Regarding the causes of sleep-related
accidents, this study assesses whether
individuals can anticipate sleep onset
accurately and how individuals acknowledge and
use physiological and cognitive cues to make
judgments related to sleep onset.
A group of 41 partially sleep-deprived
subjects predicted the likelihood of sleep in 30
consecutive two-minute intervals and noted
physiological and cognitive signs of sleepiness,
including involuntary eye closure, head-nodding,
wandering thoughts, yawns, and instances of
sleep, collectively referred to as "sleep
complaints". Continuous polysomnographic
recording compared these predictions to actual
instances of sleep.
Subjects varied in their ability to predict
sleep onset. For all subjects, the mean
prediction of the likelihood of sleep prior to
sleep was significantly higher than the mean
prediction of the likelihood of sleep prior to
intervals in which no sleep occurred (78% vs.
42%; p<.05). However, subjects tended to
predict much lower likelihoods of sleep onset
before their first sleep event (55%) than before
later sleep events. On average, the rate at
which subjects reported miscellaneous sleep
complaints (such as head-nodding, eye closure,
and wandering thoughts) was higher prior to
sleep than prior to intervals in which sleep did
not occur.
Subjects who acknowledged a limited number
of physiological and cognitive indicators of
sleepiness tended to be poor predictors.
Subjects whose physiological and cognitive signs
of sleepiness failed to provide a strong
indication of whether or not sleep onset would
occur also tended to be poor predictors.
Inability to judge sleep onset and, hence,
susceptibility to sleep-related accidents, may
be attributable to both a scarcity of meaningful
warning signs and a failure to acknowledge the
importance of physiological and cognitive
indicators.
One of the most practical questions sleep
research can hope to answer is how and to what
extent sleep-related traffic accidents might be
avoided. Researchers estimate that sleep plays a
role in 1&endash;4% of all automobile accidents
in theUnited States, although as many as 19%of
accidents involving mortality or serious injury
may be attributed to acute sleepiness. Many
aspects of this problem have been investigated
in previous studies, and a link between
sleepiness and accidents is well established.
While previous research has focused on
performance degradation as a cause of
sleep-related accidents, many more recent
studies have investigated the relationship
between awareness of sleepiness and vehicular
accidents in both simulated [9,10] and
prospective designs. Although research has
investigated awareness of sleepiness in general,
no study to our knowledge has identified the
physiological and cognitive cues that constitute
this awareness of sleepiness or the way in which
these cues relate to accurately predicting sleep
onset. This study aims to identify such
physiological warning signs and investigate
their relationship to the prediction of sleep
onset.
It is easy to imagine a scenario in which a
prudent driver with adequate information about
the consequences of his or her actions can make
a decision to avoid an accident. What is it,
then, that makes the sleepy driver continue to
drive in the face of such risk? Several
explanations are possible. The driver may fail
to make the connection between his or her
physiological state and his or her dangerously
high risk of falling asleep. Previous research
into an alertness indicator, a device that might
detect and advise drivers of decreases in
alertness, is based on the assumption that lack
of attention to physiological and cognitive cues
indicating an elevated sleep tendency is the
cause of sleep-related accidents. An alternative
explanation is that the driver may fail to make
the connection between a microsleep and the
corresponding high risk of having an accident.
Both driving while fatigued and dozing while
driving are relatively commonplace occurrences.
According to the National Sleep Foundation's
2005 Sleep in America poll, 60% of drivers
admitted to driving while fatigued in the past
year, and 37% reported falling asleep at the
wheel. Because instances of driving drowsy and
falling asleep at the wheel far exceed the
number of sleep-related accidents, however,
sleepy drivers may not always consider the risk
of having an accident significant. Finally, the
driver may have adequate information about
sleepiness and the consequences of his or her
actions but remain highly motivated to continue
driving such that, measuring the risks of an
accident against the benefits of arriving at the
destination, he or she decides to continue
driving despite the risk. Some researchers
emphasize this point in criticism of an
alertness indicator, arguing that drivers may
continue to drive despite awareness of
sleepiness . Thus, awareness of sleepiness,
awareness of the consequences of sleepiness, and
motivation to continue in spite of sleepiness
are all factors that can contribute to the
safety of the sleepy driver. This study
concentrates on the first factor: to what extent
are drivers aware of their sleepiness, and what
are the mechanisms behind this awareness?
Studies suggest that it is nearly impossible to
fall asleep without any ''warning'' whatsoeve.
Stated differently, certain physiological and
cognitive cues invariably precede sleep,
including straining and closure of the eyes,
excessive yawning, wandering thoughts,
head-nodding, and struggling to fight sleep.
However, the question remains open as to whether
these and other physiological and psychological
cues provide sufficient information for an
individual to determine accurately when sleep
will occur.
In order to address these issues, we
partially deprived college student volunteers of
one night's sleep and had them complete a
one-hour computerized exercise the following
morning. This exercise involved making repeated
judgments about the likelihood of sleep onset
and the frequency of sleep complaints to
understand the relationship between attention to
physiological and cognitive signals and
awareness of sleep onset.
Ability of subjects to predict sleep onset
In general, subjects displayed some ability to
discern future sleep from no-sleep with their
predictions. However, subjects were frequently
surprised by their first sleep event;
predictions prior to the first sleep event were
much worse than predictions prior to subsequent
sleep events. There was a wide variation among
individual subjects' ability to predict sleep
onset, with a notable number of subjects
exhibiting a limited ability to predict
sleep.
As described earlier, all subjects were
asked to predict the likelihood of falling
asleep in the next two minutes on a scale from
0% to 100%; if subjects were perfect predictors
of sleep, they would always predict 100% prior
to sleep and 0% prior to no-sleep. For all
subjects, the mean prediction of sleep
likelihood prior to sleep was 78% ± 22%,
significantly different from the mean prediction
of sleep likelihood prior to no-sleep (42%
± 25%; p < .05). The mean prediction
prior to subjects' first sleep (55% ± 22%)
was not significantly different from mean
predictions prior to episodes of no-sleep (p
> .05). Higher predictions generally
corresponded to a higher frequency of sleep
occurrence.
The relationship between mean predictions
and sleep was also examined on a
subject-by-subject basis. We evaluated the
degree of difference between subjects'
predictions prior to periods in which sleep did
and did not occur. This difference can be taken
as a measure of subjects' ability to
discriminate sleep from no-sleep, with larger
differences representing better discrimination.
For example, a subject who always predicts a
100% likelihood of sleep before sleeping and a
0% likelihood before not sleeping is considered
maximally able to discern sleep likelihood.
Table 2 summarizes the results, listing subjects
by the difference between their mean predictions
prior to sleep and prior to no-sleep. For 19
subjects, the gap between their mean prediction
prior to sleep and prior to no-sleep was less
than 30%, indicating a relatively poor degree of
accuracy in predictions.
Relationship between sleep complaints and
sleep onset In general, the rates at which
subjects reported feeling signs of sleepiness
were significantly higher before sleep than
before no-sleep (p < .05). A notable
exception occurred in the frequency of yawns:
complaints of yawning were significantly
more frequent prior to intervals in which sleep
did not occur (p < .05), indicating that
yawning is a less accurate indicator of
imminent sleep. Subject-by-subject analyses
revealed that subjects differ widely in the
relationship between sleep complaints and
predictions of sleep onset. In particular, some
subjects showed very little difference in
complaint rates between sleeps and
no-sleeps.
Subjects clicked on computer icons during
the course of the study to record physiological
and cognitive indicators of sleepiness,
including sensations of the eyes, nods of the
head, episodes of sleep, yawns, wandering
thoughts, and any other indicators of sleepiness
they felt (''other'' complaints). For each
subject, each interval, and each complaint type,
a complaint rate was calculated, measuring the
number of complaints recorded per minute the
subject was awake. A total complaint rate for
each subject was calculated by summing the
individual complaint rates for each subject per
interval, thereby gathering a rate of all
complaints experienced per subject in any given
interval. To illustrate how complaint rates for
various physiological and cognitive indicators
relate to sleep onset, we compared complaint
rates prior to intervals in which sleep occurred
to those prior to intervals in which no sleep
occurred. We found that the mean total complaint
rate was significantly higher prior to sleep
(3.50 complaints/min) than prior to no-sleep
(1.82 complaints/min; p < .05). The rates at
which subjects selected the ''sleep,'' ''head,''
''eyes,'' and ''thought'' icons were
significantly higher before sleep than before
no-sleep (p < .05). There was no significant
difference, however, in the mean rate at which
subjects selected the ''other'' icon between
sleep and no-sleep intervals, and, as noted
above, yawning was not a good predictor
of sleep likelihood. Table 3 summarizes the
means and standard deviations of complaint rates
prior to sleep (S) and no-sleep (NS).
Fig. 4 illustrates the relationship between
total complaint rate and subsequent sleep across
all subjects and intervals, with the upward
trend in the data indicating that higher
complaint rates corresponded to a higher
frequency of sleep occurrence 4. Discussion
The results of this study establish that
people do have a limited ability to predict the
onset of sleep. The wide variation in
predictions both before episodes of sleep and
before episodes of no-sleep establishes that
subjects certainly do fall asleep at times when
they think sleep is highly unlikely and fail to
fall asleep at times when they think sleep is
highly likely. The mean prediction before
subjects' first sleep was low, suggesting that
subjects in general judged sleep to be fairly
unlikely on first occurrence. Mean predictions
prior to future episodes of sleep, however, were
significantly higher than mean predictions prior
to future episodes of no-sleep. Stated
differently, subjects learned how to more
accurately predict sleep onset over the course
of the study. This improvement over time
suggests that individuals may be able to
increase their awareness of imminent sleep
onset, or ''learn'' to be better predictors,
through targeted education and self-awareness
focused on the internal cues that precede
sleep.
Since there is a significant chance that a
sleepy driver will not be alive to experience
his or her second, third, or fourth sleep
attack, predictions made prior to the first
sleep event are of particular importance in the
current study. There was a wide variation in
subjects' ability to predict their first sleep
onset. Predictions prior to the first sleep
event ranged from as high as 93.3% to as low as
4.9% on the prediction scale. The possibility
that subjects can predict a very low likelihood
of sleep and subsequently fall asleep easily
translates into a potential to make misjudgments
on the road during a prolonged state of
sleepiness. Indeed, subjects may have even
greater difficulty accurately predicting the
likelihood of sleep in real-world situations
such as driving. The current protocol required
subjects to focus and report on feelings of
sleepiness and sleep onset in a distraction-
free environment. Drivers, by contrast, face
greater distraction on the road and less impetus
to pay attention to physiological cues; these,
in turn, could hinder drivers' ability to
predict first sleep onset. There are many
reasons why the ability to predict sleep onset
may be limited. First, the information people
receive by way of physiological and cognitive
signs of sleepiness is imperfect. There is some
positive relationship between the sleep
complaints measured in the study (eyes, head,
thoughts, and prior sleep) and imminent sleep
onset (see Table 3), but this relationship is
imperfect in the sense that there are times,
particularly during extended periods in which an
individual is struggling to stay awake, when
sleep can occur suddenly and without clear
preemptive warning signs. In addition,
individual variation exists in the strength of
the relationship between indicators of
sleepiness and imminent sleep onset. Fig. 6
indicates that subjects whose sleep complaints
were not strongly correlated with sleep onset
tended to predict sleep onset rather poorly.
Poor predictions can thus be viewed as caused by
poor quality of information (i.e., physiological
and cognitive signs) available upon which to
base predictions. Finally, the information
available to sleepy individuals may be
misleading or misrepresentative of imminent
sleep onset. As illustrated in Table 3,
yawning was more frequent prior to
no-sleep intervals than to intervals in which
sleep occurred, and the experience of
miscellaneous sleep complaints did not differ
between sleep and no-sleep episodes. Thus,
physiological and cognitive cues may not always
accurately predict sleep onset, even when
acknowledged by the sleepy individual.
Subjects' ability to predict sleep onset may
also be limited because they may not adequately
consider the physiological and cognitive
indicators of sleepiness that they notice and
record. Stated differently, although individuals
have information that could provide a good
indication that sleep will occur, they may
ignore this information in judging the
likelihood of sleep onset. The results of this
study show that subjects who paid little heed to
physiological and cognitive indicators in making
their predictions tended to be poor predictors.
Inattention to the warning signs of sleep onset
is a particularly difficult problem to address
when conflated with the aforementioned problem
of inadequate or misleading warning signs.
However, it is our contention that both problems
need to be solved, and that the best solution
for the first is education. People need to
understand that their ability to judge sleep
onset is, in fact, limited, and that they may
not experience a drastic change in their
physiological state immediately prior to falling
to asleep. Some researchers have suggested that
the validity of an alertness indicator is
questionable because it fails to address this
problem of ignoring warning signs of sleepiness.
An alertness indicator is a worthy approach
given the low quality of information normally
available for sleep-related decision-making, but
it is only worthwhile to the extent that people
can understand their inability to judge sleep
onset accurately without it. Closely related to
the problem of limited information is the
problem of overconfidence regarding the ability
to anticipate sleep onset. Though this study
does not specifically investigate the phenomenon
of ability estimation, some attempts have been
made to fill in the gaps. After the study, a
questionnaire was administered to a separate
group of subjects participating in a different
experiment. The protocol of our current
experiment was explained to this separate group,
and participants were asked if they thought they
could accurately judge sleep onset in this
context. Most subjects indicated that they would
be able to predict above 90% before every sleep
and below 10% before every no-sleep. In our
study, no subjects were able to predict sleep
onset with anywhere near this degree of
accuracy. The fact that people expect to discern
sleep onset very accurately seems to make them
more willing to test the limits of their ability
to stay awake, even in risky situations such as
driving. As suggested above, a partial solution
to this problem is education. People should
understand that they often do not have adequate
information vis-a`-vis signs of sleepiness, or
that they often disregard signs of sleepiness,
and may not always be able to avoid
sleep-related accidents if driving while tired.
Even if people do have the ability to predict
the likelihood of sleep onset in a two-minute
period with great accuracy, it is still unclear
whether they would take appropriate action to
avoid sleep-related accidents. An individual
might continue to drive until sleep is imminent,
expecting a clear psychological warning sign
moments before this sleep onset occurs. However,
at least one study found that drivers do not
appear to have any last-second warning but that
''fatigue. . . develops insidiously, and even
those who know they are tired may actually lose
consciousness suddenly and without trying to
decelerate'' [20]. We are able to
control our sleep impulse to some extent, but in
some situations it can easily be mistaken as an
ability to control sleep in all environments; in
monotonous tasks, however, our study suggests
that the chance of falling asleep increases with
time, and our ability to sustain wakefulness
diminishes over this period.
The probability scale methodology in this
study provides a consistent short-term
subjective measure of anticipated sleep
occurrence. Although two-minute prediction
intervals were chosen as the upper limit of
sleep onset in the current paradigm, future
research might evaluate predictive success over
longer intervals (e.g., five minutes) to
evaluate various risks associated with longer
periods of sleepy driving. Future research
should examine the extent to which people can
improve their abilities to predict sleep onset
through practice and/or training.