Sequence analysis of sickness absence and disability pension days in 2012–2018 among privately employed white-collar workers in Sweden: a prospective cohort study

Objective The aim of the study is to explore sequences of sickness absence (SA) and disability pension (DP) days from 2012 to 2018 among privately employed white-collar workers. Design A 7-year prospective cohort study using microdata from nationwide registers. Setting Sweden. Participants All 1 283 516 privately employed white-collar workers in Sweden in 2012 aged 18–67. Methods Sequence analysis was used to describe clusters of individuals who followed similar development of SA and DP net days/year, and multinomial logistic regression to analyse associations between sociodemographic variables and belonging to each observed cluster of sequences. Odds ratios (ORs) and 95% confidence intervals (CIs) were adjusted for baseline sociodemographics. Results We identified five clusters of SA and DP sequences: (1) ‘low or no SA or DP’ (88.7% of the population), (2) ‘SA due to other than mental diagnosis’ (5.2%), (3) ‘SA due to mental diagnosis’ (3.4%), (4) ‘not eligible for SA or DP’ (1.4%) and (5) ‘DP’ (1.2%). Men, highly educated, born outside Sweden and high-income earners were more likely to belong to the first and the fourth cluster (ORs 1.13–4.49). The second, third and fifth clusters consisted mainly of women, low educated and low-income (ORs 1.22–8.90). There were only small differences between branches of industry in adjusted analyses, and many were not significant. Conclusion In general, only a few privately employed white-collar workers had SA and even fewer had DP during the 7-year follow-up. The risk of belonging to a cluster characterised by SA or DP varied by sex, levels of education and income, and other sociodemographic factors.


INTRODUCTION
Sickness absence (SA) and disability pension (DP) have adverse consequences for individuals, their employers and welfare states.4][5] Nevertheless, they constitute a large part of the workforceapproximately half in Sweden in 2018, 6 and about half of them are privately employed. 7hus, work incapacity in this group can impose high costs for employees, employers and the welfare state.To prevent work incapacity in this population, more knowledge is needed on the determinants and the process of developing long-term work incapacity.
Previous research on SA and DP within specific occupations or occupational groups has mainly focused on so-called highrisk groups, for example, manual workers and blue-collar workers, [8][9][10][11][12][13][14] while studies on white-collar workers are scarce.8][19][20] These studies have shown that there are differences in rates of SA among white-collar workers by age,

STRENGTHS AND LIMITATIONS OF THIS STUDY
⇒ Detailed sociodemographic microdata, linked from different population-based registers, about a cohort of all privately employed white-collar workers in Sweden in 2012.⇒ Long study period (7 years) with no los to follow-up and no bias from self-reports.⇒ Use of sequence analysis to capture the heterogeneity of the different sickness absence and disability pension patterns over time.⇒ How many and which states to include in the analyses is dependent on researcher judgement and thus can be arbitrary.
Open access gender, education, occupational status, and other sociodemographic and socioeconomic factors.
Studies on white-collar workers in the private sector are even more limited.In general, large-scale studies have demonstrated that SA rates in the private sector are generally lower than in the public sector. 21 22There are several studies on SA and/or DP among private-sector employees, however, hardly any specifically among whitecollar workers, despite how many it concerns.Moreover, the few such studies are mainly based on small, selected populations, have large drop-out rates and are mainly based on self-reported data. 5 23-25So far, only three largescale studies on private sector white-collar employees have been published: two Swedish studies 3 26 and a Greek study on private sector employees (also including blue-collar employees) that found a smaller SA rate in the shipyard industry than in other industries. 27The results of the two Swedish studies showed that the risk of SA and DP-and the risk of belonging to an adverse SA/DP trajectorydiffered among white-collar workers by age, sex, education, a branch of industry, psychosocial exposures at work and other sociodemographic factors.Further, none of these studies have accounted for transitions between other labour market states in addition to SA and DP, such as employment and unemployment.More studies using full population data with a longitudinal research design are needed to increase the knowledge base.
Moreover, both SA and DP are complex phenomena affected by many factors.Both increase with age, are lower in people with higher education and non-immigrants, and differ by sex; in most occupations, women have higher SA/DP levels than men, hence it is important to include such factors in studies of future SA/DP. 3 28-30equence analysis is a good method to study developments over time.Unlike more traditionally used methods, such as event history analysis or growth curve models, sequence analysis can describe the duration and frequency of multiple categorical statuses.This holistic perspective is essential in providing an overview of the future development of SA and DP, and in identifying potential sub-groups within a population who share particular patterns in terms of such SA and DP.
The aim of this study was to identify sequences of whitecollar workers in the private sector who follow future similar sequences of SA and DP days/year and second, to analyse the sociodemographic and diagnostic composition of the observed clusters of SA and DP.

Data sources and population
We conducted a 7-year prospective population-based cohort study.We used microdata from the following three nationwide Swedish administrative registers, linked at the individual level by personal identity number (a unique 10-digit number assigned to all Swedish residents) 31  The study population consisted of all individuals aged 18-67 years who lived in Sweden on both 31 December 2011 and 31 December 2012, who had an occupational code according to the Swedish Standard for Occupational Classification (SNI) indicating a white-collar occupation, 3 were employed at a private-sector company during 2012, and had an income from work, parental benefits, SA and/or DP that amounted to at least 75% of the necessary income level to qualify for SA benefits from the Social Insurance Agency (SEK7920 in 2012, approximately €910 by the 2012 exchange rate, updated yearly in line with inflation).We excluded unemployed, self-employed, and those who were on full-time DP for the entire year 2012 (n=461).The total study cohort included 1 283 516 individuals.

Public SA insurance in Sweden
In Sweden, all residents aged at least 16 years with an income from work or unemployment benefits who have a reduced work capacity due to morbidity are covered by the national public SA insurance. 32A physician's certificate is required after 7 days.After an unpaid qualifying day, the employer pays the following 13 SA days, after which SA benefits are paid by the Social Insurance Agency.For the unemployed, the Social Insurance Agency pays after the first qualifying day.Thus, we excluded SA spells shorter than 15 days, in order not to introduce bias, since we only had information of SA spells exceeding 14 days for the employed.There was no limitation regarding how long an SA spell could be ongoing for.Residents in Sweden aged 19-64 years, whose work capacity is long-term or permanently reduced, can be granted DP from the Social Insurance Agency.SA covers about 80% and DP about 65% of lost income, both up to a certain level.Both SA and DP can be granted for parttime or full-time (25%, 50%, 75% or 100% of ordinary work hours).This means that people can be on partial SA and DP at the same time.

Sociodemographic and work-related variables
We included information on sex, age group, country of birth, educational level, family composition, type of living area and branch of industry based on the SNI categorised into the following six groups: manufacturing, services, transport, construction and installation, care and education, or commerce and hospitality.All variables were measured at the baseline year 2012.

Open access
Measures on SA and DP We used SA net days/year and DP net days/year as outcomes.Net days were calculated so that partial days of SA or DP were combined, for example, 2 days of part-time SA for 50% were summed to one net day, and a similar procedure was used for DP days.The first 14 days of SA spells (>14 days) were counted as being of the same grade as day 15 for the purpose of calculating net days.The number of SA net days in 2012 were categorised as shown in table 1.The SA diagnoses were categorised into the following seven International Classification of Disease groups 33 : cancer (C00-D48), mental diagnoses (F00-F99 and Z73), circulatory diseases (I00-I99), musculoskeletal diagnoses (M00-M99), pregnancy-related diagnoses (O00-O99), injuries (S00-T98) and other diagnostic groups (all others, including missing diagnosis (approximately 1% of all spells).In the multinomial logistic regression, pregnancy-related diagnoses were dropped, as no men could have pregnancy-related diagnoses, which made it highly correlated with sex.
In analyses of the yearly states of SA/DP, all diagnoses other than mental and musculoskeletal diseases were combined to form one status.Any DP, regardless of diagnosis, was considered as one group.

Sequence analysis and multinomial regression analysis
We used sequence analysis to examine different statuses of SA and DP days/year, and the transitions between such statuses.SA and DP status was measured on a yearly basis for each of the seven follow-up years and was coded into one of the following seven statuses: 1.No SA or DP. 2. SA due to mental diagnoses but no DP.3. SA due to musculoskeletal diagnoses but no DP.4. SA due to other diagnoses but no DP. 5.Both SA and DP. 6.Only DP. 7. Ineligible for SA and DP (due to being emigrated, dead, retired, or having no qualifying income from work or work-related benefits).Individuals who had SA in more than one diagnostic category were assigned to the diagnostic category they had the most days in that year.We illustrated the individual and proportional changes in SA/DP statuses over time with sequence index plots and status proportion plots. 34e used an optimal matching (OM) method to group similar sequences with each other.OM measures the dissimilarities through the changes needed to make two sequences identical. 35In other words, the OM algorithm creates metric distances between two sequences, which can be defined as the minimum combination of replacements, insertion and deletions to transform one sequence to another. 36We used R statistical program version V.4.1.0and packages TraMineR and nnet for the sequence analysis.
We used multinomial regression analysis to analyse how sociodemographic characteristics and branch of industry were associated with each of the obtained clusters, using the first cluster as the reference category.ORs with their 95% CIs were reported.

Patient and public involvement
Representatives from the private white-collar sector in Sweden, both for employees and employers (the labour union PTK, the Confederation of Swedish Enterprise and Alecta) were involved in selecting the research questions through joint meetings throughout the project period, and afterwards in disseminating results.

Characteristics of the study population
Table 1 shows the characteristics of the study cohort of the 1 283 516 privately employed white-collar workers.1).We called this cluster 'SA due to other diagnoses'.
Cluster 3 (n=43 871, 3.4%) consisted mostly of those who had SA mainly due to mental diagnoses (figure 1).We called this cluster 'SA due to mental diagnoses'.
Cluster 4 (n=18 150, 1.4%) was characterised by individuals who were not eligible for SA or DP since they either died, emigrated or left the labour force (figure 1).We called this cluster 'ineligible for SA and DP'.
To better understand the most common SA and DP sequences, we examined the 20 most frequent sequences (online supplemental figure 1).Most (68.4%) had no SA or DP during the follow-up.The remaining trajectories largely consisted of sequences where individuals had SA for 1 year and then returned to no SA or DP.Very few had DP during the follow-up.

The associations between individual characteristics and belonging to clusters of SA and DP
To study how individual characteristics and SA at baseline were associated with cluster membership, we used multinomial regression analysis.cluster 1 'low or no SA or DP' was used as the reference category since it was the largest and most homogeneous in its sequence content (table 2).Cluster 1 could be described as consisting of men of younger working-age, who had high levels of education and income, worked in service industry or in manufacturing and had no or only little SA in 2012 (online supplemental table 1).
In the fully adjusted models, compared with cluster 1 'low or no SA or DP', women (men having an OR of 0.47 (95% CI 0.46 to 0.47)), over or under 35-44 years (but not over 64 years), those with less than tertiary education, who were born outside EU25 countries (i.e., the 25 Women (men having an OR 0.38 (95% CI 0.37 to 0.39)), 34-44 years, who had less than tertiary education, who were single living with children, worked in education, care, nursing or social service industry, had medium low income, had more than 180 SA days in 2012, especially due to mental diagnoses, had the highest ORs of belonging to cluster 3 'SA due to mental diagnoses' (table 2).The second and third clusters could be described as consisting of working-age women, who had less than tertiary education and medium income, who worked in education, care, nursing or social service industry and had some SA in baseline year, especially due to mental diagnoses in the third cluster (online supplemental table 1).
The OR for belonging to cluster 4 'ineligible for SA and DP' was the highest in men (OR 1.13; 95% CI 1.10 to 1.17), 65-67 year, had primary education, lived without children, were born outside Sweden, had a very low income, who worked in trade, hotel or restaurant industry or transport industry, had >180 SA days in 2012 and had SA due to circulatory diagnoses (table 2).The fourth cluster could be described as consisting of men over 64 years, who had primary education and were born outside Sweden, had low income and had long-term SA in 2012, especially due to cancer (online supplemental table 1).
The OR of belonging to cluster 5 'DP' were higher in women (OR 0.69 (95% CI 0.66 to 0.72) in men), 45-64 years, who had less than tertiary education, were born in Sweden, who were single, worked in manufacturing, had low to medium low income, had at least 30 SA days in 2012 and especially those with SA due to circulatory diagnoses (table 2).This fifth cluster could be described as consisting of older working age women, with low education, working in service industry with low income and long-term SA at baseline (online supplemental table 1).

Discussion
In this large prospective cohort study of all 1.3 million privately employed white-collar workers in Sweden in 2012, we analysed the development of their future number of SA and DP days/year up through 2018.In general, most of the employees had no SA during the follow-up and DP was even rarer.We found five clusters of future SA and DP trajectories: (1) 'low or no SA or DP' (88.7% of all), (2) 'SA due to other (than mental) diagnosis' (5.2%), (3) 'SA due to mental diagnosis' (3.4%), (4) 'not eligible for SA or DP' (1.4%) and ( 5) 'DP' (1.2%).These results suggest that the majority of privately employed white-collar workers were doing well in terms of SA/DP.
We found some differences related to sociodemographic factors in terms of belonging to different sequence clusters.Many of those in cluster 1 'low or no SA or DP' were Swedish-born, 25-54 years, highly educated, and highincome earning men, who lived in a large city, and were married or cohabiting with children at home.The same sociodemographic characteristics are typically associated with lower risk of SA or DP in longitudinal nationwide studies. 28 29

Open access
ref.

Country of birth Sweden
ref.

Open access
We also found that female sex, low education, low income, and working in education, care, nursing, or social services were associated with a higher risk of belonging to clusters characterised by at least some SA or DP.Similar results were found in a previous cross-sectional study using the same data with number and prevalence of SA days as outcomes, 3 as well as studies on SA and DP among whitecollar workers in the retail and wholesale industry. 26 37In general, previous longitudinal population-based studies have consistently found that women, low educated and low-and income earners, 28 29 and those working in healthcare and service industries 22 have a higher risk of SA and/or DP.While these characteristics-low education, low income and working in the healthcare industry-are usually considered as explanations to why blue-collar workers have a higher risk of SA or DP than white-collar workers, 4 38 our results indicate that the same risk factors apply within white-collar employees working in the private sector.More knowledge is warranted regarding potential mechanisms behind this.
It is understandable that SA due to mental diagnoses constituted an independent cluster since among whitecollar workers that is the most common specific diagnostic group of SA and/or DP. 1 37 39-41 This cluster was more common among women, 34-44 years, less than tertiary educated, low-income earners who worked in education, care, nursing and social industry, and had a long SA spell in 2012, which are known risk factors for SA due to mental diagnoses in general. 42 43he cluster 'ineligible for SA or DP' had relatively many individuals aged ≥ 55 years, which makes sense since those who left paid work (eg, through old-age pension) or died during the follow-up belonged to this cluster.There were also many highly educated and high-income earners, who typically are occupationally and geographically mobile, in this cluster.Relatively many of them were born outside Sweden; hence many of them probably emigrated from Sweden.Those who had SA due to cancer in 2012, had higher OR of belonging to this cluster than to any other cluster.
We found that the estimates for associations between branch of industry and cluster attenuated in the adjusted analyses, indicating that differences between the various branches of industry were more related to other factors.The Swedish Social Insurance Agency has found that in Sweden, occupation is more closely associated with SA than branch of industry. 44However, to what extent this is true within the group white-collar workers is unknown and should be further studied.

Strength and limitations
Strengths of this study are the use of a large, population-based cohort the use of linked microdata from three high-quality nationwide registers without dropouts, the long prospective follow-up, and that all data were administrative, not self-reports with possible bias.Using sequence analysis allowed us to explore specific subgroups in the development of SA and DP.Other strengths are that all included were covered by the same public SA and DP insurances, and the high employment-frequency in Sweden, that is, the healthy-worker effect did not bias the result much.
Since the study population consisted of privately employed white-collar workers in Sweden, the results cannot directly be generalised to other types of occupational populations or to other countries with other SA/ DP systems or employment frequencies.Future studies might choose to explore other, or more specific SA states, regarding number of SA days or part-time and full-time SA/DP.As this was an observational study, no causal inferences can be drawn from the results.

CONCLUSION
In general, privately employed white-collar workers rarely had SA and even more rarely DP days during the 7-year follow-up.The risk of belonging to a cluster characterised by receiving SA varied by sex, levels of education and income, branch of industry and other sociodemographic factors.
Contributors KF and KA planned and designed the study.KF supervised the analyses.LS wrote the first draft of the paper.All authors critically revised the paper for intellectual content.All authors approved the submission of the study.KF acts as guarantor for the study.

Figure 1
Figure 1 Density plot of sickness absence (SA) and disability pension (DP) visualising the proportion of each SA and DP status for each cluster over the follow-up.

Table 1
Characteristics of the study cohort in 2012 €50 556 according to the average 2012 conversion rate) per year (35.8%).A large majority did not have any SA in 2012: only around 7% had at least one SA spell >14 days.Around 2.2% had SA due to mental diagnoses, 1.4% due to musculoskeletal diagnoses and around 3.8% due to any other diagnoses.
over half lived in a large city (51.5%) and had a tertiary education (53.7%).The majority were born in Sweden (89.7%), and almost half were married or cohabiting and having children below the age of 18 at home (47.3%).The largest group was the service industry (43.1%) and the largest income group was those who earned over SEK440 000 (around

Table 2
Multinomial regression with five clusters of sickness absence (SA) and disability pension (DP) days/year among privately employed white-collar workers, ORs with their 95% CIs, cluster 1 'low or no SA or DP' was used as reference group