The sleep of 52 healthy paid subjects (23 male) divided into three age-bands (20–34, 35–49 and 50–70 y) were recorded at night in their homes for a total of 190 subject-nights while following their normal daily activities and habitual sleep-wake schedule. There was a shortening in both nocturnal total sleep period and total sleep time (TST) with age, the oldest group sleeping 46 min less than the youngest. Also, the mid-point of sleep occurred 32 min earlier in the oldest group compared with the youngest group. The reduction in TST with age was due, in part, to increased wake periods within sleep. The youngest subjects showed more Movement Time which progressively decreased with age while the amount of stage 1 increased with age. The amount of slow-wave sleep (SWS, stages 3+4) was reduced, stage 4 was more than halved, while REM was slightly reduced with age. There were far fewer significant gender differences in the sleep variables: males, particularly in the middle and oldest age bands, had more stage 1 than females, while females had more SWS, particularly stage 3, than males. In general, despite relatively limited subject selection criteria, there was good agreement with previous laboratory-based normative sleep values for the effect of age and gender.
The sleep literature has traditionally been biased towards young adult subjects so that other age groups have not been equally represented. In recent times this has been partially addressed with the growing interest in sleep pathologies which tend to afflict older individuals. However, there is still only limited information on the healthy middle aged population. In addition, the vast majority of normative sleep data originate typically from a college student population, sleeping alone in the unfamiliar environment of a sleep laboratory while following a structured protocol. This presents a set of circumstances much different from that which most adults experience when sleeping at home under habitual conditions.
The anxiety associated with the unfamiliar environment and procedures of a sleep laboratory has long been recognised to give rise to a first-night-effect (FNE) which is offered as an explanation of increased sleep disturbance on an initial recording night in the laboratory (Agnew et al. 1966). The extent to which the FNE is reduced by recording sleep in the subject's home is controversial. Although some authors (Sharpley et al. 1988; Wauquier et al. 1992) have found an absence of FNE in the home, this finding is not universal (Wauquier et al. 1991). There is some evidence (Webb and Campbell 1979) that the FNE in a sleep laboratory is more of a problem with the elderly, which would agree with the observation that aged subjects become more ‘set in their ways’ in terms of sleep−activity patterns (Minors et al. 1989). Therefore, altering sleeping arrangements, such as in attending a sleep laboratory, may produce more disruption in the elderly. Despite the lack of unequivocal evidence, common sense would predict that sleep is more likely to be typical in the home environment while following habitual protocols than it would be in the laboratory. The availability of highly portable and relatively unobtrusive equipment readily allows for EEG-based sleep recordings in subjects’ homes (Sharpley et al. 1988; Wauquier et al. 1992; Edinger et al. 1991).
There are a large number of reports which are reasonably consistent concerning the effects of age and gender differences on sleep. Most previous studies (Williams et al. 1974; Bliwise 1994) have shown that aging from early adulthood into old age is associated with progressively less depth and continuity in sleep measures. Also, age related changes in sleep become apparent in males at an earlier age than in females. However, an exhaustive review (Bliwise 1993) concerned with the effects of age on sleep highlighted a number of issues which are controversial, in particular; (i) the age at which gender differences in slow-wave sleep (SWS) begin to be detected; (ii) whether the proportion of time spent in REM sleep varies appreciably with age and (iii) whether the duration of nocturnal sleep diminishes with age. In addition, there are reports (Weitzman et al. 1982; Czeisler et al. 1992) that the phase of the endogenous sleep-wake, REM and body temperature rhythms are altered in elderly subjects. Such observations led to the suggestion that sleep disturbance and deterioration in the circadian timing system in the elderly are causally linked (Czeisler et al. 1992). Recording sleep whilst subjects are in their own homes, rather than in a structured laboratory setting, may reveal alterations in the apparent phase of the sleep-activity patterns due to age, because subjects are allowed to follow their habitual sleep-wake patterns in a home setting.
The study reported here was designed, in part, to record sleep from healthy adults (20–70 y) in as habitual a manner as possible. For example, usual sleeping arrangements with partners were maintained, daily activities continued as normal and bed-times were not specified. The data were collected during a large-scale field study (Ollerhead et al. 1992) of possible sleep disturbance caused by aircraft noise, involving; subjective sleep reports (Reyner and Horne 1995), actigraphy (Horne et al. 1994) and close inspection for directly evoked (time-locked) perturbations in the EEG (in preparation). Some of the sleep-wake data have been published in a preliminary form (Hume et al. 1993).
Subjects and sites
The subjects were 52 paid volunteers (23 males), who were recorded for a total of 190 subject-nights in their own homes where they maintained their normal daily activities and habitual sleeping arrangements including their normal sleep-wake schedule. The subjects were drawn from a pool of about 200 at each site (Ollerhead et al. 1992). There were 8 sites, 2 sites located adjacent to each of the four major airports in the UK; London Heathrow, London Gatwick, Manchester and Stansted, together with an initial test site near Heathrow. The subjects in each subject pool took part in a social survey and were representative of the local population (Ollerhead et al. 1992). The actimetry and EEG subject samples were in turn representative of the social survey pool. Further details of sites and subject characteristics are provided elsewhere (Horne et al. 1994). The following subject restrictions applied: they were willing to participate and were available at the specified times (e.g. not working night shifts); were neither hearing impaired nor suffering from night-time pain that seriously disrupts sleep; and were not currently taking sleeping pills, large quantities of alcohol nor medication that induced drowsiness. About 6% of potential subjects were excluded from the subject pool because of either hypnotic use or illness that impaired sleep. It should be emphasized that the subjects were not screened formally for; current or past history of sleep disorders, psychiatric or neurological illness nor pharmacological treatments, other than those indicated above. They were given structured interviews which, among other aspects, checked on their general health status including a subjective estimate of their quality of sleep. In addition, no recordings were made to assess the prevalence of daytime napping. Besides the above screening, two further criteria were applied; subjects were split into three age bands (20–34, 35–49 and 50–70 y) and recruitment was aimed at obtaining equal numbers in each band plus equal numbers of each gender. The age bands were selected because the initial social survey revealed that the adult sample (n=1636) was approximately equally divided between these age bands (33, 29 & 37%, respectively). All subjects wore wrist actimeters (Horne et al. 1994) for which they were paid ?5 (about $8) per actimeter-night and a further ?15 (about US$ 25) for each EEG recording night.
Most subjects (42/52) were recorded for four successive nights, while for a variety of reasons 3, 6 and 1 subjects were recorded for 3, 2 and 1 nights, respectively. Of the 190 subject-nights; 134 involved subjects sleeping with a partner, 27 had a partner sleeping in another bed and 29 were without a partner and alone in the bedroom. The sleep recordings were obtained using Medilog 9,000–2 recorders (Oxford Instruments Ltd, UK). Electrodes were attached, the integrity of the signals checked and the recorders set in the subjects home in the evening (before 21.30 hours) prior to each night of recording. Five channels of electrophysiological data (EEG, 2xEOG, EMG & ECG) were recorded by standard methods (Rechtschaffen and Kales 1968). As most subjects were recorded for four successive nights and, to avoid a potential night-of-the-week effect, the start day for subjects was systematically varied in order to balance each night of the week (including weekends) over all subjects. The mean number, standard deviation and range of subject-nights recorded for each night of the week were 27.14, 1.86 and 24–30, respectively. All the subjects were instructed to push an event button on the Medilog recorders when they extinguished the lights and attempted to sleep, but this instruction was not followed on 17% of recording nights. Subject reports on sleep quality and possible disturbance were completed the following morning in a log (Reyner and Horne 1995). The comparison of subject sleep reports with EEG data will be reported elsewhere.
The data were analyzed in two stages (Horne et al. 1994). First, the tapes were played back through the Medilog replay system and subjected to automatic, computerized scoring (Oxford Instruments Ltd, UK Sleep Stager, 7.4 software release). All of the tapes were then replayed and visually inspected on a screen to check the sleep stage scoring and where necessary the scoring was adjusted. Reliability checks of this hybrid method of scoring were carried out with ‘blind’ rescoring of samples of the data to produce a percentage agreement for all the staged epochs. The mean agreement between the ‘visually screen scored original’ vs. the ‘visually screen scored check’ was 94% (range, 92–97%). These values are in agreement with previous typical scoring reliability findings. In all subsequent analyses, the visually checked data were used. For all calibration settings and procedures, the recommendations outlined in the Medilog manuals based on accepted techniques (Rechtschaffen and Kales 1968) were followed, with the exception that the movement time (MT) score was set to 30% of the 30 s epoch rather than the usual 50%. This was done to increase the sensitivity of sleep stage scoring to movement and thereby improve the correlation between MT and movement detected by actimetry, which was the main technique used for assessing sleep disturbance. The sleep data were analyzed using two-way anova with age band and gender as factors. Alpha was set at 0.05. Where appropriate, posthoc tests (Tukey) were performed.
The number of subject-night observations in the youngest, middle and eldest age groups were well matched (62, 64 and 64, respectively) with mean ages of 26, 40 and 60 y, respectively. The total number of observations was biased towards females (103, compared with 87 from males) and there was a slight bias towards older females. The mean age of female subjects (44 y) was significantly higher (t-test; P<0.05) than that of males (39 y). Sleep latency could only be studied in a subgroup of 157 observations, because subjects forgot on 33 occasions to depress the event marker, to indicate lights-out.
Tables 1, 2 and 3 show mean and grand mean values for measures of sleep continuity, sleep times and latencies and sleep stages, respectively, for each age band and gender, together with results of two-way ANOVA and posthoc analysis of means. These data show clearly that age is a major influence on the sleep process.
Sleep timing and continuity variables
There was a progressive decrease in total sleep time (TST) with the oldest subjects sleeping, on average, 46 min less than the youngest. In addition, there were clear significant differences between the youngest and eldest subjects, such that the eldest subjects had; a shorter total sleep period (TSP=TST plus wake and movement time after sleep onset), and they put the lights out, got to sleep and woke earlier than the younger subjects. However, values of these measures for the middle aged group were not always intermediate between the youngest and eldest groups. The reduction in TSP and TST in the elderly group was not a consequence of going to sleep later, as the oldest subjects went to sleep earlier and both woke-up and got-up earlier than the other groups. In addition, the reduction in TST with age was due partially to more wake after sleep onset (WASO) in the eldest group. A net result of these changes was a slight progressive reduction in sleep efficiency with age. However, this reduction did not reach conventional statistical significance (P=0.08).
In keeping with the above results, there was a phase advance in the sleep-wake cycle in the older subjects, as measured by the time of the mid-point of the total sleep period. The mean values for the three age bands showed a progressive advance from 03.30, 03.21 to 02.58 with advancing age. Therefore, the mid-point of sleep of the older group occurred 32 min earlier than the youngest group. However, there was an interaction between age and gender on the mid-point of sleep such that in the middle-age band the males showed the earliest, and the females the latest, mid-point of sleep for all age-by-gender groups. This was mainly due to a combination of the middle female group going to sleep later and the middle male group waking up earlier than any other age-by-gender group. Further comparisons of age-by-gender means showed a significant advance for the oldest female group when compared with the youngest male group and the middle female group.
Sleep stage structure
The youngest subjects showed the greater MT, which progressively decreased with age. However, stage 1 sleep showed the opposite pattern to MT i.e. it increased with age. There was a clear interaction between age and gender for stage 1, such that males showed a clear increase with age, while this was only slight in females. Comparison of age-by-gender group means showed stage 1, in the elderly male group, to be significantly higher than all other groups.
The amount of SWS, particularly stage 4 sleep, showed a clear reduction with age. Stage 4 was more than halved, from the youngest to the oldest age-band, when expressed as both minutes and as a percentage of TST. In parallel with the reduction in amount of stage 4 sleep, the latency to stage 4 showed a progressive increase with age. There was least stage REM sleep in the eldest age band. This was not simply a consequence of the decrease in TST with age, as this REM reduction was also found when REM was expressed as a percentage of TST. In addition, the eldest group had the shortest REM latency when compared with the two younger groups, and the duration of the first REM period progressively increased with age.
Males vs females
There were far fewer differences in sleep variables between the sexes than between the age groups. Males, particularly in the middle and eldest groups, had more shallow sleep (stage 1) than females. On the other hand, females had more deep sleep (SWS) than males, particularly stage 3 in the oldest subjects.
Serial night of recording
anova (two-way; serial-night, age) revealed no significant differences in sleep measures between different recording nights. Therefore, there was no indication that sleep on the first night was any different from that on any subsequent night. Furthermore, there was no significant interaction between recording night and age. Therefore, for sleep at home, there was no evidence of a FNE in any age group. Similarly, there was no effect (anova; one-way, nights of the week) of different nights of the week on sleep. In addition, and importantly, the actimetry results (Ollerhead et al. 1992) showed that there was no difference between the nights (usually 4) when EEG-based recordings were obtained concomitantly with actimetry and nights (an additional 11) when only actimetry recordings were taken from the same subject, indicating that EEG based recordings per se did not affect sleep.
The effect of a bed-partner on sleep was investigated by separating the sleep data, on the basis of the sleep logs, into three categories: (i) no partner; (ii) partner sleeping in a different bed and (iii) partner sleeping in the same bed. As the results may be confounded by age (because more of the older subjects slept alone) an ancova was performed with age as a covariate. As one may predict, sleep latency was longest for the subjects who were sharing their beds with a partner and shortest if the partner was in a different bed. Subjects who had a partner in another bed spent more time in stage 1 and less time in stage 4 when compared with the other two categories and, on this basis, could be described as poor sleepers, who may have chosen to sleep in separate beds.
In general, the results from this field study, in which subjects undertook their normal daily activities and maintained their habitual sleeping arrangements, are in good agreement with previous normative data for age and gender derived mainly from laboratory studies (Williams et al. 1974; Bliwise 1993). In addition, our data indicated no evidence of a FNE nor that our oldest group were more prone to disturbance due to a FNE or the serial night of recording. However, it is important to note that the mean age of the three groups reported here were 26, 40 and 60 y, which presents a greater focus on the middle years of life compared to previous studies. In most studies of sleep in the aged, the elderly subjects are usually well over 65 years. Our field data agrees with most previous laboratory studies showing that ageing from early adulthood through middle age into old age is associated with progressively less depth and continuity in sleep measures (Bliwise 1994).
Field studies are an important adjunct to laboratory based research, but can suffer because of the reduced control that can be employed, which may pose limitations on the generality of the findings. Two potential limitations present in this study were the local conditions, i.e. adjacent to airports, and the subject selection criteria.
Despite our subjects being drawn from sites near to Airports there was clear evidence (Horne et al. 1994) that only a small minority of aircraft noises (1.2%) affected sleep while there were more substantive domestic and idiosyncratic causes of disturbance such as small children and visiting the toilet. Similar, low levels of disturbance have been found for local residents in a field study (Fidell et al. 1995) which employed behavioural responses to confirm awakenings due to aircraft noise. Therefore, the EEG data gathered from these sites adjacent to airports can be considered representative of the wider population in the sense that subjects drawn from areas not exposed to aircraft noise, but similar in other respects, would exhibit similar sleep. However, it could be argued that the closeness of an airport would tend to ‘shape’ the residents by e.g. causing the more noise prone to move away and deterring potential house-buyers who may be light sleepers. This could result in a more sleep resilient type of individual residing near to airports.
However, not all the sites were very noisy, as the project design required a range of noise locations where the quietest (Stansted) had hardly any night flights and at others, night flying was restricted. In addition, the lack of sleep disturbance at the more noisy sites was most likely due, in part, to the measures taken by the residents to insulate their homes and bedrooms against sound e.g. double glazing, for which residents at the noisiest sites recieved financial aid from the airport. Also, two thirds of the subjects had lived in their houses for at least 5 years allowing potential habituation to be well established.
There were potential limitations in the selection criteria; subjects were not questioned about previous psychiatric illness e.g. depression or current sleep disorders e.g. sleep apnoea, which may differentially affect sleep measures as a function of age and/or gender. Also, subjects who suffered insomnia, from whatever cause including aircraft noise, may have resorted to the frequent use of hypnotics, and, similar to deaf subjects, would have been excluded from our subject group. However, only about 4% of potential subjects were excluded for these reasons.
Another potential effect of our selection process was the exclusion of individuals who suffered physical illness and pain which affected sleep. This could be criticised for introducing bias because the older subjects would be more likely to be excluded for this reason. This would have the effect of improving observed sleep in the older group. However, this restriction only affected about 2% of the potential sample.
The one major consideration that favours our sample being reasonably representative of a normal sleeping population was the consistent picture of the major effects of age and gender on sleep obtained in this field study with the bulk of normative sleep data in the literature from sleep laboratories. Therefore, despite our subjects living adjacent to airports, employing relatively limited selection criteria and the subjects maintaining their habitual lifestyle, their sleep was similar to that found in laboratory based control groups.
Total sleep time
The effect of age on the total amount of sleep obtained is controversial (Bliwise 1993). Although we have found, in agreement with some other EEG-based studies (Williams et al. 1974; Gillin et al. 1981), a significant reduction in sleep during the main sleep period at night, most survey data suggest that there is no change or an increase in sleep with age (Bliwise 1993). It is possible that the older subjects in our study had a differential reduction on total sleep length as a function of aircraft noise because they are more likely to be lighter sleepers and less able to resume sleep after awakening. However, there was no significant age effect observed in the small amount of sleep disturbance which was observed due to aircraft noise (Horne et al. 1994). In this study, it was not possible to draw conclusions about the total sleep obtained per 24 h because recordings were not carried out throughout the day and therefore the possibility of naps adding to the daily sleep quota existed.
Sleep latency and continuity
Similar to previous studies (Bliwise 1993) we found no clear relationship between nocturnal sleep latency and either age or gender, but the middle age group fell asleep significantly faster than the eldest group. This could be due to the fatigue associated with the combined effects of work and family commitments, which are generally more demanding below 50 y. There were some interesting interactions between sleep onset times and wake-up times for these groups which may explain this variation in sleep latency. A possible reason for this low sleep latency could be that the females in this group went to sleep later than any other age/gender group (at about midnight) when the propensity for sleep according to the circadian phase is reasonably high. Also, the males in this group woke up the earliest (at about 06.20 hours). This could lead to a long ‘day’ with increased fatigue and quicker entry to sleep. Unfortunately, no data was collected on whether ‘wake-ups’ were either spontaneous, planned via an alarm-clock or due to some other cause e.g. children.
The continuity of sleep was clearly affected by age, in terms of increased wakefulness during the sleep periods in the eldest subjects. This agrees with the increased number of awakenings reported by these subjects in their morning sleep logs (in preparation) and agrees with most studies of the sleep of the elderly in which nocturnal sleep efficiency (TST/Time in bed) is reduced to about 70–80% (Bliwise 1993). In our study, sleep efficiency was reduced with age, but just failed to reach statistical significance, and was not as great (i.e. 88%) as reported by others with older subjects. This difference between the present findings and previous reports may be due to both the relatively young age of our eldest group (mean age 60 y) and because they were sleeping at home and subjected to less unfamiliarity, and therefore possibly less anxiety, as well as being allowed the freedom to better match their perceived need for sleep with their time in bed.
One interesting result was the decrease in MT with increasing age. This agrees with the actimetry results (Reyner and Horne 1995). Here, MT was set at 30% for movement artefact, unlike the usual 50% of the 30 s epoch (Rechtschaffen and Kales 1968). Consequently, this yielded a larger than usual number of epochs of MT for all ages. Comparison of this measure with other studies is difficult because it is not frequently stated as a separate epoch score, but is more often combined as part of the wake score. It seems to be the case that young adult sleep is associated with more movement but less wake within the sleep period and they are therefore able to maintain their sleep better, while older subjects have less movement and more wake.
The progressive increase in stage 1 sleep with age, observed in this study, has been widely reported (Bliwise 1993) and parallels the increases in wake and the decrease in sleep efficiency with age. However, this stage 1 increase was hardly present in the females but was particularly striking in the males. This gender effect was suggested, to a limited extent, by a meta-analysis of polysomnographic data (Rediehs et al. 1990) which indicated that, in general, older females sleep better than older males, despite frequently reporting subjectively poorer sleep than males.
Similar to earlier reports (Bliwise 1993) the clearest age related changes were in SWS and particularly stage 4, which declined substantially with age. There was significantly more stage 4 and SWS in the youngest group than both the middle and eldest groups, indicating that if SWS propensity is a marker for a biological aging process, a significant reduction was present from the group with a mean age of 26 to the one of 40 y. Another clear pattern in our data is the differential effect of aging on the sexes where the male sleep process appears to age more quickly, with less SWS (in particular stage 3) and more stage 1 sleep, when compared with females. This gender difference in SWS decline has recently been reported (Ehlers and Kuper 1997) to emerge between 30 and 40 y. Our data reflects this trend without reaching significance. In the present study, the females were older, on average, than the males. Therefore, as females tend to show fewer effects of age on sleep than males, the effects of age on sleep reported here, cannot be attributed to there being more older males than females.
The effect of advancing age on the amount of REM sleep is controversial (Bliwise 1994). In our data, we observed a small but clear reduction in REM with age, in both absolute amount and as a percentage of TST. This agrees with a meta-analysis (Benca et al. 1992) which indicated a reduction in REM time with increasing age.
Phase shift of REM propensity and sleep-wake rhythm
There are reports that show REM shifting earlier in the night in older subjects, but some authors have observed this only with depressed patients (Bliwise 1994). Our data showed significant alterations in the distribution of REM over the night with age with a significant increase in the amount of REM during the first 120 min of sleep with age. Post-hoc comparison of these means revealed that there was significantly more REM in the two older groups than the youngest group.
There are many reports indicating that the elderly have earlier bedtimes and wake-up times than younger subjects, i.e. a phase advance of the sleep-wake rhythm (Bliwise 1993) which has been associated with a phase advance of the body temperature rhythm (Czeisler et al. 1992; Monk 1991; Campbell et al. 1989). Such findings prompted the suggestion (Czeisler et al. 1992) that changes in the output of the human circadian pacemaker, in terms of a phase advance and a reduced amplitude of rhythms, could underlie the common complaints of sleep disturbance among the elderly. In keeping with these age effects we found a progressive phase shift in the sleep-wake cycle, as measured by the mid-point of the sleep period (i.e. time from initial sleep onset to final wake-up), with the oldest group advanced by 32 min compared to the youngest group. There was a highly significant interaction between age and gender for this measure, such that the mid-point of the sleep period for females in the middle age-band was the latest for all age-by-gender groups while for the males, in this middle age-band, it was the earliest. As discussed previously, these groups showed the latest sleep onset and earliest wake up times, respectively, of all the age by gender groups.
Unfortunately, we did not record data on occupational status or whether the subjects used alarm-clocks for morning wake-ups. One might predict that the older group members were more likely to be retired and less likely to use alarm clocks to terminate their sleep than the younger groups. In which case the phase shift that we observed would be an underestimate of the underlying phase setting. In agreement with this suggestion, a recent report (Haimov and Lavie 1997) on the circadian characteristics of the sleep propensity function in healthy males showed a phase advance of 56 min in elderly (aged 65–78 y) compared to young (aged 19–25) subjects using an ultra short sleep-wake paradigm.
In conclusion, this work has provided more data on controversial issues in age and gender sleep differences and demonstrated that healthy people living under entrained conditions, following their habitual pattern of sleep and wakefulness, showed, in general, good agreement with the literature, derived mainly from the laboratory. It is essential, for the full understanding of sleep phenomena, that observations are made in the controlled environment of the laboratory but such findings must be supported by detailed study in the field.
We would like to thank the following people for their help in the planning, organisation and operation of the project: Ian Diamond, Jim Horne, Ceril Jones, Francesca Pankhurst, John Ollerhead and Louise Reyner. The data reported here forms part of a project that was financed by the UK Department of Transport as reported in Ollerhead et al. (1992).