Ia

Aa

b

a

ARRA

KFTCSP

1

pdoapieidbpptaW

dact

0d

Accident Analysis and Prevention 43 (2011) 690–697

Contents lists available at ScienceDirect

Accident Analysis and Prevention

journa l homepage: www.e lsev ier .com/ locate /aap

nvestigating the relative effects of sleep deprivation and time of day on fatiguend performance

nn Williamsona,∗, Rena Friswellb

Department of Aviation, University of New South Wales (UNSW), Sydney, NSW 2052, AustraliaNSW Injury Risk Management Research Centre, University of New South Wales, Sydney, Australia

r t i c l e i n f o

rticle history:eceived 12 June 2009eceived in revised form 7 September 2010ccepted 15 October 2010

a b s t r a c t

Time of day and the time since last sleep are acknowledged causes of fatigue, but comparatively little isknown about how they interact. This study examines the relative effects of time of day and sleep depri-vation on fatigue and performance. Two independent groups were exposed to 28 h of sleep deprivationbeginning at 06:00 h for one group (n = 39) and at 00:00 h for the other (n = 22). By varying the start time

eywords:atigueime of dayircadian rhythmleep deprivationerformance

for the two groups, but keeping constant the duration of sleep deprivation, the effects of variations in thetime of day of testing could be examined. For the 06:00 h start group the longest period without sleepoccurred close to the low point of the circadian rhythm. For the 00:00 h start group the circadian lowpoint coincided with only two to six hours of sleep deprivation. Performance was evaluated two-hourlyusing eight computer-based tests and subjective fatigue ratings. The results showed a clear interactioneffect. Both time of day and sleep deprivation affected performance but only in combination; neither had

e find

independent effects. Thes

. Introduction

Fatigue is a recognised problem in industry, especially the trans-ort industry. It has been attributed to two main influences: time ofay or the circadian rhythm and the time since last sleep. Evidencen the existence of the circadian rhythm demonstrates that there isdaily rhythm in physiological functions such as temperature, sleepropensity, alertness and performance that has a pronounced min-

mum in the midnight to dawn period and a lesser minimum in thearly afternoon (Czeisler et al., 1980; Monk, 1987). Similarly, theres considerable evidence that increasing time awake also producesecreasing alertness and performance deficits. For example a num-er of studies demonstrated that, starting from around 06:00 h, aeriod of sleep deprivation of around 18 h produces alertness anderformance deficits equivalent to that produced by blood concen-rations of alcohol at the legal limit for driving in Australia andnumber of other countries (0.05% BAC; Dawson and Reid, 1998;illiamson and Feyer, 2000; Arnedt et al., 2001; Falleti et al., 2003).While the role of these two influences in producing fatigue is not

isputed, there is less evidence and agreement on how they inter-ct with one another. For example, in the sleep deprivation studiesited above, the early morning commencement of sleep depriva-ion meant that the apparently vulnerable period of 18 h of sleep

∗ Corresponding author. Tel.: +61 2 9385 4599; fax: +61 2 9385 6637.E-mail address: a.williamson@unsw.edu.au (A. Williamson).

001-4575/$ – see front matter © 2010 Elsevier Ltd. All rights reserved.oi:10.1016/j.aap.2010.10.013

ings have implications for fatigue management.© 2010 Elsevier Ltd. All rights reserved.

deprivation coincides with the circadian low point (01:00–04:00 h)when performance capacity and safety are known to be lower. Thisconfound between time awake and time of day means that it is notpossible to interpret the relative contributions of each in producingalertness and performance effects. It is possible that the apparenteffects of sleep deprivation are due to the combination of sleepdeprivation and time of day effects (so over-estimating the effectsof sleep deprivation), to sleep deprivation effects alone or solely tocircadian influences.

It is important to resolve this question as knowing the sourceof fatigue is essential for designing work–rest schedules that willbe effective for fatigue management. Consequently, the overall aimof this study was to obtain a better understanding of the relativeeffects of time of day and sleep deprivation on fatigue and perfor-mance. The aim of this study was to disentangle the effects of 28 hof sleep deprivation from the effects of time of day on performanceby comparing sleep deprivation commencing at 06:00 and 00:00 h.

2. Method

2.1. Study design

The design involved two independent groups, one collected in1997 and the other in 2006 each of which were exposed to 28 hof sleep deprivation. For one group (n = 39) the sleep deprivationcondition began at 06:00 h and for the second group (n = 22), thesleep deprivation condition began at 00:00 h. By varying the start

A. Williamson, R. Friswell / Accident Analysis and Prevention 43 (2011) 690–697 691

Table 1Overview of approach to data analysis.

Time in sleep deprivation period Details Early (2.00–5.5 h) Late (19:00–23:00 h)

Test timeStart 06:00 Neither effect (A) SD and Circ effect (D)

Sleep deprivation (SD) Low SD High SDCircadian phase (Circ) Circ high Circ lowTest period 0800–1200 0100–0500

ttptdgor0atdeprbodn2

2

vttp

i

TSt

Start 00:00Sleep deprivation (SD)Circadian phase (Circ)Test period

ime for each of the two groups, but keeping constant the dura-ion of sleep deprivation it was possible to examine the effects onerformance of variations in the time of day of testing. These startimes were selected as they test different aspects of the time ofay (circadian rhythm) and sleep deprivation interaction. For theroup commencing sleep deprivation at 06:00 h the longest periodf sleep deprivation occurred close to the low point of the circadianhythm. A six-hour phase delay (sleep deprivation commencing at0:00 h) meant that the circadian low point coincided with onlyround two to six hours of sleep deprivation. It would be expected,herefore, that if sleep deprivation degrades performance indepen-ently of time of day, less sleep deprivation will result in a smallerffect on performance and greater sleep deprivation will produceerformance deficits, regardless of the time of day. If the circadianhythm has an independent effect on performance, there shoulde a performance decrement at the circadian low point regardlessf the amount of sleep deprivation at that point. The overall studyesign is summarised in Table 1. The results for the group begin-ing at 06:00 h were described previously (Williamson and Feyer,000; Williamson et al., 2001).

.2. Study participants

There were 61 participants in total in this study. All wereolunteers. They were recruited through word of mouth and adver-

isements. All were paid for their time either as a normal part ofheir work or as a one-off payment of $1000 at the end of theirarticipation.

Table 2 shows the demographic characteristics of study partic-pants in each study group. There was no difference between the

able 2ummary of demographic characteristics for participants beginning sleep depriva-ion in the early morning (06:00 h) and midnight.

Sleepdeprivationstart 06:00 h

Sleepdeprivationstart 00:00 h

Significance

Age<30 years 35.9% 36.4% �2

(2) = 2.71, ns30–39 years 28.2% 45.5%40+ years 35.9% 18.2%

GenderMale 94.9% 86.4% �2

(1) = 1.35, ns

Education level<7 years 2.6% 0.0% �2

(4) = 1.88, ns7–10 years 57.9% 45.5%11–12 years 15.8% 18.2%Tertiary technical 13.2% 18.2%Tertiary academic 10.5% 18.2%

Epworth scoreMean (/24) (s.d.) 6.69 (3.74) 6.40 (2.99) F(1,59) = 0.09, ns

Lag between last sleep and start testingMean hours (s.d.) 2.3 (2.74) 3.03 (2.73) t(58) = 1.55, ns

Circ effect (B) SD effect (C)Low SD High SDCirc low Circ high0200–0600 1900–2300

early morning and midnight sleep deprivation start groups on age,with very few participants in the most extreme age categories (<20and 50–59 years) so most participants were in the 30–49 years agegroup (53% and 51% for 06:00 and 00:00 start groups respectively).There were also no statistically significant differences between thegroups on gender, educational background or Epworth SleepinessScale (ESS) score. In addition ESS scores were low for both groups.

2.3. Measures

Eight different computer-based performance tests were usedin this study, totaling 30–40 min test time at each test occasion.In addition subjective fatigue rating scales were used on each testoccasion. Details of the measures are provided in Williamson et al.(2001), and are as summarized below. All tests were computer-based and involved stimulus presentation on the screen andresponses using either a keypad (Genovation Micropad 622) or astandard serial mouse.

• Simple reaction time (RT): This is a simple visual-motor responsespeed test involving a yellow circle moving in an irregular coun-terclockwise path around the computer screen. The subject’s taskwas to press a key on the keypad as quickly as possible wheneverthe circle changed colour from yellow to red. The test consistedof 40 colour change trials over a 2 min period. The time takenfor subjects to respond to the colour change and the number ofmissed colour changes were both measured.

• Mackworth clock vigilance test: This task measured the ability tosustain attention in the face of monotonous stimulation. A circle,composed of 24 equally spaced dots, was presented on the com-puter screen. Each dot flashed briefly in turn at constant 500 msintervals. At random intervals (approximately every minute) oneof the dots would be omitted from the flashing sequence. Thesubjects’ task was to respond as quickly as possible via a buttonpress on the keypad whenever a dot was omitted. The task con-tinued for 15 min during which 15 flashes were omitted. Reactiontime to missed flashes and the number of missed responses wererecorded.

• Unstable tracking: This test assessed hand–eye coordination andrequired participants to keep a small green dot inside a mov-ing yellow circle on the computer screen. The task was adaptivein that it became more difficult (movement became less pre-dictable) as the participant became more accurate in keeping thedot within the circle. The test duration was 3 min and measuredtask difficulty level achieved.

• Dual task: This test assessed the ability to attend to two taskssimultaneously by combining the unstable tracking and simpleReaction time tests. The tracking task was the same as when

conducted alone and reaction time was assessed by requiring par-ticipants to respond to colour changes in the circle being trackedusing the same keypad as in the RT test alone. The test durationwas 3 min and the same measures were collected as for the twocomponent tasks alone.

6 nalysi

•

•

•

•

•

•

•

2

s

92 A. Williamson, R. Friswell / Accident A

Symbol digit coding: This test required participants to use a key-pad to respond as quickly as possible to a symbol presentedin the bottom of the screen using a decoding key located atthe top of the screen. The key contained ten nonsense sym-bols paired with digits between zero and nine. Symbols werepresented individually and once the participant had responded,the next symbol appeared. The task duration was 90 s. Thenumber of symbols correct and the average response speedwere recorded.Visual search task: This test measured visual search in 12 tri-als. Each trial involved a set of either two or six target lettersshown at the top of the screen with a set of 60 letters shownbelow. Participants were required to as quickly as possible, searchthe 60-letter set, decide whether it contained all of the targetletters and respond using either of two buttons on the key-pad. Half of the trials involved two targets and the remainderinvolved six targets. In half of the trials the entire target setwas not contained in the larger set and in the remainder all let-ters were present. Speed and number of correct responses wererecorded.Sequential spatial memory: This test involved nine squaresarranged in a 3 × 3 grid on the screen. Individual squares wereilluminated for 250 ms one at a time in a random sequence.Participants were required to reproduce the sequence frommemory by moving and clicking the mouse cursor on eachof the squares that were illuminated in the sequence. If thesequence was reproduced correctly, it increased by one squarein the next presentation. The trial ended when the participantwas unable to reproduce the sequence correctly. Three trialswere presented and the longest correct sequence on each trialwas recorded.Logical reasoning test: This test was based on the grammaticalreasoning test (Baddeley, 1968). Participants were required todecide whether a statement presented on the screen was true orfalse about a pair of letters presented just below it (e.g., ‘B fol-lows A’ is true for AB, but false for BA) and to respond by movingfrom the ‘home’ key that they were pressing between responsesand to depress one of two buttons on the keypad, designated‘true’ or ‘false’. Thirty-two statements were presented and reac-tion times, the number of missed responses and correct responseswere recorded.Subjective fatigue rating scales: Three visual analogue scales werepresented to participants on the computer screen at the begin-ning and end of each testing session. The three scales focused ondifferent aspects of the fatigue experience and were anchoredat the ends by the terms ‘fresh–tired’, ‘clear-headed–muzzy-headed’, and ‘very alert–very drowsy’. Participants used themouse to position a cursor at some point between the anchorsto reflect their current level of fatigue. The computer recordedcursor position at one of the 20 positions along the dimension.These values were subsequently converted to percentages.Questionnaires: Participants were asked to report basic demo-graphic information including age, education level, health-related lifestyle factors; the Epworth Sleepiness Scale (Johns,1991), smoking and alcohol consumption habits and recent workand sleep history.Actigraph measurement: For the group commencing sleep depri-vation at 00:00 h, actigraphs were used to validate self-reportedsleep measures. They were worn for the 30 h period leading upto the commencement of sleep deprivation and test commence-ment.

.4. Procedure

Participants were tested in groups of three to six. Practice ses-ions were conducted for all participants at 14:00 h on the day prior

s and Prevention 43 (2011) 690–697

to the commencement of testing. These sessions involved threeconsecutive full repetitions of the battery of tests. All study partici-pants underwent a sleep deprivation regime of 28 h beginning fromeither 06:00 h or 00:00 h. Participants were encouraged to get asmuch sleep as possible in the period immediately before each con-dition, including afternoon and evening naps so they were as restedas possible for the first test session. On the test day, participantswere asked to wake at 06:00 or 00:00 and testing began approx-imately two hours afterwards (08:00 h or 02:00 h). Participantswere first asked to complete a brief questionnaire concerning theirsleep, eating and drug-taking behaviour since the previous day, andthen began the testing regime. The testing schedule involved testsevery hour for the first five tests then regularly every two hours,with the last test session commencing approximately 27 h aftertheir waking time (at 09:00 h and 03:00 h for 06:00 and 00:00 starttimes respectively). A total of 15 performance test sessions werecompleted, after which the participants were allowed to retire tosleep.

2.5. Analysis

As this study involved a comparison of two independent groups,one commencing sleep deprivation in the early morning (06:00)and the other at midnight, it was important to ensure that theperformance of the two groups was not statistically significantlydifferent before the sleep deprivation testing commenced. For thisreason, the performance on the practice sessions was comparedfor the two groups. Multivariate analysis of variance (MANOVA)was used to compare the two groups across the three practicetests. Measures or tests that showed a statistically significantbetween groups main effect were not included in the main anal-ysis of the effects of time of day and sleep deprivation. Rathersecondary analysis was conducted of difference scores from base-line where baseline was calculated as the mean score of thesecond and third practice sessions. The purpose of this analy-sis was to attempt to control for known fundamental differencesin performance between the early morning and midnight startgroups.

For the main and secondary analysis two blocks of tests werechosen for each sleep deprivation condition, corresponding to testsessions 1–4 (low sleep deprivation) and 11–13 (high sleep depri-vation). These test blocks were also chosen as they correspondedto different time windows in the circadian rhythm and provideda counterbalance of sleep deprivation (low and high) and circa-dian phase (low and high). These time windows were based onthe evidence of the circadian trough in the midnight to dawnperiod and again in the early afternoon (Monk, 1987). For the06:00 h sleep deprivation start condition, the 1–4 test session blockcorresponded to 08:00 to 11:30 h time window (low sleep depriva-tion/circadian high) and the 11–13 test session block correspondedto 01:00–05:00 h time window (high sleep deprivation/circadianlow). For the 00:00 h sleep deprivation start condition, the 1–4test session block corresponded to 02:00–05:30 h time window(low sleep deprivation/circadian low point) and the 11–13 test ses-sion block corresponded to 19:00–23:00 h time window (high sleepdeprivation/circadian high). The test performance and fatigue rat-ings were averaged within each of these four sleep deprivation andcircadian phase time windows and were compared using MANOVAfollowed by post hoc t-tests when interaction effects were statis-tically significant. For this analysis the four post hoc comparisonsof interest were the effects of high sleep deprivation at high and

low circadian points and the effects of the low circadian rhythm atlow and high levels of sleep deprivation. These contrasts are shownin Table 1. For the post hoc analysis the error rate was adjusted forthe number of comparisons using a Bonferroni adjustment so alphawas set at 0.013.

A. Williamson, R. Friswell / Accident Analysi

Tab

le3

Test

san

dm

easu

res

that

show

edgr

oup

mai

nef

fect

sfo

rp

arti

cip

ants

com

men

cin

gat

06:0

0an

d00

:00

h.M

ean

san

dst

and

ard

dev

iati

ons

for

each

ofth

eth

ree

pra

ctic

ese

ssio

nfo

rea

chst

art

tim

e.

Test

/mea

sure

Beg

inat

06:0

0h

Beg

inat

00:0

0h

Sign

ifica

nt

grou

pm

ain

effe

ct

12

31

23

Trac

kin

gLe

velo

fdif

ficu

lty

ach

ieve

d37

.9(2

2.5)

41.2

(24.

7)42

.7(2

2.7)

54.5

(18.

1)59

.5(2

3.6)

54.3

(22.

8)F (

1,56

)=

9.85

,p<

0.00

3,�

2 p=

0.15

Du

alta

skLe

velo

fdif

ficu

lty

ach

ieve

d35

.9(2

4.0)

36.1

(23.

5)44

.7(2

4.0)

52.9

(24.

9)55

.8(2

2.9)

61.2

(18.

9)F (

1,56

)=

10.7

3,p

<0.

002,

�2 p

=0.

16R

eact

ion

tim

e66

9.8

(162

.2)

639.

2(1

02.3

)65

5.8

(128

.2)

572.

1(6

8.4)

584.

8(7

9.9)

593.

3(5

9.0)

F (1,

56)=

6.63

,p<

0.01

3,�

2 p=

0.11

Stan

dar

dd

evia

tion

ofre

acti

onti

me

242.

8(1

87.1

)19

6.2

(153

.5)

193.

6(1

48.1

)12

2.9

(113

.1)

123.

6(9

7.9)

138.

0(1

09.9

)F (

1,56

)=

6.79

,p<

0.01

2,�

2 p=

0.11

Dig

itsy

mbo

lR

eact

ion

tim

e27

58(7

79)

2544

(596

)24

23(5

88)

2298

(488

)21

92(4

79)

2115

(418

)F (

1,56

)=

6.06

,p<

0.01

7,�

2 p=

0.10

Stan

dar

dd

evia

tion

ofre

acti

onti

me

1208

(936

)93

9(7

33)

911

(590

)68

3(2

58)

693

(317

)64

1(1

95)

F (1,

56)=

7.18

,p<

0.01

,�2 p

=0.

11N

um

ber

corr

ect

resp

onse

s33

.1(8

.2)

35.6

(7.1

)37

.0(8

.3)

38.1

(8.3

)39

.8(8

.1)

41.1

(8.6

)F (

1,56

)=

4.6,

p<

0.03

6,�

2 p=

0.08

Spat

ialm

emor

ysp

an4.

7(2

.2)

4.8

(2.4

)5.

0(2

.2)

6.6

(2.0

)6.

5(2

.5)

5.7

(1.7

)F (

1,56

)=

7.98

,p<

0.00

7,�

2 p=

0.13

Gra

mm

atic

alre

ason

ing

Tim

edou

tre

spon

ses

(lap

ses)

3.9

(4.0

)2.

1(2

.0)

1.3

(1.7

)1.

4(1

.8)

1.0

(1.5

)0.

8(1

.4)

F (1,

56)=

7.33

,p<

0.00

9,�

2 p=

0.12

s and Prevention 43 (2011) 690–697 693

3. Results

There were a number of differences between the early morningstart group (06:00 h) and midnight start groups in the amount andpattern of sleep they obtained prior to commencement of the firsttest session. There was no significant difference between the twogroups in the time lag between completing their last sleep and thestart of testing (see Table 2), however, the midnight start group hadless sleep in the most recent period before testing began. The earlymorning start group had more recent sleep, with 7.19 h sleep onaverage between 18:00 the previous night and commencement oftesting at 08:00. The midnight start group had 5.45 h (s.d. = 2.31)sleep on average in their last sleep (all between 15:45 and 02:00),but had 8.26 h (s.d. = 1.77) in their most recent night sleep and13.74 h (s.d. = 2.20) total sleep over the 28 h from 20:00 the previousfull night before.

A summary of the analysis of the a priori group differences dur-ing practice is shown in Table 3. Significant group differences wereshown for the Unstable tracking task, both components of the Dualtask, all measures of the Digit symbol task, Spatial memory spanand the number of timed out responses measure of the Grammati-cal reasoning task. In all cases, the group commencing at midnighthad better performance than the early morning start group acrossall three repetitions of the task and there were no practice by groupinteraction effects. The raw results for these measures were there-fore not analysed further. Instead, for these tests/measures, analysiswas conducted on differences from baseline scores.

3.1. Analysis of raw performance

The main results for the Reaction time test are shown in Fig. 1and statistical test results are shown in Table 4. For the reactionspeed measure there was a significant interaction effect and amedium to large effect size (Cohen, 1977). Post hoc testing showedsignificantly slower responding in the high SD/circadian low timewindow compared to low SD/circadian low window and in the highSD/circadian low window compared to the high SD/circadian highwindow. Neither of the other two comparisons was statisticallysignificant. These findings indicate that there was no influence ofthe circadian rhythm low point on reaction speed during low sleepdeprivation and high levels of sleep deprivation were influentialonly at the circadian low point.

Similar results were found for the number of missed responsesand the variability of reaction speed in the Reaction time test (seeFig. 1 and Table 4). For the number of missed responses the inter-action effect was statistically significant with a medium effect sizeand post hoc comparisons showed the same pattern as for responsetime with no effect of circadian influence on missed responseswhen sleep deprivation levels were low and high levels of sleepdeprivation only showing significantly more missed signals duringthe circadian low period. Again, the other two comparisons werenot significant. For the variability of reaction speed measure therewas a significant interaction effect between study start time andtime of day and post hoc contrasts again showed no effect at thecircadian low point alone or high levels of sleep deprivation alone,but the combined effects were statistically significantly different.

Fig. 2 shows the results for the Mackworth clock vigilance testand the results of the statistical analysis are shown in Table 4.Similar results were found for this test as the Reaction time test.For the Reaction speed measure, a significant interaction effect fol-lowed by post hoc comparisons showed the same pattern of effects

of no effect on speed of response during the circadian low periodwhen sleep deprivation levels were low but significant slowing ofresponse speed when high levels of sleep deprivation coincidedwith the circadian low point. The missed signals measure showedvery similar results with again, a significant interaction effect and

694 A. Williamson, R. Friswell / Accident Analysis and Prevention 43 (2011) 690–697

Table 4Results of analysis of raw performance scores for measures that showed significant interaction effects between circadian phase and sleep deprivation effects.

Test/measure Significant interaction effect Low SDa

Circb low vshigh (A vs B)

High SDCirc low vshigh (C vs D)

Circ highLow vs high SD(A vs C)

Circ lowLow vs high SD(B vs D)

Simple reaction timeReaction speed F(1,59) = 13.18, p < 0.001, �2

p = 0.18 ns p < 0.002 ns p < 0.001Missed responses F(1,59) = 10.14, p < 0.002, �2

p = 0.15 ns p < 0.001 ns p < 0.001Variability of reaction speed F(1,59) = 9.06, p < 0.004, �2

p = 0.13 ns p < 0.001 ns p < 0.001

Mackworth clock vigilance testReaction speed F(1,59) = 16.79, p < 0.001, �2

p = 0.22 ns p < 0.001 ns p < 0.001Missed responses F(1,59) = 29.59, p < 0.001, �2

p = 0.33 ns p < 0.001 ns p < 0.001Variability of reaction speed F = 7.24, p < 0.009, �2 = 0.11 ns p < 0.001 ns p < 0.001

< 0.00

swclt

Fvdiva

(1,59) p

Fatigue ratings F(1.59) = 39.47, p < 0.001, �2p = 0.40 p

a SD = Sleep deprivation effect.b Circ = Circadian or time of day effect.

ignificantly more missed signals during the high SD/circadian low

indow compared to the low SD/circadian low window. Again, the

omparisons for low SD and the window containing the circadianow point alone showed no significant effects. For both reactionime and missed signals the effect sizes were large. The results for

ig. 1. Simple Reaction time test results showing reaction time, missed signals andariability of reaction time at increasing times since the commencement of sleepeprivation for test sessions starting at 06:00 and 0:00 h (means and 95% Confidence

ntervals shown). Letters A–D corresponding to the test time/time in sleep depri-ation groups shown in Table 1 are included in the first graph to aid interpretationnd are the same in the other two graphs.

3 p < 0.001 p < 0.001 p < 0.001

the variability of reaction speed followed the same pattern with a

statistically significant interaction effect and the same pattern forpost hoc contrasts as for the other measures.

The results for the remaining performance measures did notshow statistically significant interaction effects between start time

Fig. 2. Results of the Mackworth vigilance test showing reaction time, missed sig-nals and variability of reaction time at increasing times since the commencementof sleep deprivation for test sessions starting at 06:00 and 0:00 h (means and 95%Confidence intervals shown). Letters A–D corresponding to the test time/time insleep deprivation groups shown in Table 1 are included in the first graph to aidinterpretation and are the same in the other two graphs.

A. Williamson, R. Friswell / Accident Analysis and Prevention 43 (2011) 690–697 695

Fig. 3. Results of the Dual task showing reaction time, missed signals and variabilityof reaction time at increasing times since the commencement of sleep deprivationfor test sessions starting at 06:00 and 0:00 h (means and 95% Confidence intervalsssi

asvt

3

mptD

Fig. 4. Results of the Digit symbol test showing reaction time, missed signals andvariability of reaction time at increasing times since the commencement of sleep

3.3. Analysis of subjective fatigue ratings

TR

hown). Letters A–D corresponding to the test time/time in sleep deprivation groupshown in Table 1 are included in the first graph to aid interpretation and are the samen the other graph.

nd time of day. These included Dual task-missed responses, Digitymbol-percentage of correct responses, MAST-response time andariability of response time and Grammatical reasoning-responseime and number correct.

.2. Analysis of performance adjusting for baseline differences

Analysis of differences from baseline scores was conducted for

easures which demonstrated intergroup differences during the

ractice session (see Figs. 3 and 4). As shown in Table 5, onlyhe reaction time measure and level of tracking difficulty in theual task and the reaction time measure and number of correct

able 5esults of analysis of difference from baseline scores for measures that showed group ma

Test/measure Significant interaction effect LoCh

TrackingLevel of difficulty achieved n.s.

Dual taskLevel of difficulty achieved F(1,59) = 10.19, p < 0.002, �2

p = 0.15 nReaction time F(1,59) = 8.49, p < 0.005, �2

p = 0.13 nStandard deviation of reaction time ns

Digit symbolReaction time F(1,58) = 9.48, p < 0.003, �2

p = 0.14 nStandard deviation of reaction time nsNumber correct responses F(1,58) = 10.9, p < 0.002, �2

p = 0.16 nSpatial memory span ns

Grammatical reasoningTimed out responses (missed signals) ns

a SD = Sleep deprivation effect.b Circ = Circadian or time of day effect.

deprivation for test sessions starting at 06:00 and 0:00 h (means and 95% Confidenceintervals shown). Letters A–D corresponding to the test time/time in sleep depri-vation groups shown in Table 1 are included in the first graph to aid interpretationand are the same in the other graph.

responses in the Digit symbol tasks showed significant start time(group) by time of day interaction effects with medium effect sizes.In each case, post hoc tests showed significantly poorer perfor-mance in the high SD/circadian low time window compared to thehigh SD/circadian high window and in all measures except the levelof tracking difficulty achieved also showed significantly poorer per-formance in the high SD/circadian low time window compared tothe high SD/circadian high window.

The results for subjective ratings of fatigue at each of the sleepdeprivation and circadian time windows are shown in Fig. 5. Asfor the performance measures, there was a statistically signifi-

in effects during practice sessions.

w SDa

ircb low vsigh (A vs B)

High SDCirc low vshigh (C vs D)

Circ highLow vs high SD(A vs C)

Circ lowLow vs high SD(B vs D)

s ns ns p < 0.038s p < 0.011 ns p < 0.0001

s p < 0.009 ns p < 0.003

s p < 0.002 ns p < 0.004

696 A. Williamson, R. Friswell / Accident Analysi

FvCd

ccJhlwUhSt

4

dBoepIwnpr

hocsrlsibosmtbtrta(ansc

ig. 5. Fatigue ratings at increasing durations of time since the start of sleep depri-ation for test sessions starting at 06:00 and 00:00 h (showing means and 95%onfidence intervals). Letters A–D corresponding to the test time/time in sleepeprivation groups shown in Table 1 are included in the graph to aid interpretation.

ant interaction effect and a large effect size, however post hocomparisons showed somewhat different findings (see Table 4).ust as for the performance tests fatigue ratings were significantlyigher in the high SD/circadian low time window than in the

ow SD/circadian low window and in the high SD/circadian lowindow compared to the high SD/circadian high time window.nlike the performance tests, fatigue ratings were also significantlyigher in the low SD/circadian low window compared to the lowD/circadian high time window and in the high SD/circadian highime window compared to the low SD/circadian high window.

. Discussion

This study demonstrated a clear interaction between sleepeprivation and circadian influences on cognitive performance.oth time of day and sleep deprivation factors were determinantsf performance but only in combination, neither had independentffects. The poorest performance occurred when both an extendederiod of wakefulness and circadian influences occurred together.

f the circadian rhythm was not at its low point or trough, thereere no effects of sleep deprivation. Circadian influences also didot show independent effects. Performance at the circadian lowoint was not adversely affected when the study participant wasested.

Interaction between circadian and sleep deprivation effectsave been shown in some previous studies. A number of previ-us forced desynchrony studies also showed combined effects ofircadian rhythm and sleep deprivation. These studies attempt toeparate the synchrony between time awake and the circadianhythm by extending the awake period of study participants toonger than 24 h (usually 28 h) for a number of weeks. Over thetudy period, this experimental manipulation has the effect of mak-ng waking times occur at almost all times in the circadian rhythm,ut minimizing the extent of sleep deprivation. Using this method-logy studies have found that circadian and time awake influenceshow a non-linear interaction with one another for measures ofood (Boivin et al., 1997) alertness and performance on a con-

inuous addition task (Dijk et al., 1992) and the number of errorsut not performance on a modified version of the sustained atten-ion response test (SARTm) (Harrison et al., 2007). Furthermore, aeanalysis of previously collected data in a laboratory sleep depriva-ion task also showed a non-linear interaction between circadiannd time awake influences using the psychomotor vigilance task

Van Dongen and Dinges, 2003). Modeling studies of subjectivelertness and cognitive throughput also demonstrated a similaronlinear interaction between the level of the sleep homeosta-is component (hours of wakefulness) and the amplitude of theircadian component (Jewett and Kronauer, 1999). Evidence from

s and Prevention 43 (2011) 690–697

shiftwork studies also showed an interaction effect of time of dayand acute sleep deprivation. Santhi et al. (2007) showed deteri-oration in sensitivity measures in a selective attention task butonly on the first night of two successive night shifts which theyattributed to the combined effects of circadian influences, acutesleep deprivation and task difficulty.

Clearly the results of our study reinforce the role of the inter-action of the time of day and time awake processes in producingpoorer alertness and performance at the high end of sleep depri-vation and the circadian trough. These previous studies have notdemonstrated, however, that the interaction is the only effect ofthe two influences, at least on the cognitive functions measuredin this study. Most previous studies have not directly comparedthe independent and interaction effects of sleep deprivation andcircadian phase. As this study showed, circadian effects on perfor-mance were only statistically significant when in combination withsleep deprivation and sleep deprivation effects on performancewere only seen at the circadian low point. The forced desynchronystudies shed light on theoretical processes governing sleep, wakeand fatigue effects on performance and so might be expected toallow the comparison of independent and interaction effects, butthis methodology sets up an artificial and continually changingsleep–wake cycle which may not apply to real sleep–wake con-ditions. For example, lengthening waking time to 28 h over manysleep–wake cycles may produce a chronic buildup of fatigue whichmakes the separation of sleep deprivation and circadian effectsdifficult.

In the current study not all performance measures showedeffects on performance although those that did all showed the samecombined effect of time awake and time of day. Harrison et al.(2007) also found that only one error measure of the SARTm showedcombined circadian and time awake effects.

In the current study subjective fatigue ratings did not show thesame patterns as the performance tests. Fatigue ratings showedindependent effects of sleep deprivation and circadian influencesin addition to a larger interaction effect. A number of previousstudies have failed to find consistency between fatigue ratings andneurobehavioural performance (Williamson et al., 2000; Harrisonet al., 2007). This is not entirely surprising as it is likely that peo-ple respond to body status and the external cues about how tiredthey should feel and make judgments that are consistent with theseinfluences. Performance, on the other hand, has different origins.This disconnect between fatigue ratings and performance effectsmeans that fatigue ratings may not be a reliable indicator of poten-tial performance effects.

A possible limitation to this study is the difference in night sleepopportunity afforded to the two groups of study participants inthe period before the commencement of testing. While the groupthat commenced at 08:00 h had the opportunity to get good qual-ity night sleep and obtained just over seven hours on average, thegroup commencing at 02:00 h could only obtain night sleep up tomidnight. Nevertheless, the latter group was well rested as theytook advantage of a previous full night of sleep and had an addi-tional nap or medium length sleep just before the start of testing.In addition, there was no difference between the two groups atthe commencement of testing in the time since last sleep. Nor didthey differ significantly in fatigue ratings at the start of testing.A second limitation to the study is a possibility that the resultscould reflect fundamental differences between the groups. Evenusing differences from baseline analysis may not adequately con-trol for individual differences in sensitivity to time of day and time

awake affects. Further research is needed to distinguish possibleeffects of individual differences. Another possible limitation of thisstudy is its laboratory nature. It is certainly possible these per-formance effects may not remain in real-world environments ashighlighted by evidence from previous studies (Reed and Green,

nalysi

1rpoobsa

wpoeeppApnspWoptmltcfi

C

o

A

DN

R

A

B

A. Williamson, R. Friswell / Accident A

999; Philip et al., 2003; Oran-Gilad and Ronen, 2007). Furtheresearch is needed to test this possibility. It is also possible that theattern of results in this study reflect the sensitivity of the tests;nly detecting effects when two influences combine, but not whennly one influence is present. If so, this still suggests that the com-ined effects of sleep deprivation and circadian lows are of moreerious concern for safe performance than either of these influenceslone.

The finding that significant performance deficits only occurredhen high levels of sleep deprivation coincided with the lowoint of the circadian rhythm has clear implications for the designf work–rest schedules. Rested participants did not show thexpected circadian effects on performance and sleep deprivationffects on performance were not seen outside the circadian lowoint. For shift design, these results suggest that night work can beerformed safely and without error if the person is properly rested.lthough it is often argued that performance during night shift isoorer compared to day shift, some studies have demonstratedo differences between day and night work. Gillberg et al. (2003)howed that control room performance in a simulated thermalower plant did not differ between day and night shift. Similarly,illiamson et al. (2004) found no differences in the performance

f drivers doing 11–12 h day and night shifts. These studies sup-ort the findings of the current study and reinforce the conclusionhat night work may not necessarily involve higher risk of perfor-

ance failure. On the contrary, however, this study indicates thatong work hours that extend into the midnight to dawn, circadianrough period should be avoided as this combination of factors haslear adverse effects on performance. Of course, these laboratoryndings remain to be confirmed in real-world settings.

onflict of interest

The authors declare that they have no actual or potential conflictf interest in the subject matter of this paper.

cknowledgements

This project was funded by an Australian Research Counciliscovery project (DP0559532). Ann Williamson is funded by anHMRC Senior Research Fellowship.

eferences

rnedt, J.T., Wilde, G.J.S., Munt, P.W., MacLean, A.W., 2001. How do prolonged wake-fulness and alcohol compare in the decrements they produce on a simulateddriving task? Accident Analysis and Prevention 33, 337–344.

addeley, A.D., 1968. A 3 minute reasoning task based on grammatical transforma-tion. Psychonomic Science 1, 341–342.

s and Prevention 43 (2011) 690–697 697

Boivin, D.B., Czeisler, C.A., Dijk, D.-J., Duffy, J.F., Folkard, S., Minors, D.S., Totterdell, P.,Waterhouse, J.M., 1997. Complex interaction of the sleep–wake cycle and circa-dian phase modulates mood in healthy subjects. Archives of General Psychiatry54, 145–152.

Cohen, J., 1977. Statistical Power Analysis for the Behavior Sciences. Erlbaum, Hills-dale, NJ.

Czeisler, C.A., Weitsman, E.D., Moore-Ede, M.C., Simmerman, J.C., Krauer, R.S., 1980.Human sleep: its duration and organisation depend on its circadian phase. Sci-ence 210, 1264–1267.

Dawson, A., Reid, K., 1998. Fatigue, alcohol and performance impairment. Nature388, 235.

Dijk, D.-J., Duffy, J.F., Czeisler, C.A., 1992. Circadian and sleep/wake dependentaspects of subjective alertness and cognitive performance. Journal of SleepResearch 1, 112–117.

Falleti, M.G., Maruff, P., Collie, A., Darby, D.G., McStephen, M., 2003. Qualitativesimilarities in cognitive impairment associated withy 24 h of sustained wake-fulness and a blood alcohol concentration of 0.05%. Journal of Sleep Research 12,265–274.

Gillberg, M., Kecklund, G., Goransson, B., Akerstedt, T., 2003. Operator performanceand signs of sleepiness during day and night work in a simulated thermal powerplant. International Journal of Industrial Ergonomics 31, 101–109.

Harrison, Y., Jones, K., Waterhouse, J., 2007. The influence of time awake and cir-cadian rhythm upon performance on a frontal lobe task. Neuropsychologia 45,1966–1972.

Jewett, M.E., Kronauer, R.E., 1999. Interactive mathematical models of subjectivealertness and cognitive throughput in humans. Journal of Biological Rhythms14, 588–597.

Johns, M.W., 1991. A new method for measuring daytime sleepiness: the Epworthsleepiness scale. Sleep 14, 540–545.

Monk, T.H., 1987. Subjective ratings of sleepiness—the underlying circadian mech-anisms. Sleep 10, 343–353.

Oran-Gilad, T., Ronen, A., 2007. Road characteristics and driver fatigue: a simulatorstudy. Traffic Injury Prevention 8, 281–289.

Philip, P., Taillard, J., Klein, E., Sagaspe, P., Charles, A., Davies, W.l., Guilleminault,C., Bioulac, B., 2003. Effect of fatigue on performance measured by a drivingsimulator in automobile drivers. Journal of Psychosomatic Research 55, 197–200.

Reed, M.P., Green, P.A., 1999. Comparison of driving performance on-road and in alow-cost simulator using a concurrent telephone dialing task. Ergonomics 42,1015–1037.

Santhi, N., Horowitz, T.S., Duffy, J.F., Czeisler, C.A., 2007. Acute sleep depriva-tion and circadian misalignment associated with transition onto the firstnight of work impairs visual selective attention. PLoS ONE 11, e1233,doi:10.1371/journal.pone.0001233.

Van Dongen, H.P.A., Dinges, D.F., 2003. Investigating the interaction between thehomeostatic and circadian processes of sleep–wake regulation for the predictionof waking neurobehavioural performance. Journal of Sleep Research 1, 181–217.

Williamson, A.M., Feyer, A.-M., 2000. Moderate sleep deprivation produces compre-hensive cognitive and motor performance impairments equivalent to legallyprescribed levels of alcohol intoxication. Occupational and EnvironmentalMedicine 57, 649–655.

Williamson, A.M., Feyer, A.-M., Friswell, R., Finlay-Brown, S., 2000. Demon-stration project for fatigue management programs in the road transportindustry—summary of findings. Federal Australian Transportation Safety BureauReport, CR192, pp. 1–20.

Williamson, A.M., Feyer, A.-M., Mattick, R., Friswell, R., Finlay-Brown, S., 2001. Devel-

oping measures of fatigue using an alcohol comparison to validate the effects offatigue on performance. Accident Analysis and Prevention 33, 313–326.

Williamson, A.M., Feyer, A.-M., Friswell, R., 2004. Fatigue and perfor-mance in heavy truck drivers working day shift, night shift or rotatingshifts. National Transport Commission Australia (Research Report)http://www.ntc.gov.au/filemedia/Reports/FatiPerfHVDayNightDec2004.pdf.