Naturalistic field study of the restart break in US commercial motor vehicle drivers: Truck driving, sleep, and fatigue

Accident Analysis and Prevention 93 (2016) 55–64

Contents lists available at ScienceDirect

Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap

Naturalistic field study of the restart break in US commercial motor vehicle drivers: Truck driving, sleep, and fatigue Amy R. Sparrow a , Daniel J. Mollicone b , Kevin Kan b , Rachel Bartels b , Brieann C. Satterfield a , Samantha M. Riedy a , Aaron Unice b , Hans P.A. Van Dongen a,∗ a b

Sleep and Performance Research Center and Elson S. Floyd College of Medicine, Washington State University, P.O. Box 1495, Spokane, WA 99224, USA Pulsar Informatics, Inc., 3401 Market Street, Suite 318, Philadelphia, PA 19104, USA

a r t i c l e

i n f o

Article history: Received 15 December 2015 Received in revised form 13 April 2016 Accepted 18 April 2016 Keywords: Circadian rhythm Drowsy driving Fatigue risk management Hours of service Recovery sleep Recycling

a b s t r a c t Commercial motor vehicle (CMV) drivers in the US may start a new duty cycle after taking a 34-h restart break. A restart break provides an opportunity for sleep recuperation to help prevent the build-up of fatigue across duty cycles. However, the effectiveness of a restart break may depend on its timing, and on how many nighttime opportunities for sleep it contains. For daytime drivers, a 34-h restart break automatically includes two nighttime periods. For nighttime drivers, who are arguably at increased risk of fatigue, a 34-h restart break contains only one nighttime period. To what extent this is relevant for fatigue depends in part on whether nighttime drivers revert back to a nighttime-oriented sleep schedule during the restart break. We conducted a naturalistic field study with 106 CMV drivers working their normal schedules and performing their normal duties. These drivers were studied during two duty cycles and during the intervening restart break. They provided a total of 1260 days of data and drove a total of 414,937 miles during the study. Their duty logs were used to identify the periods when they were on duty and when they were driving and to determine their duty cycles and restart breaks. Sleep/wake patterns were measured continuously by means of wrist actigraphy. Fatigue was assessed three times per day by means of a brief psychomotor vigilance test (PVT-B) and a subjective sleepiness scale. Data from a truck-based lane tracking and data acquisition system were used to compute lane deviation (variability in lateral lane position). Statistical analyses focused on 24-h patterns of duty, driving, sleep, PVT-B performance, subjective sleepiness, and lane deviation. Duty cycles preceded by a restart break containing only one nighttime period (defined as 01:00–05:00) were compared with duty cycles preceded by a restart break containing more than one nighttime period. During duty cycles preceded by a restart break with only one nighttime period, drivers showed more nighttime-oriented duty and driving patterns and more daytime-oriented sleep patterns than during duty cycles preceded by a restart break with more than one nighttime period. During duty cycles preceded by a restart break with only one nighttime period, drivers also experienced more lapses of attention on the PVT-B and increased lane deviation at night, and they reported greater subjective sleepiness. Importantly, drivers exhibited a predominantly nighttimeoriented sleep schedule during the restart break, regardless of whether the restart break contained only one or more than one nighttime period. Consistent with findings in laboratory-based studies of the restart break, the results of this naturalistic field study indicate that having at least two nighttime periods in the restart break provides greater opportunity for sleep recuperation and helps to mitigate fatigue. © 2016 Elsevier Ltd. All rights reserved.

1. Introduction

∗ Corresponding author at: Sleep and Performance Research Center,Washington State University Spokane, P.O. Box 1495, Spokane, WA 99210-1495, USA. E-mail addresses: [email protected] (A.R. Sparrow), [email protected] (D.J. Mollicone), [email protected] (K. Kan), [email protected] (R. Bartels), satterfi[email protected] (B.C. Satterfield), [email protected] (S.M. Riedy), [email protected] (A. Unice), [email protected] (H.P.A. Van Dongen). http://dx.doi.org/10.1016/j.aap.2016.04.019 0001-4575/© 2016 Elsevier Ltd. All rights reserved.

Hours-of-service (HOS) regulations for the trucking industry in the Unites States require commercial motor vehicle (CMV) drivers to take a 34-h restart break at the end of a duty cycle before they can begin another duty cycle. This “restart rule” takes effect when drivers accumulate 60 h on duty in a rolling 7-day period or 70 h on duty in a rolling 8-day period. The restart break provides an

56

A.R. Sparrow et al. / Accident Analysis and Prevention 93 (2016) 55–64

opportunity for sleep recuperation between duty cycles and may help to mitigate the build-up of fatigue. The 34-h restart rule does not take into account the start and end times of the prior duty schedule and, therefore, ignores the influence of circadian rhythmicity on sleep and on waking alertness (Monk, 1990; Satterfield and Van Dongen, 2013). Fig. 1 illustrates why this is relevant in the context of the HOS regulations. That is, a daytime driver (or any driver whose duty cycle ends in the evening hours) automatically has two nighttime periods for sleep recuperation in the restart break. However, a nighttime driver (or any driver whose duty cycle ends after 01:00 and before 19:00) only has one complete nighttime period for sleep recuperation in the restart break, unless the duration of the restart break is extended beyond the minimum 34 h. Here a nighttime period is defined by the Federal Motor Carrier Safety Administration (FMCSA) as the period from 01:00 until 05:00 (Department of Transportation, 2011). This definition is based on the time zone of the driver’s home terminal (e.g., for a driver whose home terminal is in the Eastern Time Zone, the nighttime period is defined as the period from 01:00 until 05:00 Eastern Time even if the driver is en route in another time zone). In laboratory research, a 34-h restart break was found to be sufficient to maintain optimal alertness from one simulated duty cycle to the next when duty periods were scheduled during the day, but insufficient to prevent the build-up of fatigue across duty cycles when duty periods were scheduled during the night (Van Dongen and Belenky, 2010; Van Dongen et al., 2011). It was also shown that when duty periods were scheduled during the night, extending the restart break to include a second nighttime period helped to mitigate the build-up of fatigue across duty cycles (Van Dongen et al., 2010). Based in part on these findings, the FMCSA implemented new HOS regulations for CMV drivers, effective February 27, 2012, with a compliance date of July 1, 2013 (Department of Transportation, 2011). The new regulations included a provision requiring CMV drivers to include at least two nighttime periods (01:00–05:00) in their restart break, potentially extending the duration of the restart break beyond 34 h.1 When the new regulations were promulgated, stakeholders raised concern about the practical effectiveness of the new restart provision. For example, the requirement to include two nighttime periods in the restart break was based on the implicit assumption that the nighttime periods would help to mitigate fatigue because drivers use these periods for sleep recuperation. However, it has been suggested that nighttime drivers may choose to maintain a daytime sleep schedule during their restart break, which would render the new requirement ineffective. The FMCSA suspended enforcement of the new restart provision and the regulations reverted back to the original restart provision, effective December 16, 2014, pending further research. Here we present the findings of the first naturalistic study in the field to systematically investigate the restart break in CMV drivers. Our study design resembles that of earlier naturalistic field studies in CMV drivers focused on other aspects of the HOS regulations in the US. Specifically, Hanowski et al. (2007) investigated the efficacy of a new rule in a 2003 revision of the HOS regulations, which required CMV drivers to extend their time off between duty periods from 8 h to 10 h. The investigators collected sleep data by means of wrist actigraphy in 73 CMV drivers for 7 consecutive days. Drivers obtained more sleep when time off was extended to 10 h, which was interpreted as supportive of the new HOS rule for time off duty. Similarly, Hanowski et al. (2009) investigated the safety implications of a separate rule in the 2003 revision of the HOS regulations,

1

The new HOS regulations also limited the use of a restart break to no more than once every 168 h. That aspect is irrelevant in the current field study and beyond the scope of this paper.

which allotted drivers an extension of driving time from 10 h to 11 h in a 14-h duty period. They collected driving data by means of a truck-based data acquisition system for approximately 12 weeks and assessed critical incident risk in 103 CMV drivers. It was found that extending driving time from 10 h to 11 h did not significantly increase the critical incident risk, which was interpreted as supportive of the new HOS rule for maximum driving time per duty period. Our naturalistic field study of the restart break in CMV drivers used similar methodology, and the primary objective of the study was to investigate the efficacy of the new HOS restart provision requiring CMV drivers to include at least two nighttime periods in their restart break. The study was conducted prior to the compliance date for the new restart provision of July 1, 2013, while the original restart provision was still in effect. The original restart provision required a 34-h restart break at the end of a duty cycle regardless of how many nighttime periods were included in the restart break. We measured sleep, psychomotor vigilance performance, self-reported sleepiness, and driving performance across two duty cycles and the intervening restart break. Data were collected from 106 drivers, covering a grand total of 30,241 field study hours (25,185 h during duty cycles and 5056 h during restart breaks), comprising 1260 duty days, capturing 414,937 mi (8049 h) of driving, and including 3169 assessments of psychomotor vigilance performance and self-reported sleepiness. 2. Methods 2.1. Participants US truck drivers utilizing the restart provision of the HOS regulations for CMV drivers were recruited for the study. Participating drivers were fit for duty by regulatory standards and had a valid commercial driver’s license. A total of N = 106 drivers completed the study. They were 100 men and 6 women, ranging in age from 24 to 69 years (mean ± SD: 45.4 ± 10.7 years). They reported to have up to 39 years of experience as a CMV driver (mean ± SD: 12.4 ± 8.7 years). Three drivers were owner-operators independently contracting with a carrier. The others were employed with one of three carriers. These drivers were with their current carrier for up to 25 years (mean ± SD: 6.3 ± 6.4 years). The sample consisted of 44 local drivers, 26 regional drivers, and 36 over-the-road (long-distance) drivers. The study was approved by the Institutional Review Board of Washington State University. All drivers gave written, informed consent. Drivers were compensated for their study participation. They were informed that their study participation would not affect their employment or their relationship with their carrier and their data would be kept strictly confidential. Data were protected from disclosure by means of a Certificate of Confidentiality issued for this study by the National Institutes of Health. 2.2. Procedures Drivers’ duty and driving schedules were governed by the HOS regulations for CMV drivers in the US, which include the following provisions: • Duty period: Drivers may drive 11 h within a 14-h window after coming on duty2 following 10 consecutive hours off duty;

2 In the HOS regulations for CMV drivers, on-duty time is defined as all time from the moment a driver begins to work or is required to be in readiness to work until the moment the driver is relieved from work and all responsibility for performing work.

A.R. Sparrow et al. / Accident Analysis and Prevention 93 (2016) 55–64

57

Fig. 1. Illustration of a 34-h restart break in a daytime driver (top) and a nighttime driver (bottom) relative to the timing of the nighttime period. Note that the daytime driver has two nighttime periods in the 34-h restart break, whereas the nighttime driver only has one (unless the duration of the restart break were to be extended).

• Duty cycle: Drivers may drive with no more than 60 h on duty in the most recent 7 days and no more than 70 h on duty in the most recent 8 days; • Restart break: Drivers may restart a duty cycle after taking a restart break of 34 or more consecutive hours off duty (i.e., the 34-h restart rule). For study purposes, any off-duty period of 34 or more consecutive hours was considered to be a restart period. A duty cycle was operationally defined as any period of on-duty time between restart periods. The study compared sleep and fatigue between duty cycles preceded by a restart break with only one nighttime period (01:00–05:00) versus duty cycles preceded by a restart break with more than one nighttime period. During the study, drivers performed their normal driving and duty tasks and managed their schedules and restart breaks naturalistically. Every driver contributed data from two duty cycles. Measurements of sleep and fatigue were taken during the first duty cycle, the subsequent restart break, and the second duty cycle; no measurements were taken during the restart break preceding the first duty cycle. The average duration of study participation, not counting the restart break preceding the first duty cycle, was 11.9 days (SD: 1.5 days). The average duration of the first duty cycle was 5.1 days (SD: 1.2 days), and the average duration of the second duty cycle was 4.8 days (SD: 1.0 days). The total time on duty per duty cycle was on average 54.9 h (SD: 7.1 h). The number of nighttime periods in the restart break preceding each duty cycle varied within and between drivers. Thus, the number of nighttime periods preceding a driver’s first duty cycle could be the same or different than the number of nighttime periods preceding a driver’s second duty cycle. Of the 106 drivers: • 20 drivers had a restart break with only one nighttime period preceding both of their duty cycles; • 5 drivers had a restart break with only one nighttime period preceding their first duty cycle and a restart break with more than one nighttime period preceding their second duty cycle; • 26 drivers had a restart break with more than one nighttime period preceding their first duty cycle and a restart break with only one nighttime period preceding their second duty cycle; • 55 drivers had a restart break with more than one nighttime period preceding both of their duty cycles. In total, 51 drivers (48.1% of the sample) had a restart break with only one nighttime period preceding at least one of their two duty cycles. The average duration of the restart break preceding the first duty cycle was 57.8 h (SD: 25.8 h), and the average duration of the restart break preceding the second duty cycle was 47.7 h (SD: 14.0 h). In the condition where the restart break contained more than one nighttime period, the average number of nighttime periods was 2.3 (SD: 0.4) both when the restart break preceded the first duty cycle and when it preceded the second duty cycle. In this condition, the restart

break contained exactly two nighttime periods in 73.8% of cases and more than two nighttime periods in the remaining 26.2% of cases. 2.3. Measurements Drivers were provided with a wrist actigraph (Actiwatch 2, Philips Respironics, Bend, OR, USA), which they were asked to wear continuously throughout the study to measure their sleep/wake patterns. The wrist actigraph recorded cumulative activity (movement) counts and average light exposure levels in 1-min intervals. Drivers were also provided with a smartphone (Galaxy S3 i9300, Samsung Electronics, South Korea) with custom-made application software (Pulsar Informatics, Philadelphia, PA, USA). The smartphone was set up with keypad and screen touch sounds turned off, display brightness at maximum, automatic screen rotation and power-saving mode turned off, and calling, texting, e-mail and internet access disabled. Drivers used the smartphone to keep a sleep/wake log throughout their study participation. The smartphone was also used for testing of psychomotor vigilance performance and self-reporting of sleepiness. The typical time commitment for these assessments was less than 30 min per study day. Psychomotor vigilance performance was tested by means of a brief (3-min) psychomotor vigilance test (PVT-B). This simple reaction time task with high stimulus density is a sensitive, validated assay of fatigue from sleep loss and circadian misalignment (Basner et al., 2011). From each PVT-B, the number of lapses of attention (defined as reaction times greater than 355 ms) was extracted. Following each PVT-B assessment, drivers reported their subjective sleepiness on a smartphone implementation of the Karolinska Sleepiness Scale (KSS; Åkerstedt et al., 2014). Sleepiness scores ranged from 1 (extremely alert) to 9 (extremely sleepy). Anchor words were provided for scores 1, 3, 5, 7 and 9: “extremely alert,” “alert,” “neither alert nor sleepy,” “sleepy, but no effort to stay awake,” and “extremely sleepy—fighting sleep,” respectively. During duty cycles, drivers completed the PVT-B and KSS three times in each of their duty periods: once immediately before starting the duty period, once during an off-duty break about halfway through the duty period, and once immediately after ending the duty period (testing was blocked when the truck was in motion as detected through the Global Positioning System (GPS) receiver in the smartphone). During the restart break, drivers completed the PVT-B and KSS three times per day: once within 2 h after awakening, once in the middle of the day, and once within 2 h before going to bed. At the completion of each PVT-B assessment, the smartphone prompted the driver to indicate whether there had been any distractions during testing, and a text box allowed for the driver to enter any comments. Prior to the first duty cycle, participating drivers met with a member of the research team to have the study procedures and the operation of the smartphone explained to them. Drivers also practiced the PVT-B under supervision, and they were instructed to give their best effort on the test throughout the study. During the study, the smartphone data were transmitted to a secure computer server via cellular network and reviewed daily

58

A.R. Sparrow et al. / Accident Analysis and Prevention 93 (2016) 55–64

by the research team. Drivers maintained daily telephone contact with the research team to review compliance with the smartphone assessments and to discuss any issues relating to distractions during PVT-B testing or other unusual events that might have impacted their schedules or data collection. Any apparent outliers in the PVTB data and/or KSS scores, as well as any PVT-B assessments with more than a few false starts, were also discussed. PVT-B and KSS assessments for which the driver had indicated potential confounds on the smartphone or were flagged as such during the daily telephone conversations were discarded. The number of PVT-B and KSS assessments contributed to the data set by each driver was on average 29.9 (SD: 6.2). Note that PVT-B and KSS data collected during the restart break were not of primary interest, and are not reported here. At the end of the second duty cycle, when study participation ended, each driver met with a member of the research team once more. Feedback on the study and on any potentially relevant issues was solicited. Wrist actigraph data were downloaded and the sleep/wake log was reviewed with the driver to resolve any discrepancies between the actigraph record and the sleep/wake log kept on the smartphone. Periods of immobility in the actigraph record were marked as sleep after checking them against the sleep/wake log and against records of any time periods when the actigraph was not worn. Periods of sleeping were extracted with 1-min resolution. Drivers’ official duty logs for the period of the study were downloaded from their carriers’ duty log databases. From each driver’s duty log, on-duty status and driving status were extracted in 1min intervals. The beginning and end dates/times of the two duty cycles and the restart periods preceding them were also assessed. For proper alignment of data sets, all data were expressed in terms of each driver’s home terminal time zone. 2.4. Truck data For the duration of the study, participating drivers were assigned a study truck of the type they were driving routinely – either a Freightliner Cascadia (82 drivers) or an International ProStar (24 drivers). This study truck was equipped with a data acquisition system (Pulsar Informatics, Philadelphia, PA, USA), which made continuous, passive recordings while the truck was in use (i.e., when the ignition switch was activated). The data acquisition system recorded distance traveled, speed, acceleration, lateral lane position relative to left and right lane markers, steering wheel angle, headway distance, fuel use, and a range of other truck-based parameters and driving metrics. These data were taken from the truck’s SAE J1939 network through the controller area network (CAN) bus; from sensors measuring steering wheel angle, lateral and longitudinal acceleration, and yaw rate; from a global positioning system (GPS) device; and from a lane tracking system. The data acquisition system was imperceptible to drivers with the exception of the lane tracking system camera, which was mounted on the windshield facing forward in order to capture images of the road. The data acquisition system did not generate any feedback or alerts. Recorded data were encrypted and transmitted to a secure computer server via cellular network. Speed was sampled at 10 Hz and lateral lane position at approximately 13 Hz. The average and SD of speed were extracted in 1-min intervals. Lane deviation, computed as the SD of the distance of the center of the truck to the left lane marker averaged with the distance of the center of the truck to the right lane marker, was also extracted in 1-min intervals. The lane deviation data were filtered to capture stretches of unhindered driving on straight segments of highway, where lane deviation results are not confounded by road curvature or interaction with (e.g., braking for) other road users. If the sum of the distance to the left and right lane markers varied with a SD of 10 cm

or more within a given 1-min interval, this indicated that the lane was broadening, narrowing, shifting, or curving, or the driver was changing lanes. The lane deviation data for such 1-min intervals were not retained. Further, if the average speed was not between 45 and 65 mph or the SD of speed was greater than 1 mph within a given 1-min interval, this indicated that the road was not a highway, there was heavy traffic, or the truck was accelerating or braking. The lane deviation data for such 1-min intervals were also not retained. The lane tracking system provided a 4-point measure reflecting the degree of confidence in detecting the lane markers. Lane position samples lacking the highest degree of confidence were discarded, and 1-min intervals for which the number of discarded samples was 50% or more were not retained. Collectively, these data reduction steps retained a total of 235,575 usable 1-min records of lane deviation. For one driver, the lane tracking system failed, so the effective sample size for lane deviation analysis was 105 drivers. 2.5. Data analysis For statistical analysis, two naturally occurring study conditions were distinguished: • One night restart (1NR) condition: Restart breaks with only one complete nighttime period (01:00–05:00); • More than one night restart (MT1NR) condition: Restart breaks with more than one complete nighttime period (01:00–05:00). We investigated differences in patterns of duty and driving, sleep, and fatigue during duty cycles preceded by restart breaks corresponding to each of these two conditions. Each driver contributed data from two duty cycles, and since this was a naturalistic study, the condition associated with the first duty cycle could be the same or different than the condition associated with the second duty cycle. Drivers were therefore compared to themselves and each other in a mixed within- and between-subjects statistical design – with condition (1NR versus MT1NR) as the independent variable of primary interest. We also investigated differences in sleep patterns during restart breaks of each of the two conditions. Each driver contributed sleep/wake data from only one restart break. Thus, for this analysis, drivers were compared to each other in a between-subjects statistical design – with condition (1NR versus MT1NR) again as the independent variable of primary interest. A straightforward repeated-measures statistical design was not suitable for this study because of the variation among drivers in the duration of the duty cycles and restart breaks and in the timing of their PVT-B and KSS assessments. However, laboratory research focused on the restart break had previously revealed the importance of investigating systematic differences in 24-h patterns (Van Dongen et al., 2011). Therefore, the statistical approach for the study was based on comparison of 24-h patterns of outcome measures – between subjects as well as within subjects between duty cycles. To implement this statistical approach, the data for duty status, driving status and sleep/wake status collected at 1-min intervals were first averaged over 1-h bins, for each hour of the day, across days, in each duty cycle of each driver. The resulting 24 hourly bins for each of the 2 duty cycles of each of the 106 drivers (i.e., 24 × 2 × 106 = 5088 data points) were then analyzed with mixedeffects analysis of variance (ANOVA; Van Dongen et al., 2004). The statistical design had fixed effects for condition (1NR or MT1NR), duty cycle (1 or 2), hour of the day (0–23), and their two- and threeway interactions; and a random effect on the intercept over the 106 drivers. For each driver and each duty cycle, the data in each 1-h bin were weighted by the number of contributing 1-min intervals.

A.R. Sparrow et al. / Accident Analysis and Prevention 93 (2016) 55–64

The same procedure was implemented for the lane deviation data, except that the root mean square was used to combine the 1min intervals into 1-h bins (so that the variance rather than the SD of lateral lane position was averaged). Otherwise, the same mixed-effects ANOVA was performed. The same procedure was also implemented for the PVT-B and KSS assessments, except that these were performed less frequently than the other measurements discussed thus far, and at discrete time points rather than continuously. The PVT-B and KSS data were therefore aggregated (not averaged) into 4-h bins spanning the 24 h of the day, and collapsed (not averaged) across days, in each duty cycle for each driver. The same mixed-effects ANOVA was performed again, but this time without any need for weighting the data. Essentially the same methodology was applied to the sleep/wake data collected during the restart break between the two duty cycles. The sleep/wake data collected at 1-min intervals were first averaged over 1-h bins, for each hour of the day, across days, in the restart break of each driver. The resulting 24 hourly bins of each of the 106 drivers (i.e., 24 × 106 = 2544 data points) were analyzed with mixed-effects ANOVA, with fixed effects for condition, hour of the day, and their interaction; and a random effect on the intercept over the 106 drivers. For each driver, the data in each 1-h bin were weighted by the number of contributing 1-min intervals. The key statistical outcomes of these analyses are the main effects of condition and hour of the day and their interaction. Accordingly, the graphs in the figures show population marginal means over drivers and duty days (or restart days) as a function of hour of the day for each condition, with error bars indicating ±1 SE, as derived from the mixed-effects ANOVAs. Summary statistics are indicated as grand mean ± SE.

59

Fig. 2. 24-h patterns of drivers logged as on duty (top) and as driving (bottom), expressed as a percentage of total driver hours for each hour of the day during duty cycles. Black: 1NR condition. Gray: MT1NR condition.

3. Results 3.1. Duty and driving The duration of duty cycles preceded by the 1NR condition was 5.1 ± 0.1 days, and the duration of duty cycles preceded by the MT1NR condition was 4.9 ± 0.1 days. Drivers were on duty 11.2 ± 0.3 h per 24 h during duty cycles preceded by the 1NR condition, and 11.1 ± 0.2 h per 24 h during duty cycles preceded by the MT1NR condition. There was a significant main effect of hour of the day (F23,4887 = 5.78, P < 0.001), and a significant interaction of condition by hour of the day (F23,4887 = 10.66, P < 0.001). Fig. 2 (top) shows the 24-h patterns of on-duty status, as a percentage of total driver hours in the data set for every hour of the day, in each of the two conditions. As an example to help interpret the graph (Fig. 2, top), the black curve pertains to the 1NR condition, when the restart break preceding the duty cycle contained only one nighttime period (01:00–05:00). In the graph, the first shown value of the curve for that condition, in the 00:00–01:00 bin, is 53.4%. This percentage means that on any duty day in any duty cycle preceded by the 1NR condition, the chance of finding any of the 106 drivers in the study logged as on duty for any given minute between midnight and 01:00 was 53.4%. Thus, the propensity to be on duty between midnight and 01:00 during a duty cycle preceded by a restart break with only one nighttime period was 53.4% (and the propensity to be off duty between midnight and 01:00 in that condition was the remaining 46.6%). Drivers were driving 8.5 ± 0.2 h per 24 h during duty cycles preceded by the 1NR condition, and 7.9 ± 0.2 h per 24 h during duty cycles preceded by the MT1NR condition. There was a significant main effect of condition (F1,4887 = 4.09, P = 0.044), a significant main effect of hour of the day (F23,4887 = 5.06, P < 0.001), and a significant interaction of condition by hour of the day (F23,4887 = 12.55,

P < 0.001). Fig. 2 (bottom) shows the 24-h patterns of time logged as driving, as a percentage of total driver hours in the data set for every hour of the day, in each of the two conditions. These results show that time spent on duty and time spent driving were distributed more or less evenly across the hours of the day during duty cycles preceded by the MT1NR condition, whereas they were distinctly more oriented toward the night in duty cycles preceded by the 1NR condition. In other words, a driver was likely to be driving primarily at night during a duty cycle following a restart break with only one nighttime period (01:00–05:00). 3.2. Sleep Drivers were sleeping 6.0 ± 0.2 h per 24 h during duty cycles preceded by the 1NR condition, and 6.2 ± 0.1 h per 24 h during duty cycles preceded by the MT1NR condition, as measured by means of wrist actigraphy. This near-equivalence between the two conditions must be considered in the context of a significant main effect of hour of the day (F23,4887 = 6.51, P < 0.001), and in particular a significant interaction of condition by hour of the day (F23,4887 = 11.09, P < 0.001). Fig. 3 (left) shows the 24-h patterns of sleep during duty cycles, as a percentage of total driver hours in the data set for every hour of the day, in each of the two conditions. During the restart break between the two duty cycles of the study, drivers were sleeping 8.8 ± 0.3 h per 24 h in the 1NR condition, and 8.9 ± 0.2 h per 24 h in the MT1NR condition. There was a significant main effect of hour of the day (F23,2392 = 45.51, P < 0.001), and a significant interaction of condition by hour of the day (F23,2392 = 2.45, P < 0.001). Fig. 3 (right) shows the 24-h patterns of sleep during the restart break, as a percentage of total driver hours in the data set for every hour of the day, in each of the two conditions.

60

A.R. Sparrow et al. / Accident Analysis and Prevention 93 (2016) 55–64

Fig. 3. 24-h patterns of sleep, expressed as a percentage of total driver hours for each hour of the day, during duty cycles (left) and during the intervening restart break (right). Black: 1NR condition. Gray: MT1NR condition.

These results show that sleep occurred primarily during the night during duty cycles preceded by the MT1NR condition, whereas it occurred primarily during the day during duty cycles preceded by the 1NR condition. While this difference is just as striking as that for duty and driving (see Fig. 2), the 24-h patterns of sleep in the two conditions were not simply the reverse of the 24-h patterns of duty and driving. In both conditions, the propensity for sleep during duty cycles was relatively increased during the night and reduced during the early evening hours. The 24-h patterns of sleep were considerably different during the restart break than during the duty cycles (Fig. 3). Notwithstanding relatively small differences between the 1NR and MT1NR conditions, sleep during the restart break occurred predominantly at night, and almost no sleep was obtained in the early evening hours, in both conditions. 3.3. Fatigue outcomes Overall, drivers exhibited 2.0 ± 0.3 lapses of attention per PVTB assessment during duty cycles preceded by the 1NR condition, and 1.7 ± 0.3 lapses of attention per PVT-B assessment during duty cycles preceded by the MT1NR condition. There was a significant main effect of condition (F1,2526 = 5.93, P = 0.015), and a significant main effect of time of day (F5,2526 = 2.99, P = 0.011). Fig. 4 (top) shows the 24-h patterns of lapses of attention in the PVT-B during duty cycles, over consecutive 4-h periods of the day, in the two conditions. Compared to the MT1NR condition, the 1NR condition was associated with greater fatigue, as measured objectively with the PVT-B, throughout the 24 h of the day except the late evening hours. The difference between the two conditions was greatest – up to 0.8 lapses of attention more (i.e., 51.5% more) in the 1NR condition – at night, which was also when most of the driving occurred in the 1NR condition (see Fig. 2, bottom).3 On the KSS, drivers rated their sleepiness as 3.3 ± 0.1 during duty cycles preceded by the 1NR condition, and as 3.1 ± 0.1 during duty cycles preceded by the MT1NR condition. There was a significant main effect of condition (F1,2489 = 4.63, P = 0.032), a significant main effect of time of day (F5,2489 = 15.94, P < 0.001), and a

3 The analysis of PVT-B lapses was repeated with inclusion of duty cycles preceded by a restart with either one or two, but not more than two, nighttime periods. The difference between conditions remained, and continued to be greatest at night, but emerged as a significant interaction of condition by time of day (F5,2090 = 2.49, P = 0.030).

Fig. 4. 24-h patterns of lapses of attention on the 3-min PVT-B (top) and subjective sleepiness scores on the KSS (bottom), by 4-h period of the day, during duty cycles. Black: 1NR condition. Gray: MT1NR condition.

significant interaction of condition by time of day (F5,2489 = 21.18, P < 0.001). Fig. 4 (bottom) shows the 24-h patterns of KSS scores during duty cycles, over consecutive 4-h periods of the day, in the two conditions. Compared to the MT1NR condition, the 1NR condition was associated with less subjective sleepiness in the evening hours, i.e., near the beginning of driving activity typical for that condition (see Fig. 2, bottom). However, the 1NR condition was associated with greater subjective sleepiness toward the late morning hours, i.e., toward the end of driving activity typical for that condition. 3.4. Driving performance Driving performance was measured in terms of lane deviation during the two duty cycles of the study. Overall, lane deviation was 18.8 ± 0.4 cm during duty cycles preceded by the 1NR condition, and 18.7 ± 0.4 cm during duty cycles preceded by the MT1NR condition. There was a significant main effect of time of day (F1,2706 = 11.05, P < 0.001), and a significant interaction of condition by time of day (F23,2706 = 2.17, P = 0.001). Fig. 5 shows the 24-h patterns of lane deviation, over the hours of the day, in the two conditions. Post-hoc contrasts revealed that lane deviation in the 1NR condition exceeded lane deviation in the MT1NR condition in the 04:00–05:00, 08:00–09:00, and 15:00–16:00 bins, and the opposite occurred in the 19:00–20:00 bin. The 24-h patterns of lane deviation in the two conditions were consistent with the 24-h patterns observed for the PVT-B (see Fig. 4, top). However, the number of nighttime periods in the restart break had a comparatively minor effect on lane deviation during the subsequent duty cycle. Lane deviation was between 1.2 cm (7%) greater

A.R. Sparrow et al. / Accident Analysis and Prevention 93 (2016) 55–64

Fig. 5. 24-h patterns of lane deviation during duty cycles. Black: 1NR condition. Gray: MT1NR condition.

and 1.1 cm (6%) less during duty cycles preceded by the 1NR condition than during duty cycles preceded by the MT1NR condition. 4. Discussion 4.1. Duty, driving and sleep In this naturalistic field study, we measured sleep and fatigue through two consecutive duty cycles and the intervening restart break in a sample of 106 CMV drivers. During the study, drivers managed their duty and driving schedules and performed their tasks as they would normally have done; there was no experimental intervention to change their schedules or their behavior. In each of the two duty cycles measured during the study, a driver could be in one of two naturally occurring conditions: the one night restart (1NR) condition, in which the duty cycle was preceded by a restart break with only one complete nighttime period (defined as 01:00–05:00); or the more than one night restart (MT1NR) condition, in which the restart break was preceded by a restart break with more than one complete nighttime period. The study compared 24-h patterns of duty, driving, sleep, and fatigue between these two conditions (see Figs. 2–5). The results are summarized in Table 1. As shown in Fig. 2, drivers’ duty logs revealed that their time spent on duty and driving was distributed more or less evenly across the hours of the day when the preceding restart break contained more than one nighttime period, whereas their time spent on duty and driving was distinctly more oriented toward nighttime hours when the preceding restart break contained only one nighttime period. In other words, following a restart break of the 1NR condition (which was more likely to end in the afternoon or evening), a driver was likely to be driving primarily at night. Comparable to results from an earlier naturalistic driving study (Hanowski et al., 2007), sleep during duty cycles averaged 6.0 h per 24 h in the 1NR condition and 6.2 h per 24 h in the MT1NR condition, as measured with continuous wrist actigraphy. Fig. 3 (left) shows that sleep occurred primarily during the daytime hours in duty cycles preceded by the 1NR condition. Not surprisingly, the timing of duty and driving hours (Fig. 2) was a major determinant of the timing of sleep during duty cycles. Thus, in the 1NR condition, drivers were likely to be driving at night and sleeping during the day. Even so, the 24-h patterns of sleep were not simply the reverse of the 24-h patterns of duty and driving, regardless of study condition (cf. Figs. 2 and 3, left). The neurobiology of sleep regulation

61

produces a “gate for sleep” during the night and a “forbidden zone for sleep” or “wake maintenance zone” during the early evening (Lavie, 1986; Strogatz et al., 1987). Accordingly, in both conditions, the propensity for sleep was relatively increased during the night and reduced during the early evening hours. Drivers’ 24-h sleep patterns were substantially different during the restart break than during duty cycles (see Fig. 3), in agreement with what has been reported by others (Pylkkönen et al., 2015). Sleep during the restart break averaged 8.8 h per 24 h in the 1NR condition and 8.9 h per 24 h in the MT1NR condition. Thus, drivers obtained substantively more sleep per 24 h during the restart break than during duty cycles, suggesting that during the restart break they needed to recover from prior sleep insufficiency. The overall amount of restricted sleep per 24 h during duty cycles was not much different between the two conditions, nor was the overall amount of recovery sleep per 24 h during the restart break. However, in the MT1NR condition, as contrasted with the 1NR condition, drivers had more cumulative opportunity during the restart break to obtain recovery sleep before they began another duty cycle. Importantly, in both the 1NR condition and the MT1NR condition, drivers predominantly slept at night during the restart break (Fig. 3, right). This finding refutes the idea that nighttime drivers generally maintain a schedule with nighttime wakefulness and daytime sleep during their restart breaks, whether out of habit, preference, or otherwise. Instead, nighttime drivers tend to transition back to a daytime wakefulness and nighttime sleep schedule when unconstrained by duty demands, consistent with the strong neurobiological drives that regulate sleep (Borbély, 1982; Daan et al., 1984). This finding corroborates the premise that the number of nighttime periods in the restart break (i.e., 1 by definition in the 1NR condition, and 2.3 on average in the MT1NR condition) is critically relevant for sleep recuperation. Indeed, drivers gravitated toward the same nighttime-oriented pattern of sleep during their restart break whether they were in the 1NR condition or in the MT1NR condition (Fig. 3, right). Thus, it is reasonable to assume that deliberately extending the restart break from one to more than one nighttime period would substantively increase drivers’ cumulative opportunity to obtain recuperative nighttime sleep before beginning a new duty cycle. An HOS rule with a requirement for CMV drivers to include at least two nighttime periods (01:00–05:00) in their restart break (as implemented by the FMCSA prior to the rule’s suspension) may therefore be efficacious in mitigating driver fatigue. A field intervention study would be needed to confirm the behavioral effectiveness of such a requirement. 4.2. Objective and subjective fatigue To measure fatigue objectively during the study, drivers performed the 3-min PVT-B three times per duty period (immediately before starting a duty day, about halfway through the duty day, and after ending a duty day). The PVT-B is a validated assay of fatigue (Basner and Dinges, 2011; Basner and Rubinstein, 2011; Grant et al., in press). As shown in Fig. 4 (top), the number of lapses of attention on the PVT-B was significantly greater during duty cycles preceded by the 1NR condition than during duty cycles preceded by the MT1NR condition. The magnitude of the difference during the night, when this difference was greatest, is comparable to the daily increase in lapses of attention on the PVT-B documented in a study involving sleep restriction to 4 h per day (Basner and Rubinstein, 2011). To further put the nighttime performance impairment into context, we converted lapses on the 3-min PVT-B as observed in the first nighttime bin (00:00–04:00) to estimated lapses on a closely related 10-min psychomotor vigilance test (PVT; Lim and Dinges, 2008), based on a head-to-head comparison of the 3-min PVT-B

62

A.R. Sparrow et al. / Accident Analysis and Prevention 93 (2016) 55–64

Table 1 Summary of study results by outcome measure. Outcome Measure

Summary of Differences between Study Conditions

Duty time (Fig. 2)

Time spent on duty occurred predominantly at night in the 1NR condition, whereas it was more evenly distributed across the hours of the day in the MT1NR condition Time spent driving was greater and occurred more typically at night in the 1NR condition, as compared to the MT1NR condition Sleep was restricted in both conditions and occurred predominantly during the day in the 1NR condition and predominantly at night in the MT1NR condition during duty cycles Sleep duration was longer and sleep occurred predominantly during the night in both conditions during the restart break Drivers exhibited more lapses of attention, especially at night, in the 1NR condition, as compared to the MT1NR condition Drivers reported greater sleepiness, especially towards the end of their duty periods, in the 1NR condition, as compared to the MT1NR condition There was a small increase in lane deviation, especially at night, in the 1NR condition, as compared to the MT1NR condition

Driving (Fig. 2) Sleep (Fig. 3)

PVT-B lapses (Fig. 4) KSS sleepiness (Fig. 4) Lane deviation (Fig. 5)

Fig. 6. Comparison of fatigue in CMV drivers in the current naturalistic field study (black) with fatigue in other studies and populations (gray), as quantified using lapses of attention on the 10-min PVT.4 Source references: (1) Van Dongen et al. (2011); (2) Honn et al. (2016); (3) Waggoner et al. (2012); (4) Whitney et al. (2015).

and the 10-min PVT in a recent validation study (Grant et al., in press). The 10-min PVT has been employed in many other studies of shift work and fatigue, which may be drawn from here for comparison. Fig. 6 shows such a comparison, which suggests that nighttime fatigue in CMV drivers during duty cycles preceded by the 1NR condition was substantial, even when judged against operationally relevant conditions in other populations typically exposed to fatigue. It has been argued that increased lapses of attention are associated with elevated risk of crashes, by increasing the likelihood of

4 In Fig. 6, from left to right, the populations sampled and conditions investigated are: (A) healthy young adults in a laboratory study involving a simulated day shift schedule; (B) commercial aviation pilots flying a single long-range flight in a highfidelity flight simulator; (C) commercial aviation pilots flying multiple short-range flights in a high-fidelity flight simulator; (D) police officers working nights, studied in the morning at the end of a 3-day period off duty; (E) healthy young adults in a laboratory study involving a simulated night shift schedule with prior daytime sleep; (F) CMV drivers in the current field study during duty cycles preceded by a restart break with more than one nighttime period, as measured at night (00:00–04:00; see Fig. 4, top) and converted from 3-min PVT-B lapses to estimated 10-min PVT lapses; (G) police officers working nights, studied in the morning at the end of a 5-day duty cycle; (H) CMV drivers in the current field study during duty cycles preceded by a restart break with only one nighttime period, as measured at night (00:00–04:00; see Fig. 4, top) and converted from 3-min PVT-B lapses to estimated 10-min PVT lapses; (I) healthy young adults in a laboratory study involving a simulated night shift schedule without prior daytime sleep (i.e., during total sleep deprivation).

safety-critical events to coincide with periods of inattention (Van Dongen and Hursh, 2010). Lapses of attention are periods when a driver could be missing information, such as a road sign, a traffic signal, or another road user (Rumar, 1990); or could be less able to maintain a stable lane position (Verster et al., 2013). A crash may occur when hazardous circumstances align and a driver has a lapse of attention and does not take mitigating action (Van Dongen and Hursh, 2010). As a person experiences increased fatigue and more lapses of attention, the average duration of these lapses increases (Lim and Dinges, 2008). Thus, it is plausible that more lapses of attention on the PVT-B are associated with elevated crash risk. The link between fatigue and transportation safety is well established (Philip and Åkerstedt, 2006; Raslear et al., 2011), but to what extent lapses on the PVT-B are a reliable correlate of crash risk in truck drivers has not been formally established. Driving performance during the study was measured in terms of lane deviation (i.e., SD of lateral lane position). Lane deviation data were analyzed only if they were captured with the highest degree of confidence in the detection of the lane markers, on straight roads with constant lane width, when the driver was not changing lanes, and when the driving speed was near-constant and between 45 mph and 65 mph. This reduced the lane deviation data to a set comparable to that analyzed in the earlier laboratory research of the restart break (Van Dongen et al., 2011; Forsman et al., 2013), where

A.R. Sparrow et al. / Accident Analysis and Prevention 93 (2016) 55–64

lane deviation was found to be sensitive to fatigue in a personal automobile driving simulator. In the current field study, the differences between conditions across the 24 h of the day for lane deviation (Fig. 5) were similar to those for lapses of attention on the PVT-B (Fig. 4, top). However, for lane deviation the differences between conditions were much more moderate in magnitude. Given the physical inertia of a heavy truck, drivers’ lane keeping ability should perhaps be evaluated using a metric other than lane deviation computed at 1-min intervals. In a driving simulator study of CMV drivers, it was found that crashes involving fatigue were preceded by sporadic short intervals with no significant steering activity mixed with large steering angle corrections (Mortazavi et al., 2009). These phenomena are potentially related to lapses of attention such as those seen on the PVT-B. On the other hand, it has been shown that the effects of fatigue on driving performance are much greater in a driving simulator than in the real world (Hallvig et al., 2013). As with lapses of attention on the PVT-B, the extent to which lane deviation is a reliable correlate of crash risk in truck drivers in the real world has yet to be determined. Self-reported sleepiness was marginally higher overall during duty cycles preceded by the 1NR condition as compared to the MT1NR condition (Fig. 4, bottom). During duty cycles preceded by the 1NR condition, self-reported sleepiness was relatively increased during the late morning hours, toward the end of driving activity typical for that condition. Interestingly, another naturalistic field study of CMV drivers found evidence of such an effect on safety-critical events, which appeared as an interaction between driving hours and work hours. That is, if a driver began the day with several hours of non-driving work, followed by driving late into the 14-h duty period, then the risk of safety-critical events was observed to be increased (Soccolich et al., 2013). Whether the increase in subjective sleepiness during the late morning hours during duty cycles preceded by the 1NR condition reflects the same effect cannot be determined from the data at hand. Although self-reported sleepiness scores are subject to a range of possible confounds and should not be interpreted in an absolute sense, a relative comparison between study conditions of the temporal changes in these scores can be meaningful (Oonk et al., 2008). That said, the 24-h patterns of subjective sleepiness in the current field study (Fig. 4, bottom) were not congruent with those observed for the PVT-B (Fig. 4, top) and for lane deviation (Fig. 5). Discrepancies between subjective and objective measures of fatigue have been found previously in laboratory studies of simulated night work (Van Dongen et al., 2011) and sustained sleep restriction (Van Dongen et al., 2003), and in a field study of professional automobile drivers assigned to morning, afternoon or evening driving shifts (Zhang et al., 2014). The present field replication of this finding is an important reminder that self-monitoring of fatigue may not be a reliable strategy for fatigue risk management (e.g., Satterfield and Van Dongen, 2013). 4.3. Conclusion In summary, we found that when drivers had only one nighttime period (01:00–05:00) in their restart break, this was associated with driving primarily at night. Also, when drivers had only one nighttime period in their restart break, they experienced greater nighttime fatigue during duty cycles. As such, when drivers were most likely to drive at night, they were also at the greatest risk of fatigue, and thus had the greatest need for sleep recuperation. Regardless of the number of nighttime periods in the restart break, drivers adopted a predominantly nighttime-oriented sleep schedule during the restart break. Taken together, these results suggest that including more than one nighttime period (01:00–05:00) in the restart break is a viable approach to mitigating fatigue in the

63

subsequent duty cycle, both objectively and subjectively. The beneficial effects of having more than one nighttime period in the restart break are likely to be a straightforward consequence of increasing the cumulative amount of sleep obtained during the restart break, i.e., of increasing the opportunity for recovery from fatigue before recycling back to the work force. The results of this naturalistic field study are in line with the earlier laboratory research of the restart break (Van Dongen and Belenky, 2010; Van Dongen et al., 2010). Those studies demonstrated that simulated nighttime duty schedules are associated with reduced sleep and increased fatigue, which can be mitigated by extending a simulated restart break to include a second biological night. In the current field study, we found that nighttime duty schedules are most prevalent in duty cycles preceded by a restart break containing only one nighttime period, and such duty cycles are associated with increased fatigue. Seen in the context of the earlier laboratory research, the results of the field study again support the idea that a restart break with more than one nighttime period, relative to a restart break with only one nighttime period, helps to mitigate fatigue. It is important to point out that this field study focused on sleep and fatigue, not on road safety. While it is well recognized that fatigue erodes the safety margin in truck driving (Williamson et al., 2011), we did not measure safety-critical events or crashes – and no crashes were reported during the study. Also, several drivers used their sleeper berth for some or all of their sleep during duty cycles. Stratification of the results by sleeper berth use would be of interest but was not possible, as there was considerable variation in drivers’ documentation of sleeper berth use. Lastly, an important limitation of the study is that our sample, although relatively large for a naturalistic field study with extensive measurements of sleep, fatigue and driving, was drawn from only three carriers and also included only three independent owner-operators. We do not know to what extent the results are representative of the full breadth of US trucking operations utilizing the restart provision of the HOS regulations for CMV drivers.

Acknowledgments The FMCSA was charged to have this field study conducted (US Government Printing Office, 2012) in order to expand on the results of an earlier laboratory study (Van Dongen and Belenky, 2010). It was stipulated that the field study should be representative of drivers affected by the maximum driving time requirements for CMV drivers in the US. Additionally, the field study was to be consistent with the methodology of the earlier laboratory study, as well as with that of another study titled “Scheduling and Fatigue Recovery Project.” The FMCSA provided funding for the field study (award DTMC75-12-J-00049), and required the research to constitute an independent investigation. Carriers contributed resources and made trucks available for study instrumentation and use. An independent expert panel reviewed the study design, methodology, data collection, analyses and interpretation. We gratefully acknowledge these parties, as well as the CMV drivers who volunteered to participate in the field study. We also thank Martin Walker and Richard Hanowski for their support of the study. A technical report produced for the FMCSA (Van Dongen and Mollicone, 2013) provided a basis for this paper. However, the FMCSA was not involved in writing the paper, nor were the carriers or the expert panel, and the paper does not necessarily reflect their views.

64

A.R. Sparrow et al. / Accident Analysis and Prevention 93 (2016) 55–64

References Åkerstedt, T., Anund, A., Axelsson, J., Kecklund, G., 2014. Subjective sleepiness is a sensitive indicator of insufficient sleep and impaired waking function. J. Sleep Res. 23 (3), 240–252. Basner, M., Dinges, D.F., 2011. Maximizing sensitivity of the Psychomotor Vigilance Test (PVT) to sleep loss. Sleep 34 (5), 581–591. Basner, M., Rubinstein, J., 2011. Fitness for duty: a 3-minute version of the Psychomotor Vigilance Test predicts fatigue-related declines in luggage-screening performance. J. Occup. Environ. Med. 53 (10), 1146–1154. Basner, M., Mollicone, D., Dinges, D.F., 2011. Validity and sensitivity of a brief Psychomotor Vigilance Test (PVT-B) to total and partial sleep deprivation. Acta Astronaut. 69 (11–12), 949–959. Borbély, A.A., 1982. A two process model of sleep regulation. Hum. Neurobiol. 1, 195–204. Daan, S., Beersma, D.G.M., Borbély, A.A., 1984. Timing of human sleep: recovery process gated by a circadian pacemaker. Am. J. Physiol. 246, R161–R178. Department of Transportation, 2011. Hours of Service of Drivers; Final Rule. 49 CFR Parts 385, 386 390, and 395, http://www.gpo.gov/fdsys/pkg/FR-2011-12-27/ pdf/2011-32696.pdf (accessed 02.12.15.). Forsman, P.M., Vila, B.J., Short, R.A., Mott, C.G., Van Dongen, H.P.A., 2013. Efficient driver drowsiness detection at moderate levels of drowsiness. Accid. Anal. Prev. 50, 341–350. Grant, D.A., Honn, K.A., Layton, M.E., Van Dongen, H.P.A., 2016. 3-minute smartphone-based and tablet-based psychomotor vigilance tests for the assessment of reduced alertness due to sleep deprivation. Behav. Res. Methods (in press). Hallvig, D., Anund, A., Fors, C., Kecklund, G., Karlsson, J.G., Wahde, M., Åkerstedt, T., 2013. Sleepy driving on the real road and in the simulator—a comparison. Accid. Anal. Prev. 50 (1), 44–50. Hanowski, R.J., Hickman, J., Fumero, M.C., Olson, R.L., Dingus, T.A., 2007. The sleep of commercial vehicle drivers under the 2003 hours-of-service regulations. Accid. Anal. Prev. 39 (6), 1140–1145. Hanowski, R.J., Hickman, J.S., Olson, R.L., Bocanegra, J., 2009. Evaluating the 2003 revised hours-of-service regulations for truck drivers: the impact of time-on-task on critical incident risk. Accid. Anal. Prev. 41 (2), 268–275. Honn, K.A., Satterfield, B.C., McCauley, P., Caldwell, J.L., Van Dongen, H.P.A., 2016. Fatiguing effect of multiple take-offs and landings in regional airline operations. Accid. Anal. Prev. 86, 199–208. Lavie, P., 1986. Ultrashort sleep-waking schedule: III. ‘Gates’ and ‘forbidden zones’ for sleep. Electroencephalogr. Clin. Neurophysiol. 63, 414–425. Lim, J., Dinges, D.F., 2008. Sleep deprivation and vigilant attention. Ann. N. Y. Acad. Sci. 1129, 305–322. Monk, T.H., 1990. The relationship of chronobiology to sleep schedules and performance demands. Work Stress 4 (3), 227–236. Mortazavi, A., Eskandarian, A., Sayed, R.A., 2009. Effect of drowsiness on driving performance variables of commercial vehicle drivers. Int. J. Automot. Technol. 10 (3), 391–404. Oonk, M., Tucker, A.M., Belenky, G., Van Dongen, H.P.A., 2008. Excessive sleepiness: determinants, outcomes, and context. Int. J. Sleep Wakefulness 1 (4), 141–147. Philip, P., Åkerstedt, T., 2006. Transport and industrial safety, how are they affected by sleepiness and sleep restriction? Sleep Med. Rev. 10 (5), 347–356. Pylkkönen, M., Sihvola, M., Hyvärinen, H.K., Puttonen, S., Hublin, C., Sallinen, M., 2015. Sleepiness, sleep, and use of sleepiness countermeasures in shift-working long-haul truck drivers. Accid. Anal. Prev. 80, 201–210.

Raslear, T.G., Hursh, S.R., Van Dongen, H.P.A., 2011. Predicting cognitive impairment and accident risk. Prog. Brain Res. 190, 155–167. Rumar, K., 1990. The basic driver error: late detection. Ergonomics 33 (10–11), 1281–1290. Satterfield, B.C., Van Dongen, H.P.A., 2013. Occupational fatigue, underlying sleep and circadian mechanisms, and approaches to fatigue risk management. Fatigue: Biomed. Health Behav. 1 (3), 118–136. Soccolich, S.A., Blanco, M., Hanowski, R.J., Olson, R.L., Morgan, J.F., Guo, F., Wu, S.C., 2013. An analysis of driving and working hour on commercial motor vehicle driver safety using naturalistic data collection. Accid. Anal. Prev. 58, 249–258. Strogatz, S.H., Kronauer, R.E., Czeisler, C.A., 1987. Circadian pacemaker interferes with sleep onset at specific times each day: role in insomnia. Am. J. Physiol. 253, R172–R178. US Government Printing Office, 2012. Moving Ahead for Progress in the 21st Century Act. 112th Congress Public Law 141, http://www.gpo.gov/fdsys/pkg// PLAW-112publ141/html/PLAW-112publ141.htm (accessed 02.12.15.). Van Dongen, H.P.A., Belenky, G., 2010. Investigation into Motor Carrier Practices to Achieve Optimal Commercial Motor Vehicle Driver Performance: Phase I. Report No. FMCSA-RRR-09-057. Department of Transportation, Washington, D.C. Van Dongen, H.P.A., Hursh, S.R., 2010. Fatigue, performance, errors, and accidents. In: Kryger, M.H., Roth, T., Dement, W.C. (Eds.), Principles and Practice of Sleep Medicine. , 5th ed. Elsevier Saunders, St. Louis, pp. 753–759. Van Dongen, H.P.A., Mollicone, D.J., 2013. Field Study on the Efficacy of the New Restart Provision for Hours of Service. Report No. FMCSA-RRR-13-058. Department of Transportation, Washington, D.C. Van Dongen, H.P.A., Maislin, G., Mullington, J.M., Dinges, D.F., 2003. The cumulative cost of additional wakefulness: dose-response effects on neurobehavioral functions and sleep physiology from chronic sleep restriction and total sleep deprivation. Sleep 26 (2), 117–126. Van Dongen, H.P.A., Olofsen, E., Dinges, D.F., Maislin, G., 2004. Mixed-model regression analysis and dealing with interindividual differences. Meth. Enzymol. 384, 139–171. Van Dongen, H.P.A., Jackson, M.L., Belenky, G., 2010. Duration of Restart Period Needed to Recycle with Optimal Performance: Phase II. Report No. FMCSA-RRR-10-062. Department of Transportation, Washington, D.C. Van Dongen, H.P.A., Belenky, G., Vila, B.J., 2011. The efficacy of a restart break for recycling with optimal performance depends critically on circadian timing. Sleep 34 (7), 917–929. Verster, J.C., Bervoets, A.C., de Klerk, S., Roth, T., 2013. Lapses of attention as outcome measure of the on-the-road driving test. Psychopharmacology 231 (1), 283–292. Waggoner, L.B., Grant, D.A., Van Dongen, H.P.A., Belenky, G., Vila, B., 2012. A combined field and laboratory design for assessing the impact of night shift work on police officer operational performance. Sleep 35 (11), 1575–1577. Whitney, P., Hinson, J.M., Jackson, M.L., Van Dongen, H.P.A., 2015. Feedback blunting: total sleep deprivation impairs decision making that requires updating based on feedback. Sleep 38 (5), 745–754. Williamson, A., Lombardi, D.A., Folkard, S., Stutts, J., Courtney, T.K., Connor, J.L., 2011. The link between fatigue and safety. Accid. Anal. Prev. 43 (2), 498–515. Zhang, H., Yan, X., Wu, C., Qiu, T.Z., 2014. Effect of circadian rhythms and driving duration on fatigue level and driving performance of professional drivers. Transp. Res. Rec. 2402, 19–27.