Subjective sleepiness is a sensitive indicator of insufficient sleep and impaired waking function

Transcription

1 J Sleep Res. (2014) 23, Subjective sleepiness Subjective sleepiness is a sensitive indicator of insufficient sleep and impaired waking function TORBJÖRN ÅKERSTEDT 1,2, ANNA ANUND 3, JOHN AXELSSON 2 and GÖRAN KECKLUND 1 1 Stress Research Institute, Stockholm University and Clinical Neuroscience, Karolinska Institute, Stockholm, Sweden, 2 Clinical Neuroscience, Karolinska Institute, Stockholm, Sweden and 3 Swedish Road and Transport Research Institute, Linköping, Sweden Keywords driving, EEG, EOG, ratings, shift work, sleep, sleepiness, sleep deprivation, work Correspondence: Torbjorn Akerstedt, Stress Research Institute, Stockholm University, Stockholm, Sweden. Tel.: ; fax: ; Torbjorn.Akerstedt@ki.se Accepted in revised form 25 March 2014; received 20 February 2014 DOI: /jsr SUMMARY The main consequence of insufficient sleep is sleepiness. While measures of sleep latency, continuous encephalographical/electro-oculographical (EEG/EOG) recording and performance tests are useful indicators of sleepiness in the laboratory and clinic, they are not easily implemented in large, real-life field studies. Subjective ratings of sleepiness, which are easily applied and unobtrusive, are an alternative, but whether they measure sleepiness sensitively, reliably and validly remains uncertain. This review brings together research relevant to these issues. It is focused on the Karolinska Sleepiness Scale (KSS), which is a nine-point Likert-type scale. The diurnal pattern of sleepiness is U-shaped, with high KSS values in the morning and late evening, and with great stability across years. KSS values increase sensitively during acute total and repeated partial sleep deprivation and night work, including night driving. The effect sizes range between 1.5 and 3. The relation to driving performance or EEG/EOG indicators of sleepiness is highly significant, strongly curvilinear and consistent across individuals. High (>6) KSS values are associated particularly with impaired driving performance and sleep intrusions in the EEG. KSS values are also increased in many clinical conditions such as sleep apnea, depression and burnout. The context has a strong influence on KSS ratings. Thus, physical activity, social interaction and light exposure will reduce KSS values by 1 2 units. In contrast, time-on-task in a monotonous context will increase KSS values by 1 2 units. In summary, subjective ratings of sleepiness as described here is as sensitive and valid an indicator of sleepiness as objective measures, and particularly suitable for field studies. INTRODUCTION Sleepiness affects a large part of the population and is usually related to work hours or disturbed sleep (Ohayon, 2012), and is linked to increased accident risk (Philip and Akerstedt, 2006). Sleepiness may be defined as a drive to fall asleep (Dement and Carskadon, 1982) and polysomnographical (PSG) measurement, such as the multiple sleep latency test (MSLT), has been a gold standard for sleepiness assessment (Carskadon and Dement, 1982). However, this test is difficult to implement in real-life contexts. Performance measures or electrophysiological/ocular variables indicative of sleepiness have similar problems and also require equipment. Simpler alternatives are needed in many studies, and one such is repeated subjective ratings of sleepiness. This measure need not interrupt other activities for more than s, may be used frequently during the day and has virtually no learning curve and is, therefore, used extensively in work-hours research and many other areas. One problem, however, is that self-reported sleepiness is subjective and therefore less convincing than objective measures. Subjective measures are assumed to be influenced by expectations, mind-sets and intentional manipulation (which may also be true of objective tests). It should also be emphasized that subjective ratings of sleepiness are of interest in their own right, as the self-awareness of sleepiness reflects a 242

2 Subjective sleepiness 243 perceived state of the individual that is likely to be related to wellbeing. It is also likely to determine behaviours such as when to go to bed or when to cease certain activities (such as reading, driving, watching TV or interacting socially). Few empirical data on the behavioural effects of sleepiness are available, however. The present review attempts to bring together data on subjective sleepiness to show to what extent field and laboratory studies provide coherent patterns in sleepiness responses to different types of sleep alterations and whether subjective ratings and objective measures coincide. Historically, sleepiness ratings have been obtained via visual analogue scales (VAS) (Monk, 1989) or Likert-type scales, such as the Stanford Sleepiness Scale (SSS) (Hoddes et al., 1973) or the Karolinska Sleepiness Scale (KSS) (Akerstedt and Gillberg, 1990). VAS consists of a 100- mm line with anchors such as very sleepy and very alert. The numerical value is the number of millimetres measured from one end of the scale to a mark placed by the respondent that corresponds to his perceived level of sleepiness. The SSS ranges from very alert to completely exhausted, cannot function efficiently in seven steps, of which several do not refer to sleepiness but rather to fatigue or boredom. The KSS varies from very alert to very sleepy, fighting sleep, an effort to keep awake in nine steps. The latter scale was developed to obtain a measure that focused mainly on the propensity to fall asleep. Such information may be the event of interest in many studies, and it seems important to identify this stage ( fighting sleep or effort to keep awake ) and to separate it from milder forms ( sleepy but no effort to keep awake ) or absence of sleepiness ( neither sleepy, nor alert ). It was also thought that clear behavioural criteria ( fighting sleep, etc.) would be easier to recognize, and also might be used in the same way across individuals. The latter would be an advantage when interpreting results from different groups. The purpose of the present paper is to review the behaviour and validity of the KSS scale. Other subjective sleepiness scales have not been discussed in the present review, other than in a few cases when they have been used together with the KSS. The Karolinska sleepiness scale The scale spans nine levels and asks the user to circle the number that represents the sleepiness level during the immediately preceding 5 min (Fig. 1). The 5-min retrospective period was introduced because the act of being asked to rate may, in itself, reduce sleepiness somewhat. The scale was first tested in a 36-h sleep deprivation experiment with encephalography (EEG) and electro-oculography (EOG) measured continuously, but also measured during standardized 7-min test situations (Akerstedt and Gillberg, 1990). The latter involved the Karolinska Drowsiness Test, in which the participant is sitting down, focusing on a mark on the wall for 5 min, and thereafter sits for 2 min with closed eyes. The 1 Extremely alert 2 Very alert 3 Alert 4 Rather alert 5 Neither alert nor sleepy 6 Some signs of sleepiness 7 Sleepy, but no effort to keep awake 8 Sleepy, some effort to keep awake 9 Very sleepy, great effort to keep awake, fighting sleep Figure 1. The Karolinska Sleepiness Scale. original scale (A) had labels on every second step. However, Baulk et al. (2001) added labels to smooth the scale. In a comparison ( Akerstedt et al., in preparation), it was demonstrated that the two versions were highly correlated (mean individual r = , P < 0.001), with a mean regression coefficient of b = (P < 0.001) and a Y intercept (B) of The deviation from a regression coefficient of b = 1 is due probably to the difference in timing, scale B always having been preceded by 1 h longer time awake. KSS version A was 5.62 [standard error (SE) = 0.29] and 5.89 (SE = 0.20) for version B (P < 0.10). Version A had a clear tendency towards preferences for using the labelled steps, particularly 3, 5 and 7, rather than the intermediate steps. Version B essentially eliminated this tendency. Approach The approach has been to summarize data showing to what extent the KSS scale responds to sleep manipulations such as total or (acute or repeated) partial sleep deprivation and circadian influences in the laboratory, as well as in field situations such as shift work. It was also of interest to study the relation between KSS and electrophysiological and performance measures. Other questions concerned the effects of context, activity, work/no work and age. An important issue is also whether individuals are consistent across situations and in what situations the scale lacks sensitivity. We have selected studies answering these questions, but the selection is not exhaustive. To estimate effect sizes, we have used Cohen s d (Cohen, 1988), the difference between means divided by the standard deviation (SD), using our own raw data, but also estimated values from the data presented in papers from other groups. In such cases the figures were enlarged, the standard errors read off, transformed to standard deviation via n, and then used to compute the effect size. IS THERE A TIME-OF-DAY PATTERN UNDER DAYTIME WORK? Sleepiness ratings have seldom been used in the context of normal daytime work, but are an appropriate starting-point

3 244 T. Åkerstedt et al. and reference for the other laboratory and real-life situations raised here. The studies are summarized (with references in the legend) in Fig. 2. The participants have typically been instructed to rate their sleepiness on awakening, at 07:00 or 08:00, 10:00, 13:00, 16:00, 19:00, 22:00 hours and bedtime, although the exact times may differ somewhat between studies. In such cases interpolation has been used. The figure shows a pronounced pattern for all working-week data with high morning values (KSS 5 6), daytime values of 3 4 and an evening rise towards 5 7. The figure also shows that days off yield lower sleepiness levels. In the rightmost panel day-work sleepiness is depicted for the same group, with a 1.2-year interval. Here the curves are identical, suggesting strong stability of the sleepiness pattern. The pronounced evening increase is not seen under constant routine conditions (Cajochen et al., 1999). It is not clear what the cause may be, but laboratory conditions in a semi-reclining position, lack of knowledge of time and very controlled activity differ both from daily life and in most laboratory sleep contexts. The above results suggest that daytime ratings of 3 5 are common during a day of day work, but the studies are too small to provide reference material for the occurrence of how common are different ratings. We therefore selected a data set from a study of 226 participants with 3-h sleepiness ratings (07:00 22:00 hours) every working day during a week (Åkerstedt et al., in preparation) to present a distribution of ratings. The scale used was the earlier version A, with labels on every two steps. Fig. 3 shows that the most common score for a working day was 5, followed by 3 and 7. Clearly, the labelled steps, particularly 5, 3 and 7, attracted more ratings than intermediate scores. Figure 3. Distribution of Karolinska Sleepiness Scale (KSS) scores (07:00 22:00 hours) during 5 working days; n = 226 = 6789 values ( Akerstedt et al., in preparation). DOES EXTENDED TIME AWAKE (UP TO H) AFFECT SLEEPINESS? The obvious setting in which to test the sensitivity of a sleepiness indicator would be a 24-h sleep loss design; that is, being awake from 07:00 hours in one morning to 07:00 hours (or later) the next morning. Fig. 4 summarizes four such laboratory studies in which KSS assessment started in the evening hours. It shows that sleepiness at the beginning is typically approximately 3 4 or 4 5, depending on time, and then increases to in the early morning. In study (b), a VAS scale showed an almost identical pattern to the KSS (not in graph), with a mean intra-individual correlation of r = The effect size between 23:00 and 09:00 hours for studies (a), (b) and (d) was 3.2, 2.8 and 3.0, respectively (all represent very high effect sizes). The latter study used dim light melatonin onset as reference and Figure 2. Karolinska Sleepiness Scale (KSS) values from a working day starting at approximately 08:00 hours (a d) (Akerstedt et al., 2013a; Dahlgren et al., 2005; Ekstedt et al., 2009; Söderström et al., 2004) and a day off (a,c,e) (Axelsson et al., 2004). All are daytime workers, except for those in study (e), who are rotating shift workers on the second day off. A comparison of working days with a 1.2-year interval (b1,b2) for the same group. Mean standard error. Waking span indicated at the bottom. The right y-axis provides the verbal labels of the KSS. Figure 4. Karolinska Sleepiness Scale (KSS) ratings from four laboratory studies of h of time awake (meanstandard error) after rising approximately 070:0 08:00 hours): (a) (Krauchi et al., 2006), (b) (Akerstedt and Gillberg, 1990), (c) (Gillberg et al., 1994a), (d) (Lo et al., 2012).

4 Subjective sleepiness 245 the clock times have been estimated. It also provided 5 days of extended sleep before sleep deprivation, which may have somewhat counteracted increased sleepiness. The extended wake designs include combined effects of time awake and circadian regulation. Under forced desynchronization (life on a 20-h day) the circadian and wake-dependent components are clearly separated, with an amplitude peakto-peak of approximately 1.2 units and a fall of 0.7 units across the 13.7 h of time awake) (Wyatt et al., 1999). Using another approach, Krauchi et al. (2006) removed the time awake effect by giving repeated short naps. The result was a reduction of peak KSS from 7.1 in the early morning to 5.3 with naps. The latter would correspond to maximum sleepiness during the 24 h, controlling for the homeostatic effects of sleep. Moreover, in study (d) the circadian effect size for sleep deprivation during the circadian night was 3.5 for KSS, 1.6 for the psychomotor vigilance test (PVT) and for the memory tests involved. From the available data it appears that the KSS is highly sensitive to extended wakefulness. For comparison, one may use the PVT, which in study (d) showed an effect size of 3.0, the same size as the KSS. In contrast, working memory tests showed effect sizes below zero. DOES SLEEPINESS RESPOND TO ACUTE PARTIAL SLEEP LOSS? Partial sleep deprivation leads to more modest sleepiness responses than total sleep deprivation. H arm a et al. (1998) allowed subjects to sleep for 0, 2, 4 or 8 h (with awakening at 07:30 hours) and KSS values at 11:00 hours (with SDs) were 5.4 (1.4), 5.1 (1.8), 4.6 (1.3) and 4.1 (1.6) (highly significant difference across conditions), respectively, for the sleep durations, with an effect size of 0.9 for 0 versus 8 h. The effect size is considerably smaller than that seen in the previous studies of h of time awake, which contained both sleep loss and the effect of the circadian trough. In the H arm a et al. study comparisons were made close to the circadian acrophase, which may have attenuated the effects considerably. The effect on maintenance of wakefulness test (MWT, time taken to fall asleep while trying to resist sleep) values paralleled those of the KSS (7.5, 15, 18 and 19 min, respectively, for the 20 min allotted). Lo et al. (2012) found an increase from 3.2 to 5.3 or 5.6 after a night awake. The effect size of the KSS was 3.0 during the circadian day and 3.5 during circadian night (corresponding values for the PVT speed was 3.0 and 1.7, respectively). The conclusion is that it takes a relatively large acute reduction (down to 4 h) to affect sleepiness ratings or MSLT values. DOES SLEEPINESS INCREASE IF PARTIAL SLEEP RESTRICTION IS REPEATED? Fig. 5 summarizes three studies with restricted sleep (ending at 07:00 or 08:00 hours) across 5 days. All show significant increases in sleepiness across days, although the increase in Figure 5. Sleepiness under partial sleep deprivation across 5 days (meanstandard error) from study (a) (Lo et al., 2012) with 6 h time in bed (5.8 h of actual sleep), (b) (Banks et al., 2010) with 4 h of time in bed, (c) (Akerstedt et al., 2008) with 4 h time in bed (3.95 h of actual sleep), (ae) with extended time in bed of 10 h ( h of actual sleep). Each point represents the mean of ratings during that day. study (a) is modest, due probably to allowing 6 h of time in bed instead of the 4 h in the other two studies. For comparison, the extended sleep group of study (a) shows no increase at all. The effect size (mean of day 5 versus baseline) was estimated to 0.43 for study a (3.0 for 0 h of sleep; the PVT had an effect size of 0.28 and 3.0, respectively), 1.9 for study (b) (PVT as a reference had an effect size of approximately 1.1) and 1.7 for study (c). The results suggest that subjective sleepiness responds to repeated partial sleep restriction in the expected way. IS SLEEPINESS INCREASED IN SHIFT WORK? The most obvious real-life setting associated with sleepiness is night shift work. It usually means that the time awake is extended to 24 h, from a relatively normal awakening (07:00 08:00 hours) after a day off, across the ensuing day and through the ensuing night shift (from approximately 22:00 to 06:00 hours). This also means that the circadian trough (in the window 04:00 06:00 hours), which is a pronounced source of sleepiness, contributes to night shift sleepiness. Fig. 6 presents sleepiness during the first night shift for a number of studies. The night shift pattern is very similar across studies; early values of approximately KSS = 2 4 develop into values of between 5.5 and 7.5 towards the end of the night shift. The pattern is very similar to that seen for the laboratory studies of extended wakefulness, although the maximum values during night shift work seem to be about 1 unit lower. Possibly, this is due to the spontaneous activity with social interaction at work (see discussion below), and/or physical activity or posture (see below). Note the high sleepiness during the start of study (f), probably because it involves a cockpit crew after layover after a 6-h westward transatlantic flight (pilots), with reduced sleep as a result.

5 246 T. Åkerstedt et al. Figure 6. Night (22:00 06:00 hours), morning (06:00 14:00 hours) and afternoon shifts (14:00 22:00 hours for study (a) (Lowden et al., 1998), (b) (Axelsson et al., 2004), (c) (Waage et al., 2012) (no standard error given), (d) (Härmä et al., 1994) (combined age groups), (e) (Kecklund and Åkerstedt, 1993), (f) (Eriksen et al., 2006), (g) (Lowden and Akerstedt, 2012), (h) (Ingre et al., 2004), (i) (Kecklund et al., 1997). Meanstandard error from start, middle and end of shifts. Interpolation was used when measurement times differed from these time-points. Numbers indicate typical time for start/end of shifts. Effect sizes are possible to estimate for only a few studies, but that from study (a) for the first night shift was 1.5 from start (22:00 hours) to end (06:00 hours) of the shift. With the preceding noon value as reference (not shown in the figure), the effect size was 1.9. That of study (a) was 1.8 from start (22:00 hours) to end (06:00 hours) and that of (g) was 2.3 from start (23:00 hours) to end (07:00 hours) of the night shift. The figure also depicts sleepiness during the morning shift (starting at 06:00 hours or earlier) in several studies (some, but not all values, were obtained from the same studies as the night shifts). The values vary between KSS 4 and 6 and differ in shape and level from the night shifts. The last studies in Fig. 5 were obtained from afternoon shifts (typically ending at 22:00 hours), where sleepiness typically varies between KSS 2.5 and 4.5; that is, values similar to those at the start of the night shifts and to day work values. The latter is due probably to the fact that the time of finishing the night shift (typically 22:00 hours) permits relatively normal sleep hours. Most morning and afternoon shift values did not vary significantly during the shift, and no effect sizes were estimated. In summary, subjective sleepiness behaves in the expected way on the different shifts: high sleepiness towards the end of night shifts and intermediate sleepiness during morning shifts. DOES SLEEPINESS ADJUST TO NIGHT WORK? Working several night shifts in a row might be expected to result in adjustment of the circadian system, including sleepiness, across night shifts. This effect adjustment seems very modest, however. Fig. 7 illustrates this from a study of Figure 7. Karolinska Sleepiness Scale (KSS) (meanstandard error) during four consecutive night shifts, of which the three last are preceded by a sleep that ends approximately at noon (Lowden et al., 1998). factory workers in the chemical industry. On the first day, after awakening from night sleep the sleepiness level is low during the day and then starts to rise in the evening. The other 3 days involve wakefulness after a sleep between approximately 07:00 and 12:00 hours. No adjustment is seen. This is in line with the lack of adjustment of melatonin in night workers (Folkard, 2008). The reason for the lack of adjustment is thought to be that the commute home takes place in daylight, which in the morning prevents a delay of the circadian system, and that the other two shifts (morning and afternoon) and days off are day-orientated (Folkard, 2008). This works against adjustment, and even permanent night workers are unlikely to adjust their melatonin rhythm (a circadian phase marker) more than marginally. HOW DOES SLEEPINESS ADJUST TO NIGHT WORK UNDER OPTIMAL CIRCUMSTANCES? In contrast to the lack of adjustment of sleepiness in real-life night shift situations, it seems to occur under special circumstances, such as night work on oilrigs in the North Sea. It usually takes place indoors without the influence of outdoor light/dark changes and provides a laboratory-like setting while still being a real-life work situation. Thus, the conditions may be close to optimal for adjustment to night work. In one study the particular adjustment pattern of sleepiness was measured across 12 days. The pattern shows a gradual adjustment across the 12 days and then a dramatic increase in sleepiness during the first days off (Fig. 8). Apparently, the return to day life is as difficult to accomplish as the switch to night work. This suggests that the adjustment to night work may have been complete. The reason for the adjustment is very probably the lack of day/ night alternation of light (work was carried out indoors), with indoor light being adjusted to the sleep/wake cycle. The effect size from start to end of the first night shift was 2.5,

6 Subjective sleepiness 247 Figure 8. Daily pattern of Karolinska Sleepiness Scale (KSS) ratings during 12 days of night work (19:00 07:00 hours) and 6 days off (ashore) (Bjorvatn et al., 1998). Oil rig workers. Meanstandard error. diminishing across subsequent days. In a second study by the same group (Bjorvatn et al., 2006), 1 week of night work and 1 week of day work showed essentially the same pattern. Interestingly, melatonin secretion, an established biological circadian marker, also adjusts rapidly during night work periods on oil rigs (Barnes et al., 1998). In summary, subjective sleepiness appears to adjust to night work in the way expected in this particular context. WHAT IS THE EFFECT OF NIGHT DRIVING? A situation similar to night work is night driving, but very few studies of real driving are available. Fig. 9 summarizes four studies with min of driving between 01:30 and 05:00 hours, as well as day driving (between 09:00 and 16:00 hours). Towards 04:00 hours (end of the drive) KSS levels of are reached, compared to during the day. In (a) to (c) EEG and EOG measures (blink duration) were also included, both of which showed strongly increased levels during night driving compared to day driving. The distribution of KSS ratings in study (b) showed no ratings below 5 during night driving, but 50% of ratings between 1 and 3 during day driving. High sleepiness values between 8 and 9 characterized 15% of the night drive and 4% of the day drive. Effect sizes for day versus night drive was 2.6 for study (f), 2.5 for study (a) and 2.7 for study (c). We have found no other real driving studies with sleepiness ratings, but a number of driving simulator studies show similar effects (Fig. 9). End of drive values seem to reach to approximately KSS = 7.5 or 8, which is higher than seen in real driving. This observation is supported by the direct comparison of real and simulated driving, showing a large difference in KSS between the two settings (Hallvig et al., 2013). The same individuals participated in both studies and the road driven was approximately the same. The day night effect size for the simulator studies was between 2.1 and 2.5. The reason for the difference between real and simulated driving may be that the former includes real risk. In summary, subjective sleepiness shows the expected response to night driving TIME-ON-TASK Figure 9. Karolinska Sleepiness Scale (KSS) reported every 5 min (meanstandard error) in studies of real driving and simulated driving. Real driving studies include (a) (Sandberg et al., 2011), b+c) (Akerstedt et al., 2013b), (d) (Sandberg et al., in preparation), (e) (Sagaspe et al., 2008), (f) (Schwarz et al., 2012), (g) (Akerstedt et al., 2005), (h) (Anund et al., 2008a), (i) (Anund et al., 2009) (no standard error reported), (j) (Horne and Baulk, 2004). Time-on-task effects is a classic area of vigilance research. Performance usually falls with time spent on tasks that require increased attention such as, for example, a vigilance test (Mackie, 1977) or a reaction-time test, such as the psychomotor vigilance test (PVT) (Dinges, 1992). One might argue that latent sleepiness is unmasked in this type of task, or that the task itself adds a workload that increases

7 248 T. Åkerstedt et al. sleepiness. However, ratings of subjective sleepiness have seldom been part of time-on-task studies in the laboratory, except for one study with ratings every 10 min (Fig. 10). The results show a significant increase in sleepiness during a 40- min Mackworth clock test after 4 h of the previous night s sleep, and with an effect size of 1.6 (start to end of test). Alpha and theta power and number of misses increased significantly across the test (not presented in the figure) and the changes in both variables were highly correlated. A similar study showed that the increase of KSS and PVT reaction-times over 10 min (in two separate but similar sessions) were similar and highly correlated, while the initial KSS value was not associated with the change in PVT (Horne and Burley, 2010). It was concluded that the two measures yield very similar results if obtained under similar conditions. In addition, most of the real road studies cited above (Fig. 9) show a clear time-on-task effect of KSS of units during min of night driving and approximately 1.5 units during day driving. In the driving simulator (Fig. 9) the effects are very similar. In addition, Reyner and Horne (1997) showed that the time-on-task effect was repeatable twice in a row, after a break, using a stationary full-car simulator. The pattern during the first 30 min was one of a continuous increase from KSS = 4.5 (after a 5-h sleep) to 7.8 (start at 14:00 hours). After a break of 30 min the same time-on-task effect was seen again for 30 min, followed by a levelling-out for another hour of driving. In that study, caffeine or a nap eliminated the sleep loss effect, but not the time-on-task effect. Fig. 11 shows a highly significant increase in subjective sleepiness across driving for 2, 4 or 8 h at night with control for time of day and amount of previous sleep (the correlation with line crossings was: rho>0.60). The time-on-task effect means that sleepiness before a task will not be the same as sleepiness towards the end of the activity; the task itself has induced additional sleepiness. This raises the question of what context represents the true level of sleepiness. Is it a situation with social interaction, or a Figure 10. Meanstandard error of Karolinska Sleepiness Scale (KSS) ratings during a 40-min Mackworth clock vigilance test (Kaida et al., 2007). Figure 11. Time-on-task and Karolinska Sleepiness Scale (KSS) (meanstandard error). Ratings of sleepiness before the last hour of driving (at 04:00 hours) for 8-, 4- or 2-h driving spells, as well as a reference value from an evening drive (Sagaspe et al., 2008). situation with quiet relaxation or, perhaps, a sleepinessinducing work situation? The answer probably depends upon the way the question is formulated. There may be no correct or true context that can be used as a reference for all situations. The increase of sleepiness during time-on-task also means that correlations between performance and sleepiness at the start of the task may be low. Sleepiness ratings after, or during, a task may often be higher, as discussed above. One way of improving the relation between sleepiness and performance may be to introduce a minute of quiet relaxation before rating sleepiness (Yang et al., 2004). As demonstrated by Horne and Burley (2010), correlations may improve further with a longer period of quiet relaxation, but this might interfere with work tasks. In summary, subjective sleepiness shows the expected increase with time-on-task. DOES THE SITUATION MATTER? Apart from work tasks and time-on-task, other types of activity are likely to affect sleepiness ratings. Thus, one study (Eriksen et al., 2005) showed that sleepiness varied across the day and was acutely reduced by >2 units after a 20-min walk and by 1.5 units after 20 min of social interaction during a scheduled break, whereas quiet work in the office showed a (non-significant) slight increase during the 20 min. In a training simulator at a power plant (running the plant from computers and monitoring screens with performance data) night shift sleepiness was reduced by 1.5 units during a 15-min break (Gillberg et al., 2003) and thereafter returned rapidly to pre-break sleepiness. An even stronger reduction in sleepiness (from 7.4 to 5.2) was seen during a 30-min break during a car-driving simulator test the day after a 5-h night sleep (Horne and Reyner, 1996). Here the return to pre-break levels was rapid. Summed alpha and theta activity in that study fell in a similar way during the break, as did the number of line crossings. Similarly, during 5 days of 4 h of night

8 Subjective sleepiness 249 sleep, sleepiness was 1.1 units higher when measured after a reaction-time test compared to after 10 min of a break with social interaction (Akerstedt et al., 2008). While time-on-task is strongly sleepiness-inducing, a boring (low stimulus) environment will also increase subjective sleepiness. Driving a train through long stretches of homogenous forest is, for example, associated with a temporary increase in sleepiness of approximately 1.2 units (Ingre et al., 2004). Another influence on sleepiness is light. Even modest amounts of light (>200 Lux) may suppress sleepiness by several KSS units from a high night-time level (Cajochen et al., 2000). Also, the number of slow eye movements, and alpha plus theta EEG activity, was reduced strongly in that study. The correlations between the three parameters were >0.75. In particular, the blue end of the spectrum reduced sleepiness by 1.4 KSS units in the evening (as well as melatonin) and improved reaction-time performance (Chellappa et al., 2011). In a field study, using 1000 Lux lamps during night shift work in a nuclear power station, sleepiness was reduced from 4.8 at 04:00 hours to 3.9 (Lowden and Akerstedt, 2012). Napping will reduce subjective sleepiness by 0.5 units (compared to an increase of 0.7 units across a 30-min period with a nap (after a night of 3.7 h of sleep) (Gillberg et al., 1996). Napping shows similar results in, for example, simulated driving studies (Horne et al., 2008). The strong activity effects indicate that ratings of sleepiness may vary considerably depending on the immediately preceding context. One may speculate that many field studies of work situations may yield lower sleepiness levels because of high levels of physical activity (e.g. in manual work), whereas sedentary work in transportation may yield higher levels because of monotony and of working sitting down. This, again, raises the question of what is the true level of sleepiness. Is it the manifest level reflected in the sleepiness level now? This would seem to be the logical choice for most work situations in post-industrial society, with its self-paced occupations and variable work tasks. Latent sleepiness is reflected in, for example, a sleep latency test or neuropsychological test. This may uncover the sleepiness levels that occur in real-life situations that involve similar long-duration attentional loads. Such tasks are probably uncommon in modern work situations, except for tasks such as driving and other types of monitoring/operating work situations. Thus, the choice of measure should be based on an analysis of the character of the task. In summary, subjective sleepiness shows increases and decreases depending on the type of activity. DO CLINICAL GROUPS SHOW INCREASED SLEEPINESS? Only limited KSS data are available on clinical groups, but sleep apnea shows 1-unit higher levels of sleepiness for patients throughout the day (Wong et al., 2008), reaching 4.3 during the daytime for patients, compared to 3.0 among the controls. Another study of simulated driving found a 1.5-units (effect size = 1.4) higher sleepiness compared to healthy individuals, but also an increase in KSS after a night without continuous positive airway pressure (CPAP) (Filtness et al., 2012). Stress has differential effects on sleepiness. While acute stress may reduce sleepiness, patients suffering from longterm exposure to stress (exhaustion syndrome) showed a KSS = during the daytime versus in controls (Ekstedt et al., 2009). The effect size was 1.8. A year later, when half the group had returned to work, KSS values during daytime were with an effect size of 0.6 (compared to the previous year). Czeisler et al. studied a group of workers with shift work disorder and found a mean KSS during simulated night work of 7 while MSLT values were on average 2.5 min) (Czeisler et al., 2004). Both values indicate very high sleepiness. Treatment with armodafinil brought the sleepiness level down to 5.4 (effect size = 0.5). In addition, intolerance to night work, defined as having a very negative attitude to the shift schedule, was associated with a KSS = 7.3 towards the end of the night shift compared to the 5.7 for those with a positive attitude (effect size = 1.3) (Axelsson et al., 2004). Sleepiness is likely to be a key factor behind the attitude towards work hours. In summary, a number of clinical states exhibit increased sleepiness. Insomnia does not seem to have been studied in this respect, however. ARE INDIVIDUALS CONSISTENT? An important characteristic is individual consistency (stability) across conditions. It has not been addressed very often, but in one systematic evaluation across two bouts of 36 h total sleep deprivation, Van Dongen et al. (2004) found the intraclass correlation coefficient (ICC) for KSS to be 0.90 between two occasions of 36 h total sleep deprivation. The value indicates very high individual stability. The PVT and other performance tests had somewhat lower ICC values. In a field study of day and night driving on real roads the ICC for KSS was 0.76 across conditions and 0.52 for time-ontask (Sandberg et al., 2011). The ICC values for KDS (alpha/ theta presence in the EEG and/or slow eye movements in the EOG) were 0.53 and 0.03, respectively, and 0.63 and 0.31, respectively, for blink duration. Individual sleepiness values have seldom been included in published papers. However, Fig. 12 shows data (not previously published) for 10 participants driving a simulator home after a night shift at 08:00 hours (for 120 min), compared to driving at the same time after a full night s sleep. The results show that most individuals increase in sleepiness gradually, but strongly, even after full sleep, due presumably to the timeon-task effect and the relative monotony. The start of the increase differs between individuals and two participants seem to be resilient (do not show the increase). During the

9 250 T. Åkerstedt et al. DOES SUBJECTIVE SLEEPINESS REFLECT PHYSIOLOGICAL OR BEHAVIOURAL SLEEPINESS? Figure 12. Individual Karolinska Sleepiness Scale (KSS) ratings (5- min intervals) in a driving simulator at 08:00 10:00 hours after a night shift and after a full night s sleep (Akerstedt et al., 2005). post-night shift drive all individuals already increase to 8 or 9 after 15 min, including the two participants who were resilient in the full sleep condition. It appears that the sleep-deprived condition is forceful enough to impact upon all individuals in a similar way, while individual differences are given more room in the full sleep conditions. Genetic aspects of subjective sleepiness have not seen a great deal of research, but one of the clock genes (PER3) presents a tandem-repeat polymorphism in its coding, and the PER3 (5/5) homozygotes show larger increases in sleepiness during partial sleep restriction than the 4/4 homozygotes or 4/5 heterozygotes (Lo et al., 2012). It also seems that that group also falls asleep more rapidly (Maire et al., 2014). Subjective sleepiness is of interest in its own right, as discussed above. Nevertheless, a very central issue is to what extent subjective sleepiness measures objective sleepiness; that is, physiological or performance parameters that may reflect sleepiness. Traditional, correlative approaches have shown midsized to strong correlations with reaction-time and vigilance performance (mean r > 0.65) (Kecklund et al., 1994) or lapses (r = 0.57) (Kaida et al., 2006) during a night awake in the laboratory, or with line crossings (rho>0.60) during (real) night driving (Sagaspe et al., 2008) or with EEG measures of sleepiness (alpha/theta activity). Eye blink duration and lateral position in real driving also showed a very strong (mixedmodel regression) relation to KSS ratings (Sandberg et al., 2011). The shape of the relation Despite the robust relationship between KSS and performance or physiology, the shape of the association does not seem to be linear. The left panel of Fig. 13 shows the duration of presence of different symptoms of sleepiness symptoms during a night awake during a vigilance test at different times of day. Subjects were asked to rate the proportion of time elapsed with a particular sleepiness symptom. The presence of heavy eyelids relates strongly to KSS values, but also perceptions of being gravel-eyed and having difficulty keeping one s eyes open. While the first signs may start to occur at KSS level 7, it is not until level 8 Figure 13. Karolinska Sleepiness Scale (KSS) versus physiological and behavioural measures. Left: eye-related symptom ratings (Gillberg et al., 1994b), second from left: slow eye movements and alpha power density (Akerstedt and Gillberg, 1990), second from right: blink duration (Ingre et al., 2006), rightmost, standard deviation of lateral position (Ingre et al., 2006). Largest standard deviation indicated where available.

10 Subjective sleepiness 251 that symptoms increase, to peak at level 9. The gradual increase of heavy eyelids with increased subjective sleepiness suggests that this may be the basis for individuals sleepiness ratings. The second panel from the left (Fig. 13) depicts the relation between KSS and EEG alpha power density and time with slow eye movements during the Karolinska Drowsiness Test (5 min of eyes open every 2 h during the day and night). No sleepiness signs are seen until level 7, and thereafter a strong increase occurs. The second panel from the right depicts the relation between eye blink duration and KSS during simulated driving approximately 08:00 hours in a group of shift workers having worked all night. Again, the curvilinear shape is apparent and highly significant (mixed-model regression), with an increase starting at approximately KSS = 7. The last panel depicts a similar pattern for the standard deviation of lateral position (SDlat), a measure that is very sensitive to sleep loss, as indicated in many studies of simulated driving. Almost exactly the same shape was found for unintentional line crossings in the same study What happens at the highest level of sleepiness? In a study of awareness of sleepiness, Reyner and Horne (1998) had 28 participants drive for 2 h during the day after a shortened sleep (5 h, ending at 07:00 hours). KSS ratings were carried out every 200 s and KSS values of 1 3 showed one line crossing per hour while KSS = 9 yielded 19 line crossings (and KSS = 8 yielded 12); 83% of the line crossings with four wheels occurred at KSS 8 or 9. In all cases, a line crossing had been preceded by a rating of at least 7. It was concluded that there was always awareness of sleepiness before line crossings. In the previously mentioned simulator study (Ingre et al., 2006), the risk of line crossings with four wheels was increased 28 times at KSS = 8 and 185 times at KSS = 9, compared to the risk at KSS = 5. Using another approach, Anund et al. (2008b) investigated simulator driving in the morning after a night awake, and showed that the first time of being in contact with a rumble-strip (1 m from the lane markers) occurred at KSS = 8.1 (KSS = 6.0 at the start). All subjects reached high sleepiness levels, but those with >56% (median) ratings of 9 had 8.1 hits and those with <56% had 4.3 hits. Thus, there was a strong link between sleepiness ratings and hits. There is only one real-life study that has come close to measuring sleepiness leading up to a potentially dangerous event. In this study, 42% of the drivers during late night driving on a motorway had to be taken off the road because of dangerous driving (nodding-off, veering). This occurred at a sleepiness level of 8.5, while sleepiness in the drivers who managed to finish the drive reached 7.0 (Akerstedt et al., 2013b;). Increased KDS levels also characterized those who had to be taken off the road. Comments The studies discussed above indicate a close relation between KSS ratings and physiological and behavioural indicators of sleepiness and that values of approximately 8 or 9 are critical. The appearance of EEG-related sleep intrusions, long eye blinks and perception of heavy eyelids at levels 8 9 suggests that the perception of sleepiness may be a result of attempts by the brain to switch to sleep mode. Because the feeling of heavy eyelids would disappear if one closes one s eyes (a common, undocumented observation), it is tempting to draw the conclusion that sleepiness results from the effort of keeping one s eyes open. Is sleepiness, then, an intentional signal to cease waking activity when central nervous system (CNS) resources have been depleted, or is it merely a by-product of a closing-down process? Regardless of which of these applies, sleepiness is a warning signal of upcoming performance impairment and probably also a motivational signal for the host to go to bed. The increased evening sleepiness seems to suggest such an influence, but it does not seem to have been investigated formally. INCONSISTENCIES In some studies sleepiness ratings have not responded as sensitively as expected. One example is the repeated partial sleep deprivation mentioned above, in which the increase in sleepiness was modest compared with lapses on the PVT (Van Dongen et al., 2003). Possibly, pre-rating activities may have prevented a stronger response. In addition, in a study of sleep deprivation Van Dongen et al. (2004) showed that individuals scoring high on the KSS (or other scales) were not the same individuals who performed poorly on the PVT or on other performance measures (Van Dongen et al., 2004), but this was also true among several performance tests. In a forced desynchrony study, Zhou et al. (2012) showed that subjective ratings (a VAS scale) were less affected by sleep loss than the PVT during the night. As discussed previously, the reason for the difference in response between sleepiness ratings and the PVT could be the sleepiness-inducing effects of the test itself (due to the burden of sustained attention and lack of stimulation). Possibly, the correlation coefficient between subjective ratings and performance may be improved through sitting quietly for a minute (or more) before the rating, thus presumably establishing some similarity to the context of the performance test (Yang et al., 2004). An alternative is to rate sleepiness immediately after the test (Axelsson et al., 2008). Among other conflicting findings, Tassi et al. (2013) found no difference in KSS (2.2 versus 2.7 min) [or PVT or Epworth Sleepiness Scale (ESS)] between snorers and non-snorers, whereas the MSLT differed (12 versus 18 min). Both the KSS and MSLT values are very much in the alert range, and it may be that the low overall level and the restricted range of variation made it difficult to obtain differences.

11 252 T. Åkerstedt et al. Some correlative studies have also failed to find significant links between rated sleepiness and physiology or performance. Thus, Greneche et al. (2008) found no significant correlation between KSS and alpha or theta activity in apneics (12) or controls (8). Conversely, the correlations were somewhat high (r = 0.60) so poor power (n = 8 or 12) is a probable explanation in combination with the extremely low KSS values (1.2 in controls and 2.0 in apneics). Matsumoto et al. (2002) found that physical activity increased the dissociation between subjective sleepiness and objective performance levels during extended wakefulness in humans. A related issue concerns intentionally misleading ratings of sleepiness. This may be a problem in connection with health certification for driving licences, in which high sleepiness would risk disqualification. It might also be an issue in court cases involving sleepiness-related accidents. The use of subjective ratings in such contexts need careful consideration. Conversely, physiological sleepiness or performance measures may also be manipulated. In addition, admission of sleepiness may be discouraged in some occupational cultures. This is based mainly on anecdotal evidence, but the issue needs consideration. In summary, in certain situations subjective sleepiness may deviate from, for example, performance data. At least some of the deviation may be due to the addition of a sleepinducing factor and lack of control of the situation in which the rating is made. A SUMMARY ON EFFECT SIZES To obtain an overall impression of the effect sizes (Cohen, 1988) of how key conditions affect sleepiness ratings, means were computed from studies (at least three) on similar topics. This resulted in rather strong effect sizes for all conditions (Table 1) according to common criteria for effect sizes, with highest effect size for extended wakefulness in the laboratory setting and the lowest for repeated partial sleep deprivation (4 or 6 h of sleep per night for 5 days). One interesting observation in the search for effect sizes was that the standard deviation within groups and conditions showed an average of Thus, the individual differences in most groups seem very consistent, at least when non-clinical groups are considered. In summary, effect sizes are very high. Figure 14. Mean across three studies each of shift work (morning, afternoon and night shifts), day work, extended waking, as well as mean for one study of long-term stress (Ekstedt et al., 2009). Dispersions represent the most common standard error for studies with approximately 16 participants. CONCLUSIONS Fig. 14 illustrates some of the findings. Daytime ratings vary between 3 and 4 (alert) for healthy individuals, with values between 5 and 6 (neither/nor) being characteristic of the time of rising, bedtime, long-term exposure to stress, sleep apnea and morning work. Late night work is associated with values slightly below 7 (sleepy, but no effort), whereas a monotonous situation (e.g. driving) in a similar situation shows values of 7 9 (critical for safety). Levels 8 and 9 seem related consistently to accident risk or to increased sleep intrusions in the EEG and EOG measures. Furthermore, clinical states are associated with increased sleepiness. This possibility of using absolute values to characterize sleepiness across individuals and across situations seems a unique property of ratings of sleepiness compared to other indicators of sleepiness. In summary, subjective ratings of sleepiness are administered easily, sensitive to manipulations known to affect sleepiness, correlate with impaired waking function and appear to be used consistently across individuals. Thus, we conclude that the KSS is a reliable and valid method for measurement of sleepiness in many situations when standard objective methods may not be feasible, such as in working life. Table 1 Mean effect sizes for certain conditions Condition Effect size mean SD ACKNOWLEDGEMENTS This work was supported by Forte and Stockholm Stress Center. We thank the editor and reviewers for valuable input. Extended wakefulness (laboratory) Night drive (night versus day) Night shift (start end) Partial sleep deprivation (repeated) Means computed for at least three studies in each category. SD, standard deviation. AUTHOR CONTRIBUTIONS TÅ wrote the manuscript. The co-authors commented on the manuscript in several steps. They also contributed to the contents through numerous discussions on subjective sleepiness.