look before you (s)leep: evaluating the use of fatigue detection technologies within a fatigue risk...

at SciVerse ScienceDirect

Sleep Medicine Reviews xxx (2013) 1e12

Contents lists available

Sleep Medicine Reviews

journal homepage: www.elsevier .com/locate /smrv

CLINICAL REVIEW

Look before you (s)leep: Evaluating the use of fatigue detectiontechnologies within a fatigue risk management system for the roadtransport industry

Drew Dawson a,*, Amelia K. Searle b,d, Jessica L. Paterson a,c

aCentral Queensland University, Appleton Institute, PO Box 42, Goodwood, SA 5034, AustraliabUniversity of Adelaide, School of Population Health, Centre for Traumatic Stress Studies, 122 Frome St, Adelaide, SA 5001, Australia

a r t i c l e i n f o

Article history:Received 7 December 2012Received in revised form1 March 2013Accepted 18 March 2013Available online xxx

Keywords:FatigueFatigue detectionFatigue risk management systemsFRMS

* Corresponding author. Tel.: þ61 8 8378 4517; faxE-mail addresses: [email protected] (D

adelaide.edu.au (A.K. Searle), [email protected] Tel.: þ61 8 8378 4517; fax: þ61 8 8378 4532.d Tel.: þ61 8 8313 0870; fax: þ61 8 8313 5368.

1087-0792/$ e see front matter Crown Copyright � 2http://dx.doi.org/10.1016/j.smrv.2013.03.003

Please cite this article in press as: Dawson D,risk management system for the road trans

s u m m a r y

Fatigue is a significant risk factor in workplace accidents and fatalities. Several technologies have beendeveloped for organisations seeking to identify and reduce fatigue-related risk. These devices purport-edly monitor behavioural correlates of fatigue and/or task performance and are understandablyappealing as a visible risk control. This paper critically reviews evidence supporting fatigue detectiontechnologies and identifies criteria for assessing evidence supporting these technologies.

Fatigue detection devices, and relevant reliability and validation data, were identified by systematicallysearching the scientific, grey and marketing literature. Identified devices typically assessed correlates offatigue using either psychophysiological measures or embedded performance measures drawn from theequipment being operated. Critically, the majority of the ‘validation’ data were not found within thescientific peer-reviewed literature, but within the quasi-scientific, grey or marketing literature.

Based on the validation evidence available, none of the current technologies met all the proposedregulatory criteria for a legally and scientifically defensible device. Further, none were sufficiently wellvalidated to provide a comprehensive solution to managing fatigue-related risk at the individual level inreal time. Nevertheless, several of the technologies may be considered a potentially useful element of abroader fatigue risk management system. To aid organisations and regulators contemplating their use,we propose a set of evaluative and operational criteria that would likely meet the legal requirements forexercising due diligence in the selection and use of these technologies in workplace settings.

Crown Copyright � 2013 Published by Elsevier Ltd. All rights reserved.

Introduction

Fatigue and driving impairment

Fatigue is a major safety issue in transport.1 In recent decades,various studies have documented a significant association betweenfatigue and increased risk of accident and injury.2 In road transport,estimates of the contribution of fatigue range from 10% to as high as60% of heavy vehicle crashes.3 Given the high risk posed by fatiguein the transport sector, and the relatively sedentary nature ofdriving, fatigue detection devices have typically been designed forand marketed to transport organisations.

For the purposes of this paper, fatigue is defined as ‘sleepinessresulting from the neurobiological processes regulating sleep and

: þ61 8 8378 4532.. Dawson), [email protected] (J.L. Paterson).

013 Published by Elsevier Ltd. All

et al., Look before you (s)leepport industry, Sleep Medicine

circadian rhythms’e that is, ‘the drive to sleep’.4e6 According to thisview, fatigue is influenced by three main factors: 1) prior sleep, 2)prior wake, and 3) time of day.2 First, numerous laboratory studieshave demonstrated that restricting sleep by small amounts (belowthreshold values of around five hours for one night or six hours overmultiple nights) results in cumulative daytime performance deficitson neuro-behavioural measures, including tracking, vigilance andreaction time.2,7,8 Second, fatigue increases monotonically from themoment of awakening. Research has shown that 17 h of sustainedwakefulness following a good night’s sleep produces performancedeficits (measured in the early morning hours) that are comparableto those seen at a level of 0.05% blood alcohol concentration.9 Third,fatigue and alertness follow a circadian (24-h) rhythm, with thehighest levels of fatigue generally seen in the early hours of themorning (e.g., 02:00 he06:00 h), and with a second smaller dip inthe early afternoon (e.g., 13:00 he16:00 h).10 Prior sleep, prior wakeand time of day interact to influence fatigue.2 Within the wakingperiod fatigue due to ‘time-on-task’ may also influence perfor-mance, however, this issue is beyond the scope of this paper.

rights reserved.

: Evaluating the use of fatigue detection technologies within a fatigueReviews (2013), http://dx.doi.org/10.1016/j.smrv.2013.03.003

mailto:[email protected]




www.sciencedirect.com/science/journal/10870792

http://www.elsevier.com/locate/smrv

http://dx.doi.org/10.1016/j.smrv.2013.03.003



Abbreviations

ANU Australian National UniversityASTiD advisory system for tired driversDSS driver state sensorEDVTCS engine driver vigilance telematic control systemEEG electroencephalographyFIT fitness impairment testerFRMS fatigue risk management systemGPS global positioning systemGSR galvanic skin resistanceMWT maintenance of wakefulness testNHTSA National Highway Traffic Safety AdministrationOCPT online continuous performance testOSLER Oxford sleep resistance testOSPAT occupational safety performance assessment testPERCLOS percentage of time that eyes are 80e100% closed

across a given time periodPVT psychomotor vigilance testSMS safety management systemUK United KingdomUS United States

D. Dawson et al. / Sleep Medicine Reviews xxx (2013) 1e122

Fatigue risk management controls

Quantifying and controlling fatigue-related risk requires amulti-faceted approach. One of the most common multi-factorialapproaches, the ‘defences-in-depth’ model,11 outlines a five levelmodel of hazard control. According to this approach, a fatigue-related incident (level 5) is the consequence of a fatigue relatederror (level 4), which is typically preceded by the signs andsymptoms of fatigue (level 3). An individual exhibiting the signsand symptoms of fatigue has typically had insufficient sleep (level2), which may be due to an insufficient sleep opportunity (level 1).Because a fatigue-related accident is a relatively low frequency/high consequence event, an effective fatigue risk managementsystem (FRMS) needs to focus on identifying lead indicators that arehigh frequency/low consequence events. By focussing on high fre-quency lead indicators, risk may be identified more effectively andcontrols implemented at all four levels of this risk trajectory. Thismay prevent fatigue-related incidents more effectively.11

Most organisations implement controls at the first and/or sec-ond levels of the defences-in-depth hierarchy.12 Specifically, orga-nisations may employ ‘hours-of-service’ and ‘rules-of-rostering’ toensure an adequate sleep opportunity between shifts. Some com-panies also employmathematical fatiguemodelling tools13 to assistrostering, for the same purpose. Organisations may also monitordriving hours through the use of highway surveillance cameras andactual sleep obtained using sleep diaries or wrist actigraphy mon-itors. However, first and/or second level controls may not be suf-ficient to prevent fatigue-related incidents. Organisations cannotalways guarantee that employee working time arrangements arecompliant with policy or that self-report sleepewake data sup-porting ‘fitness-for-duty’ policies are reliable.

Thus, there is a potential benefit for fatigue-detection technol-ogies that identify fatigued workers and/or notify an organisation,or the workers themselves, when fatigue-related risk has reachedan unacceptable level. These technologies are typically designed todetect behavioural indicators of fatigue (i.e., a level 3 control).Several technologies are already in use in the transport, health andmining industries. Devices may be based on neurobehavioural andphysiological correlates of fatigue (e.g., reaction time or frequency,duration and rate of eye closures), or embedded performance

Please cite this article in press as: Dawson D, et al., Look before you (s)leeprisk management system for the road transport industry, Sleep Medicine

measures (e.g., vehicle dynamics such as variability in velocity orsteering lane position). While fatigue detection devices may oftenbe marketed as effective solutions for managing fatigue-relatedrisk, there is currently little systematic evidence regarding theirscientific reliability or validity or legal defensibility. There are nocurrent regulatory guidelines regarding the appropriate use ofthese technologies and how they contribute to the effectiveness ofan FRMS.

Scope of review

This report aims to critically review currently available andemerging fatigue technologies (level 3 controls). There are severalexisting reviews on fatigue detection devices, which vary inscope.14e20 We update this literature, to include new devices andemerging research, and to exclude devices that are no longercommercially available. Our scope is purposely broader than mostpre-existing reviews. That is, to address issues regarding the legaland scientific defensibility of fatigue detection technologies andtheir role within the broader context of a multi-faceted FRMS.

Method

Literature was obtained by searching 1) academic search engines(e.g., ISI, PsycInfo, Google scholar), 2) government/industry websites(e.g., AustRoads, Australian Transport Safety Bureau, TransportResearch International Documentation, Australian Road ResearchBoard, US and Europeanwebsites including USDOT, Federal RailroadAdministration FMCSA), and 3) a substantial online ‘grey literature’,using the search terms fatigue, sleepiness, drowsiness, alertness,detection,monitor,management, technology, and countermeasure (andvariations thereof).

Key academic peer-reviewed journals (e.g., TransportationResearch Part F, Accident Analysis and Prevention, and Safety Science)were also searched. References within relevant articles were sub-sequently obtained. Finally, companies that developed fatiguedetection devices were contacted directly and asked to providetechnical specifications, reliability and/or validation data.

Ideally, a legally and scientifically defensible fatigue detectiondevice would be capable of measuring fatigue and performance inan individual in real time, as well as predicting future fatiguelevels.21 It must be valid (by measuring a fatigue sensitive behav-iour such as blink velocity), reliable (doing this consistently, asemployees and managers may come to depend on it), sensitive(predicting unacceptable fatigue levels, and minimising missedevents), specific (minimising false alarms, as drivers may thendistrust the device) and generalisable (to all users, by accountingfor individual differences).22 Sensitivity is especially important, asthe device must be able to detect signs of fatigue that precede theoccurrence of fatigue-related incident, so that appropriate coun-termeasures can be put in place. Devices must also demonstrate“operational validity”,23 in that it is desirable that they are suitablyrobust and reliable for use in industrial settings; they must collecthigh quality data with minimal interference (sweat, sunlight etc.),be portable, minimally intrusive and accepted by users.

Evidence of the devices’ capabilities should be demonstrated inlaboratory and field studies, using large samples and undertaken in asample of the population of interest (e.g., heavy vehicle drivers).Devices should discriminate between normal alertness and fatigue,resulting from either partial or total sleep restriction, long shifts,time of day, or a combination of these factors. Ideally, performance ofthe technology should be compared to several other fatigue-detection devices, including the current gold standard, the psycho-motor vigilance test (PVT),24 as well as to real or simulated taskperformance. Devices must be validated by independent third


D. Dawson et al. / Sleep Medicine Reviews xxx (2013) 1e12 3

parties as well as by test developers, with results essentially repli-cated. Finally, this evidence must be published in peer-reviewedjournals. Each device was reviewed in light of these criteria.

Results

Discussions of fatigue detection technologies are presentedbelow and are grouped according to the nature of the device(fitness-for-duty tests, continuous operator monitoring and per-formance based monitoring). Within these categories, each deviceis grouped according to the fatigue correlate measured (e.g., neuro-behavioural performance, pupilometry). Issues common to mostdevices will be addressed more generally within the Discussion, toavoid repetition. Validity evidence for each device is summarised intables, with ratings in each category based on both the quality andquantity of scientific evidence available (given that, as previouslymentioned, both high-quality and replicable evidence is needed todemonstrate the defensibility of a device). It is worth noting thepossibility of a bias against the publication of studies with negativefindings. This limits our ability to compare devices based solely onthe number of published reports.

Fitness-for-duty tests

These tests are typically performed before an employee com-mences work, to determine if their current alertness level is suffi-ciently safe to work the duration of the shift. Generally these testsassess performance on neuro-behavioural tasks, particularly exec-utive functions such as vigilance or handeeye coordination.

These devices are generally computer-based, relatively portable,take up to 10 min, are unlikely to interfere with the work task andhave short or no practice/learning effects. However, these tests onlygauge fatigue at time of testing and not during shifts. There iscurrently no evidence as to whether these tests can or cannotpredict subsequent fatigue over the course of a shift. Thus, todetermine if fatigue has since increased to a level where it is nolonger advisable to continue working, drivers would need to peri-odically re-test themselves. This may not always be feasible withinthe road transport industry, especially if companies use non-portable devices, and drivers are not near the check-in depotsthat house them. Thus, research is needed regarding how far inadvance ‘fail’ scores on these tests can predict the likelihood offatigue related accidents. It is also possible that stimulant use (e.g.,caffeine) may improve performance in the short term andmake thetest-taker appear less fatigued than in reality.19 Table 1 summarisesthe validity evidence available for each fitness-for-duty device.

Neuro-behavioural performancePsychomotor vigilance test (PVT). The psychomotor vigilance testassesses sustained attention and is currently the gold standard in

Table 1Fitness for duty tests.

Device Developervalidation

Independentvalidation

Laboratorystudies

Fieldstudies

Tested against afatigue measure

PVT **** **** **** *** e

OSPAT e ** ** * **OCPT ** e e * e

EyeCheck e * * * *FIT e *** *** ** **Safety Scope e ** ** e e

**** ¼ Strong evidence; *** ¼ good evidence; ** ¼ weak evidence; * ¼ little evidence; eNote: Device ratings were based on two criteria, considered in conjunction: 1) the qualityhighly if it had been tested in multiple published scientific studies with poor results, noAbbreviations: EEG, electroencephalography; FIT, fitness impairment tester; OCPT, onlinetest; PVT, psychomotor vigilance task.


fatiguedetection.24 Individuals give a button-press response to visualstimuli on a computer screen over a 5- or 10-min period. Reactiontime and ‘lapses’ (response time �500 ms) are measured. Severalcommercial versions of the PVT are available (e.g., PalmPVT, http://www.corware.com/Default.aspx; PVT-192: Ambulatory Monitoring,http://www.ambulatory-monitoring.com/pvt192.html). The PVT hasconsistently and reliably detected performance decrements resultingfrom sleep restriction, extended wakefulness or time-of-day effectsacross numerous industries, including rail, aviation, mining, anddefence. Validation evidence is available for both the 5-min25e28 and10-min versions.7,8,24,29e32 However, PVT performance may notequate to poor driving performance, perhaps due to large individualdifferences in driving performance. Thus the PVTmay be better usedto predict fatigue-related driving incidents as part of a broader testbattery.33 This possibility may be explored further through researchcomparing the PVT with driving performance. Cut-off scores thatpredict impaired driving performance have also not been estab-lished. Despite its promise, until ‘fail’ or ‘warning’ levels are estab-lished, this device is unlikely to be useful for commercial drivers.

Occupational safety performance assessment test (OSPAT Pty Ltd.,Western Australia, http://ospat.com/). The OSPAT is a computer-based, unpredictable tracking task lasting 60e90 s assessing reac-tion time, sustained attention, and handeeye coordination. Thetask becomes progressively more difficult if the test-taker provesproficient. A baseline performance level is recorded for eachparticipant, to account for individual differences. At the end of thetask, participants are given a score of ‘pass’, ’caution’ or ‘alert’. TheOSPAT is currently being used by mining, transport and metalprocessing industries in Australia, Brazil, New Zealand andMalaysia. However, no data have been released from the organi-sations currently using the OSPAT.27

OSPAT performance has been demonstrated as sensitive to fa-tigue during 24 h of prolonged wakefulness within laboratoryconditions, and has been validated for sleep deprivation.9,27,34

OSPAT performance also demonstrated a moderate positive corre-lation with the PVT but was not related to subjective sleepiness.27

The only published field study using the OSPAT found non-significant correlations with subjective fatigue, suggesting theOSPAT may be valid in well-controlled laboratory conditions butless so in workplace settings.35 As this device (being a desktop-mounted computer) is not portable, it is currently not well suitedto regular use within the road transport industry. This is becausedrivers often do not start their shift from bases or depots, butinstead from roadhouses and highway truck stops.

Online continuous performance test (OCPT; eAgnosis Inc., Newark,DE, http://www.checkadhd.com/onlineCPTresearch.php). This test issimilar to the PVT, measuring sustained attention and vigilanceusing a 19-min internet-based task. However, the task is more

n established(e.g., PVT, EEG)

Tested againstdriving performance

Sensitivity/specificity End useracceptability

* e e

e e e

e e e

* e e

* e e

e e e

¼ no evidence., and 2) the quantity of validation evidence available. Thus, a device would not rater if it was tested in only one scientific study with excellent validation evidence.continuous performance test; OSPAT, occupational safety performance assessment


http://www.corware.com/Default.aspx

http://www.corware.com/Default.aspx

http://www.ambulatory-monitoring.com/pvt192.html

http://ospat.com/

http://www.checkadhd.com/onlineCPTresearch.php


complex than a simple visual stimulus response task and requiresparticipants to respond to triangles, and ignore circles, presentedon-screen in a pseudo-random manner. The only study to examinethis device found that errors of omission, but not response time,significantly increased following sleep restricted to approximately3e4 h.36 This test needs significant further investigation before itsuse in an industrial setting.6

PupilometryThis technology assesses the pupils’ involuntary response to

flashes of high-intensity bright light, a demonstrated biomarker offatigue.37 Specifically, the pupil shows longer constriction latencyand slower constriction velocity when an individual is fatigued.These parameters are typically assessed using bench top mountedor portable binocular-style instruments, where participants lookinto an eyepiece for a short time period (i.e., one to two minutes).Advantages of these devices are their quick administration andalerting time, the lack of training needed and the lack of learningeffects. Additionally, as pupillary responses are involuntary, the testcannot be ‘gamed’ in the same way performance tests can be.However, pupilometry may not be suitable for people with certainhead/eye conditions, or for people over 50 y, due to age-relatedchanges to pupil size and response.38 This point is speculative, asnone of the pupilometry validation studies (detailed below) havesystematically tested devices within these populations.

EyeCheck (MCJ Inc., Rockford, IL, http://www.eyecheck.com/).This hand-held device measures pupil diameter and constrictionlatency. The EyeCheck has been trialled by the United States po-lice as a detector of illicit drug use and fatigue in motorists, aswell as by an Australian mining company.38,39 There is no pub-lished peer-reviewed validity evidence for the EyeCheck. The testdevelopers demonstrated that EyeCheck scores significantlydistinguished between truck drivers who reported getting five orless hours sleep the previous night.40 However, neither self-reported fatigue nor hours of sleep were correlated with Eye-Check scores. Richman38 found that EyeCheck results predictedthe fatigue resulting from 18 h of continued wakefulness with asensitivity of 82% and specificity of 94%. In another study, how-ever, EyeCheck scores did not vary across 24 h of sustainedwakefulness, and were unrelated to PVT scores and drivingsimulator performance.41 Thus, the limited results regardingEyeCheck’s effectiveness are equivocal.

Fitness impairment tester (FIT; Pulse Medical Instruments Inc., Rock-ville, MD, http://www.pmifit.com/). The FIT measures both eyetracking (saccadic velocity) and pupilometry (pupil diameter,constriction amplitude and latency), with total FIT scores calculatedfrom these four parameters. These scores are relative, as they arealways compared to an individual’s baseline performance. Total FITscores have only been tested in three of the 10 existing validationstudies; most studies instead test the individual FIT parameters.Total FIT scores, as well as components of saccadic velocity andpupil constriction latency, have shown increases in response tosleep deprivation and subjective sleepiness in healthy volun-teers.42e49 These parameters have also been found to be signifi-cantly higher after finishing long work shifts, and during nightshifts, among small samples of truck drivers and defence pilots.50,51

Further, saccadic velocity and, to a lesser extent, constriction la-tency, have also been significantly associated with PVT lapses,driving simulator accidents, and performance on several neuro-cognitive tests.42e44,46e49 The parameters of pupil diameter andconstriction amplitude are not sizeably or consistently associatedwith fatigue, which may be at least partly due to their sensitivity toambient light and time of day effects.42


Greater laboratory and field research is needed to examine thepredictive validity of the total FIT scores, as ultimately theweightedalgorithm dictates when users are given warnings, and only two ofits four constituent parameters (saccadic velocity and constrictionlatency) are consistently related to fatigue. Of the few studiesconducted, it appears that the remaining ‘noise’ in the total FITscore has not compromised its validity.42,51 However, moreresearch is needed to verify this finding. Additionally, results ofsome studies suggest that the FITmay not be useable among peoplewith excessive blinking and/or thick corrective lenses.44,49 Thus,while potentially useful for the road transport industry, this deviceis currently limited in scope, as it cannot be assumed that it can beused successfully among the ageing demographic of the workforce.

Safety Scope (Torrance, CA, http://www.eyedynamics.com/index.htm).The Safety Scope measures pupillary reflex and eye movementparameters. Only one study has examined this device as a fatiguedetector. The proof-of-concept pilot study found that Safety Scoperesults were sensitive to fatigue produced by extended wakeful-ness, with all ‘fail’ results occurring either in the early morning, orin the early afternoon circadian dip.52 No sensitivity or specificitydata were provided against another technology. Given the absenceof any information found on the Safety Scope following this pilotstudy, this device seems to have ceased being used eithercommercially or for research purposes.

Continuous operator monitoring

These devices continuously monitor physiological correlates offatigue during work, including eye movements, brain waves, heartrate, posture, and galvanic skin response/electro dermal activity.Evidence for these devices is summarised in Table 2.

Oculomotor measurementTechnologies based on eye movements are by far the most

common, with several different products available. These productsmeasure the frequency, duration and/or rate of eye closure. Thesevariables have been linked to electroencephalography (EEG) andother measures of fatigue.53,54

Eyemovement data are typically obtained in one of twoways: 1)the structured illumination approach, which measures the reflec-tion of infrared light off the retina, and 2) facial feature recognition,which detects eye movement by tracking a set of facial featuresincluding the eyelids, iris, nose, and mouth.55 Some devices mea-sure ‘PERCLOS’e the percentage of time that a driver’s eyes are 80e100% closed across the measurement period.56,57 PERCLOS scoreshave been demonstrated as positively correlated with PVT lapses,as well as task-based measures associated with fatigue such as lanedepartures and subjective sleepiness. PERCLOS scores performwellagainst EEG activity and both eye-blink and head nodding tech-nology.23,58e60 However, devices that use PERCLOS alone may notidentify all fatigued drivers, as a significant subset of subjectsexperience micro-sleeps with eyes open.18,61 Other devices mea-sure various other aspects of eyelid closure, including saccadicvelocity, and blink velocity and duration.61 Advantages of oculo-metric devices are that they are relatively non-invasive and feasiblefor use within static operator environments such as truck cabs.However, many of the devices using infrared reflectance are limitedby factors such as ambient light and the use of prescription glassesor sunglasses. Further, some camera devices are unable to adjustappropriately to gross body movements.

Optalert (http://www.optalert.com/). Optalert uses infrared reflec-tance oculography to detect blink frequency, velocity and duration.Small sensors and light emitting diodes are mounted on the bottom


http://www.eyecheck.com/

http://www.pmifit.com/

http://www.eyedynamics.com/index.htm

http://www.optalert.com/

Table 2Continuous operator monitoring tests.



Laboratorystudies

Fieldstudies

Tested against an establishedfatigue measure (e.g., PVT, EEG)



Optalert *** *** *** ** **** *** *** *CoPilot DD850 e ** ** * e e e *Seeing Machines DSS e ** ** e e ** e e

Eye-Com e e e e e e e e

Smart Eye e e e e e e e e

Smart Cap e ** ** e *** e ** **B-Alert *** e *** e *** *** e

NapZapper e * * e e * e e

StayAwake e e e e e e e e

Driver Fatigue Alarm e e e e e e e e

NoNap e e e e e e e e

Dozer’s alarm e * * e e e e **MicroNod e e e e e e e e

TravelMate e e e e e e e e

StayAlert e e e e e e e e

EDVTCS e * * e * * e e

**** ¼ Strong evidence; *** ¼ good evidence; ** ¼ weak evidence; * ¼ little evidence; e ¼ no evidence.Note: Device ratings were based on two criteria, considered in conjunction: 1) the quality, and 2) the quantity of validation evidence available. Thus, a device would not ratehighly if it had been tested in multiple published scientific studies with poor results, nor if it was tested in only one scientific study with excellent validation evidence.Abbreviations: DSS, driver state sensor; EDVTCS, engine driver vigilance telematic control system; EEG, electroencephalography; PVT, psychomotor vigilance task.


of normal spectacle frames, pointing directly at the wearer’s eye. Amicroprocessor is housed in the glasses arm, and a cable connectsthe glasses to a dashboard-mounted processing unit. Aural andvisual warnings occur at scores of 4.5 (cautionary) and 5 (critical).61

According to the manufacturer, Optalert has been validated bythe developers and several other research laboratories, and hasbeen tested in the field among small samples of truck drivers, andnurses on nightshift.62,63 Fatigue scores are significantly higherfollowing acute and partial sleep deprivation, higher at the circa-dian nadir, and correlate significantly with PVT lapses, slow eyemovements, EEG theta power, response time on a hazard percep-tion task, lane departure on driving simulators, and PERCLOS scoresderived from the CoPilot device (which is described in detailbelow).61e70 Scores have also been demonstrated as sensitive to thealertness-promoting effects of caffeine.71

AlthoughOptalert cut-off scoreswere developedwith a sample ofonly eight people, an independent study found these scores to havemoderate sensitivity (69e75%) and reasonable though lower speci-ficity (43e75%) in predicting large lane departures in the 15 minfollowing an alert. Based on the initial study, a score of 4.5 predicted96%ofoff-roadevents occurringwithin15minof thewarning, andnooff-road events occurred without a warning.68 However, this devicemaynotperformoptimally forpeoplewithsleeporvisualdisorders. Italso cannot be worn with corrective glasses, although prescriptionscan apparently be fitted to the Optalert lenses.

Optalert has undergone trial implementations in industriesincluding aviation, mining and road transport.72,73 There is someevidence to suggest that users will tolerate the Optalert system quitewell, and that the glasses are comfortable and non-invasive. Only asmall number of users reported the connecting cable as annoying,and the glasses frames as limiting their visual field.61,63,68,73 Giventhat the warnings are designed to predict driving impairment15min in advance, resulting in relatively low specificity, it is possiblethat some drivers may perceive Optalert as having low face validityandmay disregard the high frequency of warnings. One recent studyreported that data quality degraded significantly between laboratoryand driving simulator testing, due to factors including long eyelashesand deviant blink behaviour.63 We are unaware of any equivalentfield-based studies.While this device is currently available, validatedand appears to have utility for the road transport industry, itmay notbe suitable for the significant proportion of the truck driver popu-lation suffering from sleep disorders.74


CoPilot/DD850 (Attention Technologies Inc., 2005 e http://www.driverfatiguemonitor.com/). These dashboard-mounted camera de-vices use structured illumination and are second-generation com-mercial prototypes developed by the original PERCLOS researchersfrom Carnegie Mellon University. Visual and auditory warnings aregenerated once a fatigue threshold is reached. Only one validationstudy exists. In this study, CoPilot scores significantly decreasedover 24 h of sustained wakefulness and correlated with bothdriving simulator crashes and PVT lapses.65 Both first- and second-generation prototypes have been trialled in field studies of long-haul truck drivers, but information relating to validity (e.g., speci-ficity and sensitivity) has not been presented.75,76 In field trials,drivers gave CoPilot neutral to low acceptance ratings and indicatedthey would not choose for the device to be installed in their truck.This device can only be used under low ambient light levels, and sois only suited to night-time driving. Additionally, the device wasshown to have a high false alarm rate (65%) when subjects werecompletely motionless, suggesting it is highly sensitive to move-ment.77 As this device has not performed well in uncontrolled fieldconditions, it is currently not recommended by the NationalHighway Traffic Safety Administration (NHTSA).23,78

Seeing machines driver state sensor (DSS; Canberra, Australia, www.seeingmachines.com). This dashboard-mounted device uses eyemovement analysis and was developed through the AustralianNational University (ANU) in Canberra. This device is currentlybeing trialled in Asian and American mining companies. The DSSuses facial tracking and absolute eyelid position rather thaninfrared reflectance and can be used in low light conditions, as wellas with prescription glasses or sunglasses. This device also claims tocope with head movements and partial occlusion of the face. Aswell as emitting auditory warnings, the DSS can transmit alerts to acentral dispatcher.55,79 The only validation evidence available forthis device comes from an industry report. Scores on the DSS werefound to increase across the early morning hours during anextended wakefulness protocol, and showed strong correlationswith driving performance and subjective sleepiness.20

Several other eye-monitoring technologies exist, including Eye-Com (http://eyecomcorp.com/), and SmartEye (SmartEye AB, Gothen-burg, Sweden, http://www.smarteye.se/). However, technical infor-mation or validation evidence could not be found for either of thesedevices.


http://www.driverfatiguemonitor.com/

http://www.driverfatiguemonitor.com/

http://www.seeingmachines.com

http://www.seeingmachines.com

http://eyecomcorp.com/

http://www.smarteye.se/


Electroencephalography (EEG)Changes in brain wave activity have been shown to be valid

biomarkers of fatigue. Specifically, theta (3.5e7 Hz), alpha (8e13 Hz) and beta (14e30 Hz) activity, and to a lesser extent deltaactivity, significantly increase due to fatigue.80e82 Historically, EEGmeasurement of fatigue has required time-consuming and intru-sive EEG sensor application, which is not feasible for road transportoperations. However, two portable devices that do not require scalppreparation have been recently developed.

Smart Cap (Co-operative Research Centre for Mining, http://www.smartcap.com.au/). This technology is not yet commercially avail-able, with the first production version released across an entireAustralian mine site (with approximately 200 miners) in late 2011.EEG sensors are embedded under either a baseball cap with sweat-band, or a headband, with an electronic processor card attached.Information iswirelessly transmitted to a small dashboard-mountedmonitor, and can transmit to a central dispatcher for remote moni-toring. Threshold scores were derived from the Oxford sleep resis-tance test (OSLER). Smart Cap alarms at a threshold of four, which isthe equivalent of four consecutive OSLER lapses and indicative ofmicro-sleeps. An alarmwill also sound if the cap is connected but notbeing worn for an extended period. Preliminary (unpublished)validation results by Monash University researchers suggest highsensitivity (.95) and specificity (.82) for detecting four consecutivelapses on the OSLER test (Kevin Greenwood, personal communica-tion). Further research is examining whether the Smart Cap cansensitively detect lower levels of fatigue (KevinGreenwood, personalcommunication). Anecdotally, users have reported the Smart Cap iscomfortable and easy to wear (Kevin Greenwood, personalcommunication). It should be noted that Kevin Greenwood is amember of the Smart Cap commercialisation team.

B-Alert (Advanced Brain Monitoring Inc., http://www.b-alert.com/).EEG sensors are embedded in a headband-like device, with infor-mation transmitted by radio waves via a small unit attached at theback of the head. EEG signals are classified into one of four levels ofalertness (sleep onset, relaxed wakefulness, low engagement, highengagement), derived from cognitive and maintenance of wake-fulness tests (MWT) among alert and sleep-deprived subjects.83,84

Validation studies by the device developers show that B-Alertlevels significantly decrease as a function of prolongedwakefulnessand are positively correlated with PVT lapses, subjective sleepiness,objective sleepiness on MWT, memory test performance, drivingsimulator performance, human scored EEG activity and observedsleepiness (e.g., facial expressions, head nodding) in sleep-deprivedhealthy subjects.84e89 It has also discriminated alertness levelsbetween healthy subjects and sleep apnoea sufferers.89 As thisdevice was developed for clinical research and not industry use,there is no in-cab monitor available and EEG sensors must be re-applied every four hours. According to the manufacturers, there ispotential to modify this device to suit industry requirements.

Posture/head noddingThese devices detect the posture changes that accompany fa-

tigue, in particular, head nodding as the neck muscles relax.Generally, they come in the size and shape of a hearing-aid which isworn behind the ear. An electronic angular position sensor detectswhen the head nods forward and an alarm will sound when thehead nods to a pre-set angle, generally around 15 degrees. Thesedevices are cheap and relatively non-intrusive. However, they are alate warning device, as they only detect fatigue at around the pointof sleep onset, when a fatigue-related incident is already likely tohave occurred.14,16,90 Thus, they are arguably not sensitive enoughfor use in the road transport industry. Several devices are on the


market, including the Nap Zapper (WELKIN Electronics Co. Ltd,http://zjwelkin.en.ec21.com/Nap_Zapper_Anti_drowsy_Alarm–

659656_659831.html), the Stay Awake (http://stayawakedevice.com/), the Driver Fatigue Alarm (www.driverfatiguealarm.com.au)and the NoNap (www.thenonap.com). There are several other de-vices which may not still be commercially available including theDozer’s Alarm,91 MicroNod detection system (MINDS), TravelMateand the Stay Alert (http://www.inventionconnection.com/BOOTHS/booth182.html). However, there is no validation evidence availablefor these devices. One small study tested the Dozer’s Alarm andfound the alarms on the device were not related to the number oftracking errors.91 Additionally, a field study reported the NapZapper suffered from a high rate of false alarms as a result of normalhead movement during driving tasks, with only 1% of alarmsactually predicting fatigue-related incidents.20

Galvanic skin resistance (GSR)Galvanic skin resistance (GSR) has been demonstrated to increase

with fatigue.92 However, GSR also varies in response to other in-fluences such as stress and sweat.17,30 Thus, it is not yet clear if GSR isable to identify changes that are specific to work-related fatigue.

Engine driver vigilance telematic control system (EDVTCS; Neurocom2002, http://www.neurocom.ru/en2/product/edvtcs.html). Thisfinger- or wrist-worn device contains a sensor that measures GSR.This information is sent to a dashboard-mounted unit, and an alarmsounds when alertness levels fall below a critical level, reportedlyone minute prior to sleep onset. This device has been used inRussian railway operations.93,94 Although the test developersreport that it has been extensively validated, and associated withEEG activity, there is no published evidence to verify this asser-tion.93 The only peer-reviewed published validation showed thatGSR levels did not significantly change across a 28-h period ofsustained wakefulness despite changes in subjective sleepiness,driving simulator performance and PVT performance.30 Further, alarge degree of data dropout from the wrist-worn device wasobserved, which had to be used in combination with the finger-worn version (with a smaller degree of dropout) to ensure validdata were obtained.30 Data quality may be reduced even further inreal-world conditions, but this is currently unclear as no field trialshave been published.

Performance based monitoring

These devices monitor performance indicators that are associ-ated with fatigue-related driving incidents. Validation evidence forthese devices is summarised in Table 3.

Embedded performance measuresThere has been a recent shift toward embedded performance

measures. Embedded performance measures monitor task perfor-mance directly and identify performance impairment consistentwith operator fatigue. These measures have the advantage of beingrelatively non-intrusive, embedded within the actual task (e.g.,steering or speed variability) and have high face validity, as theydirectly measure behaviours critical to job performance and safety.It is possible that drivers will be more accepting of these devicesthan of physiological measures, as the focus of surveillance is on thetask and not the employee.95

Multiple studies (using laboratory-based driving simulators,and not actual marketable devices) have found lane variability to beone of the more sensitive driving-related measures of fatigue.96,97

Specifically, fatigued drivers are found to show greater variabilityin their steering and lateral lane position, showing fewer micro-corrections and more macro-corrections in their steering.96 Lane


http://www.smartcap.com.au/

http://www.smartcap.com.au/

http://www.b-alert.com/

http://zjwelkin.en.ec21.com/Nap_Zapper_Anti_drowsy_Alarm--659656_659831.html

http://zjwelkin.en.ec21.com/Nap_Zapper_Anti_drowsy_Alarm--659656_659831.html

http://stayawakedevice.com/

http://stayawakedevice.com/

http://www.driverfatiguealarm.com.au

http://www.thenonap.com

http://www.inventionconnection.com/BOOTHS/booth182.html

http://www.inventionconnection.com/BOOTHS/booth182.html

http://www.neurocom.ru/en2/product/edvtcs.html

Table 3Performance based tests.



Laboratorystudies

Fieldstudies

Tested against an establishedfatigue measure (e.g., PVT, EEG)



SafeTrak e e e ** e e e **MobilEye e e e e e e e e

AutoVue e e e e e e e e

Delphi e e e e e e e e

Roadguard e * * e e e e e

ASTiD e e e e e e e e

**** ¼ Strong evidence; *** ¼ good evidence; ** ¼ weak evidence; * ¼ little evidence; e ¼ no evidence.Note: Device ratings were based on two criteria, considered in conjunction: 1) the quality, and 2) the quantity of validation evidence available. Thus, a device would not ratehighly if it had been tested in multiple published scientific studies with poor results, nor if it was tested in only one scientific study with excellent validation evidence.Abbreviations: ASTiD, advisory system for tired drivers; EEG, electroencephalography; PVT, psychomotor vigilance task.


departure warning devices consist of small forward-looking cam-eras that are dashboard-mounted, facing the windshield and theroad ahead. Algorithms interpret lane markings and other roadfeatures captured on video, including white lines, and alert driversto unintentional road departures (e.g., when not using the indicatorsignal).98 While these devices are generally robust to different roadand weather conditions, they do not cope well with rainy night-time driving, due to increased road reflectance or with non-bitumen surfaces in rural and remote settings.

Devices currently available as options in commercial vehicles inthe United States (US), the United Kingdom (UK) and Europe includeSafeTrak (AppliedPerceptionandAssistWareTechnology, Inc., http://www.safetrak.takata.com/), MobilEye (www.driverfatiguemonitor.com), AutoVue (Iteris Inc., www.iteris.com/av/passenger.html), andDelphi (www.delphi.com). No validation evidence was found forthese devices. SafeTrak has been trialled in a large sample of US andCanadian truck drivers who considered the device to be under-standable, while 42e69% agreed that the device helped them drivesafely.99 There were no published validity or reliability data.

Subsidiary tasksSome devices assess subsidiary tasks donewhilst driving, such as

vigilance and reaction time tests. For example, devices that involvepressing a button in response to a randomly presented stimulus(usually a flashing light) are used within the rail and mining in-dustry.17,26,30 However, subsidiary tasksmay be considered intrusive,given the continuous engagement required by the driving task.Drivers may easily circumvent this device by automatically pressingthe button at random intervals rather than as a stimulus response,which can be done under low alertness levels and possibly evenduring micro-sleeps.14,41 One small study evaluated the Roadguarddevice, which is fitted to truck dashboards and turns on afterreaching top gear.91 In the five subjects studied, reaction time on theRoadguard increased significantly and the device gave more warn-ings with increasing time on a driving simulator. The five subjectsrated this device as moderately annoying, and more annoying thanglasses-mounted oculomotor and head nodding devices.91 Clearly,with such limited evidence available, significantly more supportingevidence is needed for this device to be considered acceptable foruse in the road transport industry.

Combined measures

Advisory system for tired drivers (ASTiD; http://www.fmig.org/astid.html)

The ASTiD was developed from research conducted atLoughborough University. The dashboard-mounted device as-sesses sleep history, current driving conditions and steeringdynamics. Based on users’ input of their prior sleep, time of dayand driving time, a fatigue modelling algorithm calculates thelikelihood of falling asleep. If users do not enter prior sleep


information, default levels are assigned by the manufacturer. Thealgorithm also considers monotonous driving conditions andfine and gross steering wheel movements, including the numberof total movements and large corrections. Drivers receiveaudible and visual alarms once a fatigue threshold has beenreached.20

The ASTiD has been trialled in several mining and road transportcompanies with over 1000 h of field data collected.17,20 Initial re-sults were reportedly promising and drivers considered that ASTiDalerts were consistent with their subjective fatigue levels.17,20

However, there is currently no published validation evidence forthe ASTiD.

Discussion

In principle, current fatigue detection technologies promise apotentially objective approach to the detection andmanagement offatigue. In reality, the devices described above can be considered‘developmental’ at best. Indeed, only one device (Optalert) had anavailable evidence base addressing all assessment criteria. In manycases, organisations have adopted and implemented these tech-nologies with little consideration of the consequences of riskmitigation based on a technology that is not yet legally and/orscientifically defensible.

The devices identified vary considerably in the quality of vali-dation data. Several devices were commercially available despitehaving little or no evidence of scientific effectiveness (e.g., headnodding and lane tracking devices; EyeCheck, Safety Scope,EDVTCS). Other devices were developed using lab-derived corre-lates of fatigue (e.g., ASTiD, CoPilot), but had not yet demonstratedthat they could validly detect fatigue in a field-based setting.Finally, only a few devices have been evaluated by independentthird party researchers (e.g., FIT, OSPAT, Optalert). In addition,several devices that had not yet been commercially releasedshowed promising preliminary results (e.g., Smart Cap, B-Alert) butappear to be moving rapidly to implementation without sufficientvalidation and reliability studies.

Importantly, none of these devices met all predefined criteriafor a legally and scientifically defensible device.19,21e23 First, thesample size, demographics and criteria used to initially derivedevices’ warning scores were rarely apparent. If large andrepresentative samples were not used, subsequent warninglevels may not apply to all users. These studies were also oftenlimited by small samples of mostly young and healthy volun-teers. Further research with larger and more varied samplesmust be conducted in order to verify any initially promisingresults. Relatedly, studies must demonstrate whether these de-vices can be used reliably in drivers with certain health issues,including eye problems or sleep apnoea, both of which arearguably more common in heavy vehicle drivers than in thegeneral population.74


http://www.safetrak.takata.com/

http://www.safetrak.takata.com/

http://www.driverfatiguemonitor.com

http://www.driverfatiguemonitor.com

http://www.iteris.com/av/passenger.html

http://www.delphi.com

http://www.fmig.org/astid.html

http://www.fmig.org/astid.html


Even the devices that were supported by independent peer-reviewed research were limited in two key ways: 1) even whenvalidated in lab-based or simulator settings they had rarely beentested in the field, and 2) warning levels had not been tested todetermine if they could change behaviour in ways that led toimproved safety. These limitations have significant implications forthe ultimate effectiveness of these devices. First, field studies arecrucial for determining not only whether devices can validly detectfatigue in uncontrolled environments (e.g., truck cabs), but alsowhether they can be feasibly implemented within a specific in-dustry. No matter how well the technology may detect fatigue, if itis not well tolerated by workers, then it may not be used correctlyor regularly and ultimately may not be effective. As an additionalsafeguard, all real-time technologies might reasonably be expectedto alert supervisors if they are not being used correctly (i.e., turnedoff, muted, or not being worn). Second, regardless of whetherscores on devices have been demonstrated to co-vary with fatiguelevels, the sole cue available to drivers is the pre-defined alarmlevel. Thus, drivers must be able to be confident that the device willconsistently alert them before a fatigue-related incident occurs.This leads to the issue of how far in advance of a fatigue-relatedincident a driver should be alerted.14,19 The criteria used in thefew studies that examined sensitivity/specificity of warningsincluded full lane deviation on a simulator, and four consecutive‘lapses’ e both of which suggest the presence of micro-sleeps. Toprevent fatigue-related incidents, devices must alarm well beforemicro-sleeps occur.

Overall, this review reveals a trend for developers to releasefatigue-detection technologies prior to thorough scientific exami-nation of their effectiveness.6 To a certain extent this practice isunderstandable. The companies developing and marketing thesetechnologies are usually small and under-capitalised. There is sig-nificant pressure to produce sales and it can take a long time tobuild an evidence base that would ensure the technology wasscientifically and legally defensible. However, this ‘rush to market’may hurt developers in the long term. This may be particularlydamaging if accidents or fatalities occur where these devices are inuse and organisations cannot demonstrate due diligence in theirpre- and post-implementation assessments.

A final issue for these technologies is the question of therelationship between an estimated level of fatigue and task-related risk. It has generally been the case that people haveassumed a monotonic relationship between fatigue, risk andsafety. That is, as fatigue increases so does the risk of error, inci-dent and/or adverse outcome. This may not always be the case.Neuro-behavioural performance (e.g., on the PVT) is not the sameas complex task performance in the workplace. Indeed, neuro-behavioural performance tests are deliberately designed to elim-inate the potentially adaptive behavioural responses that mediatereal world behaviour. Fatigue impaired individuals can exhibitadaptive behaviours in workplace tasks that will significantlyreduce the likelihood of an error and/or an adverse outcome evenwhen extremely fatigued. A detailed discussion of error man-agement in the context of fatigue related impairment has beenpublished elsewhere.100

Due diligence guidelines for device implementation

In order to ensure a fatigue detection technology is imple-mented appropriately we suggest that organisations use thefollowing three guidelines to assess the suitability of a fatiguedetection technology. First, the technology should be supported byindependent, peer-reviewed, laboratory studies demonstrating thevalidity of the underlying behaviour (e.g., eye blink rates, steeringlane variability) compared with standardised benchmarks of


fatigue (e.g., EEG, neuro-behavioural measures, self-report fatiguescales).

Second, the technology vendor should provide customers withdetailed specificity and sensitivity measures for the fatigue ‘warn-ing’ thresholds along with the sample characteristics used to derivethese values. Ideally, this should be provided using laboratory andsimulator studies, as well as field evaluation studies. These data arecritical in determining the optimum implementation strategy. Forexample, if warning levels have low specificity (a high level of falsenegatives), then a significant proportion of fatigue-related in-cidents will not be detected. This may result in over-reliance on thetechnology and a paradoxical increase in the risk of crashes andfatalities. Alternatively, low sensitivity (high number of false posi-tives) will result in a high number of ‘false alarms’ that are at oddswith drivers’ subjective experience. As a result, driversmay come todistrust and ignore the device entirely, potentially resulting inmore‘missed events’ in the longer term.14 Detailed information onspecificity and sensitivity analyses should be considered so anorganisation can determine whether these are relevant, appro-priate and acceptable for use within their specific operationalcontext. A purchaser should also ensure that the specificity andsensitivity data are derived using studies from which it is appro-priate to generalise to their specific operational circumstances.

Third, as devices must ultimately operate in the field, theyshould preferably be supported by published field data that supportlaboratory or simulator studies. Few devices have such evidenceavailable at present. This is a major weakness across the class oftechnologies. It may also be appropriate to ensure that these dataare collected as part of the device implementation strategy. In somecases, companies may need to undertake a post-implementationassessment to determine local validation issues such as effective-ness, data quality and user compliance/tolerance. This should formpart of the established audit and compliance functions within anorganisation’s safety management system (SMS).

In our review of the evidence typically provided by vendors,marketing materials often referred to quasi-scientific evidence thatwould not normally be considered independent peer reviewedliterature (e.g., abstracts published in non-peer reviewed confer-ence proceedings, in-house studies or studies commissioned by thedevelopers and not reported in the independent peer-reviewedliterature). This is often ambiguous for purchasers of the technol-ogy, who may not be discipline experts as to what constitutes le-gally defensible scientific evidence.

Ultimately, the best evidence for a devices’ effectivenesswill comefrom high quality studies conducted by independent researchers,published in highly-ranked peer-reviewed journals, where multiplestudies provide converging evidence. Given that many organisationsmay lack thediscipline expertise to assess the evidencequality, itmaybe appropriate to seek guidance from a third-party subject matterexpert who can provide independent advice on the quality of thesupporting evidence provided by the vendor.

Specific recommendations for the evaluation and implementation offatigue detection technologies

In the section abovewehave identified that for a fatigue detectiondevice to be legally and scientifically defensible it should meet anumber of due diligence guidelines. It would be tempting to makespecific recommendations on the appropriate thresholds value forsome of these criteria (i.e., sensitivity and specificity >90%). Indeed,in regulatory environments characterised by a high degree of pre-scription, this is likely to be the case. However, current trends insafety management are moving towards a performance-basedapproach whereby an organisation is required to present a safetycase outlining how they will manage and mitigate fatigue risk. As



such, theways inwhich a fatigue detection device could be usedmayvary significantly from organisation to organisation and the appro-priate threshold values may also vary. In general, the more reliancean organisation places on the fatigue detection technology thegreater the need for higher levels of sensitivity and specificity,ecological validity and user uptake and compliance. On the otherhand, where a fatigue detection device was only an element of anoverall SMS, it may be acceptable to implement a technology eventhough it may not have the same performance characteristics aswould be required in settings in which the fatigue detection tech-nology was the sole defence. For example, where a fatigue detectiontechnology is used to identify the frequency of fatigue related eventsacross a workforce it would be possible, based on the law of largenumbers, to tolerate lower specificity and sensitivity. In contrast,where the technology was used to identify micro-sleeps in an indi-vidual while driving, considerably higher specificity and sensitivityvalue would be necessary. Given the general view that fatigue is acomplex hazard and is best mitigated using a defences-in-depthapproach, that is multiple, overlapping and redundant controls, it isour opinion that organisations should be required to demonstratethat the performance characteristics of the fatigue detection tech-nology employed are appropriate to the context in which it isimplemented.101

Costebenefit of implementation

Technology purveyors typically want to increase product sales,and will often argue to equip entire fleets to monitor every worker.While this approach has obvious safety advantages, it may beprohibitively expensive. Until fatigue detection technology be-comes more affordable, companies may need to implement thesedevices more strategically, using a risk-targeting approach. Forexample, 10% of fleet vehicles could be fitted with the technology(e.g., eye blink and lane deviation devices), and these vehiclesassigned to drivers for whom there are reasonable grounds tosuspect elevated fatigue levels (e.g., those rostered on for earlymorning hours or long shifts or with ‘borderline’ subjective fitness-for-duty evaluations or with a history of susceptibility to fatigue).

Implementation within a broader FRMS

Fatigue detection technology has a useful role to play in man-aging fatigue, and will be most effective when implemented as onecomponent of a broader safety risk management system withmultiple levels of control.11,12,21 No one level of control will everprevent all fatigue-related incidents. Fatigue detection technologycannot prevent fatigue from occurring, nor can it mitigate fatigueonce it is detected. However, multiple-level controls in combina-tion have a much greater sensitivity for detecting and preventing afatigue related incident.12

The hierarchy of controls for fatigue related risk must beaddressed in order. Devices may detect high levels of driver fatigueif companies have not yet developed work shifts using level 1controls (such as hours of service guidelines or fatigue modelling).Implementing these controls first will reduce the likelihood ofdriver fatigue occurring during working hours. As an additionalsafeguard, companies may verify that employees have hadadequate sleep before starting shifts (level 2 control).

What happens if and when devices detect driver fatigue? Com-panies must take responsibility for this information, and providereasonable options for on-the-job fatigue management to preventincidents from occurring.14 Currently, many fatigue detection de-vices are only set up to identify and alert individuals at risk. Thus, inthe short term, only approved rest breaks where drivers may obtainsleep will alleviate their fatigue. However, it is important for


employers to realise that it is not necessary to identify individualemployees in order to manage fatigue. Data generated from thesetechnologies could also be aggregated across individuals to identifypatterns of fatigue within the workforce. For example, such datamay identify that 30% of aworkforce is fatigued during certain shifts(e.g., 22:00 he05:00 h). This information has the potential to informmore individualised rostering practices and to be used within ed-ucation programs promoting sleep hygiene.14 Currently, manytechnologies are not equipped to do this, perhaps partly due to theirdevelopmentwithin aUS cultural contextmodelled on drug/alcoholdetection. In actuality, a combination of individual alerts andaggregated data may be the best approach.

For these practices to work, they must occur within a workplaceculture of risk management, rather than of punishment. In an in-dustrywith little supervision, it is ultimately the drivers’ decision tostop driving and they must perceive that employers support thisdecision, without fear of reprisal.14 Employees will especially resentthe possibility of losing hours (and money) if fatigue occurs as aresult of on-the-job conditions (e.g., long shifts).14,17 Companysupport will ultimately improve the effectiveness of these devices,as employees may be more accepting and less likely to circumventthe devices or ignore their warnings. For this to work, fatigue mustbe treated as just another risk to be managed. As the Australianroad transport industry relies heavily on 24-h operations, fatiguewill never be completely prevented. Thus, to maximise both safetyand productivity, a culture shift needs to occur. Employers do notneed to know why an employee is fatigued in order to manage it.Thus, in a culture of full disclosure and risk management, an em-ployer’s response to devices alarming should be “how can wemanage it?” rather than “what were you doing last night?” Onlywhen devices alarm regularly for particular employees is it neces-sary for employers to engage in root cause analysis.

Legal implications

Finally, companies must also consider the legal implicationsassociated with implementing these devices, including those thatarise if a drowsy driver were to crash while being monitored by oneof these devices. Choice of technology may only be considereddefensible if companies can demonstrate they used a well-developed and reasoned rationale for implementing the device.As companies have a duty of care to their employees, selecting adevice that has not been the subject of systematic and thoroughvalidation may not stand up in court.

It is also important to considerwhere the responsibility for fatiguemanagement rests once these devices are implemented. If companieshave implemented defensible devices, then it is ultimately thedriver’s responsibility to 1) use the device correctly, and 2) heed thewarnings and stop driving. However, current ‘chain of responsibility’laws require companies to facilitate employees to engage in safepractices and consider it illegal for companies to coerce employeesinto making unsafe decisions. Thus, simply paying lip-service to theprinciples of fatigue management by installing devices withoutsetting up driver supports, perhaps in an attempt to appease regu-latory bodies or insurance companies, will not be defensible.12,102

Future directions

Heavy vehicles have become increasingly computerised withinthe road transport industry, making them ideally placed to integratesystems-based fatigue detection technology that monitors driverperformance. For example, global positioning system (GPS) devicescould be modified to monitor speed variability and lane departure.This also applies forother industries; for example, blackbox recorderscould be modified to assess how fatigued pilots approach landing.


Research agenda

1) Extensive, independent reliability and validity in-

vestigations need to be undertaken on nearly all fatigue

detection devices, preferably using large samples from

a variety of demographic backgrounds and including

individuals with specific health issues such as eye

problems or sleep apnoea.

2) The effectiveness of fatigue detection devices for

improving safety outcomes needs to be established.

3) Devices using laboratory-derived correlates of fatigue

need to be validated in field settings.

4) Both laboratory and field based investigations should

test the assumption of a monotonic relationship be-

tween fatigue, risk and safety.

5) The ideal time for a device to alarm prior to a fatigue

related incident occurring is yet to be established.

6) The development, testing and implementation of inte-

grated, systems-based fatigue detection technology to

monitor driver performance may be a priority research

area moving forward.


Thus, the next generation of fatigue detection devices may shift frompredominatelymonitoring the operator, to a focus on their commandof the vehicle. Such devices provide a less invasive alternative tophysiological measures and there is some evidence to suggest theyare more accepted by users. Embedded performance technology isalso likely to produce higher sensitivity and specificity, given that itmeasures driver safety directly (e.g., lane departure), rather thanthrough proxy measures (e.g., eye movement). It is also likely, how-ever, that vehicle manufacturers will develop more advanced driverassistance systems, supporting lane guidance and distance keepingfunctions. This may mask the effects of fatigue to an extent. As such,specific monitoring of the driver may always be required.

Furthermore, recent marketing by some of the majorautomotive manufacturers has indicated the likely inclusion of fa-tigue detection technologies as an optional extra in bothpersonal and commercial vehicles. Details of these technologiescan be obtained from the following websites (http://www.daimler.com/dccom/0-5-1210218-1-1210332-1-0-0-1210228-0-0-135-0-0-0-0-0-0-0-0.html; http://www.zercustoms.com/news/Volvo-Driver-Alert-Control-and-Lane-Departure-Warning.html; http://www.volkswagen.co.uk/#/new/passat-vii/explore/experience/driver-assistance/driver-alert-system/; http://media.ford.com/article_print.cfm?article_id¼34562). We have not included a detailed anal-ysis of these technologies, as there have not yet been published,scientific data describing their specific performance character-istics. It is likely that the performance characteristics of thesetechnologies will remain commercial-in-confidence. This does,however, raise potential issues related to decisions about thelikely efficacy of these technologies and the extent to which anuninformed purchaser should be able to rely on the technology.

Conclusion

Fatigue detection technology has the potential to improve themanagement of fatigue generally, but particularly within thetransport industry. However, there are many challenges to addressbefore the promise of these technologies can be realised. Funda-mentally, the effectiveness of these devices must be supported withhigh quality, independent studies providing converging evidence.Of equal importance is the organisational safety culture withinwhich these devices are embedded. It will be critical for thosedeveloping and regulating these technologies to provide andrequire supporting evidence and guidelines for use that acknowl-edge both the technical strengths and weaknesses of these tech-nologies and the temptation to ‘privilege’ these technologies overother perhaps less seductive elements of an integrated FRMS.

Practice points

1) Fatigue is a significant risk factor for workplace in-

cidents, injuries and fatalities.

2) Commercially available fatigue detection technologies

are promoted as a way to mitigate fatigue risk.

3) A majority of these technologies, however, lack defen-

sible reliability or validity data, despite being founded

on reasonable scientific evidence.

4) None of the current fatigue detection technologiesmeet

all proposed regulatory criteria for a legally and scien-

tifically defensible device.

5) Some of these technologies have promise as useful

elements of a broader fatigue risk management system.

6) In line with legal requirements for exercising due dili-

gence, evaluative and operational criteria should be

met before these technologies are adopted in work-

place settings.


Conflict of interest

All authors declare no conflicts of interest.

References

1. House of Representatives Standing Committee on Communication Trans-port and the Arts. Beyond the midnight oil: an enquiry into managing fatiguein transport. Canberra: Commonwealth of Australia; 2000.

2. Williamson A, Lombardi DA, Folkard S, Stutts J, Courtney TK,Connor JL. The link between fatigue and safety. Accid Anal Prev2011;43:498e515.

3. Curnow G, editor. Australian Transport Safety Bureau heavy truck crash da-tabases: what do the statistics tell us? Melbourne: National Heavy VehicleSafety Seminar; 2002.

4. Dement WC, Carskadon MA. Current perspectives of daytime sleepiness:the issues. Sleep 1982;5:56e66.

5. Dinges DF. An overview of sleepiness and accidents. J Sleep Res 1995;4:4e14.

6. Dawson D. Validation in the context of commercialization: the white rabbiteffect. Sleep Med 2011;12:205e6.

7. Dinges DF, Pack F, Williams K, Gillen KA, Powell JW, Ott GE, et al. Cumu-lative sleepiness, mood disturbance, and psychomotor vigilance perfor-mance decrements during a week of sleep restricted to 4e5 hours pernight. Sleep 1997;20:267e77.

8. Van Dongen HP, Maislin G, Mullington JM, Dinges DF. The cumulative costof additional wakefulness: doseeresponse effects on neurobehavioralfunctions and sleep physiology from chronic sleep restriction and totalsleep deprivation. Sleep 2003;26:117e26.

9. Dawson D, Reid K. Fatigue, alcohol and performance impairment. Nature1997;388:235.

10. Härmä M, Sallinen M, Ranta R, Mutanen P, Müller K. The effect of anirregular shift system on sleepiness at work in train drivers and railwaytraffic controllers. J Sleep Res 2002;11:141e51.

*11. Dawson D, McCulloch K. Managing fatigue: it’s about sleep. Sleep Med Rev2005;9:365e80.

12. Gander P, Hartley L, Powell D, Gabon P, Hitchcock E, Mills A, et al. Fatiguerisk management: organizational factors at the regulatory and industry/company level. Accid Anal Prev 2011;43:573e90.

13. Dawson D, Noy YI, Härmä M, Åkerstedt T, Belenky G. Modelling fatigueand the use of fatigue models in work settings. Accid Anal Prev 2011;43:549e64.

*14. Hartley L, Horberry T, Mabbott N, Krueger GP. Review of fatigue detectionand prediction technologies. Melbourne: National Road Transport Commis-sion; 2000.

*15. Liu CC, Hosking SG, Lenné MG. Predicting driver drowsiness usingvehicle measures: recent insights and future challenges. J Saf Res 2009;49:239e45.

* The most important references are denoted by an asterisk.


http://www.daimler.com/dccom/0-5-1210218-1-1210332-1-0-0-1210228-0-0-135-0-0-0-0-0-0-0-0.html



http://www.zercustoms.com/news/Volvo-Driver-Alert-Control-and-Lane-Departure-Warning.html

http://www.zercustoms.com/news/Volvo-Driver-Alert-Control-and-Lane-Departure-Warning.html

http://www.volkswagen.co.uk/#/new/passat-vii/explore/experience/driver-assistance/driver-alert-system/



http://media.ford.com/article_print.cfm%3farticle_id%3d34562




*16. May JF, Baldwin CL. Driver fatigue: the importance of identifying causalfactors of fatigue when considering detection and countermeasure tech-nologies. Transportation Res F 2009;12:218e24.

*17. Wright NA, Stone BM, Horberry TJ, Reed N. A review of in-vehicle sleepinessdetection devices. TRL Limited; 2007.

*18. Barr L, Popkin S, Haworth H. An evaluation of emerging driver fatiguedetection measures and technologies: final report. Washington, D.C: FederalMotor Carrier Safety Administration; 2009.

*19. Balkin TJ, Horrey WJ, Graeber RC, Czeisler CA, Dinges DF. The challengesand opportunities of technological approaches to fatigue management.Accid Anal Prev 2011;43:565e72.

*20. Caterpillar Inc. Driver fatigue detection technology review. Peoria, IL: Cater-pillar; 2008.

*21. Horrey WJ, Noy YI, Folkard S, Popkin SM, Howarth HD, Courtney TK.Research needs and opportunities for reducing the adverse safety conse-quences of fatigue. Accid Anal Prev 2011;43:591e4.

*22. Dinges DF, Mallis MM. Managing fatigue by drowsiness detection: cantechnological promises be realised? In: Hartley LR, editor. Managing fatiguein transportation: proceedings of the third international conference on fatigueand transportation. Oxford: Elsevier; 1998.

23. Dinges DF, editor. Fatigue: where biology meets technology 2008. Keynotepresentation at the Aviation Fatigue Management Symposium: Partner-ships for Solutions.

24. Dinges DF, Powell JW. Microcomputer analyses of performance on aportable, simple visual RT task during sustained operations. Behav ResMethods Instrum Comput 1985;17:652e5.

25. Ferguson SA, Paech GM, Dorrian J, Roach G, Jay SM. Performance on asimple response time task: is sleep or work more important for miners?Appl Ergon 2011;42:210e3.

26. Lamond N, Dawson D, Roach G. Fatigue assessment in the field: validationof a hand-held electronic psychomotor vigilance task. Aviat Space EnvironMed 2005;76:486e9.

27. Petrilli RM, Jay SM, Dawson D, Lamond N. The impact of sustained wake-fulness and time-of-dayonOSPATperformance. Ind Health 2005;43:186e92.

28. Thorne DR, Johnson DE, Redmond DP, Sing HC, Belenky G. The Walter Reedpalm-held psychomotor vigilance test. Behav Res Methods 2005;37:111e8.

29. Balkin TJ, Bliese PD, Belenky G, Sing H, Thorne DR, Thomas M, et al.Comparative utility of instruments for monitoring sleepiness-related per-formance decrements in the operational environment. J Sleep Res 2004;13:219e27.

30. Dorrian J, Lamond N, Kozuchowski K, Dawson D. The driver vigilancetelemetric control system (DVTCS): investigating sensitivity to experi-mentally induced sleep loss and fatigue. Behav Res Methods 2008;40:1016e25.

31. Dorrian J, Rogers NL, Dinges DF. Kushida C, editor. Psychomotor vigilanceperformance: neurocognitive assay sensitive to sleep loss. New York: MarcelDekker; 2004.

32. Pilcher JJ, Band D, Odle-Dusseau HN, Muth ER. Human performance undersustained operations and acute sleep deprivation conditions: toward amodel of controlled attention. Aviat Space Environ Med 2007;78:B15e24.

33. Baulk S, Biggs S, Reid K, van den Heuvel C, Dawson D. Chasing the silverbullet: measuring driver fatigue using simple and complex tasks. Accid AnalPrev 2008;40.

34. Lamond N, Dawson D. Quantifying the performance impairment associatedwith fatigue. J Sleep Res 1999;8:255e62.

35. Fletcher A, Dawson D. Field-based validations of a work-related fatiguemodel based on hours of work. Transportation Res F 2001;4:75e88.

36. Sadeh A, Dan O, Bar-Haim Y. Online assessment of sustained attentionfollowing sleep restriction. Sleep Med 2011;12:257e61.

37. De Gennaro L, Ferrara M, Urbani L, Bertini M. Oculomotor impairment after1 night of total sleep deprivation: a dissociation between measures ofspeed and accuracy. Clin Neurophysiol 2000;111:1171e8.

38. Richman JE. Use of the Eyecheck� pupillometer technology effectivenessas a non-invasive technique to detect and measure possible impairmentresulting from the suspected usage of alcohol, drugs, and/or detect fatiguefrom sleep deprivation; 2002.

39. Eyes the windows to detecting fatigue & drugs. The Aust Mining Times 2003.40. Dal Santo JP, Tousman SA, Shaw DL, Gerstein JK, Falzone JJ, Dal Santo NJ,

editors. Pupillometry and trucking fatigue 1997. 22nd pupil colloquium.Point Clear, Alabama.

41. Lamond N, Dorrian J, Kozuchowski K, Hussey FJ, Dawson D. Neurocom driver/operator alertness control system and eyecheck infrared pupillometer device:validation study. South Australia: University of South Australia; 2005.

42. Goldich Y, Barkana Y, Pras E, Zadok D, Hartstein M, Morad Y. The effects ofsleep deprivation on oculomotor responses. Curr Eye Res 2010;35:1135e41.

43. McClelland LE, Pilcher JJ, Moore DD. Oculomotor measures as predictors ofperformance during sleep deprivation. Aviat Space Environ Med 2010;81:833e42.

44. Rowland LM, Thomas ML, Thorne DR, Sing HC, Krichmar JL, Davis HQ, et al.Oculomotor responses during partial and total sleep deprivation. AviatSpace Environ Med 2005;76:C104e13.

45. Caldwell JA, Caldwell L, Brown DL, Smith JK. The effects of 37 hours ofcontinuous wakefulness on the physiological arousal, cognitive perfor-mance, self-reported mood, and simulator flight performance of F-117Apilots. Milit Psychol 2004;16:163e81.


46. Chandler JF, Arnold RD, Phillips JB, Lojewski RA, Horning DS. Preliminaryvalidation of a readiness-to-fly assessment tool for use in naval aviation.Pensacola, FL: Naval Aerospace Medical Research Laboratory; 2010.

47. Hirvonen K, Puttonen S, Gould K, Korpela J, Koefoed VF, Müller K.Improving the saccade peak velocity measurement for detecting fatigue.J Neurosci Methods 2010;187:199e206.

48. Thomas ML, Sing HC, Belenky G, Holcomb HB, Mayberg HS, Dannals RF,et al. Neural basis of alertness and cognitive performance impairmentsduring sleepiness II. Effects of 48 and 72 h of sleep deprivation on wakinghuman regional brain activity. Thalamus Relat Syst 2003;2:199e229.

49. Russo M, Thomas M, Thorne D, Sing H, Redmond D, Rowland L, et al. Oc-ulomotor impairment during chronic partial sleep deprivation. Clin Neu-rophysiol 2003;114:723e36.

50. LeDuc PA, Greig JL, Dumond SL. Involuntary eye responses as measures of fa-tigue in U.S. army Apache aviators. Aviat Space Environ Med 2005;76:C86e91.

51. Morad Y, Barkana Y, Zadok D, Hartstein M, Pras E, Bar-Dayan Y. Ocularparameters as an objective tool for the assessment of truck drivers fatigue.Accid Anal Prev 2009;41:856e60.

52. Heitmann A, Guttkuhn R, Aguirre A, Trutschel U, Moore-Ede M, editors.Technologies for the monitoring and prevention of driver fatigue 2001. TheFirst International Driving Symposium on Human Factors in DriverAssessment, Training and Vehicle Design.

53. Wierwille WW, Ellsworth LA, Wreggit SS, Fairbanks RJ, Kirn CL. Research onvehicle-based driver status/performance monitoring: development, validation,and refinement of algorithms for detection of driver drowsiness. Final report(DOT HS 808 247). Washington, DC: National Highway Traffic SafetyAdministration; 1994.

54. Caffier PP, Erdmann U, Ullsperger P. Experimental evaluation of eye-blinkparameters as a drowsiness measure. Eur J Appl Physiol 2003;89:319e25.

55. Longhurst G. Understanding driver visual behaviour 2002. Available from:http://www.seeingmachines.com/pdfs/media_coverage/2002/2002_TTI_p92.pdf; 7 November 2011.

56. A user-centered drowsy-driver detection and warning system [database on theInternet] [cited 11th July 2011]. Available from: http://www.ri.cmu.edu/pub_files/pub4/ayoob_ellen_m_2003_2/ayoob_ellen_m_2003_2.pdf; 2003.

57. Weirville WW, Wreggit SS, Kirn CL, Ellsworth LA, Fairbanks RJ. Research onvehicle-based driver status/performance monitoring: development, validation,and refinement of algorithms for detection of driver drowsiness. Washington,DC: National Highway Traffic Safety Administration; 1994.

58. Dinges DF, Mallis MM, Maislin G, Powell JW. Final report: evaluation oftechnologies for ocular measurement as an index of fatigue and as the basis foralertness measurement. Washington, DC: National Highway Traffic SafetyAdministration; 1998.

59. Mallis M. Evaluation of techniques for drowsiness detection: experiment onperformance-based validation of fatigue-tracking technologies. Philadelphia,PA: Drexel University; 1999.

60. Weirville WW, Lewin MG, Fairbanks RJ. Final reports: research on vehicle-based driver status/performance monitoring, part III. Washington, DC: Na-tional Highway Traffic Safety Administration; 1996.

61. Johns MW, Tucker A, Chapman R, Crowley K, Michael N. Monitoring eyeand eyelid movements by infrared oculography to measure drowsiness indrivers. Somnologie 2007;11:234e42.

62. Ftouni S, Sletten T, Howard M, Lenne M, Lockley S, Rajaratnam S. Nightshift work and drowsy driving in Australian nurses. Sleep Biol Rhythm2010;8:A57.

63. Ahlstrom C, Andersson J, Anund A, Borjesson E, Johansson H, Johnsson J.Detecting sleepiness by Optalert: final report. Linkoping, Sweden: VirtualPrototyping and Assessment by Simulation; 2010.

64. Crowley K, Johns MW, Chapman RJ, Tucker AJ, Patterson J. An ocularmeasure of drowsiness and the EEG: changes with sleep deprivation. Sleep2008;31:A119.

65. Howard M, Clarke C, Gullo M, Johns M, Swann P, Pierce R, et al. Evaluationof two eyelid closure monitors for drowsiness detection. Sleep BiologRhythms 2006;4:A13.

66. Johns MW, Chapman R, Crowley K, Tucker A. A new method for assessingthe risks of drowsiness while driving. Somnologie 2008;12:66e74.

67. Johns MW, Tucker AJ, Chapman RJ, Michael NJ, Beale CA. A new scale ofdrowsiness based on multiple characteristics of blinks: the Johns Drowsi-ness Scale. Sleep 2006;29:A365.

68. Stephan K, Hosking S, Regan M, Verdoorn A, Young K, Haworth N. Drivingperformance and the Johns Drowsiness Scale. Victoria: Monash UniversityAccident Research Centre; 2006.

69. Howard M, Wilkinson V, Jackson M, Stevens B, Westlake J, Barnes M, et al.The utility of different eyelid movement metrics for detecting drowsiness-related lapses. J Sleep Res 2011;20:38.

70. Anderson C, Chang A-M, Ronda J, Czeisler C. Infra-red reflectance oculog-raphy reveals changes in drowsiness and is commensurate with goldstandard laboratory measures. J Sleep Res 2011;20:36e7.

71. Michael N, Johns M, Owen C, Patterson J. Effects of caffeine on alertness asmeasured by infrared reflectance oculography. Psychopharmacology2008;200:255e60.

72. BHP Billiton. Cutting-edge technology to warn of approaching drowsiness.BHP Billiton sustainability report 200870e1.

73. Corbett MA. A drowsiness detection system for pilots: Optalert. Aviat SpaceEnviron Med 2009;80:149.


http://www.seeingmachines.com/pdfs/media_coverage/2002/2002_TTI_p92.pdf

http://www.seeingmachines.com/pdfs/media_coverage/2002/2002_TTI_p92.pdf

http://www.ri.cmu.edu/pub_files/pub4/ayoob_ellen_m_2003_2/ayoob_ellen_m_2003_2.pdf

http://www.ri.cmu.edu/pub_files/pub4/ayoob_ellen_m_2003_2/ayoob_ellen_m_2003_2.pdf


74. Sharwood LN, Elkington J, Stevenson M, Grunstein R, Meuleners L, Ivers RQ,et al. Assessing sleepiness and sleep disorders in Australian long-distancecommercial vehicle drivers: self-report versus an “at home” monitoringdevice. Sleep 2012;35:469e75.

75. Blanco M, Bocanegra JL, Morgan JF, Fitch GM, Medina A, Olson RL, et al.Assessment of a drowsy driver warning system for heavy-vehicle drivers: finalreport. Washington, DC: US Department of Transportation; 2009.

76. Mallis MM, Dinges DF. Monitoring alertness by eyelid closure. In:Stanton N, Hedge A, Brookhuis K, Salas E, Hendrick H, editors. Handbook ofhuman factors and ergonomics methods. Boca Raton, Florida: CRC Press LLC;2005. p. 254e61.

77. Aboukhalil A. Systems approach to the design of locomotive fatigue man-agement technologies. Massachusetts Institute of Technology; 2006.

78. Oman CM, Liu AM. Development of locomotive in-cab alerter technologyassessment. Cambridge, MA: Man Vehicle Laboratory, Department ofAeronautics and Astronautics, Massachusetts Institute of Technology; 2007.

79. Williamson A, Chamberlain T. Review of on-road driver fatigue monitoringdevices. NSW: NSW Injury Risk Management Research Centre, University ofNew South Wales; 2005.

80. Lal SKL, Craig A, Boord P, Kirkup L, Nguyen H. Development of an algorithmfor an EEG-based driver fatigue countermeasure. J Saf Res 2003;34:321e8.

81. Lal SKL, Craig A. A critical review of the psychophysiology of driver fatigue.Biol Psychol 2001;55:173e94.

82. Craig A, Tran Y, Wijesuriya N, Nguyen H. Regional brain wave activitychanges associated with fatigue. Psychophysiology 2012;49:574e82.

83. Berka C, Levendowski DJ, Cvetinovic M, Petrovic MM, Davis G, Lumicao MN,et al. Real-time analysis of EEG indexes of alertness, cognition and memoryacquired with a wireless EEG headset. Int J Human-Computer Interact2004;17:151e70.

84. Johnson RR, Popovic D, Olmstead RE, Stickic M, Levendowski DJ, Berka C.Drowsiness/alertness algorithm development and validation using syn-chronized EEG and cognitive performance to individualize a generalizedmodel. Biol Psychol 2011;87:241e50.

85. Berka C, Levendowski DJ, Westbrook P, Davis G, Lumicao MN, Olmstead RE,et al., editors. Implementation of a closed-loop real-time EEG-based drowsi-ness detection system: effects of feedback alarms on performance in a drivingsimulator. 1st International Conference on Augmented Cognition 2005. LasVegas, Nevada.

86. Levendowski DJ, Olmstead RE, Konstantinovic ZR, Berka C, Westbrook P.Detection of electroencephalographic indices of drowsiness in real timeusing a multi-level discriminant function analysis. Sleep 2000;23:A243e4.

87. Levendowski DL, Berka C, Olmstead RE, Jarvik M, editors. Correlations be-tween EEG indices of alertness, measures of performance and self-reportedstates while operating a driving simulator. Miami Beach, Florida: Societyfor Neuroscience Meeting; 1999.


88. Levendowski DL, Berka C, Olmstead RE, Konstantinovic ZR, Davis G,Lumicao MN, et al. Electroencephalographic indices predict futurevulnerability to fatigue induced by sleep deprivation. Sleep 2001;24:A243e4.

89. Mitler MM, Westbrook P, Levendowski DJ, Ensign WY, Olmstead RE,Berka C, et al. Validation of automated EG quantification of alertness:methods for early identification of individuals most susceptible to sleepdeprivation. Sleep 2002;25:A147e8.

90. Mabbott N, Lydon M, Hartley L, Arnold P. Procedures and devices to monitoroperator alertness whilst operating machinery in open-cut coal mines. Ver-mont South, Victoria: ARRB Transport Research Ltd.; 1999.

91. Haworth NL, Vulcan P. Testing of commercially available fatigue monitors.Victoria: Monash University Accident Research Centre; 1991.

92. Miró E, Cano-Lozano MC, Buela-Casal G. Electrodermal activity during totalsleep deprivation and its relationship with other activation and perfor-mance measures. J Sleep Res 2002;11.

93. Whitlock A. Human performance driver vigilance devices: systems review.London: Rail Safety and Standards Board; 2002.

94. Kamensky G. A guardian angel called ‘DVTCS-L’. Railw Dig 2004;November:22e3.

95. Edwards DJ, Sirois B, Dawson T, Agguire A, Davis B, Trutschel U, editors.Evaluation of fatigue management technologies using weighted featurematirx method. Fourth International Driving Symposium on Human Factorsin Driver Assessment, Training and Vehicle Design 2007. Stevenson,Washington.

96. Mortazavi A, Eskandarian A, Sayed RA. Effect of drowsiness on drivingperformance variables of commercial vehicle drivers. Int J Automot Technol2009;10:391e404.

97. Rudin-Brown CM, Williamson A, Lenne MG. Can driving simulation be usedto predict changes in real-world crash risk?. Victoria: Monash UniversityAccident Research Centre; 2009.

98. Houser A, Murray D, Shackelford S, Kreeb R, Dunn T. Analysis of benefits andcosts of lane departure warning systems for the trucking industry. Washing-ton, D.C.: Department of Transportation, Federal Motor Carrier SafetyAdministration; 2009.

99. Dinges DF, Maislin G, Krueger GP, Brewster R, Carroll RJ. Pilot test of fatiguemanagement technologies. Washington, D.C.: Federal Motor Carrier SafetyAdministration; 2005.

100. Dawson D, Chapman J, Thomas M. Fatigue-proofing: a new approach toreducing fatigue-related risk using the principles of error management.Sleep Med Rev 2012;16:167e75.

101. Reason J. Managing the risks of organizational accidents. Surrey: Ashgate;1997.

102. Jones C, Dorrian J, Dawson D. Legal implications of fatigue in the Australiantransportation industries. J Ind Rel 2003;45:344e59.


look before you (s)leep: evaluating the use of fatigue detection technologies within a fatigue risk...

Documents