SEARCH

SEARCH BY CITATION

Keywords:

  • blinks;
  • car;
  • electroencephalograph;
  • fatigue;
  • Karolinska Drowsiness Score;
  • Karolinska Sleepiness Scale;
  • lateral variability

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

A large number of accidents are due to the driver falling asleep at the wheel, but details of this link have not been studied on a real road. The purpose of the present study was to describe the development of sleepiness indicators, leading to the drive being terminated prematurely by the onboard expert driving instructor because of imminent danger. Eighteen individuals participated during a day drive and a night drive on a motorway (both 90 min). Eight drivers terminated (N) prematurely (after 43 min) because of sleep-related imminent danger [according to the driving instructor or their own judgement (two cases)]. The results showed very high sleepiness ratings (8.5 units on the Karolinska Sleepiness Scale) immediately before termination (<7 at a similar time interval for those 10 who completed the drive). Group N also showed significantly higher levels of sleep intrusions on the electroencephalography/electro-oculography (EEG/EOG) than those who completed the drive (group C). The sleep intrusions were increased in group N during the first 40 min of the night drive. During the day drive, sleep intrusions were increased significantly in group N. The night drive showed significant increases of all sleepiness indicators compared to the day drive, but also reduced speed and driving to the left in the lane. It was concluded that 44% of drivers during late-night driving became dangerously sleepy, and that this group showed higher perceived sleepiness and more sleep intrusions in the EEG/EOG.


Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Accidents on the roads are related closely to sleepiness (and alcohol) (Philip and Akerstedt, 2006; Sagberg, 1999). Most of the evidence, however, is based on retrospective self-reports of falling-asleep events, or on the timing of the drive, or on previous sleep loss; that is, through inference. Very little is known about the details of physiological, behavioural changes or changes in awareness leading up to a crash during real driving.

Simulator studies of night or morning driving (after a night awake) show that alpha/theta activity, slow eye movements or lateral variability are increased before driving off the road (Anund et al., 2008a; Horne and Reyner, 1996; Lal and Graig, 2002; Otmani et al., 2005). In these studies, sleepiness apparently becomes overpowering. However, it is not clear if one can generalize from simulators to real driving, and there are no studies of real driving and sleepiness leading up to crashes or other serious adverse events. The existing studies of real driving and sleepiness have shown that reported sleepiness increases strongly during night driving (Sagaspe et al., 2008; Sandberg et al., 2011), as does inadvertent line crossing (Philip et al., 2005; Sagaspe et al., 2008), electroencephalograph (EEG) alpha and theta activity and blink duration (Sandberg et al., 2011).

Studying sleepiness indicators leading up to an accident is, obviously, not feasible in a well-controlled study. Under conditions without control of the context the so-called 100-car study linked crashes to inattention/fatigue using video recordings of driver and road for a long period of time (Klauer et al., 2006). One might also conceive of using proxies such as, for example, the first signs of dangerous driving. The latter could be based on close monitoring of the drive and driver by an expert on driving safety. The present study is a first attempt to use such an approach to describe the physiological, self-reported and behavioural indicators of sleepiness leading up the driver having to terminate the drive because of signs of unsafe driving, as judged by an onboard expert on driving safety. To the best of our knowledge, these questions have not been addressed previously.

Methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Participants and design

The experiment took place in September 2010. Eighteen individuals participated. The participants were recruited through random selection from the Swedish register of vehicle owners living in the region of Linköping (in southern Sweden). The main inclusion criteria were: age, 30–60 years, no spectacles needed (because of use of an eye-tracking system; not reported here), good health, normal weight, not shift workers, not professional drivers. Ten men and eight women participated. The average age was 46.6 years [standard deviation (SD) = 8.8]. The subjects received an honorarium of SEK 3000, approximately $450.

The participants carried out two driving sessions on a motorway, one in the afternoon (15:30–19:30 hours) and one during the night (00:15–04:30 hours), with 7 h between the end of one drive and the start of the next one. The afternoon drive took place in full daylight, the night drive in darkness. The data used in the present study were collected for the purpose of fine-tuning the parameters of an eye-tracking system for drowsy driving detection. The intention was to drive during the late night, where some sleepiness could be expected. Extreme or dangerous sleepiness was not sought out.

The study was ethically approved by the regional ethical committee in Linköping, Sweden (#2010/153-31). Special permission to conduct driving on public roads between midnight and 05:00 hours for research purposes was given by the government (#N2007/5326/TR). The participants signed written informed consent forms.

The study was carried out within the project ‘Virtual Prototyping and Assessment by Simulation’ (ViP), which is a Center of Excellence at the Swedish National Road and Transport Research Institute (VTI), funded by the government research fund Vinnova and ViP industrial partners Volvo Cars AB and Smarteye AB.

Procedure

One to 2 weeks before the experiment the participants were mailed information, questionnaires and a sleep–wake diary to be completed during the 3 days preceding the day of the drives. The participants were instructed to sleep at least 7 h per night during the three nights prior to the study, to avoid alcohol 72 h before the study and not to use caffeine from 01:00 hours on the experimental day.

Two subjects participated on each experimental day. The first participant arrived at 14:00 hours and the second at 16:00 hours. On arrival the participants were given written and oral information about the test and were then asked to complete an informed consent form, a responsibility form and a short questionnaire. The drivers' licence was inspected and a breathalyzer test was made. The experimenter then applied electrodes for physiological measurements. After this, the experiment began. The early group drove between 15:30 hours and 17:00 hours and between 00:15 hours and 01:45 hours. The corresponding times for the late group was 17:45–19:15 hours and 02:45–04:15 hours.

During the drive the participants rated their sleepiness every 5 min, prompted by the test leader (using the single word: ‘KSS’), and the response was written into a log. Electroencephalography (EEG) and electro-oculography (EOG) were recorded continuously, as were various parameters from the vehicle.

The time between the sessions was spent at the laboratory, where the participants could read, watch TV and interact socially. The participants were served dinner after the first driving session and fruits and sandwiches during the night, but no caffeine-containing drinks.

The participants were instructed to drive as they would do in ‘real life’. They were not allowed to speak, listen to the radio or do anything else that would counteract their sleepiness. They were also required not to exceed speed limits and were told that they were allowed to stop driving if they felt safety may be compromised, or for any other reason. After the night drive the electrodes were removed and the participants were sent home in a taxi.

The car used in the experiment was a Volvo XC70, with an automatic gearbox. The road was the motorway E4 from Linköping (Sweden) to exit 128 and back, a distance of 2 × 79 km, which took approximately 90 min to drive. The posted speed limit was 110 km h−1 during the entire drive, except for a road section of 750 m in the city of Norrköping, where the posted speed limit was 90 km h−1.

During the drive, an experienced driving instructor (the test leader) sat in the right front seat. The car had dual controls and there was a small screen in front of the test leader showing the driver's face, to allow for a rapid response in case the driver fell asleep. The driver had the ultimate responsibility for safety. He/she was instructed to stop driving whenever any potential threat to safety was observed. It was also made explicit that monetary compensation was in no way dependent upon completion of the drive. The test leaders had been involved in several studies of real driving and were experts on recognizing unsafe driving, including such due to sleepiness.

The test leader provided extra safety, and was instructed to take control of the car and to terminate the drive if there were any signs of safety being compromised by continued driving. The latter included any kind of illegal or otherwise dangerous driving behaviour. Because this study was expected to induce sleepiness, particular attention was paid to indicators of potential risk due to sleepy driving. The potential risk was determined by the test leader, based on whether or not the driver showed signs of sleep intrusions (unusually long eye closures or loss of muscle tension in the neck (nodding), in combination with signs of losing control of the car. In addition, a request for a break led to termination of the drive, as that was interpreted as sign of not being able to continue driving. The driver was not informed of this outcome before the start to avoid the possible temptation of hiding sleepiness in order to complete the drive.

Measurements

Vehicle data, such as speed and lateral position, were logged with a frequency of 10 Hz from the Controller Area Network (CAN) of the car. The speed was measured in km h−1. The lateral position was defined as the distance between the left line and the middle of the car (in metres), where a lane-tracker camera was mounted. The car was 176 cm wide. The variability of lateral position (SDlat) was measured as the standard deviation of the lateral position (metre). A line crossing was defined to occur when the side of the car crossed the left or right lines. To identify inadvertent line crossings, all line crossings that were part of a lane-changing process were removed. Video recordings were made of the front and rear views from the vehicle. The driver's face and feet were also recorded. All data acquisition systems were connected to each other in order to facilitate synchronization of data.

Physiological data – EEG, EOG, electromyography (EMG) and electrocardiography (ECG) – were recorded by a Vitaport 3 (TEMEC Instrument BV, Kerkrade, the Netherlands). The horizontal and vertical (left and right) EOG was recorded on three channels (F1, C1, P1 versus reference), EMG on one channel, ECG on one channel and EEG on three channels. The sampling frequency was 256 Hz for the EEG, ECG and EMG, and 512 Hz for EOG. Silver cup electrodes were used for EEG signals and disposable self-adhesive electrodes were used for all other signals.

Blink duration was extracted from one of the vertical EOG channels by using an automatic blink detection algorithm (Jammes et al., 2008). The algorithm low pass filters the EOG data, calculates the derivative and searches for sequences where the derived signal exceeds a threshold and falls below another threshold within a short time-period. If the amplitude of the (original, low pass-filtered) EOG signal in such a sequence exceeds a subject-specific threshold, the sequence is assumed to be a blink. To reduce problems with concurrence of eye movements and blinks, blink duration was calculated at half the amplitude of the upswing and the downswing of each blink and defined as the time elapsed between the two. Blink duration data were log-transformed in order to obtain a normal distribution of data.

To determine intrusions of sleep in the waking EEG/EOG patterns it was decided to score the recordings visually for sleep-related patterns, using conventional criteria (Rechtschaffen and Kales, 1968), rather than using spectral analysis or filtering. Each 20-s epoch was divided into 10 units of 2 s, each scored with respect to whether alpha (8–12 Hz) or theta waves (4–8 Hz), or slow rolling eye movements occurred. Each epoch was assigned a Karolinska Drowsiness Score (KDS) (Gillberg et al., 1996) between 0 and 100%, based on the proportion of signs of physiological sleepiness. For example, an epoch, which includes three 2-s segments with physiological sleepiness, would be represented by the KDS value 30% (Gillberg et al., 1996). For the current analysis the maximum and mean KDS during a 5-min period were used. The KDS has been validated against performance and subjective data (Anund et al., 2008a).

To obtain measures of awareness of sleepiness, drivers rated their sleepiness on the Karolinska Sleepiness Scale (KSS) (Åkerstedt and Gillberg, 1990). The KSS ranges from 1 to 9, where 1 = very alert, 5 = neither sleepy nor alert, 7 = sleepy but no effort to remain awake and 9 = very sleepy, an effort to stay awake, fighting sleep. The scale was modified to also have labels on intermediate steps. Before the drive the participants gave information on sleep behaviour during the last 24 h (bedtime, time of rising, time to fall asleep, time spent awake during sleep period, among others). After each session, the participants completed a debriefing questionnaire, including questions of heavy eyelids, having fallen asleep, etc.

The set of questionnaires completed before the drive included background data, questions on health, the Karolinska Sleep Questionnaire (KSQ: Åkerstedt et al., 2008), the Epworth Sleepiness Scale (ESS) (Johns, 1991) and the Karolinska Sleep Diary (KSD: Åkerstedt et al., 1997).

Data analysis

The data were averaged across 5-min intervals. The focus was on analysing the difference between those who completed the drive (C) and those who did not (N). Most of the drivers in the N-group terminated driving after approximately 40 min, which was also approximately the mid-point of the drive.

The statistical analysis involved analyses of variance (anovas) with repeated measures for the first 40 min during the night. The factors in the analyses were duration of the drive and group. A repeated-measures anova was also computed for the full day drive, as well as for the first 40 min of the night and day drives together (to establish whether there was an effect of time of day). The results were corrected for sphericity using the Huynh–Feldt method. All analyses were carried out with stata version 12.1 (StataCorp LP, Collage Station, TX, USA). All tests used a significance level of α = 0.05. Multi-level mixed-effects logistic regression analysis with a subject-specific intercept was performed for KDS due to the skewness and zero-inflation on this variable (KDS maximum or mean value of >0 versus KDS = 0 as reference value).

To focus on the last 5 min for those who did not complete the drive, a comparison (t-test) was made with the 5-min period at 35–40 min for those who completed the drive. Line crossings were analysed for the last 10 min, as they were rather rare. Such tests were also used to compare the last 5 min of both groups, as well as to test differences with respect to a few debriefing questions, including previous sleep amounts and sleep quality, as well as some background variables. In order to understand the relations between the predictors, correlations were carried out.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

No drivers terminated the drive prematurely during day driving. During the night drive, eight of 18 participants terminated the drive prematurely after a mean of 44 min (SD = 7.2). In five cases the test leader terminated the drive because of sleep-related potential compromise of safety. In two cases the driver decided to stop for the same reason. In one case the driver requested a break (because of sleepiness), which led to the test leader terminating the drive.

Table 1 and Figs 1 and 2 show the results of the analysis of the night and day drives, respectively. The night drive showed significant results for time on task for all variables except KDSmax. KSS, blink duration and KDSmean increased. Line crossings showed a pattern of initial increase and subsequent decrease. Speed showed an irregular pattern; lane position moved leftwards after 10 min and remained left for 15 more min before returning to the middle of the road. With respect to group, both KDS values were significantly higher for the group that did not complete the drive (N). Because the number of line crossings also varied across time, a total sum across 40 min was tested for difference between groups. This yielded C = 3.8 ± 1.4 versus N = 6.6 ± 2.6 [not significant (NS)].

Table 1. Results (F- and P-values) from the repeated-measures analysis of variance for the first 80 min of the day drive and the first 40 min of the night drive, respectively
 TimeGroupTime × group
  1. KDS, Karolinska Drowsiness Score; KSS, Karolinska Sleepiness Scale; Lat pos, lateral position; SDlat, standard deviation of the lateral position. Time: time into the drive (5-min intervals); group: completers and non-completers. aP<0.05; bP<0.01; cP<0.001. Degrees of freedom (df) for time during the night drive = 7/112 (except for the last three variables, which had a df = 7/99), and of 15/238 for the day drive (except for the last three variables, which had a df 15/211). Exactly the same df values were obtained for time × group. The loss of dfs was due to technical problems. The df for group was 1/16. Significance levels after Huyhn–Felt correction for repeated measures.

Night drive
KSS38.3c1.62.2
KDSmin2.8a15.3b1.9
KDSmax1.814.9b0.5
Line crossing4.1b1.81.1
Blink duration (log)8.2c0.52.2
SDlat9.0c1.41.2
Speed4.7b0.10.6
Lateral pos20.7c2.22.0
Day drive
KSS12.4c0.70.9
KDSmin1.721.3c1.7
KDSmax2.123.0c1.6
Line crossings0.90.01.5
Blink duration (log)4.7c1.00.9
SDlat4.4b0.00.9
Speed4.4a0.00.7
Lat pos6.1c0.30.6
image

Figure 1. Mean ± standard error for physiological and self-reported variables in 5-min intervals. Continuous line: group that terminated prematurely (N); broken line: those who completed the drive.

Download figure to PowerPoint

image

Figure 2. Mean ± standard error for driving parameters in 5-min intervals. Continuous line: group that terminated prematurely (N); broken line: those who completed the drive.

Download figure to PowerPoint

As the time of termination in group N varied, the last 5 min of the drive (10 min for line crossings, since they were relatively rare) was also compared with the value during 35–40 min for the group that completed the drive. This time slot was chosen as it corresponded to the last 5-min period before group N started to terminate prematurely. Table 2 shows a significant difference for KSS, KDSmean and KDSmax. The level was increased for group N for all three variables. When the last 5 min for group C was used as a comparison, no t-tests reached significance.

Table 2. Results from t-tests of the last 5 min of the night drive (10 min for line crossings) for group N and 35–40 min for group C
PredictorC Mean ± SE n N Mean ± SE n t-values, P-values
  1. KDS, Karolinska Drowsiness Score; KSS, Karolinska Sleepiness Scale; Lat pos, lateral position; SDlat, standard deviation of the lateral position; SE, standard error. aP < 0.01; bP < 0.001. N: group that did not complete the drive; C: group that completed the drive. The loss of data was due to technical problems.

KSS units7.0 ± 0.3108.5 ± 0.383.38a
KDSmax%5 ± 1.71022.5 ± 3.784.67b
KDSmin%0.47 ± 0.6105.83 ± 1.583.90a
Blink duration (log)–2.17 ± 0.0610–2.14 ± 0.0680.35
Speed km h−1104 ± 1.38102 ± 1.771.14
Lat pos m1.81 ± 0.0581.73 ± 0.0460.13
SDlat m0.224 ± 0.01880.250 ± 0.03760.71
Line crossings, number0.03 ± 0.0390.22 ± 0.1171.82

For the day drive, KSS, blink duration, SDlat, speed and lateral position showed significant effects of time on task. All variables increased with time except for speed, which showed an irregular pattern, and for lateral position, which moved to the left after the first 10 min. Only the two KDS variables differed between groups, with higher values for group N. No interactions were significant.

To compare night and day driving, a similar analysis as shown in Table 1 was repeated, with day and night driving entered as condition in the analysis, and also including time (first 40 min of both day and night drive) in the analysis. Table 3 shows that the effect of time was significant for all variables except KDSmax and KDSmean. Group was significant only for the two KDS variables, with higher levels for the N group. The effect of condition (night versus day) was significant for all variables. During the night drive, KSS, both KDS variables, line crossings, blink duration and SDlat were significantly higher. For speed, the night value was lower and for lane position the night drive involved a significant move to the left in the lane. There was also a significant condition × time interaction for line crossings and speed. For group × condition the effects were significant for KDSmean (night levels were especially increased for group N) and SDlat. The three-way interactions were not significant.

Table 3. Results (F- and P-values) for the repeated-measures analysis of variance, day and night drive (40-min duration) for the group that completed the drive and those who did not
 GroupTimeConditionGroup*TimeCondition*TimeGroup*ConditionGTC
  1. SDlat, standard deviation of the lateral position; Lat pos, lateral position; GTC, Group*Time*Condition. a< 0.05; b< 0.01; c< 0.001. F: F-ratio; condition: day/night. Huyhn–Felt correction was used for repeated-measures. Degrees of freedom: 7/112 for time, 1/16 for group, 1/16 for condition and 7/112 for GT, 7/112 for CT (except for the last three variables, which had df 7/91), 1/16 for GC and 7/112 for GTC (except for the last three variables, which had df = 7/91).

KSS0.132.6c90.8c0.70.82.01.35
KDSmean22.9c2.47.4a1.51.95.0a1.5
KDSmax22.3c2.17.2a0.70.74.00.3
Line crossings1.14.2b7.9a1.63.2a2.60.8
Blink duration (log)0.811.7c30.9c2.41.40.00.8
SDlat0.19.8c8.1a1.10.711.2b0.5
Speed0.05.9b10.1b0.53.8a0.40.8
Lat pos1.422.8c98.6c1.11.40.31.0

Because the distribution of the KDS variables were skewed and zero-inflated, a logistic regression analysis was also carried out (KDS maximum or mean value of >0 versus KDS = 0 as reference value). This approach yielded identical results for the two measures. The results for the night drive showed a significant group effect, with an odds ratio (OR) of 8655 and confidence interval (CI): 14–5353171, < 0.01. Time was not significant, nor was the time × group interaction. For the day drive, group was also significant (OR = 28.7, CI: 4.0–207, < 0.001). Neither the effect of time nor the time × group interaction were significant. When only the first 40 min for day and night driving were analysed together, group was again significant (OR = 157, CI: 9–2695, < 0.0001). No other factors were significant, including the effect of condition (night versus day).

The two KDS variables were highly intercorrelated (= 0.99, < 0.001) during the night drive and KDSmean correlated with blink duration (= 0.55, < 0.05) and lane crossings (= 0.55, < 0.05), while KDSmax correlated = 0.56 (< 0.05) and = 0.53 with the same variables, respectively. The correlations between KSS and the KDS variables were = 0.43 (< 0.10) for the mean and = 0.47 for the maximum (< 0.05), while that with blink duration was = 0.42, < 0.10).

After the drive, a debriefing questionnaire was completed. This showed that the group that did not complete the drive (N) reported significantly more difficulties in keeping their eyes open (C = 1.9 ± 0.3 versus N = 3.1 ± 0.3, = 2.3; < 0.05, range 1–5 very much) and rated driving performance (5.2 ± 0.5 versus 3.6 ± 0.4, = 2.7, < 0.05, range 1–7 good). Three of eight drivers in group N admitted to falling asleep, versus one of 10 drivers group C who did so. No differences were seen during the day drive.

The questions about sleep on the night prior to the drive showed a significantly higher sleepiness on awakening for group N (C = 5.1 ± 0.1.6 versus N = 6.8 ± 1.1, < 0.05, range 1–9 very sleepy), but no difference for total sleep time (TST) (7.6 ± 0.3 versus 7.2 ± 0.3, NS), sleep quality (4.2 ± 0.5 versus 4.2 ± 0.3, NS) or time of awakening (C = 05:42 h ± 12 min versus N = 06:24 h ± 30 min, = 1.9, < 0.10). The relation between start group and termination group was analysed using chi-square analysis. In group C, eight of 10 drove in start group 1. In group N, two of eight did so (χ2 = 5.4, < 0.02).

The two groups did not differ on age (44 ± 3 versus 40 ± 5 years), body mass index (BMI) (25.0 ± 1.1 versus 26.8 ± 2.1), habitual sleep duration (7.5 ± 0.20 h versus 7.7 ± 0.20 h), habitual sleep quality index (2.12 ± 0.22 versus 1.71 ± 0.17, range 1–5 poor), habitual restorative sleep index (1.80 ± 0.11 versus 2.03 ± 0.24, range 1–5, poor), ESS (7.4 ± 1.2 versus 7.9 ± 1.7, 11 = excessive), years with driving licence (24.4 ± 2.5 versus 24.2 ± 3.3), kilometres driven the preceding year (1732 ± 350 versus 1985 ± 349) or need for sleep (7.0 ± 0.25 versus 7.2 ± 0.15 h). With respect to gender, there were four women and six males versus four women and four males (NS) in the two groups, respectively.

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

For eight of 18 participants, night driving led to a termination of driving because of sleep-related signs of potential risk in continued driving. No premature terminations occurred during the day drive. To the best of our knowledge, similar results have not been reported previously and the results were unexpected, as the purpose of the study was not to reach maximum sleepiness, only to sample normal sleepiness in the average driver during late-night driving. The finding that 44% of drivers in the present study had to be taken off the road (or decided to stop on their own accord) strongly supports register and interview studies that link accidents with sleepiness (Philip and Akerstedt, 2006; Sagberg, 1999). It should be emphasized, however, that the results were obtained under conditions which excluded common countermeasures, such as taking a break, intake of caffeine, social interaction and others. The proportion of cases taken off the road may therefore contain some exaggeration.

The test leaders' impression of sleepy driving was supported strongly by the drivers' sleepiness ratings. The KSS value was very high (8.5) during the last 5 min before being stopped. This is significantly more than the 6.9 units (at the 40-min point) among those who completed the drive and higher than the ratings at the end of our previous real road study (Sandberg et al., 2011), in which no one terminated the drive prematurely. The latter study ran for 45 min on a winding rural road and showed 6.2 units for an early group that ended at 01:45 hours and 7.6 units for the late group that ended at 03:45 hours. Possibly, the shorter drive and the more demanding road may have prevented premature termination in that study. The present final KSS value of 8.5 for group N is somewhat higher than the 8.1 units seen immediately before hitting a rumble-strip in a moving base simulator (Anund et al., 2008b). Another study of real driving at night showed 6.8 units at 04:00 hours after 1 h of driving (sleepiness at the end of the drive was not reported) and in which all completed the drive (Sagaspe et al., 2008). Taken together, the results indicate that the reported sleepiness is higher among those who terminate their drive prematurely than in those who complete the drive, as well as higher than in comparative studies that did not lead to premature termination. Indeed, comparable studies show sleepiness ratings somewhat similar to those of the group that completed the present study. The debriefing after the drive also showed awareness of poor driving performance and three of eight drivers (36%) admitted to have fallen asleep compared to one of 10 (10%) in group C.

Among the other results that distinguished those who did not complete the drive were the increased levels of KDSmean and KDSmax. These were also increased during the night drive compared to the day drive, similar to what was seen in the previous real road study (Sandberg et al., 2011) and in several simulator studies (Anund et al., 2008a; Horne and Baulk, 2004), although in the former it was a combined measure of alpha and theta activity. In the study by Anund et al. (2008a), maximum KDS levels were reached immediately before driving off the road in a simulator and hitting a rumble-strip. Furthermore, the KDS value in the present study was correlated significantly with KSS (and with blink duration), and systematic studies of similar measures show a strong relation to sleep loss (Åkerstedt and Gillberg, 1990; Cajochen et al., 1995). In contrast to the peak in subjective sleepiness before terminating, KDS measures did not show a peak. These parameters also lacked the gradual increase seen for the sleepiness ratings. Rather, they were increased during most of the drive. This was also seen in the previous real driving study (Sandberg et al., 2011) and in the previous simulator study (Anund et al., 2008a).

Blink duration did not differ between the groups but differed between conditions, as demonstrated in other studies on real roads (Sandberg et al., 2011) and in the simulator (Anund et al., 2008a). Also, the number of line crossings failed to differentiate between the groups but differed between conditions, as found in previous on-road studies (Philip et al., 2005; Sagaspe et al., 2008) and in a number of simulator studies (Anund et al., 2008a; Horne and Reyner, 1996; Lal and Graig, 2002; Otmani et al., 2005). The other well-established sleepiness indicator, SDlat, also failed to differ between the groups but differed between the conditions. Thus, the expected increase during night-driving occurred in all the sleepiness-related variables. Two variables were more difficult to interpret. One was speed, which was lower during night driving, and the other was lateral position, which was moved to the left compared with day driving. The reason for the latter two effects could be related to increased caution due to sleepiness, but could also be the effect of the darkness, making the driver avoid the right lane marker for safety.

The present study was not designed to identify candidate variables for drowsiness detection (Dinges and Mallis, 1998), but the pattern of development in the key variables up to the point of terminating prematurely does not seem to identify any sudden increase just before the critical event, other than possibly in subjective sleepiness ratings. A similar lack of a ‘final warning’ was observed in a simulator study (Anund et al., 2008a). Rather, the impression gained is that several indicators of sleepiness are increased for a long period of time before a critical event on a real road or in the simulator. Such a state of instability at high levels of sleepiness have been suggested by Doran et al. (2001). Nevertheless, there might exist other, more sensitive, variables that might provide such warnings.

Apart from the main purpose, the present study also sought to investigate whether the ability to sustain alertness during a night drive would be already signalled during the day drive. The anova of the day drive, and of the combined day and night drives, indicated clearly that those who terminated the night drive prematurely had significantly increased levels of KDSmax and KDSmean values, but no other variables differed. There are no similar studies to compare with and the results need confirmation, but it might be the case that vulnerability to sleepiness may present itself as increased alpha and theta activity, or slow eye movements already occurring during day driving. These observations need confirmation in further studies.

The reason for the increased vulnerability in group N is unclear, and the present study was not designed to provide an answer to such a question. However, the most obvious potential cause, sleep duration during the night before the drive, did not differ between the groups, nor did the time of awakening, and there was even a tendency towards later awakening in group N, which should have worked against increased sleepiness. However, sleepiness on awakening was still significantly higher for group N. The time of the drive differed significantly between the groups, such that subjects in group N were more frequent in the late-start group. Thus, later timing of the drive probably contributed to higher sleepiness. Such a contribution of driving nearer to the circadian trough was found in our previous study of sleepiness during real road driving (Sandberg et al., 2011).

Among other individual factors of possible interest, habitual sleep duration or quality, age, gender, BMI, need for sleep and ESS did not differ significantly between the groups. Possibly, a larger sample may have revealed differences. No genotyping took place, but this would seem to be an important next step.

An increase was also seen in time on task during day and night driving for KSS, blink duration, number of line crossings and SDlat. For the KDS variables the effect was significant only during night driving. The previous study (Sandberg et al., 2011) showed similar results. The effect of time on task has also been demonstrated with control for previous sleep for KSS ratings and line crossings (Sagaspe et al., 2008). Obviously, driving a car on a real road causes sleepiness indicators to increase both during day and night driving, and is a factor that needs to be taken into account when estimating risk.

An important issue in the present study concerns the criteria used by the test leaders for determining when danger may be at hand if no intervention is made. This judgement was, of necessity, based on the test leader's appreciation of risk. The use of criteria such as inability to keep one's eyes open, ‘nodding-off’ and loss of control of the vehicle would be involved in all judgements, but the relative importance of each is not possible to determine. Whether non-intervention would have led to a full-blown accident must remain speculation – the driver could, theoretically, have recovered.

In the present study, two derived EEG measures of sleepiness were used, based on the presence of sleep indicators in the EEG and slow eye movements in the EOG. What value is derived that might be optimal is not clear from the present study. The mean level per scoring epoch may be a reasonable summary, but it could be hypothesized that 5 min with 10 scoring epochs, each with a 10% value, would yield an average of 10%, whereas a single epoch of 100% would yield the same value. Logically, one would deem the latter to be more of a risk and to represent severe sleepiness. However, the present study could not differentiate between the two.

The present study had several limitations, the most important of which may be the modest size of the two groups. This may have led to an underestimation of effects. Thus, the results must be considered tentative. In addition, the naturalistic and non-random seletion of the groups constitutes a problem. However, the selection must, of necessity, be non-random in a study with the present type of purpose; that is, to identify the characteristics of those who stop driving prematurely because of risk related to sleepiness. It should also be emphasized that the time-of-day effect cannot be evaluated properly, as day drives always preceded night drives (for logistic purposes).

In conclusion, this study demonstrated that late-night driving (without the use of countermeasures) leads to premature termination because of dangerous sleepiness in 44% of the drivers, and that the latter are characterized by particularly high levels of reported sleepiness and sleep intrusions in the EEG/EOG. It is also suggested that these individuals may be identifiable during day driving through sleep intrusions in the EEG/EOG during driving.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

This study received support from Volvo AB and Smart Eyes AB. T. Akerstedt has received support from Astra Zeneca AB.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References