Abstract
Background Sick leave rates in Norway and in the Western world in general are quite high, and despite the identification of a wide array of risk factors, much of the individual variation in sick leave remains unexplained. Health, education, occupation, individual abilities and traits, are all in part determined by conditions during the early lifecourse, and are also strong predictors of sick leave. However, few studies take a lifecourse approach to studying sick leave, instead often focusing only on more contemporary risk factors. One line of research has studied the importance of physical fitness on sick leave risk, though focusing primarily on adult fitness levels; poor fitness in adolescence could arguably have more far-reaching consequences, through its impact on educational and occupational attainment. A second line of research has focused on identifying pathways linking adult social position to sick leave, though these studies have rarely taken into account the individual and social conditions that precede adult conditions. Thus, it is difficult to say whether the observed findings are causal because early lifecourse factors could be confounding the association, and it also obscures the lifelong processes that lead to adult sick leave risk. Another line of work has focused on the role that social interaction may play in accounting for the patterns of clustering of sick leave risk, with some suggesting that this mechanism could account for observed sex differences in sick leave behavior. This research has mostly focused on the influence of neighborhoods and workplaces, but this norm-based mechanism could also be operating within families, with parents influencing offspring, and siblings influencing one another.
Aim In this thesis, I sought to first identify pathways linking exposures to sick leave risk through occupational and educational trajectories while taking into account contributions of early lifecourse factors. In Paper I, we aimed to assess the impact of adolescent aerobic fitness on sick leave in adulthood, through the mediating variables education and work-related factors (industry and enterprise-mean sick leave level). We restricted the analysis to musculoskeletal diagnoses. In Paper II, the main exposure is adult social position, and the mediating variable is physical workload. Here we focused on taking into account early personality development and childhood and adolescent social conditions. Second, we sought to investigate possible familial social interaction, focusing in Paper III on intergenerational transmission of sick leave, and sex-specific differences, and on sibling transmission from an older to a younger sibling in Paper IV. To account for the reverse causality problems and correlated effects in Paper IV, we used a dynamic Cox survival model that allowed for the incorporation of frailty and lagged time-dependent exposures.
Study population We used four different study samples of employed individuals for the four papers based on certain selection criteria. All participants were identified from a national birth cohort study comprising all individuals born alive in Norway from 1967 through 1976. In Paper I, the sample was restricted to 227,201 males with available aerobic fitness information data from the National Conscription Database. In Paper II, the sample was restricted to 3,328 individuals from the birth cohort who had also participated in the Nord-Trøndelag Health Study (HUNT3, 2006–2008). In Paper III, we restricted the sample to 78,887 individuals, born 1974–1976, whose parents had been employed when the participants were 18. In Paper IV, our sample was 19,634 participants with one older sibling, where both siblings had been employed for at least four months.
Statistical methods Data on exposures, outcomes and covariates were obtained from national registries, including the event-history database FD-Trygd (with data from 1992 through 2009) and from HUNT3. Our data thus spanned several decades, with some containing information on daily eventhistories. The natural methodological choice for such a data structure is survival analysis, which was used for Papers I, II and VI. Due to the exposures of interest in the four papers having been being measured at different time-points, the respective follow-up periods also differed, ranging from 1 to 15 years. In Paper I, we used a traditional epidemiological approach to mediation analysis by fitting a Cox regression model with and without the mediators, while controlling for possible confounders (parental education, intelligence, BMI, and musculoskeletal fitness). In Paper II, we applied a counterfactual approach to mediation, based on an Aalen’s additive hazards model. The model controlled for possible confounders (childhood and adolescent social position and neuroticism). In this analysis, we sought to identify the proportion of sick leave cases that could be reduced among individuals in the lowest adult social positions, if one could intervene to improve their physical workload so that it was the same as the physical workload of those with the highest adult social position. In Paper III, we estimated the additive risk difference in those exposed to parental sick leave at age 18, compared to those not exposed, using binomial regression, in a model controlling for several early lifecourse confounders. To elucidate whether the association was likely attributable to social interaction, we considered the strength of the parental-offspring associations in different constellations of parent-offspring sex and diagnostic categories. In Paper IV, we fit a dynamic Cox regression model for recurrent events, to study whether sick leave hazard increased following exposure to an older sibling’s sick leave episode. The model included both sibling sick leave and past sick leave history as time-dependent covariates. The latter was included in order to try to incorporate propensity to sick leave into the model. The model also controlled for a wide array of early life course and other time-dependent confounders.
Results Aerobic fitness at age 18 was moderately associated with sick leave hazard 5 to 15 years later. Poor and medium fitness increased the rate of non-injury sick leave, but decreased the rate of injury sick leave. While the association between aerobic fitness and sick leave appeared to be mediated through education and work-related factors, the indirect pathway had a negative sign for non-injury sick leave, but a positive sign for injury-sick leave. In Paper II we found that if we could perform an intervention that would change the physical workload of the lowest social group to that of the highest, we could reduce 24% of the extra sick leave episodes due to the social gradient for women, and 30% for men. To our knowledge, this study is the first to show that the link between social position and sick leave through physical workload was confounded by neuroticism and childhood and adolescent social position, especially for women. In Paper III, we found evidence of a parent-offspring association of sick leave. Parental sick leave in adolescence was associated with offspring sick leave 15 years later, though we found no evidence of a stronger additive association for women than for men. The sick leave risk was generally stronger for exposure to sick leave in same-sex parent, and sick leave in same-diagnostic category. In Paper IV, we found that exposure to sibling sick leave was followed by an increased sick leave rate. The hazard also increased with increasing levels of exposure in a dose-response manner. Controlling for a dynamic covariate of past sick leave history only reduced the estimates slightly. However, we found that the pattern was substantially weakened when we re-analyzed the data from time-of-first sick leave episode to time-to-recurrent episodes, suggesting that most of the association could be due to an unobserved propensity to sick leave.
Conclusion In conclusion, a greater understanding of the ways in which educational and occupational pathways interact with individual factors across the lifecourse is required. Adolescent aerobic fitness level was a risk factor for sick leave, and while this link appeared to be mediated through educational and work-related factors, and possibly also moderated, future studies should clarify this relation using causal mediation approaches that suited for survival outcomes and that can incorporate interactions between the exposure and mediator. This thesis further suggests that interventions aimed at reducing physical workload could reduce the social gradient in health. Additionally, we found that the link between adult social position and sick leave, and the mediating path through work conditions, may in part be due to personality factors and early life social conditions. This has implications for policy, but also for which variables future studies ought to consider in their analysis, since leaving them out will tend to show an inflated effect of adult social position on sick leave. In addition, a greater understanding of possible social influence within families is warranted. While we found that adolescent exposure to parent sick leave was associated with sick leave in adulthood, and sibling exposure was followed by an increased sick leave hazard, we could not rule out confounding. Studies that are able to adequately control for shared genetic vulnerability in family studies of sick leave, or studies that use a causal inference approach that evades this issue (i.e., natural experiments), are needed. Findings from the sibling study suggest that social interaction studies that do not take into account frailty may be biased. Further, it is still unaddressed which mechanisms are accounting for this possible social interaction—norms, information, or health behavior—which would also be important for intervention efforts.