Identifying patterns in cortisol secretion in an olderpopulation. Findings from the Whitehall II study

Meena Kumari a,*, Ellena Badrick b, Amanda Sacker c, Clemens Kirschbaumd,Michael Marmot a, Tarani Chandola a

aDepartment of Epidemiology and Public Health, UCL, 2-18 Torrington Place, London, WC1E 6BT, UKbCentre for Health Sciences, Barts and the London School of Medicine and Dentistry, London, UKc Institute for Social and Economic Research, University of Essex, Essex, UKdBiological Psychology, Technical University of Dresden, Dresden, Germany

Received 10 June 2009; received in revised form 14 January 2010; accepted 18 January 2010

Psychoneuroendocrinology (2010) 35, 1091—1099

KEYWORDSSalivary cortisol;Epidemiology;Mixture model;Age;Sleep duration;Smoking;Gender;Walking/gait speed;Cohort study

Summary Alterations in the patterning of diurnal cortisol secretion are associated with poorhealth in clinical populations with ‘flat’ patterns a particular risk. Flatter patterns in cortisolsecretion may reflect impaired negative feedback in the hypothalamic-pituitary-adrenal axis.The correlates of discrete clusters of patterns in the diurnal secretion of cortisol have not beenwell described in large community dwelling populations. We describe discrete clusters of patternsof cortisol secretion and examine the correlates of these patterns using a latent variable mixturemodelling approach. Analyses use data from 2802 participants with complete information oncortisol secretion, age, walking/gait speed, stress, waking up time and sleep duration. Cortisolwas assessed from six saliva samples collected at waking, waking plus 30 min, 2.5 h, 8 h, 12 h andbedtime. We find two patterns (‘‘curves’’) of diurnal cortisol secretion. These curves aredescribed as ‘normative’ [prevalence 73%] and a ‘raised’ [27%] curve differentiated by a lowercortisol awakening response in the normative group, a higher diurnal cortisol and ‘flatter’ patternof release in the raised group. Older age, being male, a smoker, stress on the day of sampling,slower walking speed and shorter sleep duration increased the odds of being in the raised curve,relative to the normative curve. In conclusion, two patterns of cortisol secretion occur in middleaged men and women. Raised pattern of secretion, which occurs in 27% of our participants isassociated with demographic variables, adverse health behaviours, psychosocial environmentand impaired physical functioning.# 2010 Published by Elsevier Ltd.

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

journa l homepage: www.e l sev ie r.com/locate/psyneuen

* Corresponding author. Tel.: +44 0 20 353 3313;fax: +44 0 20 7813 0280.

E-mail address: m.kumari@ucl.ac.uk (M. Kumari).

0306-4530/$ — see front matter # 2010 Published by Elsevier Ltd.doi:10.1016/j.psyneuen.2010.01.010

1. Introduction

The use of salivary sampling as a noninvasive tool for theassessment of free cortisol and therefore as a marker foractivity of the hypothalamic-pituitary-adrenal (HPA) axis is

1092 M. Kumari et al.

well established in human stress research (Kirschbaum andHellhammer, 1989). This sampling method captures the diur-nal variations in cortisol secretion with the large rise asso-ciated with the awakening response and the subsequentdecline in levels across the day and has provided the oppor-tunity to examine predictors of variation in these patterns ina naturalistic setting.

A number of ‘patterns’ in cortisol release have beendescribed. For instance, studies of serum cortisol variationsin patients with severe long lasting psychiatric depressionhave shown that these patients have raised evening levels ofcortisol which correspond to an inability to lower appropri-ately serum cortisol during the dexamethasone test (Rubinet al., 1987). Conversely, those with pronounced symptomsof exhaustion such as the chronic fatigue syndrome areunable to raise their cortisol level in challenging situationsand they also show very small diurnal variation (‘‘low flatcurves’’) (Demitrack et al., 1991). Cortisol hyposecretion isassociated with post traumatic stress disorder (Yehuda et al.,1991) and a ‘flat pattern’ is associated with fatigue in cancerpatients (Abercrombie et al., 2004) and in mal-treatment inchildhood (Tarullo and Gunnar, 2006). These patterns incortisol secretion may reflect disturbances of the capacityto regulate cortisol secretion. Flatter patterns in cortisolsecretion, in which evening levels of cortisol secretion areraised, may occur due to impaired feedback in the hypotha-lamic-pituitary-adrenal (HPA) axis or to hypersensitivity tocortisol stimulation later in the day with evidence suggestinga role for the former rather than latter mechanism (Spiegelet al., 2006).

Patterns in the diurnal release of cortisol secretion areusually described from visual inspection of means or popula-tion specific arbitrary cut points in slope of cortisol levels(Smyth et al., 1997; Ice et al., 2004; Ranjit et al., 2005;Cohen et al., 2006). A recent study used growth curvemodelling and described three curves in cortisol secretionin children aged 3 years (van Rysin et al., 2009). Previousanalyses have not formally described how the differentpatterns of diurnal cortisol secretion group or clustertogether in older groups because few studies have assesseddiurnal cortisol secretion in populations sufficiently largeenough, that is, with many hundreds of participants, toexamine clustering of these patterns.

Cortisol secretion is hypothesized to be etiological in thedevelopment of a number of conditions including heartdisease (Rosmond et al., 2003), osteoporosis (Raff et al.,1999), cognitive decline (Karlamangla et al., 2005) andrecently with frailty (Varadhan et al., 2008). These areconditions that are manifest in older age groups, and under-standing how the patterns of diurnal cortisol secretion grouptogether in an older population and also the predictors ofthese groups could help us identify normative and ‘‘abnor-mal’’ patterns of diurnal cortisol secretion and hence helpidentify the role of changes in cortisol secretion in thedevelopment of these morbidities and disease. Correlatesof cortisol secretion in older age groups are mixed for basicdescriptive correlates such as age in community dwellingpopulations. For example, previous reported associations ofcortisol secretion with age and gender have been equivocal.Thus, associations with age are reported to be positive(Deuschle et al., 1997; Powell et al., 2002) or null (VanCauter et al., 1996). Others have suggested that an associa-

tion is apparent for those with depression only (Kudielkaet al., 2000). Ice et al., describe a flatter slope in cortisolsecretion associated with increasing age in 48 communitydwelling volunteers aged 65 years and older. In this study, 2%of participants were described as having flat slopes in corti-sol secretion, while Smyth et al. (1997) describe ‘flat’patterns in 17% of younger participants.

Here we examine patterns of cortisol secretion in a largecommunity dwelling population use a latent variable mixturemodelling (LVMM) approach. The primary objective of LVMM isto uncover groups of individuals who share similar character-istics on a set of observed variables (in this paper cortisol).The unobserved patterns of cortisol release are described bya mixture of components, identified by categorical latentvariables (latent classes): the object of the analysis is to findthe smallest number of latent classes that can describe theassociations among the set of observed continuous cortisolvalues observed across the day. The analysis adds classesstepwise until the model fits the data well.

We use the LVMM to examine whether different patterns inthe diurnal variation of salivary cortisol can be identified inthe population. Specifically,

1. What are the patterns (clusters) of diurnal cortisol secre-tion?

2. With which variables are the LVMM patterns associated?The predictors examined are those which have beendemonstrated to modulate cortisol secretion in nonclini-cal studies and include biological (age and sex), beha-vioural factors (current smoking (Badrick et al., 2007),waking up time (Williams et al., 2005) and sleep duration(Spiegel et al., 1999), psychosocial (perceptions of stress)and a physical measure of functioning relevant to olderage groups, gait/walking speed (Melzer et al., 2003).

2. Materials and methods

2.1. Study population

Data reported here are from phase 7 (2002—2004) of theWhitehall II study. The cohort was initially recruited between1985 and 1988 (phase 1) from 20 London-based civil servicedepartments, 10,308 people participated. Eight phases ofthe study have been completed, details of the study havebeen reported elsewhere (Marmot and Brunner, 2005). Thenumber participating at phase 7 was 6968, of these 6484 hada clinical assessment. The collection of saliva samples wasinstigated part way through phase 7 and 4967 were asked toprovide a cortisol sample out of whom 90.1% (n = 4609)returned samples. This group had fewer participants in thelowest civil service employment grades compared to phase 1of the study, however this difference was small. Ethicalapproval for the Whitehall II study was obtained from theUniversity College London Medical School committee on theethics of human research. Informed consent was gained fromevery participant.

2.2. Cortisol collection and analysis

The protocol has been described previously (Badrick et al.,2007). Briefly, participants were requested to provide six

Identifying patterns in cortisol secretion in an older population 1093

saliva samples in salivettes over the course of a normalweekday at waking, +30 min, +2.5 h, +8 h, +12 h and bed-time. Participants were instructed to avoid caffeine andacidic drinks, not brush teeth, eat or drink anything for15 min prior to a sample collection. Participants used aninstruction booklet to record information on the day ofsampling including wake time (participants were instructedthat this should be time of waking and not the time at whichthey got out of bed) and time each sample was taken. Thesalivettes and booklet were returned via post. Salivetteswere centrifuged at 3000 rpm for 5 min resulting in a clearsupernatant of low viscosity. Salivary cortisol levels weremeasured using a commercial immunoassay with chemilumi-nescence detection (CLIA, IBL-Hamburg, Hamburg, Ger-many). The lower concentration limit of this assay is0.44 nmol/l; intra and interassay coefficients of variancewere below 8%. Any sample over 50 nmol/l was repeated.During analysis a total of 1002 individual samples were notassayed due to loss of sample in transport between Londonand Germany or insufficient saliva.

2.3. Covariates

Age and sex were assessed by questionnaire. Smoking statuswas assessed as previously described (Badrick et al., 2007).Current smokers, at phase 7, were defined as those thatreported smoking cigarettes, cigars or a pipe, social oroccasional smoking or taking nicotine replacement products.Waking up time was assessed by asking participants to recordthe time of waking on the day of the collection of saliva. Thatis the time at which the participants were aware of beingawake for the day and were not going to go back to sleep.Participants were also asked to record the time of fallingasleep the night before and sleep duration the night beforesample collection was calculated from these responses. Inaddition, participants were asked to rate the most stressfulevent on the day of sample collection using the categories‘not at all stressed’, ‘somewhat stressed’, ‘moderatelystressed’, ‘very stressed’ or ‘the most stressed I have everfelt’. Participants classified as having a stressful experienceif they responded that they were ‘very stressed’ or ‘the moststressed I have ever felt’. Walking/Gait speed, a measure ofphysical functioning was assessed by a nurse over a clearlymarked 8 ft walking course using a standardised protocol(Guralnik et al., 1994).

2.4. Statistical methods

Missing data: A delay in taking sample 1 results in a reducedcortisol awakening response (CAR) (Kudielka et al., 2003).Therefore, data from 726 participants who reported takingsamples later than 10 min after waking and 123 whoreported that they ate, drank, exercised or brushed theirteeth before the first sample were removed from theanalyses. Additionally, those taking steroid medication(n = 260), with incomplete data on all six cortisol samples(n = 329) and outliers (3 standard deviations or greaterfrom the mean, n = 220) were excluded. This left 3133participants with valid and complete cortisol data (includ-ing time of sampling). Additional missing values for cov-ariates (in particular, reported stress and walking speed)reduced this to 2802 participants.

2.5. Latent variable mixture modelling (LVMM)

The aim of the LVMM is to use the inherent variability in thedata to find groups (latent classes) of individuals who aresimilar to each other, which is then represented by anunobserved (latent) categorical variable (Lazarfeld andHenry, 1968; McLachlan and Peel, 2000). For the analysesin this paper, as the six cortisol measurements are collectedfrom the same person on the same day, we adopt the leastrestrictive assumption for a multivariate Gaussian mixturemodel so that the six cortisol measurements may covary indifferent ways in each latent class.

There are a number of considerations used to decide onthe number of latent classes. The analysis adds classesstepwise until the model fits the data well starting withthe simplest (a one class) solution. The number of classesto include is assessed first by an examination of the modelevaluation statistics (see Appendix A for more detail). Thesecond consideration is summarized using an entropy mea-sure based on the class membership probabilities. Entropymeasures how well the model is able to predict class mem-bership given the observed cortisol values for each indivi-dual. The third consideration is the usefulness of the latentclasses in practice. This can be determined by examining thetrajectory shapes for similarity, the number of individuals ineach class, and whether the classes are associated withobserved characteristics in an expected manner (Muthenand Muthen, 2006; Nylund et al., 2007; Lo et al., 2001).

Once a class solution has been found, background vari-ables which predict membership into the categorical latentclasses by multinomial logistic regression are simultaneouslymodelled in the LVMM. These background variables arehypothesized predictors of diurnal cortisol release (biologicaland behavioural factors that alter HPA axis activity).

Full details of the latent variable mixture model (withMPlus command syntax) are specified in Appendix A.

Analysis steps:

1. Use latent variable mixture modelling (LVMM) to decideon the number of patterns (latent classes) in diurnalcortisol secretion.

2. Add in covariates as background variables and see if thepatterns remain similar.

3. Results

Compared to all those who were asked to complete samplecollection at phase 7, the subsample of 2802 participants thatwere analyzed were very similar (Table 1).

3.1. Prevalence of discrete patterns of cortisolsecretion

The fit statistics of the 1—6-class solutions for the latentvariable mixturemodel (LVMM) are displayed in Table 2. Fig. 1graphs the Bayesian Information Criteria (BIC) fit statistic andthis shows that there was little improvement in model fitafter the 4-class solution. This suggests that a 2—4-classsolution may be appropriate. We can further assess whetherwe have chosen the right number of classes using the Vuong—Lo—Mendell Rubin (VLR) test (MPlus tech11, Lo et al., 2001).

Table 2 Model evaluation statistics for the 1—6-class latent variable mixture models (no covariates).

1-class 2-class 3-class 4-class 5-class 6-class

Log likelihood �51,318 �49,024 �48,543 �48,188 �48,003 �47,841AIC 102,702 98,170 97,265 96,610 96,296 96,027BIC 102,900 98,535 97,798 97,311 97,166 97,064Sample size adjusted BIC 102,795 98,342 97,515 96,939 96,705 96,515Entropy 0.78 0.71 0.72 0.70 0.70Lo—Mendell—Rubin adjusted LRT p-value 0.00 0.00 0.02 0.01 0.08

Table 1 Characteristics of participants at phase 7 (2002—2004) of the Whitehall II study.

Of all participantswho attended phase7 (n = 6968)

Of all participantsasked to completecortisol collection (n = 4967)

Participants includedin patterns analysis(n = 2802)

% Male 70.2 73.3 73.3Mean age (SD) 61.2 (6.0) 61.0 (5.9) 60.9 (5.9)% Non-White ethnicity 7.0 7.0 6.4% Smokers 8.3 8.2 8.1% Stressed 31.2 31.1Walking speed test in seconds (SD) 2.1 (0.7) 2.1 (0.7) 2.1 (0.6)Average wake time Hours:Minutes 06:46 06:49Number hours slept Hours:Minutes 06:58 07:00

1094 M. Kumari et al.

This test compares the model with K classes to a model withK � 1 classes. The VLR test has a p-value of <0.05 for the 4-and 5-class models, and <0.01 for the 2- and 3-class models.This suggests that two or three classes are sufficient and thatmore than three classes are not needed. Furthermore, theentropy for the 3-class model (0.71) was considerably worsethan for the 2-class model (0.78). Additional tests usingparametric bootstrapped likelihood ratio tests (MPlustech14) were carried out, but these results were likely tobe inaccurate as the log likelihood was not replicated inrepeated bootstrap draws, indicating local maxima (Nylundet al., 2007).

The mean cortisol values for the 1—3-class LVMM solutionsare shown in Fig. 2. Comparing the 2-class solution with the 1-class solution, there appears to be a ‘‘normative group’’(class 1, N = 2042) with a similar profile to the 1-class solu-tion. Furthermore, there is a ‘‘raised’’ cortisol profile group(class 2, N = 760) with higher cortisol awakening response andhigher mean day and evening time values compared to thenormative class. The two groups differed in terms of theslope; the linear decline in cortisol from awakening to bed-time (ignoring the cortisol awakening response).

Figure 1 Model fit statistics for the 1—6-class latent variablemixture models.

For the 3-class solution, in addition to the two groupsdescribed in the 2-class solution, there is another normativegroup (class 2, N = 1374) with a ‘‘cortisol awakeningresponse’’ and average cortisol over the day in betweenthe raised cortisol profile group (class 3, N = 377) and thefirst normative group (N = 1051), which has the ‘‘flattest’’profile in terms of diurnal cortisol slope.

As a result of inspecting the model fit statistics and thepattern of the cortisol profiles, we chose to investigate a 2-class solution further, using covariates to predict membershipof these 2 latent classes. The mean cortisol values of the 2-class solution with covariates are displayed in Fig. 3. Theprofile is remarkably similar to the 2-class solution withoutcovariates (Fig. 2). There is a normative profile class and araised profile class with the latter class having a highercortisol awakening response as well as higher day and eveningtime cortisol secretion. Further in Fig. 3, additional analysesanalyzing the 3-class solution with covariates were carriedout: the VLR test has a p-value of<0.05 for the 3-class model,suggesting two classes are sufficient.

3.2. Predictors of discrete patterns of cortisolsecretion

Table 3 displays the results of the logistic regression relatingthe binary 2-class pattern of cortisol release with a set ofpredictor variables, age, sex, smoking status, waking up timeand sleep duration entered simultaneously. The normativegroup was chosen as the reference group. Participants with araised cortisol profile (who had on average a greater CAR of1.8 nmol/l and greater total cortisol of 2.7 nmol/l) tended tobe older, men, smokers, had shorter sleep duration, reportedevents as more stressful, woke up earlier and walked moreslowly compared to the normative group.We tested for 2-wayinteractions between all the covariates and did not find any

Figure 2 Figures (a—c) 1,2 and 3 class solutions for latent variable mixture models of diurnal cortisol: no covariates.

Figure 3 Two class solution (with covariates predicting latent class membership).

Identifying patterns in cortisol secretion in an older population 1095

Table 3 Odds ratios (95% confidence interval) of the logistic regression of binary latent class membership regressed on backgroundvariables.

Odds of being in raised profileclass vs. normative profile class

N

Age (years) 1.01 (0.99, 1.03) 2802Sex (1 = men) 1.90 (1.48,2.45) 2802Smoker (1 = smoker) 2.02 (1.45, 1.90) 2802Sleep length (hours) 0.76 (0.69, 0.84) 2802Wake up time 0:500 h or earlier 1.94 (0.99,3.79) 72Wake up time 05:00—06:00 h 1.70 (1.11,2.60) 399Wake up time 06:00—07:00 h 1.34 (0.95, 1.90) 1073Wake up time 07:00—08:00 h 1.13 (0.79, 1.60) 901Wake up time 08:00 h or later (ref.) 1.00 357Stress on day of sampling (1 = stressed) 1.37 (1.04, 1.80) 2802Walking speed test (seconds) 1.21 (1.05, 1.39) 2802

1096 M. Kumari et al.

significant associations. We also repeated the analyses,restricting the sample to those people who had taken theirsecond saliva sample between 30 and 45 min after awakening(N = 2755), and found the same pattern as for the wholesample.

4. Discussion

We find two common curves in this large community dwellingpopulation using a latent class analysis approach. Thesecurves represent a ‘normative curve’, and a ‘raised curve’.The raised curve was composed of an increased CAR andhigher day time cortisol and flatter slope compared to nor-mative curve.

The epidemiology of the HPA axis is relatively unexploredin large community dwelling populations, here we find innearly 3000 participants, that those with a raised cortisolprofile tend to be older, be men, have shorter sleep duration,wake earlier, report events as more stress and walk moreslowly than those with normative curves. Those with a raisedprofile in cortisol secretion have flatter slopes in secretionthan the normative profile in release.

A number of techniques are used to examine the circadianprofile of cortisol when assessed by multiple sampling meth-ods. These include measures that utilize information on theslope of the decline from the peak to trough levels (De Beurset al., 2001), separate assessments of morning and/or eve-ning cortisol levels (Steptoe et al., 2000) or total cortisolconcentration over the day, using a measure such as area-under-the-curve (Pruessner et al., 2003; Dekker et al., 2008).Several studies focus on the rhythm profile of cortisol, as theCAR, as well as the extent of decline to the evening nadir(Smyth et al., 1997; Sephton et al., 2000; Schmidt-Reinwaldet al., 1999; Adam et al., 2006).

Alternative approaches seek to describe the cortisolrhythm in all its complexity, and use non-parametricapproaches to fit smoothing splines (Wang and Brown,1996) which provide better fit but, while useful for explora-tion, do not yield quantitative parameters to allow for groupcomparison. Multilevel modelling approaches may be usedbecause of the hierarchical structure of the data (Schmidt-Reinwald et al., 1999; Adam et al., 2006). However, only one

study has used a method that describes how different pat-terns of diurnal cortisol secretion in a population clustertogether (Van Ryzin et al., 2009). The latent variable mixturemodelling method used in this and our paper is different frommost methods in terms of letting the patterns of diurnalcortisol secretion be identified from the data themselves,rather than arbitrarily imposing cut off points in terms ofslope or area under curve parameters.

4.1. Discrete patterns of cortisol secretion

The analysis showed two discrete patterns of diurnal cortisolrelease with clear separation between these patterns.

It could be argued that the CAR and decline of cortisolsecretion throughout the day should be examined separatelyas predictors may be different for these aspects of cortisolsecretion (Clow et al., 2004). We did this in further analysis tolook at the differences between the two curves in terms ofthe diurnal decline in cortisol. The decline in cortisol wasshallower in the raised curve, suggesting a ‘flatter slope’.The raised curve may represent groups in which cortisol ischronically raised due to a failure in negative feedback orhypersensitivity to cortisol stimulation at the end of day.Spiegel et al. (2006) suggested a flat curve in association witha raised CAR represented impaired negative feedback. More‘extreme’ curves, for example, a pattern of low CAR and aflat curve as found in post traumatic stress disorder (Yehudaet al., 1991) are not apparent in the curves identified usingthis method suggesting that this population is healthy or thatthe population is a particularly compliant group. However, inour population, we have a prevalence of 27% of participantswho have ‘flatter’ curves than the rest of the population. Thisprevalence of flatter curves is higher than previouslyreported levels; for example Ice et al. (2004) reported 2%of flat curves, while Smyth et al. (1997) reported 17%. Thisdiscrepancy may relate to the nature of cohort study popula-tions to volunteer populations. It remains to be determinedwhich aspect of cortisol secretion, these curves or moretypically described parameters (Adam and Kumari, 2009),the CARor slope in cortisol secretion are predictive of diseasewhen participants are followed up for health events andother outcomes.

Identifying patterns in cortisol secretion in an older population 1097

4.2. Predictors of discrete patterns in cortisolsecretion

Previous reported associations of cortisol secretion with ageand gender have been equivocal. Thus, associations with ageare reported to be positive (Deuschle et al., 1997; Powellet al., 2002) or null (Van Cauter et al., 1996). Others havesuggested that an association is apparent for those withdepression only (Kudielka et al., 2000). Our findings fail toclarify this relationship as they suggest that in a relativelyhealthy community-based sample, participants with increas-ing age have increased risk of being the elevated groupsupporting the hypothesis that negative feedback in thehypothalamic-pituitary-adrenal axis may change with aging(Jacobs et al., 1984) but this association failed to reachsignificance and requires clarification in additional studies.Reported associations of sex with cortisol secretion aremixed. Our data do not accord with studies indicating raisedcortisol in women (Wust et al., 2000). We (Badrick et al.,2007) and others (Steptoe and Ussher, 2006) have shown thatcigarette smoking is associated with raised total cortisolsecretion and these findings are confirmed in the latent classanalyses which indicate that smoking is associated withelevated patterns of cortisol secretion. Similarly, we findthat in accordance with previous reports (Spiegel et al.,1999), short sleep duration is associated with elevated pat-terns of cortisol secretion. Short sleep duration is predictiveof mortality associated with cardiovascular disease (Ferrieet al., 2007) and our findings accord with the notion thatcortisol may play a role in mediating this relationship (Ros-mond et al., 2003; Smith et al., 2005). Stress on the day ofsampling is associated with increased membership of raisedcurve militating against a definitive conclusion on whetherraised curve represents hypersensitivity to cortisol increasein the day or with impaired negative feedback. Our findingssupport the notion that cortisol secretion is associated withmeasures of physical functioning (Varadhan et al., 2008) andsuggest that these associations are pertinent to younger,relatively healthy populations and to men. In particularour finding of a cross sectional association of these measuresof cortisol with walking/gait speed points to a potential rolefor cortisol in the development of poor physical function andother outcomes associated with ageing.

The advantages of our study are that we conducted thecollection of saliva from a large group of well characterizedparticipants and achieved a high response rate. We see noevidence of participants not understanding the instructions.However, our participants are high functioning, currently orrecently in white collar occupations and thus may not repre-sent the population as a whole. Additionally, we assessedcompliance by self report, we did not objectively verifyreports but our rates of ‘late’ collection are not differentto studies in which reporting was assessed objectively(Kudielka et al., 2003). Recent reports suggest that partici-pants are generally accurate in their records of time of datacollection (Dockray et al., 2008; Desantis et al., 2009). Wemeasured cortisol in less than half of the original cohort(phase 1), we found that participants were generally healthyin comparison to the rest of the cohort that attended clinic atphase 7. Finally, cortisol was assessed on a single day as itwould have been impractical to collect on more than one dayin a sample this size and age and this may have resulted in

some bias towards situational rather than trait factors aspredictors of cluster membership (Hellhammer et al., 2007).Collection of cortisol in a single day also precluded us fromcomparing our findings to previously reported patterns insecretion which measure consistency of cycles across days ofcollection such as in Smyth et al. (1997). Cortisol is beingreassessed in this cohort, which will allow us to examine thelong term stability of group membership.

In summary, we find two common curves in this largecommunity dwelling population using a latent class analysisapproach. We find a normative curve and a ‘raised curve’.The raised curvemay relate to impaired negative feedback orincreased sensitivity to cortisol stimulation later in the day.We find that compared with the normative curve, the riskfactors associated with the raised cortisol curve includesolder age, male gender, smoking, short sleep duration,increased reporting of a stressful event and slow walking/gait speed. Follow up of this cohort will allow us to determineand compare whether the clusters ormore typically collectedaspects of cortisol secretion such as the cortisol awakeningresponse or the slope in cortisol secretion predict the devel-opment of hypothesized morbidities and diseases.

Role of the funding source

The Whitehall II study has been supported by grants fromthe Medical Research Council; Economic and SocialResearch Council; British Heart Foundation (RG/02/005;PG/03/029); Health and Safety Executive; Department ofHealth; National Heart Lung and Blood Institute (HL36310),US, NIH: National Institute on Aging (AG13196), US, NIH;Agency for Health Care Policy Research (HS06516); and theJohn D. and Catherine T. MacArthur Foundation ResearchNetworks on Successful Midlife Development and Socioeco-nomic Status and Health. These funding bodies had nofurther role in the gathering of data, writing of the reportor decision to submit the paper for publication. ProfessorsSacker, Chandola’s and Dr. Kumari’s time on this manuscriptwas partially supported by the Economic and Socialresearch council (RES-596-28-0001).

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 theWhitehall II study;all members of the Whitehall II study team. The Whitehall IIStudy team comprises research scientists, statisticians, studycoordinators, nurses, data managers, administrative assis-tants and data entry staff, who make the study possible.

Appendix A. Supplementary data

Supplementary data associated with this article can befound, in the online version, at doi:10.1016/j.psy-neuen.2010.01.010.

1098 M. Kumari et al.

References

Abercrombie, H.C., Giese-Davis, J., Sephton, S., Epel, E.S., Turner-Cobb, J.M., Spiegel, D., 2004. Flattened cortisol rhythms inmetastatic breast cancer patients. Psychoneuroendocrinology29, 1082—1092.

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.

Adam, E.K., Kumari, M., 2009. Assessing salivary cortisol in large-scale epidemiological research. Psychoneuroendocrinology 34,1423—1436.

Badrick, E., Kirschbaum, C., Kumari, M., 2007. The relationshipbetween smoking status and cortisol secretion in the WhitehallII study. J. Clin. Endocrinol. Metab. 92, 819—824.

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

Cohen, S., Schwartz, J.E., Epel, E., Kirschbaum, C., Sidney, S.,Seeman, T., 2006. Socioeconomic status, race and diurnal cortisoldecline in the Coronary Artery Risk Development in Young Adults(CARDIA) study. Psychosom. Med. 68, 41—50.

De Beurs, E., Beekman, A., Geerlings, S., Deeg, D., Van Dyck, R., VanTilburg, W., 2001. On becoming depressed or anxious in late life:similar vulnerability factors but different effects of stressful lifeevents. Br. J. Psychiatry 179, 426—431.

Dekker, M.J., Koper, J.W., van Aken, M.O., Pols, H.A., Hofman, A., deJong, F.H., Kirschbaum, C., Witteman, J.C., Lamberts, S.W.,Tiemeier, H., 2008. Salivary cortisol is related to atherosclerosisof carotid arteries. J. Clin. Endocrinol. Metab. 93, 3741—3747.

Demitrack, M.A., Dale, J.K., Straus, S.E., Laue, L., Listwak, S.J.,Kruesi, M.J., Chrousos, G.P., Gold, P.W., 1991. Evidence forimpaired activation of the hypothalamic-pituitary-adrenal axisin patients with chronic fatigue syndrome. J. Clin. Endocrinol.Metab. 73, 1224—1234.

Desantis, A.S., Adam, E.K., Mendelsohn, K.A., Doane, L.D., 2009.Concordance between self-reported and objective wakeup timesin ambulatory salivary cortisol research. Int. J. Behav. Med.(July), 14.

Deuschle, M., Gotthardt, U., Schweiger, U., Weber, B., Korner, A.,Schmider, J., Standhardt, H., Lammers, C.H., Heuser, I., 1997.With aging in humans the activity of the hypothalamus-pituitary-adrenal system increased and its diurnal amplitude flattens. LifeSci. 61, 2239—2246.

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.

Ferrie, J.E., Shipley, M.J., Cappuccio, F.P., Brunner, E., Miller, M.A.,Kumari, M., Marmot, M.G., 2007. A prospective study of change insleep duration: associations with mortality in the Whitehall IIcohort. Sleep 30, 1659—1666.

Guralnik, J.M., Simonsick, E.M., Ferrucci, L., Glynn, R.J., Berk-man, L.F., Blazer, D.G., Scherr, P.A., Wallace, R.B., 1994. A shortphysical performance battery assessing lower extremity func-tion: association with self-reported disability and prediction ofmortality and nursing home admission. J. Gerontol. 49, M85—M94.

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

Ice, G.H., Katz-Stein, A., Himes, J., Kane, R.L., 2004. Diurnal cyclesof salivary cortisol in older adults. Psychoneuroendocrinology 29,355—370.

Jacobs, S., Mason, J., Kosten, T., Brown, S., Ostfeld, A., 1984.Urinary-free cortisol excretion in relation to age in acutely

stressed persons with depressive symptoms. Psychosom. Med.46, 213—221.

Karlamangla, A.S., Singer, B.H., Chodosh, J., McEwen, B.S., Seeman,T.E., 2005. Urinary cortisol excretion as a predictor of incidentcognitive impairment. Neurobiol. Aging 26 (Suppl. 1), 80—84.

Kirschbaum, C., Hellhammer, D.H., 1989. Salivary cortisol in psycho-biological research: an overview. Neuropsychobiology 22, 150—169.

Kudielka, B.M., Broderick, J.E., Kirschbaum, C., 2003. Compliancewith saliva sampling protocols: electronic monitoring revealsinvalid cortisol daytime profiles in noncompliant subjects. Psy-chosom. Med. 65, 313—319.

Kudielka, B.M., Schmidt-Reinwald, A.K., Hellhammer, D.H., Schur-meyer, T., Kirschbaum, C., 2000. Psychosocial stress and HPAfunctioning: no evidence for a reduced resilience in healthyelderly men. Stress 3, 229—240.

Lazarfeld, P.F., Henry, N.W., 1968. Latent Structure Analysis.Houghton Mifflin Co., Boston, MA.

Lo, Y., Mendell, N., Rubin, D., 2001. Testing the number of compo-nents in a normal mixture. Biometrika 88, 767—778.

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

McLachlan, G., Peel, D., 2000. Finite Mixture Models. John Wiley,New York.

Muthen, L., Muthen, B., 2006. Mplus User’s Guide, fourth edition.Muthen & Muthen, Los Angeles, CA.

Melzer, D., Lan, T.Y., Guralnik, J.M., 2003. The predictive validity formortality of the index of mobility-related limitation–—resultsfrom the EPESE study. Age Ageing 32, 619—625.

Nylund, K.L., Asparouhov, T., Muthen, B., 2007. Deciding on thenumber of classes in latent class analysis and growth mixturemodelling. A Monte Carlo simulation study. Struct. EquationModel..

Powell, L.H., Lovallo, W.R., Matthews, K.A., Meyer, P., Midgley, A.R.,Baum, A., Stone, A.A., Underwood, L., McCann, J.J., JanikulaHerro, K., Ory, M.G., 2002. Physiologic markers of chronic stressin premenopausal, middle-aged women. Psychosom. Med. 64,502—509.

Pruessner, J.C., Kirschbaum, C., Meinlschmid, G., Hellhammer, D.H.,2003. Two formulas for computation of the area under the curverepresent measures of total hormone concentration versus time-dependent change. Psychoneuroendocrinology 28, 916—931.

Raff, H., Raff, J.L., Duthie, E.H., Wilson, C.R., Sasse, E.A., Rudman,I., Mattson, D., 1999. Elevated salivary cortisol in the evening inhealthy elderly men and women: correlation with bone mineraldensity. J. Gerontol. A Biol. Sci. Med. Sci. 54, M479—M483.

Ranjit, N., Young, E.A., Raghunathan, T.E., Kaplan, G.A., 2005.Modeling cortisol rhythms in a population-based study. Psycho-neuroendocrinology 30, 615—624.

Rosmond, R., Wallerius, S., Wanger, P., Martin, L., Holm, G., Bjorn-torp, P., 2003. A 5-year follow-up study of disease incidence inmen with an abnormal hormone pattern. J. Intern. Med. 254,386—390.

Rubin, R.T., Poland, R.E., Lesser, I.M., Winston, R.A., Blodgett, A.L.,1987. Neuroendocrine aspects of primary endogenous depression.I. Cortisol secretory dynamics in patients and matched controls.Arch. Gen. Psychiatry 44, 328—336.

Schmidt-Reinwald, A., Pruessner, J.C., Hellhammer, D.H., Federenko,I., Rohleder, N., Schurmeyer, T.H., Kirschbaum, C., 1999. Thecortisol response to awakening in relation to different challengetests and a 12-hour cortisol rhythm. Life Sci. 64, 1653—1660.

Sephton, S.E., Sapolsky, R.M., Kraemer, H.C., Spiegel, D., 2000.Diurnal cortisol rhythm as a predictor of breast cancer survival.J. Natl. Cancer Inst. 92, 994—1000.

Smith, G.D., Ben-Shlomo, Y., Beswick, A., Yarnell, J., Lightman, S.,Elwood, P., 2005. Cortisol, testosterone, and coronary heartdisease: prospective evidence from the Caerphilly study. Circu-lation 112, 332—340.

Identifying patterns in cortisol secretion in an older population 1099

Smyth, J.M., Ockenfels, M.C., Gorin, A.A., Catley, D., Porter, L.S.,Kirschbaum, C., Hellhammer, D.H., Stone, A.A., 1997. Individualdifferences in the diurnal cycle of cortisol. Psychoneuroendocri-nology 22, 89—105.

Spiegel, D., Giese-Davis, J., Taylor, C.B., Kraemer, H., 2006. Stresssensitivity in metastatic breast cancer; analysis of the hypotha-lamic-pituitary-adrenal axis. Psychoneuroendocrinology 31,1231—1244.

Spiegel, K., Leproult, R., Van Cauter, E., 1999. Impact of sleepdebt on metabolic and endocrine function. Lancet 354, 1435—1439.

Steptoe, A., Cropley, M., Griffith, J., Kirschbaum, C., 2000. Job strainand anger expression predict early morning elevations in salivarycortisol. Psychosom. Med. 62, 286—292.

Steptoe, A., Ussher, M., 2006. Smoking, cortisol and nicotine. Int. J.Psychophysiol. 59, 228—235.

Tarullo, A.R., Gunnar, M.R., 2006. Child maltreatment and thedeveloping HPA axis. Horm. Behav. 50, 632—639.

Varadhan, R., Walston, J., Cappola, A.R., Carlson, M.C., Wand, G.S.,Fried, L.P., 2008. Higher levels and blunted diurnal variation of

cortisol in frail older women. J. Gerontol. A Biol. Sci. Med. Sci. 63,190—195.

Van Cauter, E., Leproult, R., Kupfer, D., 1996. Effects of gender andage on the levels and circadian rhythmicity of plasma cortisol. J.Clin. Endocrinol. Metab. 81, 2468—2473.

Van Ryzin, M.J., Chatham, M., Kryzer, E., Kertes, D.A., Gunnar, M.R.,2009. Identifying atypical cortisol patterns in young children: thebenefits of group-based trajectory modeling. Psychoneuroendo-crinology 34, 50—61.

Wang, Y., Brown, M.B., 1996. A flexible model for human circadianrhythms. Biometrics 52, 588—596.

Williams, E., Magid, E., Steptoe, A., 2005. The impact of timeof waking and concurrent subjective stress on the cortisol re-sponse to awakening. Psychoneuroendocrinology 30, 139—148.

Wust, S., Wolf, J., Hellhammer, D.H., Federenko, I., Schommer, N.,Kirschbaum, C., 2000. The cortisol awakening response–—normalvalues and confounds. Noise Health 2, 79—88.

Yehuda, R., Giller, E.L., Southwick, S.M., Lowy, M.T., Mason, J.W.,1991. Hypothalamic-pituitary-adrenal dysfunction in posttrau-matic stress disorder. Biol. Psychiatry 30, 1031—1048.