Sleep and Sickness Absence: A Nationally Representative Register-Based Follow-Up Study

Abstract

Study Objectives:

We aimed to examine various sleep measures as determinants of sickness absence while considering confounders.

Design:

Nationally representative Health 2000 Survey linked with sickness absence data from the Finnish Social Insurance Institution.

Setting:

Finland.

Participants:

Working-aged women (n = 1,875) and men (n = 1,885).

Interventions:

N/A.

Measurements and Results:

Insomnia-related symptoms, early morning awakenings, being more tired during daytime than other people of same age, use of sleeping pills, excessive daytime sleepiness, probable sleep apnea (4 items about snoring/apnea), and reporting that sleep duration varies between different seasons were examined as determinants of sickness absence over a 7.2 year follow-up. Poisson and gamma regression models were fitted. After adjusting age, all examined sleep disturbances except excessive daytime sleepiness were associated with sickness absence among men (RRs 1.3-2.5). Among women, after adjusting for age, insomnia-related symptoms, early morning awakenings, being more tired than others, and use of sleeping pills were associated with sickness absence (RRs 1.4-1.8). After further adjustments for education, working conditions, health behaviors, and objectively measured mental and somatic health, the associations somewhat attenuated but mainly remained. The optimal sleep duration with the lowest risk of sickness absence was 7.6 hours for women and 7.8 hours for men. Although persistence of other health problems could affect the estimates, direct costs due to sickness absence could decrease by up to 28% if sleep disturbances could be fully addressed.

Conclusions:

This study highlights the need for prevention of sleep disturbances and promotion of optimal sleep length to prevent sickness absence.

Citation:

Lallukka T, Kaikkonen R, Härkänen T, Kronholm E, Partonen T, Rahkonen O, Koskinen S. Sleep and sickness absence: a nationally representative register-based follow-up study. SLEEP 2014;37(9):1413-1425.

INTRODUCTION

Sleep disturbances are among the most common public health problems, with serious consequences to subsequent morbidity, work disability, and even mortality.¹ Sleep disturbances are increasingly prevalent among the ageing working-aged adults in Finland as well as elsewhere,^2–4 which highlights the significance of sleep also from the societal point of view related to, for example, vast costs attributable to insomnia and apnea in the workforce.^5,6 Short sleep is also prevalent, although changes in average sleep duration over previous decades have been minor and inconsistent across different countries, in contrast to common assumptions.⁷ Except for insomnia, associations between sleep disturbances and sickness absence have been little studied. Previous evidence on the associations between sleep and work disability is briefly summed here below.

Several previous studies have specifically addressed insomnia as a determinant of subsequent work disability and delayed return to work.^8–14 The evidence derived from these studies consistently suggests that insomnia symptoms are independently associated with subsequent work disability, even after considering potential confounders such as social and work-related factors and prior health status. Furthermore, the associations are similar for all insomnia symptoms, including difficulties initiating and maintaining sleep (nocturnal awakenings and waking up too early in the morning) as well as nonrestorative sleep.¹² However, these studies have been conducted in relatively restricted study populations, such as a specific age group or one employer only. Thus, nationally representative studies focusing on all working-age groups and comprising different employment sectors are needed to confirm the contribution of insomnia to subsequent sickness absence. Inclusion of health-related confounders has also varied. As these confounders are rarely from clinical examination, they have tended to be prone to reporting bias.

Additionally, although the U-shaped associations between sleep duration and health outcomes, including mortality, have been examined in numerous studies,^1,15,16 surprisingly little evidence exists regarding the association between sleep duration and sickness absence. Short and long sleep duration have been connected to sickness absence and disability retirement in some studies,^17–20 but the U-shaped associations tend to be dominated by insomnia^19,20 or sleep duration have not been the main focus,¹⁷ and only crude estimates have been reported. Furthermore, these few studies have focused on middle-aged and ageing public sector employees from specific areas, while sleep duration is largely affected by age and other sociodemographic factors.⁴

Sleep apnea has also been associated with sickness absence and disability retirement.^6,21–26 A previous register-based Finnish study examined the associations between obstructive sleep apnea syndrome and subsequent first sickness absence episode as well as risk of disability retirement.²⁴ As compared to controls, cases had an increased risk of both sickness absence and disability retirement, with the risks being higher among women than men. The findings further confirmed those of an earlier Norwegian study that focused on self-reported apnea symptoms.²⁵ Additionally, in a recent study, separate and joint effects of insomnia and apnea on subsequent sickness absence among Norwegian employees were examined.²³ The results suggest that both apnea and insomnia are independently and jointly associated with sickness absence, but there were no synergistic effects after adjustments. Similar to prior studies on insomnia, the evidence is mainly from cohorts with a narrower age range or focus on public sector employees or a patient population.

Evidence regarding the significance of other sleep disturbances such as excessive daytime sleepiness and seasonal variation in sleep duration to work disability is more limited. One might assume that the prevalence of sleep disturbances varies between lighter and darker seasons. In Nordic studies, sleep quality has been reported to be worse in summer, and in Finland, a deterioration of sleep in either summer or winter was associated with sleep dissatisfaction.²⁷ Earlier studies have, however, yielded conflicting results regarding such variation in sleep.^28,29 It could be still assumed that the risk of sickness absence is affected by the seasonal variation in sleep duration, as the presence of such variation is indicative of mental or other medical conditions.^30,31 However, the association with sickness absence has not been previously studied. Excessive daytime sleepiness has been shown to contribute to the risk of sickness absence, reduced work productivity, and work disability in some studies,^6,21,26 and the risk has been suggested to be pronounced in combination with sleep apnea.²¹ Again, no nationally representative studies have been done.

To our knowledge, the association between sleep medication and sickness absence has not been addressed in any prior study. Previous studies have mainly focused on one or a few sleep disturbances, while we included a diverse set of sleep disturbances simultaneously as determinants of subsequent sickness absence. Moreover, previous studies have not consistently adjusted for various covariates, and these covariates tend to be self-reported. In this study, in addition to subjective health assessment, objective physical and mental health data gathered by a field physician were used for more robust evidence about the significance of sleep to subsequent sickness absence. Previous studies have also typically analyzed specific populations such as middle-aged and ageing employees from public sector or a single employer, while nationally representative studies comprising employees from various workplaces and different employment sectors, as well as all working-aged age groups, are lacking. Previous studies on sleep disturbances and sickness absence have mainly focused on relative risks, with limited opportunities for implications in terms of cost and assessment of concrete increments in the risk of sickness absence among those suffering from disturbed sleep. Thus, more sophisticated methods are needed to shed light on independent effects of various measures of sleep on sickness absence and implications of such effects at workplaces.

The aim of this study was to examine the associations between sleep and sickness absence over a register-based follow-up of seven years. More specifically, the aim was to focus on a wide array of sleep disturbances as determinants of sickness absence while considering a range of key confounders including extensive objective health examination data gathered by a field physician.

METHODS

Data

Baseline data for this study were derived from the nationally representative Health 2000 Survey, where field work took place between September 11, 2000, and March 2, 2001.^32,33 The participation rate was 89%. A 2-stage stratified cluster sampling design was planned by Statistics Finland. The main focus was on key public health problems such as cardiovascular diseases as well as musculoskeletal and mental health, and health care use. Participants were selected to represent the demographic distributions of the Finnish adult population. The main survey was conducted among those 30 years or older, while a separate smaller survey was conducted among those who were 18 to 29 years. Those ≥ 80 years were oversampled with a double sampling fraction to have sufficient number of participants for this group. The 2-stage stratified sample obtained from the Social Insurance Institution comprised all 10,000 participants ≥ 18 years. In addition to the in-home interviews, baseline data were collected using several questionnaires comprising data on sleep, health, working conditions, and covariates. The sample included in this study comprised working-aged participants 30-64 years old at baseline. This age range was selected because the Health 2000 Survey contained health examination by a field physician only in the age group ≥ 30 years, and sickness absence can be examined only among those working-aged. We included those who had been working at any time during one year before participation (1,885 men and 1,875 women). Among men, the mean age was 44.7 years, and among women 44.0 years. More details of the data collection and methods can be found elsewhere³³ and at the project website (http://www.terveys2000.fi/indexe.html).

The study received ethical approval from the Ethics Committee for Epidemiology and Public Health of the hospital district of Helsinki and Uusimaa in Finland. Survey data were combined with register based follow-up data based on signed informed consent forms from the participants.

Sleep Measures

Insomnia-related symptoms were derived from an extensive Symptom Checklist-90 (SCL-90) questionnaire.³⁴ In the questionnaire, the respondents were asked if they had been bothered by disturbed sleep or insomnia. The frequency of these symptoms during previous 30 days was requested, using 5 response alternatives ranging from never to very often. Following previous procedures,⁴ these symptoms were reclassified into 3 groups: no or very rare insomnia-related symptoms, occasional symptoms, and frequent or very frequent symptoms.

In addition to general insomnia-related symptoms or disturbed sleep, the respondents were asked to separately report if they had had nocturnal awakenings or had woken up very early in the morning, and if they had felt exceptional tiredness (“Are you usually more tired during the daytime than other people of your age?”). Those who reported that they never or only sometimes woke up too early in the morning were compared to those who responded that they often or nearly every night woke up too early. Those who reported that they were not more tired than others were compared to those who almost always or at least weekly felt more tired than others, and those who responded “cannot say.”

Use of hypnotics was requested by an item asking whether the respondents had difficulties falling asleep without sleep medication. There were 4 response alternatives ranging from not at all to nearly always. These were reclassified into 3 groups for the analyses: not at all, sometimes, and often or nearly always.

Excessive daytime sleepiness was measured with the Finnish modification of the original Epworth Sleepiness Scale (ESS).³⁵ The respondents were asked about falling asleep or dozing in different situations during daytime. These situations covered sitting and reading, watching television, sitting as a passenger in a car for one hour without a break, lying down to rest in the afternoon, sitting and talking to someone, sitting quietly after a lunch, and in a car while stopped at a traffic light. In the original ESS, the answering modes express the probability or chance of dozing in different situations (e.g., “slight chance of dozing”), whereas in the modified Finnish version the answering modes express the actual frequency of dozing in different situations (e.g., “seldom”). Each item had 4 response alternatives: I never fall asleep, I seldom fall asleep, I fall asleep quite often, and I fall asleep nearly always. The respondents were further instructed to answer “I never fall asleep,” if they think they would feel tired in a certain situation, but would not fall asleep. The sum measure was dichotomized using the highest quartile as a cutoff point to reflect sleepiness. Analyses were also repeated using a continuous measure, i.e., a sum score of all responses (data not shown).

Probable sleep apnea was based on responses to 4 items asking about snoring and breathing at night. First, respondents were asked to report whether they snored and to ask others if they did not know. Those who responded “yes” were asked to answer more specific questions related to sleep apnea. These questions asked about the frequency of snoring (once a month or rarer, 1-2 nights a week, 3-5 nights a week, and almost every night/every night), and loudness of snoring as well as breaks in snoring. Lastly, the respondents were asked whether they themselves or others had noticed breaks in their breathing at night while they were asleep, and how frequent those breaks were (similar response alternatives as those for the frequency of snoring).

Seasonal variation in sleep duration was assessed by a question asking if respondents had noticed whether their sleep duration varied between different seasons of the year. Four response alternatives were given: no change, some variation, clear variation, and substantial variation. Those who reported no change were used as a reference group, and the 2 upper categories were combined for the analyses.

Sleep duration was measured by a single item asking how many hours the respondents slept on average during 24 hours. No time frame was specified, and the response was requested as full hours without predefined categories.

Covariates

Key covariates of sleep and sickness absence were included for more robust evidence. Age was adjusted for in all the analyses, as both sleep disturbances and sickness absence vary as a function of age. Additionally, sector of employment was taken into account, alongside educational level derived from the Statistics Finland register data. Further covariates comprised a range of working conditions, health behaviors, obesity, and a variety of indicators of mental and physical health.

Regarding working conditions, physically demanding work, chemical and physical hazards, threat of mental violence or bullying, job insecurity, job demands, job control, job respect at working community, no support at work, negative work climate, and possibilities for developing educational skills at work were included. These data were collected by interviews and self-administered questionnaires. Detailed descriptions of the measures of working conditions will be published in a separate report (unpublished data).

Smoking status was derived from the health interview, while heavy drinking was based on self-reported alcohol consumption derived from the questionnaire (grams per week). Physical exercise was also derived from the questionnaire. Exercising ≥ 4 times per week for ≥ 30 minutes at a time was defined as adequate. Obesity was calculated from measured height and weight (weight/height²).

Arterial hypertension was based on the field physician's diagnosis at health examination or derived from the hospital discharge register or Social Insurance Institution's (SII) register covering all entitlements to special reimbursement for medication. Coronary heart disease, specific findings in electrocardiogram, stroke, cardiac failure, arrhythmia, and other cardiovascular diseases were also based on the field physician's diagnosis, hospital diagnosis, or reimbursed medication. Diabetes was also based on field physician's diagnosis, fasting blood glucose > 11.1 mmol/L, hospital diagnosis, or reimbursed medication. Pulmonary disease was defined as forced expiratory volume in one second percentage, FEV% < 70%. Musculoskeletal disorders included low back syndrome, chronic neck and shoulder syndrome, knee osteoarthritis, hip osteoarthritis, and musculoskeletal injuries, which were again all based on the field physician's diagnosis. Other somatic diseases comprised hyperlipidemia, chronic skin diseases, cataract, and glaucoma as diagnosed by the field physician. Mental disorders included alcohol disorder, anxiety, or depressive disorder during the past 12 months. Mental disorders were based on WHO's composite international diagnostic interview (CIDI)³⁶ and general health questionnaire (GHQ-12).^37–38 GHQ-12 was included as a continuous measure, i.e., sum of responses to each question.

Sickness Absence

Data on sickness absence were derived from the Social Insurance Institution of Finland. These register data comprise all sickness absence spells > 10 days, as for shorter spells, no sickness allowance can be applied for. The register data were prospectively merged with the survey data. In addition, data comprising any other absence such as unemployment were also derived from the SII, while retirement data were derived from the Finnish Centre for Pensions and mortality data from the Statistics Finland. Follow-up started from the date the respondents participated in the baseline health survey and lasted until the end of 2008. Those who exited the labor market due to unemployment, retirement, or death were censored. Thus, the average follow-up was for men 7.16 years (range: 0.003–8.375) and for women 7.18 years (range: 0.002–8.342).

Statistical Analyses

Poisson and gamma regression analyses were conducted to examine sleep measures as determinants of sickness absence days. The results are reported as rate ratios (RR) and arithmetic mean ratios for the Poisson and gamma regression models, respectively. Additionally, Härkänen and Kaikkonen recently developed a new method to predict adjusted average sickness absence days per working year (DWY, unpublished data). This method allows for a deeper understanding of the relative importance of different work-related and health-related factors on the associations between sleep and sickness absence. The burden due to sickness absence depends both on the number and length of the episodes. Some factors can be associated with only one of these components, and some with both of them. In the latter case, the associations can amplify the burden or, if the associations with the components are positive and negative, even cancel each other out. Using the method, both of the components and their joint effects on sickness absence can be analyzed.

The survey versions of the Poisson and the GLM, which was used for the gamma regression models using the gamma family and log link options, were used for the regression analyses on the number and the lengths of the sickness absence episodes, respectively. The margins procedure was applied to estimate the predicted margins (PM)³⁹ of the number and the lengths of the sickness absence episodes in the sleep groups. For each individual, the expected number of sickness absence days per working year (DWY) was defined as the product of the expected number of sickness absence episodes and the expected length of an episode using the predicted values based on the regression models as in the predicted margins above. Multiple imputation was used to take into account item missing (20 datasets). Multiple imputation including all the covariates was conducted with the Stata software, version 11 using the Markov chain Monte Carlo Method. The imputation model contained all covariates. An imputation model, which also contained the outcome variables, appeared to have virtually no effect on the results. The imputed binary covariate values were transformed back to the original binary scale using the midpoint of the values of the binary covariate as the threshold. Complex sampling design³³ was taken into account in all the analyses, and the unit non-response was accounted for using post-stratification weights.⁴⁰

Association between sleep duration (hours) and subsequent sickness absence were examined by conducting 2nd degree polynomial linear regression analyses. In Figures 1 and 2, rate ratios and their 95% (nonlinear) confidence intervals (CI) are displayed. The curves further show the optimal sleep duration in these data with the lowest risk of sickness absence.

Figure 1.

The association between sleep duration and sickness absence among men. 7.76 h the lowest risk. RR, rate ratios; 95% CI, 95% confidence interval.

Figure 2.

The association between sleep duration and sickness absence among women. 7.63 h the lowest risk. RR, rate ratios; 95% CI, 95% confidence interval.

Hypothetical calculations regarding the direct costs of sickness absence to the state and employers, and estimated savings achieved if sleep disturbances could be addressed as modelled in this paper, were made using one-year values for both genders in sickness absence benefits based on the statistics of SII,⁴¹ the size of the employed 30-64 year old target population from Statistics Finland,⁴² the educational structure of the study population, and the direct costs to employers.⁴³ In calculations gender, age, and educational structure were accounted for. Only the hypothetical direct costs of sickness absence are reported, with likely variance between individuals depending on the severity of any remaining illness and disabilities, for example.

RESULTS

Descriptive

Different sleep disturbances as well as short sleep were relatively prevalent in the examined cohort representative of Finnish working-aged women and men (Table 1). Men reported more frequent occasional use of sleeping pills and probable sleep apnea as well as tended to report shorter sleep; while women reported more frequent the seasonal variation in sleep duration and longer sleep. Some gender differences were also found for other indicators of sleep disturbances.

Table 1.

Descriptive statistics: distribution (%) of sleep disturbances and sleep duration among men (n = 1,885) and women (n = 1,875).

Sleep and Sickness Absence

The associations (relative risks) between sleep disturbances and subsequent sickness absence are displayed in Table 2 for men and Table 3 for women, while the absolute differences in sickness absence days per working year are displayed in Table 4 for both genders. After adjusting for age, insomnia-related symptoms showed strong associations with sickness absence among men (Model 1, Tables 2 and 4). A gradient in the associations was also observed, with the most frequent symptoms showing the strongest associations with sickness absence. Absolute difference in sickness absence days per working year (DWY) was 5.58 when comparing those with no or rare insomnia-related symptoms to those with frequent symptoms (Model 1, Table 4). The associations were mainly unaffected by adjustment for education (Model 2), and attenuated but remained after adjusting for health behaviors and health (Model 3). The effects of working conditions (Model 4) on the associations were similar to those of the health behaviors and health.

Table 2.

Associations of sleep disturbances with subsequent sickness absence among men (rate ratios and their 95% confidence intervals from Poisson regression analyses).

Table 3.

Associations of sleep disturbances with subsequent sickness absence among women (rate ratios and their 95% confidence intervals from Poisson regression analyses).

Table 4.

Sickness absence days per working year (DWY) among men and women.

Early morning awakenings were similarly associated with sickness absence among men in a dose-response manner. The age-adjusted associations (Model 1, Tables 2 and 4) were stronger and attenuated particularly after adjustment for health behaviors and health and for working conditions. Absolute difference in sickness absence was 5.61 DWY after adjusting for age between those who reported early morning awakenings almost every night and those who did not suffer from such awakenings (Model 1, Table 4).

Corresponding associations were also found for reporting being more tired than others and subsequent sickness absence (Tables 2 and 4). After adjusting for age, men who were more tired than others had 8.96 days of sickness absence per year, as compared to 4.81 days among those who reported that they were not more tired than the others and 7.96 days among those who did not know if they were more tired than the others (Model 1, Table 4). All the associations remained throughout the modeling, and even those who did not know if they were more tired than others had a higher risk for subsequent sickness absence.

After adjusting for age, reported use of sleeping pills was associated with subsequent sickness absence among men (Model 1, Tables 2 and 4). The associations were stronger with the more frequent the use of pills. Regarding days of sickness absence per working year, those who reported using sleeping pills often or almost every night had 8.14 days more sickness absence per working year than those who did not use sleeping pills. Adjustment for education had again a negligible contribution to the examined association (Model 2). The associations somewhat attenuated but remained when adjusting for health behaviors and health (Model 3) as well as for working conditions (Model 4).

No associations could be confirmed between excessive daytime sleepiness and subsequent sickness absence among men (Tables 2 and 4). Probable sleep apnea was, in turn, weakly associated with sickness absence among men after adjusting for age (Model 1). This association reduced after further adjustments particularly after adjusting for health behaviors and health (Model 3), but remained after adjusting for working conditions (Model 4). Seasonal variation in sleep duration was mainly unassociated with sickness absence, although a weak association was observed only in the age-adjusted model (Model 1, Tables 2 and 4).

Among women, the patterns of the associations were similar to men, although effect sizes tended to be weaker (Tables 3 and 4). Thus, insomnia-related symptoms, early morning awakenings, feeling more tired than others, and use of sleeping pills were associated with subsequent risk of sickness absence among women after adjusting for age (Model 1). As among men, the associations attenuated after adjustments, particularly after adjusting for health behaviors and health (Model 3). Furthermore, those reporting frequent insomnia-related symptoms had 5.69 more days of sickness absence per working year than those not reporting such symptoms (Model 1, Table 4). The corresponding difference in days of sickness absence per working year was 2.27 for early morning awakenings almost every night, 5.48 for being more tired than others, and 3.16 for the use of sleeping pills often or almost every night, as compared to those not reporting such sleep disturbances or use of sleeping pills. Excessive daytime sleepiness, probable sleep apnea, and seasonal variation in sleep duration were unassociated with sickness absence among women.

Optimal sleep duration among men was 7.76 hours per day (Figure 1). The association between sleep duration and sickness absence tended to be U-shaped, with those who slept less or more than 7.76 hours having a higher risk of sickness absence. Those reporting the optimal sleep duration had on average 5.93 days of sickness absence per year, while the corresponding figures in the extreme groups of very short and long sleep duration were 14.83 DWY and 12.01, respectively (Table 5). Those sleeping ≥ 11 h had 25.17 days of sickness absence per year, but the group was too small for separate analyses.

Table 5.

The association between sleep duration (hours) and sickness absence days per working year (DWY) among men and women.

Among women, optimal sleep duration was 7.63 hours per day (Figure 2). Similar to men, the association between sleep duration and sickness absence tended to be U-shaped among women. Those reporting the optimal sleep duration had on average 7.64 days of sickness absence per year, while the corresponding figures in the extreme groups of very short and long sleep duration were 13.09 DWY and 12.21 DWY, respectively (Table 5). As among men, the 11-h sleepers had the highest number of sickness absence days per working year (19.59), but also this group was too small for separate analyses.

Finally, using the estimates from the models further analyses we were able to calculate difference in days of sickness absence between good sleepers and those suffering from sleep disturbances that information could be used to (roughly) estimate the hypothetical direct costs of sickness absence that are attributable to sleep disturbances, and estimate the cost due to sickness absence that could be saved per year if sleep disturbances could be fully addressed, as modelled in this paper. Thus, if sickness absence days among those reporting frequent insomnia-related symptoms were on the same level as those not reporting such symptoms, approximately 501 million euros among men and 341 million euros among women (together 17% decrease of overall costs of sickness absence) could be saved. Corresponding figures for frequent early morning awakenings would be approximately 428 million euros among men and 260 million euros among women (together 14% decrease of overall costs), for tiredness approximately 703 million euros among men and 751 million euros among women (together 28% decrease of overall costs), for frequent use of sleeping pills approximately 154 million euros among men and 252 million euros among women (together 8% decrease of overall costs), for excessive daytime sleepiness (highest quartile) approximately 95 million euros among men and 87 million euros among women (together decrease of 4% of overall costs), and for probable sleep apnea 4 million euros among men and 260 million euros among women (together 5% decrease of overall costs).

Regarding seasonal variation, results differed between men and women. Among men, 237 million euros among men could be saved, if sickness absence levels among those with seasonal variation in sleep duration were on the same level as those not reporting such variation. Among women, those with no variation in sleep duration had the highest number of sickness absence episodes. Finally, from variation of sleep duration 109 million euros among men and 99 million euros among women (together 4% of overall costs) could be saved yearly if those with short or long sleep reached the level of sickness absence of those sleeping on average 8 hours. That figure corresponds to 0.1% to 0.8% (depending on the sleep disturbance) of the gross domestic product in Finland, and 4% to 28% potential of decreasing the costs caused by sickness absence overall.

DISCUSSION

Main Findings

This study sought to examine the associations between various sleep disturbances, sleep duration, and subsequent sickness absence in a population-based cohort representative of working-aged Finnish women and men. In addition, to focus on various key sleep-related measures, this study had an extensive set of confounders included to establish independent effects of sleep on an objectively measured work disability outcome, i.e., sickness absence. Insomnia-related symptoms, waking up too early, being more tired than others, and using sleeping pills were most consistently associated with sickness absence, while excessive daytime sleepiness, probable sleep apnea, and seasonal variation in sleep duration were mainly unassociated with sickness absence. However, the associations varied somewhat between women and men and were attenuated, particularly after health behaviors and health and working conditions had been taken into account. Sleep duration showed U-shaped associations with sickness absence, with the risk increasing towards the extreme ends of sleep duration distribution, i.e., among those sleeping < 6 h or > 9 hours. The results suggest notable potential savings if sleep disturbances and subsequent sickness absence could be prevented.

Interpretation

Although sleep disturbances^4,44 and sickness absence^45,46 tend to be more prevalent among women than men, sleep disturbances were particularly strongly associated with the risk of sickness absence among men. Overall, associations between insomnia-related symptoms and sickness absence are in line with previous evidence.^9,11 These results thus confirm and add generalizability to those from earlier studies that have focused on a particular age group or a single employer. Furthermore, these results show that the associations between insomnia and sickness absence are not sensitive to the selected measure, as similar results have been produced using various single and multi-item measures.

Although the association between sleep medication and subsequent sickness absence has not been studied previously, effects of sleep medication on health and mortality might be assumed to reflect similar patterns. The associations in this study were, however, weak and only partly in line with the associations observed earlier between sleep medication and other health outcomes.⁴⁷ Additionally, as those with the most severe and persistent insomnia-related symptoms probably use more medication,⁴⁸ the association between sleep medication and sickness absence probably partly reflects frequent insomnia. The recent results regarding the association between sleep medication and mortality⁴⁷ have also been criticized and have caused controversy, as the associations similarly attenuated after considering health-related baseline risk factors in accordance with our findings.^49,50 It is possible that medication use reflects poor health and thereby subsequent mortality. Alternatively, severe sleep disturbances could lead to poor health, suggesting that adjusting for health is not applicable.

In contrast to earlier evidence,^6,21,26 excessive daytime sleepiness made little contribution to the risk of sickness absence among women and men. However, difference in validity between the original and our measure needs to be considered. While our questionnaire was focused on the actual frequency of dozing in different situations, the original questionnaire asks the respondents to estimate their likelihood of dozing or falling asleep.³⁵ As we have suggested earlier, the results derived using our measure could be easier to interpret, and the concrete frequency could be seen as a more direct estimate of daytime sleepiness.⁵¹ Another difference was that in contrast to the original item about likelihood of dozing when sitting quietly after a lunch without alcohol, our version only asked whether respondents fell asleep when sitting quietly after a lunch, without mentioning alcohol. The effect of this difference on the score is likely negligible, as it is uncommon to drink alcohol at lunch in Finland. Furthermore, the respondents were all working, and thus they can be assumed to have had lunch at workplace without alcohol.

Our study further failed to find consistent associations between probable sleep apnea and sickness absence, although the effects have been reported earlier and have been particularly strong among women.^24,25 Lack of associations could again be partly due to our measures, as we only had self-reported proxy measures of probable sleep apnea and ESS available. In line with earlier evidence, however, our sensitivity analyses showed that probable apnea is associated with sickness absence among younger women but not older women (see supplemental material for details). It could be assumed that as our measure of probable apnea asked snoring, this could have produced more false positive responses among older women. Among younger women, snoring is rarer and could be related to clinical apnea, while other reasons could contribute to the increased snoring with ageing. This could dilute the associations and explain differing results between age groups.

High ESS scores were originally found among those suffering from moderate or severe obstructive sleep apnea syndrome (OSAS), idiopathic hypersomnia, or narcolepsy.³⁵ Alternatively, other methodological issues or differences between cohorts might explain the discrepancy and inconsistent evidence. We focused on a large population-based cohort, representative of all working-aged adult Finns, while other studies have focused on patients or had an otherwise narrower study population. Prevalence of probable sleep apnea was also relatively low in this cohort compared to other sleep disturbances, such as insomnia-related symptoms. The prevalence is, nonetheless, higher in this population than the original study sample (8%), which included those outside the labor market.⁵² The gender difference in the cost of sickness absence due to probable apnea likely reflects the nonexistent difference in days of sickness absence between men who did and did not report probable apnea. In contrast, women with probable sleep apnea had around six days more sickness absences per working year than women who did not report apnea. Additionally, as our apnea measure focused on snoring, it could have lacked specificity particularly among men, and this could also explain the findings. Thus, among men, snoring is likely overall more prevalent than actual clinical apnea syndrome. One might assume that by being able to better distinguish those with the clinical apnea, the cost of sickness absence would become apparent and increased among both men and women.

Although many of the observed associations were attenuated after adjustments, these results nonetheless showed independent effects of several indicators of disturbed sleep on sickness absence. Among men, early morning awakenings almost every night and being more tired than others were associated with a significant increase in workdays lost due to sickness absence. The former is a cardinal symptom of depression, while the latter is unspecific and associated with a range of medical and nonmedical conditions. It is of note here that the association of these early morning awakenings remained among men even after adjustment covering the most common mental and somatic disorders. Among women, occasional insomnia-related symptoms and occasional use of sleep medication were associated with increased number of workdays lost due to sickness absence.

The health-related confounders were based on register data; while the inclusion of these measures is a clear strength, we cannot rule out that we might have overadjusted effects. As repeated measurements of sleep were unavailable and sleep disturbances tend to be recurrent and persistent,^53–55 some baseline health problems might have been caused by sleep disturbances or comorbid with them. Therefore, adjusting for such conditions could lead to overadjustment and subsequent in the associations, if these covariates were on the path between sleep disturbances and sickness absence.⁵⁶ For these reasons, we preferred not to include results after adjusting simultaneously for all of the confounders. Our sensitivity analyses suggest, however, that the associations largely attenuate after a full adjustment; however, among men early morning awakenings and being more tired than others, and among women insomnia-related symptoms, being more tired than others, and use of sleeping pills remained associated with sickness absence (data not shown). The results of heterogeneous models with a large number of different covariates are, nonetheless, difficult to interpret, and in addition to further increasing the risk of overadjustment, the model fit can be questioned.

The association between sleep duration and sickness absence has been little studied. Our results regarding a U-shaped association between reported sleep length and sickness absence are in line with an earlier study among middle-aged public sector employees from Finland,¹⁷ where those who reported that they slept 7.5 hours had the lowest risk of sickness absence. A recent study among middle-aged public sector employees showed that the association is dominated by insomnia.⁵⁷ Although the evidence regarding the association between sleep duration and sickness absence is limited, numerous studies have shown the U-shaped association between short and long sleep and various health-related outcomes as well as mortality.^16,58 Additionally, few studies have focused on disability retirement, and found some evidence to support that short and long sleep are associated with an increased risk of work disability.^18,19

Finally, it is well known that insomnia and apnea result in huge costs at workplaces, not only due to sickness absence but also because of lost work productivity and accidents.^6,59,59 This study provided novel evidence in the Finnish population showing that preventing sleep disturbances would result in notable reductions in the direct costs of sickness absence at workplaces, if days of sickness absence and health-related factors were similar to those among good sleepers.

Methodological Considerations

This study had some further limitations and strengths to be acknowledged here. First, only self-reported data on sleep were used. However, none of the previous studies have been able to simultaneously focus on a range of sleep disturbances and sleep duration, and objective measurements are difficult to obtain in large epidemiological surveys. Because data are self-reported, we cannot rule out bias. For example, those reporting insomnia might have underestimated their sleep duration.^18,19 More detailed methodological limitations related to our measures of sleep are summed here. In order to obtain more accurate estimates, international definitions of insomnia would have been preferable, such as those based on the Diagnostic and Statistical Manual of Mental Disorders.⁶⁰ Nonetheless, the item on insomnia-related symptoms has been shown to have relevant psychometric properties in several studies, and even self-reported single items have been shown to have predictive value for various physical and mental health outcomes.^61–63 Additionally, as the associations with sickness absence were similar to those found for more specific, multi-item, and validated measures,^9–11 this suggests that our measure at least broadly captured those suffering from insomnia-related symptoms. As the ability of a single item to discriminate those suffering from insomnia from good sleepers is likely poorer than multi-item measures,^64,65 one might assume that the results would be stronger if we had a more robust multi-item measure for insomnia. We further need to acknowledge that the question about “being more tired than others” has not been validated. For example, those with clinical insomnia typically are more tired— although not sleepy—during daytime, but other sleep disturbances could also cause tiredness. The question was included to capture this exceptional tiredness, and it reflects respondents' own perception and comparison to other people of same age.

Sleep medication was an indirect measure, and there was no opportunity to separate prescribed medication and over-the-counter drugs; also, the question did not specify any hypnotic classes. However, the item likely distinguished those who had used sleep medication from those not using any medication to aid their sleep, although the actual frequency of hypnotic use was unavailable.

Because the ESS items were not identical to the original ones,³⁵ the originally suggested cutoff points were not applicable as such, and the results are not directly comparable to earlier studies. We conducted sensitivity analyses and examined the ESS as a continuous measure as well as using different cutoff points, for example focusing on the more severe symptoms only (data not shown). The association was weak in all the analyses, suggesting that the modest impact of sleepiness on sickness absence is not sensitive to the selected cutoff point. As our measure of apnea is only a screening instrument based on questions on snoring and its dimensions, we preferred to use the term “probable sleep apnea.” The measure has also been used similarly in previous studies.⁵² Although sleepiness or non-restorative sleep dimensions were not included in our apnea measure, we preferred not to combine two different measures (excessive daytime sleepiness and probable sleep apnea). Moreover, the correlation between ESS and probable apnea was weak (0.144), suggesting no clear association between these measures of sleep disturbances. This is not surprising because it is well known that daytime symptoms of OSAS (including excessive daytime sleepiness, fatigue and unintentional sleep episodes) correlate poorly with the severity of OSAS as measured by the frequency of respiratory breathing events during sleep (AHI index).⁶⁶ In addition, severe cases of apnea are likely rare in this cohort of working people, which could explain the lack of an association. Thus, only those with a very severe apnea could be likely to fall asleep in the examined social situations. Our classification of probable apnea was based on snoring and related breathing resistance, and although this measure likely was sensitive (those with apnea snore), it could be assumed to produce false positives, since not all those who snore have clinical apnea. To allow comparability to previous (and future) studies, we preferred to retain the original measures intact.

Seasonal variation in sleep duration could be seen distinct from the other determinants of disturbed sleep. It has been little studied in population-based health examination surveys, and as our idea was to have a comprehensive view on a range of sleep-related phenotypes, we included this measure in the analysis. It needs to be noted that the main focus regarding this determinant is on major variation, while minor variation in sleep duration between seasons is regarded as a normal phenomenon and is expected to be found among healthy people. If there is major variation that is perceived as a severe problem, it likely relates to mood disorders.

Finally, sleep duration was measured with a single item without the ability to separate for example workdays from weekends or other leisure time. While we acknowledge that sleep diaries and other detailed measured would provide a more accurate picture of the actual sleep time, collecting such data is not feasible in a large-scale epidemiological health survey. As described in the Introduction, sleep duration measured with similar single items has predicted morbidity and even mortality in numerous previous studies over decades, and it is also associated with work disability.^1,19 Thus, despite the limitations, its use can be seen as justified following the procedures from the previous studies.

Second, diagnosis-specific sickness absence could not be examined, and shorter spells were also unavailable. Further studies could aim to disentangle whether the examined sleep disturbances similarly contribute to the risk of sickness absence due to mental and somatic conditions. However, earlier studies on short and long sleep duration and insomnia suggest that they are a risk factor for disability retirement due to both mental and physical health, although the risks are higher for disability retirement due to mental disorders.^12,19 Further studies could also aim to confirm whether ESS and apnea contribute to shorter sickness absence spells, as these data only comprised medically confirmed longer spells. It could be that the effects regarding seasonal variation in sleep duration are only observed for shorter periods of absence.

Third, as we focused on those continuously employed, we cannot rule out that healthy worker effect contributed to our findings. Thus, those with poorest health were likely out of labor market already at baseline,⁶⁷ as work-related covariates and sickness absence outcome concerned only those who were working at baseline and remained employed during follow-up.

Fourth, although we took into account age in all the analyses, one might assume that the associations could differ between younger and older participants, as sleep disturbances and sickness absence increase with age. We therefore conducted sensitivity analyses to examine all the associations between sleep disturbances and sickness absence in two age cohorts (30-44 and 45-64 years). All the models were fitted including the same confounders as the reported ones, and the analyses were done for both women and men. The patterns of the associations were broadly similar, although the magnitude somewhat varied. Due to the lower numbers in these stratified models, the results are also more unstable. As referred earlier in the discussion, a noteworthy difference was that there was a strong association between probable sleep apnea and sickness absence in younger women (RR 1.98; 95% CI 1.24-3.17), while a corresponding association was nonexistent for the older women (RR 0.99; 95% CI 0.66-1.49). No such age difference was found among men, and thus these age stratified analyses did not help shed light on the gender difference in the cost of sickness absence due to probable apnea. These four additional tables with full modeling, relative risks, and their 95% confidence intervals are available in the supplemental material.

Fifth, comorbidity and further focus on health could affect the results, and age might play a further role in those questions as well. However, as we have included a variety of somatic and mental health conditions and sleep disturbances, such an analysis would be very complicated, and also beyond the focus of this paper. Thus, we examined the association between a variety of sleep disturbances and sickness absence, while considering prior health problems as part of the key confounders. Moreover, as health status and sleep were measured at the same time, it was not possible to distinguish whether baseline sleep disturbances had caused the health problems or whether disturbed sleep was due to or comorbid with poor health. However, being able to consider health prior to sickness absence provides more robust results.

There were also several further strengths in this study. First, we were able to focus on all key indicators of sleep disturbances as well as sleep duration. Second, the data were nationally representative and comprised participants from all employee groups, different employment sectors, as well as women and men. Third, sickness absence data were derived from national registers and are thus objective and comprised complete days of reimbursed sickness absence. Furthermore, sophisticated methods provided evidence about the absolute and relative significance of sleep disturbances to subsequent risk of sickness absence, as well as evidence regarding cost. Fourth, a range of key confounders of the association between sleep and sickness absence were considered, and thus the results provide more robust evidence about the independent effects of sleep disturbances and short and long sleep on subsequent sickness absence. The ability to extensively control for example for several working conditions, social factors, health behaviors, and health was a special strength, and enabled us to produce novel evidence about robustness of the associations, as the included factors are strongly connected with both exposures and outcome.^4,48,68–70 Furthermore, the included health-related confounders covered both somatic and psychiatric conditions and were mainly derived from objective register data or based on field physician's diagnosis and were therefore less prone to reporting bias.

CONCLUSION

The results highlight the need to focus on prevention of sleep disturbances and promotion of optimal sleep duration to help employees maintain their work ability and prevent sickness absence among both women and men. The savings achieved by such prevention, if successful, would also be notable. In particular, efforts should be made to prevent insomnia-related symptoms and need for sleep medication; contribution of other sleep disturbances such as daytime sleepiness, apnea, and seasonal variation in sleep duration to sickness absence appears to be minor in the Finnish working aged population.

DISCLOSURE STATEMENT

This was not an industry supported study. The study was supported by the National Institute for Health and Welfare, the Academy of Finland (#128180) and the Finnish Work and Environment Fund (#108305). RK has been supported by the Doctoral Programs in Public Health (DPPH). This work was performed at the Finnish Institute of Occupational Health, and at National Institute for Health and Welfare. The authors have indicated no financial conflicts of interest. There was no off-label and investigational use.

SUPPLEMENTAL MATERIAL

Table S1

Associations of sleep disturbances with subsequent sickness absence among men aged 30-44 years (n = 1,046).

Table S2

Associations of sleep disturbances with subsequent sickness absence among men aged 45-64 years (n = 927).

Table S3

Associations of sleep disturbances with subsequent sickness absence among women aged 30-44 years (n = 953).

Table S4

Associations of sleep disturbances with subsequent sickness absence among women aged 45-64 years (n = 922).