounding the relation between adult social position and sick leave. In addition, childhood and adolescent social position has an effect on occupational choice (19), and consequently the level of physical workload associated with a given occupation. Childhood and adolescent social position could thus also confound the indirect pathway from adult social position to sick leave, both by confounding the pathway between adult social position and physical workload, and between physical workload and sick leave.

The second body of evidence is based on findings rooted in personality psychology. Neuroticism—a personality trait characterized by negative emotions in response to threat, frustration, or loss (20)—has been an area of wide interest in the work environment literature (21). Individuals high on neuroticism tend to report more adverse working conditions (21), including higher physical workload (22), and there has been a concern that this reporting pattern is the result of a bias, as opposed to an objective categorization of the work environment (21). At the same time, neuroticism has a well-established link with overall health (23) and sick leave (24), as such, the pathway from physical workload to sick leave could be confounded by neuroticism, as shown in figure 1. Personality has also received some attention as a possible factor driving selection both into lower educational attainment and poor health outcomes (25) and mortality (26). Negative emotionality in particular, which is closely related to neuroticism, has in many studies been shown to be associated with both socioeconomic status and health (27). Given the aforementioned association neuroticism has with health and sick leave, the direct pathway between adult social position and sick leave could therefore be confounded by the contribution of neuroticism through its relation with adult social position, and sick leave, as shown in figure 1.

In the present study, we investigate how the effect of adult social position on sick leave is mediated through physical workload, while controlling for childhood and adolescent social position and neuroticism. Direct effects of adult social position and indirect effects through physical workload is estimated within a formal framework for causal inference suggested by Lange & Hansen (8), based on the additive hazards survival model (28).

Methods

Study population

The source population for this study is a national birth cohort initiated with the purpose of studying the impact of early conditions and events on health and employment outcomes later in life. The birth cohort includes all 626 928 live-born individuals (304 953 females and 321 975 males) registered in the Medical Birth Registry of Norway (29) between 1967 and 1976. Parental relationships have been obtained from the Medical Birth Registry, enabling us to identify the mother with certainty in all but six cases, and the father in all but 7.1% of cases. Demographic, educational, employment, and sick leave data has been collected from national registers for both the individuals and their parents. For the current study, we limited the analysis to a subset of individuals from this birth cohort who also participated in the Nord-Trøndelag Health Study, 2006–2008 (HUNT3) (30). For HUNT3, all individuals residing in Nord-Trøndelag county aged 20 or older were invited to participate. Information on personality, life style factors, and work factors were collected by interview and two questionnaires. The first questionnaire was sent with the invitation to participate and was returned at the time of the interview. The second questionnaire was distributed at the interview and was completed at home and returned by mail. Overall, 93 860 individuals were invited to HUNT3, with 50 807 (52%) choosing to participate, and about 80% of these individuals returning the take-home questionnaire (30). A study of non-participants in HUNT3 suggests that individuals who had a lower social position, poorer health, and who received disability pension were less likely to participate (31).

Demographic data obtained from the database FD-Trygd (32) indicated that as of 2006, 15 053 (2.3%) individuals from the birth cohort that had not died or emigrated, were residing in Nord-Trøndelag, and were thus eligible to be recruited to the HUNT3 study; of these, 6913 (46%) individuals were linked to participants in the HUNT3 study. We further restricted participation to only include individuals who were employed at the time of the interview (excluding self-employed individuals), and who had valid responses in HUNT3 on occupation, physical workload (assessed both at the interview and in the take-home questionnaire), and neuroticism. In total, 5416 individuals, 3222 women and 2194 men reported being employed. Due to long lines at the interview site, 631 participants (12%) were randomly selected to be given a short version of the interview that did not include an item on physical workload; therefore, these individuals were also excluded. Of those completing the full interview, 3328 individuals, 2099 women and 1229 men, had completed the physical workload questions and the neuroticism scale in the take-home questionnaire and had an identifiable occupational group. This comprised our final study population. The selection process is shown in figure 2.

Figure 2

Flow chart showing selection into the study, starting with all individuals in a Norwegian birth cohort (born 1967–1976). Individuals who also participated in the Nord-Trøndelag Health Study 2006–2008 (HUNT3) were matched to this cohort, and provided the study base. Study participation was further restricted to those employed at the time of HUNT3 participation, and who had valid responses in HUNT3 on physical workload and neuroticism

Results

The study population included 2099 women and 1229 men, with a mean age of 36 years (range 30–41) at the time of study inclusion. Baseline characteristics according to adult social position, physical workload, and sick leave are shown in table 1. Both for women and men there was an overrepresentation in social group II (86% for women, and 68% for men). Lower adult social position was associated with higher levels of physical workload, lower childhood and adolescent social position (parental education level, parental income and parental disability benefits), and higher levels of neuroticism. For neuroticism, this gradient was especially pronounced for women.

Table 1

Descriptive characteristics by adult social position, physical workload, and sick leave: HUNT3 participants, 2006–2008.

Mean physical workload increased from 2.8 [standard deviation (SD) 2.2] in social group I to 5.8 (SD 2.4) in group III for women, and from 2.2 (SD 1.6) to 6.6 (SD 2.3) for men. Individuals from a lower childhood and adolescent social position and with high neuroticism had higher levels of physical workload (table 1).

During the follow-up from date of participation in HUNT3 (2006–2008) through 2009, women and men contributed a total of 3725 and 2431 years of follow-up time, respectively. The mean number of follow-up years was 1.8 (SD 0.9) for women, and 2.0 (SD 0.8) for men. The number of women who had at least one sick leave spell was 992 (47%) compared to 359 men (29%). The incidence rate of a first-sick-leave-spell per 100 000 person-years at risk was 72 for women and 40 for men. Individuals with a lower adult social position had a higher incidence rate, increasing from 43 in group (I) to 109 in group (III) for women, and in corresponding groups, from 24 to 70 for men. High physical workload, lower childhood and adolescent social position, and high neuroticism were also associated with a higher incidence rate (table 1).

Table 2 shows the estimates obtained from the linear regression analysis, where we regress physical workload on adult social position. In a model controlling for childhood and adolescent social position, neuroticism, age, and marital status, an increase in physical workload was observed with decreasing adult social position. Women in group III had, on average, a physical workload score 2.45 units higher than the women in group I. The corresponding number for men was 4.06 units. For women, the inclusion of childhood and adolescent social position and neuroticism reduced the estimates, compared to a model not adjusting for these variables, by 16% (2.45 versus 2.91) for group III and by 11% (1.47 versus 1.64) for group II. For men, the estimates were reduced by around 5% in both groups II and III. By including childhood and adolescent social position and neuroticism in the models separately (not shown), it became apparent that for women, the reduction was due to both these factors, with the former weighing slightly more heavily than the latter. For men, the slight reduction was attribu table only to childhood and adolescent social position.

Table 2

Parameter estimates with 95% confidence intervals (95% CI) from the linear regression analysis ^a of physical workload on adult social position: HUNT3 participants, 2006–2008

The direct effect of adult social position and physical workload on sick leave rate is derived from the additive hazards survival model and is presented in table 3. A change in social position from group I to group III for women, resulted in 40.3 (95% CI 12.7–67.9) additional spells per 100 000 person-days at risk. The corresponding rate increase for men was 29.4 (95% CI 8.6–50.2). For women, including childhood and adolescent social position and neuroticism in the model accounted for a 16% reduction (40.3 versus 47.9) of the estimates in group III and 9% reduction (17.6 versus 19.4) in group II. The reduction in the estimates was primarily due to neuroticism, though including childhood and adolescent social position reduced the estimates further. For men, the change in estimates was negligible when both childhood and adolescent social position and neuroticism were included simultaneously. However, neuroticism did reduce the effect estimates when included separately, but the effect was mitigated when childhood and adolescent social position was also included.

Table 3

Parameter estimates with 95% confidence intervals (95% CI) from the Aalen’s additive hazard regression analysis of days to first sick leave spell: HUNT3, 2006–2008.

A 1-unit increase in physical workload resulted in a rate increase of 4.6 (95% CI 2.6–6.6) for women and 2.9 (95% CI 1.1–4.7) for men (table 3). When including childhood and adolescent social position and neuroticism, the physical workload estimates were reduced by around 17% for women (4.6 versus 5.4) and again the reduction was mostly attributable to the inclusion of neuroticism. For men the change was negligible.

The total (TE) and indirect effects (IE) are presented in table 4. For women, a hypothetical change in adult social position from highest to lowest social group would result in a total of 51.6 (95% CI 24.7–78.5) additional sick leave spells per 100 000 person-days at risk. For men, there would be 41.1 (95% CI 21.4–60.8) additional spells. For women, 24% (95% CI 10%–49%) of these extra spells could be attributed to the mediating pathway of physical workload, while for men the corresponding figure was 30% (95% CI 10%–63%). That is, an intervention that could have reduced the physical workload of group III so that it was at the same level as group I would have reduced the number of sick leave spells by 24% for women and 30% for men. The total effect of adult social position on sick leave was reduced when childhood and adolescent social position and neuroticism were included in the model; for women in group III, the number of extra spells was reduced by 20% (51.6 versus 63.7), and for those in group II, by 14% (24.3 versus 28.2). For men, the reduction in total effect was substantially lower: 1% and 5% in group II and III, respectively. For men and women, both childhood and adolescent social position and neuroticism accounted for the reduction in effect estimates, though with neuroticism accounting for the largest reduction. The indirect effect estimates were similarly reduced when childhood and adolescent social position and neuroticism were included in the model; 24% (6.7 versus 8.8) for women in the middle group and 29% (11.2 versus 15.7) for women in the lowest group. For men, the reduction was around 5% in both middle and lowest group. Finally, the estimates of proportion mediated were slightly reduced for women, by 6% (0.31 versus 0.33) and 8% (0.24 versus 0.26) in the middle and lowest social group, respectively, when childhood and adolescent social position and neuroticism were included in the model. For men, the proportion mediated was reduced by 4% for the middle social group, and remained unchanged in the lowest social group.

Table 4

Total effects (TE) of adult social position on sick leave and indirect effects (IE) through physical workload: HUNT3, 2006–2008. Note the direct effect can be found in table 3. [95% CI=95% confidence interval.]

Discussion

This study of a Norwegian working population indicates that an intervention that could reduce the physical workload of social group III to the level of group I has the potential of reducing the number of sick leave episodes attributable to adult social position by 24% among women and 30% among men. Similarly, improving the physical workload conditions of group II to the level of group I could reduce 31% of the number of sick leave episodes attributable to adult social position amongst women, and 48% amongst men.

Advantages of our study are that we were able to take into account childhood and adolescent social position and neuroticism in our analysis; that we had complete, register-based sick leave data; and that we used an analytic approach that enables a causal interpretation. However, several limitations to our study are apparent. First, the causal interpretation of our findings rests on the assumptions of the mediation analysis not being violated (8). The assumption of no confounding on the direct and indirect pathway is threatened by our lack of indicators of childhood and adolescent health, as early health is associated with adult health (42) and educational attainment in adulthood (15); as a result, the association between adult social position on sick leave, via physical workload, could still be somewhat overestimated. We did have data on overall health measured at the time of study inclusion in HUNT3, but, in our opinion, poor health measured at this time point could have been the result of ongoing exposure to high physical workload rather than a cause of it. Thus, by adjusting for this variable, we could be blocking a pathway from physical workload (experienced prior to HUNT3 participation) to sick leave. A clear advantage would have been to have repeated measurements of physical workload, employment status and health, ideally from the start of employment history, which would enable us to disentangle the complex relation between health and physical workload over time. Furthermore, the expression for indirect and direct effect developed in Lange & Hansen focuses on situations in which the exposure and mediator do not interact (8). We assessed for interaction between the exposure and mediator, though since our study sample was relatively small, we may not have had adequate power to detect a possible interaction.

Second, our measure of physical workload was based on just three self-report questions that to our knowledge have not been validated. Further, in order to accommodate the mediation analysis, which has so far only been defined for a continuous variable, we created a sum score based on these three items. However, the physical workload items were arguably on an ordinal scale. Additionally, the questions appeared to assess slightly different aspects of physical workload (eg, the degree of heavy physical workload as opposed to perceived exhaustion from said workload). As a result, the interpretation of what a 1-unit increase on the physical workload scale means in terms of real-life physical work exposure remains unclear. Additionally, objective measurements of work exposure, such as direct measurements of mechanical exposure, would be more optimal. However, a review study suggests that self-report measures tend to underestimate the effect mediated through work, compared to more objective measures (12), which suggests that, if anything, the mediating role of physical workload would be somewhat underestimated in our study.

Third, our study assumes that personality traits measured in adulthood are representative of personality in childhood and adolescence, which is consistent with the tendency to view personality as relatively fixed (20). Recent advances have challenged this perspective by demonstrating personality changes across the lifespan (43). Moreover, these personality changes have been linked to environmental influences (44), including adverse experiences at the workplace (45). As such, personality could in part be a result of adverse conditions throughout the life course, rather than the ultimate cause of them. If this is the case, then including neuroticism might tend to underestimate the role of adult social position and physical workload on sick leave. Another challenging aspect regarding neuroticism is that it is strongly linked to depression (46). If depression is in part a result of high physical workload, then by controlling for neuroticism, we could in part be blocking a causal pathway from physical workload to sick leave which would lead to an underestimation of the true effect of physical workload.

Last, there were a fairly high number of non-participants in the HUNT3 study, and 30% of those who participated in the first part of the study and completed the full interview either did not return or had incomplete responses on the second questionnaire (as shown in figure 2). A recent study showed that individuals with low social position, poor health, and higher rates of disability were less likely to participate in HUNT3 (31). Participation in our study is thus in part determined by adult childhood position and also most likely sick leave, suggesting that the observed association between these variables could in part be induced by selective study participation. Furthermore, our study participants were restricted to people living in the county of Nord-Trøndelag, so our findings may not be generalizable to other populations. On a more general level, sick leave rates vary within geographical region, according to economic changes and according to the welfare system (39).

Despite these limitations, our study provides support that physical workload is an important mediator between adult social position and sick leave. Our study replicates findings from other research showing that the social gradient in sick leave is reduced when factoring in the role of the physical exposures at work (6–10). Direct comparison with these studies is difficult, due to the different analytic approaches in measuring mediation. The paper by Lange & Hansen (8) is the only other study we are aware of that estimates the mediating effect of physical work environment in a survival context with a causal interpretation. In their study, they found that the proportion mediated via physical work environment ranged from 24–48% for men and 26–71% for women, which is slightly higher than what we found. However, direct comparison is difficult due to the use of different measures of adult social position and physical work environment. Another aspect of our study that makes direct comparison with other studies difficult is that most of them included other aspects of the working environment and a host of lifestyle factors, including smoking, alcohol use, and physical activity, in their model. The relation between working conditions and other life-course factors could quite likely be the result of a complex causal chain, whereby working conditions may be prohibitive of a healthy lifestyle, and reduced health resulting from a poor lifestyle may influence the ability to handle physical workload. Therefore, while several life-course factors were assessed in the HUNT3 study, we chose not to control for them because they could in part be a result of coping with high physical workload, rather than a pathway operating directly from adult social position to sick leave.

Our study makes a unique contribution in that we were able to include childhood and adolescent social position and neuroticism in assessing the mediating role of physical workload on sick leave. The total number of extra sick leave spells attributable to adult social position became lower when these factors were included, and the reduction was more prominent for women (24–29%) than men (5%). For women, the reduction in number of extra spells was accounted for by a reduction both in number of extra spells attributable to the indirect pathway through physical workload and the direct pathway from adult social position to sick leave. For men, there was only a reduction in the number of extra spells attributable to the indirect pathway through physical workload.

The finding that the total number of extra spells in group III compared to group I are reduced when factoring in childhood and adolescent social position and neuroticism is compatible with a life-course perspective, whereby adult social position and health is influenced by earlier factors in life. This interpretation is in line with several studies showing that social position early in life is important for health (47) and sick leave (48), even while accounting for adult social position. Fewer studies have investigated the role of personality as an early life-course factor accounting for the attainment of adult social position and health outcomes, but there is evidence from a study by Hampson et al that personality traits can in part explain the social gradient in self-reported health through its influence on both education and health-promoting behaviors (25). Our study suggests that childhood and adolescent social position and neuroticism are important confounders between adult social position and sick leave, and by taking into account these early factors, we get a more conservative estimate of the rate increase attributable to physical workload. Future research would therefore benefit from considering these variables when studying adult social position and sick leave, in particular there is still a need to elucidate to what extent early-life conditions could be influencing the development of personality over time and vice versa.

The implications of these findings should also be considered. Sick leave is a large problem both in terms of its cost to society and employers, but also the individual. Unlike adult social position, which would be very difficult to modify, it is possible to intervene at workplaces in order to reduce exposure to hazardous physical working conditions. Such interventions could be based on requirements and recommendations set up in national regulations and guidelines. While our study provides an estimate of the maximum number of sick leave episodes that could be removed under an intervention in which the physical workload of group II and III were reduced to that of group I – arguably a target that is difficult to meet in real life – this finding is nonetheless informative in that it quantifies how much of the social gradient in adult social position could possibly be removed by such an intervention. The pathway through physical workload represents one of many possible targets for an intervention, and the finding that a substantial gradient persists after taking into account the role of physical workload suggest that other causal pathways, such as through psychosocial work factors and the range of lifestyle factors, are also linking adult social position and sick leave. These pathways could be separate and distinct, but might also be interacting with one another in complex ways over time. Recent developments in mediation analysis rooted in a causal framework have made advances that enables the analysis of multiple mediating pathways and interactions (49). Future studies would benefit from applying these methods to datasets that include repeated measurements of life-course factors over time in order to disentangle the complex relation between social position, work exposures, and sick leave.

In summary, the present study quantifies the mediating effect of physical workload in explaining the social gradient in sick leave while taking into account the contribution from personality and social position earlier in life. It also shows that neuroticism and early life-course factors contribute to the total effect of social position on sick leave.