Assessment of Sleep in the National Social Life, Health, and Aging Project

Abstract

Objectives.

The relationship of sleep to health has been an active area of research in recent years, and the National Social Life, Health, and Aging Project (NSHAP) expanded sleep data collection in Wave 2 with enhanced core questions and a novel sleep module that included an objective measure of sleep duration and quality.

Method.

A randomly selected one-third of Wave 2 participants and their spouses or coresident partners were invited to participate in the sleep module. Objective sleep data were collected using wrist actigraphy, an accelerometer that records an integrated measure of motion over short epochs (15 s each). This information is stored and subsequently analyzed to determine sleep and wake periods by epoch. Individuals were instructed to wear the actiwatches for 72hr. Several sleep parameters were derived from the accelerometer. Individuals concurrently kept a sleep diary.

Results.

Sleep actigraphy data were successfully collected from 780 individuals. Many of the survey-based and the actigraph-estimated sleep parameters varied by age and gender. However, age and gender patterns often differed for sleep characteristics that were both asked and measured, such as sleep duration.

Discussion.

The survey and actigraphy data provide different information about sleep characteristics. The opportunity to examine actigraph-estimated sleep characteristics in a nationally representative sample of older adults allows novel analyses of the associations of sleep parameters with social and health data.

Sleep is a complicated behavior with multiple dimensions that include timing, duration, consolidation, depth, and restfulness. Sleep characteristics vary from person to person (Lauderdale et al., 2006; Nunes et al., 2008; Unruh et al., 2008), and complaints about difficulty sleeping are common among older adults (Ancoli-Israel, 2003). Although it seems intuitively obvious that sleep is essential to life, the physiological functions of sleep are not fully understood; this has been an active area of research in recent years (Siegel, 2005; Xie et al., 2013).

Sleep has not historically been one of the lifestyle risk factors routinely queried in population-based surveys and studies. For example, the annual National Health Interview Survey just added sleep questions to the core Adult Health Behaviors Section in 2004 (Schoenborn & Adams, 2010). However, there were a few large epidemiologic cohorts that included one or more survey questions about sleep prior to 2000, notably the American Cancer Society’s Cancer Prevention Study I and II, the Nurses’ Health Study, and the First National Health and Nutrition Examination Survey Epidemiologic Follow-up Study (Ayas et al., 2003; Gangwisch, Malaspina, Boden-Albala, & Heymsfield, 2005; Kripke, Garfinkel, Wingard, Klauber, & Marler, 2002). Analyses from these studies have found that survey-based sleep characteristics correlate with many health outcomes, including mortality (Cappuccio, D’Elia, Strazzullo, & Miller, 2010; Gallicchio & Kalesan, 2009). Most sleep research in nonclinical populations has been laboratory-based and included small numbers of volunteers. Such studies have identified potential mechanisms that could explain some of the correlations found between sleep and health in the cohort studies. For example, laboratory-based experimental studies of younger adults have indicated that there may be deleterious consequences of inadequate sleep across multiple physiological domains, including cognitive, metabolic, and immunologic (Faraut, Boudjeltia, Vanhamme, & Kerkhofs, 2012; Van Cauter, Spiegel, Tasali, & Leproult, 2008; Walker, 2008).

Wave 2 of National Social Life, Health, and Aging Project (NSHAP) added a novel sleep module that included an objective measure of sleep as well as additional survey questions; these are the first objective sleep data available for a nationally representative sample of the population in the United States. This paper will describe the survey questions about sleep in the Wave 1 and Wave 2 core interviews, the measurements included in the sleep module in Wave 2, and some suggestions for using them in analyses.

Wave 1 Sleep Data

Survey questions about sleep have most often assessed quantity, quality, or general satisfaction, frequently posed as “feeling rested” during the day. In the first wave of data collection (2005–2006), the NSHAP asked about sleep with four questions in the core survey. These questions were asked of all participants. Sleep quantity was assessed with the single question “How many hours do you usually sleep at night?” and response options were in whole hours. Sleep quality was included as part of the depression instrument, in which respondents were asked about the frequency of several conditions, including “My sleep was restless.” Overall satisfaction was assessed with the question “How often do you feel really rested when you wake up in the morning?” with four levels of response endorsement. Finally, in the social relationships section, there was a question about how often individuals with spouses or romantic partners slept in the same bed as that person during the past month. Mean reported sleep duration from the single question was just under 7hr and did not vary greatly by age or sex (Table 1). Only about 60% reported feeling rested in the morning most of the time, and this also did not vary greatly by age or sex. Relatively few, however, reported that sleep was restless most of the time (12%), and this also did not vary strongly by age or sex. The one sleep question with a strong age effect in Wave 1 was sleeping with a partner, which decreased by age, not only because fewer individuals had partners, but also because those with partners were less likely to sleep in the same bed. Data are publicly available (NSHAP Wave 1: Waite, Linda J., Edward O. Laumann, Wendy Levinson, Stacy Tessler Lindau, and Colm A. O’Muircheartaigh. National Social Life, Health, and Aging Project (NSHAP): Wave 1. ICPSR20541-v6. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], 2014-04-30. doi:10.3886/ICPSR20541.v6.).

Table 1.

Sleep Questions in Wave 1 of the National Social Life, Health, and Aging Project (2005–2006)

Measure	Overall	Male			Female
		Aged 57–64	Aged 65–74	Aged 75–85	Aged 57–64	Aged 65–74	Aged 75–85
How many hours do you usually sleep at night?
Number of observations	2,984	527	544	378	489	540	506
Mean (hours)	6.90	6.71	6.97	7.19	6.88	6.92	6.89
How often do you feel really rested when you wake up in the morning?
Number of observations	3,000	528	546	378	491	545	512
Never	0.04	0.04	0.03	0.04	0.06	0.05	0.03
Rarely	0.09	0.09	0.10	0.06	0.08	0.08	0.11
Sometimes	0.26	0.26	0.25	0.25	0.28	0.25	0.29
Most of the time	0.61	0.61	0.63	0.65	0.58	0.62	0.57
My sleep was restless
Number of observations	2,998	527	545	378	490	545	513
Rarely/none of the time	0.45	0.45	0.49	0.47	0.43	0.44	0.46
Some of the time	0.30	0.31	0.27	0.28	0.32	0.30	0.28
Occasionally	0.13	0.12	0.15	0.12	0.10	0.14	0.13
Most of the time	0.12	0.12	0.10	0.13	0.15	0.12	0.13
In the last month, how often did you sleep in the same bed with your spouse or romantic partner?
Number of observations	2,417	425	443	292	410	442	405
Never	0.19	0.11	0.18	0.23	0.16	0.23	0.27
Some of the time	0.11	0.16	0.14	0.17	0.06	0.05	0.06
All/most of the time	0.55	0.68	0.62	0.50	0.62	0.48	0.28
No spouse/partner	0.16	0.05	0.06	0.10	0.16	0.24	0.38

Note. All are proportions except where the mean is specified. Means and proportions are calculated using the sample weights, while the “Number of observations” is the unweighted number of respondents.

Wave 2 Sleep Data

Survey Questions in the Core Data Collection

Growing interest in sleep as a chronic disease risk factor contributed to a decision to enhance sleep data collection in Wave 2 (2010–2011) of NSHAP. There were two types of enhancements: (a) changes in the core sleep questions including a different approach to querying typical sleep duration and a few additional questions and (b) inclusion of a new module, with an objective sleep measurement and sleep log, offered to a randomly selected one-third of respondents and their spouses or coresident partners. Data are publicly available (NSHAP Wave 2: Waite, Linda J., Kathleen Cagney, William Dale, Elbert Huang, Edward O. Laumann, Martha K. McClintock, Colm A. O’Muircheartaigh, L. Phillip Schumm, and Benjamin Cornwell. National Social Life, Health, and Aging Project (NSHAP): Wave 2 and Partner Data Collection. ICPSR34921-v1. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], 2014-04-29. doi:10.3886/ICPSR34921.v1.).

The core sleep duration question was modified in response to accumulating evidence that single questions about usual sleep duration are not highly correlated or well calibrated with measured sleep duration for older adults, although self-reported duration does nonetheless predict many health outcomes (Ananthakrishnan et al., 2014; Gangwisch et al., 2006; Martínez-Gómez, Guallar-Castillón, León-Muñoz, López-García, & Rodríguez-Artalejo, 2013). Self-reported duration is typically longer than sleep as measured by either actigraphy or polysomnography, and the correlation between self-reported duration and measured sleep is in the low to moderate range (Patel, Blackwell, Ancoli-Israel, & Stone, 2012; Silva et al., 2007; van den Berg et al., 2009). There is some evidence of systematic biases in self-reported duration: for example, those with more education have a higher correlation between reported and measured sleep compared to those with less education (Lauderdale, Knutson, Yan, Liu, & Rathouz, 2008; Silva et al., 2007).

A few epidemiological cohorts of older adults have recently added or included objective sleep measurements, either actigraphy, which estimates sleep based on arm motion, or polysomnography, which defines sleep based on electrical activity in the brain recorded by an electroencephalogram (Blackwell et al., 2008; Nieto et al., 2000; Patel et al., 2008; van den Berg et al., 2007). Neither is a perfect measure of routine sleep behavior. They are imperfect because: (a) actigraphy measures a physical state that is highly correlated with sleep but does not actually measure sleep and (b) polysomnography may alter sleep behavior due to its intrusiveness and its need to be set up by a technician.

There has been very little methodological work about how best to ask about sleep duration to optimize validity and reliability. This is likely because asking about sleep duration in more than one way in the same interview causes respondents, remembering their most recent answer, to be more consistent in their subsequent answers than they might be if differently worded questions were asked at different times. There are understandable reasons why habitual sleep duration questions may be challenging for individuals to answer. A valid answer draws on several cognitive domains. First, individuals must realize that they need to carry out a mental arithmetic problem in order to give an accurate answer. Then they need to figure out the time when they typically fall asleep and when they usually wake up—both of which may have great night-to-night variability. Finally, they must hold those numbers in mind and subtract in their head, a calculated difference that often spans midnight. It seems likely that many older individuals may have difficulty doing this or simply chose not to, and instead respond with a number that seems plausible. In Wave 2, NSHAP removed two potentially problematic aspects of the duration question: realizing it is an arithmetic problem and mentally subtracting. Respondents were asked “What time do you usually go to bed and start trying to fall asleep?” and “What time do you usually wake up?” These questions were asked separately for “weekdays or work days” and for “weekends, or days off.” Individuals responded with clock time, including hours, minutes, and a.m./p.m. The answers not only allow the calculation of usual sleep duration by the analyst, but also make it possible to consider the circadian timing of usual sleep as an additional sleep characteristic.

The remaining three sleep questions from Wave 1 were repeated in the Wave 2 core interview. There were also new questions about sleep medications and naps in the Wave 2 core interview (Table 2). Average sleep duration calculated from the usual bedtime and waketime questions for weeknights and weekend nights (weighted 5/7 and 2/7) was 8.2hr; this increased with age for men but not for women. Overall, 20% had taken sleep medications in the previous 2 weeks—this was higher for women than men but did not vary with age. A majority of respondents had taken at least one nap during the past week, and the frequency of naps increased with age.

Table 2.

New Sleep Questions in Wave 2 of the National Social Life, Health, and Aging Project (2010–2011)

Measure	Full sample	Male			Female
		Aged 62–69	Aged 70–79	Aged 80–90	Aged 62–69	Aged 70–79	Aged 80–90
Mean sleep duration in hours (calculated from usual bedtime and waking time)	8.20	7.93	8.19	8.36	8.32	8.25	8.18
In the past 2 weeks, have you taken any medications or used other treatments to help you sleep?
Number of observations	3,115	562	585	322	672	611	363
Yes	0.20	0.16	0.13	0.17	0.24	0.22	0.25
No	0.80	0.84	0.87	0.83	0.76	0.78	0.75
Were these medications or other treatments recommended to you by a doctor?
Number of observations	3,115	562	585	322	672	611	363
Not applicable	0.81	0.84	0.87	0.83	0.76	0.78	0.75
Yes	0.12	0.09	0.08	0.10	0.13	0.16	0.17
No	0.07	0.07	0.05	0.07	0.10	0.06	0.08
During the past week, on how many days did you nap for 5min or more?
Number of observations	2,720	488	510	277	598	549	298
Never	0.27	0.27	0.18	0.17	0.35	0.32	0.21
1 or 2 days	0.33	0.32	0.30	0.24	0.36	0.36	0.34
3 or 4 days	0.21	0.24	0.27	0.25	0.16	0.17	0.20
5 or more days	0.19	0.16	0.26	0.34	0.13	0.14	0.26
During the past week, on how many days did you nap for an hour or 2?
Number of observations	2,720	488	510	277	598	549	298
Never	0.58	0.60	0.52	0.45	0.65	0.63	0.53
1 or 2 days	0.27	0.26	0.28	0.31	0.24	0.27	0.29
3 or 4 days	0.08	0.09	0.11	0.14	0.06	0.07	0.08
5 or more days	0.06	0.05	0.09	0.10	0.05	0.03	0.10

Note. All are proportions except where the mean is specified. Means and proportions are calculated using the sample weights, while the “Number of observations” is the unweighted number of respondents.

The Sleep Module

A randomly selected one-third of the Wave 2 respondents were invited to participate in the Sleep Module during their in-home interview (Wave 1 respondents and their spouses/partners were selected independently). Altogether, 808 individuals in the original sample and 309 spouses (or coresident partners) were invited to participate; of these, 897 agreed to participate. The Module had two components: 72hr of wrist actigraphy and a “Sleep and Activity Booklet.” For those who agreed, a date was set for sleep data collection and the booklet and actigraph, along with a prepaid return mailer, were mailed out, usually a few days after the in-home interview; 823 were successfully recontacted and received the packet. The booklet had two parts: additional questions about sleep and a 3-day sleep log to be filled out concurrently with wearing the wrist actigraph. Seven hundred and eighty-five booklets were ultimately returned. Because the booklet was received a few days after the interview, NSHAP had the opportunity to ask respondents about sleep duration in two different ways without having both in the same interview. The Wave 1 sleep duration question was repeated in the Sleep and Activity Booklet: “How many hours do you usually sleep at night?” There were three additional sleep quality questions in the booklet: (a) the frequency of having trouble falling asleep, (b) waking up during the night, and (c) waking up too early (Table 3). These questions have previously been used in many studies, including the Established Populations for Epidemiologic Studies of the Elderly (EPESE) and the Health and Retirement Survey (HRS). The mean sleep hours reported in the booklet, using the same wording as the Wave 1 core question was 7.5hr, which decreased with age for women, but not for men. Women were more likely than men to report having trouble falling asleep. With increasing age, both men and women were more likely to report waking up during the night and waking up too early.

Table 3.

Booklet Questions of the Sleep Substudy in Wave 2 of the National Social Life, Health, and Aging Project (2010–2011)

Measure	Full sample	Male			Female
		Aged 62–69	Aged 70–79	Aged 80–90	Aged 62–69	Aged 70–79	Aged 80–90
How many hours do you usually sleep at night?
Number of observations	606	115	111	64	141	115	60
Mean (hours)	7.47	7.35	7.77	7.63	7.57	7.40	7.06
How often do you have trouble falling asleep?
Number of observations	629	118	114	65	143	126	63
Most of the time	0.13	0.10	0.04	0.17	0.18	0.12	0.15
Sometimes	0.46	0.33	0.49	0.35	0.50	0.46	0.63
Rarely or never	0.42	0.57	0.47	0.48	0.31	0.42	0.22
How often do you have trouble with waking up during the night?
Number of observations	629	118	114	65	143	126	63
Most of the time	0.30	0.30	0.37	0.31	0.28	0.31	0.24
Sometimes	0.44	0.40	0.41	0.57	0.45	0.37	0.55
Rarely or never	0.26	0.30	0.21	0.12	0.27	0.32	0.21
How often do you have trouble with waking up too early and not being able to fall asleep again?
Number of observations	625	117	112	65	142	125	64
Most of the time	0.13	0.12	0.15	0.12	0.11	0.13	0.17
Sometimes	0.47	0.42	0.37	0.63	0.44	0.56	0.59
Rarely or never	0.40	0.47	0.48	0.25	0.47	0.32	0.24

Note. All are proportions except where the mean is specified. Means and proportions are calculated using the sample weights, while the “Number of observations” is the unweighted number of respondents.

The booklet also contained a 3-day sleep log. For each of the three 24-hr periods (within the 72hr) when the actigraph was worn, individuals were asked to write down when they started trying to fall asleep, when they awoke, whether they napped and for how long (within categories; variables described in Table 4).

Table 4.

Sleep Log Questions of the Sleep Substudy in Wave 2 of the National Social Life, Health, and Aging Project (2010–2011)

Measure	Response categories
What is the total amount of time that you spent napping during the day? (N = 730)	No answer
	No naps
	Less than 15 min
	15min to 1 hr
	More than 1 hr
When you went to bed, what time did you start trying to fall asleep? (N = 730)	O’clock – hour
	O’clock – minutes
	a.m./p.m.
What time did you wake up? (N = 730)	O’clock – hour
	O’clock – minutes
	a.m./p.m.

Note. Each question is asked for each of three 24-hr periods, concurrent with wearing an Actigraph.

Actigraphy

Wrist actigraphy data were collected using the Actiwatch Spectrum model from Phillips Respironics; the manufacturer loaned Actiwatches to NSHAP for this study through their grant review process. Actiwatches look like a digital wristwatch, but they include a highly sensitive accelerometer that digitally records gross motor activity. Counts of activities are recorded within prespecified epochs (15-s epochs were used for NSHAP). Upon return of the Actiwatch, data were downloaded and analyzed using manufacturer-developed and tested algorithms that estimate several sleep parameters using the patterns of activity counts. This Actiwatch Spectrum model includes additional features: measures of ambient and colored light and a detector of when the Actiwatch is on the wrist. There have been many validation studies of individuals concurrently wearing an actigraph (various manufacturers and models) and undergoing polysomnography. However, most of these studies only included a small number of subjects, often fewer than 20, and focused on either a clinical population or a much younger population. There have been several review articles about the validation studies (Ancoli-Israel et al., 2003; Sadeh & Acebo, 2002; Van de Water, Holmes, & Hurley, 2011). Ancoli-Israel and colleagues (2003) concluded that the overall correlation in sleep duration for adults is about 90%. Van de Water and coworkers (2011) pointed out that the accuracy of actigraphy, compared with polysomnography, depended on the sleep variable, the manufacturer of the device and the study population. Among the few studies of older, community-dwelling adults, a study of 36 postmenopausal women reported correlations over 0.9 for total sleep time and sleep efficiency (Jean-Louis, Kripke, Cole, Assmus, & Langer, 2001). More recently, a very large study of older men compared different actigraphy scoring algorithms to polysomnography and found only moderate correlations for the best of the algorithms (Blackwell, Ancoli-Israel, Redline, Stone, & Osteoporotic Fractures in Men (MrOS) Study Group, 2011). In general, actigraphy overestimates sleep relative to polysomnography, as still awake time tends to be scored as sleep. In addition, Phillips Respironics has compiled a bibliography of validation studies using their models (accessed on 18 December, 2012, from http://www.healthcare.philips.com/us_en/homehealth/sleep/actiwatch/default.wpd). Details about the Actiware 5.5 software are available from http://www.learnactiware.com/tutorials/ and the user manual is available online at www.camntech.com/files/The_Actiwatch_User_Manual_V7.2.pdf.

Individuals were asked to wear the Actiwatch for 72hr; the Actiwatch records the time and date. The devices were sent out on all weekdays, and the participants were asked to start wearing the watch the day they received it, which was always a weekday because of the type of delivery. The model also includes an event marker and participants were asked to push it both when they started trying to fall asleep each night and when they awoke fully for the day. The event marker leaves a time stamp but does not affect recording, which was set to begin automatically during the morning of the day when the participant received the instrument and record continuously thereafter. The actigraphy data may be fully analyzed whether or not the event marker was pressed; the event markers supply additional information about the individual’s sleep-related behavior. Six hundred and fifty-five individuals remembered to use the event marker at the beginning and 674 remembered to use it at the end of at least one night. Some analyses may need to be restricted to nights with event markers. For example, an analysis of sleep latency would be restricted to nights with an evening event marker, whereas an analysis of consistency of bedtime might be restricted to individuals with all three evening event markers. The main rest interval (which contains not only the sleep interval but also the time spent falling asleep and waking up) for each individual was first set by the manufacturer’s software to capture the longest sleep interval each day, based on the activity pattern. The rest intervals were then manually reexamined and reset if the interval set by the software failed to fully include the longest sleep interval, defined by the activity pattern. For example, the software sometimes only included part of the night when there was a long awake episode during the night. Additional data used for setting the rest interval were the event marker and the light level, which typically indicated when the lights were turned off. When present, the event marker was prioritized unless it was inconsistent with the other pieces of evidence, because it appeared that individuals occasionally forgot to press the event marker and then thought to press it if they awoke during the night.

Actigraphy Results

Actigraphs with data were returned by 793 individuals. However, for 13 of them, the data could not be used to determine sleep statistics, for reasons such as no discernable rest interval, resulting in usable data from 780 individuals. Primary respondents had birth years 1920–1947 and were approximately aged 62–90 during Wave 2 data collection, whereas the 227 spouses ranged in age from 36 to 91. The data presented in Table 6 includes only the primary respondents and the spouses/partners whose birth years were in the range included in the primary respondent sample (174 spouses/partners). However, analyses with actigraphy data could include a more extended age range among the spouses. For the small number of individuals with fewer than three usable nights of actigraphy data, the summary means include one or two nights, as available. For each individual, the public release dataset includes sleep parameters for up to three individual nights and also provides means across nights.

The definitions of the actigraphy-estimated sleep characteristics are given in Table 5, and the averages for the sleep characteristics included in the public release data are given in Table 6. The public use file also includes the nightly data. Several variables are based on the definition of each epoch as sleep or not, which is a function of the activity count in that epoch and surrounding epochs. “Assumed sleep” is the length of the interval that begins with the first epoch scored as “sleep” in the major sleep interval and ends with the last epoch scored as “sleep.” “Actual sleep” is the total duration of the epochs scored as sleep within the assumed sleep interval. Wake after sleep onset, referred to as WASO, is the total minutes awake during the assumed sleep interval. “Percent sleep” is the percent of the assumed sleep interval that is actual sleep. Additional variables are the clock times at the beginning and end of the assumed sleep interval, the day of the week of each sleep interval and the fragmentation index. Fragmentation is an indicator of sleep disruption and is the sum of two percentages: the percentage of the sleep interval spent moving and the percentage of immobile periods (i.e., contiguous epochs with no movement) that are no more than one minute long. Although there is no clear definition of what “sleep quality” is, WASO, fragmentation, and percent sleep are each thought of as quality indicators.

Table 5.

Variable Names and Definitions of Actigraph-Estimated Sleep Characteristics From the Sleep Substudy in Wave 2 of the National Social Life, Health, and Aging Project (2010–2011)

Measure	Variable name	Definition
Assumed sleep	assumed_sleep	The length of the sleep interval from the first epoch counted as sleep to the last epoch counted as sleep in the main sleep interval, in hours.
Actual sleep	actigraph_sleep	The total number of minutes counted as sleep in the main sleep interval, in hours.
WASO	waso	Minutes of wake after sleep onset.
Fragmentation	fragmentation	The sum of two percentages: the percentage of the sleep interval spent moving and the percentage of immobile periods (i.e., contiguous epochs with no movement) that are only 1min long.
Percent sleep	percent_sleep	The ratio of acti_sleep to assumed_sleep, as a percentage.

Note. WASO = wake after sleep onset.

Table 6.

Actigraph-Estimated Sleep Characteristics From the Sleep Substudy in Wave 2 of the National Social Life, Health, and Aging Project (2010–2011) for Individuals Aged 62–90 and by Gender and Age Groups

Measure	Full sample (n = 727)	Male			Female
		Aged 62–69 (n = 137)	Aged 70–79 (n = 127)	Aged 80–90 (n = 79)	Aged 62–69 (n = 165)	Aged 70–79 (n = 146)	Aged 80–90 (n = 73)
Assumed sleep (hours)	7.90 (1.64, 13.17)	7.49 (1.64, 13.17)	7.82 (3.39, 11.94)	8.27 (4.42, 11.14)	8.02 (2.50, 11.70)	8.00 (2.03, 12.09)	8.11 (3.30, 12.07)
Actual sleep (hours)	7.25 (1.52, 12.60)	6.90 (1.52, 12.60)	7.14 (2.91, 10.94)	7.46 (3.81, 10.28)	7.38 (1.74, 11.08)	7.37 (1.95, 11.29)	7.43 (3.13, 10.54)
WASO (minutes)	39.07 (3.17, 198.50)	34.96 (5.17, 198.50)	41.07 (6.75, 141.33)	48.20 (7.67, 130.00)	38.33 (3.17, 129.50)	37.41 (4.83, 153.67)	40.88 (3.42, 178.50)
Fragmentation	14.36 (1.19, 50.28)	13.80 (4.04, 50.28)	15.85 (3.47, 45.81)	16.94 (5.52, 36.28)	13.42 (3.57, 30.93)	13.50 (3.73, 36.96)	14.82 (1.19, 41.25)
Percent sleep	0.82 (0.29, 1.00)	0.86 (0.35, 1.00)	0.80 (0.32, 1.00)	0.75 (0.34, 1.00)	0.85 (0.45, 1.00)	0.81 (0.36, 1.00)	0.78 (0.29, 1.00)

Note. The numbers in parenthesis represent the smallest and largest values observed in the gender-age group. Means are calculated using the sample weights, while the “Number of observations” is the unweighted number of respondents. WASO = wake after sleep onset.

For the full sample of individuals with birth years 1920–1947, the average assumed sleep was 7.9hr (Table 6). This was higher for women than men and there was a slight trend toward longer assumed sleep with increasing age. Actual sleep averaged 7.3hr, and this increased with age for men but not for women, with the mean being higher for women than men at younger ages but similar in the oldest age group. WASO averaged 39min, and this increased with age for men but not for women. The mean fragmentation index was higher for men than women in each age group and increased with age for men. Percent sleep averaged 82% overall and decreased with age for both men and women. Figure 1 shows the distribution of actigraphy-estimated average sleep durations (“actual sleep”) by hours of self-reported sleep (from the single question “How many hours do you usually sleep at night?”). This graph shows that there is not a high correlation between the self-reported and actigraph-estimated durations.

Figure 1.

The distribution of the actigraph-estimated sleep duration (“actual sleep”) by hours of self-reported sleep in response to the question “How many hours do you usually sleep at night?” For each boxplot, the central line is the median and the top and bottom of the box represent the interquartile range. Outliers are marked as small circles. These data are unweighted.

Discussion

NSHAP greatly expanded the opportunities to investigate hypotheses related to sleep in the Wave 2 data collection. A few additional questions were added to the Wave 2 core questionnaire, but the more important additions were in the randomly selected one-third sleep module that successfully collected an objective measure of sleep, wrist actigraphy. The sleep module also included a 3-day sleep log and several additional survey questions. There is, however, no perfect way to measure sleep, and the actigraphy-derived variables should be thought of as estimates of sleep parameters. Because NSHAP also collected survey-based sleep characteristics, the data provide diverse opportunities to compare both ways of measuring sleep characteristics and examine how each relates to social factors, biomarkers, and health. Finding differences between survey responses and an objective measure is not unique to sleep: Differences have similarly been found between self-reported physical disabilities and performance-based measures, with both independently predicting short-term mortality and nursing home admission (Guralnik et al., 1994).

There are important limitations to the actigraphy data. We only collected data during a 72-hr interval based on concerns that compliance would suffer with a longer protocol. Given the amount of day-to-day variability in sleep, a longer interval would provide a better estimate of the individual’s underlying mean and also permit additional within-respondent analyses of variability (Rowe et al., 2008). However, 3-day data collection is probably less of a limitation in this mostly retired population than among younger, working adults. For example, we did not find any evidence of a day-of-the-week effect, which is usually seen in working age or school age populations. Another limitation is that the data do not support the calculation of sleep latency for all individuals and all nights. Sleep latency is the amount of time it takes an individual to fall asleep initially after she or he begins to try falling asleep. It is difficult to determine when an individual began trying to fall asleep unless the event marker was used, which was only true 84% of the nights. Finally, while there are 195 couples where both partners provided actigraphy data but the data were not collected contemporaneously, thereby precluding an hour-by-hour analysis of how one partner’s sleep or sleep quality affects the other. This was done to avoid the possibility of the actiwatches being inadvertently switched were they sent to both members of a couple at the same time.

Suggestions for Analysis

The public use actigraphy data file includes the day of the week for each for the three nights.

For each night, there are five variables derived from actigraphy, listed in Table 5, which also includes the corresponding variable name. In addition, the sleep log bedtime and waketime are included for each night, both on a 1,440min basis beginning at noon (log_bed and log_wake). Finally, there is a nap variable for each day (log_nap).

Most analyses will want to use the mean values across nights, although estimates of variability across the three nights may also be of interest. Caution should be used when including more than one of the actigraph sleep variables in the same statistical model because they are all interrelated and almost all are strongly correlated with each other. Specifically, note that acti_sleep and waso sum to assumed_sleep, and that percent_sleep is the ratio of acti_sleep and assumed_sleep.

Analysis of changes in sleep quantity and quality from Wave 1 to Wave 2 are limited by the changes in how the sleep data were collected, and the relative paucity of sleep questions in Wave 1. Comparisons of the three questions asked identically in the two waves are straightforward. However, comparisons of self-reported sleep hours across waves are necessarily limited because the duration data were collected differently in the Wave 2 core. Investigators may want to consider whether a crosswalk can be developed between the single question and the sleep duration derived from the usual bedtime/waketime questions in the Wave 2 core by investigating the responses to the single question in the Sleep and Activity Booklet for in Wave 2.

The enhanced sleep data in Wave 2 of NSHAP, combined with the rich survey and biomarker data, should allow researchers to advance our knowledge of how objective and survey-based measures of sleep relate to health and aging.

Key Points .

• In Wave 1, sleep data are limited to four survey questions.

• In Wave 2, sleep data collection was enhanced for all respondents in two ways: one way, by including additional questions about naps and sleep medications and another way, by using a different method to determine usual sleep duration—separately querying usual bedtimes and waking times—which may be more precise and less subject to reporting bias.

• A modularized sleep substudy in Wave 2, in which 785 of the Wave 2 respondents participated, included a 3-day sleep log, additional survey questions used in other population-based studies of older individuals, and an objective sleep measure, wrist actigraphy.

• Wrist actigraphy provides three nights of data that give several estimated sleep characteristics about duration and sleep quality.

Funding

The National Social Life, Health, and Aging Project is supported by the National Institutes of Health, including the National Institute on Aging (R37AG030481; R01AG033903), the Office of Women’s Health Research, the Office of AIDS Research, and the Office of Behavioral and Social Sciences Research (R01AG021487), and by NORC which was responsible for the data collection. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health, or NORC. In addition, the sleep data collection received support from Philips Respironics and the Health and Retirement Survey.