It is well known from laboratory studies that fatigue can cause shortfalls in performance, including slower response times, attention failures, and poor decision making (see Chapter 3). It also is well known from empirical data collection that fatigue can result in an increased risk of crashes, which is due to these decreases in performance. Further, it is reasonable to believe that restrictions on hours of service lead to a reduction in the percentage of fatigued drivers. However, this linkage is complicated by other dynamics that argue against such a simple causal statement. Given the current hours-of-service (HOS) regulations, there would be primary interest in assessing the difference in fatigue between driving for 10 hours and driving for 11 hours in a day, since that is the current policy-relevant question relative to setting HOS regulations. This causal question is framed somewhat narrowly, and the panel believes it is important to view this issue more broadly for the following reason. The nation is experiencing increased use of technology in vehicles that could be effective for fatigue detection (see Chapters 9 and 11), and other improvements are being made in the design of trucks and buses, in the driving environment, possibly in commercial motor vehicle (CMV) drivers’ personal habits, and in the scheduling policies of carriers. Therefore, the question of greatest importance is what factors mediate the above causal path from fatigue to performance shortfalls to crashes, and therefore how accidents can be prevented in a variety of future circumstances, not simply those being faced today.
For present purposes, then, the panel is interested in whether this causal chain is in operation for truck and bus drivers and in the char-
acteristics of its operation. The laboratory and empirical evidence for passenger cars is clear, but given the few rigorous studies for truck and bus drivers and their different driving circumstances, it remains possible that truck and bus drivers, as professionals, may be different. For instance, they may be better able to judge when their performance has been compromised and take some action in response to reduce the risk. Such action might include pulling over and resting; consuming caffeine; or in the longer term, changing one’s sleeping habits or a carrier’s changing the scheduling of drivers. In addition, when trying to address causal factors such as fatigue in a system as complex as CMV driving, one must be aware of the possibility of unintended consequences. For example, the existence of rumble strips may give drivers a false sense of security and so may encourage them not to pull over. Also, having restart provisions that require no driving between specified ranges of hours may increase the density of driving just outside the boundaries. For this reason, it may be useful to examine the literature on sociotechnical systems, which can be helpful in identifying such unintended consequences (Carayon, 2006; Hanowski, 2013).
After providing an introduction to crash risk due to fatigue, this chapter summarizes the techniques that have been used and the data sets to which they have been applied in some of the leading research on how increases in hours of service and increases in fatigue are linked to increases in crash risk for CMV drivers. There is also a relevant literature on the factors that underlie fatigue in CMV drivers. The review in this chapter is not meant to be comprehensive but is based on a selection of key reports that highlight the methods and data used in this research and the kinds of results such methods provide. Given the large number of confounding factors involved (see Table 10-1 in Chapter 10), which include health and other characteristics of the driver, the vehicle used, the driving environment, and policies of the carrier, along with the difficulty of collecting data on many of these factors, it should not be surprising that much remains to be learned about the relationship among hours of service, fatigue, and crash risk. One of the key research issues is the need to separate out the impacts of hours worked, time of day, and number of hours slept to determine the extent to which each affects fatigue and therefore crash risk. This kind of analysis will require more sophisticated statistical models than have routinely been applied in this area. Some of these techniques are discussed in Chapter 6, and Chapter 10 provides a conceptual approach for moving forward.
This chapter concludes by identifying questions that need additional research. The hope is to both help policy makers understand the complexity of the issue and to help guide future researchers in deciding where to focus their efforts to reduce the remaining uncertainties.
It should be noted that all but three of the studies included in the literature review in this chapter (Connor et al. , Guo et al. , and Tefft [2010, 2014]) involved data on commercial motor vehicles rather than passenger cars. The panel understands that there exists an extensive literature on passenger car crashes and the relationship between crash risk and fatigue for automobiles that these three sources only touch on. The reason for the deemphasis on this related literature in the present review is that the causal model for truck and bus drivers is likely considerably different from that for passenger car drivers with respect to the strength of the individual causal factors that need to be accounted for, but perhaps also with regard to which causal factors are involved. Relative to driving a passenger vehicle, CMV driving often involves longer periods of continuous driving, greater fractions of a day and of a week spent driving, the resulting lifestyle, the economic pressures to continue driving when fatigued, the physical demands of loading and unloading, and the differences in driving a truck or bus compared with a passenger car, not to mention the lack of an analogue to HOS regulations. All of these factors contribute to the panel’s view that the emphasis here should be on research on the relationship among fatigue, hours of service, and crash risk for CMV drivers.
In the United States, 3,921 people were killed and 104,000 were injured in crashes involving large trucks in 2012. The analogous statistics for bus crashes for 2011 were 283 and 24,000, respectively. Crash databases compiled from police reports, such as the Fatality Analysis Reporting System (FARS), are sometimes used to provide estimates of the number of crashes involving trucks or buses that were associated with a fatigued driver. In particular, FARS has been used to estimate that 1.5 percent of crashes involving large trucks that resulted in a fatality in 2013 were due to the truck driver being asleep or fatigued.
Many view this estimate, and similar estimates for nonfatal crashes, as biased low because driver fatigue is difficult to detect during police accident investigations (the basis for FARS reporting) (see Chapter 5). Additionally, police investigators, not usually trained in how to recognize fatigue post hoc, are somewhat reluctant to identify it as such on crash reports because they subsequently will be expected to explain in court why they labeled a crash as related to driver fatigue. If a vehicle is not instrumented for the purpose, it is extremely difficult to determine whether fatigue contributed to a crash. And as argued in Chapter 5, even if a vehicle is instrumented, there remain situations in which it is unclear whether fatigue was a primary contributing factor.
An often-quoted, relatively high estimate of the percentage of fatal truck crashes associated with a fatigued driver resulted from a 1990 report by the National Transportation Safety Board (NTSB) (National Transportation Safety Board, 1990). As discussed in greater detail below, the NTSB conducted an in-depth examination of 182 crash reports on fatal-to-the-driver large-truck crashes that occurred in eight states between October 1, 1987, and September 30, 1988. The analysis considered information available on the number of hours recently driven, the type of accident (such as colliding with a vehicle ahead of the truck or gradually veering out of one’s lane), and time of day. Driver fatigue was identified as a principal cause in 31 percent (56 of 182) of those fatal-to-the-driver crashes (National Transportation Safety Board, 1990). Indeed, of all the factors investigated, driver fatigue turned out to be the most frequent cause. The NTSB estimated at the time that the sample of crash reports it examined in depth represented about 25 percent of such fatal-to-the-driver reports nationwide.
Since fatal-to-the-driver crashes are a subset of fatal crashes, which in turn are a subset of crashes involving trucks (or buses), this percentage should not be applied to larger crash populations. (For example, Knipling and Wang  estimated that 1-4 percent of truck crashes were related to driver fatigue.)
A more realistic estimate of the percentage of serious truck crashes linked to driver fatigue comes from the Large Truck Crash Causation Study (LTCCS), conducted by the Federal Motor Carrier Safety Administration (FMCSA) and the National Highway Traffic Safety Administration (NHTSA) between 2001 and 2003. This study entailed conducting in-depth investigations of 963 crashes involving a truck that resulted in a fatality or an injury to determine the critical reasons for these crashes, each of which was assigned to one or more reasons (see below and Chapter 5 for more details on this study). Truck driver fatigue was found to be associated with 13 percent of the crashes (Starnes, 2006). This means that one of the drivers involved was found to be fatigued, but it was not established whether that fatigue was an important contributor to the crash.
There also has been some international work on this issue. In England, for example, Horne and Reyner (1995) found that 16 percent of all vehicle crashes were related to sleep insufficiency. More recently, Garbarino and colleagues (2001) determined that 21.9 percent of highway crashes in Italy between 1993 and 1997 were related to sleepiness. While these studies are not specific to truck crashes, they provide some indication of the importance of driver fatigue as a cause of crashes.1
1 Another study (Stevenson et al., 2014) was brought to the panel’s attention when this report was nearly finalized. It is a careful case-control study of the causes of nonfatal, nonsevere crashes involving heavy vehicles in Australia.
The wide range of estimates of the degree to which sleepiness or fatigue can be shown to be associated with truck collisions is due in part to the many differences in what is being estimated. These differences include the degree of severity of the crash, whether cars or trucks and buses were involved, the time and location of data collection, and the definition of operator fatigue. In the end, despite the various attempts that have been made to estimate the incidence of truck driver fatigue contributing to crashes on the nation’s roadways, the panel simply did not find convincing enough evidence that at this time would support a reliable estimate.
A literature review conducted by Belenky and colleagues (2013) to assess the prevalence of fatigue in bus drivers and its association with crash risk found that none of the studies included in the review investigated the impact of nonpathologic fatigue on the driving ability of these drivers. Putcha and colleagues (2002) analyzed FARS data looking for fatal crashes in which bus drivers were involved during the period 1995-1999. The FARS database includes the contributing factor “drowsy, asleep, or fatigued,” so the authors were able to determine the fatal crashes in which this factor was indicated as occurring. They found this to be the case for only 5 of a total of 1,483 fatal crashes involving buses over the 5-year period. The panel believes that the resulting estimate of 0.3 percent is almost certainly an underestimate, for the reasons discussed earlier. Clearly, this is an area in need of further research.
Following are summaries and critiques of some of the key research examining the relationship among CMV driver fatigue, HOS regulations, and crash risk. For further information on most of these studies, see Knipling (2015) at http://sites.nationalacademies.org/DBASSE/CNSTAT/CMV_Driver_Fatigue_Long-Term_Health_and_Highway_Safety/index.htm.
Crash Involvement of Large Trucks by Configuration:
A Case-Control Study
(Stein and Jones, 1988)
This was a prospective case-control study of the causes of large-truck crashes. According to the authors, “For a two-year period, large truck crashes on the interstate system in Washington State were investigated using a case-control method. For each large truck involved in a crash, three trucks were randomly selected for inspection from the traffic stream at the same
time and place as the crash but one week later. The effects of driver and truck characteristics on crash risk was assessed by comparing their relative frequency among the crash-involved and the comparison sample trucks.” The data set represented 676 crashes involving 734 trucks that occurred between 1984 and 1986. The characteristics assessed for crash and control trucks were truck configuration, age of driver, weight of load, hours of driving, truck body type, and fleet size. The trucks also were inspected to check on the condition of the brakes, steering, and tires. (Continuous variables were classified into three groups of equal size to define matching cases.) To determine whether a variable was distributed differently in the crash versus the control population, the percentage of trucks with that characteristic in the crash population was divided by the percentage of trucks with that characteristic in the control population. In addition, to deal with simultaneous effects of the various characteristics, a logistic regression model was used to estimate the adjusted odds ratio for each of the factors included. Analyses also were stratified by the following factors individually: crash type (single vehicle, multiple vehicle), day/night, route (Interstate 5 or 90), and roadway alignment.
The researchers found that the risk of crashes was higher for double trailer trucks and single units pulling trailers, and it was also higher for younger drivers, longer hours of driving, and operation of empty trucks. The fact that long hours of driving raised the crash risk is of course of interest in the present context. Two criticisms of this study are (1) that data from drivers’ logbooks on the number of hours driven per day may not be of sufficiently high quality, and (2) such analyses rest on the assumption that any characteristics not measured and not equally distributed in the crash and control populations are themselves not causal or correlated with causal factors.
A related study of the same data by the same two researchers (Jones and Stein, 1989), found that driving in excess of 8 hours compared with driving 2 hours or less resulted in a 1.8 times higher risk of a crash. In addition, driver logbook violations raised the unadjusted odds ratio by 3.0. The authors also found that a decrease in the quality of the steering raised crash risk.
Fatigue, Alcohol, Other Drugs, and Medical Factors in
Fatal-to-the-Driver Heavy Truck Crashes, Safety Study
(National Transportation Safety Board, 1990)
As mentioned above, the NTSB investigated 182 crashes in which the driver of a heavy truck was fatally injured, occurring in eight states during October 1, 1987, to September 30, 1988. The purpose of this study was to identify the principal reason(s) why each crash occurred. The
great majority were single-vehicle crashes in which only the truck driver died. The sample represented about 25 percent of the crashes of this type nationally during this period. To assess the contributing factors, the NTSB developed information to augment the crash reports of police crash investigators, making it possible to describe more completely the operator(s), vehicle(s), and roadway at the time of the crashes. In addition, the NTSB interviewed representatives of the carrier, available witnesses, and reachable family members to obtain more detailed information on hours of service, fatigue, carrier operations and maintenance, safety programs, training and testing, preemployment screening, and other factors. The researchers also received the case files developed by the crash investigators and driver logs, as well as blood and urine specimens from the fatally injured drivers obtained from local coroners and medical examiners.
While at the time, the country and the NTSB were still focused on alcohol-related driving fatalities on the nation’s roadways, the somewhat surprising finding of this study was that so many of these truck driver fatalities were more likely attributable to the influence of driver fatigue. The major findings relevant to driver fatigue and exceeding HOS regulations were as follows: (1) fatigue was cited as a probable cause 31 percent of the time, which made it the most frequently cited cause; (2) of the 57 drivers who were fatigued, 19 were also impaired by alcohol or other drugs; and (3) there was a strong association between HOS violations and drug use.
The primary goal of the Driver Fatigue and Alertness Study (DFAS) was “to observe and measure the development and progression of driver fatigue and loss of alertness, and to develop countermeasures to address it, through a field study.…” Beginning in 1993, 80 truck drivers aged 25-65 with at least 1 year of experience in the United States and Canada driving long-haul less-than-truckload cargo in tractor-semi-trailers were monitored for 16 weeks each as part of a naturalistic driving study. Data were collected on work-related factors “thought to influence the development of fatigue, loss of alertness, and degraded driving performance in commercial motor vehicle drivers.” As is typical of naturalistic driving studies (see Chapter 5), the DFAS was carried out within an operational setting of real-life, revenue-generating trips. The work-related factors examined included the amount of time spent driving during a work period, the number of consecutive days of driving, the time of day when
driving took place, and schedule regularity. For each driver, data collection lasted 4-5 days.
The drivers were divided into four groups of 20, and each group was asked to follow one of the following schedules: (1) 10 driving hours turnaround route, starting around 10 AM for five consecutive trips; (2) 10 driving hours turnaround route starting 3 hours earlier each successive day, with more night driving time than in schedule (1); (3) 13 driving hours turnaround route starting late each evening for four consecutive trips, with more night driving than in schedule (2); and (4) 13 driving hours starting in the late morning and early afternoon for four consecutive trips. Schedule (1) provided 11 hours off between trips, while the other three schedules provided only 8 hours off between trips. Measures collected for each subject included lane tracking, steering wheel movement, driving speed, distance monitoring, response vigilance tests, continuous video monitoring of the driver’s face and the road ahead, and polysomnography during sleep and while driving.
The strongest factor influencing driver fatigue was determined to be time of day of driving. Drowsiness was greatest during night driving. Hours of daytime driving was not a strong predictor of observed fatigue. (Hours of nighttime driving could not be assessed as a predictor of fatigue given the study design.) Finally, there was some evidence of cumulative fatigue across days of driving. The fact that there was more than one difference among the schedules confounded attempts to interpret comparisons of means across the four groups.
Effects of Sleep Schedules on Commercial Motor Vehicle Driver Performance
(Balkin et al., 2000): Study 1, Actigraphic Assessment of Sleep of CMV Drivers over 20 Days
In this field study (the report also describes a simulator study), wrist actigraphy was used to determine the hours of sleep obtained by 25 long- and 25 short-haul CMV drivers over 20 consecutive days, both on and off duty. It was found that both long- and short-haul drivers averaged 7.5 hours of sleep per 24 hours.
Stress and Fatigue Effects of Driving Longer Combination Vehicles
(Battelle-Seattle Research Center, 2000)
The authors describe this study as follows:
Between October 31, 1994, and January 21, 1995, 24 experienced longer-combination vehicle drivers drove approximately 2700 miles
each in specially-equipped and loaded single- and triple-trailer commercial vehicles under controlled experimental conditions. The tractors were equipped with video and digital equipment to gather data on the drivers’ performance. . . . Each driver who participated in the study was assigned to a specific tractor and drove it for the entire study week, using it to alternatively pull each of the three different trailer configurations: a single 48-foot trailer, a triple-trailer combination with three 28-foot trailers and standard converter (A-dollies), and a triple-trailer combination with three 28-foot trailers and double-drawbar, self-steering converter dollies (C-dollies).
With three possible configurations, there were six orderings of trailers, and four drivers were randomly assigned to each permutation. Fatigue-related measures included lane deviation assessments. Each driving day consisted of 10 hours on duty, including 8 hours of driving.
The researchers determined that driving the triple trailers contributed to increased fatigue as measured primarily by lane departures. In addition, there was substantial heterogeneity in the outcomes, including lane departures, representing 32 percent of the variability in outcomes.
Driver Sleepiness and Risk of Serious Injury to Car Occupants:
Population Based Case Control Study
(Connor et al., 2002)
This study examined data from 571 car drivers involved in crashes in the Auckland region of New Zealand between April 1998 and July 1999 in which at least one occupant was admitted to a hospital or killed. It also looked at 588 controls who were randomly selected to mimic the distribution of people driving on the region’s roads during the study period. The cases were “identified by cluster sampling of drivers at 69 randomly selected sites on the road network. The day of the week, time of day, and direction of travel for each survey site were randomly assigned.” The goal was to determine whether the relative risk for injury was associated with various driver characteristics, especially fatigue, through case-control methods. For each driver admitted to the hospital, interviews were conducted, often within 48 hours of the crash. For crashes that were fatal to the driver, proxies were interviewed. Questions about sleep obtained made up a small portion of the interview to disguise the intent. To employ the Stanford sleepiness scale, the researchers had respondents select one of seven statements that most closely described their alertness immediately before the crash. Controls were similarly interviewed around the time of their selection.
Confounding factors considered in the analysis included age, gender,
socioeconomic status, ethnicity, alcohol consumption, use of recreational drugs, time spent driving per week, vehicle speed, average traffic speed, type of road, and how long the person had been driving on the day of the crash. The analysis involved the estimation of odds ratios using logistic regression, with the complication that the cluster sample was accommodated using the SUDAAN statistical analysis software. The change-in-estimate method was used to assess potential confounders (Greenland, 1989). The confounders that passed this test were included in the logistic regression model.
The results showed a “strong association between the level of acute driver sleepiness, as measured by the Stanford sleepiness score, and the risk of injury. . . . The two direct determinants of acute sleepiness . . . sleep deprivation and time of day, were also strongly associated with the risk of an injury crash. Drivers who reported five hours or less of sleep in the previous 24 hours were at significantly increased risk compared with those who had more than five hours.”
In addition to a considerable amount of missing data, this study was subject to recall bias. It is well known, as discussed earlier, that self-reports about the amount of sleep received are of uncertain quality. Finally, this was not a study of truck drivers but of automobile drivers, and it was for travel in New Zealand.
In the Motor Carrier Safety Improvement Act of 1999, Congress mandated “a study to determine the causes of, and contributing factors to, crashes involving commercial motor vehicles.” As a result, FMCSA and NHTSA conducted a “multiyear, nationwide study of factors that contribute to truck crashes. . . . A nationally representative sample of large truck fatal and injury crashes was investigated during 2001 to 2003 at 24 sites in 17 states. Each crash involved at least one large truck and resulted in at least one fatality or injury. Data were collected on up to 1,000 elements in each crash. The total sample involved 967 crashes, which involved 1,127 large trucks, 959 nontruck motor vehicles, and included a total of 251 fatalities, and 1,408 injuries.”
Data collection was carried out at each crash site by a two-person team consisting of a trained researcher and a state truck inspector. They collected data on the crash scene, including information about the roadway and the weather; vehicle rollover, fire, jackknife, or cargo shift; problems with brakes, tires, steering, engine, or lights; driver credentials, method of payment, physical condition, fatigue (based on sleep pattern, work schedule, and recreational activities), and inattention/distraction;
trip start time, purpose, and intended length; and driver’s familiarity with the route.
One and only one “critical event” was designated for each crash—the event that immediately led to the crash. Likewise, one and only one “critical reason” (the immediate reason for the critical event) was assigned to each event. In addition, crashes were coded with associated factors, which were indicated as being present but not necessarily causal, and more than one of these could be assigned to an individual crash. The findings relevant to fatigue were that shortage of sleep was given as the critical reason in 7 percent of the crashes, and partial sleep deprivation was given as an associated factor in 13 percent of the crashes.
The LTCCS had some design flaws. First, fatigue was assessed either indirectly or by self-report, so the quality of that information likely is not high. In addition, the requirement to find a critical event and the critical reason for that critical event could have biased the observers toward factors in immediate physical or temporal proximity. Even with these flaws, however, and the fact that the data are now 10 years old, the LTCCS is considered an important source of quality information on the relative likelihood of various causes of truck crashes.
In this study, three samples of long-distance truck drivers were interviewed face to face as they passed through roadside weigh stations on Interstate highways in Pennsylvania and Oregon immediately before and after the 2003 change in the HOS regulations, which increased the limit on daily driving from 10 to 11 hours. The first sample of responses was collected prior to the change, from November to December 2003; the second sample was collected 1 year after the change, from November to December 2004; and the third sample was collected 2 years after the change, from November to December 2005. A total of 1,921 drivers participated in one of the three groups of interviews.
To encourage participation, drivers were given an incentive payment of $10. Participation rates ranged from 88 to 98 percent. Questions were asked about work schedules, rule violations, and fatigued driving, with the differences between 2003 and 2004 responses and between 2003 and
2 The rule changes at issue between 2003 and 2004 were (1) the daily minimum off-duty requirement was changed from 8 hours to 10 hours, (2) the maximum hours of driving prior to going off duty was changed from 10 to 11 hours, (3) the maximum tour of duty was 14 hours, and (4) the 34-hour restart period was initiated.
2005 responses providing the statistics to be interpreted. The authors report the following findings:
The large majority (72-76% in 2004 and 69-70% in 2005) said that their current daily driving times were about the same as before the rule change. But in both 2004 and 2005, about one-fifth of drivers said they were driving more hours daily under the new rule. . . . In the 2004 and 2005 surveys, a sizeable percentage of drivers in both states reported they typically got more daily sleep under the new work rule than under the old rule. . . . At least 72% said the restart was part of their regular schedules. . . . The percentage of drivers interviewed in Pennsylvania who said they drove their trucks while sleepy at least once during the past week increased from 43% in 2003 to 48% in 2004 and then declined to 43% in 2005. . . . In Oregon, the percentage was who reported sleepy driving was 36% in both 2003 and 2004 and 41% in 2005. The percentage who reported dozing at the wheel of a truck on at least one occasion during the past month increased over time in each state, with the percentage difference between 2004 and 2005 being statistically significant.
Finally, compliance with the rules decreased in Pennsylvania over the 2-year time period but went up in Oregon.
There are two main criticisms of this study. First, as noted earlier, driver reports of sleepiness are not always of high quality. Second, there may have been other dynamics between 2003 and 2004 and 2005 that were not controlled for.
Analysis of Risk as a Function of Driving-Hour:
Assessment of Driving-Hours 1 Through 11
(Hanowski et al., 2008)
This project was a naturalistic driving study of 98 CMV drivers (97 males, 1 female, age range of 24-60). Data collection started in May 2004 and was completed in September 2005. Study participants drove company trucks on their usual routes. Equipment, most of which was unobtrusive, was installed in 46 trucks to record the driver, the road ahead, and other data. The average number of weeks the drivers participated was 12.4. The final data set consisted of 2.3 million miles of driving data. Driving performance was assessed through the occurrence of critical incidents, which included crashes, near-crashes, and crash-relevant conflicts. In addition, for some of the analyses, only those incidents in which the driver was viewed as being at fault were included.
Given the potential for subjectivity in the assessment of near-crashes and crash-relevant conflicts, the number of critical incidents varied in each of eight analyses carried out. Also, to adjust for the differences
in opportunities across driving hours, the frequency of critical incidents in any given driving hour was divided by the total opportunities for that hour.
One direct analysis of the frequency of critical incidents as a function of driving hour showed a visible spike in the relative frequency of critical incidents in the first driving hour, and this finding was consistent across the various analyses. There was no evidence of a time-on-task effect. A second analysis computed odds ratios using logistic regression models. The assumption of independence of incidents was not made for this analysis; instead, generalized estimating equations were used to account for correlations that might exist between and within drivers. This was done for all trips and conditional on trips that lasted the full 11 hours. In addition, there was evidence of a traffic-density effect.
Investigation into Motor Carrier Practices to Achieve Optimal Commercial Motor Vehicle Driver Performance: Phase I
(Von Dongen and Belenky, 2010)
The objective of this project was to determine the effectiveness of the 34-hour restart provision in the HOS regulations for CMV drivers. To this end, a sample of 27 healthy subjects were subjected to one of two protocols involving two 5-day work periods (14 hours per day), separated by a 34-hour restart period during which the driver transitioned back to a daytime wake, nighttime sleep schedule. The first group drove during the day and slept at night, while the second group drove at night and slept during the day. The primary outcome measure was the comparison between 10-minute psychomotor vigilance tests evaluated before and after the 34-hour restart. In addition, lane deviations were measured. Under both outcome measures, the 34-hour restart provision was more effective at mitigating the sleep loss for those working during the day than for those working at night.
Near Crashes as Crash Surrogates for Naturalistic Driving Studies
(Guo et al., 2010)
This analysis of the 100-car study (a large-scale naturalistic driving study described in Dingus et al. ) examined whether safety-critical events are useful as crash surrogates. The 100-car study collected data on 2 million vehicle-miles and 43,000 hours of driving. Crashes were defined as any contact with an object at any speed in which kinetic energy is measurably transferred or dissipated. A near-crash was defined as “any circumstance that requires a rapid, evasive maneuver by the participant vehicle, or any other vehicle, pedestrian, cyclist, or animal, to avoid a
crash.” The researchers detected such maneuvers by looking at the vehicle kinematic data. The primary measure on which they relied to assess the impact of a factor on traffic safety was the odds ratio, that is, the odds of the presence of a factor for safety events divided by the odds of its presence for baseline events. The authors state that a measure, in this case safety-critical events, is to be viewed as a surrogate measure when (1) the causal mechanisms are the same or similar for crashes, and (2) there is a strong association between the frequency of surrogate and primary measures. Based on results of a variety of analytic techniques, the report suggests that safety-critical events are only somewhat effective as surrogates.
Hours of Service and Driver Fatigue: Driver Characteristic Research
(Jovanis et al., 2011)
This study compared the effect of different driver HOS regulations on the odds of a crash. The analysis was based on crashes reported by the trucking companies that cooperated with the researchers involving either a fatality, an injury requiring medical treatment away from the scene of the crash, or a towaway. Driver logs for periods of 1-2 weeks prior to the crash were compared with those for two noncrash-involved drivers that were randomly selected from the same company, terminal, and month, using case-control logistic regression. Other covariates included cumulative hours driving, driving patterns over multiple days, time of day, breaks during driving, and use of the 34-hour restart policy. Data from 2004-2005 and 2010 were collected from a total of 1,564 truck drivers. Separate analyses were carried out for truckload and less-than-truckload freight modes of operation. The main results were as follows:
- Driving time was a statistically significant predictor of crash risk for the less–than-truckload drivers.
- Less-than-truckload data showed a pattern of increasing crash odds as driving time increased, with a consistent increase from hour 5 through hour 11.
- Truckload data showed significant interactions between some multiday driving patterns and increased crash risk between the seventh and eleventh hours.
- Driving breaks reduced crash risk for both types of drivers.
- Driving times that would have been a violation of the 34-hour restart provision—but were not since they occurred in 2010—were associated with an increased risk of a crash.
Issues with this study include the unclear validity of driver logs and
the omission from the logistic regression model of other covariates that are predictive of crash frequency, such as driver age and experience.
An Assessment of Driver Drowsiness, Distraction, and Performance in a Naturalistic Setting
(Barr et al., 2011)
Barr and colleagues (2011) conducted a reanalysis of a naturalistic driving study carried out by Hanowski and colleagues (2000) on local and short-haul truck drivers to determine causes of fatigued driving. A major difference was that in the original analysis, attention was given only to safety-critical events. In this study, the researchers identified incidents of driver fatigue or drowsiness that occurred during all periods of driving.
The data examined consisted of 908 hours of footage (from five video cameras) on 41 drivers. The researchers reviewed the entire video library to code 3-minute durations defined either as baseline periods or as periods corresponding to the occurrence of a characteristic drowsiness behavior. The coding comprised “not drowsy, slightly drowsy, moderately drowsy, very drowsy, or extremely drowsy,” and was based on evidence of yawning, rubbing eyes, closing eyes, slow blinks, bobbling one’s head, and verbal announcement of drowsiness. The analysis encompassed 1,000 such baseline and fatigue events. Baseline events were matched to fatigue events using road, weather, time of day, traffic, and other conditions and served as a control group in the statistical analysis. The potential predictors available included years of driving experience, time of day of event, amount of time on duty, number of hours slept the previous night, actual sleep as measured by actigraphy, traffic density, number of lanes, the lane the driver was in, road type, road geometry, road conditions, weather, visibility, and illumination outside the vehicle.
The analysis was stratified by the coded degree of fatigue. Logistic regression models (and some contingency table and regression models) were used to determine which factors explained the difference between the baseline and fatigued events. The findings were as follows:
- Higher levels of fatigue were associated with younger drivers.
- Drowsiness was twice as likely to occur between 6:00 and 9:00 AM.
- Drivers were affected by fatigue and drowsiness in the early morning and near the end of their shift.
- Thirty percent of all severe drowsy events occurred in the first hour of the work shift.
- The relationship between sleep quality or quantity and driver fatigue was fairly weak.
- Undivided highways and poor visibility increase driver attention and reduce fatigue.
Besides the small sample size of volunteers, an important limitation of this study was its focus on local/short-haul truck drivers, who face different challenges from those faced by long-haul drivers. In particular, short-haul drivers typically do not drive at night, so the analysis could not address the impact of nighttime driving.
The Impact of Driving, Non-Driving Work, and Rest Breaks on Driving Performance in Commercial Motor Vehicle Operations
(Blanco et al., 2011)
This naturalistic driving study took place between November 2005 and March 2007. The study included 97 drivers (aged 21-73) with an average of 9 years of experience driving commercial motor vehicles. The drivers were employees of four for-hire trucking companies, and represented both long-haul operations and drivers that returned home most nights.
The drivers drove instrumented trucks during their usual routes for 4 weeks. Nine trucks were fitted with video cameras trained on the driver’s face, on the steering wheel, and outside of the truck, and additional sensors measured other aspects of the driver’s actions. Driver performance was assessed through the occurrence of safety-critical events, which were defined as crashes, near-crashes, and crash-relevant conflicts, as well as unintentional lane deviations. In addition, each driver was asked to fill out a daily register that included on-duty and off-duty activities for the 4 weeks of the study.
The results, based on descriptive statistics, odds ratios, and negative binomial regression models, were as follows. First, drivers spent 66 percent of their workday driving and 23 percent doing paperwork, loading or unloading, or performing other work activities. Second, using driving hour as a continuous variable in a mixed-effects negative binomial regression model to model the number of safety-critical events, Blanco and colleagues (2011) showed a statistically significant effect for time on task. Looking at the effects of individual driving hours pairwise, some of this effect stemmed from the fact that the eleventh hour had significantly more safety-critical events than the first and second hours, although there was no statistically significant difference between the effects for the tenth and eleventh hours. On the other hand, an analysis that involved counting only whether a safety-critical event had or had not occurred found no significant differences as a function of driving hour, although an analysis of the rate of occurrence of such events as a function of shift duration did
show a significant increase as shifts grew longer, an effect that lasted into the fourteenth work hour. In addition, when nondriving activities were introduced during the driver’s shift, such breaks significantly reduced the risk of safety-critical events for the first hour after the break. Finally, it should be noted that the spike observed during the first hour of driving in Hanowski et al. (2008) was not seen in this study.
Motorcoach Driver Fatigue Study, 2011
(Belenky et al., 2012)
This study examined whether commercial motorcoach drivers were working within the limits set by the HOS regulations. Data were collected on duty start times, total duty time per 24 hours, and total sleep time per 24 hours. Driver performance and degree of fatigue were measured and related to those predictors.
Eighty-four motorcoach drivers working for charter, tour, regular route, or commuter express carriers were studied for 31 consecutive days. During this time, they kept to their normal work/rest schedules. All drivers maintained a duty/sleep diary. Actigraphy devices were used to measure sleep/wake times, so that sleep variables could be assessed. To measure performance, drivers were administered a psychomotor vigilance test when going on and off duty and before and after breaks during the day. The Samn-Perelli Fatigue Scale and the Karolinska Sleepiness Scale were used to measure subjective sleepiness when the drivers were going on and off duty and before and after breaks. Unfortunately, baseline measures of fatigue were not available.
It was found that drivers drove an average of 43 hours per week. Total time on duty per day averaged slightly more than 9 hours and rarely exceeded the regulatory limit of 15 hours. Mean actigraphically collected total sleep duration was 7 to 9 hours. In general, drivers worked within the limits of the current HOS regulations and balanced the demands of work and family to obtain sufficient sleep.
Effect of Circadian Rhythms and Driving Duration on Fatigue Level and Driving Performance of Professional Drivers
(Zhang et al., 2014)
This naturalistic study examined the interplay among circadian rhythms, drive time, drive duration, fatigue, and driver performance. Fifteen middle-aged professional daytime drivers (using private vehicles with automatic transmission) were randomized to driving one of three schedules: (1) one group started driving at 9 AM, (2) a second group started at noon, and (3) a third group started at 11 PM. Each group drove
for 6 hours. Drivers reported their rating on the Karolinska Sleepiness Scale every 5 minutes while driving so that fatigue could be assessed. Measurements of steering and lane position were taken regularly as well.
The results were as follows: (1) both circadian rhythms and increasing drive duration (between 0 and 6 hours) had significant effects on fatigue levels, and fatigue levels increased more rapidly in the evening group; (2) drivers were most likely to feel tired between 2 and 4 AM and between 2 and 4 PM; and (3) the group that was most tired was the evening group.
NHTSA’s National Automotive Sampling System Crashworthiness Data System (NASS CDS) is a nationally representative sample of police-reported crashes. The first of these studies (Tefft, 2010) used NASS CDS data for 1999 to 2008 on crashes involving passenger vehicles that were towed from the crash scene, which represented 80,821 vehicles involved in 47,597 crashes. The researchers found that “3.9 percent of all those crashes, 7.7 percent of those that resulted in at least one person being admitted to a hospital, and 3.6 percent of those that resulted in death involved a driver who was coded as drowsy. However, the attention status of 45 percent of the drivers in the data was unknown.” To address this degree of uncertainty, imputation was used to estimate the drowsy status of the remaining drivers. The result was “an estimated 7.0 percent of all crashes in which a passenger vehicle was towed, 13.1 percent of crashes that resulted in a person being admitted to a hospital, and 16.5 percent of fatal crashes involved a drowsy driver.” Drowsiness was determined on the basis of information from “interviews conducted by NASS CDS investigators with crash-involved occupants from police reports.” The imputation used the following covariates: maximum injury, driver injury severity, number of vehicles in crash, pre-event maneuver, crash type, day of week, hour of day, traffic flow, number of passengers, driver age, driver gender, light condition, relation to intersection, roadway alignment, speed limit, number of lanes, surface conditions, precrash critical event, vehicle disposition, year, stratum, and primary sampling unit. Among crashes in which the driver was fatally injured, information on attention was missing for 92 percent. Therefore, for inferences about crashes fatal to the driver, 92 percent of fatigue status was imputed.
This study has the following weaknesses. First, it involved only passenger vehicles (a weakness only for drawing inferences about CMV crashes). Second, the validity of assessing drowsiness from interviews is unclear. Finally, the imputation model was not validated, and it was extensively employed.
The second of these studies (Tefft, 2014) is an update of the previous study using data for 2009 to 2013. Instead of the above percentages related to crash severity of 7.0, 13.1, and 16.5 percent, the respective percentages in this later study were 6.0, 13, and 21 percent.
Table 7-1 summarizes the above studies.
HOS regulations need to take into account the trade-off between the economic advantages of transporting goods more quickly and the disadvantages of increasing crash risk. Given that, it would be helpful to make this trade-off as explicit as possible by linking increases in crash risk to increases in the number of hours of service permitted. However, given the multivariate causal structure of crashes, such a construct can be provided only by fixing all the other causal factors at levels that rarely if ever obtain, and therefore, such a statement would not be useful to support the development of policies.
Two Relevant Meta-Analyses Concerning Obstructive Sleep Apnea as a Fatigue-Related Risk Factor
Two important recent meta-analyses provide useful summaries of the literature on how obstructive sleep apnea (OSA) affects crash risk.
Tregear and colleagues (2009b) performed a meta-analysis to understand the degree to which CMV drivers with OSA are at an increased risk of crashes compared with drivers without OSA. They selected 18 studies satisfying a variety of criteria for analysis. They found that drivers with OSA had a 1.21 to 4.89 times higher crash risk compared with drivers without OSA. In addition, the 18 studies indicated that factors associated with increased crash risk for drivers with OSA are body mass index, apnea/hypopnea index, and oxygen saturation.
Tregear and colleagues (2009a) performed a meta-analysis of the impact of the use of continuous positive airway pressure (CPAP) treatment on motor vehicle crash risk for automobile and CMV drivers with OSA. Nine studies met their criteria. They found that CPAP use significantly reduced crash risk following treatment, with a 95 percent confidence interval of 0.22, 0.35.
TABLE 7-1 Summary of Studies on Hours of Service, Fatigue, and Crash Risk
|Study Author(s)||Data Source/Data Collection Method||Corresponding Years||Dependent Variable|
|Stein and Jones (1988)||Crash reports from Washington State and random inspections||1984-1986||Crash rate|
|National Traffic Safety Board (1990)||Crash reports in 8 states augmented by investigators||1987-1988||Fatal-to-the-driver crash rate|
|Wylie et al. (1996)||Naturalistic driving study||1993||Driver fatigue as measured by lane tracking, etc.; EEG; PVT|
|Balkin et al. (2000)||Actigraphy||1997||Potential fatigue as measured by actigraphy; also included simulator study|
|Battelle-Seattle Research Center (2000)||Naturalistic driving study||1994-1995||Lane deviations|
|Connor et al. (2002)||Crash investigations, interviews||1998-1999||Driver ratings of sleepiness; risk of serious crash|
|Federal Motor Carrier Safety Administration (2006), Starnes (2006)||Crash investigations, interviews||2001-2003||Fatal truck crash data, critical event for each crash|
|McCartt et al. (2008)||Personal interviews||2003-2005||Interview data on fatigued driving, sleep obtained|
|Hanowski et al. (2008)||Naturalistic driving study||2004-2005||Frequency of safety-critical events|
|Logistic regression case control||Driving in excess of 8 hours results in a 1.8 times greater crash risk||Driver fatigue measured indirectly by time on task through logbook data|
|Frequencies||Fatigue cited as cause in 31 percent of sample||Driver fatigue assessed by investigators’ reconstruction|
|Means, frequencies||Time of day important; drowsiness greatest during night driving||Confounded design|
|Analysis of variance||Both long- and short-haul drivers often get 7.5 hours of sleep per 24 hours||Some driver populations not subject to substantial sleep loss|
|Analysis of variance||Driving triple trailers adds to fatigue||Small sample; many possible confounders omitted|
|Logistic regression case control||Strong association between sleepiness and crash rate||Much missing data; driver recall bias re sleep obtained|
|Frequencies||7 to 13% of crashes associated with sleep shortage||Driver fatigue is assessed indirectly as critical events at crash sites|
|Frequencies, means||Drivers reported more sleep obtained under new HOS regulations; dozing increased with change in HOS regulations||Interview self-reporting on fatigue|
|Frequencies, logistic regression with generalized estimating equations||First driving hour is most risky; also time-of-day effect and traffic density effect||Some safety-critical events may not be evidence of fatigue|
|Study Author(s)||Data Source/Data Collection Method||Corresponding Years||Dependent Variable|
|Van Dongen and Belenky (2010)||Naturalistic driving||2008-2009||Lane deviations, PVT|
|Guo et al. (2010)||Naturalistic driving study||2008||Safety-critical events and crashes in cars|
|Jovanis et al. (2011)||Company crash data||2004-2005, 2010||Crash risk|
|Barr et al. (2011)||Naturalistic driving study||2000||Degree of fatigue assessed by observation|
|Blanco et al. (2011)||Naturalistic driving study||2005-2007||Safety-critical events|
|Belenky et al. (2012)||Naturalistic driving study (quasi)||2011||PVT, total time on duty, total time asleep|
|Zhang et al. (2012)||Naturalistic driving study, random allocation into treatments||2011 (China)||Karolinska Sleepiness Scale, lane position, steering behavior|
|Tefft (2010, 2014)||Automobile crash data||1999-2008||Police reports of crash frequency in automobiles|
NOTES: EEG = electroencephalogram; HOS = hours-of-service; PVT = psychomotor vigilance test.
|Frequencies, means||34-hour restart, day work shift, nighttime sleep effective at mitigating sleep loss, but work at night, sleep in day not so effective||Small sample size; many potential confounding factors|
|Odds ratios and Poisson regression||Safety-critical events sometimes are effective surrogates, sometimes not||Combining safety-critical events of different types may complicate inference|
|Logistic regression case control||Driving time was a significant predictor of crash risk for less-than-truckload trucks||Paper logs are of uncertain quality; no treatment of several confounding factors|
|Reanalysis of previou study; logistic regression case contr||s Fatigue associated with young drivers, driving ol between 6 and 9 AM, and when starting out||Small sample size; focus on local/short-haul truck drivers|
|Frequencies, odds ratios, negative binomial regression||Rate of safety-critical events increases with time on task; breaks are beneficial||Small sample size; safety-critical events are uncertain surrogates|
|Means||Motorcoach drivers function well under current HOS regulations||Mixed types of motorcoach drivers; no accounting for some confounding factors|
|Means||Fatigue is affected by circadian rhythms and by drive duration||Small sample size|
|Frequencies||16.5% of fatal crashes involved a drowsy driver||Assessment of fatigue by interview; much missing data|
Chapter 10 suggests methodological directions for research on the relationship between fatigue among CMV drivers and highway safety. Foreshadowing that discussion, the following is a list of deficiencies that need to be addressed, some methodological and some substantive (and some not covered in Chapter 10):
- Either more experimental control of important variables is necessary for various confounding factors, or these factors need to be addressed after data collection using techniques such as propensity scoring.
- More research is needed on the relationship among HOS regulations, driver fatigue, and crash risk for bus drivers.
- More research is needed to disentangle the effects of different causes of fatigue—sleep deprivation; chronic lack of sleep; circadian displacement; times on task; and medical conditions, including OSA.
- More research is needed on how drivers decide how to address their sleepiness while driving, that is, what do they do while driving after determining that they are fatigued.
- More research is needed on the benefits of CPAP as treatment for CMV drivers with OSA; perhaps safe studies might be conducted in driving simulators.