American Journal of Epidemiology Advance Access published December 9, 2014

American Journal of Epidemiology

© The Author 2014. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of DOI: 10.1093/aje/kwu245
Public Health. All rights reserved. For permissions, please e-mail: [email protected].

Original Contribution

Fragmentation and Stability of Circadian Activity Rhythms Predict Mortality

The Rotterdam Study

Downloaded from http://aje.oxfordjournals.org/ at University of California, San Francisco on December 11, 2014
Lisette A. Zuurbier, Annemarie I. Luik, Albert Hofman, Oscar H. Franco, Eus J. W. Van Someren, and
Henning Tiemeier*
* Correspondence to Henning Tiemeier, Department of Epidemiology, Erasmus Medical Center, P.O. Box 2040, 3000 CA Rotterdam,
The Netherlands (e-mail: [email protected]).

Initially submitted May 13, 2014; accepted for publication August 13, 2014.

Circadian rhythms and sleep patterns change as people age. Little is known about the associations between
circadian rhythms and mortality rates. We investigated whether 24-hour activity rhythms and sleep characteristics
independently predicted mortality. Actigraphy was used to determine the stability and fragmentation of the 24-hour
activity rhythm in 1,734 persons (aged 45–98 years) from the Rotterdam Study (2004–2013). Sleep was assessed
objectively using actigraphy and subjectively using sleep diaries to estimate sleep duration, sleep onset latency, and
waking after sleep onset. The mean follow-up time was 7.3 years; 154 participants (8.9%) died. Sleep measures
were not related to mortality after adjustment for health parameters. In contrast, a more stable 24-hour activity
rhythm was associated with a lower mortality risk ( per 1 standard deviation, hazard ratio = 0.83, 95% confidence
interval: 0.71, 0.96), and a more fragmented rhythm was associated with a higher mortality risk ( per 1 standard
deviation, hazard ratio = 1.22, 95% confidence interval: 1.04, 1.44). Low stability and high fragmentation of the
24-hour activity rhythm predicted all-cause mortality, whereas estimates from actigraphy and sleep diaries did
not. Disturbed circadian activity rhythms reflect age-related alterations in the biological clock and could be an indi-
cator of disease.

circadian activity rhythm; elderly; mortality; sleep

Abbreviations: ADL, activities of daily living; CI, confidence interval; HR, hazard ratio; PSQI, Pittsburgh Sleep Quality Index; SD,
standard deviation.

Most physiological processes, including body tempera- investigated were mainly focused on sleep duration. The re-
ture, hormone secretion, and sleep-wake timing, are regulated sults suggested that the association between sleep duration
in cycles that last approximately 24 hours, called circadian and mortality is U-shaped; both subjective short and long
rhythms. Circadian rhythms and sleep patterns change as sleep durations are predictors of all-cause mortality (9, 10).
people age (1). Elderly people sleep less during the night, Few studies have investigated the associations of circadian
have more fragmented sleep, have more difficulty in falling rhythms with mortality. In the elderly, the amplitude of sev-
asleep, tend to fall asleep and wake up earlier, take more naps, eral physiological circadian rhythms is reduced compared
and report a lower sleep quality (2–6). The longitudinal asso- with that of younger people, as is the stability of the day-night
ciations of these changes with adverse health consequences rhythm (1, 4, 11). This could be explained by an age-related
and mortality are not well understood. In previous studies, in- decline in circadian organization. The aging process affects
vestigators found that people who slept for short durations central and peripheral oscillators differently, possibly leading
each night (≤6 hours) and had poor sleep quality had higher to suboptimal peripheral physiological functioning (12). The
risks of diabetes and cardiovascular diseases (7, 8). Studies in circadian rhythm of physical activity in elderly is better char-
which the association between sleep and mortality have been acterized by 2 nonparametric variables that do not assume the

1
2 Zuurbier et al.

24-hour cosine-like shape that is present in, for example, core these, 2,063 (78%) agreed. There were no exclusion criteria
body temperature and hormones. Two nonparametric vari- besides being able to understand the instructions for this
ables quantify stability and fragmentation. A stable activity study. After exclusion of recordings that failed because of
rhythm is characterized by a 24-hour profile that remains technical problems or that contained fewer than 4 consecutive
very similar from day to day. This gives an indication of the days and nights, 1,734 (84%) recordings (mean recording du-
strength of synchronization between the activity rhythm and ration, 138 (standard deviation (SD), 14) hours) were avail-
zeitgebers, which are environmental cues with a 24-hour pat- able for analyses (17). No participants were lost to follow-up.
tern. Fragmentation gives an indication of the frequency of
alterations between rest and activity relative to its 24-hour Actigraphy
amplitude.
In 2 previous studies in which 24-hour activity rhythms We measured the 24-hour activity rhythm using an acti-
were analyzed parametrically, investigators found that older graph unit (Actiwatch model AW4, Cambridge Technology
men and women with the least robust 24-hour activity rhythms Ltd., Cambridge, United Kingdom) worn on each subject’s

Downloaded from http://aje.oxfordjournals.org/ at University of California, San Francisco on December 11, 2014
had a 1.5–2-times higher risk of all-cause mortality (13, 14). nondominant wrist, as described previously (18). In brief,
In addition, abnormal sleep-wake cycles were associated with participants were asked to wear the actigraph unit for 7 con-
a 3-times higher mortality rate in elderly persons older than secutive days and nights and to remove it only before bathing.
85 years of age (15). These studies assessed 24-hour activity Recordings were obtained in 30-second intervals. Because
rhythms in very old persons and had relatively short follow- the elderly can have less distinct circadian rhythms, subjects’
up periods (average follow-up of 4.1, 3.5, and 2 years, respec- 24-hour activity rhythms were analyzed nonparametrically;
tively). In these activity rhythm and mortality studies, a few thus, no assumptions were made about the underlying shape
subjective sleep parameters were taken into account. One of the circadian rhythm. The actigraph unit was used to cal-
study considered sleep medication and disturbed sleep due culate two 24-hour activity rhythm variables (interday stabil-
to pain as potential confounders (14). Anderson et al. (15) ity and intraday variability (19)) and 3 sleep variables (sleep
assessed sleep quality using the Pittsburgh Sleep Quality duration, sleep onset latency, and waking after sleep onset
Index (PSQI) and daytime sleepiness using the Epworth (20)). Interday stability is a measure of how stable the rhythm
Sleepiness Scale in relation to mortality but did not find an is over multiple days, that is, how similar the individual day-
association. However, none of these studies accounted for night patterns are over time. It is calculated as the ratio of the
objectively assessed sleep characteristics, such as sleep dura- variance of the average activity patterns around the mean and
tion, sleep onset latency, and waking after sleep onset. In a the overall variance (21). Intraday variability reflects the frag-
large population-based study with a longer follow-up period, mentation of the rhythm, that is, the rate of shifting between
we aimed to evaluate the association between nonparametric rest and activity. It is calculated as the ratio between the mean
measures of the 24-hour activity rhythm, stability, and frag- squares of the difference between all successive hours (first
mentation, and mortality in middle-aged and elderly people. derivative) and the mean squares around the grand mean
We also ran analyses in which we excluded deaths that oc- (21). The variables have been shown to be sensitive in obser-
curred in the first 2 years to minimize the risk of reversed cau- vational and experimental studies on aging (22, 23). Exam-
sality. Furthermore, actigraphic and subjective sleep diary ples of activity rhythms characterized by a high or low
estimates of sleep duration, sleep onset latency, waking after stability and fragmentation are given in Figure 1.
sleep onset, and sleep quality were studied in relation to mor-
tality to investigate whether circadian activity rhythms and Sleep diaries
sleep characteristics predicted mortality independently.
During the week of actigraphy assessment, each partici-
pant kept a sleep diary comprising data on sleep characteris-
METHODS tics, sleep quality, use of sleep medications, napping, and
Participants alcohol consumption. To assess subjective sleep duration
and sleep onset latency, participants were asked to answer
The present study was conducted within the Rotterdam the questions, “In total, how many hours did you sleep last
Study, a prospective study of persons 45 years of age or older night?” and “How long did it take you to fall asleep?” We
living in Rotterdam, The Netherlands, that started in 1990. averaged a weeks’ worth of daily values. Sleep quality was
The aim of the study was to examine the incidence of and measured using 3 dichotomous questions: “Do you think
risk factors for neurological, cardiovascular, psychiatric, and you slept well last night?”, “Do you think the amount of
other chronic diseases. Details of the study have been pub- sleep was not enough?”, and “Did you feel rested after getting
lished previously (16). The Rotterdam Study was approved out of bed?” We reverse-coded the answers to the second
by the medical ethics committee according to the Wet question so that a score of 1 indicated good sleep quality.
Bevolkingsonderzoek ERGO (Population Study Act Rotter- The score for sleep quality was created by summing the 3 di-
dam Study), executed by the Ministry of Health, Welfare and chotomous questions assessed each day (range, 0–3), taking
Sports of the Netherlands, and conforms to the Declaration the daily average of this score (range, 0–1), summing these
of Helsinki. All participants provided written informed scores over the days of participation, and taking into account
consent. the total number of days a person participated (range, 0–7).
From December 2004 to April 2007, we invited 2,632 suc- Higher scores represent a better sleep quality. Each day, par-
cessive participants to participate in the actigraphy study; of ticipants specified which sleep medications were used, if any.
 Circadian Activity Rhythms, Sleep, and Mortality 3

A)
1,500

Activity Counts per 30 Seconds

1,200

900

600

300

0 00
0: 0

00

00

00

00

00

00
:0

:0

:0

:0

:0

:0
0:

0:

0:
0:

0:

0:
12

12

12

12
18

12

12
Time of Day

Downloaded from http://aje.oxfordjournals.org/ at University of California, San Francisco on December 11, 2014
B)
1,500
Activity Counts per 30 Seconds

1,200

900

600

300

0
0: 0
00

00

00

00

00

0
:0

:0

:0

:0

:0

:0

0
0:

0:

0:
0:

0:

0:
12

12

12

12
18

12

12

Time of Day
C)
1,250
Activity Counts per 30 Seconds

1,000

750

500

250

0
0: 0
00

00
0

00

00

00

00

00

00
:0

:0

:0

:0

:0
:0
0:

0:

0:
0:

0:

0:
:
12

12

12

12
18

12

12

Time of Day
D)
1,250
Activity Counts per 30 Seconds

1,000

750

500

250

0
0: 0

00

00

0
00

00

00

00

0
:0

:0

:0

:0

:0
0

0
:

0:

0:

0:
0:

0:

0:
12

12

12
12
18

12

12

Time of Day

Figure 1. Four examples of activity plots, the Rotterdam Study, the Netherlands, 2004–2007. Plots show activity rhythms of participants from the
Rotterdam Study. The x-axis represents time (0:00, midnight; 12:00, noon) and the y-axis represents activity counts per 30 seconds. Participants
started wearing the actigraph unit at 18:00 hours on day 1. These activity counts are scaled relative to the individual means and cannot be compared
easily across persons. A) Activity rhythm with high stability and low fragmentation in a 73-year-old man; B) activity rhythm with high fragmentation in
an 84-year-old woman; C) activity rhythm with low stability in a 47-year old man; D) activity rhythm with low stability and high fragmentation in a
62-year-old man.
4 Zuurbier et al.

In all analyses, use of sleep medication was dichotomized variables were put into a separate category. Interday stability
into no use of sleep medication or any use of sleep medication and intraday variability were standardized and Winsorized at
during the week of actigraphy. Napping was evaluated by 4 standard deviations from the mean. We analyzed the curvi-
asking whether the participant had taken 1 or more naps. linear association between sleep duration and mortality by
The total number of days during which the participant took adding a squared term of sleep duration to the model. In ad-
a nap, adjusted for the total number of days for which the par- dition, we tested a nonlinear association by defining 3 catego-
ticipant contributed data, was used in analyses. Alcohol con- ries of sleep duration (<6, 6–7.5 (reference group), and >7.5
sumption was evaluated as the sum of cups of alcohol after hours).
18:00 hours during the week of actigraphy. Cox proportional hazards models were used to determine
the hazard ratios and 95% confidence intervals for the asso-
Pittsburgh Sleep Quality Index ciations of circadian rhythm and sleep parameters with mor-
tality. We included a covariate in the model if it changed the
The PSQI was used to measure subjective sleep quality estimate of the main determinants by more than 10%, if the

Downloaded from http://aje.oxfordjournals.org/ at University of California, San Francisco on December 11, 2014
(global PSQI score) and possible sleep apnea (24). Higher covariate predicted mortality (P < 0.05), or if it was an impor-
scores represent poorer sleep. We considered participants to tant a priori confounder. On the basis of these criteria, educa-
have sleep apnea if they reported that they snored loudly at tional level, employment status, and alcohol consumption
least 2 nights per week and if they reported occasional respi- were not included in the full model. We tested 2 different
ratory pauses or respiratory pauses during sleep at least 1–2 models. Model 1 was adjusted for age and sex. Model 2 was
nights per week (25). adjusted for age, sex, ADL, current smoking, diabetes, myo-
cardial infarction, stroke, cognitive functioning, depressive
Assessment of outcome symptoms, body mass index, use of sleep medication, pos-
sible sleep apnea, and napping. All statistical tests were
Records of general practitioners and hospitals were used to 2-sided, and a P value <0.05 was considered statistically sig-
continuously assess death from any cause. In addition, infor- nificant. We tested the proportional hazards assumption using
mation on vital status was acquired bimonthly from death cer- Schoenfeld residuals (29). The residuals did not significantly
tificates from the municipality. The number of person-years deviate from zero slope. Analyses were performed in SPSS,
was calculated from the date of actigraphy start to the date of version 20 (SPSS Inc., Chicago, Illinois).
death or end of follow-up on September 27, 2013. The mean
follow-up time was 7.3 years. RESULTS

Covariates Baseline characteristics

We assessed the following variables as possible confound- The average follow-up time for the 1,734 participants was
ers based on previous literature (14, 17): sex, age, use of sleep 7.3 (SD, 1.3) years. The mean age was 62.2 (SD, 9.3) years,
medication, possible sleep apnea, napping, activities of daily and 47% of subjects were male. In total, there were 154
living (ADL), educational level, cognitive functioning, de- deaths (8.9%) during the follow-up period. The participants’
pressive symptoms, body mass index, employment status, baseline characteristics stratified whether they were alive at
current smoking, alcohol consumption, myocardial infarc- the end of follow-up are summarized in Table 1.
tion, diabetes, and stroke. Use of sleep medication, napping, Interday stability and intraday variability were moderately
and alcohol consumption were estimated using the sleep di- correlated (r = −0.49, P < 0.001). The global PSQI score was
aries. Information on possible sleep apnea, ADL, educational moderately correlated with sleep quality as assessed using
level, depressive symptoms, employment status, and current a sleep diary (r = −0.45, P < 0.001). Circadian rhythm and
smoking (cigarettes, cigars, or pipe) was obtained in a home sleep variables were only weakly to mildly correlated, with
interview. ADL were measured with the Stanford Health As- the highest correlation between interday stability and objec-
sessment Questionnaire and were used to indicate general tive sleep duration (r = 0.31, P < 0.001). All correlations be-
health (26). Depressive symptoms were measured using the tween 24-hour activity rhythm and sleep parameters can be
Center for Epidemiologic Studies-Depression scale (27). found in Web Table 1 (available at http://aje.oxfordjournals.
During a visit to our research center, cognitive functioning org/).
was measured using the Mini Mental State Examination
(28); height and weight were measured with the participants Cox proportional hazards model
wearing light clothing and no shoes to calculate body mass
index (weight (kg)/height (m2)). Myocardial infarction, dia- Both circadian rhythm variables were significantly related
betes, and stroke were determined using medical records. to mortality (Table 2). After full adjustment, interday stability
was associated with a lower mortality risk ( per SD, hazard
Statistical analyses ratio (HR) = 0.83, 95% confidence interval (CI): 0.71, 0.96),
and intraday variability (i.e., fragmentation) was associated
The number of missing values for a variable never ex- with a higher mortality risk ( per SD, HR = 1.22, 95% CI:
ceeded 5% (the maximum amount of missing data was 4.2% 1.04, 1.44). To show the cumulative survival graphically,
for ADL). Missing values for continuous variables were re- interday stability and intraday variability were divided into
placed with the median, and missing values for categorical quartiles (Figure 2). Because interday stability and intraday
 Circadian Activity Rhythms, Sleep, and Mortality 5

Table 1. Baseline Characteristics of Study Participants by Status at End of Follow-up (n =1,734), the Rotterdam Study, the Netherlands, 2004–
2007

Total (n = 1,734) Status at End of Follow-up

Characteristic Alive (n = 1,580) Dead (n = 154) Test Statistica
Mean (SD) No. %
Mean (SD) No. % Mean (SD) No. %

Age, years 62.2 (9.3) 61.2 (8.6) 72.6 (10.4) −13.1b

Male sex 808 46.6 723 45.8 85 55.2 5.0c
Employed 574 33.1 558 35.3 16 10.4 39.9b
Educational level
Low 264 15.2 227 14.4 37 24.0
Intermediate 1,097 63.3 1,000 63.3 97 63.0 14.5d

Downloaded from http://aje.oxfordjournals.org/ at University of California, San Francisco on December 11, 2014
High 341 19.7 322 20.4 19 12.3
Activities of daily living score 0.29 (0.4) 0.25 (0.4) 0.66 (0.6) −8.3b
MMSE score 28.0 (1.8) 28.0 (1.7) 27.3 (1.9) 5.0b
CES-D scale score 5.5 (7.1) 5.4 (7.0) 6.6 (7.9) −2.1c
Myocardial infarction 67 3.9 52 3.3 15 9.7 15.7b
Diabetes 205 11.8 167 10.6 38 24.7 26.8b
Stroke 45 2.6 30 1.9 15 9.7 34.1b
e
Body mass index 27.9 (4.2) 27.9 (4.1) 27.6 (4.1) 0.82
Current smoker 358 20.6 328 20.8 30 19.5 0.29
Alcohol, cups per week 5.7 (1.3) 5.8 (7.3) 5.3 (6.9) 0.80
Interday stability score 0.80 (0.1) 0.80 (0.1) 0.77 (0.1) 3.2d
Intraday variability score 0.43 (0.1) 0.42 (0.1) 0.52 (0.2) −7.7b
Objective sleep duration, hours 6.4 (0.9) 6.4 (0.9) 6.4 (1.0) −0.58
Objective sleep duration, hours
<6 523 30.2 480 30.4 43 27.9
6–7.5 1,069 61.6 977 61.8 92 59.7 3.9
>7.5 142 8.2 123 7.8 19 12.3
Objective sleep onset latency, 14.6 (12.4) 13.8 (11.8) 22.6 (14.7) −8.6b
minutes
Objective waking after sleep onset, 69.5 (25.9) 68.9 (25.2) 74.5 (30.4) −2.6c
minutes
Sleep medication 252 14.5 218 13.8 34 22.1 8.6c
Apnea 507 29.2 448 28.4 59 38.3 6.8c
Napping, days per week 1.7 (2.0) 1.6 (2.0) 2.7 (2.4) −5.6b
Subjective sleep duration, hours 6.9 (0.9) 6.9 (0.9) 6.8 (1.1) 0.40
Subjective sleep duration, hours
<6 277 16.0 244 15.4 33 21.4 4.1
6–7.5 1,048 60.4 964 61.0 84 54.5
>7.5 409 23.6 372 23.5 37 24.0
Subjective sleep onset latency, minutes 17.7 (11.7) 17.7 (11.7) 17.8 (11.7) −0.09
Sleep quality sleep diary score 5.5 (1.6) 5.6 (1.6) 5.5 (1.6) 0.74
Global PSQI score 3.7 (3.5) 3.6 (3.5) 3.9 (3.7) −0.78
Poor sleep (PSQI score >5) 398 23.0 353 22.3 45 29.2 −1.80

Abbreviations: CES-D, Center for Epidemiologic Studies-Depression; MMSE, Mini Mental State Examination; PSQI, Pittsburgh Sleep Quality
Index; SD, standard deviation.
a
Statistical test to compare the means of the alive or dead status at end of follow-up. Student t test was used for continuous variables and χ2 was
used for categorical variables.
b
P < 0.01.
c
P < 0.05.
d
P < 0.001.
e
Weight (kg)/height (m)2.
6 Zuurbier et al.

Table 2. Associations of 24-Hour Activity Rhythm and Sleep Parameters With All-Cause Mortality in Study
Participants (n = 1,734), the Rotterdam Study, the Netherlands, 2004–2013

Adjusted for Age and Sex Fully Adjusteda

Determinant
HR 95% CI HR 95% CI

Activity rhythm
Interday stability score 0.75b 0.65, 0.86 0.83c 0.71, 0.96
b
Intraday variability score 1.37 1.20, 1.57 1.22c 1.04, 1.44
Objectively assessed sleep
Continuous sleep duration
Sleep duration, hours 0.42 0.11, 1.57 0.69 0.19, 2.53
Sleep duration squared, hours2 1.06 0.95, 1.18 1.02 0.92, 1.13

Downloaded from http://aje.oxfordjournals.org/ at University of California, San Francisco on December 11, 2014
Categorical sleep duration, hours
<6 1.29 0.89, 1.87 1.12 0.77, 1.65
6–7.5 0.00 Referent 0.00 Referent
>7.5 hours 1.33 0.81, 2.18 1.18 0.70, 1.98
Sleep onset latency, minutes 1.01c 1.00, 1.02 1.01 1.00, 1.02
Waking after sleep onset, minutes 1.01c 1.00, 1.01 1.01 1.00, 1.01
Subjectively assessed sleep
Continuous sleep duration
Sleep duration, hours 0.23c 0.07, 0.74 0.40 0.11, 1.39
2 c
Sleep duration squared, hours 1.12 1.02, 1.22 1.07 0.97, 1.17
Categorical sleep duration, hours
<6 1.45 0.97, 2.17 1.41 0.93, 2.13
6–7.5 0.00 Referent 0.00 Referent
>7.5 1.13 0.77, 1.67 1.10 0.74, 1.64
Sleep onset latency, minutes 0.99 0.98, 1.01 0.99 0.98, 1.01
Sleep quality sleep diary score 0.92 0.83, 1.02 0.97 0.87, 1.09
Sleep quality PSQI score 1.01 0.97, 1.06 0.99 0.94, 1.04
Poor sleep (PSQI score >5) 1.28 0.90, 1.83 1.13 0.76, 1.69

Abbreviations: CI, confidence interval; HR, hazard ratio; PSQI, Pittsburgh Sleep Quality Index.
a
Adjusted for age, sex, activities of daily living score, current smoking, diabetes, myocardial infarction, stroke,
cognitive functioning, depressive symptoms, body mass index, sleep medication, napping, and apnea.
b
P < 0.001.
c
P < 0.05.

variability were moderately correlated, we also analyzed a results did not change meaningfully (Web Table 2). Other sub-
model adjusted for age and sex that included both 24-hour jective sleep parameters were not related to mortality (Table 2).
activity rhythm variables (per SD of intraday variability, HR =
1.25, 95% CI: 1.07, 1.47; per SD of interday stability, HR = Sensitivity analysis
0.84, 95% CI: 0.71, 1.00).
Actigraphically measured sleep onset latency and waking Because circadian rhythms can be influenced by undiag-
after sleep onset were marginally related to mortality in an nosed morbidity, we also ran the analyses excluding deaths
age- and sex-adjusted model, but these associations were that occurred in the first year and in the first 2 years after
nonsignificant in the fully adjusted analysis ( per minute of the week of actigraphy. These exclusions reduce the possible
sleep onset latency, HR = 1.01, 95% CI: 1.00, 1.02; per mi- effect of reversed causality. In these analyses, 148 and 128
nute of waking after sleep onset, HR = 1.01, 95% CI: 1.00, deaths occurred during the remaining follow-up periods,
1.01). The actigraphic estimates of sleep duration, both con- respectively. In the fully adjusted model that excluded deaths
tinuous and categorical, were not related to mortality. Subjec- in the first year, the observed hazard ratio was essentially
tive sleep duration was quadratically associated with mortality unchanged compared with the previous analyses in which
in the age- and sex-adjusted model. However, this association all participants were included ( per SD of interday stability,
was not significant after further adjustment ( per hour, HR = HR = 0.83, 95% CI: 0.71, 0.97; per SD of intraday variabil-
1.07, 95% CI: 0.97, 1.17). When different cut-offs were cho- ity, HR = 1.23, 95% CI: 1.05, 1.45). Again, very similar
sen to categorize objective and subjective sleep duration, the results were observed when deaths that occurred in the first
 Circadian Activity Rhythms, Sleep, and Mortality 7

A) and might differ per level of circadian organization (12).

100
For example, changes may occur in the suprachiasmatic nu-
cleus (the central clock of the brain), in peripheral oscillators,
Cumulative Survival, %

95 or in the ability of the suprachiasmatic nucleus to drive pe-

ripheral oscillators. In humans, postmortem studies demon-
90 strated a reduction in the number of vasopressin-expressing
neurons in the suprachiasmatic nucleus at old age (>80
85 Quartile 1
years) (30). This may underlie a smaller amplitude of sev-
Quartile 2
eral circadian rhythms, a more fragmented 24-hour activity
80 Quartile 3 rhythm, and a temperature and melatonin phase that occur
Quartile 4 earlier than in younger people (12, 30–35). This loss of tem-
0 poral organization between different rhythms can lead to sub-
0 2 4 6 8
optimal physiological functioning because physiological
Follow-Up Time, years

Downloaded from http://aje.oxfordjournals.org/ at University of California, San Francisco on December 11, 2014
processes do not all take place at their optimal time of day
B)
100 (36). As a result, people who suffer from temporal disorgani-
zation have a higher susceptibility to disease. Second, nap-
ping increases the fragmentation of the rhythm (17) and
Cumulative Survival, %

95
can also be an indicator of bad health (37). However, previ-
90 ous literature on the association of napping with mortality
were inconsistent. It was found that people who take naps
85 regularly might have a higher mortality rate (38, 39), espe-
cially those who sleep more than 9 hours per night (40).
80 On the other hand, another study found no significant benefit
or harm of napping (41), whereas yet another study found a
0
protective association between napping and mortality for
0 2 4 6 8 people with short sleep durations (40). In these studies, cir-
Follow-Up Time, years cadian rhythm parameters were not taken into account. In
our study, self-reported naps could not explain the associa-
Figure 2. Crude cumulative survival plots per quartile of A) interday tion between the stability and fragmentation of the 24-hour
stability and B) intraday variability, the Rotterdam Study, the Nether- activity rhythm and mortality. Third, the disturbed 24-hour
lands, 2004–2013. Quartile 1 is the lowest quartile and quartile 4 is
the highest. Survival was lower in participants with a low interday activity rhythm might be an indicator of poor health. Occur-
stability or a high intraday variability (fragmentation). rence of disease has been related to disrupted circadian
rhythms, for example in persons with cardiovascular disease
and Alzheimer’s disease (21, 42). Also, more fragmented and
less stable 24-hour activity rhythms have been related to
sleepiness, depression, cognitive deficits, high body mass
2 years were excluded ( per SD of interday stability, HR = index, smoking, high blood pressure, and obesity (17, 23,
0.86, 95% CI: 0.73, 1.02; per SD of intraday variability, 43–45).
HR = 1.18, 95% CI: 0.98, 1.41). Although we controlled for several health measures, such
as ADL, depressive symptoms, and diabetes, residual con-
DISCUSSION founding by disease might explain part of the results. We
also ran analyses in which we excluded deaths that occurred
In the present prospective, population-based cohort study, in the first 2 years to test for reverse causality. The observed
more fragmented and less stable 24-hour activity rhythms hazard ratio was very similar to that from the analyses that
were associated with a 20% increase in all-cause mortality included all participants. This suggests that the association
risk in a middle-aged and elderly population. These asso- between the 24-hour activity rhythm and mortality is not ex-
ciations remained after adjustment for health parameters, pos- clusively driven by short-term mortality.
sible sleep apnea, and napping. After adjustment for age and Circadian rhythms and sleep patterns change as people
sex only, subjective sleep duration showed a U-shaped mar- age. For example, elderly people sleep less during the night
ginal association with mortality risk. However, after full ad- and tend to fall asleep and wake up earlier (2–6). Neverthe-
justment, no sleep parameter, whether estimated objectively less, in the present study, the associations between sleep and
using an actigraph unit or subjectively using a sleep diary, 24-hour activity rhythm parameters were weak. We found
predicted all-cause mortality. This suggests that although that more fragmented and less stable circadian activity rhythms
the circadian rhythm and sleep both change during aging, predicted mortality but none of the sleep variables, whether
the circadian rhythm is independently related to mortality. measured objectively or subjectively, predicted mortality
Our finding that fragmentation and low stability of the in the fully adjusted model. During our follow-up period,
24-hour activity rhythm predict mortality has several possible 154 participants died, which implies that we had sufficient
explanations. First, the biological aging processes may be in- power to detect the moderate associations of 24-hour activity
volved. Although our analyses were adjusted for age, age- rhythm parameters on mortality. Arguably, our study might
related changes to the circadian organization are complex not have been powered to detect mild associations between
8 Zuurbier et al.

sleep characteristics and mortality. Yet, the findings suggest such as restless leg syndrome and sleep apnea, was limited.
that the circadian rhythm is more strongly and independently Second, in our population-based study, 154 participants
associated with mortality than is sleep duration. died. Therefore, we may not have been able to detect the
There have been few studies on the association between ac- mild associations of sleep parameters on mortality.
tivity rhythms and mortality (13–15, 46, 47). Mortality risk To conclude, in a representative middle-aged and elderly
was found to be higher in older men and women with less population, fragmentation and low stability of the 24-hour ac-
robust or abnormal 24-hour activity rhythms (13–15). In pa- tivity rhythm predicted all-cause mortality independent of
tients with metastatic colorectal cancer, a higher mortality and better than sleep estimates. Changes in the regulation
rate was associated with disturbed circadian rhythms; in of circadian rhythms could indicate disease and reflect age-
dementia patients, it was associated with abnormal timing related alterations in the biological clock of the brain. Future
of the rhythm (46, 47). To our knowledge, the association of research must show whether improving circadian activity
fragmentation and stability of the circadian activity rhythm rhythm disturbances can improve health and survival.
with all-cause mortality has not been described before.

Downloaded from http://aje.oxfordjournals.org/ at University of California, San Francisco on December 11, 2014
Our estimates showed a significant U-shaped relationship
between continuously analyzed subjective sleep duration and
mortality. However, these associations were nonsignificant ACKNOWLEDGMENTS
after adjustment for health parameters. A U-shaped relation-
Author affiliations: Department of Epidemiology, Erasmus
ship between sleep duration and mortality was found in sev-
Medical Center, Rotterdam, The Netherlands (Lisette
eral previous studies (for 2 meta-analyses, see Cappucio et al.
A. Zuurbier, Annemarie I. Luik, Albert Hofman, Oscar H.
(9) and Gallicchio et al. (10)). In general, stronger associa-
Franco, Henning Tiemeier); Department of Sleep and Cogni-
tions between long sleep duration and all-cause mortality
tion, Netherlands Institute for Neuroscience, an Institute
were observed in the more extreme sleep categories (≥9
of the Netherlands Royal Academy of Arts and Sciences,
hours). In our study, few persons were extreme short or long
Amsterdam, The Netherlands (Eus J. W. Van Someren); De-
sleepers. Consequently, in our analyses, the power regarding
partment of Integrative Neurophysiology, Center for Neuro-
extreme sleep duration was limited, which might explain the
genomics and Cognitive Research, Neuroscience Campus
attenuation of the curvilinear relation between sleep duration
Amsterdam, VU University, Amsterdam, The Netherlands
and mortality after further adjustment. Previous studies in
(Eus J. W. Van Someren); Department of Medical Psy-
which sleep duration stratified on health status was examined
chology, VU University Medical Center, Amsterdam, The
were inconsistent (48, 49). In 1 study, Magee et al. (48) ob-
Netherlands (Eus J. W. Van Someren); Department of Child
served an association between sleep duration and mortality
and Adolescent Psychiatry, Erasmus Medical Center, Rotter-
in persons with preexisting disease only; Mesas et al. (49)
dam, The Netherlands (Henning Tiemeier); and Department
also found this association in healthy people. In addition,
of Psychiatry, Erasmus Medical Center, Rotterdam, The
part of the association of short sleep duration with mortality
Netherlands (Henning Tiemeier).
can be explained by sleep apnea (50). We adjusted for possi-
This work was supported by a Netherlands Organization
ble sleep apnea based on 2 questions from the PSQI (24, 25).
for Scientific Research grant (017.106.370) awarded to
Possible apnea was not a significant predictor of mortality in
H.T. The Rotterdam Study is funded by Erasmus Medical
our fully adjusted model. However, the PSQI cannot be used
Center and Erasmus University Rotterdam; Netherlands
to diagnose sleep apnea. We cannot rule out that sleep apnea,
Organization for the Health Research and Development;
if assessed more in detail, might explain part of the observed
the Research Institute for Diseases in the Elderly; the Minis-
associations.
try of Education, Culture and Science; the Ministry for
In the present study, as in previous studies, perceived sleep
Health, Welfare and Sports; and the European Commission.
quality was not related to all-cause mortality, whether it was
O.H.F. works in ErasmusAGE, a center for aging research
measured using the sleep diary or with the PSQI (51). Previ-
across the life course funded by Nestlé Nutrition (Nestec
ously, sleep disturbances were associated with higher all-
Ltd.), Metagenics Inc., and AXA.
cause mortality risk only in men younger than 45 years of
We thank the staff of the Rotterdam Study and the partic-
age and not in women or men older than 45 years (52).
ipating general practitioners and pharmacists.
One strength of our study is that it is embedded in the
Nestlé Nutrition (Nestec Ltd.), Metagenics Inc., and AXA
Rotterdam Study, a prospective population-based cohort
had no role in design and conduct of the study; collection,
study. This increases the generalizability of our results, and
management, analysis, and interpretation of the data; and
we were able to adjust for many covariates. A second strength
preparation, review or approval of the manuscript.
was our use of both subjective and objective measurements to
Conflict of interest: none declared.
estimate sleep duration and sleep onset latency. Because sub-
jective and objective sleep variables are not strongly associ-
ated (18), it is important to analyze both and to test whether
they predict mortality independently. A third strength was the
complete follow-up of the death date of all participants. REFERENCES
Fourth, the 24-hour activity rhythms were analyzed nonpara- 1. Van Someren EJ. Circadian and sleep disturbances in the
metrically, so no assumptions were made about the underly- elderly. Exp Gerontol. 2000;35(9-10):1229–1237.
ing shape of their circadian rhythms. This study also has 2. Avidan AY. Sleep changes and disorders in the elderly patient.
some limitations. First, data collection on sleep disorders, Curr Neurol Neurosci Rep. 2002;2(2):178–185.
 Circadian Activity Rhythms, Sleep, and Mortality 9

3. Espiritu JR. Aging-related sleep changes. Clin Geriatr Med. 23. Oosterman JM, van Someren EJ, Vogels RL, et al.
2008;24(1):1–14. Fragmentation of the rest-activity rhythm correlates with
4. Huang YL, Liu RY, Wang QS, et al. Age-associated difference age-related cognitive deficits. J Sleep Res. 2009;18(1):
in circadian sleep-wake and rest-activity rhythms. Physiol 129–135.
Behav. 2002;76(4-5):597–603. 24. Buysse DJ, Reynolds CF 3rd, Monk TH, et al. The Pittsburgh
5. Monk TH. Aging human circadian rhythms: conventional Sleep Quality Index: a new instrument for psychiatric practice
wisdom may not always be right. J Biol Rhythms. 2005;20(4): and research. Psychiatry Res. 1989;28(2):193–213.
366–374. 25. Fogelholm M, Kronholm E, Kukkonen-Harjula K, et al.
6. Buysse DJ, Browman KE, Monk TH, et al. Napping and Sleep-related disturbances and physical inactivity are
24-hour sleep/wake patterns in healthy elderly and young independently associated with obesity in adults. Int J Obes
adults. J Am Geriatr Soc. 1992;40(8):779–786. (Lond). 2007;31(11):1713–1721.
7. Yaggi HK, Araujo AB, McKinlay JB. Sleep duration as a risk 26. Fries JF, Spitz P, Kraines RG, et al. Measurement of
factor for the development of type 2 diabetes. Diabetes Care. patient outcome in arthritis. Arthritis Rheum. 1980;23(2):
2006;29(3):657–661. 137–145.
8. Hoevenaar-Blom MP, Spijkerman AM, Kromhout D, et al. 27. Radloff LS. The CES-D scale: a self-report depression scale for

Downloaded from http://aje.oxfordjournals.org/ at University of California, San Francisco on December 11, 2014
Sleep duration and sleep quality in relation to 12-year research in the general population. Appl Psychol Meas. 1977;
cardiovascular disease incidence: the MORGEN Study. Sleep. 1(3):385–401.
2011;34(11):1487–1492. 28. Folstein MF, Folstein SE, McHugh PR. “Mini-mental
9. Cappuccio FP, D’Elia L, Strazzullo P, et al. Sleep duration and state”. A practical method for grading the cognitive state
all-cause mortality: a systematic review and meta-analysis of of patients for the clinician. J Psychiatr Res. 1975;12(3):
prospective studies. Sleep. 2010;33(5):585–592. 189–198.
10. Gallicchio L, Kalesan B. Sleep duration and mortality: a 29. Schoenfeld D. Partial residuals for the proportional hazards
systematic review and meta-analysis. J Sleep Res. 2009;18(2): regression model. Biometrika. 1982;69(1):239–241.
148–158. 30. Swaab DF, Fliers E, Partiman TS. The suprachiasmatic nucleus
11. Hofman MA, Swaab DF. Alterations in circadian rhythmicity of of the human brain in relation to sex, age and senile dementia.
the vasopressin-producing neurons of the human Brain Res. 1985;342(1):37–44.
suprachiasmatic nucleus (SCN) with aging. Brain Res. 1994; 31. Yoon IY, Kripke DF, Elliott JA, et al. Age-related changes of
651(1-2):134–142. circadian rhythms and sleep-wake cycles. J Am Geriatr Soc.
12. Yamazaki S, Straume M, Tei H, et al. Effects of aging on central 2003;51(8):1085–1091.
and peripheral mammalian clocks. Proc Natl Acad Sci U S A. 32. Czeisler CA, Dumont M, Duffy JF, et al. Association of
2002;99(16):10801–10806. sleep-wake habits in older people with changes in output of
13. Paudel ML, Taylor BC, Ancoli-Israel S, et al. Rest/activity circadian pacemaker. Lancet. 1992;340(8825):933–936.
rhythms and mortality rates in older men: MrOS Sleep Study. 33. Duffy JF, Zeitzer JM, Rimmer DW, et al. Peak of
Chronobiol Int. 2010;27(2):363–377. circadian melatonin rhythm occurs later within the sleep of
14. Tranah GJ, Blackwell T, Ancoli-Israel S, et al. Circadian older subjects. Am J Physiol Endocrinol Metab. 2002;282(2):
activity rhythms and mortality: the Study of Osteoporotic E297–E303.
Fractures. J Am Geriatr Soc. 2010;58(2):282–291. 34. Van Someren EJ, Riemersma-Van Der Lek RF. Live to the
15. Anderson KN, Catt M, Collerton J, et al. Assessment of sleep rhythm, slave to the rhythm. Sleep Med Rev. 2007;11(6):
and circadian rhythm disorders in the very old: the Newcastle 465–484.
85+ Cohort Study. Age Ageing. 2014;43(1):57–63. 35. Harper DG, Stopa EG, Kuo-Leblanc V, et al. Dorsomedial SCN
16. Hofman A, Darwish Murad S, van Duijn CM, et al. The neuronal subpopulations subserve different functions in human
Rotterdam Study: 2014 objectives and design update. Eur J dementia. Brain. 2008;131(Pt 6):1609–1617.
Epidemiol. 2013;28(11):889–926. 36. Gibson EM, Williams WP 3rd, Kriegsfeld LJ. Aging in the
17. Luik AI, Zuurbier LA, Hofman A, et al. Stability and circadian system: considerations for health, disease prevention
fragmentation of the activity rhythm across the sleep-wake and longevity. Exp Gerontol. 2009;44(1-2):51–56.
cycle: the importance of age, lifestyle, and mental health. 37. Asplund R. Daytime sleepiness and napping amongst the
Chronobiol Int. 2013;30(10):1223–1230. elderly in relation to somatic health and medical treatment.
18. Van Den Berg JF, Van Rooij FJ, Vos H, et al. Disagreement J Intern Med. 1996;239(3):261–267.
between subjective and actigraphic measures of sleep duration 38. Bursztyn M, Stessman J. The siesta and mortality: twelve years
in a population-based study of elderly persons. J Sleep Res. of prospective observations in 70-year-olds. Sleep. 2005;28(3):
2008;17(3):295–302. 345–347.
19. Witting W, Kwa IH, Eikelenboom P, et al. Alterations in the 39. Stone KL, Ewing SK, Ancoli-Israel S, et al. Self-reported sleep
circadian rest-activity rhythm in aging and Alzheimer’s disease. and nap habits and risk of mortality in a large cohort of older
Biol Psychiatry. 1990;27(6):563–572. women. J Am Geriatr Soc. 2009;57(4):604–611.
20. Kushida CA, Chang A, Gadkary C, et al. Comparison of 40. Cohen-Mansfield J, Perach R. Sleep duration, nap habits, and
actigraphic, polysomnographic, and subjective assessment of mortality in older persons. Sleep. 2012;35(7):1003–1009.
sleep parameters in sleep-disordered patients. Sleep Med. 2001; 41. Lan TY, Lan TH, Wen CP, et al. Nighttime sleep, Chinese
2(5):389–396. afternoon nap, and mortality in the elderly. Sleep. 2007;30(9):
21. van Someren EJ, Hagebeuk EE, Lijzenga C, et al. Circadian 1105–1110.
rest-activity rhythm disturbances in Alzheimer’s disease. Biol 42. Paudel ML, Taylor BC, Ancoli-Israel S, et al. Rest/activity
Psychiatry. 1996;40(4):259–270. rhythms and cardiovascular disease in older men. Chronobiol
22. Scherder EJ, Van Someren EJ, Swaab DF. Transcutaneous Int. 2011;28(3):258–266.
electrical nerve stimulation (TENS) improves the rest-activity 43. Maaskant M, van de Wouw E, van Wijck R, et al. Circadian
rhythm in midstage Alzheimer’s disease. Behav Brain Res. sleep-wake rhythm of older adults with intellectual disabilities.
1999;101(1):105–107. Res Dev Disabil. 2013;34(4):1144–1151.
10 Zuurbier et al.

44. Matthews KA, Kamarck TW, Hall HM, et al. Blood pressure 48. Magee CA, Holliday EG, Attia J, et al. Investigation of the
dipping and sleep disturbance in African-American and relationship between sleep duration, all-cause mortality, and
Caucasian men and women. Am J Hypertens. 2008;21(7): preexisting disease. Sleep Med. 2013;14(7):591–596.
826–831. 49. Mesas AE, López-García E, León-Muñoz LM, et al. Sleep
45. van den Berg JF, Knvistingh Neven A, Tulen JH, et al. duration and mortality according to health status in older adults.
Actigraphic sleep duration and fragmentation are related to J Am Geriatr Soc. 2010;58(10):1870–1877.
obesity in the elderly: the Rotterdam Study. Int J Obes (Lond). 50. Grandner MA, Hale L, Moore M, et al. Mortality associated
2008;32(7):1083–1090. with short sleep duration: the evidence, the possible
46. Gehrman P, Marler M, Martin JL, et al. The timing of activity mechanisms, and the future. Sleep Med Rev. 2010;14(3):
rhythms in patients with dementia is related to survival. 191–203.
J Gerontol A Biol Sci Med Sci. 2004;59(10):1050–1055. 51. Hublin C, Partinen M, Koskenvuo M, et al. Sleep and mortality:
47. Mormont MC, Waterhouse J, Bleuzen P, et al. Marked 24-h a population-based 22-year follow-up study. Sleep. 2007;
rest/activity rhythms are associated with better quality of life, 30(10):1245–1253.
better response, and longer survival in patients with metastatic 52. Rod NH, Vahtera J, Westerlund H, et al. Sleep disturbances and
colorectal cancer and good performance status. Clin Cancer cause-specific mortality: results from the GAZEL cohort study.

Downloaded from http://aje.oxfordjournals.org/ at University of California, San Francisco on December 11, 2014
Res. 2000;6(8):3038–3045. Am J Epidemiol. 2011;173(3):300–309.