In contemporary societies, sickness absence is an important economic challenge because of its high costs to employers and society.1 Sickness absence has also been increasingly used in occupational epidemiology as an outcome reflecting reduced work ability.2 Typically, the goal has been to search risk factors for the occurrence of individual sickness-absence episodes and their durations. Various individual, occupational, and social determinants for sickness absence have been identified.3,4 Nevertheless, less attention has been paid to the fact that sickness absence can be presumed to be a recurrent phenomenon where absence episodes tend to occur repeatedly among individuals who have already experienced absence.
There are various reasons to assume that sick leave may occur repeatedly for some individuals. For many diseases, the usual course of illness comes and goes over time. Working conditions and other extra-individual factors may also affect sickness absence, which can recur if poor conditions persist. Furthermore, some people may be more vulnerable to illnesses, leading to repeated periods of absence. Sickness-absence spells of different durations may also be affected by different factors. Longer sickness-absence episodes have been shown to predict early retirement and mortality and they are more likely to reflect chronic morbidity compared with shorter sickness-absence episodes.5,6 For shorter absence spells, the reasons may be more heterogeneous. Short sickness absences are likely to mainly reflect minor morbidities but their risk factors are thought to be more variable than are those of longer episodes, and these can include non–health-related reasons. Besides working conditions, it is possible that short sickness-absence spells reflect the cultures and norms that shape sickness-absence behaviors. Although short absence spells are more common, the majority of sickness-absence days accumulate from long-term absence and thus also incur the largest costs.
Few previous studies have focused on the recurrence of sickness absence. A study from the Netherlands found that those who had frequent sickness absences during a baseline period of 1 year had higher risks of recurrent sickness absences during a 4-year follow-up period.7 They were also more likely to have long-term sickness absences of 6 weeks or more. A study comparing the incidence of first sickness-absence episode because of common mental disorders with the recurrence density of subsequent absence episodes concluded that prior absence spells were associated with an increased risk of future absence in a Dutch Post and Telecommunications company.8 Another study among Dutch hospital employees found that the predictive ability of absence history to future absence was limited to 2 years.9 Studies among Brazilian hospital employees also found that the risk of sickness absence was higher for those who had already had some sickness absence.10,11 Furthermore, the prospective cohort studied showed that the risk of sickness absence increased progressively by the number of prior sickness-absence episodes irrespective of the diagnosis. Nevertheless, these previous studies have not separately considered sickness-absence episodes of different lengths and have paid insufficient attention to the factors that might shape these associations.
The aim of this study was to examine whether past sickness-absence episodes are associated with increased risk of subsequent sickness absences. Short (1 to 3 days), intermediate (4 to 14 days), and long (>2 weeks) sickness-absence spells were examined separately using a prospective cohort consisting of City of Helsinki employees. We also examine whether a history of short absence spells increases the risk of future longer sickness absences after controlling for various measured covariates and unmeasured factors by applying frailty models. Furthermore, we examine whether working conditions and health behaviors modify the found associations.
The present study included 6934 municipal employees of the City of Helsinki who during 2000 to 2002 participated in a questionnaire survey on working conditions and health. The data were linked to the employers' sickness-absence records from the day of returning the questionnaire (response rate 67%) onwards for those respondents who had given written permission for the linkage (78%). Consistent with the gender ratio in most Finnish municipalities, the large majority of the employees in our cohort were women. Nonresponse analysis showed only small differences in responding and consenting to register linkages by age, occupational class, income, type of employment contract and employment sector, and the associations of these characteristics with sickness absence by responding and consenting.12
Employment and absence records from the City of Helsinki were used to extract all periods of absence from the day of returning the questionnaire to the end of 2007. If the employee's work contract terminated before that, follow-up was stopped at that point. Using the exact dates from the absence records, we separated all periods into working, sickness absence, and absence for non–health-related reasons. The last category included, for example, maternal and parental leave and absence because of the sickness of a dependent child. Consecutive sickness-absence periods were combined, but if two successive absences were separated by at least 1 day, they were considered distinct spells. The data include altogether 35.347 person-years, and the average follow-up time was 5.1 years.
Questionnaire data were available on occupational class, working conditions, and health behaviors. Occupational class was categorized into managers and professionals, semiprofessionals, and routine nonmanual and manual workers. Working conditions included psychosocial working conditions measured by nine questions on job control and nine questions on job demands from the Karasek job content questionnaire13 and social support measured by the work-related items of the Sarason brief inventory for social support.14 Physical working conditions were based on an 18-item inventory15 that was summarized in three measures on the basis of factor analysis: hazardous exposures at work consisting of nine questions, physical workload consisting of six questions, and computer work consisting of three questions. All working conditions were dichotomized from the highest quartile to evaluate the effect of poor working conditions. Smoking was divided into current smokers and nonsmokers. Body mass index (BMI) was calculated from self-reported height and weight and those with a BMI greater than 30 kg/m2 were determined to be obese. Further details of the measurement of working conditions and health behaviors can be found elsewhere.16,17
The data were recurrent events data in which subjects experience repeated occurrences of several different types of events.18 At any given time during follow-up, employees were classified as working, absent because of sickness, or absent for other reasons. From the beginning of follow-up, employees were tracked from their first sickness-absence episodes, after which they became at risk of the other, and so on.
We first estimated the rates of sickness absences per person-year during follow-up stratified by the accumulated number of past sickness-absence episodes during follow-up. These calculations were conducted separately by the length of past sickness absences and gender. The number of past episodes was truncated at seven because, in particular for long spells, the person-years at risk became increasingly sparse.
Proportional hazards models were then applied to estimate the association of sickness absences with the number of previous absence spells. To cope with recurrent absences, the Andersen and Gill19 model was adopted along with the consideration of individual random effects.20 The number of previous absence spells was used as a time-varying predictor in the Andersen and Gill regression model. All breaks in working not caused by the employee's illness, such as maternal or parental leaves, were taken into account as censoring intervals, during which time employees were not at risk of sickness absence. The latent frailty term in the Andersen and Gill model summarizes the impact of omitted variables on the hazard rate and this can be regarded as unobserved person characteristics, such as genetic background and stable environmental factors. The frailty model accounts for variation in the hazard of persons for repeated absence spells not accounted for by the observed covariates and produces estimates of the relative risk.21 Recurrent events are assumed to be conditionally independent given the frailty and measured covariates.
We performed separate analyses for short (1 to 3 days), intermediate (4 to 14 days), and long (>2 weeks) sickness-absence episodes based on the assumption that they occur more or less independent of each other. This categorization was based on the requirement of a medical certificate among Finnish municipal employees for sickness absences lasting 4 days or more. Because absence periods of 4 days are still relatively short in terms of the severity of an illness, a further cutoff point of 2 weeks was also used in line with previous studies.22,23 We also examined whether prior short and intermediate absence spells were related to subsequent absence spells of longer durations. Furthermore, we tested whether this association was modified by working conditions and health behaviors. Our aim here was not to examine the associations of these factors with the risk of sickness absence (presented previously elsewhere from these data) but to examine whether they modify the effects of previous absence episodes on subsequent ones. All analyses were performed using R version 2.12.2 (The R Foundation for Statistical Computing).
Table 1 presents the absolute rates of sickness absence stratified by gender and the number of prior absence spells during the follow-up period. There were 5470 women in the data of whom 4697 experienced a short sickness-absence spell during the follow-up period and they thus shifted to be at risk of a second short absence spell. The total number of person-years for women following the first short sickness-absence spell was 7435, giving a rate of 0.63 sickness-absence spells per person-year among women with no previous short absence spells during follow-up. When the number of previous sickness-absence spells increased, the number of short sickness-absence events relative to the person-years of follow-up became more common. Among women with six previous short sickness-absence spells, the rate of the next one was 1.94 sickness-absence spells per year of follow-up. Altogether, the data included 44,186 short sickness-absence episodes among 5470 women and thus 1.57 short sickness-absence events per year on average. Among the 1464 men in the data, 1068 experienced at least one short sickness-absence spell during follow-up, giving a rate of 0.36 first short sickness-absence spells per year. Similar to women, the rate of short sickness-absence spells increased with the increasing number of previous spells. Similar patterns were also observed for intermediate and long sickness-absence spells, which nevertheless were less common than were short ones.
The hazard ratios and 95% confidence intervals (CIs) of a new sickness-absence spell by the number of preceding sickness-absence spells, adjusted for occupational class, working conditions and health behaviors, are shown in Figure 1. The risk of a next sickness-absence spell was higher with the accumulated number of previous sickness-absence spells. The increase was steeper for longer than it was for shorter sickness-absence spells and for men than it was for women. In general, the risk of a new sickness-absence episode increased monotonically and rather linearly by increasing number of past absence spells. Nevertheless, in short sickness-absence spells, there was a threshold for the first absence, and the increase was somewhat less steep after that. For example, among women the risk of experiencing a new short absence spell was 1.29 (95% CI: 1.23 to 1.36) among those who had already experienced one previous absence spell compared with those with no previous absence spells. Among those with six previous short absence spells, the relative risk of experiencing a new one was 1.49 (95% CI: 1.39 to 1.60). Among those with six previous intermediate or long absence spells, the relative risk of experiencing a new one was 1.72 (95% CI: 1.55 to 1.90) and 2.01 (95% CI: 1.63 to 2.48), respectively.
Table 2 presents the hazard ratios and 95% confidence intervals (CIs) of prior short and intermediate absence spells with subsequent absence spells of longer durations. Among women, the risk of intermediate sickness-absence spells increased rather monotonically by the number of preceding short sickness-absence spells. Among those with six preceding short absence spells, the risk of intermediate sickness-absence spells increased to 3.15 (95% CI: 2.86 to 3.46). The results for short sickness-absence spells as predictors of future long absences were relatively similar. For intermediate sickness-absence spells, the risk of long absence spells also increased monotonically, but somewhat more steeply. For men, similar associations were seen, but the associations tended to be slightly stronger than they were among women. The results shown have been adjusted for the covariates but these adjustments only minimally affected the findings.
We also tested for interactions between working conditions, smoking and BMI measured at baseline, and the number of previous sickness-absence spells on the risk of future spells (results not shown). Only weak and casual associations were found, and in unexpected directions: if anything, among those with poorer working conditions, the association of previous sickness-absence spells and the risk of future spells was weaker than it was among those who reported better working conditions. The most consistent associations were found for physically heavy work, which tended to decrease the effect of previous sickness absences on the risk of future sickness absence irrespective of the length of the sickness absence.
A better understanding of the dynamics of sickness absence and the factors affecting its repetition may be beneficial for understanding the reasons for frequent sickness absences. We used a large prospective cohort sample with accurately recorded dates of absences to examine whether past sickness-absence spells predict new sickness-absence episodes of different lengths. We also examined whether a history of short absence spells increases the risk of future longer sickness-absence spells.
Previous studies examining whether previous absence spells predict future absences have not paid attention to the lengths of such absences. A study examining whether employees who during a reference year had frequent absence episodes or at least one long-term absence spell exceeding 6 weeks have increased risk of sickness absence found that employees with prior long-term absences of more than 6 weeks had an increased risk of recurrent absences.7 Another study using prospective cohort data found that the higher the number of prior sickness-absence spells, the greater the risk of presenting a new one.10 As the median length of absence episodes was 2 days, these results were in practice strongly dominated by short absence spells. Thus, the knowledge of recurrent absence spells of different lengths has so far been fragmentary.
We systematically examined whether short (1 to 3 days), intermediate (4 to 14 days), and long (>2 weeks) sickness-absence spells predicted new sickness-absence episodes of corresponding lengths. The risk of sickness absence increased with the number of preceding absence spells. Nevertheless, the recurrence was higher for longer sickness-absence spells. Longer sickness-absence spells indicate more severe morbidity, which often may be recurrent in nature. Therefore, it is plausible that the relative risk of recurrence is higher for longer absence spells.
Recurrent sickness absences were also more common among men than they were among women. Nevertheless, this should not obscure the fact that in general sickness absence is more frequent among women, in particular short spells.24 In these data, the risk of experiencing a first short sickness-absence episode was nearly twofold among women compared with men. Gender differences diminished for the risk of subsequent absence spells; however, for short spells, the difference even remained in the last category examined (seven or more sickness-absence spells). Previous studies have not generally examined the recurrent sickness absences separately among women and men. A Dutch study found no gender difference in recurrent sickness absences because of common mental disorders.25
In general, the risk of a new absence episode increased constantly on the basis of the number of past absence spells. Nevertheless, in short absence spells, the risk was substantially increased after one preceding absence. In this study population, a medical certificate was required from all employees for sickness absences lasting 4 days or more, whereas absences up to 3 days were possible without consulting a physician. Nevertheless, it is possible that some employee subgroups can have short absence spells without being sick-listed. Therefore, the group of people without any absences may be artificially small, and the gradient may in reality be less steep than was observed. Nevertheless, the risk of new absence still gradually increased after the first absence spell, so this is not a sufficient explanation for the found association.
We also examined whether shorter absence spells predict longer ones. Long-term sickness absence has been shown to indicate chronic morbidity and to predict early retirement and mortality.5,6 Long-term sickness absence can thus be considered a well-established indicator for ill-health. By contrast, the role of short sickness absence as a health indicator is questionable. Almost everyone suffers minor health problems from time to time and has to be absent from work for a day or two. Furthermore, short sickness absence could be assumed to relate to factors other than health. Because diagnoses and medical reasons are typically available only for long absence spells, the reasons for short sickness-absence spells are less well known. Shorter absence spells are often possible without consulting a physician and if the reason for the absence is recorded, these may be less reliably registered than are longer absence spells that require a medical visit. Nevertheless, we found that shorter absence spells strongly predicted subsequent longer spells. The found associations were again fairly linear, and even though preceding short absence spells were already a predictor of longer ones, these associations were even stronger for intermediate spells. This suggests that short absence spells are not trivial for health. It is possible that people may try to cope with emerging illness by having short sickness-absence spells but if they do not properly recover, a longer absence episode will follow.
In previous studies examining the recurrence of sickness absences, the question about short absence spells predicting long ones has not been posed. A study from the Netherlands showed that frequent sickness absences (at least four absence spells in a year) predicted uninterrupted sickness-absence spells of at least 6 weeks.7 These frequent absence spells could have been long or a combination of short and long spells. Nevertheless, as the length of the frequent absence spells was not determined, they are likely to have been predominantly short. This finding thus is in agreement with ours.
The interactions with working conditions and health behaviors were further analyzed to examine whether they modify the risk of recurrent sickness absence. The results suggest that these factors do not markedly modify the recurrence of sickness absences.
The strengths of this study include its large study population and prospective study design. Sickness-absence data were derived from a register and, because they are used as a basis for salary payments and social security benefits, they can be considered to be reliable. In addition, questionnaire data from various covariates were used. These data included municipal employees aged 40 to 60 years at baseline.
We used advanced methods to prospectively track absence during follow-up. Tracking started from the day of answering a questionnaire survey and thus the starting point can be considered to be arbitrary with respect to sickness absence. We were not aware of the full absence history before receiving the questionnaire. However, the frailty can be assumed to some extent to also control for past absence history if frequent absenteeism before the beginning of follow-up is also reflected in absence patterns during follow-up. The covariates were measured only once, and it is possible that they changed during follow-up.
Medical diagnoses for sickness-absence episodes were not available. The recurrence of sickness absence due to different diseases could be assumed to be different. Previous studies have shown that past sickness-absence episodes predict new ones in all diagnostic groups but the strength of the association may differ. A study examining sickness-absence spells longer than 3 weeks among Dutch Post and Telecom employees found that recurrence was highest for sickness absences due to musculoskeletal and mental disorders.26 A study among Brazilian hospital employees found the highest recurrence for sickness absences of any length was caused by mental and behavioral disorders, diseases of the skin and subcutaneous tissue, and diseases related to the nervous system.10
We found that previous sickness-absence spells were predictive of future absence spells. This applied to absence spells of all lengths. The risk of absence spells increased monotonically with the number of preceding absence spells. Recurrence was higher for longer sickness-absence spells than for shorter sickness-absence spells and more common among men than among women. Shorter absence spells were also associated with longer absence spells. The modifying effects of working conditions and health behaviors on these associations were modest.
1. Henderson M, Glozier N, Holland Elliott K. Long-term sickness absence. BMJ. 2005;330:802–803.
2. Allebeck P, Mastekaasa A. Swedish Council on Technology Assessment in Health Care (SBU). Chapter 3—causes of sickness absence: research approaches and explanatory models. Scand J Public Health Suppl. 2004;63:36–43.
3. Beemsterboer W, Stewart R, Groothoff J, Nijhuis F. A literature review on sick leave determinants (1984–2004). Int J Occup Med Environ Health. 2009;22:169–179.
4. Duijts SF, Kant I, Swaen GM, van den Brandt PA, Zeegers MP. A meta-analysis of observational studies identifies predictors of sickness absence. J Clin Epidemiol. 2007;60:1105–1115.
5. Head J, Ferrie JE, Alexanderson K, et al. Diagnosis-specific sickness absence as a predictor of mortality: the Whitehall II prospective cohort study. BMJ. 2008:a1469.
6. Vahtera J, Pentti J, Kivimäki M. Sickness absence as a predictor of mortality among male and female employees. J Epidemiol Community Health. 2004;58:321–326.
7. Koopmans PC, Roelen CA, Groothoff JW. Risk of future sickness absence in frequent and long-term absentees. Occup Med (Lond). 2008;58:268–274.
8. Koopmans PC, Bultmann U, Roelen CA, Hoedeman R, van der Klink JJ, Groothoff JW. Recurrence of sickness absence due to common mental disorders. Int Arch Occup Environ Health. 2011;84:193–201.
9. Roelen CA, Koopmans PC, Schreuder JA, Anema JR, van der Beek AJ. The history of registered sickness absence predicts future sickness absence. Occup Med (Lond). 2011;61:96–101.
10. Reis RJ, Utzet M, La Rocca PF, Nedel FB, Martín M, Navarro A. Previous sick leaves as predictor of subsequent ones. Int Arch Occup Environ Health. 2011;84:491–499.
11. Navarro A, Moriña D, Reis R, Nedel FB, Martín M, Alvarado S. Hazard functions to describe patterns of new and recurrent sick leave episodes for different diagnoses. Scand J Work Environ Health. 2012;38:447–455.
12. Laaksonen M, Aittomaki A, Lallukka T, et al. Register-based study among employees showed small nonparticipation bias in health surveys and check-ups. J Clin Epidemiol. 2008;61:900–906.
13. Karasek RA. Job Content Questionnaire and User's Guide. Revision 1.1. Lowell, MA: University of Massachusetts, Department of Work Environment; 1985.
14. Sarason IG, Sarason BR, Shearin EN, Plerce GR. A brief measure of social support: practical and theoretical implications. J Soc Pers Relat. 1987;4:497–510.
15. Piirainen H, Rasanen K, Kivimaki M. Organizational climate, perceived work-related symptoms and sickness absence: a population-based survey. J Occup Environ Med. 2003;45:175–184.
16. Laaksonen M, Piha K, Martikainen P, Rahkonen O, Lahelma E. Health-related behaviours and sickness absence from work. Occup Environ Med 2009;66:840–847.
17. Laaksonen M, Pitkaniemi J, Rahkonen O, Lahelma E. Work arrangements, physical working conditions, and psychosocial working conditions as risk factors for sickness absence: Bayesian analysis of prospective data. Ann Epidemiol. 2010;20:332–338.
18. Cook RJ, Lawless JF. The Statistical Analysis of Recurrent Events, Chapter 6. New York, NY: Springer; 2007.
19. Andersen PK, Gill RD. Cox's regression model for counting processes: a large sample study. Ann Stat. 1982;10:1100–1120.
20. Navarro A, Reis RJ, Martin M. Some alternatives in the statistical analysis of sickness absence. Am J Ind Med. 2009;52:811–816.
21. Christensen KB, Andersen PK, Smith-Hansen L, et al. Analyzing sickness absence with statistical models for survival data. Scand J Work Environ Health. 2007;33:233–239.
22. Labriola M, Christensen KB, Lund T, Nielsen ML, Diderichsen F. Multilevel analysis of workplace and individual risk factors for long-term sickness absence. J Occup Environ Med. 2006;48:923–929.
23. Lidwall U, Marklund S, Voss M. Work-family interference and long-term sickness absence: a longitudinal cohort study. Eur J Public Health. 2010;20:676–681.
24. Laaksonen M, Martikainen P, Rahkonen O, Lahelma E. Explanations for gender differences in sickness absence: evidence from middle-aged municipal employees from Finland. Occup Environ Med. 2008;65:325–330.
25. Koopmans PC, Roelen CA, Bultmann U, Hoedeman R, van der Klink JJ, Groothoff JW. Gender and age differences in the recurrence of sickness absence due to common mental disorders: a longitudinal study. BMC Public Health. 2010;10:426.
26. Roelen CA, Koopmans PC, Anema JR, van der Beek AJ. Recurrence of medically certified sickness absence according to diagnosis: a sickness absence register study. J Occup Rehabil. 2010;20:113–121.