ARTICLE IN PRESS

Available at www.sciencedirect.com

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

Psychoneuroendocrinology (2008) 33, 530–539

0306-4530/$ - see frdoi:10.1016/j.psyne

Abbreviation: t1,measure; SEP, socioCI, confidence inter�Corresponding au

E-mail address: c

Cognitive development and cortisol patternsin mid-life: Findings from a British birth cohort

Chris Powera,�, Leah Lia, Clyde Hertzmanb

aUCL Institute of Child Health, Centre for Paediatric Epidemiology and Biostatistics, 30 Guilford Street,London WC1N 1EH, UKbHuman Early Learning Partnership, Faculty of Graduate Studies, University of British Columbia,Vancouver, BC, Canada V6T 1Z3

Received 18 June 2007; received in revised form 18 January 2008; accepted 21 January 2008

KEYWORDSCortisol;Cognition;Education;Childhood;Adulthood;Birth cohort

ont matter & 2008uen.2008.01.017

time 1 cortisol m-economic positival.thor. Tel.: +44 207

.power@ich.ucl.a

SummaryLife-course associations among cortisol, cognitive development and educational attain-ment in the general population are not well understood. Using the 1958 British birthcohort, our aim was to establish whether cortisol patterns at age 45 y are associated withchildhood cognition and qualification level by adulthood. We measured salivary cortisol in6527 individuals, 45min after waking (t1) and 3 h later (t2). To identify lack of morningcortisol peak and diurnal rhythm, we defined groups with: (a) t1 cortisol in the bottom 5%of the distribution, or (b) ‘flat’ t1–t2 cortisol. Data on cognitive tests at ages 7, 11 and 16 yand educational level were used. All childhood cognitive tests (maths, reading, verbal andnon-verbal ability) were inversely associated, although not always significantly, with low t1and flat t1–t2 cortisol. For example, at age 11 for males, a standard deviation (SD) increasein maths score was associated with a 28% decreased odds for lowest t1 cortisol, and with a13% decreased odds of flat t1–t2 cortisol. Associations for lowest t1 and flat t1–t2 cortisolwere attenuated after adjustment for qualification level at 33 y among males, althoughadjustment for childhood socio-economic position had little effect. Weaker associationsfor lowest t1 cortisol among females were either unchanged or strengthened afteradjustment for qualification level. Our results for males, but less so for females, arecompatible with a causal relationship in either direction, namely from cortisol to cognitiveability or vice versa.& 2008 Elsevier Ltd. All rights reserved.

Elsevier Ltd. All rights reserved.

easure; t2, time 2 cortisolon; SD, standard deviation;

905 2106.

c.uk (C. Power).

1. Introduction

There are several reasons to expect that cortisol andcognition are related. The brain is a major target organfor corticosteroids (Belanoff et al., 2001). Gluccorticoidsmodulate neuro-transmitter systems, regulate the plasticity

ARTICLE IN PRESS

Cognitive development and adult cortisol patterns 531

and circuitry of many brain regions, and influence theattention and concentration functions of the pre-frontalcortex (Davis et al., 2002; Lupien et al., 2005). In the short-term cortisol influences memory as well as the acquisitionand consolidation of information. Children exhibiting in-creases in cortisol in response to a challenge have beenshown to have higher performance on measures of executivefunction (a peg tapping inhibitory control task and an itemselection task) than those who do not (Blair et al., 2005).Despite these insights, there is little population-basedevidence linking the acquisition of cognitive skills andcortisol. However, from the limited evidence available todate, based on studies of children aged 6–14 years, therelationship appears to be bi-directional; that is, witheffects of gluccorticoids on cognitive function and, con-versely, effects of cognitive processing on gluccorticoidsecretion (Lupien et al., 2005). Disruption of the normaldiurnal rhythm of cortisol may also be important forcognition. The normal diurnal rhythm is characterised by apost-waking peak and subsequent decline over the day, butwith approximately 10% of individuals lacking the post-waking peak (Stone et al., 2001). It has been argued that alack of an early morning cortisol peak under non-experi-mental conditions could reduce a child’s capacity to remainalert during the day (Gunnar and Vazquez, 2001). Haleyet al. (2006) provide evidence that the ability to mount acortisol response to a learning task is related to memoryconsolidation, at least in newborns. Brandtstadter et al.(1991) demonstrated that high salivary cortisol at 8am wasassociated with higher educational attainment. Severalinvestigators have suggested that there is an invertedU-shape relationship, such that extremely low and highcortisol levels impair memory consolidation (Herbert et al.,2006). Lower levels are thought to lead to under-stimulationof hippocampal mediated learning and memory whilst highercortisol levels lead to suppression as opposed to potentia-tion (Belanoff et al., 2001; Davis et al., 2002; Haley et al.,2006; Lupien et al., 2005).

Our interest here is in the relationship between cortisoland cognition across the life-course. Research on stimulationin early life (licking and suckling of baby rats during a criticalperiod of development) shows alterations in the expressionof genes that influence the development of responsepatterns in the HPA axis, and also of higher-order executivefunctions in the brain (Weaver et al., 2005). In humans,levels of environmental stimulation are known to vary bysocial group, such that lower social classes receive lessstimulation (Hart and Risley, 1995). Low socio-economicposition (SEP) is also related to higher levels of basal cortisolin children, emerging between ages 6 and 10 years (Lupienet al., 2000). However, this relationship may be unstablethereafter: between age 10 and 14 years, the directassociation between basal cortisol and SEP is supersededby an indirect effect whereby SEP is associated withdifferences in cognitive processing style that, in turn, areassociated with basal cortisol (Lupien et al., 2005). Inrespect of this latter proposed sequence, cortisol secretioncould be an outcome rather than a precursor of cognitivefunction. In sum, the studies of SEP and cortisol at differentstages in childhood suggest a dynamic bi-directionalrelationship between cortisol and cognition evolving overtime. A SEP gradient in cortisol metabolism has been

reported in adults, from the most to the least favourableSEP, with those in lower SEP having a slower rate of declinein cortisol over the day, resulting in higher evening levels(Cohen et al., 2006). This raises the prospect that reducedcognitive ability in childhood may lead to lower SEP inadulthood and, through that, to altered cortisol patterns.Finally, with respect to cortisol and cognition at later stagesof the lifecourse, chronic exposure to high levels of cortisolis associated with memory impairments in the elderly(Seeman et al., 1997; Karlamangla et al., 2005; Lupienet al., 2005; Li et al., 2006).

Despite these insights, the life-course associations amongcortisol, cognitive development and educational attainmentin the general population are not well understood. This isnot surprising because understanding the direction ofassociation requires information on cortisol patterns in earlylife, followed by data on cognitive trajectories and educa-tional outcome some decades later. To our knowledge, suchtrue prospective data do not exist. Yet, cognitive trajec-tories have been delineated in several population-basedstudies (Jefferis et al., 2002; Richards and Sacker, 2003),providing an opportunity to examine the cognition, educa-tion and cortisol relationship retrospectively.

We anticipate that the social environment in early lifeaffects cortisol patterns, which in turn, influence cognitivedevelopment in childhood, setting in train dynamic pro-cesses linking school success, and thence qualification levelachieved by adulthood. We are not able to evaluate thisprocess prospectively, rather we assess it indirectly, byinvestigating whether cortisol patterns in mid-adulthood areassociated with cognition at different ages in childhood andwith later educational achievement in the 1958 British birthcohort (Power and Elliott, 2006). Although we are unable todetermine the temporal sequence, that is, whether cogni-tive function leads to cortisol patterning or vice versa, wecan establish whether any associations between childhoodcognition and adult cortisol are dependent on early socialorigins and adult educational achievement. Our previouswork showed that less advantaged SEP in childhood isassociated with poorer cognitive development and educa-tional outcome (Jefferis et al., 2002), and also that adultcortisol patterns vary by SEP (Li et al., 2007), thushighlighting the need to take account of early social originsin studies of cognition and adult cortisol.

2. Methods

2.1. Study population

The 1958 cohort includes all children born in England,Scotland and Wales, in one week in March 1958. A populationof about 17,000 live births were followed-up at ages 7, 11,16, 23, 33, 42 y (Power and Elliott, 2006). More recently, atarget sample of 11,971 participants identified as still incontact with the study, and at age 42 had not required aproxy interview (e.g. due to learning disability) were invitedto a clinical examination undertaken in their home by atrained nurse; 9377 (78%) participants were seen September2002–March 2004 the majority aged 45 y (68%), some aged44 y (31%) and only a minority aged 46 y (o1%). Ethical

ARTICLE IN PRESS

C. Power et al.532

approval for the 45-year biomedical survey was given by theSouth East Multi-Centre Research Ethics Committee.

The majority of 45-year participants (n ¼ 9165) con-sented to saliva collection for cortisol measurement, butprobably because the protocol specifications for salivacollection (described below) relied on participants collect-ing and returning saliva samples, we obtained usable data oncortisol level at 45 y for 6527 (3211 male and 3316 female)participants. Those with data on cortisol had higher mathand reading scores in childhood (for example, at age 7 meanmath and reading scores were 5.38 and 24.36) compared tothe original birth sample (respectively, 5.11 and 23.34).They were also more likely to have a higher education levelby age 33 y (32.2% men and 29.6% women had higher thanA-level, compared to 29.2% and 25.9% in the originalsample, respectively), and were less likely to originate fromunskilled manual social class: 24.2% were from classes IV&Vat age 7 y, versus 26.7% in the original sample.

2.2. Measures

2.2.1. Salivary cortisolAt the end of the nurse interview, participants wererequested to collect two saliva samples on the nextconvenient day, the first at 45min after awakening (time1) and the second, 3 h later on the same day (time 2).A reminder was sent to 53% of those consenting to return asaliva sample, if they had not done so within 2 weeks of thenurse visit. Samples were received from 6568 participants,6527 yielding information on at least one cortisol measure(6470 with time 1 and 6509 with time 2 measure).

Participants were instructed to avoid brushing or flossingtheir teeth, or eating or drinking for 15min before takingeach sample. They were asked to chew on a salivette until itwas soaked, record the date and time of collection, andstore the sample at room temperature until mailed to thelaboratory. Salivary cortisol is stable at room temperaturefor up to 30 days but samples were frozen after reaching thelaboratory to reduce microbial growth. Cortisol levels weremeasured at the University of Dresden with a commercialimmunoassay kit with chemiluminescence detection (CLIA,IBL-Hamburg, Hamburg Germany). The lower sensitivity ofthis assay is 0.44 nmol/L, with intraassay and interassayprecision of o10% for a wide range of cortisol concentra-tions. High cortisol levels (450 nmol/l) were rerun in asecond assay for confirmation. Participants also reported: (i)whether they regularly worked at night (shift-work), (ii)wakefulness during the previous night, (iii) dental workwithin the last three days, (iv) cuts inside their mouth thatmay bleed, and (v) current medications.

2.2.2. Cognition and educationAt ages 7, 11, and 16 y, participants took age appropriatetests at school for maths and reading ability. The arithmetictest at age 7 comprised 10 problems with graded levels ofdifficulty (range 0–10); teachers read the questions to poorreaders. At age 11, the mathematics test was constructed bythe National Foundation for Educational Research in Englandand Wales (range 0–40). At 16 years, a mathematicscomprehension test was constructed at Manchester Uni-versity (range 0–31). The Southgate test was used to detect

poor readers at age 7 (range 0–30) (Southgate, 1962):children selected from several words the one correspondingto a picture, teachers also read out words that the childrenidentified from a list. Reading tests at ages 11 and 16 wereparallel to the Watts Vernon comprehension test (range0–35). A general ability test at age 11 approximated theconventional intelligence test, with verbal (range 0–40) andnon-verbal (range 0–40) components (Douglas, 1964). High-est qualifications achieved by age 33 were categorised asfive groups: no qualification, less than O level (or equiva-lent), O level (or equivalent), A level (or equivalent), orhigher qualification; broadly equivalent to US categories: noqualifications, ograde 10; grade 10; high school diploma/grade 12; 4high school diploma. Education level wastreated as a categorical measure in all analyses.

2.2.3. SEP in childhoodSocial class in childhood was based on father’s occupation atbirth and at age 7 and categorised as classes I&II (profes-sional/managerial), IIINM (skilled non-manual), IIIM (skilledmanual), and IV&V (semi-unskilled manual). We derived acumulative measure of childhood SEP by summing acrosssocial class at birth and age 7 and then dividing by two.Social class at one age was used when the other one wasmissing. The measure of SEP (range 1–4) was used as acontinuous variable.

2.3. Analysis

Extreme outliers for time 1 (t1) and time 2 (t2) weretruncated: at 2 nmol/l for o2 nmol/l (n ¼ 24 at t1; n ¼ 123at t2) and also at 100 nmol/l for 4100 nmol/l (n ¼ 22 at t1;n ¼ 20 at t2) in order that potentially implausible values didnot exert a disproportionate influence on the analyses.Cortisol values were then transformed using log 10 toreduce the skewness in the distribution. Not all sampleswere collected at the specified times, leading to variationin times around the target for t1 (mean (sd) of 49(15)min)and t2 (mean (sd) of 3 h 5min (23min)). Cortisol levelwas influenced by both the time of awaking and timesince awaking, and in order to take account of thisvariability, the transformed cortisol values for each indivi-dual were centred at 0808 h (45min after mean waking timeof 0723 h) and t2 at 1108 h (3 h 45min after mean awakeningtime) by back transforming the predictions from linearregression models at specified times, to the original scale(nmol/l). Thus, t1 and t2 cortisol values used in all analysesare adjusted for both time of awaking and time sinceawaking.

We examined associations for potential confoundingfactors in relation to t1 cortisol, including gender, shift(night) work, awake during the night prior to salivacollection, cuts inside mouth, recent dental treatment,current medication, and day of the week. On average, menhad a lower t1 cortisol level, but a higher t2 level thanwomen (Table 1). Regular shift (night) working (8%), dentaltreatment within the previous three days (1.5%), cuts insidemouth (3.3%), and current medication were not associatedwith t1 cortisol level. Sleep disturbance during the previousnight, i.e. awake between 2400 and 0600 h (37.3%) wasweakly associated with a reduced t1 level, but only in

ARTICLE IN PRESS

Table 1 Summary of cortisol values and time of measurement and cognitive measures.

Males Females p�

Mean (sd) n Mean (sd) n

Cortisol measuresTimes of measures (h:min)Waking up time 7:23 (1:22) 3015 7:24 (1:12) 3200Time 1 8:12 (1:22) 3173 8:13 (1:12) 3298Time since waking 0:49 (0:15) 3014 0:49 (0:15) 3197Time 2 11:17 (1:26) 3061 11:17 (1:13) 3170Interval (time 2–time 1) 3:06 (0:26) 3059 3:04 (0:18) 3166

Time 1 cortisol (nmols/l)y 21.01 (11.94) 3187 21.88 (12.15) 3283 0.001Median 18.80 19.60

Time 2 cortisol (nmols/l)y 9.20 (9.02) 3200 8.24 (7.63) 3309 o0.001Median 7.10 6.60

AUCz (range 6–300 nmols/l) 44.82 (24.82) 3176 44.75 (22.58) 3276 0.28Median 40.11 40.68

% n % n

Flat t1–t2 cortisoly 11.2 355 8.7 285Lowest 5% time 1 cortisol 4.9 157 5.2 170Highest 5% time 1 cortisol 5.1 164 5.2 171

Mean (sd) n Mean (sd) n

Cognitive measuresMath scoreAt 7 y (range 0–10) 5.22 (2.50) 7646 5.00 (2.48) 7253At 11 y (range 0–40) 16.81 (10.60) 7255 16.44 (10.08) 6874At 16 y (range 0–31) 13.39 (7.27) 6104 12.09 (6.64) 5817

Reading scoreAt 7 y (range 0–30) 22.44 (7.43) 7674 24.29 (6.70) 7257At 11 y (range 0–35) 15.93 (6.55) 7257 16.03 (6.02) 6876At 16 y (range 0–35) 25.35 (7.34) 6129 25.27 (6.83) 5858

Verbal ability at 11 y (range 0–40) 21.05 (9.46) 7256 23.13 (9.13) 6878

Non-verbal ability at 11 y (range 0–40) 20.76 (7.68) 7256 22.01 (7.54) 6878

% n % n

Education level at 33 yNo qualification 9.2 502 10.8 612oO level 13.6 742 16.8 957O level 23.9 1301 36.2 2060A level 24.1 1317 10.3 588Higher 29.2 1592 25.9 1471

�p-value for male and female difference.yTruncated at 2 and 100 nmols/l.zCalculated using centred t1 and t2 cortisol values.yFlat ¼ t2 cortisol level is within 20% of t1 value, or t2 value o120% of t1 value if t1 value is o7.5 nmols/l.

Cognitive development and adult cortisol patterns 533

women. Therefore, no adjustments were made in theanalyses. Smoking was strongly associated with increasedt1 and t2 cortisol levels, and therefore we adjusted forsmoking. All analyses were conducted for men and womenseparately, for three reasons: first, because of gender

differences in educational attainments in the era of thiscohort (Kerckhoff, 1990); second, because of evidence thatcortisol patterns differ for males and females; and third, toachieve comparability with parallel studies of growth(Power et al., 2006).

ARTICLE IN PRESS

C. Power et al.534

As expected, the most common pattern was of relativelyhigh t1 cortisol with a steep decline to t2, but other patternswere observed. For example, for some individuals there waslittle difference between t1 and t2. We constructed threecortisol measures based on the t1 and t2 samples: (1) log 10of t1 cortisol level (used to indicate the individual’s capacityto generate a morning peak); (2) area under the curve(AUC), derived from t1 and t2 values (adjusted, as describedabove, for both time of awaking and time since awaking andback-transformed) which were summed, multiplied by 3 h,and divided by 2. Thus, AUC represents the 3 h average of t1and t2, after allowing for the variation in collection timesand the non-normality of the cortisol distribution, and it isused here to indicate total 3 h exposure to cortisol); and (3)flat t1 to t2 level, used to differentiate individuals lacking adiurnal rhythm. ‘Flat’ was defined as a t2 cortisol within 20%of the t1 value, or when the t1 value was o7.5 nmols/l, flatt1–t2 was defined as a t2 value o120% of t1. Both log t1cortisol and log AUC were analysed as continuous variables.In order to examine whether both extreme high and low t1values were associated with cognition, as suggested bystudies showing U-shaped relationships, we identified thehighest and lowest 5% extremes of the t1 cortisol distribu-tion: the cut-offs used to identify the highest 5% were40.73 nmols/l for males and 41.69 nmols/l for females; cut-offs for the lowest 5% were 7.24 and 7.94 nmols/l,respectively.

Because the range of test scores differed between agesand cognitive measures, all childhood cognitive tests (i.e.scores for maths and reading, general verbal ability, andgeneral non-verbal ability) were transformed to internallyderived standard deviation (SD) scores (z-score) in order tocompare their effects. Associations between cognitivedevelopment and t1 cortisol and AUC were examined usinglinear regression; logistic regression was used for extremesof t1 cortisol and flat t1–t2 change, from which odds ratios(OR) and 95% confidence intervals (CI) were obtained. Stagesof analyses were as follows. Effects of each cognitivemeasure, including z-scores for maths and reading at ages 7,11, and 16 y, verbal and non-verbal test at age 11 y, andhighest educational level by age 33 y were estimated inunivariate models; non-linear relationships were tested withinclusion of a quadratic term. Repeated measures of mathsand reading z-scores allowed assessment of change in therelative position of an individual between successive ages.Conditional models were constructed by adding the priormeasure of cognition to univariate models to test effects ofchange in cognition z-scores during intervals in childhood(i.e. between ages 7–11 and 11–16 y).

Associations between cognitive measures and cortisolwere adjusted, first for childhood SEP, second for qualifica-tion level at age 33 y, and third for both childhood SEP andadult qualification. The rationale for these adjustments wasto establish whether any associations for childhood cogni-tion were independent of early social origins and adulteducational achievement. Further adjustments of thecognition/cortisol association were also made for smokingand adult SEP.

Finally, we repeated analyses to test whether methodo-logical considerations could account for our results, asfollows: (i) because scores for reading tests were skewed wealso analysed these data using quartiles of the scores.

Results were similar to those obtained for continuous scores,hence we present the latter only; (ii) because of variation inage of testing in childhood and the influence of this factor ontest score, we examined whether age at test affected thecognition/cortisol relationship. Effects of adjustment forthis factor were negligible (data not presented); (iii)because of the small sample biases mentioned above, weperformed analyses with corrections for the better cognitivescores among those with cortisol data at 45 y compared tothe original sample. We achieved this correction byweighting the population with cortisol data to the distribu-tion of cognitive scores at age 11 y. Compared to the resultsfrom unweighted analyses the cognition/cortisol associa-tions were slightly stronger, suggesting that the resultsreported here are conservative estimates of the childhoodcognition/cortisol patterns association.

3. Results

Men had a lower t1 cortisol (median 18.8 nmol/l), but ahigher t2 level (median 7.1 nmol/l) at 45 y, compared towomen (19.6 and 6.6 nmol/l, respectively) (Table 1). Someparticipants had a higher t2 measure than t1, but onaverage, cortisol levels declined, with women having agreater decline than men. AUC was similar for men andwomen; for both the range of AUC values is wide, indicatinglarge differences in total 3 h ‘‘free’’ cortisol. 11.2% of menand 8.7% of women were defined as ‘‘flat t1–t2’’, i.e. with at1–t2 change of o20% (Table 1). Descriptive statistics for thecognitive tests and educational qualifications by 33 y arealso given in Table 1.

No associations were found with cognitive measures andt1 cortisol and AUC as continuous variables, nor for thehighest 5% of t1cortisol. Therefore, we report results forlowest 5% t1 cortisol and flat t1–t2 pattern.

3.1. Lowest t1 cortisol

In unadjusted analyses, an increase in maths scores at allages was associated with a decrease in odds for having alowest t1 cortisol at 45 y: for males the odds decreased by20–28% for a SD increase in maths score; in femalesassociations were in the same direction as in males,although 95% CI include 1.0 (Table 2). Relative change inmaths score between ages 7 and 11 y was associated withlowest t1 cortisol in males. Figure 1 suggests that males withthe lowest t1 cortisol at 45 y gained skills in maths at aslower rate between ages 7 and 11 y relative to others. Thissuggestion was supported by conditional models in which a 1SD decrease in maths score between the two ages wasassociated with an increased odds of lowest t1 cortisol(OR ¼ 1.41, 95% CI 1.12, 1.79). Similarly for reading, a SDincrease in the test score was associated with decreasedodds of lowest t1 cortisol, with a 25–26% reduction for males(Table 2). With regard to changes in scores over time, a 1 SDdecrease in reading score between ages 11 and 16 y wasassociated with an increased odds (OR ¼ 1.59, 95% CI 1.14,2.22) for a low t1 cortisol. For females the associations withreading scores were weaker and 95% CIs include 1.0.Increased verbal and non-verbal ability was associated withdecreased odds of lowest t1 cortisol, by 29% for 1SD increase

ARTICLE IN PRESS

Table 2 Associations (odds ratios, 95% CI) between cognitive tests (z-scores) in childhood and lowest t1 cortisol at 45 y.

Lowest 5% t1 cortisol OR (95% CI)

Unadjusted Adjusted for childhood SEPand qualifications by 33 y�

Maths z-score Males 7 y 0.80(0.67,0.96)y 0.94(0.77,1.14)11 y 0.72(0.59,0.86)y 0.81(0.64,1.03)16 y 0.78(0.64,0.94)y 0.93(0.72,1.22)

Females 7 y 0.89(0.75,1.05) 0.91(0.76,1.10)11 y 0.86(0.72,1.02) 0.83(0.67,1.04)16 y 0.88(0.74,1.06) 0.79(0.63,1.00)y

Reading z-score Males 7 y 0.74(0.63,0.88)y 0.87(0.71,1.08)11 y 0.75(0.63,0.91)y 0.87(0.68,1.10)16 y 0.75(0.63,0.90)y 0.89(0.69,1.16)

Females 7 y 0.87(0.74,1.03) 0.92(0.75,1.13)11 y 0.90(0.76,1.07) 0.94(0.75,1.17)16 y 0.94(0.77,1.14) 0.90(0.69,1.18)

Verbal z-score Males 11 y 0.71(0.59,0.85)y 0.80(0.64,1.00)y

Females 0.87(0.73,1.03) 0.79(0.63,0.99)y

Non-verbal z-score Males 11 y 0.71(0.59,0.85)y 0.85(0.68,1.05)Females 0.82(0.69,0.98)y 0.79(0.64,0.98)y

�Cumulative measure of childhood SEP; qualification at 33 y was treated as a categorical measure.yStatistically significant at 0.05 level.

Cognitive development and adult cortisol patterns 535

on both test scores in males; and by 13% and 18%,respectively, in females. Table 3 shows that for educationalqualifications at 33 y, the odds of lowest t1 cortisol werereduced by more than half for males with any qualification,compared to those without a qualification; in females, theassociation was U-shaped with reduced odds for those withintermediate qualification levels (p ¼ 0.05 for a quadraticterm). Associations for cognitive tests, 7–16 y, and lowest t1cortisol were adjusted for childhood SEP and qualificationlevel at 33y. After adjustment, associations attenuatedamong males (Table 2), largely due to adjustment forqualifications rather than for childhood SEP; among femalesassociations were either unchanged or strengthened afteradjustment (Table 2). Further adjustments for adult SEP andsmoking did not alter the associations between childhoodcognitive tests and t1 cortisol (data not presented).

3.2. Flat t1 to t2 cortisol

In unadjusted analyses for males, there was a 12–14%reduction in odds of flat t1 to t2 cortisol associated with a SDincrease in maths and reading at ages 11 and 16 y and verbaland non-verbal ability at age 11 y (Table 4). A decline inmaths score during childhood was associated with increasedodds of flat t1–t2 cortisol: for example the OR for a 1SDdecline age 11–16 y was 1.27 (95% CI 1.01, 1.59). Forfemales, weaker associations for flat t1–t2 cortisol were inthe same direction as for males. Table 3 shows that asqualification level increased at 33 y the odds of having a flatt1–t2 cortisol decreased linearly for males, whilst forfemales there was no clear trend (p40.05). All associations

with childhood cognitive measures were abolished afteradjustment for childhood SEP and qualifications at 33 y,almost entirely through the latter (Table 4). Additionaladjustments for adult SEP and smoking did not affect thisfinding (data not presented).

4. Discussion

This study investigated associations between cognitivedevelopment and cortisol in a large birth cohort. Its maincontribution is the demonstration of inverse associations ofcognition in childhood with low morning t1 and flat t1–t2cortisol in mid-adult life, independent of time of waking,sample collection and smoking. This inverse association withcortisol was found for all cognitive tests in childhood and foreducational qualifications by adulthood for males and, forsome cognitive tests, a greater risk of low morning t1 andflat t1–t2 cortisol was evident for males who declined inability relative to others during childhood. These findingsare consistent with a scenario in which males who do notscore well on childhood cognitive tests end up with fewerqualifications and, several decades later, are more likely tobe in the lowest 5% at t1 and have a flat t1–t2 cortisolpattern. Associations for females were consistent with thispattern but weaker. In general, for both sexes associationswere seen for the dichotomous cortisol variables (the lack ofa morning peak and diurnal rhythm), whereas associationswere absent when cortisol was represented by continuousvariables. It is remarkable that any associations could bedetected over an interval of three decades betweencognitive tests and cortisol measurement.

ARTICLE IN PRESS

Males

-0.35

-0.3

-0.25

-0.2

-0.15

-0.1

-0.05

0

0.05

Age (y)

Mea

n m

ath

scor

e (S

DS

)

Other Bottom 5% t1 value

Females

-0.35

-0.3

-0.25

-0.2

-0.15

-0.1

-0.05

0

0.05

Age (y)

Mea

n m

ath

scor

e (S

DS

)

7 11 16

7 11 16

Figure 1 Mean math standardised deviation scores (SDS), age7–16 y for the lowest 5% t1 cortisol versus others at 45 y.

C. Power et al.536

4.1. Strengths and limitations

As we acknowledged at the outset, the major limitation isthe temporal ordering of cortisol, cognition and educationlevel. An early childhood measure of cortisol would haveallowed a prospective analysis from cortisol to cognition.Given that cortisol patterns may not be stable over anindividual’s lifespan, associations between childhood mea-sures of cognition and cortisol may differ from thosereported here. Set against this limitation is the availabilityof repeat cognitive measures in childhood, together withadult educational outcome in an unselected nationwidesample.

Since salivary measures are relatively easy and inexpen-sive to collect they are well-suited for population studies.Ideally, an individual’s diurnal cortisol rhythm is obtainedfrom multiple saliva collections throughout a day, repeatedover several days to address the full diurnal rhythm andintra-individual variability. In our large population study a

maximum of two samples on one day was feasible andaffordable. The literature on diurnal rhythm has establishedthat decline in cortisol from the post-waking level occursmainly within the first few hours of the day and thisrepresents the normative pattern (Stone et al., 2001). In theabsence of unambiguous guidance from the literature onoptimal timing for two samples, we selected a measure soonafter waking followed by a pre-lunch measure 3 h later tocapture the pattern over this important period. Our analyticstrategy for two samples was directed towards abnormaldiurnal pattern, in respect of lack of morning peak andsubsequent decline. Because of the study size, precision inthe estimation of effects is gained at the group level eventhough our estimates for individuals may be less reliable.Absolute levels of cortisol in saliva are significantlylower than in blood, but are strongly correlated withserum cortisol (r ¼ 0.71–0.96) and are more closely corre-lated with the ‘‘free’’ cortisol fraction (Kirschbaum andHellhammer, 1994).

Due to attrition, the sample with complete data was lessthan half of the original birth cohort, leading to an under-representation of participants with lower cognitive scores inchildhood. Analyses weighted to allow for this bias con-firmed results presented here and suggested that estimatedcortisol/cognition associations are likely to be conservative.Moreover, we were able to establish that several factorspotentially affecting the validity of the cortisol measuresdid not affect our findings (see methods). Because our aimwas limited to establishing whether there was a life-courserelationship between cognition and cortisol, we argue that itis premature to further adjust for an exhaustive list ofpotential factors from different stages of the life-course.The analyses described in this paper are intended to be afirst step towards building a more complex understanding ofcortisol and human development.

4.2. Interpretation of findings

Studies to date provide insight into the associations betweencortisol and cognition at different life stages, but few haveexamined the relationship between cortisol and cognitionlongitudinally, over an extended period of life. For bothlowest t1 and flat t1–t2 cortisol in males, the increased riskof having poor childhood cognition scores was followed by anincreased risk of no educational qualifications by earlyadulthood. Because cognition in childhood is a determinantof final qualification level, it would not be surprising to findthat the cognition/cortisol association attenuated withadjustment for educational qualifications. In our adjustedanalyses, the associations with childhood cognition tendedto reduce, although not consistently for females amongwhom cortisol (lowest t1 and flat t1–t2)/qualificationassociations were non-linear. Our results are thereforeconsistent with opposing causal directions. First, the lowand flat cortisol patterns may have preceded cognitivetesting and educational outcome; consistent with theobservation that high salivary cortisol at 8am was associatedwith higher educational attainment in adulthood (Brandt-stadter et al., 1991). Second, the cortisol patterns may havefollowed cognitive testing and educational outcome, suchthat difficulties with cognition could result in more stressful

ARTICLE IN PRESS

Table 3 Associations (odds ratios, 95% CI) between qualification at 33 y and cortisol (lowest t1 and flat t1–t2) patterns at 45 y.

Qualification(33 y)�

Lowest 5% t1 cortisol OR (95% CI) Flat t1–t2 cortisol OR (95% CI)

Unadjusted Adjusted forchildhood SEPy

Unadjusted Adjusted forchildhood SEPy

MalesoO-level 0.36(0.17,0.74)z 0.34(0.17,0.72)z 1.01(0.61,1.67) 0.99(0.59,1.65)O-level 0.45(0.25,0.81)z 0.38(0.21,0.70)z 0.78(0.49,1.25) 0.70(0.43,1.13)A-level 0.44(0.24,0.78)z 0.41(0.22,0.74)z 0.66(0.42,1.06) 0.61(0.38,0.99)z

4A-level 0.37(0.20,0.65)z 0.34(0.18,0.62)z 0.49(0.31,0.78)z 0.42(0.25,0.69)z

FemalesoO-level 0.85(0.44,1.66) 0.96(0.48,1.91) 1.43(0.80,2.58) 1.51(0.82,2.75)O-level 0.63(0.34,1.17) 0.72(0.38,1.36) 1.13(0.65,1.97) 1.14(0.64,2.02)A-level 0.59(0.27,1.29) 0.67(0.30,1.52) 0.94(0.48,1.82) 0.91(0.45,1.81)4A-level 0.86(0.46,1.58) 1.08(0.56,2.08) 0.93(0.52,1.64) 0.90(0.49,1.66)

�Baseline ¼ no qualification.yCumulative measure of childhood SEP.zStatistically significant at 0.05 level.

Table 4 Associations (odds ratios, 95% CI) between cognitive tests (z-scores) in childhood and flat t1–t2 cortisol pattern at45 y.

Flat t1 to t2 cortisol OR (95% CI)

Unadjusted Adjusted for childhood SEPand qualifications by 33 y�

Maths z-score Males 7 y 0.99(0.88,1.12) 1.10(0.95,1.26)11 y 0.87(0.77,0.98)y 0.95(0.81,1.11)16 y 0.86(0.76,0.98)y 0.98(0.81,1.17)

Females 7 y 0.97(0.85,1.10) 1.01(0.87,1.17)11 y 0.90(0.79,1.03) 0.93(0.79,1.11)16 y 0.96(0.84,1.11) 1.05(0.87,1.26)

Reading z-score Males 7 y 0.90(0.79,1.02) 0.95(0.82,1.11)11 y 0.88(0.78,1.00)y 0.97(0.83,1.14)16 y 0.88(0.77,1.01)y 1.01(0.84,1.22)

Females 7 y 0.97(0.84,1.12) 0.95(0.80,1.12)11 y 0.96(0.84,1.10) 0.99(0.83,1.19)16 y 0.95(0.82,1.11) 1.05(0.84,1.31)

Verbal z-score Males 11 y 0.86(0.76,0.97)y 0.93(0.80,1.08)Females 0.89(0.78,1.02) 0.86(0.72,1.02)

Non-verbal z-score Males 11 y 0.88(0.78,1.00)y 0.97(0.83,1.13)Females 0.93(0.82,1.07) 0.93(0.79,1.11)

�Cumulative measure of childhood SEP; qualification at 33 y was treated as a categorical measure.yStatistically significant at 0.05 level.

Cognitive development and adult cortisol patterns 537

life circumstances and eventually disrupted patterns incortisol. This direction of association is consistent withevidence that the diurnal pattern may flatten with increas-ing age (Herbert et al., 2006). The stresses of daily living arethought to exert ‘wear and tear’ on the HPA axis, such thatcortisol metabolism becomes dysregulated (McEwen, 1998;Kristenson et al., 2004). One interpretation of our results is

that lacking qualifications would be associated with accel-erated ‘wear and tear’ on the HPA axis, which for someindividuals may result in a flattening of the diurnal rhythm(Herbert et al., 2006). (Notable in this regard, our findingsapply to a low t1 and flat t1–t2 cortisol pattern, but not tocontinuous measures, which highlights the need investigatepatterns that may indicate dysregulation of the HPA.)

ARTICLE IN PRESS

C. Power et al.538

Findings for males in the 1958 cohort are consistent withboth causal directions. A further possibility is that anassociation between cortisol and cognition is due to acommon cause, such as inherited predisposition, althoughthis cannot be addressed in the present study.

Associations between childhood cognition and cortisol inmid-life were weaker for females and findings for educa-tional qualifications varied between cortisol measures.Hence, there was little support for either direction ofcausation in females. Sex differences in cortisol patternshave been reported in many studies (Goodyer et al., 2001;Steptoe et al., 2003; Polk et al., 2005; Herbert et al., 2006;Kajantie and Phillips, 2006). These differences have beenattributed to the influence of sex hormones. Uncertaintiesremain in the literature on how sex hormones affect therelationship of cortisol with other factors, such as growth,psychopathology and cognition. A further possibility is thatthere were ‘gender’ differences in educational expectationsduring the childhood of this generation, such that femaleshad less pressure to achieve academically than males, withconsequent lower demand on the HPA axis.

In conclusion, the association between cognition andcortisol has been established in our large population-basedstudy, at least in males, but the causal direction remainsuncertain.

Role of the funding sources

Data collection at age 45 years was funded by the MedicalResearch Council, Grant G0000934. Analysis was funded bythe MRC and the Human Early Learning Partnership (HELP),Vancouver, Canada. Research at the Institute of Child Healthand Great Ormond Street Hospital for Children NHS Trustbenefits from R&D funding received from the NHS Executive.

Conflict of interest

We confirm that there is no conflict of interest associatedwith any contribution to this paper.

Acknowledgements

Cortisol levels were measured under the direction ofProfessor Kirschbaum (Biological Psychology, Departmentof Psychology, University of Dresden, Germany). Dataproviders: Centre for Longitudinal Studies, Institute ofEducation and National Birthday Trust Fund, NationalChildren’s Bureau, City University Social Statistics ResearchUnit.

References

Belanoff, J.K., Gross, K., Yager, A., Schatzberg, A.F., 2001.Corticosteroids and cognition. J. Psychiatr. Res. 35, 127–145.

Blair, C., Granger, D., Razza, R.P., 2005. Cortisol reactivity ispositively related to executive function in preschool childrenattending head start. Child Dev. 76, 554–567.

Brandtstadter, J., Baltes-Gotz, B., Kirschbaum, C., Hellhammer, D.,1991. Developmental and personality correlates of adrenocor-tical activity as indexed by salivary cortisol: observations in theage range of 35–65 years. J. Psychosom. Res. 35, 173–185.

Cohen, S., Doyle, W.J., Baum, A., 2006. Socioeconomic status isassociated with stress hormones. Psychosom. Med. 68, 414–420.

Davis, E.P., Bruce, J., Gunnar, M.R., 2002. The anterior attentionnetwork: associations with temperament and neuroendocrineactivity in 6-year-old children. Dev. Psychobiol. 40, 43–56.

Douglas, J., 1964. The Home and the School. MacGibbon and Kee,London.

Goodyer, I.M., Park, R.J., Netherton, C.M., Herbert, J., 2001.Possible role of cortisol and dehydroepiandrosterone in humandevelopment and psychopathology. Br. J. Psychiatry 179,243–249.

Gunnar, M.R., Vazquez, D.M., 2001. Low cortisol and a flattening ofexpected daytime rhythm: potential indices of risk in humandevelopment. Dev. Psychobiol. 13, 515–538.

Haley, D.W., Weinberg, J., Grunau, R.E., 2006. Cortisol, contin-gency learning, and memory in preterm and full-term infants.Psychoneuroendocrinology 31, 108–117.

Hart, B., Risley, T.R., 1995. Meaningful Differences in EverydayExperience of Young American Children. Paul H Brooks PublishingCompany, Baltimore.

Herbert, J., Goodyer, I.M., Grossman, A.B., Hastings, M.H., deKloet, E.R., Lightman, S.L., Lupien, S.J., Roozendaal, B., Seckl,J.R., 2006. Do corticosteroids damage the brain? J. Neuroendo-crinol. 18, 393–411.

Jefferis, B.J., Power, C., Hertzman, C., 2002. Birth weight,childhood socioeconomic environment, and cognitive develop-ment in the 1958 British birth cohort study. Br. Med. J. 325, 305.

Kajantie, E., Phillips, D.I., 2006. The effects of sex and hormonalstatus on the physiological response to acute psychosocial stress.Psychoneuroendocrinology 31, 151–178.

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

Kerckhoff, A.C., 1990. Getting Started: Transition to Adulthood inGreat Britain. Westview Press, Boulder, US.

Kirschbaum, C., Hellhammer, D.H., 1994. Salivary cortisol inpsychoneuroendocrine research: recent developments and ap-plications. Psychoneuroendocrinology 19, 313–333.

Kristenson, M., Eriksen, H.R., Sluiter, J.K., Starke, D., Ursin, H.,2004. Psychobiological mechanisms of socioeconomic differ-ences in health. Soc. Sci. Med. 58, 1511–1522.

Li, G., Cherrier, M.M., Tsuang, D.W., Petrie, E.C., Colasurdo, E.A.,Craft, S., Schellenberg, G.D., Peskind, E.R., Raskind, M.A.,Wilkinson, C.W., 2006. Salivary cortisol and memory function inhuman aging. Neurobiol. Aging 27, 1705–1714.

Li, L., Power, C., Kelly, S., Kirschbaum, C., Hertzman, C., 2007.Life-time socio-economic position and cortisol patterns in mid-life. Psychoneuroendocrinology 32, 824–833.

Lupien, S.J., King, S., Meaney, M.J., McEwen, B.S., 2000. Child’sstress hormone levels correlate with mother’s socioeconomicstatus and depressive state. Biol. Psychiatry 48, 976–980.

Lupien, S.J., Fiocco, A., Wan, N., Maheu, F., Lord, C., Schramek, T.,Tu, M.T., 2005. Stress hormones and human memory functionacross the lifespan. Psychoneuroendocrinology 30, 225–242.

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

Polk, D.E., Cohen, S., Doyle, W.J., Skoner, D.P., Kirschbaum, C.,2005. State and trait affect as predictors of salivary cortisol inhealthy adults. Psychoneuroendocrinology 30, 261–272.

Power, C., Elliott, J., 2006. Cohort profile: 1958 British birthcohort (national child development study). Int. J. Epidemiol. 35,34–41.

Power, C., Li, L., Hertzman, C., 2006. Associations of early growthand adult adiposity with patterns of salivary cortisol inadulthood. J. Clin. Endocrinol. Metab. 91, 4264–4270.

Richards, M., Sacker, A., 2003. Lifetime antecedents of cognitivereserve. J. Clin. Exp. Neuropsychol. 25, 614–624.

ARTICLE IN PRESS

Cognitive development and adult cortisol patterns 539

Seeman, T.E., McEwen, B.S., Singer, B.H., Albert, M.S., Rowe, J.W.,1997. Increase in urinary cortisol excretion and memorydeclines: MacArthur studies of successful aging. J. Clin.Endocrinol. Metab. 82, 2458–2465.

Southgate, V., 1962. Southgate Group Reading Tests: Manual ofInstructions. University of London Press, London.

Steptoe, A., Kunz-Ebrecht, S., Owen, N., Feldman, P.J., Willemsen,G., Kirschbaum, C., Marmot, M., 2003. Socio-economic statusand stress-related biological responses over the working day.Psychosom. Med. 65, 461–470.

Stone, A.A., Schwartz, J.E., Smyth, J., Kirschbaum, C., Cohen, S.,Hellhammer, D., Grossman, S., 2001. Individual differences inthe diurnal cycle of salivary free cortisol: a replication offlattened cycles for some individuals. Psychoneuroendocrinology26, 295–306.

Weaver, I.C., Champagne, F.A., Brown, S.E., Dymov, S., Sharma, S.,Meaney, M.J., Szyf, M., 2005. Reversal of maternal programmingof stress responses in adult offspring through methyl supple-mentation: altering epigenetic marking later in life. J. Neurosci.25, 11045–11054.