Subjective sleepiness, simulated driving performance and blink duration: examining individual differences.
ABSTRACT The present study aimed to provide subject-specific estimates of the relation between subjective sleepiness measured with the Karolinska Sleepiness Scale (KSS) and blink duration (BLINKD) and lane drifting calculated as the standard deviation of the lateral position (SDLAT) in a high-fidelity moving base driving simulator. Five male and five female shift workers were recruited to participate in a 2-h drive (08:00-10:00 hours) after a normal night sleep and after working a night shift. Subjective sleepiness was rated on the KSS in 5-min intervals during the drive, electro-occulogram (EOG) was measured continuously to calculate BLINKD, and SDLAT was collected from the simulator. A mixed model anova showed a significant (P < 0.001) effect of the KSS for both dependent variables. A test for a quadratic trend suggests a curvilinear effect with a steeper increase at high KSS levels for both SDLAT (P < 0.001) and BLINKD (P = 0.003). Large individual differences were observed for the intercept (P < 0.001), suggesting that subjects differed in their overall driving performance and blink duration independent of sleepiness levels. The results have implications for any application that needs prediction at the subject level (e.g. driver fatigue warning systems) as well as for research design and the interpretation of group average data.
- SourceAvailable from: Kenneth Holmqvist[Show abstract] [Hide abstract]
ABSTRACT: Driver sleepiness contributes to a considerable proportion of road accidents, and a fit-for-duty test able to measure a driver’s sleepiness level might improve traffic safety. The aim of this study was to develop a fit-for-duty test based on eye movement measurements and on the sleep/wake predictor model (SWP, which predicts the sleepiness level) and evaluate the ability to predict severe sleepiness during real road driving. Twenty-four drivers participated in an experimental study which took place partly in the laboratory, where the fit-for-duty data were acquired, and partly on the road, where the drivers sleepiness was assessed. A series of four measurements were conducted over a 24-h period during different stages of sleepiness. Two separate analyses were performed; a variance analysis and a feature selection followed by classification analysis. In the first analysis it was found that the SWP and several eye movement features involving anti-saccades, pro-saccades, smooth pursuit, pupillometry and fixation stability varied significantly with different stages of sleep deprivation. In the second analysis, a feature set was determined based on floating forward selection. The correlation coefficient between a linear combination of the acquired features and subjective sleepiness (Karolinska sleepiness scale, KSS) was found to be R = 0.73 and the correct classification rate of drivers who reached high levels of sleepiness (KSS ⩾ 8) in the subsequent driving session was 82.4% (sensitivity = 80.0%, specificity = 84.2% and AUC = 0.86). Future improvements of a fit-for-duty test should focus on how to account for individual differences and situational/contextual factors in the test, and whether it is possible to maintain high sensitive/specificity with a shorter test that can be used in a real-life environment, e.g. on professional drivers.Transportation Research Part C Emerging Technologies 01/2013; 26:20-32. · 2.82 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Research suggests that between 10 and 20% of all road crashes are caused by sleep deprivation. This has led to an interest in finding different countermeasures and one of them is technical solutions like "Sleepiness Warning Systems" that identify sleepiness as a factor for increasing the risk of a crash. Such systems often consist of sensors for measuring physiological and behavioural changes, as well as algorithms used to quantify such changes and predict risk. Not so much development or research has addressed the warning strategies and how to give the driver feedback/warning in a way that will make a sleepy driver aware of the dangerous situation and act accordingly. In Europe there are projects like AWAKE and SENSATION with the aim to develop a system that predict/detect and support the sleepy drivers or operators. Also the Swedish government has recognised the problem and try to find solutions. A big step was taken with the project DROWSI. This three year project started in 2006. It is part of the national research and development programme IVSS (Intelligent Vehicle Safety Systems). The project is conducted in cooperation with the authorities, research institutes, universities and industries. The knowledge and experience from these activities will be the starting point for the future systems.
- [Show abstract] [Hide abstract]
ABSTRACT: Sleepiness and fatigue are important risk factors in the transport sector and bio-mathematical sleepiness, sleep and fatigue modeling is increasingly becoming a valuable tool for assessing safety of work schedules and rosters in Fatigue Risk Management Systems (FRMS). The present study sought to validate the inner workings of one such model, Three Process Model (TPM), on aircrews and extend the model with functions to model jetlag and to directly assess the risk of any sleepiness level in any shift schedule or roster with and without knowledge of sleep timings. We collected sleep and sleepiness data from 136 aircrews in a real life situation by means of an application running on a handheld touch screen computer device (iPhone, iPod or iPad) and used the TPM to predict sleepiness with varying level of complexity of model equations and data. The results based on multilevel linear and non-linear mixed effects models showed that the TPM predictions correlated with observed ratings of sleepiness, but explorative analyses suggest that the default model can be improved and reduced to include only two-processes (S+C), with adjusted phases of the circadian process based on a single question of circadian type. We also extended the model with a function to model jetlag acclimatization and with estimates of individual differences including reference limits accounting for 50%, 75% and 90% of the population as well as functions for predicting the probability of any level of sleepiness for ecological assessment of absolute and relative risk of sleepiness in shift systems for safety applications.PLoS ONE 10/2014; 9(10):e108679. · 3.53 Impact Factor
Subjective sleepiness, simulated driving performance and blink
duration: examining individual differences
MICHAEL INGRE1 , 2, TORBJO¨RN A˚KERSTEDT1 , 3, BJO¨RN PETERS4,
ANNA ANUND4and GO¨RAN KECKLUND1
1National Institute for Psychosocial Medicine (IPM), Stockholm, Sweden,2Department of Psychology, Stockholm University, Stockholm, Sweden,
3Karolinska Institute, Stockholm, Sweden and4Swedish Road and Transport Research Insitute, Linko ¨ ping, Sweden
Accepted in revised form 21 November 2005; received 15 March 2005
The present study aimed to provide subject-specific estimates of the relation between
subjective sleepiness measured with the Karolinska Sleepiness Scale (KSS) and blink
duration (BLINKD) and lane drifting calculated as the standard deviation of the lateral
position (SDLAT) in a high-fidelity moving base driving simulator. Five male and five
female shift workers were recruited to participate in a 2-h drive (08:00–10:00 hours)
after a normal night sleep and after working a night shift. Subjective sleepiness was
rated on the KSS in 5-min intervals during the drive, electro-occulogram (EOG) was
measured continuously to calculate BLINKD, and SDLAT was collected from the
simulator. A mixed model anova showed a significant (P < 0.001) effect of the KSS for
both dependent variables. A test for a quadratic trend suggests a curvilinear effect with
a steeper increase at high KSS levels for both SDLAT (P < 0.001) and BLINKD
(P ¼ 0.003). Large individual differences were observed for the intercept (P < 0.001),
suggesting that subjects differed in their overall driving performance and blink duration
independent of sleepiness levels. The results have implications for any application that
needs prediction at the subject level (e.g. driver fatigue warning systems) as well as for
research design and the interpretation of group average data.
lane drifting, mixed models, standard deviation of lateral position
It is well known that prolonged wakefulness, partial sleep
deprivation and high sleepiness levels have negative effects on
many performance measures (Rogers et al., 2003). Sleepiness
has also been identified as a major risk factor for road accidents
(Dinges, 1995). Even though subjective sleepiness has shown
high intra-individual correlations with electroencephalogram
indicators of sleep (A˚kerstedt and Gillberg, 1990; Cajochen
et al., 1999; Horne and Baulk, 2004; Torsvall and A˚kerstedt,
1987) and subjective sleepiness has been found to predict
performance (Dorrian et al., 2000; Gillberg et al., 1994), it has
been pointed out that subjective sleepiness is far from perfect in
predicting performance (Rogers and Dinges, 2003).
Contextual factors might explain why subjective sleepiness
sometimes shows a weak association with performance. Yang
et al. (2004) found that an instruction to calm down with eyes
closed for 1 min substantially increased the correlations
between subjective sleepiness and performance. Another cause
of weak associations could be individual differences that
complicate the relation between sleepiness and performance.
Recent studies have reported large individual differences on
both ?objective? performance measures and subjective sleepi-
ness ratings during sleep loss (Van Dongen et al., 2003, 2004).
Thus, there is a need for further studies of subjective sleepiness
in relation to performance and physiology under well-con-
trolled conditions with individual differences accounted for.
This paper presents such a study focused on two variables with
well-established sensitivity to sleepiness.
One of these variables is eye blink duration (Stern et al.,
1984). Increased subjective sleepiness on the Karolinska Slee-
piness Scale (KSS) has been related to increased amounts of
slow eye movements (A˚kerstedt and Gillberg, 1990) and higher
Correspondence: Michael Ingre, National Institute for Psychosocial
Medicine (IPM), Box 230, 171 77 Stockholm, Sweden. Tel.: +46 (0) 8
524 820 48; fax: +46 (0) 8 32 05 21; e-mail: email@example.com
J. Sleep Res. (2006) 15, 47–53
? 2006 European Sleep Research Society
levelsofsleepiness onavisual analoguescalehas beenrelated to
longer eye blink durations (Caffier et al., 2003). None of these
studies provided any direct estimates of individual differences.
Lane drifting while driving a vehicle is an established
indicator of sleepiness related performance (O’Hanlon and
Kelly, 1974). It can be calculated as the standard deviation
(SD) of the lateral position of the vehicle, which has been
shown to be a very sensitive measure of performance decre-
ments after treatment with hypnotics and anti-depressant
drugs (O’Hanlon, 1984). The SD of the lateral position has
also been used to assess decrements in driving performance
after prolonged wakefulness and alcohol intake in a driving
simulator (Arnedt et al., 2000, 2001). None of these studies
provided any information about individual differences; how-
ever, O’Hanlon and Kelley (1977, p. 88) reported large
individual differences in an early study on lane drifting and
sleepiness: ?However, we were impressed at the time by the
wide range of individual differences with respect to changes in
both performance and physiology as a function of time on the
road: Some drivers never appeared to tire, whereas others
behaved in a progressively more erratic manner, and a few lost
control within a relatively short time span…?.
Other researchers have also recognized the presence of
individual differences. For example, Hanowski et al. (2003)
reported that 5% of the subjects accounted for 26% of the
incidents in a study of truck drivers. Similarly, Mitler et al.
(1997) reported that 10% of the drivers accounted for 54% of
M. Ingre et al. (unpublished data) found large individual
differences in accident propensity, independent of sleepiness
levels, by applying a Generalized Linear Mixed Model
approach on accident data recorded in the present study.
In the presence of individual differences, group average
estimates, which dominate the literature, may lead to inaccur-
ate conclusions if they are used to infer effects in individual
subjects (i.e. the ?ecological fallacy?). When large individual
differences are expected, researchers have to be very careful
when interpreting group average data, or apply statistical
methods that explicitly estimate the effects at the subject level
so that individual differences may be evaluated and described.
The present study takes the latter approach.
In a recent paper from the same study, A˚kerstedt et al.
(2005) reported increased subjective sleepiness levels, longer
blink durations and increased lane drifting with increased
driving time and after working a night shift. The main
objective of the present study was to estimate the direct
association between subjective sleepiness measured with the
KSS on the one hand and blink duration and lane drifting on
the other with special focus on individual differences. A linear
mixed model approach was used to provide subject-specific
estimates to evaluate individual differences.
The overall design of the study was a within-subject experi-
mental design with two conditions: a ?Night sleep? condition
when subjects had a normal night sleep and a ?Night work?
condition when the subjects had stayed up all night working.
In both conditions, subjects drove in a high-fidelity (HI-FI) car
simulator between 08:00 and 10:00 hours. The subjects parti-
cipated in the two conditions in a counter-balanced order with
at least 3 days between conditions.
Five male and five female shift workers were recruited
through advertisements in local companies with night work.
Most came from hospitals, newspapers and an energy plant.
They had a mean age of 37 years (SD ¼ 12), drove annually
an average of 9500 km (SD ¼ 6800) and had 5–9 years of
experience as shift workers. Three participants worked only at
night while the rest alternated between night and day work.
They received a monetary compensation of approximately
€110. The study was carried out by the Swedish National Road
and Transport Research Institute, under their study guidelines,
including the Declaration of Helsinki.
The subjects were instructed to maintain their normal/work
sleep pattern and behaviour in connection with night and day
work during experiment. Before the study proper, the subjects
had a practice drive in the simulator for 20 min and practice at
using the rating scale (described below), which had been sent
out beforehand. They arrived at approximately 07:00–
07:30 hours, directly after night work or rising respectively.
After the drive the subjects were debriefed and sent home. In
the Night work condition the drivers were brought to and from
the test centre by a taxi.
A dynamic, HI-FI, moving base driving simulator was used.
The car cab was a Volvo 850 (Volvo AB, Gothenburg,
Sweden) and the system simulated acceleration in three
dimensions through roll, pitch and linear lateral motion. The
visual system presented the scenario on a 120? wide screen
2.5 m in front of the driver. The sound system generated noise
and infrasound that resembles the internal environment in a
modern passenger car. The vibration system simulated the
sensations the driver experiences from the contact between the
road surface and the vehicle. The driving scenario was a rural
two-lane road with lanes 3.6 m wide with a 0.5 m hard
shoulder. The conditions were ?summer? with a slightly hazy
sky. The signed speed limit was 90 km h)1and there was
sparse oncoming traffic or cars to follow or pass.
The measures obtained from the system included speed
(mean + variability), lateral position (mean + variability),
time to Line Crossing, steering wheel angle (mean + variab-
ility). Driving behaviour was recorded at a frequency of
12.5 Hz. In the present study, the standard deviation of the
lateral position (SDLAT) was selected as previous studies
suggested that this may be the most sleepiness sensitive
continuous performance measure (O’Hanlon and Kelly,
1974; Arnedt et al., 2000).
To calculate blink durations (BLINKD), the EOG was also
recorded by means of a Vitaport recorder (TEMEC Instru-
ments, B.V., Kerkrade, The Netherlands) using horizontal and
vertical derivations above and below the right eye. Data were
collected with a sampling rate of 128 Hz and a band pass filter
set at 0.3–25 Hz. Raw data were analysed with a modified
48M. Ingre et al.
? 2006 European Sleep Research Society, J. Sleep Res., 15, 47–53
matlab program developed by the Centre for Applied and
Environmental Physiology (Dr A Muzet, CEPA, Strassbourg,
France). It essentially involves using a low-pass filter to
establish a stable baseline for the signal, finding a threshold
that has to be exceeded to define a blink (done visually). The
definition of the start/end of the blink is based on slope and the
measurement of blink duration is carried out at mid-slope. To
reduce problems with concurrence of eye movements and eye
blinks, blink durations were calculated by finding the half
amplitude of the upswing and downswing of each blink and
computing the time elapsed between the two.
Sleepiness was rated every 5 min prompted by an instruction
displayed on the windshield, with the response given orally,
using the scale pasted to the steering wheel. This yielded a total
of 48 measures for an individual for the two conditions. The
scale used was the KSS ranging 1–9 where 1 ¼ very alert,
3 ¼ alert, 5 ¼ neither sleepy nor alert, 7 ¼ sleepy but no effort
to remain awake, and 9 ¼ very sleepy, fighting sleep, difficulty
staying awake (A˚kerstedt and Gillberg, 1990). The scale was
modified to have labels on all nine steps (Reyner and Horne,
1998) and subjects were also allowed to rate intermediate steps
with half points yielding a highest possible rating of 9.5;
however, only the integer part was used in the analyses. The
modifications of the scale were motivated by a high-rating
frequency in a highly controlled environment where the
subjects hadcontinuous access to the scale atthe steering wheel.
Data were analysed with mixed effects regression analysis by
means of restricted maximum likelihood estimation using the
Stata procedure xtmixed (StataCorp, 2003). The fixed effect
estimates were modelled as an anova structure with a nine-level
factor variable for the KSS (1–9). The KSS was used to predict
the mean level of the dependent variable in the subsequent
5-min segment. A random intercept was included in the model
to account for individual differences in the overall level of the
dependent variable. A second random effect was modelled for
the factor of the KSS with a single variance component and
zero covariance between levels, to account for level-specific
individual differences. The significance of the fixed effect
estimates was tested by means of a Wald test and the
significance of the random effects was tested by means of a
likelihood ratio test.
06:51 hours ± 13 min
7.6 ± 0.32 h. In the Night shift condition the subjects finished
work around 07:25 hours ± 17 min in the morning and
obtained 2.2 ± 0.8 h of sleep, mainly during the prior
morning or afternoon.
From descriptive data presented in Fig. 1 it is evident that
all subjects reached high levels of sleepiness as indicated by the
KSS. The figure also shows large individual differences in all
three variables measured in the present study.
The results from the estimated model with SDLAT as the
dependent variable are presented in Table 1 and show that
(mean ± SEM)
0 30 6090120
Minutes since beginning of drive
Figure 1. Observed data for the Karolinska Sleepiness Scale (left)
standard deviation of lateral position in centimetres (middle) and blink
duration in 1/100-s (right) during the Night work (solid line) and Night
sleep (dashed line) conditions for all subjects (1–10).
Subjective sleepiness, driving performance and blink duration49
? 2006 European Sleep Research Society, J. Sleep Res., 15, 47–53
SDLAT was significantly related to the KSS. Moreover, the
significant random effect SD of the intercept shows that
subjects differed in the overall level of SDLAT independent of
the KSS levels. The significant random effect of the factor
variable KSS shows that subjects also differed with respect to
specific levels of the KSS. The fixed (group mean) effect is
illustrated in Fig. 2 which shows an increasing SDLAT with
increased sleepiness. The figure also describes large individual
differences, for example, when comparing 4 and 6 with 2 and 8.
The former show a much higher SDLAT at all sleepiness levels
when compared with the latter and also a stronger increase at
the highest levels.
The plots in Fig. 1 suggest that there might be a curvilinear
relation between the KSS and SDLAT, with a stronger
increase in SDLAT at high KSS levels when compared with
low KSS levels. To test this assumption, three models were
estimated with the method of maximum likelihood so that a
likelihood ratio test could be used to assess the increase in
model fit. The random effects were identical to the model
presented in Table 1 but the fixed effect in the first model was
only a constant. In the second model, a linear trend of the KSS
was added and in the third model, a squared component of the
KSS was also added. The result suggests that there was a linear
trend of the KSS (v2¼ 38, df ¼ 1, P < 0.001) with an average
increase of 0.032 m (SE ¼ 0.004) for each level of the KSS.
Adding a squared component further increased model fit
(v2¼ 11, df ¼ 1, P < 0.001) suggesting a curvilinear relation
between the KSS and SDLAT.
The intra-class correlation coefficient (ICC) was calculated
to be 0.49 using estimates from a final model similar to the one
presented in Table 1 but with only one random effect (the
The estimates for BLINKD showed a similar pattern as for
SDLAT but with a larger heterogeneity in the subject-specific
estimates between different levels of the KSS (Table 1, Fig. 1).
Using the procedure described above revealed a significant
linear trend (v2¼ 49, df ¼ 1, P < 0.001) with an average
increase of 0.0056 s (SE ¼ 0.0006) with each level of the KSS
but also a squared component (v2¼ 8.78, df ¼ 1, P ¼ 0.003)
and ICC ¼ 0.30.
In an earlier paper (A˚kerstedt et al., 2005), the effects of time
and condition on the KSS, blink duration (BLINKD) and
standard deviation of lateral position (SDLAT) has been
reported. The present study extends the knowledge with
estimates of the direct association between the KSS and
SDLAT/BLINKD and information about individual differ-
ences. The results have shown that there is a relation between
the KSS and ?objective? measures of driving performance (SD
of the lateral position) and sleepiness (blink duration).
However, large individual differences were observed, which
suggests a complicated relation where the absolute level, as
well as the relative effect, has to be adjusted between
individuals. The effect seems to be curvilinear with a steeper
rise at high KSS levels especially for the SD of the lateral
position. This conclusion agrees well with previous research on
subjective and objective sleepiness (A˚kerstedt and Gillberg,
1990) and suggests that serious behavioural and physiological
changes do not occur until relatively high levels of sleepiness
(KSS ‡ 7) are reached.
During the experiment the subjects were alone in a HI-FI
car simulator cab in a static environment. All subjects had
Table 1 Summary of estimated models
ParametersSD of the lateral position (m)Eye blink duration (s)
Wald test of factordf
KSS880.54 (0.000)8104 (0.000)
Random effects SDSE (P-value)SDSE (P-value)
The table shows estimated means (mean), standard error of the mean (SEM) and the standard deviation (SD) with standard error (SE) of the
random effects. A Wald test was used to test for significance of the KSS and a one-degree-of-freedom likelihood ratio test was used to test for
significance of the random effects. The models included 10 subjects and a total of 424 observations.
50M. Ingre et al.
? 2006 European Sleep Research Society, J. Sleep Res., 15, 47–53
identical conditions and they were not engaged in any other
activity than driving the car and giving oral ratings of
sleepiness at 5-min intervals. This was an ideal situation for
obtaining subjective ratings (and objective measures) uncon-
taminated by other activities or other distracting situational
factors. Thus, reliable estimates of the association between the
KSS and SDLAT/BLINKD could be obtained somewhat
similar to what Yang et al. (2004) found, even though subjects
had their eyes closed in that study.
Similar to the findings of O’Hanlon and Kelley (1977), one
of the most striking effects observed in the present study was
the individual differences. Such differences have been found in
many areas, recently reviewed by Van Dongen et al. (2005),
and they offer some challenges. Thus, they need to be
accounted for by the statistical model and they need to be
taken into account when interpreting and generalizing from
the results. In the present study, the statistical model included
random effects to account for individual differences.
The SD of the random intercept for SDLAT was estimated
at 0.10 m, which is approximately the same as the average
increase between 8 and 9 on the KSS or three times the average
increase between any two levels. This finding indicates
substantial heterogeneity between subjects in overall driving
performance independent of sleepiness. The results were
similar for BLINKD but with a SD of 1.7 times the average
increase between two levels of the KSS. Individual differences
in performance have previously been observed during sleep
deprivation (Van Dongen et al., 2004) and were expected. The
present study adds to these findings that individual differences
were to a large extent independent of subjective sleepiness. All
subjects showed impaired performance at high sleepiness levels
but their baseline levels were different which seem to reflect
individual differences in overall driving skill or driving
accuracy. Interestingly, similar differences were also observed
for blink duration, suggesting individual differences also in
basic physiological eye-blink processes.
There were also significant random effect SDs of the factor
variable (KSS) for both dependent variables. The results
suggest that subjects also differed from the fixed effect
estimates with respect to specific levels of the KSS, indicating
individual differences also in the shape of the function.
Very little is known about the mechanisms behind the
observed individual differences and these issues have only
recently been addressed in research with statistical methods
capable of modelling effects at the individual subject level (Van
Dongen et al., 2004). As of today, one might only speculate on
their underlying mechanism and origin. Further research is
needed that explicitly addresses these issues.
An important question is whether the reported individual
differences should be considered a stable trait (Van Dongen
Standard deviation of lateral position (m)
Eye blink duration (s)
Karolinska Sleepiness Scale (KSS)
Karolinska Sleepiness Scale (KSS)
Figure 2. Prediction of the standard deviation of lateral position (left) and eye blink durations (right) for KSS levels 1–9. The figure shows the
estimated fixed effect (thick lines) and best linear unbiased predictions for individual subjects (thin lines). Numbers to the right identify individual
subjects. Subject-specific lines indicate the range of the observed KSS and dots indicate actual observations.
Subjective sleepiness, driving performance and blink duration 51
? 2006 European Sleep Research Society, J. Sleep Res., 15, 47–53
et al., 2004). However, this cannot be determined in the
present study because it would require replicating the
experimental conditions several times during a longer time
period. The present study can only address the issue of
stability/reliability during the two ?snap-shots? in time that the
study captured. Moreover, the KSS is a measure with error
(like most measures) and that will add to the error variance
when reliability estimates of the dependent variables are
calculated. The reported reliability estimates, calculated as
the ICC, suggests that 49% of the total error variance in
SDLAT and 30% of the error variance in BLINKD was
explained by stable individual differences in the intercept after
removing the fixed effect of the KSS.
One possible problem with the interpretation of the random
effects in the present study is the relative lack of control over
the experimental manipulation. Subjects were allowed a
normal sleep at home before one condition and were working
the night shift before the other condition. While this approach
has several advantages in terms of external validity (it reflects
real-life conditions) it could have made subjects behave
differently so that part of the differences between subjects in
the overall level of the studied variables may be due to
differences in work content and behaviour (e.g. napping and
coffee consumption) before the driving sessions. However, the
main objective of the present study was to study the associ-
ation between sleepiness and the outcome variables (SDLAT/
BLINKD) and while a possible lack of experimental control
might have had an effect on both KSS and the outcomes, the
association between the KSS and the outcomes is less likely to
be confounded by differences in behaviour and work content
before the drive although it cannot be completely ruled out.
One possible confounder with respect to the association
could be caffeine intake. The effect of caffeine and/or hypnotics
on the association between subjective sleepiness and perform-
ance was studied by Johnson et al. (1990) in four groups given
hypnotics or placebo the night before and placebo or caffeine
in the morning. The correlations were negative in all groups
and weaker in the caffeine conditions but none was significant.
The results suggest that neither caffeine nor hypnotics at
bedtime had a significant effect on the association between
subjective sleepiness and performance. However, statistical
power was rather low because of the small group sizes
(n ¼ 10–40). Moreover, the results from the present study
indicate that the between-subjects design may be problematic.
The large random variance observed in the intercept for
driving performance (SDLAT) will affect the between-subjects
association. Small groups may show very different between-
subjects correlations even before treatment, just by chance. A
within-subject design that explicitly models subject-specific
effects is needed to adequately address this issue.
One major implication of individual differences like those
reported in the present study, is that they will make predictions
at the subject level suffer from systematic error if they are
made from the same parameters for all subjects. This is an
important issue in any application that needs precision at the
subject level, for example, driver fatigue warning systems. If
only the group (fixed effect) estimates were used in such a
system, driving performance (SDLAT) would be approxi-
mately predicted from the KSS for subjects that behave like
subject #9 but systematically underestimated for #4 and with
frequent false alarms for #8. Predictions for #6 should
probably have a steeper rise at high sleepiness levels than
predictions for #8. To achieve predictions at the subject level,
random effects have to be added to the fixed effects model, i.e.
a mixed effects model. This issue has also been discussed in
relation to predictions from bio-mathematical models of
fatigue and performance (Olofsen et al., 2004; Van Dongen,
Finally, the present study made use of an anova structure of
the fixed effect, treating the KSS as a nine-level factor variable.
This allowed the slope of the KSS to vary between all levels of
the scale. It would be possible to estimate a more restricted
model that describes the relation between the KSS and the
dependent variable as a function of a fewer number of linear
and/or non-linear parameters. However, any miss-specification
of the parameters in such models may inflate the estimates of
individual differences. Because exploring individual differences
was one of the main objectives of the present study, a less
restrictive and more explorative approach was chosen. The
results suggest, though, that the association between the KSS
and both dependent variables describes a curvilinear function.
Further research is needed to establish the exact shape of a
more parsimonious function of the association between the
KSS and SDLAT/BLINKD. The findings of the present study
suggest that such models should make use of a random
intercept and probably also a random coefficient of the slope
of the KSS to account for individual differences.
In conclusion, the present study has demonstrated a relation
between subjective sleepiness measured with the KSS and
driving performance as well as ?objective? indicators of
sleepiness in a highly controlled situation without disturbing
physical activities or other environmental changes. With
higher KSS levels, the SD of the lateral position increased
and eye blinks became longer in duration. Furthermore, this
relation was observed for all subjects in the present study.
However, large individual differences were also observed that
needs to be taken into account if predictions are to be made for
individual subjects. These individual differences were complex
and the mechanisms behind them are not yet understood.
Further research is needed that explicitly addresses these issues
with proper designs and statistical methods.
The results presented in this paper are based on the research
work being performed at the European co-funded, 6th FW,
Integrated project SENSATION (IST, 507231).
A˚kerstedt, T. and Gillberg, M. Subjective and objective sleepiness in
the active individual. Int. J. Neurosci., 1990, 52: 29–37.
52M. Ingre et al.
? 2006 European Sleep Research Society, J. Sleep Res., 15, 47–53
A˚kerstedt, T., Peters, T., Anund, A. and Kecklund, G. Impaired
alertness and performance while driving home from the night shift –
a driving simulator study. J. Sleep Res., 2005, 14: 17–20.
Arnedt, J. T., Wilde, G. J. S., Munt, P. W. and Maclean, A. W.
Simulated driving performance following prolonged wakefulness
and alcohol consumption: separate and combined contributions to
impairment. J. Sleep Res., 2000, 9: 233–241.
Arnedt, J. T., Wilde, G. J. S., Munt, P. W. and MacLean, A. W. How
do prolonged wakefulness and alcohol compare in the decrements
they produce on a simulated driving task? Accid. Anal. Prev., 2001,
Caffier, P. P., Erdmann, U. and Ullsperger, P. Experimental evaluation
of eye-blink parameters as a drowsiness measure. Eur. J. Appl.
Physiol., 2003, 89: 319–325.
Cajochen, C., Khalsa, S. B. S., Wyatt, J. K., Czeisler, C. A. and Dijk,
D.-J. EEG and ocular correlates of circadian melatonin phase and
human performance decrements during sleep loss. Am. J. Physiol.,
1999, 277: R640–R649.
Dinges, D. F. An overview of sleepiness and accidents. J. Sleep Res.,
1995, 4 (Suppl. 2): 4–14.
Dorrian, J., Lamond, N. and Dawson, D. The ability to self-monitor
performance when fatigued. J. Sleep Res., 2000, 9: 137–144.
Gillberg, M., Kecklund, G. and A˚kerstedt, T. Relations between
performance and subjective ratings of sleepiness during a night
awake. Sleep, 1994, 17: 236–241.
Hanowski, R. J., Wierwille, W. W. and Dingus, T. A. An on-road
study to investigate fatigue in local/short haul trucking. Accid. Anal.
Prev., 2003, 35: 153–160.
Horne, J. A. and Baulk, S. D. Awareness of sleepiness when driving.
Psychophysiology, 2004, 41: 161–165.
Johnson, L. C., Spinweber, C. L., Gomez, S. A. and Matteson, L. T.
Daytime sleepiness, performance, mood, nocturnal sleep: the effect
of benzodiazepine and caffeine on their relationship. Sleep, 1990, 13:
Mitler, M. M., Miller, J. C., Lipsitz, J. J., Walsh, J. K. and Wylie, C.
D. The sleep of long-haul truck drivers. N. Engl. J. Med., 1997, 337:
O’Hanlon, J. and Kelly, G. A psycho-physiological evaluation of devices
for preventing lane drift and run-off-road accidents. Technical Report
1736-F, Human Factors Research Inc., Santa Barbara Research
Park, Goleta, California, 1974.
O’Hanlon, J. F. Driving performance under the influence of drugs:
rationale for, and application of, a new test. Br. J. Clin. Pharmacol.,
1984, 18: 121S–129S.
O’Hanlon, J. F. and Kelley, G. R. Comparison of performance and
physiological changes between drivers who perform well and poorly
during prolonged vehicular operation. In: R. R. Mackie (Ed.)
Vigilance. Plenum Press, New York, 1977: 87–111.
Olofsen, E., Dinges, D. F. and Van Dongen, H. P. A. Nonlinear mixed
effects modeling: individualization and prediction. Aviat. Space
Environ. Med., 2004, 75: A134–A140.
Reyner, L. A. and Horne, J. A. Falling asleep whilst driving: are
drivers aware of prior sleepiness? Int. J. Legal Med., 1998, 111: 120–
Rogers, N. L. and Dinges, D. F. Subjective surrogates of performance
during night work. Sleep, 2003, 26: 790–791.
Rogers, N. L., Dorrian, J. and Dinges, D. F. Sleep, waking and
neurobehavioural performance. Front. Biosci., 2003, 8: s1056–s1067.
StataCorp. Stata Statistical Software: Release 8.0. StataCorp LP,
College Station, TX, 2003.
Stern, J. A., Walrath, L. C. and Goldstein, R. The endogenous
eyeblink. Psychophysiology, 1984, 21: 22–33.
Torsvall, L. and A˚kerstedt, T. Sleepiness on the job: continuously
measured EEG changes in train drivers. Electroencephalogr. Clin.
Neurophysiol., 1987, 66: 502–511.
Van Dongen, H. P., Maislin, G., Mullington, J. M. and Dinges, D. F.
The cumulative cost of additional wakefulness: dose–response effects
restriction and total sleep deprivation. Sleep, 2003, 26: 117–126.
Van Dongen, H. P. A. Comparison of mathematical model predictions
to experimental data of fatigue and performance. Aviat. Space
Environ. Med., 2004, 75: A15–A36.
Van Dongen, H. P. A., Baynard, M. D., Maislin, G. and Dinges, D. F.
Systematic interindividual differences in neurobehavioral impair-
ment from sleep loss: evidence of trait-like differential vulnerability.
Sleep, 2004, 27: 423–433.
Van Dongen, H. P. A., Vitellaro, K. M. and Dinges, D. F. Individual
differences in adult human sleep and wakefulness: leitmotif for a
research agenda. Sleep, 2005, 28: 479–496.
Yang, C.-M., Lin, F.-W. and Spielman, A. J. A standard procedure
enhances the correlation between subjective and objective measures
of sleepiness. Sleep, 2004, 27: 329–332.
Subjective sleepiness, driving performance and blink duration53
? 2006 European Sleep Research Society, J. Sleep Res., 15, 47–53