Self-reported sleep duration in Finnish general population

Authors


Erkki Kronholm, Department of Health and Functional Capacity Laboratory for Population Research Peltolantie 3, National Public Health Institute, FI-20720 Turku, Finland. Tel.: +358 2 331 6718; fax: +358 2 331 6720; e-mail: erkki.kronholm@ktl.fi

Summary

Self-reported short or long sleep duration has been repeatedly found to be associated with increased mortality and health risks. However, there is still an insufficient amount of detailed knowledge available to characterize the short and long sleep duration groups in general population. Consequently, the underlying mechanisms potentially explaining the health risks associated with short and long sleep duration are unclear. In the present study, the self-reported sleep duration in a sample of Finnish general population was studied, and its possible associations with such factors as self-perceived health, sociodemographic characteristics, lifestyle, sleep difficulties and daytime concomitants were analyzed. In particular, an effort was made to define mutually statistically-independent determinants of sleep duration. In the Finnish Health 2000 Survey, a representative sample of 8028 subjects of 30 years of age or older and a sample of 1894 subjects of 18–29 years of age were invited to take part in the health interview and health examination. The participation rate of the study was over 80%. The most important and statistically-independent determinants of short and long sleep duration were gender, physical tiredness, sleep problems, marital status, main occupation and physical activity. However, in the multivariable model they only accounted for approximately 16% of the variance in sleep duration in short and long sleepers, suggesting multiple sources of variance. The present study also suggests a dose–response like relationship between the sleep duration and many of its determinants within both short and long sleepers. A more detailed analysis of the clinical status of the short and long sleep duration groups is needed to evaluate the possible importance of these findings for health risks associated with sleep duration.

Introduction

Several epidemiologic prospective sleep studies suggest that self-reported sleep duration of less than 7 h or more than 8 h is associated with increased mortality and decreased health. The association has been shown at least in nine independent population or community-based random samples and in two somehow selected samples in five different countries (for references, see Youngstedt and Kripke, 2004b). Risk groups determined by the self-reported sleep duration make up 14–22% of the adult population for short sleepers and correspondingly, 7–10% for long sleepers, emphasizing the public health perspective of the issue.

Initially, researchers focused mainly on the detrimental effects of short sleep, and experimental sleep deprivation as its supposed model, but recently, long sleep has received increasing attention (see e.g. Foley, 2004; Youngstedt and Kripke, 2004a,b). However, there is not yet any consensus on the interpretation of the association between the short and long sleep duration and impaired health or longevity. Some studies have shown the association to exist for men only (Burazeri et al., 2003; Huppert and Whittington, 1995; Kojima et al., 2000). These studies are, however, relatively small leaving a possibility that the lack of significant association among women may be due to insufficient statistical power. In two samples the association disappeared after the adjustments for relevant confounders (Mallon et al., 2002; Pollak et al., 1990). The association between sleep duration and health also depends on age (Hublin et al., 2004). There are several more or less speculative hypotheses suggesting various mechanisms to explain the association under inspection (see e.g. Gottlieb et al., 2005; Taheri et al., 2004; Youngstedt and Kripke, 2004b).

Sleep duration can be considered as a lifestyle factor that is modulated and influenced by a range of background factors, or a factor determined by the underlying health condition and genotype of an individual. Sleep duration can also be considered as a factor reflecting one of the vital functions of the human body, and thus influencing bodily well-being and longevity.

Although there are some studies attempting to characterize subjects who belong to short or long sleep duration groups (Bliwise et al., 1994; Grandner and Kripke, 2004; Kojima et al., 2000; Kripke et al., 2002), several questions still remain open, such as the following: Do both short and long sleepers report similar types of sleep complaints? Do any sociodemographic factors explain why females are over-represented among long sleepers? Are depression and sleep apnea associated with long sleep? Are short sleepers actually sleep deprived?

There is a clear need to characterize, in general population, subpopulations that differ by their sleep duration. The proportions of ‘natural’ short and long sleepers in a population are unknown. In these groups of individuals, the sleep–wake patterns are likely to be genetically determined (De Castro, 2002), especially, by the circadian pacemaker (Aeschbach et al., 2003), making the sleep pattern a stable trait-like characteristic of an individual. Consequently, the genuine prevalence of deviant sleep duration that is caused by sleep disturbances or sociodemographic and lifestyle factors is unknown. A detailed examination of the variance of sleep duration would also facilitate understanding the risks associated with sleep duration.

For this study, we decided to examine self-reported sleep duration in the Finnish general population and to analyze its possible associations with a number of factors, including perceived health, sociodemographic factors, lifestyle factors, sleep difficulties and daytime concomitants. In particular, we made an effort to define which determinants of sleep duration would be mutually independent from each other.

Methods

Sampling and survey procedure

The field interviews and examinations for the Finnish Health 2000 Study were done in the years 2000–2001. The implementation, target population, sampling design, samples and methods of the survey are described in detail on http://www.ktl.fi/health2000 (see also Aromaa and Koskinen, 2004; Pirkola et al., 2005). In brief, the two-stage stratified cluster sampling frame done by Statistics Finland comprised all adults aged 30 years and older. Persons aged 80 years or older were over-sampled by doubling the sampling fraction. The Social Insurance Institution selected a sample comprised of 8028 persons aged 30 years or older. These persons were invited to take part in the health interview and physical examination.

In addition, a sample of 1894 young persons aged 18–29 years was drawn using the same sampling design. For this sample, the study protocol was a modified and abbreviated version of the one used for the main sample.

In the first phase, health interviews were conducted by Statistics Finland's interviewers. A few weeks later, health examinations were carried out by the National Public Health Institute's field units (nurses, dentists and physicians).

The interviewers met for interview 6986 persons (87%) of the adult sample and 1504 (79%) subjects of the younger sample at their homes or in an institution. During the interviews, the respondents were given an information leaflet and an informed consent form to be returned after signing. Subjects in the adult sample also received Questionnaire 1 which they were asked to fill in and bring along to the subsequent health examination. During the health examination, subjects were given Questionnaire 3 to be returned by mail. Questionnaire 1 was returned by 6460 subjects (80% of the sample). Questionnaire 3 was completed by 6269 (78% of the sample). As regards the younger sample, a shortened and modified version of Questionnaires 1 and 3 was used; it was given to the participants during the interview. (For all questionnaires, see http://www.ktl.fi/health2000). The questionnaire was returned by 1282 subjects (68% of the sample).

Measures

Sleep-related measures

Sleep duration groups

Self-reported 24-h sleep duration was used for the classification of the participants into short and long sleepers. Subjects were asked: ‘‘How many hours do you sleep in 24 hours?’’ The responses were recorded in whole numbers. According to previous studies (Breslow and Enström, 1980; Hublin et al., 2004; Kripke et al., 1979, 2002, Tamakoshi et al., 2004; Wingard et al., 1982), health risk is increased among subjects reporting sleep duration of 6 h or less and among subjects reporting sleep duration of 9 h or more. Accordingly, for the purposes of this study, subjects were divided into three groups according to their sleep duration. The groups were labeled as ‘short sleepers’ (6 h or less), ‘mid-range sleepers’ (7–8 h) and ‘long sleepers’ (9 h or more).

Deviation from mean sleep duration

If the health risk is increased among short and long sleepers, when compared to mid-range sleepers, it is reasonable to ask whether there is a dose–response like relationship between sleep duration and different exposure factors within the groups of short and long sleepers separately. The answers may yield new information and hypotheses concerning possible causative associations. Therefore, we calculated the deviation from the mean sleep duration by subtracting the reported value from the mean value of the sample. The absolute value of the difference was used for analyses. This variable is hereinafter referred to as ‘sleep duration deviation’. In addition, in the multivariable models of the whole sample this measure allows us to deal with the U-shaped (non-linear) associations that in previous studies have been suggested to exist between sleep duration and the factors under study.

Other sleep-related measures

The questionnaires applied in the Health 2000 Survey were composed using different sources (Aromaa and Koskinen, 2004). We used the following sleep-related questions: being awake during night (‘Have you recently lost much sleep over worry?’) (source: GHQ-12) (Goldberg, 1972); difficulty sleeping without sleeping pills (‘Do you have difficulties in getting sleep without sleeping medicine?’); sleeping disorders or insomnia [‘Have you had some of the following usual symptoms and troubles within the last month (30 days): … sleeping disorders or insomnia…?’] (source: SCL-90) (Derogatis, 1977). The following items related to daytime condition were used: the sum score of the Epworth Sleepiness Scale (ESS) (Johns, 1991) was used as an indicator of daytime sleepiness; questions about symptoms and troubles with ‘…feeling non-energetic or tired?’ (source: SCL-90); ‘… overexertion or exhaustion’ (source: SCL-90); and about exceptional tiredness (‘Are you usually more tired during the daytime than other people of your age?’) served as indicators of daytime tiredness.

Sociodemographic variables

The following sociodemographic variables were used: age, gender, marital status, size of the household, number of children aged <7 years in the household, number of school-aged children in the household, level of education, main occupation and dwelling place. Work exposure variables were the following: sort of working hours, number of working hours per week.

Health and lifestyle variables

For the purposes of this study, health status was operationalized as perceived health through the following question: ‘Is your present state of health: good; rather good; moderate; rather poor; poor?’ Data on clinical diagnoses and medication as well as laboratory measures will be published elsewhere. The following lifestyle variables were used: physical activity during leisure time, alcohol consumption during the past 12 months, smoking, frequency of drinking coffee or tea with sugar during a week's time.

For the phrasings and answering modes of the questions, see http://www.ktl.fi/health2000. The answering modes are also shown in the corresponding tables included in this study.

Data analyses and statistics

For the analyses of this study, the sample of subjects aged 30 years or over was pooled with the sample of subjects younger than 30 years, when possible. Statistical analyses were completed using the Statistical Analysis System (SAS) version 9.1 (SAS Institute Inc., Cary, NC, USA).

The mean values of sleep duration and the prevalence estimates for the short and long sleepers were calculated by using SAS Surveymeans procedure to take the complex sampling design into account.

Comparisons between the sleep duration groups

Chi-squared analysis was performed for bivariable (unadjusted) comparisons of each categorical variable between the groups of different sleep duration. The General Linear Models (GLM) procedure was applied to compare groups in case of continuous variables.

Possible dose–response like relationship between the sleep duration deviation and the different exposure factors in short and long sleepers

The mean values of sleep duration deviation were calculated across the response modes of the categorical exposure factors within both short and long sleepers separately. Concerning the continuous ESS sleepiness score, the sleep duration deviation was classified into four categories (<2 h; 2–2.9 h; 3–3.9 h; >4 h) and ESS mean scores were calculated across these categories. The statistical difference between the means was analyzed by one-way anova. The form of the possible dose–response function was determined visually by graphic representation. The GLM procedure was used in bivariable analyses to estimate the amount of the variance (R2) of sleep duration deviation accounted for by the factor under study.

Multivariable analyses

An effort was made to find independent determinants of sleep duration deviation and to estimate their explanatory power. Three separate multivariable GLM models were constructed: one with sleep related, one with sociodemographic and one with health and lifestyle determinants as explanatory variables and sleep duration deviation as the dependent variable. Initially, all determinants of each category were included in the model with age and gender and the interaction terms of the determinants with age and gender: then the best model (by R2 and model F) was found by dropping out, one by one, determinants not carrying any significant additive information. In the end, all three models were pooled together and the best general model was constructed.

Results

Self-reported sleep duration

The distribution of self-reported sleep duration by gender is given in Fig. 1.

Figure 1.

 Distribution of self-reported 24-h sleep duration by gender in Finnish general adult population (n = 7262).

The average sleep duration was 7.51 h (SD 1.24). Women slept (7.62 h; SD 1.31) slightly more [F(1,7262) = 66.7, P < 0.0001] than men (7.39 h; SD 1.04). Among all subjects 14.5% (16.7% of men and 12.5% of women) were short sleepers and 13.5% (10.5% of men and 16.1% of women) were long sleepers.

Because of the suggested non-linear relationships between age and sleep duration, age was categorized into seven classes. Fig. 2 was created to analyze the relationships between sleep duration, age and gender.

Figure 2.

 Self-reported sleep duration by age and gender in Finnish general adult population (n = 7262). Bars indicate SD, asterisks indicate a significant difference between genders.

As expected, a U-shaped association between age and sleep duration was detected (Fig. 2). The main effects of age (F = 23.4, P < 0.0001) and gender (F = 7.6, P = 0.006) were both independently significant. The age by gender interaction was also significant (F = 7.0, P < 0.0001). Post hoc comparisons revealed that the gender difference was significant in the following age bands: 18–29 years [F(1,1276) = 37.9, P < 0.0001]: 30–44 years [F(1,2046) = 76.6, P < 0.0001]: and 45–54 years [F(1,1538) = 9.0, P = 0.003]. In the remaining bands, there were no gender differences observed.

Sleep-related variables

Comparisons between the sleep duration groups

The differences (chi-squared test) between short, long and mid-range sleepers across the variables describing sleep and its daytime concomitants are shown in Table 1.

Table 1.   Bivariable comparisons between sleep groups across sleep-related variables
Determinant (n)Sleep groups
Short sleepersMid-range sleepersLong sleepers
  1. Paired comparisons between sleeper groups and alternative choices on the relevant predictor variable: *P < 0.05 for comparisons against mid-range sleepers, #P < 0.05 for comparisons between short and long sleepers. R2 refers to the bivariable explanatory General Linear Model of the sleep duration deviation as the dependent variable.

Feeling non-energetic or tired (5873) (R2 = 0.08)(n = 904)(n = 4235)(n = 734)
 Not at all24.9%*36.8%26.3%*
 Quite little28.4%*34.8%29.4%*
 To some extent29.2%*21.2%27.3%*
 Quite much12.6%*6.2%12.5%*
 Very much4.9%*1.0%4.5%*
Sleeping disorders or insomnia (5884) (R2 = 0.05)(n = 908)(n = 4242)(n = 734)
 Not at all21.5%*#41.2%45.2%*#
 Quite little19.1%*#28.3%24.7%*#
 To some extent27.2%*#21.5%21.0%#
 Quite much21.1%*#7.4%6.0%#
 Very much11.1%*#1.6%3.1%#
Exceptional tiredness (7237) (R2 = 0.05)(n = 1037)(n = 5179)(n = 1021)
 Yes, nearly always7.6%*2.3%6.5%*
 Yes, often18.7%*11.5%16.5%*
 No44.3%*#61.9%51.9%*#
 Cannot say29.4%*#24.3%25.1%#
Difficulties in getting sleep without sleeping medicine (7218) (R2 = 0.04)(n = 1033)(n = 5170)(n = 1015)
 No52.1%*#71.2%73.4%#
 Sometimes25.9%#23.4%18.8%*#
 Often11.0%*#2.8%2.9%#
 Nearly always10.9%*#2.6%4.9%*#
Overexertion or exhaustion (5859) (R2 = 0.03)(n = 899)(n = 4228)(n = 732)
 Not at all29.5%*#39.2%40.4%#
 Quite little29.5%*33.9%30.3%
 To some extent26.9%*#20.3%19.7%#
 Quite much10.8%*#5.5%6.6%#
 Very much3.3%*1.0%3.0%*
Being awake during night (7148) (R2 = 0.01)(n = 1021)(n = 5140)(n = 987)
 Not at all23.6%*#36.3%43.6%*#
 No more than usual48.2%#48.9%42.8%*#
 To some extent more than usual21.5%*#13.0%11.6%#
 Much more than usual6.7%*#1.8%2.0%#
Sleepiness (5452) (R2 = 0.005) (ESS score mean ± SD)4.6 ± 3.2*4.3 ± 2.74.4 ± 3.2

There was clear U-shaped association observed between several sleep-related problems and sleep duration. Both short and long sleepers reported significantly more symptoms of feeling non-energetic or tired, exceptional tiredness and severe symptoms of overexertion or exhaustion than mid-range sleepers. However, moderate symptoms of overexertion or exhaustion were more common in short sleepers than in the other groups. Short sleepers reported significantly more sleeping disorders or insomnia, difficulties in getting sleep without sleeping medicine, and being awake during night, when compared to mid-range and long sleepers. Short sleepers were also sleepier than mid-range sleepers, but they did not differ significantly from long sleepers in their average sleepiness.

Sleep duration deviation

The relationships between the amount (intensity or frequency) of exposure to a sleep problem or its daytime concomitants and the sleep duration deviation (formally considered here as a response without a true causative interpretation) were tested separately in short and long sleepers. The form of the dose–response like function was studied visually by graphic representations. The results are shown in Table 2 and an example of the graphic representation is shown in Fig. 3.

Table 2.   Dose–response like relationships (anova) between the sleep-related variables and sleep duration deviance
‘Exposure’ variabled.f.Short sleepersLong sleepers
FPForm of the functionFPForm of the function
  1. ‘Upward’ means, for the form of the function the increasing (intensity or frequency) ‘exposure’ variable is positively associated with the increasing sleep duration deviance.

Feeling non-energetic or tired421.3<0.0001Upward24.8<0.0001Upward
Sleeping disorders or insomnia427.1<0.0001Upward0.60.687 
Exceptional tiredness327.0<0.0001Upward21.9<0.0001Upward
Difficulties in getting sleep without sleeping medicine332.0<0.0001Upward4.00.008Upward
Overexertion or exhaustion414.4<0.0001Upward (from ‘to some extent’)12.2<0.0001Upward (from ‘quite little’)
Awake during night322.6<0.0001Upward1.00.399 
Sleepiness31.200.297 12.5<0.0001Upward
Figure 3.

 Graphic representation of the form of the dose–response function between the symptoms of feeling non-energetic or tired and the absolute value of the sleep duration deviation from the sample mean in short and long sleepers. Bars indicate SD.

In all cases (except for three), there was a significant upward dose–response like function between the studied sleep-related variables and the sleep duration deviation (Table 2). The exceptions were as follows: the ‘sleeping disorder or insomnia’ related symptoms and ‘being awake during night’ were not associated with sleep duration deviation in long sleepers; sleepiness as defined by the ESS was not associated with sleep duration deviation in short sleepers.

Multivariable analysis

In Table 1, the independent variables are ordered by descending explanatory power (R2), as revealed by bivariable analyses. Separately, they accounted for 8–0.5% of the variance in sleep duration deviation. Combining all sleep-related variables and their interaction terms with age and gender into the same multivariable explanatory model revealed that the ‘being awake during night’ variable is not independently significant. After dropping the insignificant predictors out of the model, one by one, the model accounted for 12% of the variance (Table 3). Consequently, there was (as expected) a strong multicollinearity between different sleep-related variables, although all the predictors were independently significant for the model. The best three predictors (feeling non-energetic or tired; difficulties in getting sleep without sleeping medicine; and sleepiness) accounted for 10% of the variance in sleep duration deviation.

Table 3.   The best explanatory model (n = 5328) of sleep duration deviance by sleep-related variables
PredictorF-valueP
  1. Model R2 = 0.120; model d.f. = 19; model F = 38.2; model P < 0.0001.

Feeling non-energetic or tired31.9<0.0001
Difficulties in getting sleep without sleeping medicine23.3<0.0001
Sleepiness16.5<0.0001
Exceptional tiredness15.7<0.0001
Overexertion or exhaustion7.4<0.0001
Sleeping disorders or insomnia6.6<0.0001

Sociodemographic variables

Comparisons between the sleep duration groups

Comparisons between short, long and mid-range sleepers across the sociodemographic variables are shown in Table 4. There were significant differences between short and long sleepers in all except four of the 11 variables analyzed. Short and long sleepers did not differ from each other in the following variables: size of the household; number of children aged <7 years old in the household; sort of working hours; and dwelling place. Short sleepers differed from mid-range sleepers in all variables, except in sort of working hours. Long sleepers differed from mid-range sleepers in all except two variables: sort of working hours and dwelling place. An additional analysis was performed for in case of weekly working hours. The weekly working time was higher among short sleepers than among mid-range sleepers who, in turn, worked longer than long sleepers. The same order was found in separate comparisons among subjects with full-time employment and subjects with part-time employment.

Table 4.   Bivariable comparisons between sleep groups across sociodemographic variables
Determinant (n)Sleep groups
ShortMid-rangeLong
  1. Paired comparisons between sleeper groups and alternative choices on the relevant predictor variable: *P < 0.05 for comparisons against mid-range sleepers; #P < 0.05 for comparisons between short and long sleepers. R2 refers to the bivariable explanatory model of the sleep duration deviation as the dependent variable and a determinant as an explanatory variable.

Main occupation (7243) (R2 = 0.04)(n = 1040)(n = 5181)(n = 1022)
 Full-time employment47.6%*#56.5%29.0%*#
 Part-time employment3.0%*#5.0%6.7%*#
 Student2.8%*#4.6%8.2%*#
 Retired37.4%*23.7%38.5%*
 Unemployed6.8%#6.6%10.9%*#
 At home1.4%#2.6%5.8%*#
 Other1.0%1.0%1.2%
Marital status (7242) (R2 = 0.03)(n = 1040)(n = 5181)(n = 1021)
 Married or cohabiting58.5%*69.5%58.7%*
 Divorced11.5%*#7.2%7.6%#
 Widowed13.6%*#6.4%10.4%*#
 Single16.4%#16.9%23.3%*#
Age (7262) (R2 = 0.02)52.6 ± 17.9*#46.6 ± 16.747.9 ± 21.7*#
 Age in men 324148.3 ± 17.0*45.4 ± 15.749.1 ± 21.4*
 Age in women 402657.1 ± 17.8*#47.7 ± 17.447.4 ± 21.8#
Level of education (7233) (R2 = 0.01)(n = 1039)(n = 5178)(n = 1016)
 Basic53.5%*#40.3%43.5%#
 Secondary28.4%29.4%26.3%*
 Higher18.1%*#30.3%30.2%#
Size of the household (7179) (R2 = 0.01)(n = 1032)(n = 5160)(n = 987)
 1 person31.2%*21.3%27.3%*
 2 persons36.7%38.0%38.9%
 3 persons13.5%*16.8%15.7%
 4 persons12.3%*15.0%11.5%*
 5 or more persons6.3%9.0%6.7%
Number of children <7 years old In the household (5485) (R2 = 0.007)(n = 709)(n = 4059)(n = 717)
 083.4%*80.0%78.9%
 19.5%*#12.4%14.5%*#
 2 or more7.2%7.6%6.7%
Number of school aged children in the household (5486) (R2 = 0.005)(n = 709)(n = 4059)(n = 718)
 072.4%*#68.6%77.2%*#
 115.0%#16.6%10.7%*#
 29.6%10.7%9.5%
 3 or more3.1%4.2%2.6%
Sort of working hours (3589) (R2 = 0.005)(n = 471)(n = 2821)(n = 297)
 Regular day-job or evening job67.1%72.2%67.3%
 Night job or shift work16.8%13.8%15.8%
 Other sort of working time16.1%13.9%16.8%
Dwelling place (7262) (R2 = 0.003)(n = 1044)(n = 5194)(n = 1024)
 Urban centers49.8%*54.0%51.4%
 Urban like districts5.5%*3.8%4.0%
 Rural regions44.6%42.2%44.7%
Weekly working time in hours (4254) (R2 = 0.001) (mean ± SD)40.6 ± 8.1*#38.9 ± 8.335.5 ± 9.3*#
Gender (R2 = NS) 7262(n = 1044)(n = 5194)(n = 1024)
 Women48.7%*#54.8%65.4%*#
 Men51.3%*#45.2%34.6%*#

Sleep duration deviation

For ordinal variables, the analysis was made analogously to the sleep-related variables (Table 5). For nominal variables, it is pointless to suppose any dose–response like association with the sleep duration deviation. In these cases, anova was only performed and the group with the highest sleep duration deviation is named in Table 5.

Table 5.   Associations between sociodemographic variables and sleep duration deviation
‘Exposure’ variabled.f.Short sleepersLong sleepers
FPForm of the function/ the group with largestFPForm of the function/ the group with the largest
  1. Dose–response like relationships (in case of ordinal variables) and the group with the largest sleep duration deviation (in case of nominal variables) (anova) are shown

    Sleep duration deviation  Sleep duration deviation
Main occupation65.0<0.0001Unemployed subjects14.1<0.0001Unemployed subjects
Marital status35.10.002Widowed subjects23.6<0.0001Widowed subjects
Age65.7<0.0001Upward (from 65 years)22.6<0.0001Upward (from 65 years)
Level of education213.8<0.0001Downward19.6<0.0001Downward
Size of the household51.10.353 4.9<0.0001U-shaped association
Number of children under 7 years old40.30.842 8.2<0.0001Downward
Number of school-aged children50.90.492 1.70.169 
Sort of working hours22.60.075 8.00.0004Night or sift work
Dwelling place42.70.281 5.90.003Living in countryside
Gender11.30.257 0.50.499 

Of the 10 variables analyzed, eight were associated with sleep duration deviation in long sleepers, but only four in short sleepers.

Multivariable analysis

In Table 4, the independent variables are ordered by descending explanatory power (R2) as revealed by bivariable analyses. Separately, they accounted for 4–0.1% of the variance in sleep duration deviation. The best model with sociodemographic variables and their interaction terms with age and gender accounted for 6.2% of the variance (Table 6). The best independent predictors were main occupation, marital status, age and the interaction of marital status with age. In long sleepers, single subjects showed the largest sleep duration deviation among those aged 30–44 years, but in short sleepers there was no association of age and sleep duration deviation when stratified by marital status.

Table 6.   The best explanatory model (n = 7241) of sleep duration deviation by sociodemographic variables
PredictorF-valueP
  1. Model R2 = 0.062; model d.f. = 13; model F = 36.9; model P < 0.0001.

Main occupation21.6<0.0001
Age×marital status25.0<0.0001
Marital status20.4<0.0001
Age16.9<0.0001

Health and lifestyle variables

Comparisons between the sleep duration groups

Comparisons between short, long and mid-range sleepers across the perceived health and lifestyle variables are shown in Table 7. There were significant differences between short and long sleepers in all studied variables except one: leisure-time physical activity. Both short and long sleepers differed from mid-range sleepers in all the five variables analyzed.

Table 7.   Bivariable analyses between sleep duration groups and the perceived health and lifestyle variables
Determinant (n)Sleep groups
ShortMid-rangeLong
  1. Paired comparisons between sleeper groups and alternative choices on the relevant predictor variable: *P < 0.05 for comparisons against mid-range sleepers; #P < 0.05 for comparisons between short and long sleepers. R2 refers to the bivariable explanatory model of the sleep duration deviation as the dependent variable and a determinant as an explanatory variable.

Self-perceived health (7223) (R2 = 0.06)(n = 1036)(n = 5169)(n = 1019)
 Good28.4%*#41.4%36.4%*#
 Rather good25.5%*29.7%24.9%*
 Average29.0%*#21.4%24.9%*#
 Rather poor11.7%*#6.0%8.4%*#
 Poor5.5%*1.6%5.3%*
Physical activity in leisure time (7141) (R2 = 0.02)(n = 1022)(n = 5132)(n = 987)
 Not much35.3%*24.6%35.3%*
 >4 hours per week47.9%*54.6%49.2%*
 Exercise >3 hours/week15.2%*18.5%13.0%*
 Competition sports1.6%2.3%2.5%
Alcohol consumption (7165) (R2 = 0.02)(n = 1029)(n = 5135)(n = 1001)
 Not once19.7%*13.8%21.9%*
 Max once/month27.7%29.3%30.4%
 2–4 times/month34.8%*38.1%34.8%*
 2–3 times/week11.8%*14.5%9.8%*
 5–7 times/week6.0%*#4.3%3.2%#
Coffee or tea with sugar (7024) (R2 = 0.002)(n = 999)(n = 5060)(n = 965)
 At least 3 or times/day24.6%*#20.5%15.2%*#
 1–2 times/day25.3%22.5%25.9%*
 2–5 times/week4.1%*#6.3%7.5%#
 More rarely15.5%15.2%16.2%
 Never30.4%*#35.5%35.2%#
Smoking (7231) (R2 = 0.001)(n = 1037)(n = 5157)(n = 1019)
 Not smoking67.7%*#71.7%77.2%*#
 Occasionally smoking27.2%*#21.6%16.8%*#
 Habitually smoking5.1%6.6%6.0%

Sleep duration deviation

The associations between the sleep duration deviation and the perceived health and lifestyle variables are described in Table 8. As expected, poor perceived health and physically passive lifestyle were associated with larger sleep duration deviation in both short and long sleepers. In both groups, the largest sleep duration deviation was found among those subjects who did not use alcohol at all. In short sleepers, non-smoking was slightly associated with larger sleep duration deviation, in comparison with occasional or habitual smoking.

Table 8.   Dose–response like relationships (anova) between the self-perceived health and lifestyle variables and sleep duration deviation
‘Exposure’ variabled.f.Short sleepersLong sleepers
FPForm of the functionFPForm of the function
Self-perceived health427.0<0.0001Upward25.6<0.0001Upward
Physical activity in leisure time35.50.001Downward11.7<0.0001Downward
Alcohol consumption frequency44.50.001U-shaped association14.9<0.0001Downward
Coffee or tea with sugar42.20.066 2.00.098 
Smoking23.70.025Downward1.50.223 

Multivariable analysis

In Table 5, the independent variables are ordered by descending explanatory power (R2) revealed by bivariable analyses. Separately, they accounted for 6–0.1% of the variance in sleep duration deviation. The strongest explanatory variable was perceived health. The best model with perceived health and lifestyle variables and their interaction terms with age and gender accounted for 6.8% of the variance (Table 9). Thus, almost all the information was carried by perceived health alone, and adding other variables increased the R2 relatively little. In the model, however, the main effect of perceived health was not significant because of interactions with age and gender.

Table 9.   The best explanatory model (n = 7064) of the sleep duration deviation by the self-perceived health and lifestyle variables
PredictorF-valueP
  1. Model R2 = 0.068; model d.f. = 24; model F = 21.3; model P < 0.0001.

Gender6.80.009
Age×physical activity6.40.0003
Alcohol consumption5.00.0005
Gender×self-perceived health4.40.002
Age×self-perceived health3.70.006
Physical activity2.20.084
Self-perceived health1.20.331
Age1.10.307

Combined model

If all the three above-described multivariable models were completely additive for explaining sleep duration deviation, they would together account for 25% of the variance. However, when the three models were combined, the best model only accounted for 16.4% of the variance. In the model, the main effects of gender, feeling non-energetic or tired, sleeping disorders or insomnia, marital status, main occupation, difficulties in getting sleep without sleeping medicine and leisure-time physical activity were independently significant. Age had an independently significant interaction with leisure time physical activity, marital status, exceptional tiredness, sleeping disorders or insomnia and sleepiness. Gender had an independently significant interaction with difficulties in getting sleep without sleeping medicine, sleepiness and perceived health.

Discussion

Prevalence of short and long sleepers

In a representative nationwide sample of the adult Finnish population, the average sleep duration was 7.5 h. The prevalences of both short (14.5%) and long sleepers (13.5%) were in line with previously published figures as observed in different cultures (range 14–29% for short sleepers and 7–10% for long sleepers) (Ayas et al., 2003; Burazeri et al., 2003; Chen and Foley, 1994; Gottlieb et al., 2002; Hublin et al., 2004; Kojima et al., 2000; Kripke et al., 1979, 2002, Liu et al., 2000; Tamakoshi et al., 2004).

On the other hand, recent data from the US National Health Interview Survey indicate that the percentage of adults sleeping 6 h of less increased from 20% in 1985 to 25% in 2004 (Cizza et al., 2005). The Finnish percentage of 14.5% then appears low. It is generally accepted that 24-h society, originally due to considerable changes in technology, economics and working life, is one of the major reasons for the increasing trend of insomnia and short sleep among the working population. The differences in demographics and especially working hours between the USA and the EU countries, the latter with more strict working hour regulations, could explain a part of this difference. Based on our results, the socioeconomic status (basic education), weekly working time and night or shift work were all related with a higher percentage of short than average length sleepers. In contrast, students and non-working subjects were over-represented among long sleepers. According to earlier studies, only very high weekly working hours – such as 50–75 h – are related with insufficient sleep (e.g. Härmäet al., 2003; Hublin et al., 2001).

Sources of variance in sleep duration

Our multivariable analysis showed, as expected, a clear multicollinearity between the sleep-related variables; even the best linear combination accounted for about 12% of the variance in sleep duration deviation. When we combined all three multivariable models, the best model accounted for about 16% of the variance. We emphasize that it is not possible to draw a conclusion that the determinants cause 16% of the variance in sleep duration deviation. We suggest that such a small proportion of explained variation is due to the highly heterogeneous sources of variance in sleep duration in a general population.

Sources of variance in reported sleep duration are many, and they have different impacts on health and well-being. In our study, gender, feeling non-energetic or tired, sleeping disorders or insomnia, marital status, main occupation, difficulties in getting sleep without sleeping medicine and leisure-time physical activity each had an independent main effect on the variance. These results are in line with a recent Norwegian study (Ursin et al., 2005).

Methodological differences make it difficult to compare individual studies. For example, there are substantial differences in the operationalization of short and long sleep. Varying definitions of short sleepers have been presented in literature: <5.5 h (Webb and Friel, 1971); <6 h (Aeschbach et al., 2001, 2003, Bliwise et al., 1994; Hartmann et al., 1972; Skinner, 1983); ≤6 h (Fichten et al., 2004; Grandner and Kripke, 2004; Hicks and Rozette, 1986; Hicks and Youmans, 1989; Hicks et al., 1991); ≤6.7 h (Wagner and Mooney, 1975). In some studies, the variable size of the extreme ends of the sleep duration distribution in a given study sample have been selected for group comparisons (Hicks and Pellegrini, 1977; Kumar and Vaidya, 1982, 1984, 1986); median or mean split has been used (Fichten et al., 2004); and also an agreement with the statement ‘five or six hours of sleep are sufficient to me’ has been used as a definition of a short sleeper group (Domino et al., 1984). Analogously, the same is true in the definition of long sleepers. Some studies have included in the comparisons mid-range sleepers, sleeping 6–8 h (Bliwise et al., 1994) or 7–8 h (Grandner and Kripke, 2004), while several studies have not included mid-range sleepers in the comparisons. The influence of these methodological differences on the results in different studies is not clearly known. We tried, in part, to address these problems by analyzing, in addition to the usual group comparisons, the dose–response relationships between the sleep duration deviation and different determinants within the groups of short and long sleepers.

Sleep-related variables

In general, short sleepers reported more insomnia-related symptoms, physical tiredness, fatigue-related symptoms and sleepy behavior than mid-range sleepers. Long sleepers reported more physical tiredness and fatigue-related symptoms than mid-range sleepers, but not reported sleepiness or insomnia-related symptoms. In comparison to long sleepers, short sleepers reported more fatigue and insomnia-related symptoms, but not physical tiredness or reported sleepiness. It is worth noting that the intensity or frequency of most of the insomnia and fatigue-related symptoms, as well as, physical tiredness were positively associated with larger sleep duration deviation, both in short and long sleepers. Behavioral sleepiness was the only exception. In short sleepers, the reported sleepiness was not associated with larger sleep duration deviation, but in long sleepers the association was found. As a whole, our results are in general agreement with the earlier studies indicating that insomnia-related symptoms, sleepiness and fatigue follow a U-shaped distribution as a function of total sleep time (Grandner and Kripke, 2004; Kripke et al., 2002). The lack of a dose–response like relationship between sleep duration deviation and reported sleepiness (as defined by ESS sum score) in short sleepers is an intriguing finding. Possible reasons for the positive link between the sleep duration deviation and reported sleepiness in long sleepers and the lack of this link in short sleepers are likely to be dissimilar. One explanation may be that the primary insomniacs are over-represented in short sleepers as suggested by the high prevalence rates of insomnia-related symptoms in this group. Accumulative evidence shows that primary insomniacs have hyperarousal which tends to prevent or decrease their daytime sleepy behavior (see e.g. Bonnet and Arand, 1997, 2000; Stepanski et al., 1988). However, it needs to be kept on mind that short sleepers are a heterogeneous group including also secondary insomniacs and sleep-deprived persons as well as short sleepers by nature. The heterogeneity of this group probably explains the low amount of the variance in sleep duration deviation accounted for by sleepiness. Therefore, a detailed clinical analysis of this group is still needed in future.

In previous studies, short sleepers have also been described to report a higher frequency of insomnia-related symptoms than long sleepers among well-sleeping older adults (Fichten et al., 2004) and in a representative sample of British adults (Groeger et al., 2004). On the other hand, in a nationally representative sample in the USA (Grandner and Kripke, 2004), both short and long sleepers reported more sleep problems than mid-range sleepers. Self-reported sleep duration may not correspond well with the sleep duration assessed with polysomnography in short and long sleepers in particular. As far as the long sleepers are concerned, it is proposed that those who are not satisfied with or successful in life (or are ill) tend to stay longer in bed (Groeger et al., 2004). Their estimates concerning self-reported sleep duration may thus reflect more time in bed than objectively defined sleep duration. Similarly, the extent to which reported sleep duration is influenced by cognitive beliefs about sleep need is not known in short sleepers. However, this does not devalue the importance of self-reported symptoms such as sleep duration as possible markers of health risks in epidemiologic settings. Notably, both long and short sleepers report similar types of sleep complaints (Grandner and Kripke, 2004). Therefore, it is important to analyze insomnia-related symptoms also in short sleepers, although it may appear as these symptoms are consequences of the sleep report.

In our study, long sleepers did not differ from mid-range sleepers for sleep disorder or insomnia symptoms or for symptoms of being awake during night. The reason for this discrepancy with Grandner and Kripke (2004) study may be a difference in the set of questions used in the studies. In addition, the different methods of operationalizing the mid-range sleeper group may have influenced the results. Grandner and Kripke defined mid-range sleepers as subjects with 8 h of sleep, whereas we included also subjects with 7 h of sleep in this group. Grandner and Kripke found that subjects with 7 h of sleep reported slightly but significantly more problems with both falling asleep and waking up non-refreshed than subjects with 8 h of sleep. Combining these two groups (as in the present study) is likely to reduce the difference between mid-range and long sleepers. This is in line with our finding that both the intensity of the symptoms of sleeping disorders or insomnia and difficulties in falling asleep without a sleeping pill were positively associated with larger sleep duration deviation in both short and long sleepers. However, being awake during night was positively associated with larger sleep duration deviation in only short sleepers and not in long sleepers. Thus, combining 7- and 8-h sleepers into one group can be considered as a limitation of our study. However, we think that the interpretation concerning the U-shaped associations would not have changed if we had had four measurement points instead of three. We do emphasize that potential differences between 7- and 8-h sleepers (especially concerning health status) needs to be analyzed in further studies, because epidemiologic surveys suggest that sleeping 8 h is associated with higher mortality than sleeping 7 h (see e.g. Kripke, 2004).

Daytime mental and physical alertness (as a consequence of nocturnal sleep) is a multidimensional concept (see e.g. Bonnet and Arand, 2001; Pilcher et al., 2003; Sangal et al., 1999; Seidel et al., 1984). We compared the different sleep duration groups in terms of tiredness, exhaustion (fatigue) and sleepiness.

Subjective feelings and symptoms of physical tiredness (feeling non-energetic or tired and/or exceptional tiredness) were more prevalent among short and long sleepers than among mid-range sleepers. However, these symptoms did not differ between short and long sleepers. We consider these symptoms to be related with the self-reported perceived daytime sleepiness factor described by Kim and co-workers (Kim and Young, 2005), a dimension of daytime alertness which is different from sleepy behavior. Our concept leaves outside some forms of fatigue, which may have a variety of medical origins.

If short sleepers were actually deprived of sleep, as suggested by Aeschbach et al. (1996, 2001), one would expect them to report a lower perceived alertness than long sleepers. Our results do not support this assumption. This is likely to be explained, in part, by the suggested increased tolerance to sleep pressure by natural short sleepers (Aeschbach et al., 2001; Richardson et al., 2002). Similarly to a previous study (Kahneman et al., 2004), short sleepers were more tired than the mid-range sleepers. We also observed that the intensity of the self-reported tiredness was positively related with larger sleep duration deviation in both short and long sleepers. This dose–response like association between the sleep duration deviation from the sample mean and the alertness-related perceived symptoms of tiredness within both the short and long sleeper groups has not been previously described. This is in line with the suggestion that more sleep is not always better (Groeger et al., 2004).

On the other hand, short sleepers reported frequent symptoms of overexertion or exhaustion more often than long sleepers who, in turn, reported these symptoms more often than mid-range sleepers. We attribute these symptoms to fatigue, reflecting a different aspect of daytime alertness than feelings of physical tiredness. It is likely that these symptoms are related to different medical conditions. Our results agree with the suggestion that habitual short sleep is associated with vital exhaustion (Van Diest, 1990). This may indicate that, among short sleepers, there are more pathologic factors than in other sleep groups. A positive dose–response like relationship between the exertion or exhaustion and larger sleep duration deviation was revealed in both groups with deviant sleep duration. It is important to note that the concept of fatigue (like perceived daytime alertness in general) is not clearly defined either. It is considered to include three to five dimensions: general fatigue (tired, bushed, exhausted), mental fatigue (cognitive impairment), physical fatigue and sleepiness (tendency to fall asleep) and sometimes lack of motivation or activity (Åkerstedt et al., 2004). We interpret the symptoms of overexertion or exhaustion measured in our study to be related with general fatigue. In their study, Åkerstedt et al. (2004) found that mental fatigue was strongly and independently associated with disturbed sleep. They emphasized that insomniacs are not characterized by sleepiness but by mental fatigue and weariness. In our multivariable analysis, all symptoms related with daytime alertness were independently associated with sleep duration deviation, which adds the evidence supporting the multidimensional structure of perceived daytime alertness in relation to sleep duration.

The third dimension of daytime alertness measured in our study was reported sleepiness which was defined independently from subjective feelings of tiredness or mental alertness. We measured this dimension by the Finnish modification of the ESS, providing a more direct way of testing the assumed association between sleep deprivation and short sleep. We found short sleepers to be sleepier than mid-range sleepers. This is in agreement with a recent study by Liu et al. (2000). However, Monk et al. (2001) reported no difference in the scores of the ESS between natural short sleepers and their controls. These results suggest that increased sleepiness among short sleepers is due to insufficient sleep because of behavioral and/or health-related factors, rather than sleep duration per se. This interpretation is further supported by the finding that, in short sleepers, the ESS-defined sleepiness was not associated with the sleep duration deviation, probably because natural short sleepers and primary insomniacs as well as persons who were sleep deprived due to lifestyle factors were all included. An alternative interpretation is that short sleepers are not able to accurately assess their physiologic sleep tendency (Richardson et al., 2002), and that the more extreme short sleeper an individual is, the more the awareness of sleep need may be altered.

We also found, in line with a recent study (Grandner and Kripke, 2004) that short sleepers did not significantly differ from long sleepers in their level of sleepiness. This result supports the criticism presented by Youngsted and Kripke against the suggestion that long sleepers would be more alert during daytime than those who only sleep 7 h (Youngstedt and Kripke, 2004a). In long sleepers, the association between sleepiness and the sleep duration deviation was positive, suggesting that extreme long sleepers were sleepier than the more moderate long sleepers. This finding is in line with a study by Richardson et al. (2002). However, when compared to the original ESS, there is a validity issue in the Finnish modification of ESS which must be taken into account. In the original ESS (Johns, 1991), 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 frequency of dozing in different situations (e.g. ‘seldom’). We have unpublished evidence (M. Partinen, oral communication) that the difference in phrasing has influence on the validity of the questions. In a sample of 32 subjects, the Finnish-modified ESS yielded a mean value that was approximately two points lower than the value yielded by the original version of the ESS. This difference in validity between the original ESS and the Finnish modification will probably explain the difference between the sample mean values (4.3 ± 3.0) in the present study and in other studies among Australian workers (5.8 ± 4.0) (Johns and Hocking, 1997); in a Spanish population (6.1 ± 2.9) (Izquierdo-Vicario et al., 1997); among elderly persons (5.9 ± 4.0) (Whitney et al., 1998); and among 18–65 years old subjects (7.9 ± 4.5) (Myers et al., 2003). The difference in validity must be considered as a limitation in our study in terms of the generalizability of our results. Keeping this in mind, it seems unlikely that the altered ability to assess their physiologic sleep tendency would explain our finding that short sleepers are not markedly sleepier than long sleepers. The Finnish version of ESS emphasizes the actual sleepy behavior (real behavioral dozing events) even more directly than the original ESS (estimated probability or likelihood to doze). Thus, one would expect the Finnish modification to be less vulnerable to altered ability to assess one's physiologic sleep tendency than the original version of the ESS.

Interestingly, in a recent study (Ng and Tan, 2005), sleepiness was assessed using the frequency phrase ‘how often’ (…would a subject doze off or fall asleep or feel sleepy in different situations?). The study did not reveal any significant association between the increased odds of excessive daytime sleepiness and reported sleep hours (Ng and Tan, 2005), suggesting, in line with our study, that short sleepers did not differ from long sleepers in terms of behavioral sleepiness. In future, it is important to define how large the proportion of naturally short sleepers is in the whole group of short sleepers in the general population. There is some experimental evidence suggesting a greater sleep debt (volitional sleep deprivation) in young subjects with shorter habitual bedrest duration (Klerman and Dijk, 2005). In our combined multivariable model, the main effect of sleepiness on sleep duration deviation was not independently significant, whereas the interaction of sleepiness with gender was significant. This suggests that the association of sleepiness with sleep duration is modulated by many factors and deserves a more detailed analysis. Future studies should also include an estimate of the insufficient sleep which may be a more important determinant of daytime sleepiness than sleep duration per se (Gillberg, 1995; Hublin et al., 1996; Ng and Tan, 2005).

Sociodemographic variables

We found that being a short sleeper was more prevalent in men (16.6%) than in women (12.6%) and being a long sleeper was more prevalent in women (16.7%) than in men (10.9%). The same gender difference has been reported in representative samples in the USA (Grandner and Kripke, 2004; National Sleep Foundation, 2003), among persons aged 40–45 years in Norway (Ursin et al., 2005), among persons aged 20–45 years in Sweden (Lindberg et al., 1997), in the elderly in Canada (Fichten et al., 2004) and in a British sample (in the age group 40–49 years only) (Groeger et al., 2004). There is also some home-based sleep monitored evidence in favor of the suggested gender difference (Jean-Louis et al., 2000b; Reyner et al., 1995). A possible interpretation is that women have a higher need of sleep than men do (Ursin et al., 2005). This might also explain the common finding of greater self-reported sleep problems in women (in spite of their longer sleep duration) when compared with men (Ursin et al., 2005). However, also contradictory gender-related results have been reported (Belloc, 1973; John et al., 2005; Liu et al., 2000).

Among short sleepers, there were more divorced and widowed subjects than in the other groups. Widowed and divorced subjects also showed larger sleep duration deviation than married or co-habiting short or long sleepers. These results are in agreement with previous studies (Groeger et al., 2004; Ursin et al., 2005).

Short sleepers were significantly older than long sleepers. In a recent study among women, short sleepers were found to be older than the other groups (Verkasalo et al., 2005).

Highly educated subjects were under-represented among short sleepers. Among both short and long sleepers, the largest sleep duration deviation was found in those with less than 9 years of education. These results are in agreement with three recent studies showing that persons with higher education levels slept longer during the week than those with lower education (John et al., 2005; Ursin et al., 2005; Verkasalo et al., 2005).

In our multivariable analysis, the sociodemographic variables revealed a strong multicollinearity, and the best combination of them accounted for 6.3% of the variance in sleep duration deviation. Ursin et al. (2005) found that multiple regression analyses, with sleep duration as the dependent variable and education, family income, marital status, urban/rural living, shift work or watches and subjective health as independent variables, explained less than 3% of the variance in men. Interestingly, Ursin et al. (2005) also reported that subjective health had no impact on sleep duration when the socioeconomic factors were controlled for.

It should be noted that the results suggesting that sociodemographic variables do not to any significant extent account for the variance in sleep duration are not necessarily in disagreement with the suggested role of sleep in translating socioeconomic factors into health (Moore et al., 2002; Van Cauter and Spiegel, 1999). This should be tested more properly by analyzing sociodemographic, health and sleep variables together. A more important factor than society may be the fact that interindividual differences are explained by numerous physiologic and genetic factors that have not been measured (Ursin et al., 2005). There is evidence suggesting that genetic differences account for a substantial 40% of the variance in sleep pattern variables, such as sleep duration (Heath et al., 1990).

Health and lifestyle variables

Self-perceived health was poorest in short sleepers; 17% of them reported rather poor or poor health, whereas 14% of long sleepers and 8% of mid-range sleepers reported rather poor or poor health. Sleep duration deviation was linearly associated with poorer perceived health within both short and long sleepers. Our results do not support the belief that sleeping for more than 8 h per night would be advantageous to individuals; this strengthens the previous evidence against the said belief (Jean-Louis et al., 2000a; Pilcher and Ott, 1998; Pilcher et al., 1997). In future, a more detailed health status analysis in samples differing by sleep duration is clearly needed.

Both short and long sleepers reported less physical activity in leisure time than mid-range sleepers, but they did not differ from each other. Factors contributing to this association are likely to be different in short and long sleepers. However, a dose–response like relationship was found between physical activity and sleep duration deviation in both short and long sleepers. These results are in line with the recent report, according to which female short sleepers were more sedentary exercisers (Verkasalo et al., 2005).

Alcohol consumption was ambiguously associated with sleep duration. Subjects not using alcohol at all were more prevalent in both short and long sleepers than in mid-range sleepers. However, subjects consuming alcohol frequently (five to seven times a week) were also more prevalent among short sleepers than in the other groups. Verkasalo et al. (2005) reported that female short sleepers consumed more alcohol than the others. Smoking was more prevalent in short sleepers than in mid-range or long sleepers. Non-smoking was more prevalent in long sleepers than in the other two groups. Other researchers have suggested that, given a sufficient severity of dependence, short sleep duration may be expected (John et al., 2005). Sleep duration may be influenced via the high comorbidity between alcohol or nicotine dependence and depressive and anxiety disorders. On the other hand, an individual may drink alcohol to induce sleep or to maintain sleep. It has been emphasized that only little evidence exists about the relationship between sleep duration and nicotine or alcohol dependence (John et al., 2005). Consequently, this question needs to be addressed in future research.

Habitual coffee drinking was more prevalent among short sleepers than in the other groups. This may indicate an attempt to compensate for the lower level of certain dimensions of daytime mental and physical alertness in short sleepers than in the other groups. However, an alternative interpretation is that habitual coffee drinking may decrease the sleep duration. It is not possible to choose between these explanations based on this cross-sectional study alone. It is, however, noteworthy that those among the long sleepers who did not drink coffee or did so seldom showed larger sleep duration deviation than coffee-drinking long sleepers.

In multivariable analysis, the self-perceived health and lifestyle variables accounted for about 6% of the variance in sleep duration deviation.

Conclusions

The main conclusions from our study are that the sleep-related, sociodemographic, self-perceived health and lifestyle factors account for a relatively small part of the variance in sleep duration in adult general population. The most important and statistically-independent determinants of sleep duration deviation include gender, physical tiredness, sleep problems, marital status, main occupation and physical activity. However, to clarify their importance for the increased health risk associated with the deviant sleep duration requires a more detailed analysis of the clinical status of short and long sleepers. Our study also emphasizes the dose–response like relationship between the sleep duration deviation and its several determinants within both short and long sleepers. For future studies, it may therefore be fruitful to make a hypothesis about the U-shaped association between sleep duration and sleep-related, sociodemographic, health and lifestyle factors may, at least in some cases, to have a causative nature.

Ancillary