Comparison between subjective and actigraphic measurement of sleep and sleep rhythms

Transcription

1 J. Sleep Res. (1999) 8, Comparison between subjective and actigraphic measurement of sleep and sleep rhythms STEVEN W. LOCKLEY, DEBRA J. SKENE andjosephine ARENDT School of Biological Sciences, University of Surrey, Guildford, UK Accepted in revised form 27 January 1999; received 28 September 1998 SUMMARY Sleep is often assessed in circadian rhythm studies and long-term monitoring is required to detect any changes in sleep over time. The present study aims to investigate the ability of the two most commonly employed methods, actigraphy and sleep logs, to identify circadian sleep/wake disorders and measure changes in sleep patterns over time. In addition, the study assesses whether sleep measured by both methods shows the same relationship with an established circadian phase marker, urinary 6- sulphatoxymelatonin. A total of 49 registered blind subjects with different types of circadian rhythms were studied daily for at least four weeks. Grouped analysis of all study days for all subjects was performed for all sleep parameters ( days data per sleep parameter). Good correlations were observed when comparing the measurement of sleep timing and duration (sleep onset, sleep offset, night sleep duration, day-time nap duration). However, the methods were poorly correlated in their assessment of transitions between sleep and wake states (sleep latency, number and duration of night awakenings, number of day-time naps). There were also large and inconsistent differences in the measurement of the absolute sleep parameters. Overall, actigraphs recorded a shorter sleep latency, advanced onset time, increased number and duration of night awakenings, delayed offset, increased night sleep duration and increased number and duration of naps compared with the subjective sleep logs. Despite this, there was good agreement between the methods for measuring changes in sleep patterns over time. In particular, the methods agreed when assessing changes in sleep in relation to a circadian phase marker (the 6-sulphatoxymelatonin (amt6s) rhythm) in both entrained (n = 30) and free-running (n = 4) subjects. KEYWORDS actigraphy, blindness, circadian, naps, sleep INTRODUCTION The assessment of sleep is often performed in circadian rhythm studies. Changes in sleep due to circadian rhythm disorders may take a long time to become apparent and thus require long-term monitoring to detect any changes. Polysomnography (PSG) is considered to be the gold standard method for assessing sleep but it is not always practical for longitudinal studies or field use. Two principal methods are used to measure sleep in field studies, as they are convenient for the subject, require minimal supervision and can easily be maintained. The first technique, subjective sleep assessment, Correspondence: Steven W. Lockley, School of Biological Sciences, University of Surrey, Guildford, GU2 5XH, UK. Tel.: ; fax: ; s.lockley@surrey.ac.uk 1999 European Sleep Research Society consists of daily sleep diaries or logs. Alternatively, actigraphy, using wrist-worn activity monitors, indirectly assesses when sleep occurs from the amount of physical activity detected. Both methods, however, have shortcomings. For subjective measurement of sleep, there is evidence to suggest that people have difficulties assessing their own sleep especially when suffering from insomnia (e.g. Carskadon et al. 1976) and that there may be gender- and age-related effects on subjective sleep assessment (Reyner and Horne 1995). Although there is obviously some difficulty in recollecting exact sleep times or duration, in long-term studies it should be easy for subjects to distinguish between major differences in sleep parameters (e.g. between 5, 30 and 60 min for sleep latency). However, the limitations of subjective sleep assessments need to be kept in mind when interpreting the results. Activity monitors have been widely used in a variety of 175

2 176 S.W. Lockley et al. clinical investigations, including identification of sleep disorders marker. The first question our study addresses is whether and drug treatments (for review see Sadeh et al. 1995). They subjective and actigraphic methods give the same absolute do not depend on subjective measurement but difficulty may values for night and day sleep parameters. As has been noted arise in the interpretation of activity levels. For example, the previously (Carskadon et al. 1976; Hauri and Wisbey 1992; monitor cannot distinguish sleep from inactivity or night- Sadeh et al. 1995), there may be discrepancies in night sleep time restlessness during sleep from night-time awakening. This measured by subjective and objective (actigraphic and/or PSG) problem may be accentuated in individuals with sleep disorders methods. To our knowledge, there has been no systematic, longwho spend time lying still but awake in bed, or those that may term comparison of actigraphic and subjective measurement of sleep restlessly (Sadeh et al. 1995). spontaneous napping under normal conditions although the There are also some problems associated with PSG. This Multiple Sleep Latency Test (MSLT) has been assessed technique usually requires individuals to sleep in research simultaneously by PSG and actigraphy (Levine et al. 1986). laboratories which are known to change habitual sleep patterns The second question is whether actigraphically assessed sleep (Reynolds et al. 1992) and induce a first-night effect (Riedel and sleep logs are able to assess similar changes in sleep et al. 1998). There may also be inter-individual differences in patterns, thus helping identification of a circadian sleep-wake the scoring of results. Even though all of these methods attempt disorder. Thirdly, the study investigates whether the established to measure sleep, they measure different things, i.e. subjective relationship between sleep timing and circadian phase is recollection of sleep, motor activity or electrical activity of the observed in both subjective and actigraphic data. brain. Previous work, using primarily subjective sleep measurement, has shown that blind individuals are likely to have circadian METHOD sleep/wake disorders (Miles et al. 1977; Okawa et al. 1987; Subjects Arendt et al. 1988; Klein et al. 1993; Lockley et al. 1997a, 1999). They are caused by insufficient light input from the A total of 49 subjects were studied (36 males, 13 females; mean retina to the central circadian oscillator, situated in the age±sd = 46.6±12.2 years). All subjects were registered as suprachiasmatic nuclei, for entrainment of rhythms to a solar blind and suffered from a range of pathologies (Lockley et al. 24-h day. Longitudinal measurement of the sleep/wake pattern, 1997b). Of these 49 subjects, 19 had light perception or better with simultaneous assessment of a reliable phase-marker of the and 30 had no conscious light perception. They were recruited circadian clock, is required before accurate identification of a from a database compiled at the Moorfields Eye Hospital, UK disorder can be made. The three major circadian sleep/wake as part of an epidemiological study of sleep disorders in the disorders, defined by the (ICSD 1990) and known to occur in blind (Tabandeh et al. 1998). Ethical permission for the study blind people, are advanced sleep phase syndrome (ASPS), was granted by the University of Surrey Advisory Committee delayed sleep phase syndrome (DSPS) and non 24-h circadian on Ethics and the Moorfields Eye Hospital Ethics Committee. sleep/wake syndrome (or free-running sleep/wake rhythm). In Subjects were excluded if they were taking any medication ASPS, the subject s sleep/wake cycle is advanced relative to a known to affect sleep and/or melatonin production (tricyclic conventional sleep time and in DSPS, the sleep/wake cycle is antidepressants, monoamine oxidase inhibitors, serotonin relatively delayed (ICSD 1990). In some blind subjects, ASPS reuptake inhibitors, benzodiazepines, neuroleptics, β-blockers and DSPS are associated with advanced and delayed amt6s or sleeping medication). Informed consent was obtained from rhythms, respectively (Lockley et al. 1999). In non 24-h sleep/ all subjects and included consent for the assessment of urine wake syndrome, the subject s clock assumes it s own for a number of drugs (opiates, amphetamine, barbiturates, endogenous period which is different from, and usually longer benzodiazepine, cocaine, cannabis, tricyclic antidepressants). than, 24 h. The sleep/wake cycle is characterised by a gradual Prior to the study, subjects underwent a full ophthalmological delay in sleep onset and offset times and frequent naps of a long examination and interview and completed the Pittsburgh Sleep duration when the circadian clock is completely desynchronised Quality Index (PSQI) questionnaire (Buysse et al. 1989). The from the 24-h day (Miles et al. 1977; Lockley et al. 1999). majority of subjects (86%) complained of a sleep disorder as Actigraphy has been used to assess free-running activity and assessed by the PSQI (score 5 indicative of a sleep disorder). sleep/wake rhythms in sighted humans in laboratory based The subjects were studied for at least four weeks. They were isolation experiments (Middleton et al. 1996). Blind children, asked to keep daily sleep and nap diaries recording sleep for whom sleep diaries may be less reliable, have also been latency, onset, offset, number and duration of night awakenings investigated by actigraphy to reveal disordered sleep and and the number and duration of daytime naps. Naps were daytime sleepiness (Tzischinsky et al. 1991). Improvement in defined as any sleep that occurred outside the subjects bedtime actigraphically derived sleep has been shown following period. The diaries were recorded in several ways according to melatonin treatment (Tzischinsky et al. 1992). the subjects preference and visual ability. The methods used The present study investigates the ability of actigraphy and were large print forms, audiotape, computer disk, and sleep logs to identify sleep/wake disorders in a group of blind Braille. Data were transcribed onto audiotape or computer subjects and whether sleep measured by both methods shows disk or dictated to the researcher. Subjects were also asked to the same relationship with an established circadian phase wear activity monitors (Motionloggers or Minimotionloggers,

3 Comparison of subjective and actigraphic sleep 177 Ambulatory Monitoring Inc., New York, USA) on the wrist range ). The same analysis was then repeated for continuously for 24 h a day for the duration of the study. The each individual. For each test, the mean difference between the monitors were preset to the zero crossing mode so that the methods was calculated by subtracting the actigraphic score signal voltage from the motion transducer is compared with a from the subjective score. reference voltage for a change in state, effectively measuring To assess whether similar trends were observed in the data activity frequency (User s Guide for the Motionlogger using the subjective and actigraphic methods, daily sleep onset Actigraph, Ambulatory Monitoring). The subjects were only and offset times were fitted with best-fit regression lines for allowed to remove the activity monitor if there was a risk of each individual. The period (tau) of any free-running rhythm it getting wet or if it interfered with any usual activity. This was calculated by adding the slope of the regression line to 24 was recorded in an activity diary. The actigraphs were [tau = slope (h)] and was considered significant if the programmed to group the data automatically into one minute 95% confidence limits of the regression line did not cross 0 (i.e. epochs (the largest epoch that can be used for automatic sleep h). Free-running periods derived by the two methods scoring). Raw data were downloaded onto computer disk at were considered non-significantly different if the 95% regular intervals and the activity monitors reset. confidence limits of the two regression lines overlapped. On For 24 or 48 h each week, subjects collected sequential 4- four occasions there was a clear jump in the sleep plots, and hourly urine samples (8 h overnight) for analysis of the major in these instances the longest continuous sequence of days urinary metabolite of melatonin. Full details of the methods without abrupt shifts was fitted with the regression line. used to determine circadian rhythm type from the 6- Correlation analysis was performed on the rhythm periods for sulphatoxymelatonin (amt6s) rhythms are described in sleep onset and offset between the two methods. Lockley et al. (1997b). Data analysis Calculation of sleep parameters Statistical analysis Using all the data for all the subjects as a group, Pearson correlations and Students t-tests were performed between subjective and actigraphic results for sleep latency, number and duration of night awakenings, onset, offset, total night sleep duration, and the number and duration of day-time naps (n For the subjective data, the sleep parameters were calculated as follows. Sleep onset was the sum of time at which sleep was attempted and estimated sleep latency. Sleep offset was taken as the wake up time. Total night sleep duration was calculated by subtracting the minutes of night sleep lost from the time between sleep onset and offset. Activity data were edited using the subjects activity diary and analysed by the Action3 software (Ambulatory Monitoring Inc., New York, USA). Days with more than 33% of data points missing were considered incomplete and excluded from the analysis. Only full nights with no data missing were used. To calculate night sleep parameters, automatic sleep/ wake scoring was performed on the activity data between bedtime and get up time derived from the subjective sleep diaries. Similarly, day sleep parameters were automatically scored for sleep between get up time and bedtime derived from the sleep logs. The sleep scoring algorithm assesses the activity of the preceding four min and the subsequent two min of each minute before defining whether each minute is awake, asleep or both. Automatic sleep scoring gave measures of night-time sleep latency, onset, offset, number and mean duration of night awakenings. The automatically derived day- time parameters were total day sleep duration and number of separate sleep episodes. amt6s analysis Urinary amt6s concentrations were measured and analysed by cosinor analysis to reveal the peak or acrophase (phi) of the amt6s rhythm. The results of these measurements are discussed in detail in Lockley et al. 1997b. Using the acrophase times, subjects were classified into four circadian rhythm categories; normally entrained to 24 h (NE; phi range = h, n = 21), abnormally entrained to 24 h (AE; phi range = h, n = 9), free-running (FR; tau range = h, n = 17) or unclassified with no discernible rhythm (UN; n = 2). One of the subjects clearly had a free-running amt6s rhythm but, due to lack of urine samples, an estimation of tau was not possible and this subject was excluded from analysis that required a tau estimation. Unclassified subjects were omitted from any further analysis involving amt6s rhythms. Sleep in relation to circadian phase In subjects that did not show free-running amt6s rhythms, the mean amt6s acrophase time (phi) was used to define circadian phase. Multiple regression analysis (MANOVA) was performed between this value, the mean subjective and actigraphic sleep parameters and age of the subject. In subjects with free-running amt6s rhythms, the regression line fitted through the amt6s acrophase times was used to calculate the daily amt6s acrophase for each person. Mean subjective and actigraphic sleep parameters were compared during the subjects normal (amt6s acrophase h) and abnormal (amt6s acrophase h) circadian phase. For those subjects that passed through a full circadian cycle during the study, sleep and nap data were firstly expressed in relation to each individual s overall mean for each parameter and then grouped into hourly amt6s acrophase times. One- way ANOVA was used to calculate whether there was an overall

4 178 S.W. Lockley et al. significant change in these parameters with circadian phase (amt6s acrophase). For details of the analysis and results of the subjective sleep data see Lockley et al. (1999). RESULTS Sleep parameters The mean (±SD) number of nights analysed per subject was 23±7 (range 6 35). At least 1062 nights of data were compared for all six night-sleep parameters (range days per sleep parameter). Two subjects failed to record the minutes of sleep lost per night in their sleep logs and therefore a subjective estimate of the total night sleep could not be calculated for these individuals. Two subjects also failed to record sleep latency but did record an estimated sleep onset time. The mean (±SD) number of days analysed per subject was 23±7 (range 6 36). In total, 1112 days of data were compared for both nap parameters. Figure 1 shows the correlation analyses for all the raw data for all subjects (n range ). Good correlations were observed when comparing the measurement of sleep timing and duration, i.e. sleep onset [r = 0.77, n = 1150 (Fig. 1C)] and sleep offset [r = 0.88, n = 1150 (Fig. 1D)] and to a lesser extent, night-time sleep duration [r = 0.57, n = 1109 (Fig. 1B)] and day-time nap duration [r = 0.48, n = 1112 (Fig. 1H)]. In contrast, the methods were poorly correlated in their assessment of transitions between sleep and wake states, i.e. sleep latency [r = 0.12, n = 1062 (Fig. 1A)], number of night awakenings [r = 0.06, n = 1122 (Fig. 1E)], duration of night awakenings [r = 0.22, n = 1072 (Fig. 1F)] and the number of day-time naps [r = 0.05, n = 1112 (Fig. 1G)]. The correlation coefficients between the two methods varied considerably between individuals and between sleep parameters. Sleep onset, offset and night sleep duration were generally highly positively correlated although the large ranges indicate the extent of the Figure 1. The correlation of all raw data for all subjects over all variation between subjects (mean r±sd = 0.68±0.37, nights (n range = ). The vertical axes show actigraphic values and the horizontal axes show subjective values. 0.59±0.43 and 0.64±0.26, respectively). The other parameters were generally less well correlated with a similar large variation between individuals (mean r±sd = 0.45±0.30 for the proportion of subjects that showed a significant difference duration of naps; 0.29±0.26 for the duration of night between the two methods than did not (χ 2 = 34.5, P < ). awakenings; 0.22±0.25 for the number of naps; 0.19±0.32 for The incidence of significant differences was greatest in the the number of night awakenings and 0.07±0.29 for sleep number of naps (41/49 subjects, 84%) and least for sleep onset latency). (19/49, 39%) (Fig. 3B). The discrepancy between the two When analysed as a group, there were statistically significant methods was largest when measuring the number or duration differences between the absolute measures of all sleep of sleep or wake episodes (duration and number of night parameters (Fig. 2). Actigraphy showed a longer sleep period awakenings, night sleep duration, number and duration of day- (i.e. shorter sleep latency, advanced sleep onset, delayed sleep time naps) and smallest when comparing sleep timing (sleep offset and longer night sleep duration), greater sleep disturbance onset, offset). To illustrate this point, Figure 4 shows plots of (increased number and duration of night awakenings) and all the sleep parameters measured by both methods in one greater day-time sleepiness (increased number and duration of subject. In all the analyses, the two methods gave results that naps) compared with subjective sleep assessments. differed significantly. Actigraphy measured shorter sleep latency Analysing individuals, there were positive and negative (Fig. 4A), increased night sleep duration (Fig. 4B), an advanced relationships between the two methods (Fig. 3A). Figure 3B sleep onset (Fig. 4C), a delayed sleep offset (Fig. 4D), increased depicts only those subjects that showed significant differences number and duration of night awakenings (Fig. 4E,F) and between the methods. Overall, there was a significantly higher increased number and duration of day-time naps (Fig. 4G,H).

5 Comparison of subjective and actigraphic sleep 179 Figure 2. The mean (±sem) values for all sleep parameters for all subjects as assessed by subjective diaries (white columns) and actigraphic automatic sleep-scoring (black columns) ( P< 0.05; paired Student s t-test). Sleep rhythms Free-running sleep onset and offset rhythms, calculated by regression analysis, were observed in some subjects. The subjective and actigraphic methods agreed in determining the period of sleep rhythms in 42 subjects for sleep onset, 38 of whom had taus that were not significantly different from h. The other four subjects had free-running taus that were statistically indistinguishable by the two methods. Of the remaining seven subjects, five had a free-running tau according to the subjective data whereas the tau derived from the actigraphic data was h. The other two subjects had the reverse relationship. Analysis of the subjective and actigraphic measures of sleep offset rhythms also agreed in 42 subjects; 40 had rhythms not significantly different from h and two subjects had freerunning taus that were statistically indistinguishable. The two methods did not agree in seven subjects, four of whom had a free-running rhythm according to the subjective data but were entrained according to the actigraphic data. In the other three subjects this relationship was reversed. When the taus assessed by both methods were compared in Figure 3. The mean difference (±SD) between subjective and actigraphic measurements of sleep. The positive direction (black bars) refers to subjective values > actigraphic values and the negative direction (grey bars) refers to subjective values < actigraphic values. Figure 3(A) shows the mean difference for all subjects. Figure 3(B) shows the mean difference for only those subjects that had a significant difference between the two methods (P< 0.05, Student s t- test). The double line delineates the left axis (minutes) from the right axis (number). The numbers above each panel show the number of subjects. all 49 subjects, there were significant correlations (P < 0.01) for sleep onset and offset, correlation coefficients (r) being 0.35 and 0.42, respectively. Sleep in relation to circadian rhythm type In order to test whether sleep was grossly affected by circadian rhythm type the mean sleep parameters were compared between NE, AE and FR subjects. There were significant differences in the subjective number of naps between the subgroups and in the subjective and actigraphic duration of day-time naps (P < 0.05, one-way ANOVA). In all cases, NE subjects had the fewest naps of the shortest duration. Both subjective and actigraphic measures of night-time sleep duration showed significantly greater sleep duration in NE subjects compared to AE and FR subjects. Sleep in relation to circadian phase Entrained subjects Thirty subjects had amt6s rhythms which were entrained to the 24 h day (n = 21 NE; n = 9 AE). When controlled for age, subjective sleep onset showed a significant relationship with amt6s phase (P < 0.05, MANOVA) with a correlation coefficient

6 180 S.W. Lockley et al. Figure 4. The plots of eight sleep parameters simultaneously measured by subjective sleep diaries (Ε Ε) and automatic actigraphy (Χ Χ) in one subject (S48) for 32 days. (r) of When the amt6s rhythm was advanced, sleep onset (subjective data, P = 0.07; actigraphic data P = 0.46) (Fig. 5A) was advanced and when the amt6s rhythm was delayed, sleep and a significant change in both measures of sleep offset onset was delayed. Actigraphic sleep onset data agreed with (P < 0.05) (Fig. 5B). Both methods confirmed that sleep onset the subjective data showing a similar but non-significant effect and offset were relatively advanced when the amt6s acrophase with amt6s acrophase (r = 0.35, P = 0.06, MANOVA). No other was advanced and delayed when the amt6s acrophase was sleep parameters showed a significant relationship with amt6s delayed. phase when controlled for age. Both methods also showed significant changes in the duration of night sleep and day-time naps in relation to Free-running subjects circadian phase (P< 0.05) (Fig. 5C,D, respectively). Night sleep duration was minimal when the amt6s rhythm was There were nine FR subjects which had both subjective out of phase and maximal when the amt6s rhythm was in and actigraphic data and passed in (amt6s acrophase phase. Conversely, day-time nap duration was maximal when h) and out ( h) of a normal circadian amt6s was out of phase and minimal when amt6s was in phase. Only subjective assessment of the number and duration phase. Only in the subjective data did the number of naps of naps showed a significant change with phase, there being change significantly with phase (Fig. 5E) reflecting the more naps of a longer duration when the amt6s phase was changes in nap duration. Of the other sleep parameters, only abnormal ( h). The actigraphic data showed the actigraphic duration of night awakenings showed a significant same trend but it was not statistically significant. change with phase (data not shown) due to a large duration Four subjects had free-running amt6s rhythms that passed of night awakenings when the amt6s acrophase was between through most (> 75%) of a full circadian cycle during the study and h. As Fig. 5 shows, there is a remarkable and had sufficient activity data for assessment of sleep across degree of similarity in the change of sleep with circadian all phases. Figure 5 shows the grouped analysis of these phase between data assessed by the two methods. However, subjective and actigraphic sleep data, plotted in relation to it should be stressed that these data are expressed in relation circadian phase (amt6s acrophase time). Both methods showed to each individual s own mean before grouping which will similar changes in sleep timing with circadian phase. There mask large variations in absolute values both between subjects was a non-significant change in both measures of sleep onset and methods.

7 Comparison of subjective and actigraphic sleep 181 Figure 5. The mean (±sem) sleep parameters for four subjects (calculated with respect to each subjects own mean) plotted against amt6s acrophase time using subjective (Ε Ε) and actigraphic (Β Β) data. The grey areas represent the mean (±2SD) amt6s acrophase time for normally entrained sighted subjects (n = 80, English and Arendt, unpublished results). DISCUSSION duration, nap duration, sleep offset, latency, duration of night awakenings and sleep onset. However, the study also demonstrates that in general, actigraphic and subjective sleep data agree when determining the rhythms of sleep and both detect similar changes in sleep in relation to circadian phase. The greatest inconsistencies between the two methods were found in the absolute measurement of the duration and incidence of sleep or wake episodes (number and duration of night awakenings, duration of night sleep and the number and duration of day-time naps). Actigraphy overestimated sleep periods (night- and day-time) and night-time disturbance compared to sleep diaries. One reason for these results could be the way in which actigraphs assess sleep, namely by measurement of motor activity. Activity monitors appear to be misinterpreting inactivity during wakefulness, both prior to and after the actual night sleep period and during the day, as sleep. In addition, they interpret night-time activity during sleep as waking periods. There is less inconsistency between the measurements of sleep timing (onset and offset) between the two methods, although in some individuals there were still large differences in either direction (even the sleep parameter with the least differences (sleep onset) showed a significant difference in 39% of individuals). We believe that sleep and nap diaries are useful tools to interpret the actigraphic measure. For example, in our experience, blind people often go to bed and listen to the radio or audiotapes before going to sleep. By asking individuals to note their bedtime and the time of trying to fall asleep separately, we can determine whether the time spent in bed is spent awake or asleep. Our results show that activity-derived sleep latency is short and sleep onsets are advanced compared to the sleep logs and therefore actigraphs are likely to be interpreting the whole in bed period as sleep. Similarly, subjective sleep offset times are likely to be accurate and again, activity-derived sleep offsets were delayed relative to the subjective measure. The nap diary is designed so that subjects only make an entry in the nap diary when a nap has occurred and therefore the periods of day-time sleeps recorded subjectively are reliable. Actigraphically-derived naps were much more frequent and of a longer duration than the subjective nap data. These results highlight the need for simultaneous use of sleep diaries with actigraphy wherever possible, in order that sleep diaries can be used to edit actigraphy data and provide points of reference for analysis. Accurate use of the subjective event marker available on some actigraphs may help reduce the inconsistency. In general, actigraphs recorded a shorter sleep latency, advanced onset time, increased number and duration of night awakenings, delayed offset, increased night sleep duration and increased number and duration of naps compared with the subjective sleep logs. The extent and magnitude of these discrepancies makes sole use of activity monitors inappropriate to assess night or day sleep in studies where absolute values for sleep periods, duration and disturbance are required. The higher degree of significant error in the assessment of day- time naps indicates that the activity monitor is unsuitable for This study compared actigraphic and subjective measurement of sleep in field conditions in 49 blind individuals living in their normal environment. The results demonstrate a large variation between the methods in the measurement of absolute sleep parameters. This difference is greatest in the number of naps, followed by the number of night awakenings, night sleep

8 182 S.W. Lockley et al. measurement of day-time sleep using the present algorithm. Unfortunately, the direction and magnitude of the differences between the two methods is not predictable and does not appear to be strongly related to any variable in these subjects. Even within a subject, the direction and significance of differences was not consistent in all the sleep parameters measured. It is important to present the direction of error because a simple average would tend towards zero as the negative and positive values cancel each other out (Chambers 1994). There is no clear consensus on the differences between subjective, PSG and actigraphic sleep data. Most previous studies have validated actigraphy against PSG recordings in an attempt to substitute actigraph-derived sleep measurement when PSG is not possible (for review, see Sadeh et al. 1995). Our conclusion that actigraphy overestimates sleep duration compared to subjective logs supports previous studies that have shown actigraphy to overestimate sleep relative to sleep diaries and/or PSG despite good correlations (Hauri and Wisbey 1992; Chambers 1994; Schmidt-Nowara et al. 1992; Cole et al. 1992) although not in all cases (Levine et al. 1986; Driver et al. 1998). The main finding of these reports was that actigraphs were only marginally acceptable when attempting to measure the exact time an individual was asleep (especially for subjects who may lie awake motionless, i.e. insomniacs) but were appropriate for documenting longitudinal changes in sleep patterns or schedules. Overall, actigraphy tended to be less accurate at times when sleep was less likely to occur. Our results confirm and extend these general conclusions. It appears that actigraphs overestimated sleep duration compared to PSG (Hauri and Wisbey 1992) to a similar extent that we have found comparing actigraphs with sleep logs. Our study found that in 66% of individuals, actigraphs overestimated night sleep duration compared to 56% reported by Hauri and Wisbey (1992). Similarly, the overall error in night sleep duration, regardless of direction, was 60 min in our study compared to 49 min (Hauri and Wisbey 1992). These comparisons suggest that sleep logs may estimate sleep more accurately than actigraphs and provide results closer to the expected PSG data. However, further work is required to assess sleep measurement (using PSG, actigraphy and sleep diaries) over prolonged periods in the field or a home environment (Broughton et al. 1996). One way to minimise the difference may be to modify the sleep/wake scoring algorithm to make it less sensitive to switches between sleep and wake states. However, this new algorithm would require extensive validation in a variety of populations prior to its routine use. There are other factors which may account for these differences. Most previous work has been undertaken in a laboratory setting which is known to affect sleep, usually by advancing the sleep offset time (Reynolds et al. 1992). The duration of the studies may also be important given that the short assessment periods used in most studies (2 3 nights) are prone to first-night effects (Riedel et al. 1998) and that reliability between subjective and actigraphic data has been shown to improve with increased number of study days (Acebo and Carskadon 1997). Comparisons between data should be interpreted with caution. Many authors report > 80% agreement between activity-derived sleep and PSG when compared on an epochepoch basis (e.g. Mullaney et al. 1980; Webster et al. 1982; Levine et al. 1986; Sadeh et al. 1989; Dunham et al. 1991; Cole et al. 1992; Hauri and Wisbey 1992; Reid and Dawson 1998). Although this suggests an acceptable level of agreement between the methods, there is no indication as to where the 20% of inaccurate comparisons occur. If, for example, the inaccuracy occurs close to the onset and offset of sleep, which our results suggest, it could give misleading results about sleep timing and duration. Similarly, studies often report a high correlation between sleep parameters assessed by different methods and conclude high degrees of consistency between the methods (e.g. Cole et al. 1992; Sadeh et al. 1995). However, whilst such a correlation may suggest a strong directional relationship between the data, it gives no indication of how similar the actual raw data are. Our study shows that despite good correlations between the data, there may still be a significant difference between the absolute sleep measurements. This discrepancy must be kept in mind if absolute, rather than relative sleep data are required. This study has also investigated the ability of both methods to detect sleep rhythms and changes in sleep in relation to circadian phase. Blind subjects with disordered circadian rhythms (Lockley et al. 1997b) are ideal to study the relationship between circadian rhythms disorders and sleep (Lockley et al. 1999). Changes in sleep were assessed by both methods in subjects with different types of circadian rhythms. Overall, there appears to be little difference in the assessment of sleep rhythms by regression analysis between subjective and actigraphic measures. In addition, the influence of permanent (in AE subjects) or transient (in FR subjects) circadian disorders on sleep is measured well by either method and could be confidently used longitudinally to assess changes in sleep in relation to phase. The choice of method may be important depending on the study population; for example, sleep diaries may be inappropriate for subjects that are not compliant or motivated, such as children or the elderly, and actigraphy may be more suitable. Subjects should always be used as their own controls to minimise the large difference in absolute values measured by subjective and actigraphic methods. In conclusion, this study demonstrates large and inconsistent errors between actigraphically-determined automatic sleep scoring and daily subjective sleep logs in a longitudinal home- based field study of sleep patterns in blind individuals. The greatest disagreement between the two methods is found in the assessment of absolute values of night and day sleep parameters with the largest error occurring in day sleep assessment. This is most likely due to misinterpretation by automatic sleep- scoring of low levels of motor activity as sleep episodes. Thus if absolute values of sleep parameters are required, the present results suggest that actigraphs are not appropriate for sole use in sleep studies, especially in determining day and night-time

9 Comparison of subjective and actigraphic sleep 183 sleep disturbance. Both methods should be employed in order Levine, B., Moyles, T., Roehrs, T., Fortier, J. and Roth, T. Actigraphic to allow cross-referencing. Despite this, our study showed good monitoring and polygraphic recording in determination of sleep and wake. Sleep Res., 1986, 15: 247. agreement between actigraphy and subjective assessment in Lockley, S. W., Skene, D. J., Arendt, J., Tabandeh, H., Bird, A. C. measuring changes in sleep patterns over time. Subjective and Defrance, R. Relationship between melatonin rhythms and diaries and actigraphs are therefore equally suited to assess visual loss in the blind. J. Clin. Endocrinol. Metab., 1997b, 82: relative alterations of sleep where absolute values are not required. Lockley, S. W., Skene, D. J., Butler, L. J. and Arendt, J. Sleep and activity rhythms are related to circadian phase in the blind. Sleep, 1999, in press. ACKNOWLEDGEMENTS Lockley, S. W., Skene, D. J., Tabandeh, H., Bird, A. C., Defrance, R. and Arendt, J. Relationship between napping and melatonin in the This work was supported by South Thames Regional Health blind. J. Biol. Rhythms, 1997a, 12: Authority, UK; Stockgrand Ltd, University of Surrey, UK; Middleton, B., Arendt, J. and Stone, B. M. Human circadian rhythms and Servier Research and Development, France. S.W. Lockley in constant dim light (8 lux) with knowledge of clock time. J. Sleep is supported by the Wellcome Trust grant /Z/96/Z. Res., 1996, 5: Miles, L. E. M., Raynal, D. M. and Wilson, M. A. Blind man living in normal society has circadian rhythms of 24.9 hours. Sci., 1977, REFERENCES 198: Mullaney, D. J., Kripke, D. F. and Messin, S. Wrist-actigraphic Acebo, C. and Carskadon, M. A. Reliability of actigraph measures in estimation of sleep time. Sleep, 1980, 3: children and adolescents. Proceedings of the Associated Professional Okawa, M., Nanami, T., Wada, S., Shimizu, T., Hishikawa, Y., Sasaki, Sleep Societies (APSS), 1997, 347. H. et al. Four congenitally blind children with circadian sleep-wake Arendt, J., Aldhous, M. and Wright, J. Synchronisation of a disturbed rhythm disorder. Sleep, 1987, 10: sleep-wake cycle in a blind man by melatonin treatment. Lancet, Reid, K. and Dawson, D. Correlation between actigraphic and EEG 1988, i: measurements of sleep in younger and older subjects during a Broughton, R., Fleming, J. and Fleetham, J. Home assessment of sleep simulated shiftwork schedule. Sleep, 1998, 21 (Suppl.): 206. disorders by portable monitoring. J. Clin. Neurophysiol., 1996, 13: Riedel, B. W., Winfield, C. and Lichstein, K. L. Anxiety and first night effect in older insomniacs. Sleep, 1998, 21 (Suppl.): 131. Buysse, D. J., Reynolds, C. F., III Monk, T. H., Berman, S. R. and Reyner, A. and Horne, J. A. Gender- and age-related differences in Kupfer, D. J. The Pittsburgh Sleep Quality Index: a new instrument sleep determined by home-recorded sleep logs and actimetry from for psychiatric practice and research. Psychiatry Res., 1989, 28: 400 adults. Sleep, 1995, 18: Reynolds, C. F., III Grochocinski, V. J., Monk, T. H., Buysse, D. J., Carskadon, M. A., Dement, W. C., Mitler, M. M., Guilleminault, C. Giles, D. E., Coble, P. A. et al. Concordance between habitual sleep G., Zarcone, V. P. and Spiegel, R. S. Self-reports versus sleep laboratory findings in 122 drug-free subjects with complaints of times and laboratory recording schedules. Sleep, 1992, 15: chronic insomnia. Am. J. Psychiatry, 1976, 133: Sadeh, A., Alster, J., Urbach, D. and Lavie, P. Actigraphically based Chambers, M. J. Actigraphy and insomnia: a closer look; part 1. Sleep, automatic bedtime sleep-wake scoring: validity and clinical 1994, 17: applications. J. Amb. Monitoring, 1989, 2: Cole, R. J., Kripke, D. F., Gruen, W., Mullaney, D. J. and Gillin, J. Sadeh, A., Hauri, P. J., Kripke, D. F. and Lavie, P. The role of C. Automatic sleep-wake identification from wrist activity. Sleep, actigraphy in the evaluation of sleep disorders. Sleep, 1995, 18: 1992, 15: Driver, H., Baker, F., Maloney, S. and Mitchell, D. A comparison of Schmidt-Nowara, W. W., Beck, A. A. and Jessop, C. A. Actigraphic objective EEG measures of sleep onset and awakenings with assessment of a treatment trial of sleep restriction in chronic subjective data in healthy young men and women. Sleep, 1998, 21 insomnia. Sleep Res., 1992, 21: 259. (Suppl.): 292. Tabandeh, H., Lockley, S. W., Buttery, R., Skene, D. J., Defrance, R., Dunham, D. W., Hoffman, R. F. and Broughton, R. J. Wrist actigraphy Arendt, J. and Bird, A. C. Disturbances of sleep in blindness. Am. and sleep-wake estimation revisited. Sleep Res., 1991, 20A: 491. J. Ophthalmol., 1998, 126: Hauri, P. J. and Wisbey, J. Wrist actigraphy in insomnia. Sleep, 1992, Tzischinsky, O., Pal, I., Epstein, R., Dagan, Y. and Lavie, P. The 15: importance of timing in melatonin administration in a blind man. ICSD-International Classification of Sleep Disorders. In: M. J. Thorpy J. Pineal Res., 1992, 12: (Ed.) Diagnostic and coding manual. Allen Press, Lawrence, Tzischinsky, O., Skene, D. J., Epstein, R. and Lavie, P. Circadian Klein, T., Martens, H., Dijk, D.-J., Kronauer, R. E., Seely, E. W. and rhythms in 6-sulphatoxymelatonin and nocturnal sleep in blind Czeisler, C. A. Circadian sleep regulation in the absence of light children. Chronobiol. Internat : perception: Chronic non-24-hour circadian rhythm sleep disorder in Webster, J. B., Kripke, D. J., Messin, S., Mullaney, D. J. and Wyborney, a blind man with a regular 24-hour sleep-wake schedule. Sleep, 1993, G. An activity-based sleep monitor system for ambulatory use. Sleep, 16: , 5: