SLEEP, Vol. 30, No. 10, 2007 1245
INTRODUCTION
ABOUT 40 YEARS AGO, THE FIRST REPORT ON THE
ASSOCIATION BETWEEN SLEEP LENGTH AND SUBSE-
QUENT MORTALITY INDICATED A U-SHAPED CURVE,
showing that those sleeping 7 hours had the lowest mortality.1 The
U-shaped association between sleep length and mortality has been
found in most of the nearly 20 published epidemiologic studies.2
The significance of short sleep has generally decreased or disap-
peared after adjustment for factors known to be associated with
mortality, such as smoking, alcohol use, and physical inactivity. As
for long sleep, Youngstedt and Kripke concluded that “studies with a
relative large number of subjects (>10,000) have without exception
shown that sleep of 8 hours or longer is associated with a significant
mortality risk.”2 While this excess risk has been rather robust to ad-
justment for lifestyle and other factors, its precise nature and extent
still remain obscure. Earlier studies have not investigated the effect
of stability in sleep length on subsequent mortality.
Other aspects of sleep behavior (such as sleep quality) and the
association to mortality have been little investigated. Increased
risk of mortality associated with the use of medication for sleep
has been reported.3,4 However, it is not clear which medications
(prescribed hypnotics or nonprescribed sleep promoting agents)
might account for this.5 The only study with verification of the
drugs being taken indicated that sedative-hypnotics are not as-
sociated with increased mortality risk.6
Our objective was to continue to probe the relationship be-
tween these 3 aspects of sleep behavior (self-report of sleep
length, sleep quality, and use of sleep promoting medication),
changes in behavior, and mortality in a large, prospective, popu-
lation-based cohort of Finnish adults with 22-year mortality fol-
low-up data and detailed information on potential confounding or
effect-modifying factors.
METHODS
The Finnish Twin Cohort
The Older Finnish Twin Cohort consists of all Finnish twin
pairs of the same sex born before 1958 with both co-twins alive in
1975. These twin pairs were selected from the Central Population
Registry of Finland in 1974.7 The Cohort includes adult twins and
individuals who were not twins, because twin candidates were
selected by identifying pairs of persons with the same surname
at birth, the same birth date, and the same community at birth.
Biological twinship was confirmed from a questionnaire and local
parish birth records.
The first questionnaire was mailed in autumn 1975, and the
response rate was 89%. The 97-item questionnaire included ques-
tions on sociodemographics, health status, lifestyle and psycho-
social factors, and sleep patterns. In autumn 1981, a second ques-
tionnaire including 100 similar items (response rate 84%) was
mailed to the twins that had responded to the first questionnaire;
Sleep and Mortality: A Population-Based 22-Year Follow-Up Study
Christer Hublin, MD, PhD1; Markku Partinen, MD, PhD2; Markku Koskenvuo, MD, PhD3; Jaakko Kaprio, MD, PhD3,4
1Brain@Work Research Center, Finnish Institute of Occupational Health, Helsinki, Finland; 2Skogby Sleep Clinic, Rinnekoti Foundation, Espoo,
Finland; 3Dept. of Public Health, University of Helsinki, Helsinki, Finland; 4Department of Mental Health and Alcohol Research, National Public
Health Institute, Helsinki, Finland
Sleep and Mortality—Hublin et al
Disclosure Statement
This was not an industry supported study. The authors have reported no
financial conflicts of interest.
Submitted for publication September, 2006
Accepted for publication June, 2007
Address correspondence to: Christer Hublin, Brain@Work Research Center,
Finnish Institute of Occupational Health, Topeliuksenkatu 41 a A, FIN-00250
Helsinki, Finland; E-mail: christer.hublin@ttl.fi
Study Objectives: Long and short sleep have been associated with in-
creased mortality. We assessed mortality and 3 aspects of sleep behavior
in a large cohort with 22-year follow-up.
Design/Setting: Prospective, population-based cohort study.
Participants: 21,268 twins aged ≥18 years responding to questionnaires
administered to the Finnish Twin Cohort in 1975 (response rate 89%), and
1981 (84%).
Interventions: N/A
Measurements: Subjects were categorized as short (<7 h), average, or
long (>8 h) sleepers; sleeping well, fairly well, or fairly poorly/poorly; no,
infrequent, or frequent users of hypnotics and/or tranquilizers. Cox propor-
tional hazard models were used to obtain hazard ratios (HR) for mortality
during 1982-2003 by sleep variable categories and their combinations.
Adjustments were done for 10 sociodemographic and lifestyle covariates
known to affect risk of death.
Results: Significantly increased risk of mortality was observed both for
short sleep in men (+26%) and in women (+21%), and for long sleep
(+24% and +17%), respectively, and also frequent use of hypnotics/tran-
quilizers (+31% in men and +39% in women). Snoring as a covariate did
not change the results. The effect of sleep on mortality varied between
age groups, with strongest effects in young men. Between 1975 and
1981, sleep length and sleep quality changed in one-third of subjects.
In men there was a significant increase for stable short (1.34) and stable
long (1.29) sleep for natural deaths, and for external causes in stable
short sleepers (1.62).
Conclusions: Our results show complicated associations between sleep
and mortality, with increased risk in short and long sleep.
Keywords: Follow-up study, mortality, population, sleep, hypnotics
Citation: Hublin C; Partinen M; Koskenvuo M; Kaprio J. Sleep and mortal-
ity: a population-based 22-year follow-up study. SLEEP 2007;30(10):1245-
1253.
SLEEP DURATION AND MORTALITY
SLEEP, Vol. 30, No. 10, 2007 1246
non-twins were not contacted. The present analyses include those
twin individuals responding to both questionnaires with informa-
tion on sleep length and sleep quality, and resident in Finland in
1981 (N = 21,268; 52.3% women; mean age in 1981, 40.7 years,
SD 13.5 years, range 24-101 years).
The study was approved by the ethical committee of the De-
partment of Public Health, University of Helsinki. Informed con-
sent was obtained from all respondents.
Questionnaire Data
Information on sleep length was obtained by asking “How many
hours do you usually sleep per 24 hours?” In 1975 seven response
alternatives were used (<4 hours, 5, 6, 7, 8, 9, and ≥10 hours), and in
1981 nine alternatives (≤6 hours, 6.5, 7, 7.5, 8, 8.5, 9, 9.5, and ≥10
hours). Information on quality of sleep was obtained both in 1975
and 1981 by asking: “Do you usually sleep well?” The 5 response
alternatives were: “well,” “fairly well,” “fairly poorly,” “poorly,”
and “cannot say.” Use of hypnotics and use of tranquilizers were
included in both questionnaires in the question “On how many days
in total during the last year have you used the following types of
medications?” with alternatives “no use,” “on less than 10 days,”
“on 10-59 days,” “on 60-180 days,” and “on more than 180 days.”
We included tranquilizers in our analyses because they are widely
used interchangeably with hypnotics in clinical practice, and the
majority of the compounds in both groups have similar pharma-
cological profile (they are benzodiazepines or benzodiazepine-like
agents acting on the GABA system in the brain).
We assessed the following sociodemographic and lifestyle co-
variates (asked both in 1975 and 1981; selected characteristics
given in Table 1): married (yes/no), social class (6 categories: up-
per or lower white collar, skilled or unskilled workers, farmers,
others), education (9 categories by years of school, high school
equals to 12 years), working status (employed yes/no), BMI
[body mass index (kg/m2) computed from self-reported weight
and height], smoking status (4 categories: never, occasional, ex-,
or current cigarette smoker), binge drinking,8 grams of alcohol
consumed daily (based on self-reported average quantities of
use of beer, wine, and spirits consumed9), conditioning physical
activity (3 categories: sedentary, intermediate, vigorous physi-
cal activity10), and life satisfaction.11 Life satisfaction correlates
highly (r = - 0.63) with depression as assessed concurrently by
the Beck Depression Inventory.12 Snoring was asked in 1981 only
with response alternatives “never,” “occasionally,” “often,” “al-
most always,” and “do not know.”
Follow-up Data
Vital status (alive in Finland on December 31, 2003, date of
death, or date of migration from Finland) was obtained from the
Population Register Centre of Finland. The follow-up for mortal-
ity was from the exact date of response (date questionnaire re-
turned) to December 31, 2003. The follow-up time is denoted in
Tables as 1982-2003. Our analyses are based on 431,782 person-
years and 3700 deaths during follow-up. Cause-of-death statis-
tics up to the end of 2003 were obtained from Statistics Finland.
Both registers cover all Finnish citizens and permanent residents.
These data were linked to the Finnish Twin Cohort data using
the unique personal identification numbers assigned to every per-
manent resident of Finland. Deaths were categorized as natural
(ICD-8 and ICD-9 codes 1-799, or ICD-10 codes A-R) or due to
external causes (violent deaths; ICD-8 and ICD-9 codes 800-999,
or ICD-10 codes S-Y.)
Data Analysis and Statistical Methods
Cox proportional hazard models were used to obtain hazard ra-
tios (HR) and their 95% confidence intervals (CI) for mortality by
Sleep and Mortality—Hublin et al
Table 1—Selected Demographic Characteristics in 1981 of the Entire Cohort with Complete Sleep Behavior Data. Percentage of Those in the
Specified Category of Potential Confounder; Number of Men/Women Given in Column Header
Use Of Hypnotics And/or
Sleep Length* Sleep Quality Tranquilizers**
Short Average Long Sleeping Sleeping Sleeping fairly No Infrequent Frequent
N (m/f) = N (m/f) = N (m/f) = well fairly well poorly/poorly N (m/f) = N (m/f) = N (m/f) =
1539 / 1500 7113 / 7247 1488 / 2381 N(m/f) = N (m/f) = N (m/f) = 8209 / 8329 454 / 776 261 / 347
4388 / 4586 4804 / 5486 948 / 1056
Age
24-39 years 52.1 / 44.7 56.8 / 56.4 52.8 / 57.3 62.7 / 64.5 52.3 / 51.4 38.3 / 32.0 60.9 / 62.3 40.8 / 46.4 35.3 / 34.3
40-54 years 30.2 / 25.3 29.5 / 26.3 25.6 / 22.3 26.0 / 23.2 30.7 / 27.0 34.4 / 25.9 27.7 / 23.7 37.9 / 29.1 36.8 / 29.1
55 years or more 17.7 / 30.1 13.7 / 17.3 21.6 / 20.5 11.3 / 12.3 17.0 / 21.6 27.3 / 42.1 11.4 / 14.0 21.4 / 24.5 28.0 / 36.6
Married 70.1 / 58.4 74.7 / 68.5 68.4 / 66.8 74.4 / 66.6 72.7 / 67.7 69.0 / 62.5 73.7 / 69.1 73.6 / 63.7 58.2 / 47.0
Employed 82.5 / 76.6 87.8 / 85.1 75.0 / 76.1 89.9 /87.0 84.5 / 81.4 65.9 / 63.7 90.0 / 86.7 76.2 / 79.6 41.0 / 51.6
Social class: skilled and
unskilled workers 63.3 / 48.9 55.7 / 42.2 51.1 / 42.6 54.1 / 39.6 57.3 / 45.0 60.0 / 49.3 56.0 / 42.3 54.2 /37.3 45.6 / 43.2
Education: high school
or more 8.2 / 10.8 12.8 / 16.4 11.8 / 13.0 13.8 / 18.9 11.1 / 13.1 7.9 / 7.4 12.9 / 16.7 15.2 / 20.5 10.3 / 12.2
Current smoker 47.6 / 24.5 35.5 / 19.9 30.3 / 16.3 35.9 / 21.7 35.2 / 18.3 43.2 / 18.5 36.7 / 20.8 42.8 / 21.5 40.1 / 26.5
Binge drinker 47.0 / 10.8 41.3 / 8.8 37.1 / 8.6 37.2 / 8.4 43.8 / 9.3 50.2 / 10.6 42.1 / 9.2 48.9 / 13.1 38.1 / 12.0
BMI*** ≥25 42.0 / 30.9 38.1 / 24.5 41.7 / 26.5 37.5 / 21.1 40.1 / 27.4 42.8 / 37.8 37.3 / 22.3 40.7 / 26.1 52.5 / 37.9
Sedentary 15.4 / 15.8 11.8 /11.3 14.5 /13.2 13.3 / 11.9 11.3 / 11.7 17.6 / 16.8 12.5 / 11.8 11.7 / 13.0 15.3 / 17.6
Low life satisfaction 23.3 / 23.9 14.1 / 14.5 18.2 / 16.0 9.8 / 10.3 17.4 / 17.1 39.0 / 36.4 13.6 / 13.0 30.0 / 28.7 46.4 / 43.6
Deceased 1982-2003 27.1 / 24.7 17.2 / 12.4 25.7 / 16.9 15.0 / 10.9 21.2 / 15.5 36.8 / 30.9 15.5 / 10.8 29.3 / 16.4 42.9 / 35.2
* short = <7 hours, average = 7-8 hours, and long = >8 hours; ** infrequent use = 1-59 days/year; frequent use = 60 or more days/year; *** body
mass index
SLEEP, Vol. 30, No. 10, 2007 1247
sleep length, sleep quality, and the use of hypnotics and/or tran-
quilizers. We ascertained that the proportional-hazards assumption
was not violated by using log-log plots, (i.e. -ln{-ln(survival)}
curves versus ln(analysis time) of survival curves of the 3 cat-
egories of sleep length, sleep quality, and use of hypnotics and/or
tranquilizers, to check that the curves were parallel. Because the
study sample included twin pairs, standard errors and CIs were
adjusted for possible within-pair correlations using robust esti-
mators of variance.13 All statistical analyses were performed with
Stata version 9.2 (Stata Corporation, College Station, TX, USA).
Sleep length was categorized in 3 classes: short (< 7 hours),
average (7-8 hours), and long (> 8 hours). Sleep quality was also
dealt with in 3 categories (well, fairly well, and fairly poorly/poor-
ly). Use of hypnotics and/or tranquilizers was similarly assessed
in 3 categories (no use of either hypnotics or tranquilizers, infre-
quent use = 1-59 days per year of either medication, frequent use
= 60 or more days per year of either medication). Because some
subjects had missing data on use of both hypnotics and tranqui-
lizers, we created a fourth category for those with missing data.
This was included in the modelling in order not to lose subjects,
but results for this class are not shown. Subjects in the reference
group had average sleep length, slept well, and used no hypnotics
and/or tranquilizers.
The association between mortality and the stability of the 3 sleep-
related variables (sleep length, sleep quality, and use of hypnotics
and/or tranquilizers) was assessed using combinations of categories
(3 alternatives both in 1975 and 1981 giving 9 subgroups in each
sleep related variable) in modeling. Age-adjusted HRs for total
mortality are given, and results for men and women are presented
separately because of significant gender differences.
Gender by sleep behavior interactions were tested by assessing
the difference in model fit between a model with gender by sleep
variable interactions (all 3 variables) compared with a model with
main effects of the sleep variables and sex alone. The difference
in model fit is chi-square distributed. This likelihood ratio test
chi-square probability for overall presence of any sex-interactions
was 0.07 in the youngest age-group; correspondingly for the age
group 40-54 years 0.34 and for 55+ years 0.59, and for the total
population 0.03 (Table 10).
In fully-adjusted models, adjustments were made for the so-
ciodemographic and lifestyle covariates (measured in 1981)
known to affect risk of death (see “Questionnaire data” above).
Subjects with missing data on any of the covariates (N given in
each Table) were excluded from the fully-adjusted models. When
a sleep-related variable was not dependent, it was included as a
covariate in the model. The effect of snoring (3 categories: never,
occasionally, and often/almost always) was also assessed by sepa-
rate models. The joint effects of sleep related variables measured
in 1981 were also assessed.
RESULTS
Descriptive data of the study population is given in Table 1 by
categories of the self-reported sleep length, sleep quality, and use
of hypnotics and/or tranquilizers. In the last row, the percentage of
deaths in each category is given; it is lowest in those with average
sleep length, sleeping well, and no use of hypnotics and/or tranqui-
lizers. Covariates have been surveyed both in 1975 and 1981 and
their stability was variable: the kappa-value of, e.g., for being mar-
ried was 0.56, level of education 0.89, binge drinking 0.57, ciga-
rette smoking 0.70, overweight (BMI ≥25) 0.67, sedentary physical
activity 0.30, low life satisfaction 0.26, and being employed 0.45.
There was information on the frequency of use of hypnotics
and/or tranquilizers in 1975 and 1981 in 86.4% of the study popu-
lation. Of all users (N = 1881) 22.9% used only hypnotics, 48.3%
only tranquilizers, and 28.8 % used both types of medication.
To assess the interrelationships between sleep length, sleep qual-
ity, and use of hypnotics and/or tranquilizers (medication) poly-
choric correlation matrices of the 3-class variables measured in
1975 and 1981 were computed, and all correlations in both genders
were statistically significant (P ≤0.02). In men the correlation be-
tween 1975 and 1981 in sleep length was 0.49 (kappa-value 0.25),
in sleep quality 0.64 (0.40), and in use of medication 0.43 (0.21); in
women 0.50 (0.27), 0.63 (0.38), and 0.44 (0.22), correspondingly.
Risk of mortality by each sleep variable category is given
in Table 2. In the fully-adjusted model, there was a significant
increase in mortality in the 2 genders in both short and in long
sleepers: 26% in men and 21% in women for short sleep, and for
long sleep 24% and 17%, respectively. Sleep quality (sleeping
worse than well) was significant only in men in the age-adjusted
model, indicating no independent association between sleep qual-
ity and mortality. Frequent use of hypnotics and/or tranquilizers
significantly increased risk of mortality by 31% in men and by
39% in women. Including snoring as a covariate or exclusion of
deaths during the first 3 years of follow-up (up to the end of 1985)
did not essentially change the HRs or the statistical significance.
Table 3 shows age-adjusted risk of total mortality by age groups
separately for men and women in different sleep variable catego-
ries. In men, short sleep was significantly associated with increased
risk in all ages, most clearly in the youngest group (+ 96%). In
women there was a similar but nonsignificant trend. Sleep quality
significantly affected the risk only in young men with an increase
of 129% in those sleeping fairly poorly/poorly. Frequent use of
hypnotics and/or tranquilizers was associated with increased risk
of mortality in all age groups in both genders (even more clearly
in men), but the effect attenuated with age (HRs in the youngest
group in men 2.90 and in women 2.57, in the oldest group 1.38
and 1.50, respectively). In the fully-adjusted model the HRs were
clearly attenuated and half of the significant hazard ratios became
nonsignificant, and the pattern of decreasing HRs related to sleep
abnormalities with increasing age was mainly lost. Due to smaller
numbers of subjects in the age-group specific analyses, the power
to detect differences was less than in the overall sample.
Table 4 shows the association between stability of sleep length
and total mortality. Length category remained unchanged in
68.8% of men and in 66.2% of women from 1975 to 1981, and re-
spectively, sleep shortened in 16.6% and 16.7% and lengthened in
14.6% and 17.1%. In both age-adjusted and fully-adjusted mod-
els in men, but not in women, stable short (HR 1.36) and stable
long (1.32) sleep was associated with a significantly increased
risk of mortality. A decrease of sleep length to short resulted in
significantly increased mortality in women (1.24-2.17), and there
was a similar trend in men. Lengthening of sleep from average
to long significantly increased risk of mortality in both genders
(about 1.20). Thus, in men there was a U-shaped association with
significantly increased risk of mortality in short and long sleepers
at the beginning of the follow-up, but in women the pattern was
less clear, but with some significant associations. Including snor-
ing as a covariate in the fully-adjusted model did not significantly
change the HRs otherwise, but in men the category average to
Sleep and Mortality—Hublin et al
SLEEP, Vol. 30, No. 10, 2007 1248
short decreased from 1.20 to 1.15 and became nonsignificant. Ex-
cluding sleep quality and use of hypnotics and/or tranquilizers as
covariates did not change HRs substantially.
Table 5 gives the mor