Labour market trajectories following sickness absence due to self-reported all cause morbidity—a longitudinal study

From September 2012 to March 2014, all new cases of sickness absence exceeding 4 weeks (n = 4541) in the Western part of Denmark were registered. They received a questionnaire about social characteristics, RTW expectations and reasons for sickness absence. The questionnaire was originally used for an RCT study evaluating the effect of psychoeducation on RTW in individuals on sick leave [24]. No difference in relative risk of RTW during the first 6 and 12 months after inclusion was found between the intervention group and the control group [25].

The questionnaire was completed by 2788 individuals (61.4 %). Those who did not provide data on reason for sickness absence (n = 20), information on education (n = 31), employment (n = 123) and RTW expectations (n = 126) were excluded. All participants were linked to The Danish National Labour Market Authority’s DREAM database [26], which provided information about economic compensation for unemployment, sickness absence, and other kinds of social transfer income. The type of transfer payment in DREAM is recorded for each week if the person has received the benefit for one day or more. Termination of registration occurs following the first full week of not receiving any type of transfer payment. If no transfer payment is registered for a specific week, the person is considered to be self-supporting and consequently as working. In Denmark, a citizen in the workforce (employed as well as unemployed) is entitled to sickness absence compensation (at the time of this study after 4 weeks), and if the employee receives normal salary during the sick leave period, the employer receives municipal reimbursement. Data from the DREAM database is increasingly applied in research and has been validated in research in individuals on sick leave [26‐28].

A total of 452 participants were not registered as being on sick leave in the DREAM database when the questionnaire was distributed and consequently they were excluded from the study. It was done to avoid misclassification and that a difference in social benefits in the study could be attributed to a difference in social benefits at baseline. Thus, the final study population consisted of 2036 individuals between 18 and 64 years of age (mean: 44.5, sd: 11.1). Data on registrations in the DREAM database was obtained from the week the questionnaire was sent and 51 weeks onwards.

The outcome variable in this paper was employment status during the 51 weeks following the questionnaire and was recorded weekly.

In the time-to-event analysis, the outcome was return-to-work, which was defined as the period (in weeks) between inclusion and the first period of 4 consecutive weeks without receiving any social benefits.

In the sequence analysis, the outcome was extended to include five different categories for labour market participation and RTW: 1) sickness absence, 2) working 3) unemployment, 4) temporary support (other than unemployment and sickness benefits), and 5) permanent support. Working was defined as the weeks with no benefits, and unemployment was defined as receiving unemployment benefits. Temporary support was defined as social benefits that are given temporarily aiming at promoting subsequent employment, e.g. public education grant, social assistance or rehabilitation benefit. Permanent support was defined as social benefits that are given on a permanent basis, where regular employment is no longer possible e.g. early retirement, public retirement pension and supported job (the Danish labour market arrangement for people with reduced ability to work and wage is partly compensated).

Self-reported reason for sickness absence was the main exposure. The participants stated in the questionnaire what they considered to be the reasons for their absence. They could report several reasons, but if they had reported anxiety, depression, other mental illness or stress and burnout, they were categorised as having “mental health reasons”, while the rest of the individuals were categorised as having “other health reasons” (e.g. musculoskeletal disorders, cancer, or chronic pain (Table 1).

Information about education, employment, age, gender and RTW expectations was retrieved from the questionnaire. RTW expectations were estimated by the participants as the probability of not being on sick leave after 6 months (as a percentage in whole tens from 0 to 100 %). The covariates were categorized as seen in Table 2.

Participation was voluntary, and the study has been registered and approved by the Danish Data Protection Agency (http://​www.​datatilsynet.​dk). The participants did not provide consent, as the data were analysed anonymously.

Initially, a comparison of the individuals from the two exposure groups was made in relation to age, gender, education, employment and RTW expectations by means of Chi2 or t-tests.

Secondly, the pseudo value-regression approach was used to examine differences in RTW between the two exposure groups by calculating relative risk (RR) and cumulative incidence proportions (CIP) at the end of the 51 weeks of follow-up [29, 30]. CIP showed the percentages of individuals in each group who had returned to work. The allocation of the RCT study was adjusted for in all steps of the analysis [25] and thereafter, different adjustment strategies were carried out based on variables that were chosen a priori; 1) adjustment for gender and age, 2) plus education and employment and 3) plus RTW expectations. Death, emigration and receiving permanent support were considered as competing risk.

Furthermore, sequence analysis was performed, which is a statistical study of successions of states or events. A sequence is defined as an ordered list of elements (e.g. labour market status) and episodes (identical successive elements) expressed on a time axis [31, 32]. In this study, sequences showed a complete event history of labour market participation in each particular week from baseline to follow-up. The relative proportion of each of the five employment status for every week was displayed in a status proportion plot [33]. In the sequence analysis, further 18 participants were excluded due to death or emigration (four from mental health reasons and 14 from other health reasons). Thus, in those analyses, the study population consisted of 2,018 participants.

In the sequence analysis, the mean duration in weeks within a given state and the mean number of episodes of different status for the exposure groups were calculated. Differences between exposure groups were performed by using the syntax ttesti in STATA by adding the n, mean and sd for each group. This syntax was used as sequence analysis was made in long format and thus regular tests were not possible to perform.

The distributions of the sequences were compared in the two exposure groups. All individuals were divided into four groups according to their sequences; 1) only sick leave, 2) moving to continuous work, 3) having at least one episode of work, and 4) sick leave and social benefits. The different distributions of sequences were tested in a chi2 test.

A volatility indicator was defined as the proportion of work and unemployment episodes in relation to total episodes. Episodes within work and unemployment reflected a positive status of RTW or readiness to RTW. The volatility indicator indicated that the higher the value of this indicator (range 0–1), the higher the quality of the transitions [34].

An integration indicator was measured as an indicator of how quickly and to what extent the individuals re-entered employment. It was assessed as the sum of number of sequence positions where status was work, which were weighted by their position within the sequence. This indicated that the longer or more episodes in work, the higher the quality of the integration process (range 0–1) [34].

Moreover, the sequences were grouped based on optimal matching algorithms and statistical cluster analysis to find and categorize observed sequences into a smaller number of clusters [31, 35]. Optimal matching was used to measure dissimilarities between sequences by applying the Levenshtein distance measure, which measured the number of operations that were needed to transform one sequence into another [31]. Similar sequences were grouped together using hierarchical cluster analysis with Ward’s linkage [34, 35]. On the basis of these results, similar sequences were merged into eight clusters, which were named based on employment status. Afterwards, the distribution of the clusters across the exposure groups was tested by means of logistic regression. The same adjustment strategies were used as in the pseudo value analysis.

Point estimates were presented with 95 % confidence intervals. STATA/IC 11.2 (StataCorp LC, College Station, TX) was used for all statistical analyses with the SQ-ADOS to perform the sequence analyses.

Individuals on sick leave due to mental health reasons spent more weeks on sickness absence and in temporary support and less weeks on work compared to individuals with other health reasons for sick leave. Moreover, fewer of the individuals on sick leave due to mental health reasons had returned to work during the 51 weeks of follow-up, compared to the individuals with other health reasons.

Also the chance of having returned to work was lower for individuals with mental health reasons when adjusting for gender, age, education and employment status but after adjusting for RTW expectations, the chance was the same in the two groups. Moreover, individuals with mental health reasons had higher odds of being in the “sickness absence” cluster and a lower odds of being in the “fast RTW” cluster, but the difference was attenuated after adjusting for RTW expectations.

The results show that RTW expectations can be considered a confounder in the effect of health reasons for RTW. Individuals with mental health reasons returned to work later than individuals with other health reasons, but after adjusting for RTW expectations both exposure groups were found to return to work at the same time. Other studies have also found RTW expectations to be a predictor of RTW in both individuals on sick leave due to mental and physical disorders i.e. a positive RTW expectation predict a shorter time to RTW [8‐13]. It has been speculated that positive RTW expectations represent the self-efficacy of the employee, i.e. the belief an individual has in his/her own capacity to perform a specific behaviour successfully, in this case in relation to RTW [8, 10]. Furthermore, bad mental health and low RTW expectations could be influenced by the same problems, i.e. problems meeting demands at work or at home, social problems at work or other work-related factors may have triggered both mental health problems and low RTW-expectations if the prospects of solving these problems seem low.

Individuals with other health reasons had a higher level of RTW expectations than individuals with mental health reasons. This has also been confirmed in a study by Huijs et al. [8]. Another possible explanation could be that the stigmatization of mental health problems in the workplace is high, and therefore the employees might avoid their workplace and receive less support from their colleagues and supervisor, making it seem less likely to return to work. A third explanation of the lower RTW expectations among individuals with mental health reasons could be influenced by their psychological symptoms like hopelessness, discourage and reduced self-confidence. These symptoms likely reduce the belief of RTW.

The maximum number of episodes for one individual was 23 in the group of mental health reasons and 26 in the group of other health reasons. This shows that the RTW process for individuals on sickness absence benefits may be long and complex [15, 36], which is in line with previous Nordic studies using multi-state models [20‐23]. It also emphasises the need to analyse RTW as a process [14, 15], and not only at a single point in time [37, 38]. The advantage of this approach is that it provides a more complete picture of RTW and employment trajectories and therefore, a more complete understanding of the impact of disability on the employee’s life and well-being [36, 39].

During the last 10 years, transitions of states have been used in the research of sickness absence by means of multi-state models [20‐23]. Pedersen et al. showed the transitions for Danish individuals on sick leave and with 4 year follow-up [22]. They included the states; work, unemployment, sickness absence, and disability pension, and identified predictors for each of the different transitions. Three Norwegian studies have used multi-state models to analyse the transitions of states [20, 21, 23]. Lie et al. applied three different states that low back pain patients could be in after an intervention; recovery (RTW), sick leave benefits, or disability pension [20], while Gran et al. also included partial sick leave and work assessment allowance [23]. Oyeflaten et al. extended the model to include eight different categories for social benefits or return to work over a 4 year period [21]. Only Oyeflaten et al. included categories on varying types of social benefits, whereas Pedersen et al., Lie et al. and Gran et al. mostly looked at disability benefits besides work and sickness absence. To be able to show a more realistic picture of the transitions, it is relevant to include all types of social benefits.

The prospective design of the study and the record linkage of the cohort data with sickness absence data from DREAM added to the strengths of this study. The study had complete follow-up of weekly employment status due to full coverage of registers of social benefits and the information is considered valid [26]. Moreover, this study included sequence analysis to look at transitions besides the more traditional time-to-event outcome. Using the method has given an overview of the life course after the start of the sickness absence period. Sequence analysis is considered an exploratory method rather than a method for hypothesis testing, which means that sequence analysis cannot answer the question of causality. Due to this, sequence analysis is best used in combination with other methods, and cannot replace methods like event history models [40].

There is no clear agreement about how long a follow-up period is needed to get the best measurement of the effect on work and benefits after sick leave [16, 41]. Previous studies using process analyses have used a longer follow-up period, i.e. 3–4 years [20‐23]. In this study, only 51 weeks of follow-up was applied which reduces the complexity of the sequences as e.g. 20 % with mental health reasons and 13 % with other health reasons were still on sick leave and thus, had not changed states. Thus, a longer follow-up period would have been preferable, as Oyeflaten et al. concluded that several years are needed to get an adequate picture of the RTW outcome [21].

The frequency of mental disorders in RTW research has been found to be underestimated [4, 42, 43]. Therefore, the grouping of exposure may cause misclassification if the individuals are not true about reporting the sickness absence reason. However, as the questionnaire was sent in relation to an RCT study for individuals with mental health problems, it is considered a minor issue. Moreover, misclassification in relation to the outcome may occur as a new sickness absence period is registered only if it is longer than 4 weeks. Thus, the short term sickness absence periods may be underestimated, which at the same time may overestimate the participation in work. Also, DREAM provides no information on whether an individual is actually working or not. When data from DREAM are used in research studies, work is categorised as those weeks which the individual does not receive any benefits. However, DREAM data have been validated in the context of sick leave [26, 28].

Some studies have divided the sickness absence reasons into mental, physical and co-morbidity and found that co-morbidity was associated with longer time until RTW than only reporting physical or mental problems [8, 44]. In this study, individuals with co-morbidity were not categorized separately, as it was not the aim of the study. Moreover, only co-morbidity that was due to the sickness absence was reported. Therefore, the degree of co-morbidity in this study is unknown.

Another limitation derives from the relatively low response rate (61.4 %). The relationship between sickness absence reasons and employment status may have been different in non-responders, and thus could change the estimates. If individuals on sick leave due to mental health reasons to a greater extent did not response to the questionnaire and at the same time were on sick leave for a longer time period, we may have underestimated the association between sickness absence for mental health reasons and RTW. Moreover, some of the clusters in Fig. 2 included few observations which will give rise to a large random variation.

Comparison between studies may be difficult due to the large variation between countries in the regulation of sick leave compensation and social benefits. Within the Nordic countries, the social security systems are relatively similar and make comparisons feasible [45]. Our findings may, therefore, be generalized to the Nordic countries. However, we see no reason why the longer sickness absence periods and lower RTW expectations for those with mental health problems than for those with other health problems should not be similar in other Western countries.