Cortisol secretion and fatigue: Associations in a community based cohort

Cortisol secretion and fatigue: Associations in acommunity based cohort

Meena Kumari a,*, Ellena Badrick b, Tarani Chandola a, Emma K. Adam c,Mai Stafford a, Michael G. Marmot a, Clemens Kirschbaumd, Mika Kivimaki a

aDepartment of Epidemiology and Public Health, UCL, 1-19 Torrington Place, London, WC1E 6BT, United KingdombCentre for Health Sciences, Barts and the London School of Medicine and Dentistry, Abernethy Building, London,E1 2AT, United Kingdomc School of Education and Social Policy and Institute for Policy Research, Northwestern University,2120 Campus Drive, Evanston, IL 60208, USAdBiological Psychology, Technical University of Dresden, D-01062 Dresden, Germany

Received 2 December 2008; received in revised form 27 March 2009; accepted 3 May 2009

Psychoneuroendocrinology (2009) 34, 1476—1485

KEYWORDSSalivary cortisol;Fatigue;Cohort study;Longitudinal;Epidemiology

Summary The association between fatigue and reduced activity in the hypothalamo-pitui-tary—adrenal (HPA) axis has been described. However the temporal association between fatigueand HPA activity is under debate. We examine whether alterations in cortisol secretion play a rolein the development of fatigue or whether changes occur later as a consequence of fatigue in alongitudinal cohort study of 4299 community dwelling adults (mean age 61). Cortisol secretionwas measured from saliva samples collected waking, waking + 0.5, 2.5, 8, 12 h and bedtime atphase 7 (2003—2004) of the Whitehall II study. Fatigue was measured at phase 6 (2001), phase 7and phase 8 (2006) of the Whitehall II study. Three elements of secretion were examined: wakingcortisol, the cortisol awakening response and diurnal slope in cortisol secretion. Fatigue wasdetermined using the vitality sub-scale of the Short Form-36. A wide variety of co-variates weremeasured. We find that fatigue measured at phase 6 was not associated with cortisol secretion atphase 7. At phase 7, low waking cortisol levels and a flat slope in diurnal cortisol secretion wereassociated with fatigue independently of co-variates. In participants low or free of fatigue atphase 7 low waking cortisol and flatter slope in cortisol secretion were associated with new-onsetfatigue at phase 8 (for example, odds ratio for lowest vs. highest tertile of waking cortisol 1.50;95% confidence intervals, 1.08, 2.09 after adjusting for all co-variates). In conclusion, we findthat low waking salivary cortisol and a flat slope in cortisol secretion is associated with fatigue.Cortisol is also associated with future onset of fatigue suggesting that changes in cortisolsecretion are etiologic or occur early in the genesis of fatigue.# 2009 Published by Elsevier Ltd.

* Corresponding author. Tel.: +44 020 7679 5637; fax: +44 020 7813 0242.E-mail address: [email protected] (M. Kumari).

ava i lab le at www.sc ienced i rect .com

journa l homepage: www.el sev ier.com/locate/psyneuen

0306-4530/$ — see front matter # 2009 Published by Elsevier Ltd.
doi:10.1016/j.psyneuen.2009.05.001
mailto:[email protected]

http://dx.doi.org/10.1016/j.psyneuen.2009.05.001

Cortisol secretion and fatigue 1477

1. Introduction

Fatigue is a symptom of Addison’s disease, or primary adrenalfailure, suggesting that diminished adrenal steroid produc-tion is linked with fatigue (Werbel and Ober, 1993). A numberof clinical and case—control studies on post-traumatic stressdisorder, chronic fatigue syndrome (CFS) and fibromyalgia,which all feature fatigue as a symptom suggest that thesesyndromes have low levels of the adrenal corticosteroidcortisol (Fries et al., 2005; Van Den Eede et al., 2007; Nateret al., 2008). These findings are not apparent in all studies(Crofford et al., 2004; Di Giorgio et al., 2005). Diurnal cortisolrelease is typically characterized by high levels upon waking,an increase of some 50—60% in cortisol concentration follow-ing awakening (called the cortisol awakening response orCAR), and a subsequent decline over the remainder of theday, reaching a low point or nadir around midnight. Studiesthat fail to show diminished cortisol secretion in associationwith fatigue have generally measured morning or serumcortisol which is not analysed in relation to waking (Croffordet al., 2004; Di Giorgio et al., 2005). Recent evidence from 77patients suggests that mild hypocortisolism found in CFS canbe reversed by cognitive behavioral therapy (Roberts et al.,2009).

Research on non-clinical populations suggests cortisol tobe associated with concepts close to fatigue, such as burnoutand vital exhaustion (Appels, 1990; Melamed et al., 2006; TerWolbeek et al., 2007). Evidence of the direction of theassociation between diminished cortisol secretion and fati-gue is limited. Thus, it is unclear whether alterations incortisol secretion play a role in the development of fatigueor whether changes in cortisol develop later as a conse-quence of co-morbidities associated with fatigue and relatedsyndromes (Wessely et al., 1995; Candy et al., 2003; Cleare,2003). For example, fatigue in burnout, which is a conceptused in a work-related context, may have a different aetiol-ogy to fatigue associated with vital exhaustion which is ahealth related concept. In these cases changes in cortisolmay be related to the syndromes rather than fatigue per se.

A recent study of 77 apparently healthy participants foundcortisol secretion to be associated with fatigue assessed bythe ‘vitality’ sub-scale of the Short Form-36 scale, a measureof health related quality of life (Lindeberg et al., 2008).These data, in which low cortisol at waking and flatteneddecline (slope) in diurnal cortisol secretion was associatedwith exhaustion, suggest that the association of cortisol withfatigue may be apparent in non-patient populations beforethe onset of overt disease. The etiology of hypocortisolism isunclear and may represent an adaptation or response tochronic stress (McEwen, 1998; Fries et al., 2005). The etiol-ogy of flattened diurnal slopes in cortisol secretion is unclear.Flattened slopes in cortisol are reportedly predictive ofmortality in cancer patients (Sephton et al., 2000).

A short term longitudinal study, conducted over the courseof several days, provides initial evidence that low morningcortisol may precede fatigue symptoms in non-patient popu-lations (Adam et al., 2006). In this study, fatigue assessed theday before was not associated with diminished cortisol secre-tion the next morning, but lowmorning cortisol secretion wasassociated with increased fatigue later in the same day(Adam et al., 2006). However, the status of cortisol as a longterm predictor of future or persistent fatigue symptom

reporting has not been described in community dwellingpopulations.

In this longitudinal epidemiological study of over 4000participants, we examine cortisol secretion from saliva sam-ples collected across one single day and fatigue, assessed byscore in the vitality sub-scale of the SF-36 approximately 2.5years before, concurrently with and 2.5 years after cortisolassessment. We address two questions: first, is fatigue anindependent predictor of future derangements in cortisolsecretion in a community dwelling population? Secondly,does cortisol secretion predict future fatigue? We also exam-ine the role of persistent fatigue in these associations andtake into account a wide variety of potential confoundingfactors, including depressive symptoms, sleep disturbanceand use of medication.

2. Methods

2.1. Study population

The Whitehall II study was originally recruited as an occupa-tional cohort and was designed to examine the mechanismsby which social position impacts health. The cohort wasinitially recruited between 1985 and 1988 (phase 1) from20 London based civil service departments, 10,308 peopleparticipated (response 73%). The cohort was predominantlymale (66% men) and participants were aged between 39 and55 at baseline. Eight phases of the study have been com-pleted, details of the cohort profile have been reportedelsewhere (Marmot and Brunner, 2005). This study used datafrom phase 6 (2001), phase 7 (2002—2004) when cortisol wasmeasured and phase 8 (2006). The number of participants atphase 7 was 6941; of these 6484 (93.4%) had a clinicalassessment, the collection of saliva samples was instigatedpart way through phase 7 and of those eligible for cortisolassessment 4609 (90.1%) returned samples. In analysesreported here, fewer participants were in the lowest employ-ment grades compared to the baseline population of thestudy, however this difference was small. Ethical approvalfor the Whitehall II study was obtained from the UniversityCollege London Medical School committee on the ethics ofhuman research.

2.2. Saliva collection and cortisol analysis

The protocol for saliva collection has been described pre-viously (Badrick et al., 2007). Briefly, participants that agreed,were requested to provide six saliva samples over the course ofa normalweekday atwaking,waking + 30 min,waking + 2.5 h,waking + 8 h, waking + 12 h and bedtime. Participants wereinstructed to record the time of sample collection in aninstruction booklet (the ‘logbook’) to record information onwake time and stressful events on the day of sampling. Thesalivettes and logbook were then returned via post. Salivettedevices were centrifuged at 3000 rpm for 5 min resulting in aclear supernatant of low viscosity. Salivary cortisol levels weremeasured using a commercial immunoassay with chemilumi-nescence detection (CLIA, IBL-Hamburg, Hamburg, Germany).The lower concentration limit of this assay is 0.44 nmol/l;intra- and interassay coefficients of variance were below 8%.Any sample over 50 nmol/l was repeated.

1478 M. Kumari et al.

2.3. Data reduction for cortisol assessment

Cortisol values outside of three standard deviations from themean were removed (values above 75 nmol/l, n = 27).Despite this, there was a skewed distribution of cortisolvalues and cortisol data were transformed by natural loga-rithm. The CAR was calculated as the difference betweencortisol at time 2 and time 1. The slope of the decline incortisol levels over the day was calculated by regressingcortisol values on time after waking samples were takenfor samples 1 and 3—6. As there is evidence that the CARand slope in cortisol secretion are under different neurobio-logical control systems (Clow et al., 2004), sample 2 was notincluded to ensure that the CAR does not obscure the slopecalculation. The diurnal slope in cortisol secretion across theday was derived from a hierarchical linear model to predictlog cortisol in which measurement occasion was used as alevel 1 identifier and person as a level 2 identifier with sampletime as the independent variable and random intercepts andrandom slopes. The slope was estimated for each person asthe overall slope (which was negative) plus the level twoslope residual; lower (more negative) slopes indicate morerapid the decline in cortisol levels, whereas slope valuescloser to zero reflect flatter diurnal rhythms. Slope estimateswere exported and sex specific tertiles of all three measuresof cortisol secretion were also created for use in descriptiveanalyses. Cut points were: 12.8 and 20.1 (men) and 11.0 and18.2 (women) for waking cortisol (nmol/l); 1.6 and 11.1(men) and 2.6 and 12.7 (women) for cortisol awakeningresponse (nmol/l) and �0.136 and �0.116 (men) and�0.137 and �0.117 (women) for the decline in log cortisolacross the day (log cortisol nmol/(l h)).

2.4. Assessments of fatigue

Fatigue was assessed using the vitality sub-scale of the SF-36(Ware and Sherbourne, 1992) at phases 6—8. The questionsare: in the past 4 weeks ‘did you feel full of life?’; ‘did youhave a lot of energy?’; ‘did you feel worn out?’; ‘did you feeltired?’. All the questions required responses on Likert-typeresponse formats (e.g. 1 = ‘‘all of the time’’ to 6 = ‘‘none ofthe time’’) and the scale was constructed by summing up theresponse scores on the individual questions. This scale wasscored such that a higher score was indicative of high vitalityand a lower score of fatigue, distribution of the scale isskewed (range is 0—100, modal score = 80) and a populationnorm cut point of 50 was used to define fatigue, as previouslydescribed (Ware and Sherbourne, 1992). The percentage ofparticipants reporting fatigue was 26.5% at phase 6; 26.8% atphase 7 and 24.2% at phase 8. Between phases 6 and 7, 438(10%) participants reported fatigue at both phases. Four-hundred-thirty-eight reported fatigue both at phase 6 andphase 7, 442 participants reported fatigue both at phase 7and phase 8, and 259 participants with no fatigue at phase 7reported new fatigue at phase 8.

2.5. Co-variates

Time of waking was recorded on the day of saliva samplecollection by the participants in their logbook. CAR analysesincluded terms for time since waking. Last known civil service

employment grade was used as an index of social position,depressive symptoms were assessed by the 20 item Center forEpidemiology Studies-Depression, using 16 as a cut point(Radloff, 1977). Smoking and alcohol intake were assessedby questionnaire and categorized as current vs. non-currentsmokers (Badrick et al., 2007) and units of alcohol consump-tion per an average week (Badrick et al., 2008) as thesecategories were previously found to be related to cortisolsecretion in our cohort. Waist circumference was measuredin the standard way, high waist was defined as 80+cm inwomen and 100+cm in men. Participants were asked to ratethe most stressful event on the day of sample collection usinga question that required responses on Likert-type responseformats (e.g. 1 = ‘‘not at all stressed’’ to 5 = ‘‘the moststressed I have ever felt’’). Due to a skewed distribution inresponse, a stressful experience was categorized if partici-pants responded that they were ‘very stressed’ or ‘the moststressed I have ever felt’. Participants reported on currentcentral nervous system medication (drugs for depression,Parkinsons, dementia, anxiolytics and antipsychotics) andcardiovascular medication (antihypertension, lipid loweringdrugs, nitrates, and antiplatelets).

2.6. Statistical analysis

Population characteristics were examined by fatigue cate-gory and also by tertile of cortisol secretion using x2 tests forcategorical variables or regression analysis for continuousmeasures. Models were tested for sex interactions and werefound not to be significant and so data were combined foranalyses. We fitted several models to explore the directionof the association between cortisol and fatigue and the roleof persistent fatigue in this association. First, the associa-tion of past fatigue (phase 6) with current cortisol (phase 7)was modeled. Second, logistic regression was used to exam-ine the onset of new fatigue at phase 7 in a sub-set ofparticipants who were not fatigued at phase 6. Third, thecross-sectional association between current fatigue andcortisol (phase 7) was examined. Fourth, cortisol secretionat phase 7 was examined in those who reported high fatigueat both phases 6 and 7 and separately logistic regression wasused to examine cortisol at phase 7 in those that reportedhigher fatigue at phase 7 or at both phases 7 and 8 using nofatigue at either phases as reference. Fifth, logistic regres-sion was used to examine the association of cortisol secre-tion at phase 7 with new fatigue at phase 8 in a sub-cohortof participants who were not fatigued or at low fatigue atphase 7.

The contribution of the co-variates to the associationsbetween cortisol and fatigue was explored in multi-vari-able adjusted linear regression models by sequentiallyincluding each of the following sets of factors: collectionfactors (time of waking, stress on day of sample collec-tion), demographic factors (social position, age, sex),behavioral factors (current smoking, alcohol intake, sleepduration as linear and quadratic function, sleep distur-bance), and measures of health (CNS medication, CVDmedication, depressive symptoms, waist circumference)with final analyses including simultaneous adjustment forall the above co-variates. Analyses were performed usingMLWin (v 2.10 beta 6) and SAS software, version 9.1 (SASInstitute, Cary, NC).


3. Results

3.1. Missed samples and exclusions

From the samples returned 168 individual samples were nottaken by participants which equates to 0.55% of the totalnumber of samples expected. During analysis a total of 1004samples were not assayed for technical and other reasons(insufficient collection of saliva, loss of samples in transitbetween London and Germany). Participants taking corticos-teroid medications were removed from the analysis n = 236.This left us with 4326 participants available for analyses,following removal of those with very high cortisol values finalnumbers used of 4299.

3.2. Participant characteristics

Table 1 shows the characteristics of the population withcomplete fatigue and cortisol data at phase 7. Increased

Table 1 Participant characteristics by the vitality sub-scale of t

Demographic characteristics at phase 7Men (%)Mean age (years)Lowest employment grades (%)

SF-36 vitalityScore �50 (fatigued) at phase 6 (%)Score �50 (fatigued) at phase 8 (%)

Health behaviors at phase 7Current smoker (%)Heavy drinker (%)b

Less than 5 h sleep per typical night (%)9 h or more sleep per typical night (%)Sleep disturbance score (Jenkins scale)

Health data at phase 7Depression (%)c

CNS medication (%)CVD medication (%)High waist circumference (%)d

Cortisol data at phase 7Mean wake up time on day of sampling (h:min)Mean cortisol at wakinge

Mean cortisol awakening response (nmol/l) e

Mean decline in cortisol across the day (nmol/(l h))e

First sample collected greater than 10 min after waking (%)Stress on day of sample collection (%)

Data limited to participants with complete cortisol measures.a p values are derived from x2 analysis for dichotomized variables and f

score, waist circumference, cortisol data).b Sex specific measure of alcohol consumption.c Score of 16 or more on the CES-D.d >102 cm in men and >88 cm in women.e Cortisol data adjusted for age, employment grade and sex.

fatigue (indicated by a low vitality score) was associatedwith younger age, female sex, low social position, currentsmoking, both low and high sleep duration, depressivesymptoms and high waist circumference. When examiningthe characteristics of the population by tertile of cortisolsecretion at phase 7 (Table 2), low cortisol at wakingpredicted fatigue at phase 8 and was associated withlow sleep duration, higher waist circumference, increasedcardiovascular medication use, with sample collectionwithin 10 min of waking and early waking time. Anincreased CAR was associated with current smoking, lowsleep duration and earlier waking time. Those with a smallCAR were more likely to have collected their first salivasample later than 10 post-waking. Flat slope in cortisolsecretion at phase 7 was associated with older age, lowsocial position, higher fatigue, current smoking, low sleepduration, use of cardiovascular medication, higher waistcircumference, earlier wake time and stress on day ofsample collection.

he SF-36 at phase 7 of the Whitehall II study.

SF-36 vitality at phase 7

Above 50(not fatigued)

Less or equal 50(fatigued)

p valuea

3489 87577.7 65.8 0.000161.4 59.8 0.00018.9 12.3 0.0001

9.8 61.1 0.00018.2 55.6 0.0001

6.2 8.7 0.0111.0 10.1 0.455.7 16.5 0.00011.6 2.9 0.00018.6 13.2 0.0001

7.3 42.4 0.00013.1 13.5 0.0001

28.7 34.8 0.00421.9 32.1 0.0001

6.48 6.49 0.8414.5 13.6 0.027.20 7.01 0.63�0.130 �0.127 0.00214.5 15.7 0.385 14.5 0.0001

rom regression analyses for continuous variables (mean age, Jenkins

Table 2 Participant characteristics by tertile of waking cortisol and diurnal slope of cortisol secretion.

Waking cortisol (sex specific tertile) a

Bottom Middle Top p valueb

Demographic characteristics at phase 7Men (%) 75.2 72.0 73.5 0.15Mean age at phase 7 (%) 60.9 61.3 61.0 0.17Lowest employment grades 10.3 9.8 8.5 0.14

SF-36 vitalityScore �50 (fatigued) at phase 6 (%) 20.5 18.8 20.6 0.97Score �50 (fatigued) at phase 7 (%) 20.7 18.5 20.6 0.85Score �50 (fatigued) at phase 8 (%) 19.2 15.9 16.9 0.05

Health behaviors at phase 7Current smoker (%) 8.2 5.5 6.1 0.13Heavy drinker (%)c 9.9 10.8 12.1 0.07Less than 5 h sleep per typical night (%) 10.0 7.0 6.2 0.00019 h or more sleep per typical night (%) 2.1 1.6 2.0 0.2Sleep disturbance score (Jenkins scale) 9.6 9.4 9.6 0.67

Health data at phase 7Depressive symptoms (%)d 16.0 13.4 14.6 0.24CNS medication (%) 5.5 5.1 4.9 0.40CVD medication (%) 32.7 29.9 26.8 0.0008High waist circumference (%)e 25.9 23.5 22.2 0.003

Collection dataMean wake up time on day of sampling (h:min) 6.44 6.49 6.50 0.0001First sample collected greater than 10 min

after waking (%)11.1 14.1 19.7 0.0001

Stress on day of sample collection (%) 7.2 7.2 6.5 0.54

Cortisol awakening response (sex specific tertile)

Low Medium High p valueb

Demographic characteristics at phase 7Men (%) 73.6 74.2 73.0 0.77Mean age at phase 7 (%) 61.2 61.2 60.9 0.03Lowest employment grades 9.6 9.8 9.4 0.85


Health behaviors at phase 7Current smoker (%) 4.8 6.4 9.0 0.0001Heavy drinker (%)c 11.6 10.4 10.7 0.43Less than 5 h sleep per typical night (%) 6.3 7.7 9.6 0.0039 h or more sleep per typical night (%) 2.4 1.5 1.8 0.06Sleep disturbance score (Jenkins scale) 9.6 9.6 9.4 0.15

Health data at phase 7Depressive symptoms (%)d 15.6 14.9 13.5 0.12CNS medication (%) 4.4 5.9 5.2 0.34CVD medication (%) 30.6 30.4 29.0 0.36High waist circumference (%)e 24.8 24.2 23.1 0.14

Collection dataMean wake up time on day of sampling (h:min) 6:55 6:49 6:38 0.0001First sample collected greater than 10 min

after waking (%)21.7 13.3 9.0 0.0001

Stress on day of sample collection (%) 7.0 7.4 6.6 0.71


Slope in cortisol (sex specific tertile)

Flat slope Medium slope High slope p valueb

Demographic characteristics at phase 7Men (%) 74.3 72.6 74.1 0.81Mean age (%) 61.5 61.3 60.6 0.0001Lowest employment grades 10.8 10.1 8.2 0.008


Health behavior data at phase 7Current smoker (%) 10.4 6.9 3.5 0.0001Heavy drinker (%)c 12.0 11.0 9.9 0.09Less than 5 h sleep per typical night (%) 10.7 7.6 5.9 0.00019 h or more sleep per typical night (%) 1.5 1.6 2.5 0.01Sleep disturbance score (Jenkins scale) 9.6 9.5 9.5 0.13

Health data at phase 7Depression (%)d 15.6 14.4 14.4 0.4CNS medication (%) 5.5 4.4 5.8 0.58CVD medication (%) 33.1 30.8 26.7 0.0008High waist circumference (%)e 25.6 24.4 22.4 0.09

Collection dataMean wake up time on day of sampling (h:min) 6.31 6.47 6.59 0.0001First sample collected greater than 10 min after waking (%) 13.3 15.2 15.0 0.27Stress on day of sample collection (%) 8.3 7.1 6.0 0.02

a Cut points were: 12.8 and 20.1 (men) and 11.0 and 18.2 (women) nmol/l for waking cortisol; 1.6 and 11.1 (men) and 2.6 and 12.7 (women)for cortisol awakening response and�0.136 and�0.116 (men) and�0.137 and�0.117 (women) for the decline in log cortisol across the day.b p values are derived from x2 analysis for dichotomized variables and from regression analyses for continuous variables (mean age, Jenkins

score, waist circumference, wake up time).c Sex specific measure of alcohol consumption.d Score of 16 or more on the CES-D.e >102 cm in men and >88 cm in women.

Table 2 (Continued )


3.3. Association of fatigue assessed at phase 6(2001) with cortisol secretion at phase 7(2002—2004)

Fatigue assessed at phase 6 was not associated with cortisolexamined at phase 7 (Table 2). While fatigue at phase 6tended to be associated with lower waking cortisol (meancortisol in fatigued group vs. low fatigued group (95% con-fidence intervals); 14.0 (13.7, 14.4) vs. 13.7 (13.1, 14.3),p = 0.24) and flatter slope in cortisol secretion (fatigued

Table 3 Mean cortisol, 95% confidence intervals (nmol/l) at each2004).

Above 50 (not fatigued)

Waking 14.0 (13.8, 14.3)Waking + 30 min 20.3 (19.9, 20.8)Waking + 2.5 h 8.23 (8.05, 8.41)Waking + 8 h 5.04 (4.93, 5.16)Waking + 12 h 2.26 (2.20, 2.33)Bedtime 1.58 (1.53, 1.63)

Numbers represent mean cortisol (nmol/l) adjusted for age, sex and w

group vs. low fatigued group �0.129 (�0.130, �0.128) vs.�0.128 (�0.129, �0.126), p = 0.09) at phase 7, these asso-ciations were not statistically significant in analyses adjustedfor sex, age and employment grade at phase 6 and wakingtime at phase 7. No difference was observed for the CAR( p = 0.7). Those participants who were free of fatigue atphase 6 but reported new fatigue at phase 7 had a non-significantly lower cortisol measure at waking ( p = 0.06) anda shallower decline in cortisol secretion across the day( p = 0.05) at phase 7 compared to other participants. After

time point across the day by fatigue status at phase 7 (2002—

Less or equal 50 (fatigued) p

13.3 (12.8, 13.9) 0.0319.3 (18.5, 20.2) 0.048.47 (8.11, 8.85) 0.245.05 (4.82, 5.29) 0.962.36 (2.23, 2.49) 0.211.74 (1.64, 1.86) 0.007

aking time.

Table 4 Multi-variable adjusted mean waking cortisol and mean decline in cortisol secretion (slope) by status of fatigue at phase7.

Status of fatigue Model 1 Model 2 Model 3

Waking cortisol (nmol/l)Sf-36 score above 50 14.03 14.08 14.01Sf-36 score below 50 (fatigue) 13.32 13.35 13.57p for difference 0.03 0.04 0.25

Slope in cortisol secretion (log nmol/(l h))SF-36 score above 50 �0.1294 �0.1295 �0.1294SF-36 score below 50 (fatigue) �0.1267 �0.1268 �0.1269p for difference 0.002 0.006 0.016

Number for waking cortisol represent back transformed geometric means. The numbers for slope in cortisol secretion represent the declinein log cortisol/h. Model 1: adjusted for age, employment grade, sex, time of waking. Model 2: model 1 + additionally adjusted for healthbehaviors: current smoking, units of alcohol intake per week, sleep disturbance (Jenkins score), sleep duration. Model 3: model2 + additionally adjusted for health data: waist circumference, CVD medication, CNS medication, depressive symptoms (CES-D score � 16).


adjustment for all co-variates, the odds ratio for reportingnew fatigue at phase 7 in those with the flattest slopecompared to those with the steepest slopes in cortisol secre-tion is 1.56 (95% confidence intervals 1.09, 2.22) in thoseparticipants free of fatigue at phase 6 (n = 3020).

3.4. Cross-sectional associations of cortisol andfatigue at phase 7 (2002—2004)

In models adjusted for age, employment grade and wakingtime at phase 7, low cortisol levels at waking ( p = 0.03) and aflatter slope in cortisol secretion across the day ( p = 0.0002)were cross-sectionally associated with fatigue (Table 2). TheCAR was not associated with fatigue. As shown in Table 3 bothwaking cortisol and cortisol 30 min post-waking were lower inthose reporting higher fatigue compared to those reportinglower fatigue. Table 3 also shows that flatter slope was notsimply a reflection of a low waking cortisol in those reportingfatigue as higher bedtime cortisol was also apparent in thisgroup (mean cortisol in fatigued group vs. low fatigued group(95% confidence intervals); 1.74 (1.64, 1.86) vs. 1.58 (1.53,1.63), p = 0.007). Table 4 shows that the low waking cortisol

Table 5 Multi-variable adjusted odds ratio [95% CI] for the asspersistent fatigue between phase 7 and phase 8 (2006).

Model 1

Waking cortisolTop tertile 1Mid tertile 0.97 [0.79, 1.21]Bottom tertile (low cortisol) 1.12 [0.90, 1.40]

Slope in cortisol secretionTop tertile 1Mid tertile 1.04 [0.86, 1.25]Bottom tertile (flat slope) 1.39 [1.13, 1.70]

Results from a cumulative logit model. Numbers represent cumulated oreference is no fatigue at phase 7 or phase 8. Model 1: adjusted for age,1 + additionally adjusted for health behaviors: current smoking, units oduration. Model 3: model 2 + additionally adjusted for health data: wsymptoms (CES-D score � 16).

in fatigued participants was largely accounted measures ofhealth. When co-variates were entered individually, lowwaking cortisol was largely accounted for depressive symp-toms. In contrast, the association of flat slope in cortisolsecretion with fatigue remained largely independent of co-variates.

3.5. Association of cortisol secretion at phase 7(2002—2004) with fatigue assessed at phase 8(2007)

Table 5 shows occurrence of fatigue between phases 7 and 8by levels of waking cortisol and slope at phase 7. Participantswith flatter slopes in cortisol secretion were more likely toreport persistent fatigue between phases 7 and 8; odds ratio(95% CI) 1.34 (1.06, 1.69) in the fully adjusted model. Whenexamining the association of cortisol at phase 7 with new-onset fatigue at phase 8, flatter slope in cortisol secretionwas found to be predictive of fatigue at phase 8 (Fig. 1). Theassociations between cortisol secretion and new-onset fati-gue were largely independent of the co-variates examined. Inthe final models, age, waist circumference, sleep distur-

ociation between cortisol secretion at phase 7 (2003—2004)

Model 2 Model 3

1 10.93 [0.76, 1.14] 0.87 [0.72, 1.05]1.03 [0.83, 1.28] 0.98 [0.81, 1.20]

1 11.01 [0.82, 1.24] 1.01 [0.82, 1.30]1.36 [1.09, 1.71] 1.34 [1.06, 1.69]

dds for fatigue at phase 7 only and fatigue at phase 7 and phase 8,employment grade, sex, time of waking at phase 7. Model 2: modelf alcohol intake per week, sleep disturbance (Jenkins score), sleepaist circumference, CVD medication, CNS medication, depressive

Fig. 1 Multi-variable adjusted odds ratio [95% CI] for theassociation between (a) waking cortisol secretion and (b) slopein cortisol secretion at phase 7 (2003—2004) and new-onsetfatigue at phase 8 (2006). Analyses are conducted followingremoval of those participants that reported low fatigue at phase7, resulting in n = 3175. 259 participants reported high fatigue atphase 8. Model 1: adjusted for age, employment grade, sex, timeof waking on day of sample collection. Model 2: additionallyadjusted for current smoking, units of alcohol intake per week,sleep disturbance (Jenkins score), sleep duration, waist circum-ference, CVDmedication, CNS medication, depressive symptoms(CES-D score �16).


bance and depressive symptoms remained independentlyassociated with new-onset fatigue.

4. Discussion

The present study is the first to examine the longitudinalassociation between cortisol secretion and fatigue in a largecommunity based population. Among over 4000 former orcurrent British civil servants, low cortisol at waking predictednew-onset of reported fatigue during an approximately 2.5-year follow-up. The association with new-onset fatigue wasindependent of demographic characteristics, behavioral andbiological risk factors, presence of health conditions, andmedication use. Persistent fatigue assessed some 2 yearsprior to cortisol secretion was not associated with cortisolsecretion. However, cortisol secretion was associated withpersistent fatigue measured after cortisol assessment andthis effect tended to remain independent of all co-variates.Thus, our findings suggest that the direction of the associa-tion is dominantly from low cortisol and flat slope in cortisolsecretion leading to fatigue.

4.1. Fatigue as a predictor of cortisol secretion

Participants who reported any fatigue prior to the assessmentof cortisol did not have altered cortisol secretion. The find-ings suggest that short term psychological and biologicalconsequences of fatigue, which are not described in theWhitehall II study and are beyond the scope of this paper,serve to mitigate the hypocortisolism associated with fatigueitself. These findings support suggestions that for conditionsthat feature fatigue, such as burnout and vital exhaustion,altered cortisol secretion may be due to the disorders or co-morbidities associated with the disorders rather than a con-sequence of fatigue per se (Cleare, 2003). Alternatively, itmay be that fatigue at phase 6, when participants wereyounger andmore likely to be in employment, is qualitativelydifferent to fatigue at phases 7 and 8, when participants wereolder and more likely to have left work.

4.2. Cortisol as a predictor of fatigue

Our finding of a cross-sectional association of low wakingcortisol and flat diurnal slope in cortisol with fatigue and ofthe association between cortisol secretion and future fatiguesuggest that an alteration in cortisol secretion occurs early inthe process of the development of fatigue or are etiological inthe development of fatigue. Our findings suggest that wherewaking cortisol is low and slope in cortisol secretion isflattened, there is increased risk of future fatigue. Thesefindings accord with clinical studies in which pathologicallylow cortisol was associated with fatigue (Werbel and Ober,1993) and with studies that suggest lower levels of cortisol inthe morning to be related to lower levels of vitality later inthe same day (Adam et al., 2006). Our findings are alsoconsistent with the notion that low cortisol may exacerbatefatigue development or influence the progression of condi-tions such as CFS (TerWolbeek et al., 2007). The causes of lowcortisol secretion are unknown but factors not included in ouranalyses, such as early childhood trauma may influencewaking cortisol (Tarulla and Gunner, 2006). We have adjustedfor a number of co-variates including measures that mayoverlap with fatigue such as depressive symptoms, sleepdisturbance and duration. By controlling for these factorswe also controlled to some extent for potential confoundingarising from depression-related factors, such as early lifeevents (Tarulla and Gunner, 2006). However, we were notable to adjust for these co-variates in a time varying mannerand so it is possible that low cortisol may predict the devel-opment of measures that overlap with fatigue which wecannot account for in our analyses. The causes of flat slopesin cortisol secretion, which may represent a decline infunctioning in the HPA axis, are not established. In cross-sectional analyses, we found that flat cortisol slopes wereassociated with stress, smoking, early waking and CVD med-ications, other studies suggest that flat slopes may indicatepathology or disease severity as they predict mortality inclinical populations (Sephton et al., 2000). Our findingsdisagree with reports that cortisol is not predictive of fatiguein high risk groups (Candy et al., 2003). These discrepanciesmay be due to differences in study populations, responserates, study size, the assessment of basal cortisol secretionrather than cortisol assessed in relation to a stimulus and the


collection of saliva samples relative to waking time ratherthan clock time. In the Whitehall II study, waking time is a keypredictor of cortisol secretion.

4.3. Limitations of these analyses

Fatigue was assessed with the SF-36 whichmaymean that ourstudy is qualitatively different from fatigue as manifested infatigue and stress-related illness. Thus, fatigue in our studymay not correspond to fatigue as assessed in burnout syn-dromes. This may explain the discrepancy with recent datasuggesting that burnout is associated in increases in inte-grated measures of cortisol secretion as assessed in hairsamples (Kirschbaum, personal communication). However,a history of unexplained fatigue precedes the developmentof CFS which implies a continuum between fatigue in thehealthy population and fatigue-related illnesses (Wesselyet al., 1995; Huibers et al., 2004). Our analyses are con-ducted in a cohort of older people to whom fatigue may besecondary to subclinical or overt disease, although adjustedfor a number of health conditions we cannot discount thepossibility there may be residual confounding. Our findingsdisagree with studies that suggest hypocortisolism is asso-ciated with better health (Cleare, 2003; Hellhammer et al.,2004). This discrepancy may be because we have examined acommunity dwelling sample characterized by large size andhigh response rate. Although statistically significant and in apreviously reported direction (Lindeberg et al., 2008), thecross-sectional associations between cortisol and fatiguewere relatively weak corresponding to those observed whenwe examined associations of cortisol with smoking (Badricket al., 2007) and alcohol intake (Badrick et al., 2008). Ourfindings may not be generalisable to the general populationas we examined participants who originally formed an occu-pational cohort that was predominantly male and are at ahealth advantage. This health advantage and the cortisolassessment, which is demanding, precluded us from includingclinical assessment of CFS. Chronic fatigue syndrome has alow prevalence in the population (Ter Wolbeek et al., 2007)and the assessment of a continuous measure of fatigue mayprovide information that is of more broad relevance to healthin the general population. Some 49% of the cohort was retiredat phase 7 precluding us from examining psychological con-structs such as burnout and vital exhaustion as these arepredominantly work-related. We relied on self-report for thetimes at which samples were collected, which could dispro-portionately affect waking cortisol and the CAR. However,participants are generally accurate in their recording ofinformation (Dockray et al., 2008). We could collect daytime cortisol only and collected cortisol on 1 day whichmay bias findings to situational rather than trait exposures(Hellhammer et al., 2007), but given the predictive associa-tions observed over the 2.5-year follow-up, a large biasseems unlikely. Assessment of fatigue at these intervalsmay not be sufficiently frequent to capture the variationsin fatigue that occur in the population (Wood et al., 1992;Jason et al., 2001). This may mean that fatigued participantsare mis-classified into non-fatigued groups, which wouldserve to bias our estimates conservatively. Many co-variateswere not available in the phases before or after cortisolassessment which precluded us from examining the effectsof change in these variables. We have planned to follow-up

our cohort further to examine the association of cortisol withlonger term fatigue.

In conclusion, our analyses suggest that altered cortisolsecretion is associated with fatigue and predicts new-onsetfatigue in a relatively healthy older cohort. These findingssupport a role for cortisol in the development of fatigue andprovide insight into how cortisol may play a secondaryexacerbating role in conditions associated with fatigue.

Role of funding sources

The Whitehall II study has been supported by grants from theMedical Research Council; Economic and Social ResearchCouncil; British Heart Foundation; Health and Safety Execu-tive; Department of Health; National Heart Lung and BloodInstitute (HL36310), US, NIH: National Institute on Aging(AG13196), US, NIH; Agency for Health Care Policy Research(HS06516); and the John D and Catherine T MacArthur Foun-dation Research Networks on Successful Midlife Developmentand Socio-economic Status and Health. M. Kivimaki wassupported by the Academy of Finland (grants no.: 117604,124322, and 124271). Funders have had no further role in thestudy design; in the collection, analysis and interpretation ofdata; in the writing of the manuscript and in the decision tosubmit the paper for publication.

Conflict of interest

The authors have no conflicts of interests to declare.

Acknowledgements

We thank all participating Civil Service departments and theirwelfare, personnel, and establishment officers; the Occupa-tional Health and Safety Agency; the Council of Civil ServiceUnions; all participating civil servants in the Whitehall IIstudy; all members of the Whitehall II study team.

References

Adam, E.K., Hawkley, L.C., Kudielka, B.M., Cacioppo, J.T., 2006.Day-to-day dynamics of experience–—cortisol associations in apopulation-based sample of older adults. Proc. Natl. Acad. Sci.U.S.A. 103, 17058—17063.

Appels, A., 1990. Mental precursors of myocardial infarction. Br. J.Psychiatry 156, 465—471.

Badrick, E., Bobak, M., Britton, A., Kirschbaum, C., Marmot, M.,Kumari, M., 2008. Alcohol and cortisol secretion in an ageingcohort. J. Clin. Endocrinol. Metab. 93, 750—757.

Badrick, E., Kirschbaum, C., Kumari, M., 2007. Smoking status andsalivary cortisol secretion. J. Clin. Endocrinol. Metab. 92, 819—824.

Candy, B., Chalder, T., Cleare, A.J., et al., 2003. Predictors of fatiguefollowing the onset of infectious mononucleosis. Psychol. Med.33, 1333—1338.

Cleare, A., 2003. The HPA axis and the genesis of chronic fatiguesyndrome. Trends Endocrinol. Metab. 15, 55—59.

Clow, A., Thorn, L., Evans, P., Hucklebridge, F., 2004. The awakeningcortisol response: methodological issues and significance. Stress7, 29—37.

Crofford, L.J., Young, E.A., Engleberg, N.C., Korszun, A., Brucksch,C.B., McClure, L.A., Brown, M.B., Demitrack, M.A., 2004. Basal


circadian and pulsatile ACTH and cortisol secretion in patientswith fibromyalgia and/or chronic fatigue syndrome. Brain Behav.Immun. 18, 314—325.

Di Giorgio, A., Hudson, M., Jerjes, W., Cleare, A.J., 2005. 24 hpituitary and adrenal hormone profiles in chronic fatigue syn-drome. Psychosom. Med. 67, 433—440.

Dockray, S., Bhattacharyya, M.R., Molloy, G.J., Steptoe, A., 2008.The cortisol awakening response in relation to objective andsubjective measures of waking in the morning. Psychoneuroen-docrinology 33, 77—82.

Fries, E., Hesse, J., Hellhammer, J., Hellhammer, D.H., 2005. A newview on hypocortisolism. Psychoneuroendocrinology 30, 1010—1016.

Hellhammer, J., Schlotz, W., Stone, A.A., Pirke, K.M., Hellhammer,D., 2004. Allostatic load, perceived stress and health: a prospec-tive study in two age groups. Ann. N. Y. Acad. Sci. 1032, 8—13.

Hellhammer, J., Fries, E., Schweisthal, O.W., Schlotz, W., Stone,A.A., Hagemann, D., 2007. Several daily measurements arenecessary to reliably assess the cortisol rise after awakening:state- and trait components. Psychoneuroendocrinology 32, 80—86.

Huibers, M.J., Kant, I.J., Knottnerus, J.A., Bleijenberg, G., Swaen,G.M., Kasl, S.V., 2004. Development of the chronic fatigue syn-drome in severely fatigued employees: predictors of outcome inthe Maastricht cohort study. J. Epidemiol. Community Health 58,877—882.

Jason, L.A., Taylor, R.R., Carrico, A.W., 2001. A community-basedstudy of seasonal variation in the onset of chronic fatigue syn-drome and idiopathic chronic fatigue. Chronobiol. Int. 18, 315—319.

Lindeberg, S.I., Eek, F., Lindbladh, E., Ostergren, P.-O., Hansen,A.M., Karlson, B., 2008. Exhaustionmeasured by the SF-36 vitalityscale is associated with a flattened diurnal cortisol profile. Psy-choneuroendocrinology 33, 471—477.

Marmot, M., Brunner, E., 2005. Cohort Profile: the Whitehall II study.Int. J. Epidemiol. 34, 251—256.

Melamed, S., Shirom, A., Toker, S., Berliner, S., Shapira, I., 2006.Burnout and risk of cardiovascular disease: evidence, possible

causal paths and promising research directions. Psychol. Bull.132, 327—353.

McEwen, B.S., 1998. Protective and damaging effects of stressmediators. N. Engl. J. Med. 338, 171—179.

Nater, U.M., Maloney, E., Boneva, R.S., Gurbaxani, B.M., Lin, J.M.,Jones, J.F., Reeves, W.C., Heim, C., 2008. Attenuated morningsalivary cortisol concentrations in a population-based study ofpersons with chronic fatigue syndrome and well controls. J. Clin.Endocrinol. Metab. 93, 703—709.

Radloff, L.S., 1977. The CES-D scale: a self report depression scale forresearch in the general population. Appl. Psychol. Meas. 1, 385—401.

Roberts, A.D., Papadopoulos, A.S., Wessely, S., Chalder, T., Cleare,A.J., 2009. Salivary cortisol output before and after cognitivebehavioural therapy for chronic fatigue syndrome. J. Affect.Disord. 115, 280—286.

Sephton, S.E., Sapolsky, R.M., Kraemer, H.C., Speigel, D., 2000.Diurnal cortisol rhythm as a predictor of breast cancer survival.J. Natl. Cancer Inst. 92, 944—1000.

Tarulla, A.R., Gunner, M., 2006. Child maltreatment and the devel-oping HPA axis. Horm. Behav. 50, 632—639.

Ter Wolbeek, M., van Doornen, L.J., Coffeng, L.E., Kavelaars, A.,Heijnen, C.J., 2007. Cortisol and severe fatigue: a longitudinalstudy in adolescent girls. Psychoneuroendocrinology 32, 171—182.

Van Den Eede, F., Moorkens, G., Van Houdenhove, B., Cosyns, P.,Claes, S.J., 2007. Hypothalamic-pituitary—adrenal axis functionin chronic fatigue syndrome. Neuropsychobiology 55, 112—120.

Ware Jr., J.E., Sherbourne, C.D., 1992. The MOS 36-item Short-formHealth Survey (SF-36) I. Conceptual framework and item selec-tion. Med. Care 30, 73—83.

Wessely, S., Chalder, T., Hirsch, S., Pawlikowska, T., Wallace, P.,Wright, D.J., 1995. Postinfectious fatigue: prospective cohortstudy in primary care. Lancet 345, 1333—1338.

Werbel, S.S., Ober, K.P., 1993. Acute adrenal insufficiency. Endocri-nol. Metab. Clin. North Am. 22, 303—328.

Wood, M.E., Magnello, Sharpe, M.C., 1992. Fluctuations in perceivedenergy and mood among patients with chronic fatigue syndrome.J. Roy. Soc. Med. 85, 195—198.

Documents

Cortisol secretion and fatigue: Associations in a community based cohort