Unemployed and disabled for work: identifying 3-year labour market pathways from the beginning of a sickness absence using sequence and cluster analyses in a register-based longitudinal study in Finland


Unemployment and work disability are associated in multiple ways. First, unemployment increases the risk for future disability pension (DP)1–5 and sickness absence (SA), especially based on mental disorders.6 7 In Finland, also the unemployed may receive sickness allowance as compensation for SA days if their weakened work ability prevents them from seeking work. While having a relatively low frequency of SA spells, the unemployed persons’ SA spells are longer than in other socioeconomic groups.8 A longer SA in turn is a predictor for DPs.7 9 10

Central to work careers and sustainable working life patterns, SA is also a risk factor for future unemployment.7 11 12 A combination of both SA and prolonged unemployment may in particular hamper future work life participation,13 14 not least because unemployment and poor health tend to reinforce each other.15–17

It is important to understand how the combination of unemployment and work disability affects future labour market pathways, and uncover the predictors of these pathways. Not surprisingly, previous unemployment is a solid predictor of recurring unemployment after SA.7 18 In their 4-year follow-up of SA, Pedersen et al7 showed how the likelihood of transitioning from SA to unemployment was increased by both previous unemployment and SA spells. A history of unemployment and SA likewise increased the risk for future unemployment-to-SA transition, showing the double-burden of employment and work disability.

Regardless of the initial attachment to work, labour market pathways after SA often include many transitions between sick leave, employment, unemployment or rehabilitation.19–21 Sequence analysis has proven to be a valuable method in capturing these phases and transitions, as well as identifying subgroups with specific pathways.21–25 In sequence analysis studies covering all-cause SA, Madsen23 and Pedersen et al21 have identified subgroups based on post-SA labour market states, such as emphases on return to work or dependency on temporary or permanent social benefits. Madsen’s study showed the association of lower education and lower occupational class with later labour market pathways characterised by unemployment, reliance on support or prolonged disability. However, both these studies lacked unemployment history as a baseline characteristic affecting the labour market pathways. More precisely, studies segmenting the labour market pathways for the unemployed on an all-cause SA with a multistate analysis such as sequence analysis has been lacking in past research.

This study concentrates on SA periods of unemployed persons and follows their labour market pathways from the start of an all-cause SA. We aim to unravel subgroups within these unemployed sickness absentees. Moreover, we aim to identify covariates of these subgroups, as in addition to employment status, the pathways after SA can depend on a wide range of demographic, socioeconomic and medical characteristics and SA diagnosis.21–26 New findings may recognise possibilities for timely prevention for unemployed subgroups and successful direction of resources to support sustainable work careers.


Study population and follow-up

All persons who started a full SA spell during 2016 were first retrieved from the register of the Social Insurance Institution of Finland (Kela). The study population was then restricted to persons aged 18–59 years with no SA days during 12 months prior to the reference spell, and who had received unemployment benefit when the work disability started (N=12 639). The age limits were set so that all the subjects would be of adult age and would not reach the lowest limit of old-age pension in Finland (63 years) during the follow-up. Subjects were followed for 36 months (3 years) from the first SA day.

Data on full and partial SA spells, demographics and the entitlements to reimbursements for medicine expenses were also retrieved from registers of Kela. Employment and unemployment benefit spells were retrieved from registers of the Finnish Centre for Pensions. Data on temporary (full and partial) and permanent (full and partial) DPs, rehabilitation spells and benefits were retrieved from Kela and the Finnish Centre for Pensions. All data on benefit and employment spells included start and end dates. Data on educational level were obtained from Statistics Finland.

Disability and rehabilitation benefits in Finland

SA was measured through compensated sickness allowance days. Kela can pay sickness allowance to non-retired persons aged 16–67 years as compensation for loss of income due to sickness or impairment. The unemployed may receive sickness benefits if their weakened work ability prevents them from seeking work. In 2016, 12% of all persons starting a sickness allowance spell were unemployed (Kela restricted database). The allowance can be paid when the SA exceeds ten weekdays (including Saturdays). For the employed, these 10 days are covered by the employer. A physician’s sickness certificate is needed for the allowance. The allowance can generally be paid up to 12 months during 2 years. If work ability is reduced but the beneficiary is able to continue working part time, partial sickness allowance can be paid.

As a rule, a DP may be granted after the 12-month statutory maximum period of full sickness allowance. A temporary DP can also be granted to compensate for earnings loss during rehabilitation or treatment. Like sickness allowance, pensions can be paid as full or partial benefits. Also, the rehabilitation benefit is meant for securing income during vocational or medical rehabilitation that is already realising or secured.

Definition of the states

Nine mutually exclusive states were constructed for each person and for each of the 36 months of the follow-up. The states were (1) permanent DP, (2) permanent partial DP, (3) rehabilitation (rehabilitation or rehabilitation benefit spells), (4) temporary DP, (5) temporary partial DP, (6) SA, (7) unemployment (8) employment and (9) if none these states could be found for a 1-month unit, the state of that month was recorded as ‘not covered by data’.

In the case of overlap, the state listed earlier in the above list dominated over the states listed later, even if that state covered a lesser amount of days in that month unit. An exception was made when there was both unemployment and employment—and no states dominating them—during a month unit. In that case the state was defined based on which of the two states had more registered days.

In contrast to the DP dimensions, full and partial SA were studied in unison. Partial sickness allowance can be used only by persons who are employed27 and it can, therefore, be presumed to play only a marginal role in our target population. Persons who died during the follow-up (3,4%, N=423) were classified to the ‘not covered by data’ state from the month of death.


The covariates were mostly measured in the beginning of 2016. Age was classified into four groups (see table 1). Marital status was categorised as married, unmarried and divorced, separated or widowed. The level of urbanisation of the home municipality was used as a covariate since it may affect the risk for work disability.10 Subjects were divided into living in an urban, semiurban or a rural municipality according to Statistics Finland.28 Immigrant background (yes/no) was also used as a covariate. Although there are no Finnish studies on the subject, first generation immigrants can have a differing risk for transitioning to DP from SA, shown in Sweden.2 Immigrant background was defined following previous Finnish studies,29 30 and using information on country of birth, language and citizenship. Educational level was categorised into upper tertiary, lower tertiary, secondary and primary education. The amount of unemployment days during 2 years before SA were categorised as less than 6 months, 6–12 months and over 12 months. Likewise, employment days during 2 years before SA were categorised as none, at least 1 day but less than 6 months, 6–12 months and over 12 months. Rejected DP applications during the follow-up (yes/no) were included as they are associated with post-SA unemployment and DP transition.25 Entitlement to reimbursements for medicine expenses was used as a proxy measure for chronic or severe diseases before the SA spell.31 These entitlements are ensured through National Health Insurance and guarantee the recipients’ access to medicines needed for the treatment of certain diseases at a reasonable cost. The study population was also classified according to the diagnosis group of their first SA spell in 2016 according to the International Classification of Diseases (ICD).32 Diagnosis groups were categorised as mental disorders (‘mental SA’, ICD-10 chapter F), musculoskeletal diseases (‘musculoskeletal diseases’, ICD-10 chapter M) and ‘other diagnoses. Disability due to mental disorders was frequent (30%) for the unemployed. In comparison, the relative frequency of mental SA spells in Finland was 18% for all recipients.

Table 1

Covariates in the study sample (N=12 639)

Statistical methods

Sequence analysis was used to study the temporal succession of states and the intertemporal variation between individuals.33 Sequences were defined as a string of 36-month units of the states. To illustrate changes in the nine states over time, status proportion plots and sequence index plots were created. The frequencies of states, total durations of each state and the average number of episodes (successive months of each state) were drawn from the sequences. Moreover, the average number of transitions in total and the average number of different states in the sequences were examined. The Stata SQ-Ados33 and SADI packages34 were used for both the analyses and graphs.

Individual sequences were grouped into clusters based on intersequence distances, applying Optimal matching analysis (OMA,).35 Substitution costs were set at double the size of indel cost, to balance between sensitivity to order, timing and duration of the episodes of states.33 Cluster analysis was conducted with Ward’s linkage.33 Point biserial correlation, average silhouette width, Hubert’s C coefficient, Calinski-Harabasz pseudo-F’s index were used to determine the number of clusters.36 37 Finally, multinomial regression analysis, with ORs and their 95%CIs, was used to analyse covariates associated with belonging to each cluster.

Patient and public involvement



Frequencies and duration of states and transitions between states

Figure 1 visualises the proportion of individuals in each state in the 36 follow-up months. The proportion of individuals on SA decreased substantially from the first follow-up months until circa the 12th follow-up month and stayed stable from then on. Unemployment was the most frequent state after the first few follow-up months. The proportion of persons employed increased slowly but remained at a rather low level. After 12 months, the frequencies of permanent and temporary full DP started to increase, as at this point many persons reach the statutory maximum period of full sickness allowance. Partial pensions were relatively rare. The proportion of persons on rehabilitation periods remained stable from the start to the end of the follow-up. The proportion of those in some state not covered by data increased rapidly from the first months and stayed rather stable over time.

Figure 1
Figure 1

Status proportion plot visualising the relative proportion of the states. DP, disability pension.

Table 2 shows aggregated characteristics of the nine states. Most (80%) had recurring unemployment, whereas 38% found employment at some time point. 13% of the study population transferred to permanent full DP during the follow-up. 13% had a temporary full DP and 21% a rehabilitation period at some point. Partial DPs were used by a much smaller proportion (2%). 40% had at least 1 month with some unknown state not covered by data during the follow-up.

Table 2

Frequency, average duration and average number of episodes of the states

Unemployment covered on average 15.1 months of the total follow-up time. Also SA (mean 6.4 months) and employment (4.8 months) showed rather long average durations. Respectively, the average number of episodes was highest for unemployment (1.7) and SA (1.5).

Overall, the average number of transitions between states during the follow-up was 4.3 (SD 2.9). While one (17%), two (15%) or three (16%) transitions were most typical, 52% had four or more transitions during the follow-up. The average number of different states in a sequence was 3.1 (SD 0.8).

Clusters of sequences

The cluster-stopping indexes supported solutions with either five, six or seven clusters (see online supplemental table S1). However, the six-cluster solution was the best in distinguishing between cluster identities with different emphases on the states and was thus chosen.

Supplemental material

Status proportion plots and sequence index plots were produced to illustrate the clusters (figure 2). The sequence index plots visualise how fragmented many subjects’ individual labour market pathways are. Within the index plots, individuals were sorted primarily by the first state and second by the second state in order to help the reader. To further characterise the six clusters, aggregated sequence characteristics (online supplemental table S2) and 10 most frequent sequence patterns (online supplemental table S3) were drawn for each cluster.

Figure 2
Figure 2

Status proportion plots (left) and sequence index plots (right) over the follow-up. DP, disability pension; SA, sickness absence.

The largest cluster 1 (44%) was dominated by recurring unemployment. After the initial SA, many faced prolonged unemployment that lasted on average 26.5 months in total during the 36-month follow-up (online supplemental table S2). A significant proportion of persons transitioned to the unemployment period directly from SA (figure 2), and for 21% that unemployment period lasted until the end of the follow-up (online supplemental table S3). Still, transitions between states were frequent, often between unemployment and either SA, employment or some unknown state not covered by our data (figure 2, online supplemental table S3). Although one in three had employment at some point (34%, online supplemental table S2), and the frequency of employment increased during the follow-up, it was relatively rare at the end of the follow-up (11%, not presented in tables). Rehabilitation spells and especially DPs were rare.

Cluster 2 (18%) was characterised by employment after a short SA, reflected both by the long total duration of employment (on average 20.4 months) and a high average number of employment episodes (2.5). At the end of the follow-up, 69% were employed (not presented in tables). On average, persons in this cluster had a shorter initial SA spell as well as a smaller total amount of SA months than other clusters (figure 2, online supplemental table S2). This cluster had the highest average amount of transitions (5.8, online supplemental table S2). People often transitioned from SA back to unemployment and then found employment in the first 18 months of the follow-up (figure 2, online supplemental table S3). Instead, transitioning to employment through rehabilitation was rare (figure 2, online supplemental table S2). DPs were also rare in this cluster.

Cluster 3 (12%) was dominated by rehabilitation, recurring SA and unemployment. Contrary to other clusters, two out of three (66%) persons in this cluster had a rehabilitation spell during the follow-up. However, rehabilitation rarely led to sustained employment: although 25% had employment at some point, only 8% were employed at the end of the follow-up (not presented in tables). Instead, recurring SA spells and unemployment were typical. Unemployment was most typical during the second year after which recurring SA spells started to increase. The frequency of full DPs also increased somewhat towards the end of the follow-up, as did the frequency of permanent partial DP. Like cluster 2, cluster 3 had an especially high average number of transitions (5.3), seen also in figure 2, and reflecting fragmentation of the labour market pathways.

Cluster 4 (11%) depicted persons transitioning to unknown sources of income, not covered by our data. As shown in figure 2, persons in this cluster often transitioned straight from SA to the state not covered by data or first back to unemployment and then to the state not covered by data. One out of five (19%) had only states not covered by data after the initial SA (online supplemental table S3). DPs, rehabilitation or employment were quite rare in this cluster.

Cluster 5 (9%) consisted of individuals transferring to permanent full DP after a prolonged SA. Before the DP transition, approximately half had a continuous 1-year long SA spell (figure 2). Almost half (46%) transitioned to permanent full DP either directly from SA or through a temporary DP (online supplemental table S3). However, a considerable proportion of individuals were unemployed or in a state not covered by data between SA and the DP (figure 2, online supplemental table S3). The unknown state was most frequent around the 10th month of the follow-up. Some also had unemployment or rehabilitation between SA and the DP transition (figure 2, online supplemental table S3). Employment spells were especially rare in this cluster (online supplemental table S2).

Finally, cluster 6 (7%) was dominated by temporary full DP after a prolonged SA. Temporary full DP increased rapidly 10–12 months after the start of the follow-up. During the last 12 months, also permanent full DPs increased, and one in four (26%) made that transition. Like in cluster 5, around 10 months into the follow-up, the unknown state not covered by data was frequent (figure 2), usually between SA and the temporary full DP (online supplemental table S3).

Associations between the covariates and cluster memberships

Finally, we conducted a multinomial logistic regression analysis to examine how the covariates were associated with cluster membership (table 3). Covariates were mutually adjusted for. Cluster 2 was chosen as the reference cluster in the model, since employment is the societally preferred status after SA. Thus, in the analysis, the other clusters were not directly compared with each other but to cluster 2.

Table 3

Associations between covariates and cluster memberships

There were four covariates that distinguished all five other clusters from the reference cluster 2—all five clusters were associated with not being married, less employment days and more unemployment days before the SA spell, having a rejected DP application during the follow-up, having a chronic illness before the SA and an SA based on a mental disorder.

In addition, the following associations were found. Cluster 1 membership was characterised by older age, not living in a rural municipality and having a primary or secondary educational level.

Cluster 3 membership was characterised by female sex, older age, not living in a rural municipality and not having an immigrant background. The association with having a rejected DP application during the follow-up was strong.

Cluster 4 was characterised by being rather 18–30 years than 41–50 years, and having only a primary educational level. In addition, cluster 4 was the only cluster associated with having an immigrant background.

Cluster 5 was characterised by male sex, older age, not living in a rural municipality, not having an immigrant background and having a primary or secondary educational level. This cluster was strongly associated with being over 50 years.

Finally, cluster 6 was characterised by male sex, older age, living in an urban municipality and having a primary or secondary educational level. The association with an SA based on a mental disorder was strong.

A comprehensive table with CIs is presented as online supplemental table S4.


This study followed the labour market pathways of unemployed persons who started an SA spell. We aimed to unravel subgroups based on nine altering labour market states and to identify covariates of these subgroups. Studies segmenting the pathways for unemployed persons on SA have been lacking in previous literature.

Our main finding is that unemployed persons with work disability are a heterogeneous group, with very different pathways after the start of SA. Sequence and clustering analyses revealed six clusters with unique emphases on the examined states. As found in previous studies,7 18 26 unemployment can be a strong indicator of recurring unemployment after SA. Indeed, those with recurring unemployment formed the biggest subgroup found in our study (cluster 1, 44%). Many in this cluster transitioned directly to unemployment from SA. From then on, some had long unemployment periods, while some alternated between other states, often unemployment and either recurring SA or employment. Interestingly, while one in three in this cluster had employment during the 3-year follow-up, they mostly transitioned away from employment, not succeeding in sustained employment. Like some other clusters not characterised by post-SA employment, this group had a lower educational level and more unemployment days before SA than those finding employment after SA (see group below).

The second biggest group (cluster 2, 18%) was persons finding employment after SA. Knowing that for many, the combination of unemployment and work disability is a poor basis for future labour market pathways13 14 this group deserves attention as a positive exception. For many in this cluster, transition to employment happened after a relatively short SA and a recurring unemployment spell. On average, they also had many transitions between states, showing how return to work after work disability is rarely a straightforward process, but includes many phases between alternating labour market positions.19–21 Compared with all other clusters, this group was characterised by a favourable background, presumably aiding the transition to work. Although unemployed at the time the SA spell started, they were more attached to work during the last 2 years before SA than others in this study. Compared with other clusters, they also had less pre-SA chronic illnesses, and their SA was more often based on somatic rather than mental disorders than in other clusters. Fewer chronic illnesses or comorbidities understandably increase the likelihood for shorter occupational disability,38 39 found in this group. As for the diagnosis behind SA, SAs based on mental disorders have been associated with a lower likelihood of return-to-work.3 21 26 In a sequence analysis study by Pedersen et al,21 mental health problems likewise predicted prolonged SA, and slower return to work among Danish long-term sickness absentees. Possibly because of the favourable background, rehabilitation did not play a significant role in transitioning to employment. Our results also show that many predictors of positive labour market pathways after SA are valid among the unemployed as well.

One cluster (cluster 3, 12%) was formed of persons not finding a settled labour market status after SA despite a high frequency of rehabilitation. Rehabilitation rarely seemed to lead to sustained employment but often to recurring SA and unemployment. This is not surprising as multiple studies have found weak effects between rehabilitation and later work participation.40–44 Instead this group had many transitions between the labour market states. Unemployment occurred especially after the first follow-up year and recurring SA especially towards the end of the follow-up. This temporal pattern is probably associated with the Finnish system: After being on SA for 12 months in total, the claimant reaches a statutory maximum of sickness allowance. After this, the claimant is entitled to sickness allowance again only after 1 year’s time if he/she again applies for the allowance based on the same diagnosis as before. Thus, this cluster can include persons transitioning back to unemployment after the statutory maximum period, and then starting a new SA spell approximately 2 years into the follow-up. In addition to the frequent transitions, this group had a high frequency of rejected DP applications during the follow-up. Previous studies have shown that a rejected DP application is a risk for future unemployment, being dependent on social assistance or exiting the labour market.25 45–47 In our study, a rejected DP was measured during the follow-up and cannot be perceived as a predictor but a covariate.

Furthermore, the clustering of labour market sequences revealed two distinct clusters (5 and 6) transitioning to DP. Almost 1 in 10 had a pathway ending relatively early to permanent full DP. For approximately half of this cluster, permanent DP occurred after a continuous year-long SA. Some transitioned then directly from SA to permanent pension, some through other states such as temporary DP or unemployment. Either way, this is a group well identified in previous research, on average older, with lower education level and relatively often long-term unemployed.2 5 9 48 Based on Madsen’s sequence analysis,23 as well as Finnish retrospective register studies,49 50 persons transitioning to permanent work disability rarely have frequent efforts to rehabilitate work ability or return to employment before the DP transition. The other DP cluster was depicted by long temporary full DPs. One in four persons in this cluster transitioned to a permanent full DP, typical for claimants of temporal DP.51 52 Besides a weaker attachment to work during 2 years before SA, this cluster was especially associated with an SA based on a mental disorder. The association between sick leave due to mental disorders and a high DP risk has been found in previous studies.10 53 54 Moreover, in Finland, mental disorders have been shown to increase the likelihood for a prolonged temporary DP.51 55

Finally, we found a segment (cluster 4) transitioning to long periods that are not covered in our data. This was a young group, often not married, with a lower educational level, and also the only cluster associated with an immigrant background. These persons may live either on other benefits not captured by our data, such as family benefits or students’ benefits or on savings, other household members’ income or on last-resort income support. Despite not being properly identified in this study, this group may be at an economic and health-related risk. Unemployment at a young age is a clear risk factor for later ill-health, SA and low frequency of labour market participation.6 56 Likewise, SA at a young age is a risk factor for unemployment and work disability later on.57

As sensitivity analyses, we conducted the OMA clustering with indel and substitution set equal and with substitution costs set 0.5 the size of the indel cost. Both analyses revealed a six-cluster solution supported by cluster-stopping indexes, and very similar cluster identities as the original analysis.

Methodological considerations

We were able to study multiple relevant labour market states for the unemployed, and to use register data on all states. Registers are considered objective sources, with no self-report bias and very little loss to follow-up. Furthermore, unlike many multistate analyses on SA, we included and examined a comprehensive set of register-based demographic, socioeconomic and disability-related covariates for the clusters. A limitation in our study is the possibility that unemployed persons on an SA do not necessarily represent well all unemployed persons with lowered work ability. For many persons on unemployment benefit, sickness allowance would not raise their income level. Therefore, they do not have a strong incentive to apply sickness allowance.58 Another limitation is that with a register-based study we could not take into account the whole range of social and economic problems and negative life events faced by many of the unemployed with health issues.59 60 These circumstances can affect the labour market pathways. Last, our data on services supporting employment were restricted to rehabilitation. Future studies should use linked data on a wider set of services (eg, health and social services) to study the labour market pathways in more detail.

Practical implications

For many, the combination of unemployment and temporary work disability means lowered chances for future attachment to work or even regained work ability. Five out of six subgroups mostly relied on social welfare benefits for years after the start of SA or left the labour market. Besides a lower health status, also long periods outside employment, a rejected DP application and a mental disorder were identified as background risk factors for all five non-employment pathways. Unemployed persons with these features could benefit from targeted support. The same applies to older individuals and those with a low educational level as they are at risk for long-term unemployment and permanent work disability.

More generally, the unemployed in Finland usually do not have access to free-of-charge occupational health services, specialised in work ability issues.61 This may cause insufficiency in early treatment and rehabilitation, early assessment of remaining work ability or direction to appropriate benefits. With that there is a risk for not only deteriorating health,15–17 but also disability pensioning and a rejected pension application on the other.62

All in all, work ability problems among the unemployed should be tackled in the early stages, just like with the employed. Efforts to maintain and support work ability among the unemployed are crucial in enabling employment and preventing disability retirements.

This post was originally published on https://bmjopen.bmj.com