Do sleep, stress, and illness explain daily variations in fatigue? A prospective study

Journal of Psychosomatic Research 76 (2014) 280–285

Contents lists available at ScienceDirect

Journal of Psychosomatic Research

Do sleep, stress, and illness explain daily variations in fatigue?A prospective study☆

Torbjörn Åkerstedt a,b,⁎, John Axelsson b, Mats Lekander a,b, Nicola Orsini c, Göran Kecklund a,b

a Stress Research Institute, Stockholm University, Stockholm, Swedenb Department of Clinical Neuroscience, Karolinska Institute, Stockholm, Swedenc Institute for Environmental Medicine, Karolinska Institute, Stockholm, Sweden

☆ This study was supported by the Swedish Science CSocial Sciences and Working Life and Stockholm Stress Ce⁎ Corresponding author at: Stress Research Institute

Stockholm, Sweden. Tel.: +46 8 5537 8947; fax: +46 8 5E-mail address: [email protected] (T. Åke

0022-3999/$ – see front matter © 2014 Elsevier Inc. All rihttp://dx.doi.org/10.1016/j.jpsychores.2014.01.005

a b s t r a c t

Article history:

Keywords:TSTDailySleep qualityStressMixed modelSubjective healthSRHKSS

Objective: Fatigue is related to a number of serious diseases, as well as to general well-being. It is also a majorcause of sickness absence and use of health facilities. Still, the determinants of variations in fatigue are littleinvestigated. The purpose of present study was to investigate the relationships between the daily variations offatigue with sleep during the previous night, stress or disease symptoms during the same day — across 42 con-secutive days of normal life.Methods: 50 individuals participated and gave diary reports and used an actigraph across the 42 days. The datawas analyzed using a multilevel approach with mixed model regression.Results: The analyses showed that the day-to-day variation in fatigue was related to (poor) sleep quality (p b .001)and (reduced) sleep duration (p b .01) the previous night, aswell as to higher stress (p b .05), and to the occurrenceof a cold or fever (p b .001) during the same day as the fatigue rating. Fatigue was also strongly related to poorersubjective health (p b .001) and sleepiness (p b .001) during the same day.Conclusion: The results indicate that prior sleep (and sleepiness) as well as stress and illness are consistentlyconnected to how fatigue is experienced during normal living conditions.

© 2014 Elsevier Inc. All rights reserved.

Introduction

Fatigue is a common medical symptom in a wide range of diseases[1,2], and a central characteristic in the chronic fatigue syndrome [3],burnout [4], depression [5], and insomnia [6]. Fatigue is also an integralpart of sickness behavior in relation to a number of inflammatory states[7]. For example, administration of a (typhoid) vaccine that causes anincrease in interleukin (IL-6) levels is paralleled by increased fatigue[8]. Along the same line, observational studies show that poor subjectivehealth, which is closely associated with fatigue, is related to both low-grade inflammation [9,10] and short sleep [11–13].

Because of its inhibitory and negative subjective quality, and itscentral role in many diseases, fatigue is also an important predictor ofconsumption of medical resources [14], lack of self-care ability [15],sickness absence [16–18], as well as of lack of return to work afterlong term sickness absence [17,18].

Outside the clinical area, fatigue is associated with high workdemands, heavy work, long hours, female gender and low age [19–22].

ouncil, the Swedish Council fornter., Stockholm University, 10691537 8900.rstedt).

ghts reserved.

Fatigue is also associated with stress, sleep problems [23–25] and exper-imental sleep deprivation [13,26]. The prevalence of fatigue varies greatlydepending on how it is measured, but, as an example, the prevalence of“often being fatigued during the last two weeks”was 32.8% in a nationalrepresentative sample of 58,000 Swedes [21].

Despite its apparent importance, the fatigue construct is not clearlydefined, although it often refers to a state of energy depletion [27–29].No physiological indicators exist, only questionnaires and rating scales.These include typical items like “fatigued”, “tired”, and “exhausted” [30]and the most used fatigue scale is probably “The fatigue severity scale”(FSS) [31]. However, most items in this scale describe effects of fatiguewhen fatigue is actually present, not how often it is present. It is, hence,not possible to estimate an amount or severity of fatigue in general.

Sleepiness, while highly related to fatigue, is usually not includedamong types of fatigue since it is conceptually different [32,33]. Thus,sleepiness reflects a drive to fall asleep [34], rather than a state of energydepletion. The relation between the two concepts has been a topic forrepeated discussions [27,28] but empirical studies of their relation arerare.

Available studies on fatigue have mainly been of a cross-sectionalnature although some of the studies above have been longitudinal andspanned long time periods. However, it is a common experience that fa-tigue varies between days although documentation seems to be lacking.If established, it seems reasonable to believe that a day-to-day variationmay be linked to variations in stress or disease the same day or to the

281T. Åkerstedt et al. / Journal of Psychosomatic Research 76 (2014) 280–285

adequacy of the immediately preceding sleep episode. No previousstudies exist, however. A better understanding of such links may giveimportant insights into the day-to-day variations in work performance,well-being, health care consumption, quality of life, and other phenom-ena. An investigation of how the daily variations of fatigue relate tosleep/health/stresswould require dailymeasures across a longer periodof time. Such an approach has been used for investigating the effect ofthe daily variation of sleep on the daily variation of mood across twoweeks [35]. Such a longitudinal approach requires multilevel modelingto handle the longitudinal covariation involved.

The aim of the present study was to investigate how fatigue duringthe day (rated at the end of the day) would be related to: sleep duringthe prior night (measured with actigraphy and a sleep diary in themorning), stress and illness (fever, cold) and subjective health duringthe same day and across a period of 42 days.We believe that the resultsof such a study would contribute.

Methods

A total of 50 subjects participated. The participants were recruitedthrough advertisements and personal contacts. The intention was toobtain a normal, healthy sample. Exclusion was based on questions onwhether the participant suffered from depression, anxiety disorders, in-somnia, cardiovascular disease, hypersomnia, diabetes, and diseases re-quiring regular medication. The response alternatives were: no/yes/yesbut more than a year ago. This was complemented by a short anamnes-tic screening by a physician, but no subject was excluded on medicalgrounds, probably because the health requirements were carefullyexplained in the invitation to participate.

The participants received an economic compensation of ap-proximately $180. The ethical committee of the Karolinska Insti-tute approved the study. All participants gave written informedconsent and the study was carried out according to the principlesof the Declaration of Helsinki.

The study covered one entire year and for each individual data wascollected daily over a period of 6 weeks. At the start a background ques-tionnaire was filled out. Each morning a sleep diary with sleep qualityratings were filled out and during the day sleepiness and stress wererated every 3 h during wakefulness. Every evening during the studythe participants filled out a diary reporting on health symptoms, includ-ing fatigue, subjective health, having had a fever, and having had a coldduring the day. During the entire measurement period, the subjectswore an actigraph for sleep recording.

Measurements

The questionnaire included questions about age, gender, mar-riage status (cohabitation/single, number of children b7 years oldliving at home, housing (own house/apartment — 1/0), education(university/high school/less than high school), employment(employed/unemployed/student), white collar/blue collar work,work hours (day work/shift work (roster) with nights, shiftwork (roster) without nights/permanent morning work, perma-nent evening work/other), smoking (yes/no), alcohol consump-tion (never/occasional/2–4 times per month/2–3 times perweek/4 days per week or more, exercise (nothing/seldom/lightexercise 1–2 times per week/N2 times per week and becomessweaty/serious exercise – almost competition level), pain killers(never/occasionally/once per week, 2–3 times per week/almostevery day). Depression and anxiety were measured through theHospital Anxiety and Depression Scale [36,37].

For fatigue measurement once a day the fatigue severity scale wasunsuited due to its more trait-like items. It is also relatively long foruse every day. Instead, we used a scale that quantified the amount of fa-tigue during the present day using the items: persevering fatigue, full ofenergy (reverse coded), mentally exhausted, and physically exhausted

[38,39]. Cronbach's alpha in that studywas 0.86 and fatigue fellmarked-ly from high levels at the start of the study after being sick-listed forlong-term exposure to stress [39]. The score ranged from 1 (not at all)to 5 (to a high degree) and theCronbach alpha of the scale in thepresentstudy was 0.83. The participants were instructed to fill out this scaleeach evening before bedtime. At the same time they reported theirself-rated health (SRH) (How would you rate your state of health forthe day? 1–7; very poor–excellent) [9], as well as occurrence of feverand/or a cold (scale 1–5, “not at all” to “to a large extent”). The latterconstituted the “illness score”.

Stress and sleepiness were rated every third hour during the timeawake on nine-graded Likert scales. The stress scale, ranged from 1(no stress at all) to 9 (maximum stress imaginable) [40] and has beenused in a previous study of daily stress [41]. Sleepiness was rated onthe Karolinska Sleepiness Scale [42]. This scale ranges from 1 (veryalert) to 9 (very sleepy, fighting sleep, an effort to keep awake). In thepresent study a mean was calculated for each day based on the intervalbetween 8 and 22 h.

The Karolinska sleep diary (KSD) [43], was filled out in the morningand included: bedtime (h), time of awakening (h), sleep latency (h),sleep quality (how did you sleep?, very well 5–very poorly 1), feelingrefreshed after awakening (completely 5–not at all 1), calm sleep(very calm 5–very restless 1), did you get enough sleep? (definitelyenough 5–definitely too little 1), ease of waking up (very easy 5–verydifficult 1), and ease of falling asleep (very easy 5–very difficult 1).Four items formed a sleep quality index (SQI): sleep quality, calmnessof sleep, ease of falling asleep, and sleep throughout the allotted time.Using bedtime, time of awakening and sleep latency, a measure ofdiary total sleep time (TST) was derived for each day.

Actigraphy was measured using the Cambridge Actiwatch®actigraph. It was attached to the non-dominant wrist and emptied andreloaded every three weeks. The actigraph is essentially a piezoelectriccrystal that senses acceleration in three dimensions and has a thresholdover which “activity” is scored. No or little activity is scored as sleepusing a proprietary algorithm that takes into account not only the activ-ity but also the pattern of activity. Actigraphs in general have a reason-able validity and reliability [44] and the actigraphy in this study has acorrelation of r = 0.70 with polysomnographical recording of totalsleep time (TST) [45]. Also bedtime and time of rising were obtainedthrough the actigraph. The subjects were instructed to use the eventmarker at lights out and at rise time.

Statistical analysis

The design of the present study involved an analysis of the longitu-dinal covariation between variables. One may envision the analysis asa traditional regression analysis for one individual. Thus, for day onethere is a pair of data with, for example, a sleep quality for the elapsednight in one column and fatigue reported in the evening later thesame day. This is then repeated for day 2 in the next row, for day 3 inthe row after that, up to day 42. Sleep quality is then regressed onfatigue for that individual and a regression coefficient is obtainedrepresenting the change in the dependent variable as a function of thechange in the independent variable (Fig. 1). When the regression coef-ficient is averaged across all individuals a mean regression coefficient(with standard errors) is obtained that can be tested for significance.However, since therewill be serial dependencies across the 42measure-ment points as well as collinearity between some independentvariables, and a second level of analysis (between groups) is added,another approach than classical regression is necessary. For this purposeamultilevel analysis was used [46]. This kind of approach has been usedto, for example, demonstrate the daily covariation of mood and sleepacross two weeks [35].

All statistical analyses were performed with the statistical packageStata 11 (StataCorp, College Station, USA) [47]. Linear mixed effect

1

2

3

4

5F

atig

ue

1 2 3 4 5Sleep Quality

1

2

3

4

5

Fat

igue

1 2 3 4 5Illness

1

2

3

4

5

Fat

igue

1 2 3 4 5Illness

1

2

3

4

5

Fat

igue

1 2 3 4 5 6 7 8 9Sleepiness

Fig. 1. Estimates of regressions between fatigue and sleep quality, illness, subjective health and subjective sleepiness, respectively. Groupmean= fat, black line. Individual regression lines =thin, gray lines.

Table 1Means ± SD or % for background variables.

Variable Mean ± SD or % Other information

Age 43.5 ± 16.4 Range: 18–61Females 59%Married/cohabiting 68%Children b7 years old 28%Own house 51%University education 30%Employed 75%Blue collar workers 66%Daytime work 97% Occasional evenings: 3%Smokers 25%Moderate/heavy exercise 81.3% Never or rarely: 18.7%Alcohol 2–4 times/month 84.4%Rather good/good health 71.9% Rather/very poor: 3.1%Pain killers ≥1/week 5.1%BMI 24.5 ± 0.45 Range: 20.5–29.7

i= index; TST= actigraphy total sleep time; SRH= subjective health; clock time is givenin decimal notation.

282 T. Åkerstedt et al. / Journal of Psychosomatic Research 76 (2014) 280–285

models [48] were estimated with the Stata procedure “xtmixed”. Lessthan 5% of the data were missing.

To evaluate the effect of individual differences in slope and intercept,a set of linearmixed effect regressionmodels, including a random effectover the intercept to allow for variation in level and slope between sub-jects,was estimated for each predictor against sleepiness. Thiswas donefor each predictor separately in a univariate model, as well as simulta-neously in amultivariatemodel. The latter reflects the effect of each var-iable, with all the other variables adjusted for. The output for fixedeffects includes a mean regression coefficient (β) from the individualcoefficients, a mean intercept (“constant”) with the Y-axis (fatigue),the inferential statistic “Z”, and its p-value. In addition, the applicationof random effects permits the regression coefficients to vary betweenindividuals. Also, the intercept with the Y-axis is permitted to vary.The key statistic her is the mean regression coefficient since it repre-sents the relation between one predictor and fatigue across the42 days. The subject-specific predictions were calculated adding to thefixed-effects (population average), the contribution of the best linearunbiased predictions (BLUPs) of the random effects, often referred toas a “semi-Bayes” prediction. All analyses were adjusted for the effectof linear time (days). None of these effects became significant, however,and are not entered into the tables. The main focus of the analysis wasthe prediction of fatigue from illness, sleep, stress and subjective health,with age, gender, and depression as level 2 (between groups) factors.

Results

Table 1 presents the mean ± SD of the background variables. The age range was18–91 years, somewhat more women than men participated, themajority was employed(with day work), and thereweremore blue than white collar workers. Most were in goodhealth, 1/4 were smokers, most exercised, few used pain-killers and the BMI rangedbetween normal and overweight.

Table 2 shows the univariate relations between fatigue and the predictors. As shownby the regression coefficients in column 3, fatigue increased with worse sleep quality andshorter total sleep time and also with stress during the day. The strength of the relationswas relatively modest. Fatigue was also strongly related to illness, increasing with 0.31units (β coefficient) per unit of illness. It was also strongly related to poorer subjectivehealth, at the rate of 0.26 fatigue units per unit of subjective health. Fatigue decreasedwith age and increased with depression. The relation between fatigue and sleep quality,illness, subjective health, and sleepiness are depicted in Fig. 1 (unadjusted). For illness, in-dividuals exhibited very similar regression lines, whereas there is more individual disper-sion for subjective health.

Table 3 shows the results of themultivariate analysis (all variables entered at the sametime and thus adjusted for) of all significant predictors from Table 2. All previously signif-icant longitudinal predictors retained a significant contribution, in particular subjective

Table 2Results from univariate mixed model regression predicting fatigue.

Sleepiness Mean ± SD Fixed effect coeff β SEM Y-intercept SEM Z, p 95% CI for β

Time of awakening (time) 7.7 ± 1.9 −.030 .007 4.32 .08 4.3c .016/.043TST (hours–decimal) 7.1 ± 1.4 −.024 .006 4.22 .10 4.4c −.012/−.066iSleep quality (1–5 high) 4.1 ± 1.3 −.165 .016 3.88 .09 10.4c .134/.024Stress average (1–9 high) 2.1 ± 1.2 .108 .014 4.75 .06 8.0c −134/− .081iIllness (1–5 high) 4.8 ± 0.49 −.307 .025 3.05 .13 12.7c .259/.354SRH (1–7high) 5.4 ± 1.3 −.256 .009 3.14 .07 45.9.c .239/.274Age 43.5 ± 16.4 −.010 .004 4.12 .16 2.8b .003/.016Gender (M/F, 1/0) 44% −.226 .122 4.43 .08 1.9 −.014/.444Depression 3.1 ± 2.6 .093 .020 4.81 .08 4.6c − .132/− .052

“i” indicates index. Fixed effect coefficient: the regression coefficient for the fixed effect; SEM = standard error of the mean, Z = z value for the regression; p = significancelevel, where a = p b .05, b = p b .01, c = p b .001 and d = p b .0001. Y-intercept; SD random = standard deviation of the intercepts (random effect). TST = total sleep time.SRH = subjective health. SEM = standard error of the mean.

283T. Åkerstedt et al. / Journal of Psychosomatic Research 76 (2014) 280–285

health. Age and depression retained their significant relations without beingmuch affect-ed. Model FI was: Chi2 = 298.2, p b .001.

The relation between sleepiness and fatiguewas not a main purpose of this paper butitwas of interest to investigate the relation between the two variables since the two termsare often used alternatively. Thus, a post-hoc analysis showed that sleepinesswas a signif-icant predictor of fatigue, with a coefficient = 0.25 ± .03 (Z = 9.5, p b .0001; and a con-stant = 5.59 ± .08 (95% confidence interval = .30–.20)). This means that for each unitof change on the sleepiness scale (1–9) fatigue increased by 0.25 units. Thus, a changefrom 1 to 9 of sleepiness would be related to a change of 2.0 on the fatigue scale (range1–5).

Discussion

The results show that the day-to-day variation in fatiguewas relatedto the day-to-day variation of sleep, stress, illness and subjective healthacross 42 days, particularly the latter two variables.

The link between sleep loss and fatigue has been demonstrated incross-sectional studies [22–24] and in studies of experimental sleep re-striction [13], but the present study shows that this relation is presentprospectively across days. The effect of sleep duration was modest, butthat of sleep quality was stronger, 0.17 fatigue units per unit of quality(1–5 scale). This corresponds to 0.85 fatigue units across the entiresleep quality scale. It is possible that the modest relation for the sleepvariables was due to the lack of poor/short sleep in the sample. The re-lationwith sleepmay have beenmore dramatic and had periods of largeamounts of sleep loss or sleep problems which occurred during the6 weeks of data collection. However, the relations are probably reason-able for a group of subjects with relatively normal sleep.

It should be emphasized that the sleep variables were rated in themorning and preceded the evening rating of fatigue by 15–16 h, sug-gesting that much of the relation may have been causative, i.e. thatpoor sleep resulted in fatigue the same day. This causality is also

Table 3Results from multivariate mixed model regression predicting fatigue.

Predictors Fixed effect coeff β SEM Z, p 95% CI for β

Time of awakening (h) −.008 .007 1.2 −.023/.006TST (hours) −.012 .002 2.6b .004/.024iSleep quality (1–5 high) −.096 .016 6.1c .065/.127Stress average (1–9high) .035 .014 2.5a − .062/− .007iIllness .070 .026 2.7b .020/.121SRH (1–7) −.221 .012 19.0c .199/.244Age (years) −.011 .003 3.8c .005/.017Depression HAD .050 .018 2.8b − .086/− .014Y-intercept 2.23 .206 10.8c

“i” indicates index. Fixed effect coefficient: the regression coefficient for the fixedeffect; SEM = standard error of the mean, Z = z value for the regression; p =significance level, where a = p b .05, b = p b .01, c = p b .001 and d = p b .0001.Y-intercept; SD random = standard deviation of the intercepts (random effect).TST =total sleep time. SRH = subjective health. SEM = standard error of the mean.

supported by experimental studies, showing that total [26] or partial,sleep deprivation [13] cause subsequent fatigue. Future studies mightshed more light on the sleep/fatigue relation by introducing occasionalshort sleeps, for example, through asking participants to rise 2–3 hearlier than usual. One might also follow occupational groups withnight or morning work.

The connection between stress ratings during the day and the eve-ning rating of the day's fatigue was expected from previous clinical orepidemiological cross sectional studies [4,19–25]. The present work,however, again adds the connection between the two longitudinallyacross days; higher stress during the day predicts higher fatigue duringthe subsequent evening. The size of the relation was moderate but itshows that for each unit of stress (scale 1–9) fatigue increased0.11 units, that is, almost one unit of fatigue across the entire stressscale. It should be pointed out that the range of stress in the study wasmodest. High levels of stress across a whole day rarely occurred in thissample of normal individuals. Thus, fatigue at bedtime after days withstrong stress remains to be investigated in future studies. The ratingsof stress were made every 3 h and the fatigue rating was made just be-fore bedtime, whichmay suggest a causal connection. Still, the twomayalso be parallel phenomena. The mechanisms behind a stress-fatiguelink are not known, but the use of local brain areas (e.g. those engagedin stress responses) depletes the energy levels and determines theamount of restitution needed and thus the homeostatic pressure forsleep [49].

The illness index, i.e. the occurrence of fever and/or a cold, wasstrongly linked to fatigue. The coefficient was 0.31 in the univariateanalysis, that is, fatigue increased with 0.31 units (scale 1–5) witheach unit of illness (scale 1–5). This relation is in linewith the role of fa-tigue in sickness behavior [7]. The latter refers to the coordinated set ofmotivated behaviors during infection to conserve energy to increase thechance for survival. During sickness, interleukin-1 in thebrain is indicat-ed to have a pivotal role for fatigue, as assessed bybehavioral tests in an-imals [7]. It is thus reasonable that the presence of a fever or a cold, wascausally related to the concomitant high fatigue ratings. The contribu-tion of the present study is, again, the demonstration of a connectionacross days. It is also evident from the figure that the regression coeffi-cients are rather similar across individuals. Thus, it is difficult not to con-clude that increased fatigue emerged as a consequence of illness.Indeed, a consistent finding in investigations of sickness is the reducedlocomotor activity, suggesting the presence of a more generalizedmotor inhibitory response [50]. Togetherwith a re-orientedmotivation-al drive towards rest and anhedonia in response to inflammation [7],these findings speak, together with the day-to-day variations as ob-served in the present study, for a strong connection between illnessand fatigue. Comparisons of strength between variablesmay bedifficult,but it seems clear that illness, when present, is a considerably strongerfactor in fatigue than sleep loss or stress — at least under the rathermodest levels of the two latter observed in the present study. Future

284 T. Åkerstedt et al. / Journal of Psychosomatic Research 76 (2014) 280–285

studies should include immune parameters in naturalistic designs toclarify a possible mediating role with fatigue. Experimental viral infec-tion (common cold) might also be used to clarify causality.

Subjective health reduced the univariate coefficients of all the otherpredictors, in particular illness. This outcome seems logical since subjec-tive health and fatigue are coupled to the sickness response [51]. Subjec-tive health clearly takes precedence over illness in the prediction offatigue, possibly because of its inclusive character that may captureboth illness and other factors associated with fatigue, such as sleep du-ration and sleep quality. Thus, all these factors showed reduced coeffi-cients when subjective health was added to the models. This suggestsa relation between the three variables, although little empirical dataexist. The exceptions include a recent study of restricted sleep acrossmany days, showing pronounced dose–response effects [13]. Otherstudies, but not the study by Lekander et al., show that proinflammatorycytokines, involved in sickness behavior, may increase during sleep loss[52].

With respect to sleepiness, the post-hoc analysis clearly demonstrat-ed that it has a close relation to fatigue, which is in line with previousobservations [27,28,32,33]. As with the other relations studied in thispaper the exact fit of the regression line might have become differenthad more severe sleepiness or fatigue occurred. The occurrence ofmore severe levels may also have revealed if the two variables mayhave dissociated at some point. Since low levels of sleepiness is labeled“very alert” or “alert” low levels of fatigue items are labeled “not at all”that end of the scales is very similar. The high end, however, is labeled,for example, “totally agree” for exhaustion and “very sleepy, fightingsleep, an effort to keep awake” for sleepiness. It seems reasonable to ex-pect considerable dissociation at the high end. This remains to be dem-onstrated, however.

The present study has several limitations. One concerns the conve-nience type of sample. However, it was a study of intraindividual covari-ation between variables, which should reduce the confounding effectsfrom the selection procedure. The present group was, although raterheterogeneous, essentially healthy, without any insomniacs or clinicallydepressed individuals. Thus, the generalization to clinical groups orother age groups should be avoided. Another limitation is that the pres-ent study was naturalistic, even if prospective, and it is difficult to attri-bute cause without experimentation and randomization. On the otherhand, the interest in the present study was specifically on the ecologi-cally valid day-to-day spontaneous covariation and prediction. An ex-perimental design would have answered another question and wouldnot have been realistic across a longer period of time, such as the pres-ent 42 days. A third point is that physical activity was not included inthe study. It is a reasonable assumption that such a variable wouldhave contributed towards explaining the day-to-day variations of fa-tigue. One may also discuss whether the drawn-out data collection pe-riod may have involved seasonal effects. Since the statistical analysisessentially is a form of regression within individuals, seasonal influ-ences should not have affected the regression coefficients.

In conclusion, the present study has shown that the day-to-day var-iation in fatigue is related to last night's sleep quality and sleep duration,to the stress during the day, and to the occurrence of illness (or reducedsubjective health). The findings may be of importance in understandingfatigue variations in working life and in clinical settings.

Conflict of interest

TÅ has received support from AstraZeneca. Otherwise no conflicts ofinterest have been declared.

References

[1] Wessely S, Chalder T, Hirsch S,Wallace P, Wright D. The prevalence andmorbidity ofchronic fatigue and chronic fatigue syndrome: a prospective primary care study. AmJ Public Health 1997;87:1449–55.

[2] Chalder T, Neeleman J, Reme SE, Power M, Wessely S. Factors associated with acutefatigue in primary care. Psychol Med 2010;40:1289–95.

[3] Sullivan PF, Pedersen NL, Jacks A, Evengard B. Chronic fatigue in a populationsample: definitions and heterogeneity. Psychol Med 2005;35:1337–48.

[4] Maslach C, Leiter MP. Stress and burnout: the critical research. In: Cooper CL,editor. Handbook of stress medicine and health. Lancaster, UK: CRC Press;2005. p. 155–72.

[5] APA. Diagnostic and Statistical Manual of Mental Disorders. Fourth Edition.(DSM-IV). Washington D.C.: American Psychiatric Association; 1994

[6] Lichstein KL, Means MK, Noe SL, Anguillard RN. Fatigue and sleep disorders. BehavRes Ther 1997;35:733–40.

[7] Dantzer R,O'Connor JC, FreundGG, JohnsonRW,Kelley KW. From inflammation to sick-ness and depression: when the immune system subjugates the brain. Nat Rev2008;9:46–56.

[8] Harrison NA, Brydon L, Walker C, Gray MA, Steptoe A, Dolan RJ, et al. Neural originsof human sickness in interoceptive responses to inflammation. Biol Psychiatry2009;66:415–22.

[9] Lekander M, Elofsson S, Neve I-M, Hansson L-O, Unden A-L. Self-rated health is relat-ed to levels of circulating cytokines. Psychosom Med 2004;66:559–63.

[10] Christian LM, Glaser R, Porter K, Malarkey WB, Beversdorf D, Kiecolt-Glaser JK.Poorer self-rated health is associated with elevated inflammatory markers amongolder adults. Psychoneuroendocrinology 2011;36:1495–504.

[11] Steptoe A, Peacey V,Wardle J. Sleep duration and health in young adults. Arch InternMed 2006;166:1689–92.

[12] Ingre M, Kecklund G, Akerstedt T, Soderstrom M, Kecklund L. Sleep length as afunction of morning shift-start time in irregular shift schedules for traindrivers: self-rated health and individual differences. Chronobiol Int2008;25:349–58.

[13] Lekander M, Andreasson AN, Kecklund G, Ekman R, Ingre M, Akerstedt T, et al.Subjective health perception in healthy young men changes in response to ex-perimentally restricted sleep and subsequent recovery sleep. Brain BehavImmun 2013 (e-publication).

[14] Watt T, Gronenvold M, Bjorner JB, Noerholm V, Rasmussen N-A, Bech P. Fatigue inthe Danish general population. Influence of sociodemographic factors and disease.J Epidemiol Community Health 2000;54:827–33.

[15] Visser MRM, Smets EMA. Fatigue, depression and quality of life in cancer patients:how are they related? Support Care Cancer 1998;6:101–8.

[16] Akerstedt T, Kecklund G, Alfredsson L, Selen J. Predicting long-term sickness absencefrom sleep and fatigue. J Sleep Res 2007;16:341–5.

[17] Akerstedt T, Kecklund G, Selen J. Disturbed sleep and fatigue as predictors of returnfrom long-term sickness absence. Ind Health 2010;48:209–14.

[18] Bultmann U, Huibers MJ, van Amelsvoort LP, Kant I, Kasl SV, Swaen GM. Psycholog-ical distress, fatigue and long-term sickness absence: prospective results from theMaastricht Cohort Study. JOEM 2005;47:941–7.

[19] Bültmann U, Kant IJ, van den Brandt PA, Kasl SV. Psychosocial work characteristics asrisk factors for the onset of fatigue and psychological distress: prospective resultsfrom the Maastricht Cohort Study. Psychol Med 2002;32:333–45.

[20] Bültmann U, Kant I, van Amelsvoort LGPM, Kasl SV. Differences in fatigue andpsychological distress across occupations: results from the Maastricht CohortStudy of fatigue at work. JOEM 2001;43:976–83.

[21] Åkerstedt T, Fredlund P, GillbergM, Jansson B.Work load andwork hours in relationto disturbed sleep and fatigue in a large representative sample. J Psychosom Res2002;53:585–8.

[22] Åkerstedt T, Knutsson A,Westerholm P, Theorell T, Alfredsson L, Kecklund G. Mentalfatigue, work and sleep. J Psychosom Res 2004;57:427–33.

[23] Thorsteinsson EB, Brown RF. Mediators and moderators of the stressor–fatigue rela-tionship in nonclinical samples. J Psychosom Res 2009;66:21–9.

[24] Theorell-Haglow J, Lindberg E, Janson C. What are the important risk factors fordaytime sleepiness and fatigue in women? Sleep 2006;29:751–7.

[25] Bultmann U, Kant I, Kasl SV, Beurskens AJ, van den Brandt PA. Fatigue and psycho-logical distress in the working population: psychometrics, prevalence, and corre-lates. J Psychosom Res 2002;52:445–52.

[26] Fröberg J, Karlsson CG, Levi L, Lidberg L. Circadian variations of catecholamineexcretion, shooting range performance and self-ratings of fatigue during sleepdeprivation. Biol Psychol 1975;2:175–88.

[27] Dement WC, Hall J, Walsh JK. Tiredness versus sleepiness: semantics or a target forpublic education? Sleep 2003;26:485–6.

[28] Horne J. The semantics of sleepiness. Sleep 2003;26:763 [author reply 4].[29] Rosenthal TC, Majeroni BA, Pretorius R, Malik K. Fatigue: an overview. Am Fam

Physician 2008;78:1173–9.[30] Shahid A, Shen JH, Shapiro CM.Measurements of sleepiness and fatigue. J Psychosom

Res 2010;69:81–9.[31] Krupp LB, Larocca NG, Muir-Nash J, Steinberg AD. The fatigue severity scale. Arch

Neurol 1989;46:1121–3.[32] Shen J, Barbera J, Shapiro CM. Distinguishing sleepiness and fatigue: focus on

definition and measurement. Sleep Med Rev 2006;10:63–76.[33] Hossain JL, Ahmad P, Reinish LW, Kayumov L, Hossain NK, Shapiro CM. Subjective fa-

tigue and subjective sleepiness: two independent consequences of sleep disorders?J Sleep Res 2005;14:245–53.

[34] Dement WC, Carskadon MA. Current perspectives on daytime sleepiness: the issues.Sleep 1982;5:56–66.

[35] McCrae CS, McNamara JP, Rowe MA, Dzierzewski JM, Dirk J, Marsiske M, et al. Sleepand affect in older adults: using multilevel modeling to examine daily associations.J Sleep Res 2008;17:42–53.

[36] Zigmond AS, Snaith RP. The Hospital Anxiety and Depression Scale. Acta PsychiatrScand 1983;67:361–70.

285T. Åkerstedt et al. / Journal of Psychosomatic Research 76 (2014) 280–285

[37] Lisspers J, Nygren A, Söderman E. Hospital Anxiety and Depression Scale(HAD): some psychometric data for a Swedish sample. Acta Psychiatr Scand1997;96:281–6.

[38] Ekstedt M, Söderström M, Åkerstedt T, Nilsson J, Sondergaard H-P, Perski A. Dis-turbed sleep and fatigue in occupational burnout. Scand J Work Environ Health2006;32:121–31.

[39] Ekstedt M, Soderstrom M, Akerstedt T. Sleep physiology in recovery from burnout.Biol Psychol 2009;82:267–73.

[40] Wang J, Rao H, Wetmore GS, Furlan PM, Korczykowski M, Dinges DF, et al. Perfusionfunctional MRI reveals cerebral blood flow pattern under psychological stress. ProcNatl Acad Sci 2005;102:17804–9.

[41] Dahlgren A, Kecklund G, Akerstedt T. Different levels of work-related stress and theeffects on sleep, fatigue and cortisol. Scand J Work Environ Health 2005;31:277–85.

[42] Åkerstedt T, Gillberg M. Subjective and objective sleepiness in the active individual.Int J Neurosci 1990;52:29–37.

[43] Åkerstedt T, Hume K, Minors D, Waterhouse J. Good sleep — its timing and physio-logical sleep characteristics. J Sleep Res 1997;6:221–9.

[44] Sadeh A. The role and validity of actigraphy in sleep medicine: an update. Sleep MedRev 2011;15:259–67.

[45] Lichstein KL, Stone KC, Donaldson J, Nau SD, Soeffing JP, Murray D, et al. Actigraphyvalidation with insomnia. Sleep 2006;29:232–9.

[46] Raudenbush SW, Bryk AS. Hierarchical linear models. Applications and data analysismethods. Second ed. Thousand Oaks, USA: Sage Publications, Inc.; 2002.

[47] StataCorp. Stata Statistical Software: Release 8.0. College Station, TX: StataCorp LP;2003.

[48] Rabe-Hesketh S, Skrondal A. Multilevel and Longitudinal Modeling Using Stata. FirstEdition. Texas: Stata Press; 2005.

[49] Vassalli A, Dijk DJ. Sleep function: current questions and new approaches. Eur JNeurosci 2009;29:1830–41.

[50] Lockey AJ, Kavaliers M, Ossenkopp KP. Lipopolysaccharide reduces tactile startle re-sponse magnitude but not prepulse inhibition in rats: a dose–response examination.Pharmacol Biochem Behav 2009;93:47–53.

[51] Andreasson AN, Szulkin R, Unden AL, von Essen J, Nilsson LG, LekanderM. Inflamma-tion and positive affect are associated with subjective health in women of the gen-eral population. J Health Psychol 2013;18:311–20.

[52] Vgontzas AN, Zoumakis E, Bixler EO, Lin HM, Follett H, Kales A, et al. Adverse effectsof modest sleep restriction on sleepiness, performance, and inflammatory cytokines.J Clin Endocrinol Metab 2004;89:2119–26.