Sex Differences in Subjective and Actigraphic Sleep Measures: A Population-Based Study of Elderly Persons

Abstract

Study Objectives:

To investigate and explain sex differences in subjective and actigraphic sleep parameters in community-dwelling elderly persons.

Design:

Cross-sectional study.

Setting:

The study was embedded in the Rotterdam Study, a population-based study.

Participants:

Nine hundred fifty-six participants aged 59 to 97 years.

Interventions:

N/A.

Measurements and Results:

Participants wore an actigraph and kept a sleep diary for an average of 6 consecutive nights. Subjective sleep quality was assessed with the Pittsburgh Sleep Quality Index. Unadjusted sex differences in sleep parameters were assessed with t tests. Women reported shorter total sleep time, a less favorable sleep-onset latency, lower sleep efficiency, and worse global sleep quality, as compared with men. When assessed with actigraphy, however, women were found to have longer and less-fragmented sleep than men. Sex differences in diary-reported sleep duration and other subjective sleep parameters were attenuated by adjustment for marital status, the use of sleep medication, and other covariates, but all sex differences remained significant in a multivariate-adjusted model. Sex differences in actigraphic sleep parameters were barely attenuated by multivariate adjustment, although the shorter actigraphically measured sleep duration in men was partly explained by their higher alcohol consumption. Some covariates (eg, sleep medication) had a different relationship with diary-reported or actigraphic total sleep time in men and women.

Conclusions:

If assessed by diary or interview, elderly women consistently reported shorter and poorer sleep than elderly men. In contrast, actigraphic sleep measures showed poorer sleep in men. These discrepancies are partly explained by determinants of sleep duration, such as sleep medication use and alcohol consumption.

Citation:

van den Berg JF; Miedema HME; Tulen JHM; Hofman A; Knuistingh Neven A; Tiemeier H. Sex differences in subjective and actigraphic sleep measures: a population-based study of elderly persons. SLEEP 2009;32(10):1367-1375.

EPIDEMIOLOGIC STUDIES HAVE CONSISTENTLY SHOWN THAT WOMEN HAVE MORE SLEEP-RELATED COMPLAINTS AND A HIGHER RISK OF INSOMNIA THAN men.^1–5 Zhang et al.⁴ stated that the trend of female predisposition for insomnia was consistent across numerous studies and progressive across age. These sex differences are also pervasive in neuropsychiatric disorders, such as depression, that are strongly related to disturbed sleep.⁶ Explaining sex differences in sleep parameters might shed light on the etiology of sleep complaints. However, when sleep parameters such as sleep duration are taken into account, findings are less consistent: some authors have found that women reported longer total sleep time (TST) than men, in studies performed in Sweden, United Kingdom, and Norway,^1,3,7 whereas others, in the Netherlands and the United Kingdom, found no substantial sex differences in self-reported TST.^2,8 Not only is sex related to self-report measures of sleep and sleep quality, but it is also an important determinant of objectively measured sleep parameters. Redline et al.,⁹ in a study in the United States, noted that sex explained the largest proportion of the variance in each sleep-architecture measure they investigated. A meta-analysis of 65 studies that investigated sleep in convenience samples of healthy participants with objective measurement methods showed that women had a modestly longer TST but also a longer sleep-onset latency (SOL) than men and no difference in sleep efficiency.¹⁰

Studies that have investigated sex differences in objectively measured sleep parameters have rarely been community based. As a result of recruitment methods and inclusion criteria, study populations often represent a limited range of demographic and comorbid conditions, and the results may not be representative of the population. One exception is a large community-based actigraphy study by Lauderdale et al.¹¹ They found longer sleep durations and a higher sleep efficiency in women than in men. However, their study did not include elderly persons, they did not investigate the effect of psychiatric disorders or sleep medication, and they did not present data obtained by self-report measures.

We investigated sex differences in self-reported and objectively measured sleep parameters in a large study of community-dwelling elderly persons. In this study, we used wrist actigraphy, which is relatively unobtrusive and enabled the participants to stay in their natural environment and adhere to their normal sleep habits. Our study also examined the extent to which the observed sex differences in sleep patterns were attributable to a range of demographic, health-related, and sleep-related variables, as this is still poorly understood. We hypothesize that sex differences in sleep parameters correspond to differences in mental health because depression and anxiety disorders are consistently found to be more common in women and these disorders are strongly related to sleep disturbance.^6,12–14 Demographic variables such as socioeconomic status, as well as physical health, have also been linked to subjective and objective sleep parameters^{7,11,13,15,16}; these variables may also account for part of the sex differences in sleep parameters. Another possibility is that sex differences arise from the measurement method. With self-report measures, men and women may report differently on the same amount of sleep. Therefore, in the present study, we relied on self-report as well as—objective—actigraphic measures.

METHODS

Study Population

This study is embedded in the Rotterdam Study, a population-based cohort study aimed at assessing the occurrence of and risk factors for chronic diseases in the elderly.¹⁷ In 1990, all of the inhabitants of a district of Rotterdam aged 55 years and older were invited to participate. In 2000, the study population was extended with a second cohort of people aged 55 years and older. From May 2004 to December 2005, these participants underwent their second examination, consisting of a home interview and 2 visits to the research center. From December 2004 onward, 1515 persons were asked to participate in the actigraphy study, 1076 (71.0%) of whom agreed. In the present study, we included only subjects who contributed at least 2 valid nights of actigraphy, at least 2 valid nights of diary data, and a valid score on the Pittsburgh Sleep Quality Index (PSQI),¹⁸ which was administered during the home interview. A total of 956 subjects fulfilled these criteria. Together they contributed 5895 valid nights: on average 6.2 (SD = 1.1) nights per participant. The Medical Ethics Committee of the Erasmus University Rotterdam approved the Rotterdam Study, and written informed consent was obtained from all of the participants.

Assessment of Sleep Parameters

We used the Actiwatch model AW4 (Cambridge Neurotechnology Ltd., Cambridge, UK), an actigraph that can be worn like a watch and is equipped with an event-marker button. Participants were instructed to wear the actigraph over a period of 5 to 7 consecutive days and nights, on the nondominant wrist while continuing their normal activities and sleep-wake rhythms in their home environment. During the actigraphy study period, approximately 3 weeks after the home interview, participants kept a sleep diary, in which they indicated, among other things, their estimated TST and SOL of the night before. Participants were asked to press the event-marker button on the actigraph each night when they began trying to fall asleep and again when they got out of bed each morning. To calculate sleep parameters from the raw actigraphy data, we used the Actiwatch algorithm that has been validated against polysomnography by Kushida et al.¹⁹ With this algorithm, a score is calculated for each 30-second epoch, taking into account the weighted value of previous and following epochs. We used a threshold of 20 to distinguish sleep from waking, since this high-sensitivity setting yielded the best agreement with polysomnography with regard to TST in Kushida et al.'s validation study.¹⁹

We applied the following rules to the data:

• Bedtime and arising time were derived from the event-marker buttons, and, if these data were not present for a certain night, we derived them from the sleep diary to determine sleep start and sleep end.

• Sleep start was defined using the first immobile period (no more than one 30-second epoch of movement) of at least 10 minutes after bedtime. The midpoint of that period was classified as sleep start.

• To define sleep end, we identified the last period of at least 10 minutes of immobility before waking time that had no more than 1 epoch of movement. The last epoch of this period was classified as sleep end.

• TST is the time between sleep start and sleep end, minus the time classified as awake by the algorithm.

• Sleep efficiency is calculated as 100% times the TST divided by the time between bedtime and rising time.

The definitions of sleep start and sleep end were derived from the Actiwatch manual and are equal to those used by the Actiwatch software.²⁰ The fragmentation index is a measure of the amount of interruption of sleep by physical movement. It is calculated as follows: 100 times the number of groups (1 or more) of consecutive immobile 30-second epochs divided by the total number of immobile epochs.²⁰

Subjective sleep quality was assessed with the PSQI.¹⁸ The PSQI is a self-rating questionnaire that measures sleep quality and disturbance retrospectively over a 1-month period, resulting in a global score between 0 and 21, with higher scores indicating poorer sleep quality.

Assessment of Covariates

Age, marital status (married/other), education attainment, employment (paid job for ≥ 20 hours per week), smoking, bereavement, the use of psychoactive medication (other than sleep medication), and joint pain were assessed in the home interview. Education attainment was categorized into “low” (primary education only), “intermediate” (secondary education or lower/intermediate vocation education) and “high” (higher vocation education or university). Weight and height were measured, and body mass index was calculated. Sleep medication use and coffee and alcohol consumption during the study period were ascertained with the sleep diary. To operationalize the occurrence of probable sleep apnea, 2 questions from the PSQI were used. In line with Fogelholm et al.,²¹ sleep apnea was considered probable in persons who reported (1) loud snoring at least 2 nights a week, with at least occasional respiratory pauses, or (2) respiratory pauses during sleep with a frequency of at least 1 to 2 nights per week. Depressive symptoms were assessed with the Center for Epidemiologic Studies Depression (CES-D) scale.²² The CES-D is a self-report scale with 20 items, with a maximum score of 60. Scores of 16 or greater on the CES-D are interpreted as suggestive of clinically significant depression. Cognitive function was assessed using the Mini Mental-State Examination.²³ Scores on this test range from 0 to 30, with higher scores indicating a better cognitive performance. We used a slightly adapted Munich version of the Composite International Diagnostic Interview²⁴ to assess the presence of anxiety disorders. The Stanford Health Assessment Questionnaire²⁵ was used to evaluate functional disability, a subjective measure of physical health with emphasis on the ability to perform daily activities in 5 different domains. Larger scores on this questionnaire represent more disability. All of the questionnaires were administered as part of the home interview.

Statistical Analysis

First, we calculated descriptive statistics for all of the variables in our study, separately for men and women. We examined the differences between men and women of these variables with Student t tests for continuous variables and χ tests for categorical variables. Prior to interpreting each t test, the Levene test for equality of variances was performed. If this test yielded a significant result, the t test without equal variance assumed was used. This procedure was followed for all of the t tests in this study.

Next, we analyzed sex differences in objective and subjective sleep parameters with unadjusted t tests. To clarify the relationships of sex and other covariates with TST, and to investigate possible interaction effects, we performed multivariate linear regression analyses that included all of the available covariates. These analyses were first carried out with diary TST as the dependent variable and subsequently with actigraphic TST as the dependent variable. Although some of the covariates were significantly correlated, there was no collinearity; all correlations were < 0.60.

Means were imputed for missing values for coffee consumption (n = 48), alcohol consumption (n = 48), body mass index (n = 23), cognitive function (n = 20), depressive symptoms (n = 3), and functional disability (n = 1). Because education attainment was missing for 23 participants, anxiety disorders for 28 participants, and probable sleep apnea for 219 participants, “missing” categories were added for these categorical variables. The high number of missing values for probable sleep apnea was due to the fact that a substantial proportion of participants slept alone in a bedroom and were therefore not aware of their snoring or respiratory pauses.

Next, we investigated the independent effect of each individual covariate on sex differences in TST. A change in effect estimate could indicate that the variable acts as an intermediate on the pathway from sex to TST, that it has a different effect on TST in men and women, or, in some cases, reversed causality may occur: variation in a certain variable may be a consequence of TST and also independently related to sex.

An exploratory analysis investigating possible sex-interaction effects predicting TST was performed as follows. We built regression models that used all 17 single covariates and then included an 18^th variable, representing a sex-by-“X” interaction term. A total of 17 separate regressions were performed for diary TST, and a total of 17 separate regressions were performed for actigraphic TST. Since we had no a priori theoretical basis for including the interaction terms, and our main interest was in explaining sex differences, rather than an attempt to build the best fitting model for predicting TST, we were interested in the possible effect of each interaction term separately. Also, building a model with all of the covariates and all of the possible interaction terms would result in an overspecified model. Forward selection, backward selection, and the combination “stepwise” regression fail as tools for model selection for both theoretical and practical reasons. They not only will generally fail to find the correct model that generated the data, but are not even likely to find the best fitting model.²⁶ If a significant sex-by-covariate interaction term appeared for any of these models, the significant sex-interaction effects were further explored in a posthoc stratified analysis. Interaction effects with sex largely fall into 2 categories. Either the variable has an effect on TST in one sex, but not in the other sex, or the effects on TST are in opposite directions, but not necessarily significant in either men or women.

To investigate the possible influence of hormone replacement therapy, we compared the 11 women in our study who used hormones to the women who did not use hormones (Student t tests for continuous variables, χ tests for categorical variables).

To conclude, we tested multivariate-adjusted analysis of covariance models for sex differences in each of the sleep parameters. These models were adjusted for all available covariates.

All of the analyses were performed with SPSS version 11.0 (SPSS Inc., Chicago, IL).

RESULTS

Table 1 presents the characteristics of the study population. The participants were, on average, 68.4 ± 6.8 years old (range 59-97), and 52.3 % were women. Of the total study population, 74.7 % were married, and 6.3 % were employed. The average diary-reported TST was 6.89 ± 0.97 hours, and the average actigraphically measured TST was 6.53 ± 0.83 hours. A nonresponse analysis showed that persons who visited the research center but refused to participate in the actigraphy study (n = 439) were, on average, 2.5 years older than responders (P < 0.001). Women were more likely than men to refuse participation (32.2 % nonresponding women vs 25.1 % nonresponding men, P = 0.002). Refusal to participate was not associated with global PSQI score, the average TST as self-reported in the home interview, nor with education level. Actigraphic marker signals for bedtime or rising time were missing on 3077 out of the 11,790 occasions (2 × 5895 nights), with, on average, 3.2 (± 3.2) missing event markers per participant.

Table 1.

Characteristics of the 956 Participants in the Study

Characteristic	Men (n = 460)	Women (n = 496)	P valuea
Age, y	68.3 ± 6.7	68.4 ± 7.0	0.88
Married	397 (86.3)	317 (63.9)	< 0.001
Education attainment			< 0.001
Low	63 (14.0)	124 (25.6)
Intermediate	282 (62.8)	313 (64.7)
High	104 (23.2)	47 (9.7)
Employmentb	47 (10.2)	13 (2.6)	< 0.001
Bereavement	57 (12.4)	117 (23.7)	< 0.001
Cognitive function, MMSE score	27.9 ± 1.7	27.8 ± 1.9	0.59
Depressive symptoms, CES-D score	4.0 ± 6.0	5.9 ± 6.9	< 0.001
Anxiety disorder	19 (4.2)	56 (11.8)	< 0.001
Medication use
Psychoactive	22 (4.8)	54 (10.9)	< 0.001
Sleepc	28 (6.1)	74 (14.9)	< 0.001
Drinks/day, no.
Coffee	1.1 ± 0.9	0.9 ± 0.8	< 0.001
Alcohol	1.0 ± 1.1	0.5 ± 0.8	< 0.001
BMI, kg/m2	27.7 ± 3.5	28.1 ± 4.4	0.14
Smoking	77 (16.7)	69 (13.9)	0.23
Joint pain	156 (33.9)	251 (50.6)	< 0.001
Functional disability, HAQ score	1.3 ± 0.4	1.5 ± 0.5	< 0.001
Probable sleep apnea	81 (21.0)	14 (4.0)	< 0.001

Note: Data are presented as mean ± SD or number (%). All of the percentages refer to the cases with information on this variable (valid percentage). Data on education were missing for 11 men and 12 women, data on anxiety disorders were missing for 5 men and 23 women, and data on probable sleep apnea were missing for 75 men and 144 women.

^aStudent t test for continuous variables, χ² test for categorical variables

^bBeing employed means working at a job for ≥ 20 h per week and being paid.

^cRefers to sleep medication use during the actigraphy study period on at least 1 night.

MMSE refers to Mini Mental State Examination; CES-D, Center for Epidemiologic Studies Depression scale; HAQ, Health Assessment Questionnaire

Table 2 shows that women reported less and poorer sleep than men on all of the subjective measures. Women reported a 0.22-hour (= 13.2 min) shorter TST, a 10.1-minute longer SOL, and a 4.2% lower sleep efficiency than men. In addition, they had a 2-point higher global PSQI score. However, when measured with actigraphy, women slept 0.26 hours (= 16 min) longer than men, had a 1.2% higher sleep efficiency, and a lower fragmentation index, indicating less-fragmented sleep.

Table 2.

Sex Differences in Subjective and Actigraphic Sleep Measures in the 956 Participants

	Unadjusted model
	Men Estimated mean	Women Estimated mean	Sex difference Men – women (95% CI)	P value
Subjective sleep parameters
Diary
TST, h	7.01	6.79	0.22 (0.10, 0.35)	< 0.001
SOL, min	20.1	30.2	-10.1 (-12.9, -7.3)	< 0.001
SE, %	85.4	81.2	4.2 (2.9, 5.5)	< 0.001
Global PSQI score	2.6	4.6	-2.0 (-2.5, -1.6)	< 0.001
Actigraphic sleep parameters
TST, h	6.40	6.65	-0.26 (-0.36, -0.15)	< 0.001
SE, %	77.8	79.0	-1.2 (-2.1, -0.2)	0.01
Fragmentation index	7.2	6.0	1.2 (0.8, 1.5)	< 0.001

Each row represents a single Student t test. TST refers to total sleep time; SOL, sleep-onset latency; SE, sleep efficiency; PSQI, Pittsburgh Sleep Quality Index; CI, confidence interval.

Table 2 also shows that, for the men and women each considered as a group, the mean difference between diary-reported and actigraphic TST was larger in men (7.01 – 6.40 = 0.61 h) than in women (6.79 – 6.65 = 0.14 h).

In Table 3, the results of the multivariate regression analysis with diary TST as a dependent variable are presented. The total R² of the multivariate model without interaction effects was 0.063, F = 2.846, P < 0.001. The regression coefficient of sex and diary TST remained significant in a multivariate-adjusted model (β = −0.15 [95% confidence interval {CI} −0.29, −0.004], P = 0.045). Cognitive function, depressive symptoms, and sleep medication use were also significantly related to diary TST in this model. Sleep medication decreased the sex difference in diary TST when added to the multivariate model. Sleep medication use was more common in women (Table 1) and associated with shorter diary TST (β = −0.34 [95% CI = −0.55, −0.13], P = 0.002).

Table 3.

Association of Sex and Other Covariates with Diary TST in 956 Participants

Covariates	Association with diary TST, ha
	β	95% CI	R2 change
Sex (female)	-0.15b	-0.29, -0.00	0.004
Age, y	-0.01	-0.02, 0.01	0.001
Marital status (married)	0.13	-0.05, 0.31	0.002
Education attainment			0.002
Low	0.09	-0.13, 0.31
Intermediate	0.02	-0.16, 0.19
Missing	0.23	-0.20, 0.65
Employment	-0.25	-0.51, 0.01	0.003
Bereavement	0.08	-0.08, 0.25	0.001
Cognitive function	-0.04b	-0.08, -0.01	0.005
Depressive symptoms	-0.01b	-0.02, -0.00	0.005
Anxiety disorders			0.001
Yes	0.10	-0.13, 0.34
Missing	0.15	-0.22, 0.52
Psychoactive medication	-0.15	-0.38, 0.09	0.001
Sleep medication	-0.34c	-0.55, -0.13	0.010
Coffee consumption	0.01	-0.07, 0.09	0.000
Alcohol consumption	0.00	-0.07, 0.07	0.000
BMI, kg/m2	0.00	-0.02, 0.01	0.000
Smoking	0.13	-0.04, 0.31	0.002
Joint pain	-0.05	-0.18, 0.08	0.001
Functional disability	0.07	-0.09, 0.24	0.001
Probable sleep apnea			0.005
Yes	0.14	-0.08, 0.35
Missing	-0.15	-0.31, 0.03
Sex × Depressive Symptomsd	-0.02b	-0.04, -0.01	0.006
Sex × Sleep Medicationd	-0.71c	-1.15, -0.27	0.010
Sex × Functional Disabilityd	-0.32b	-0.60, -0.04	0.005

^aThe analysis has been performed with 1 multivariate regression model that included all possible covariates but no interaction terms. The total R² of the multivariate model without interaction effects was 0.063, F = 2.846, P < 0.001.

^bP < 0.05.

^cP < 0.01.

^dThe exploratory analyses of interaction effects have been performed with multivariate regression models that used all 17 single covariates and then included an 18^th variable, representing a sex by “X” interaction term. A total of 17 separate regressions were performed. Only significant interactions with sex, i.e. variables that have a different effect on diary TST in men and women, are included in the table. CI refers to confidence interval.

In a multivariate model with all possible covariates, depressive symptoms, sleep medication use, and functional disability showed significant interaction effects with sex on diary TST, indicating that this variable has different effects on diary TST in men and women. When we further explored these interaction effects in a posthoc analysis, we found that depressive symptoms and sleep medication use were related to significantly shorter diary TST in women but not in men (β of depressive symptoms in women = −0.02 [95% CI −0.04 to −0.01]; β of sleep medication in women = −0.53 [-0.79 to −0.28]). The association between functional disability and diary TST was different in men and women, but the regression coefficients did not reach statistical significance in either sex.

Table 4 shows the results of the regression analyses with actigraphically measured TST. The total R² of the multivariate model without interaction effects was 0.062 (F = 3.858, P < 0.001). The regression coefficient of sex and actigraphic TST remained significant in a multivariate-adjusted model (β = 0.19 [95% CI: 0.07, 0.31]; P = 0.002). Only alcohol consumption decreased the sex difference in actigraphic TST when added to the multivariate model. Men consumed twice as much alcohol as women (Table 1), and alcohol consumption was related to shorter actigraphic TST (β = −0.08 [95% CI: −0.14, −0.03], P = 0.005). Hence, alcohol consumption explained a substantial part of the differences in actigraphic TST between men and women. Bereavement and body mass index were associated with actigraphic TST as well.

Table 4.

Association of Sex and Other Covariates with Actigraphically Measured TST in 956 Subjects

Covariates	Association with actigraphic TST, ha
	β	95% CI	R2 change
Sex, female	0.19b	0.07, 0.31	0.010
Age, y	0.01	-0.00, 0.02	0.001
Marital status, married	0.12	-0.03, 0.27	0.003
Education attainment			0.004
Low	0.14	-0.05, 0.32
Intermediate	0.11	-0.04 0.26
Missing	0.32	-0.04, 0.68
Employment	-0.21	-0.44, 0.01	0.003
Bereavement	0.18c	0.04, 0.32	0.006
Cognitive function	0.01	-0.03, 0.04	0.000
Depressive symptoms	0.00	-0.01, 0.01	0.001
Anxiety disorders			0.001
Yes	0.08	-0.12, 0.28
Missing	0.05	-0.26, 0.36
Psychoactive medication	-0.05	-0.25, 0.15	0.000
Sleep medication	-0.01	-0.19, -0.17	0.000
Coffee consumption	-0.02	-0.09, 0.04	0.001
Alcohol consumption	-0.08b	-0.14, -0.03	0.008
BMI, kg/m2	-0.03d	-0.04, -0.01	0.015
Smoking	-0.07	-0.22, 0.08	0.001
Joint pain	-0.01	-0.12, 0.11	0.000
Functional disability	0.09	-0.04, 0.23	0.002
Probable sleep apnea			0.001
Yes	-0.01	-0.19, 0.17
Missing	-0.09	-0.24, 0.06
Sex × Sleep Medicatione	-0.39c	-0.76, -0.01	0.004
Sex × Joint Paine	-0.24c	-0.45, -0.03	0.005

^aThe analysis has been performed with 1 multivariate regression model that included all possible covariates, but no interaction terms. The total R² of the multivariate model without interaction effects is 0.062, F = 3.858, P < 0.001.

^bP < 0.01.

^cP < 0.05.

^dP < 0.001.

^eThe exploratory analyses of interaction effects have been performed with multivariate regression models that used all 17 single covariates and then included an 18^th variable, representing a sex by X interaction term. A total of 17 separate regressions were performed. Only significant interactions with sex, ie, variables that have a different effect on actigraphic TST in men and women, are included in the table. CI refers to confidence interval.

In a multivariate model predicting actigraphic TST with all possible covariates, the interaction effects of sleep medication and joint pain with sex on actigraphic TST were significant. These two interaction effects fell into the second category, which means that the associations of these variables with TST were in opposite directions in men and women, although they were not statistically significant in either men or women alone.

Of the 496 women in our study population, 11 were on hormone replacement therapy. These 11 women had a significantly lower body mass index than did the women who did not use hormones (25.5 vs 28.1 kg/m², P = 0.04). The 2 groups did not differ on any of the sleep parameters nor on any of the other covariates. Excluding these women from our analyses did not change any of our results.

Table 5 presents a fully adjusted model with all available covariates. All of the sex differences, in self-reported as well as actigraphic measures, remained significant in this fully adjusted model. For actigraphic sleep efficiency and fragmentation index, the sex differences were not attenuated in the multivariate-adjusted model. Rather, if anything, the differences were larger.

Table 5.

Sex Differences in Subjective and Actigraphic Sleep Measures in 956 Subjects

	Multivariate-adjusted modela
	Men Estimated Mean	Women Estimated Mean	Sex difference Men – women (95% CI)	P value
Subjective sleep parameters
Diary
TST, h	6.97	6.82	0.15 (0.00, 0.29)	0.05
SOL, min	22.7	27.9	-5.2 (-8.3, -2.1)	0.001
SE, %	84.4	82.3	2.0 (0.6, 3.5)	0.01
Global PSQI score	3.0	4.2	-1.2 (-1.6, -0.7)	< 0.001
Actigraphic sleep parameters
TST, h	6.43	6.62	-0.19 (-0.31, 0.07)	0.002
SE, %	77.6	79.2	-1.6 (-2.7, -0.6)	0.003
Fragmentation index	7.3	5.9	1.3 (1.0, 1.7)	< 0.001

Each row represents a single analysis of covariance analysis.

^aAdjusted for age, marital status, education attainment, employment, bereavement, cognitive function, depressive symptoms, anxiety disorders, psychoactive medication use, sleep medication use, coffee consumption, alcohol consumption, body mass index, smoking, joint pain, functional disability, and probable sleep apnea. TST refers to total sleep time; SOL, sleep-onset latency; SE, sleep efficiency; PSQI, Pittsburgh Sleep Quality Index; CI, confidence interval.

DISCUSSION

This study demonstrates, in a population-based study of elderly persons, that actigraphically measured sleep is better in women, whereas—as is well known—women report less and poorer sleep than men. The sex differences in subjective sleep parameters were attenuated after adjustment for marital status, the use of sleep medication, and other covariates, but all of the differences remained significant in a multivariate-adjusted model. Sex differences in actigraphic sleep parameters were only marginally explained by adjustment for covariates, although differences in alcohol consumption clearly accounted for part of the sex differences in actigraphic TST.

The relationships of depressive symptoms, sleep medication use, and functional disability with diary-reported TST were different for men and women. This was also true for the relationships of sleep medication use and joint pain with actigraphic TST. These interaction effects also partially explained sex differences in diary-reported and actigraphically measured TST.

Before we discuss these findings, some methodologic comments have to be made. First, women were less likely to participate in the actigraphy study. This could have diminished or enlarged sex differences in sleep parameters, if nonparticipation was related differently to sleep parameters in men and women. However, nonparticipation was not associated with global PSQI score or with self-reported TST in the home interview. Second, it has been suggested that women are more likely than men to express emotional distress and to report somatic symptoms in general.^4,8 In our study, we used subjective sleep data, which may have been influenced by this sex-specific reporting bias. Data on physical and psychological symptoms that we investigated as covariates were also obtained through self-report, which may limit the reliability of these data. Third, when the event-marker button was not pressed to signal bedtime and rising time, we derived these parameters from the diary to be able to calculate the other actigraphic sleep parameters. The event-marker signal was missing in 26% of occasions. This problem has been previously reported by others; in the CARDIA study, the sleep log was needed for 40% of bedtimes and wake times.¹¹ Fourth, actigraphy is not the gold standard for distinguishing sleep from waking. The results of the study may differ with different actigraphy devices or algorithms. The Actiwatch algorithm has only been validated in a study of sleep-disordered patients¹⁹; the appropriateness of this algorithm in a normal population has not been tested. Agreement between actigraphy and polysomnography is high in normal sleepers²⁷ but can be lower in persons with poor sleep quality,²⁸ since these persons tend to lie in bed motionless, but awake, for long time periods. In these participants, the actigraphy algorithm will overestimate sleep duration, compared with polysomnography. Nevertheless, multiple authors, including the American Academy of Sleep Medicine's Standards of Practice Committee, conclude that actigraphy is a reliable method for assessing sleep-wake patterns in adults.^27,29,30 Comparing our actigraphic sleep data with those of other studies, we conclude that the mean actigraphic sleep durations for men and women that we found were remarkably similar to the mean actigraphic sleep durations reported by Patel et al. in the Osteoporotic Fractures in Men Study and the Study of Osteoporotic Fractures, respectively.³¹ Patel et al. reported a mean actigraphic sleep duration of 6.40 hours for men aged 67 to 96 years, which is similar to the 6.40 hours we found in slightly younger men. The mean actigraphic sleep duration in women aged 70 to 99 years reported by Patel et al. was 6.75 hours, whereas this was 6.65 hours in our study. Lauderdale et al. noted mean actigraphic sleep durations of 6.1 hours for White men and 6.7 hours for White women aged 38 to 50 years,¹¹ which is again similar to our study in women and somewhat shorter than our results in men.

Strengths of the present study include its population-based nature, its large size, the availability of multiple nights of actigraphic measurements, and assessment of numerous covariates. This study confirmed the results of earlier studies that showed that self-reported sleep quality was poorer in women.^1–3,5 The shorter self-reported TST that we found has been less commonly noted in previous literature.^1,3,7 We also found—in line with earlier studies—that women had longer actigraphically measured sleep duration and a higher sleep efficiency than men. Lauderdale et al.¹¹ studied the combined effect of sex and race on sleep parameters measured with 3 nights of actigraphy in a population-based study of 669 participants and found that mean sleep duration varied by race-sex group, ranging from 6.7 hours for White women to 5.1 hours for Black men. They made very elaborate adjustments for a wide range of socioeconomic, demographic, and lifestyle variables. However, they did not adjust for psychiatric disorders or the use of sleep medication, which may be important in explaining (race-) sex differences in sleep. In addition, they did not show data for subjective measures, although these differed substantially from their actigraphy measurements. Longer sleep in women had been observed earlier by Jean-Louis et al.³² in a small population-based study with actigraphic sleep parameters. However, in this study, no attempt was made to explain the differences by adjusting for covariates other than age and work status. In a meta-analysis of polysomnography studies, which are mostly performed in clinical convenience samples, Ohayon et al. confirmed a modestly higher mean TST in women than in men, but they did not find sex differences in sleep efficiency.¹⁰

We investigated whether the sex differences that we found were explained by demographic factors such as marital status, education attainment, or employment; by sleep-related factors such as sleep medication use and probable sleep apnea; or by depressive symptoms, functional disability, or alcohol consumption. Particularly, the sex differences in self-reported sleep parameters were to some extent explained by adjustment for these covariates, although none of the differences was fully accounted for. The covariates that were most strongly related to sex differences in diary-reported sleep duration were marital status, the use of sleep medication, and probable sleep apnea. The interrelatedness of these variables with sex and TST cannot be entirely elucidated in a cross-sectional study. However, we were able to shed some light on these relationships.

The relationship between depressive symptoms and sleep complaints has been described previously. Longitudinal studies have shown that the association of depression with poor sleep quality is bidirectional: on the one hand, depression strongly increases the risk of poor sleep quality, and, on the other hand, poor sleep quality is a predictor for future depressive episodes.^12,14,33 Epidemiologic studies estimate that complaints of poor sleep quality are observed in 40% to 90% of subjects with diagnosed depression.^33,34 In our study, depressive symptoms were related to shorter diary-reported TST in women but not in men and were also more common in women. Nevertheless, depressive symptoms did not explain sex differences in subjective sleep parameters. This is in line with the findings of Zhang et al.⁴ and Voderholzer et al.,³⁵ who showed that sex differences in the prevalence of insomnia persisted after the underlying psychiatric disorders had been taken into account.

The use of sleep medication was also related to shorter diary-reported TST in women but not in men and were more common in women. This could be an example of reversed causality; it is likely that women who feel that they obtain too little sleep revert to sleep medication, which may not return their (perceived) sleep duration to normal or average.

Sex differences in actigraphically measured TST were only partly explained by alcohol consumption, which was almost twice as high in men, as compared with women, and was related to shorter actigraphic sleep duration. Alcohol affects physiologic processes that occur during sleep, especially at higher doses.³⁶ This may be detectable by actigraphy if it also affects the frequency or intensity of movement during sleep. Alcohol use may be related to reported apnea symptoms because alcohol use can exacerbate obstructive sleep apnea and may precipitate other breathing disorders during sleep.³⁷

The associations of sleep medication use and joint pain with actigraphic TST were different in men and women. Although these associations did not reach statistical significance in a stratified analysis, the data suggest that these variables are related to longer actigraphic TST in men and to shorter actigraphic TST in women. However, further research is warranted to further elucidate these relationships and the possible underlying mechanisms.

We were not able to explain all of the sex differences in sleep parameters; therefore, other mechanisms must explain why women sleep longer and better than men when sleep is measured with actigraphy and why women nevertheless report less and poorer sleep than do men. We propose a few biologic, psychologic, and methodologic hypotheses that may be addressed in future research.

Women may need more sleep than men, and, therefore, the same amount of sleep may be satisfactory for men but not for women. This would offer a partial explanation of the sex differences in both self-reported and objective sleep measures. However, a higher self-reported sleep need in women has been demonstrated by Lindberg et al.¹ only in subjects aged 20 to 45 years and also by Broman et al.³⁸ in younger persons. Likewise, sex differences in circadian rhythms—morningness versus eveningness—are only apparent before the age of menopause.³⁹ Moreover, if differences in sleep need were to explain the sex differences that we found, the question arises as to why women do not simply adapt their sleep times to their needs. Because the youngest woman in our study was 59.7 years old, the vast majority of the women in our study were postmenopausal, but hormone differences may still underlie a higher sleep need in women.

Several studies in rodents and humans have shown that, in mammals, both endogenous and exogenous steroid hormones can affect the regulation of sleep and circadian rhythms.⁴⁰ Gonadectomy in male and female mice eliminates the sex difference in the amount of wakefulness and non-rapid eye movement (non-REM) sleep. This suggests that sex differences in sleep architecture depend on gonadal hormones.⁴¹ Progesterone has a sedating effect, and, in human and animal sleep, it shortens SOL and enhances sleep continuity. Estrogen suppresses REM sleep in rats but enhances REM sleep in humans.⁴⁰ During the menopausal and postmenopausal periods, estrogen and progesterone levels decrease. Compared with premenopausal women, postmenopausal women are more likely to suffer from sleep disturbance. Hot flashes often cooccur with sleep disturbance, but the causal relationship is not clear.⁴⁰ In our study, we did not find any differences in sleep parameters between women who were on hormone replacement therapy and women who were not, but it should be noted that this could be due to a lack of statistical power because our group of women using hormones was very small.

In healthy men, circulating testosterone levels decline with advancing age. This process is characterized by considerable interindividual variability.⁴² Studies of the association between testosterone levels and sleep in elderly men suggest that lower testosterone levels are associated with less TST⁴² and lower sleep efficiency.^43,44 The higher prevalence of sleep apnea in men suggests that sex hormones are involved in the pathogenesis of this disorder.⁴³ Although sex hormones clearly influence sleep parameters in both men and women, the exact mechanism by which hormone differences may explain sex differences in sleep remains poorly understood.

Both Groeger et al.⁸ and Zhang et al.⁴ hypothesized that the sex difference in reporting symptoms, which we discussed above, might be related to the greater body vigilance and awareness among women. If this is true, and if this also implies that women's perceptions of sleep are more realistic than men's, sex differences in self-reported sleep parameters may also result from an increased likelihood of overestimation of sleep, or the perception of wakefulness as sleep, in men. As can be derived from Table 2, the average discrepancy between diary-reported and actigraphic TST is indeed larger in men than in women, which supports this hypothesis. However, it could be argued that the absolute values of actigraphically measured sleep parameters may differ from the “true” values of the sleep parameters. In that case, another possible explanation for the shorter self-reported TST in women arises: women—more than men—may show a tendency to perceive sleep as wakefulness and thus underestimate their sleep duration. This phenomenon is often seen in insomniacs.^45,46 We did not exclude persons who suffered from insomnia, which is more frequent in women.⁴ However, matters of perception and reporting still do not offer an explanation for discrepancies in actigraphic sleep measures.

Finally, sex differences may also result from a methodologic issue. Actigraphy measures limb movement. It is possible that there are sex differences in the relationship between sleep and movement. For example, limb movement might be more frequent or more intense in sleeping men than in sleeping women, and women might lie quietly awake more often than do men. In this case, the differences in “true” sleep may be smaller than our study indicates. Although Van Hilten et al.⁴⁷ stated that the mean duration of nocturnal immobility periods was higher in females than males, there is no indication that the validity of actigraphic measurement, ie, the relationship between movement and sleep, differs between men and women.

To conclude, in a normal elderly population, women consistently reported shorter and poorer sleep than did men. When actigraphic measurements were considered, women showed more and better sleep than did men. In other words, sex differences in sleep parameters depended on what was measured: either perceived sleep or objectively measured sleep. Sex differences in sleep parameters in the elderly are partly, but not fully, explained by covariates that are related to both poor sleep and sex, such as sleep medication use and alcohol consumption. Other mechanisms underlying these discrepancies remain to be elucidated.

DISCLOSURE STATEMENT

This was not an industry supported study. The authors have indicated no financial conflicts of interest.