Association of Diurnal Patterns in Salivary Cortisol with All-Cause and Cardiovascular Mortality: Findings from the Whitehall II Study

A flatter slope in the decline in cortisol measures across the day is predictive of cardiovascular but not noncardiovascular-related deaths in a cohort of early old aged men and women.

Abstract

Context:

Evidence for the association of cortisol with mortality or disease events is mixed, possibly due to a failure to consider diurnal cortisol patterns.

Objective:

Our objective was to examine the association of diurnal cortisol patterns throughout the day with cardiovascular and noncardiovascular mortality in a community-dwelling population.

Design:

This was a prospective cohort study among 4047 civil servants, the Whitehall II study, United Kingdom. We measured diurnal cortisol patterns in 2002–2004 from six saliva samples obtained over the course of a normal weekday: at waking, +30 min, +2.5 h, +8 h, +12 h, and bedtime. Participants were subsequently followed for all-cause and cause-specific mortality until January 2010.

Participants:

Participants included 4047 men and women aged 61 yr on average at baseline.

Outcomes:

We assessed all-cause, cardiovascular, and noncardiovascular death.

Results:

There were 139 deaths, 32 of which were deaths due to cardiovascular disease, during a mean follow-up period of 6.1 yr. Flatter slopes in cortisol decline across the day were associated with increased risk of all-cause mortality (hazard ratio for 1 sd reduction in slope steepness 1.30; 95% confidence interval (CI) = 1.09–1.55). This excess mortality risk was mainly driven by an increased risk of cardiovascular deaths (hazard ratio = 1.87; 95% confidence interval = 1.32–2.64). The association with cardiovascular deaths was independent of a wide range of covariates measured at the time of cortisol assessment. There was no association between morning cortisol, the cortisol awakening response, and mortality outcomes.

Conclusions:

These findings demonstrate, for the first time, the relationship between a flatter slope in cortisol levels across the day and an increased risk of cardiovascular disease mortality in a nonclinical population.

Evidence from population cohort studies suggests a link between psychosocial stress and an increased risk of coronary heart disease (1). A common assumption is that this association is attributable to impacts of the neuroendocrine changes associated with the stress response. However, empirical evidence to demonstrate such adverse effects is surprisingly modest because there have been few methodologically sound prospective studies in this area, principally due to difficulties in measuring biomarkers of stress in large population studies. This is particularly apparent for the assessment of the hypothalamic-pituitary-adrenal (HPA) axis, a main neuroendocrine system that is responsive to stressful stimuli, which shows marked diurnal patterning (2). Thus, evidence from studies that have examined plasma or serum cortisol, which is a product of the HPA axis, with clinical end points is equivocal, with studies suggesting both low (3) and high (4, 5) serum cortisol levels to be predictive of clinical events or death in patient populations and no association in cohort studies (6–8). A recent study has found that total 24-h urinary cortisol output is associated with cardiovascular but not noncardiovascular mortality (9).

The diurnal rhythm in cortisol release is likely to be a contributing factor to these inconsistencies between studies. Recently, the diurnal pattern has been addressed by the assessment of cortisol in saliva samples which allows measurement across the day in a relatively noninvasive manner (2). The general pattern is characterized by relatively high cortisol level on waking followed by a rise that reaches a peak at 30 min after waking, termed the cortisol awakening response (CAR), followed by a decline across the day reaching a nadir at midnight. Studies that fail to show an association of cortisol levels with clinical events have generally measured morning cortisol or serum cortisol that was not analyzed in relation to waking.

The few studies that have examined cortisol across the day find that flatter slopes or less steep declines in cortisol throughout the day are predictive of mortality in patient populations (10). In a study of elderly people, raised saliva evening levels of cortisol were predictive of mortality in women but not men, whereas high morning levels were predictive in men but not women (7). The interpretation of these findings is problematic because the presence of severe medical comorbidity in this population may either compound or obscure the association between diurnal cortisol patterns and mortality.

The Whitehall II study is a prospective cohort study targeting a large cohort of community-dwelling men and women. A major strength of this study is the possibility to determine diurnal cortisol pattern based on six saliva samples obtained over the course of a normal weekday: at waking, +30 mins + 2.5 h, +8 h, +12 h, and bedtime. A further advantage is that the data include measurement of conventional risk factors and a follow-up of overall, cardiovascular, and noncardiovascular deaths based on comprehensive medical records. We therefore sought to examine the association of diurnal patterns of cortisol with cardiovascular and noncardiovascular mortality in the Whitehall II study. More specifically, we examine two aspects of the diurnal cortisol pattern, the CAR and the diurnal slope in cortisol across the day.

Subjects and Methods

Study population

Data reported here are from phase 7 (2002–2004) of the Whitehall II study, which was initially recruited between 1985 and 1988 (phase 1) from 20 London-based civil service departments; 10,308 people participated, and details of the clinical cohort have been reported elsewhere (11). The number participating at phase 7 was 6941 when mean age of participants was 61 yr (age range 50.5–73.9). Saliva sample collection was initiated part way through phase 7, and of those participants that were asked to collect saliva samples, 90.1% (n = 4609) returned samples. Ethical approval for the Whitehall II study was obtained from the University College London Medical School committee on the ethics of human research. Informed consent for involvement in the study was gained from every participant.

Cortisol collection and analysis

Saliva samples collection protocol has been described previously (12). The collection of saliva samples from participants was initiated part way through phase 7. Briefly, in a face-to-face interview, participants were requested to provide six saliva samples in salivettes over the course of a normal weekday at waking, +30 min, +2.5 h, +8 h, +12 h, and bedtime. Participants were instructed to not brush teeth or eat or drink anything for 15 min before sample collection. An instruction booklet was used to record information on the day of sampling including date of collection, wake time, time each sample was taken, and stressful events. The salivettes and booklet were returned via post. Salivettes were centrifuged at 3000 rpm for 5 min, resulting in a clear supernatant of low viscosity. Salivary cortisol levels were measured using a commercial immunoassay with chemiluminescence detection (CLIA; IBLHamburg, Hamburg, Germany). The lower concentration limit of this assay is 0.44 nmol/liter; intra- and interassay coefficients of variance were below 8%. Any sample over 50 nmol/liter was reanalyzed.

Mortality

Mortality follow-up was available through the National Health Services Central Registry until January 31, 2010; a mean of 6.1 yr from phase 7. Registration of death within 5 d is a legal requirement in the United Kingdom, so participants not registered can be assumed to be alive. Death certificates were coded using the 10th revision of the International Classification of Disease (ICD) and categorized as cardiovascular disease, ICD-10 codes I00-I99, and noncardiovascular disease, all remaining codes.

Assessment of covariates

We used standard protocols to assess characteristics of the participants at the time diurnal cortisol patterns were measured (phase 7, 2002–2004).

Demographic and sample collection variables

Age, sex, and current or most recent civil service employment grade, a measure of social position, were assessed by questionnaire. Waking up time was assessed by participants' records on the day of the collection of saliva. Time difference between waking and taking first sample was categorized into 5-min intervals.

Health behavior variables

Smoking status (12) was defined as current smokers vs. the noncurrent smokers. Participants were asked to record the time of falling asleep the night before, and sleep duration the night before sample collection was calculated from these responses (13).

Psychological/psychosocial variables

Stress on the day of cortisol sampling was measured by questions on whether the participant had experienced a stressful event and, if yes, how stressful this was. Responses were used as continuous response score of 1–5 or grouped into binary categories: no/not at all/moderate stress and somewhat/very stressful. Financial insecurity was assessed by the question, “Thinking of the next ten years, how financially secure do you feel?” Participants who responded that they felt fairly insecure or insecure were classified as financially insecure (14). Fatigue was assessed using the vitality subscale of the 36-item Short-Form Health Survey (SF-36) (15). The scale was scored such that a higher score was indicative of high vitality and a lower score of fatigue; distribution of the scale is skewed (range is 0–100; modal score = 80), and a population norm cut point of 50 was used to define fatigue, as previously described (15, 16).

Body mass index (BMI) was assessed by measurement of height and weight by a nurse. Height was assessed using a stadiometer with the head in the Frankfort plane, and weight was assessed using a portable digital scale (Tanita, Yiewsley, Middlesex, UK). BMI was categorized as obese (BMI > 30) or not (16). We recently described a nonlinear association of BMI with slope in cortisol secretion, and so BMI was also categorized using the cut point suggested by these analyses at 21 or less or 31 or more (17). Fasting glucose was measured from participants who had fasted for at least 5 h using the glucose oxidase method (18) on a YSI model 2300 Stat Plus analyzer (YSI Corp., Yellow Springs, OH) (mean coefficient of variation = 1.4–3.1%) (19).

Statistical analysis

Participants who reported taking steroid medications were removed from the analyses (n = 236), and cortisol values outside 3 sd from the mean were removed from the analyses (n = 14). Despite this, cortisol data were skewed and were therefore logged for analysis. Conventionally, analyses are restricted to samples that are collected within 10 min of waking (sample 1 taken >10 min; n = 615) due to a reduced CAR (20). However, we did not see a significant difference in lateness by subsequent mortality (p = 0.09), so all participants were retained, and adjustments were made by including time-delay dummies in the CAR models. The CAR was calculated by subtracting cortisol measured at time 1 (waking) from cortisol measured at time 2 (+30 min). The slope of the decline in cortisol levels over the day was calculated by regressing cortisol values on time after waking for samples 1 (waking), (+2.5 h), 4 (+8 h), 5 (+12 h), and 6 (bedtime). Because it is suggested that the CAR and slope in cortisol secretion are under different neurobiological control systems (2), sample 2 was not included to ensure that the CAR does not obscure the slope calculation. The diurnal slope in cortisol secretion across the day was derived from a hierarchical linear model to predict logarithmically transformed cortisol across the measurements in which measurement occasion was used as a level 1 identifier and person as a level 2 identifier with sample time as the independent variable and random intercepts and random slopes. The slope was estimated for each person as the overall slope (which was negative) plus the level 2 slope residual; lower (more negative) slopes indicate a more rapid decline in cortisol levels, whereas slope values closer to zero reflect flatter diurnal rhythms. The slope values were exported, and z-scores (mean = 0; sd = 1) were created for the CAR, slope, waking cortisol, and cortisol measures at bedtime. Cox proportional hazards models with follow-up period as the time scale were used to determine the hazard ratio (HR) [and 95% confidence interval (CI)], using the four standardized cortisol measures as linear terms. The adequacy of the linear terms was visually checked using plots created by fitting restricted cubic splines (21). These also allowed formal tests for nonlinearity (all P values >0.17), which indicated that all the cortisol-mortality relationships were well described using a single linear term. The proportional hazards assumption, tested by fitting interaction terms between the standardized cortisol measures and the logarithm of the follow-up period, was found to be not violated. Because results were virtually identical in women and men, pooled estimates are presented. The slope estimates were generated using 1 MLWin. version 2.10 beta 6, and other analyses were performed using SAS version 9.1.

Results

From the samples returned, 168 individual samples were not taken by participants, which equates to 0.6% of the total number of samples expected. During analysis, a total of 1002 individual samples (3.6%) were not assayed due to loss of sample in shipping to the laboratory or low saliva yield. The final number of cortisol samples for analyses was 24,121 among 4,047 participants with cortisol measures available. The average CAR was 7.31 (se = 0.179). The average diurnal slope estimated from the hierarchical linear model was −0.1288 nmol/liters·h (se = 0.023). The characteristics of participants who provided saliva samples were similar to all participants who took part at phase 7 of the study (Table 1).

Table 1.

Characteristics of participants at phase 7 (2002–2004) of the Whitehall II study

	Participants who attended phase 7 (n = 6968)	Participants included in analyses of mortality (n = 4047)
% male	70.2	73.6
Mean age (sd)	61.2 (6.0)	61.1 (5.9)
% current smokers	8.3	8.1
% lowest employment grades	12.2	10.3
BMI (sd)	26.8 (4.4)	26.7 (4.3)
% fatigue	20.4	19.2
Cardiovascular medication (%)	30.9	29.8

Fatigue is defined as a score of 50 or less in the vitality subscale of the SF-36.

Table 2 shows the characteristics of the participants who provided cortisol samples. Those who subsequently died were more likely to be obese, report less than 50 on the vitality subscale (i.e. fatigue), smoke, and have raised glucose levels at phase 7 than those who survived. Cortisol collection measures, such as waking time on day of sampling, were not different by survival status. Of the main causes of death, the increased risk of mortality was due to an increased risk of cardiovascular death rather than death due to cancer for which HR (95% CI) was 1.17 (0.93=1.49); n = 76 events per 4046 total population. The average CAR in participants who subsequently died vs. those who subsequently survived was 8.35 (95% CI = 4.40–12.30) vs. 7.31 (95% CI = 6.96–7.66); P = 0.61 after adjustment for age, sex, waking time, time since waking, and social position. The concomitant diurnal slope values were −0.1143 (95% CI = −0.1222 to −0.1065) vs. −0.1290 (−0.1297 to −0.1283); P = 0.0003.

Table 2.

Characteristics of participants at time of cortisol assessment (2002–2004) by survival status to January 2010

	Number of participants	All-cause mortality (n = 139)	Alive (n = 3908)	P value
Age (95% CI)	4047	65.2 (64.2–66.2)	60.9 (60.8–61.1)	<0.0001
Sex (% male)	4047	68.4	73.8	0.15
Lowest employment grade (%)	4047	12.3	10.3	0.4
Current smoker at phase 7 (%)	3992	10.9	6.6	0.05
Cardiovascular medication (%)	4034	49.6	29.1	<0.0001
Obese (%)	4034	33.3	18.1	<0.0001
Raised fasting glucose (%)	3966	10.1	17.1	0.011
Stress (%)	3656	7.8	6.9	0.71
Financial insecurity (%)	3995	13.9	10.0	0.14
Short sleep (%)	3822	7.6	5.6	0.31
Fatigue (%)	4015	30.9	18.7	<0.0001
Late saliva collection (%)	4047	19.4	14.3	0.09

Obese is classified as a BMI of 30 and over. Raised fasting glucose is defined as values of 6.1 or higher. Stress is a stressful event perceived as very stressful or the most stress ever experienced on day of sample collection. Financial insecurity is classified from a response of insecure or very insecure to a question on future perceived financial insecurity. Short sleep is sleep of 5 h or less per night. Fatigue is defined as a score of 50 or less on the vitality subscale of the SF-36. Late saliva collection is a reported sample collection 10 min later than reported waking time.

Table 3 shows HR for cardiovascular and noncardiovascular mortality per 1 sd increase in scores of the CAR and slope in cortisol across the day. No association was seen for cardiovascular or noncardiovascular mortality with the CAR. For slope in cortisol across the day, a linear association is observed with mortality attributed to cardiovascular disease. A flatter slope in cortisol patterns across the day can be due to low waking values or high evening values of cortisol. We examined the association of cardiovascular and noncardiovascular events with waking cortisol and cortisol measured at bedtime and found that waking cortisol levels were unrelated to subsequent mortality. In contrast, cortisol measures at bedtime were found to be predictive of subsequent cardiovascular-related mortality; HR per 1 sd increase in z-score was 1.98 (95% CI = 1.39–2.81), suggesting that constantly elevated rather stable low cortisol level across the day accounted for the increased death risk. Both slope in cortisol secretion and bedtime cortisol values were not significantly associated with an increased risk of noncardiovascular mortality.

Table 3.

HR of all-cause, cardiovascular, and noncardiovascular mortality among 4047 participants of the Whitehall II study from phase 7 (2002–2004) through to January 2010 by z-scores of measures of cortisol

	All-cause mortality	Noncardiovascular deaths	Cardiovascular deaths
Waking cortisol	0.94 (0.80–1.12)	0.93 (0.77–1.13)	0.95 (0.67–1.36)
CAR	0.94 (0.80–1.12)	0.90 (0.74–1.10)	1.12 (0.79–1.57)
Slope across the day	1.30 (1.09–1.55)	1.17 (0.96–1.43)	1.87 (1.32–2.64)
Bedtime cortisol	1.33 (1.11–1.59)	1.17 (0.96–1.44)	1.98 (1.39–2.81)

There were 139 all-cause deaths, 106 noncardiovascular deaths, and 32 cardiovascular deaths. One death from unknown cause was removed from the analysis. Numbers with bedtime cortisol were 137 all-cause deaths, 105 noncardiovascular deaths, and 32 events per 3977 total population for cardiovascular deaths. Data were adjusted for age, sex, employment grade, waking time on day of sample collection, and time of sample collection since waking at phase 7.

As shown in Table 4, the associations of slope in cortisol and bedtime cortisol level with cardiovascular mortality were robust to separate adjustments for each set of baseline covariates included in Table 2, the corresponding HR varying between 1.82 (1.29–2.57) and 2.06 (1.42–2.99) depending on the adjustment. In these analyses, obesity was independently predictive of both cardiovascular and noncardiovascular mortality [HR (95% CI) = 2.10 (1.28–3.47) and 1.72 (1.31–2.25), respectively, in analyses of bedtime cortisol]. Fatigue was also predictive of both cardiovascular and noncardiovascular deaths [for example, HR (95% CI) = 2.37 (1.12–5.04) and 1.72 (1.31–2.25) of each, respectively, in analyses of bedtime cortisol]. Interestingly, self-reported stress on day of cortisol sampling was independently associated with cardiovascular deaths [HR = 3.45 (95% CI = 1.40–8.49)]. We have previously described a nonlinear association of diurnal slope in cortisol across the day and measures of obesity (17). We therefore ran our analyses again using cut points suggested by the relation of slope with BMI (cut points 21 and 31), but there were no cardiovascular deaths in those with a BMI of less than 21 at phase 7, and using 31 as a cut point failed to influence the association of cortisol slope with cardiovascular or noncardiovascular disease. Analyses examining the association of slope in cortisol with cardiovascular deaths in participants who were not current smokers at phase 7 showed that slope remained associated with subsequent mortality [HR = 2.12 (95% CI = 1.49–3.01]. Similarly, when analyses were restricted to those that collected waking samples within 10 min of waking, results were unaffected [for example, HR for cardiovascular mortality = 1.81 (95% CI = 1.209–2.713)].

Table 4.

HR of cardiovascular and noncardiovascular mortality among participants of the Whitehall II study from phase 7 (2002–2004) through to January 2010 by z-scores of measures of cortisol before and after adjustment for covariates

	Noncardiovascular deaths		Cardiovascular deaths
	No. of events/total population	HR (95% CI)	No. of events/total population	HR (95% CI)
Slope in cortisola	106/4047	1.17 (0.96–1.43)	32/4047	1.87 (1.32–2.64)
Plus separate adjustment for
Smoking	106/3989	1.15 (0.94–1.40)	30/3974	1.96 (1.37–2.80)
Obese	103/4005	1.20 (0.98–1.47)	30/4005	1.92 (1.35–2.72)
Fasting glucose	99/3963	1.18 (0.96–1.45)	29/3963	2.00 (1.40–2.87)
Stress	96/3632	1.15 (0.93–1.42)	30/3632	1.77 (1.25–2.49)
Financial insecurity	104/3977	1.14 (0.93–1.39)	31/3977	1.96 (1.38–2.77)
Hours slept	101/3795	1.18 (0.96–1.46)	28/3795	1.86 (1.29–2.69)
Fatigue	102/3928	1.15 (0.93–1.41)	31/3997	1.93 (1.36–2.73)
Bedtime cortisola	105/3977	1.17 (0.96–1.44)	32/3977	1.98 (1.39–2.81)
Plus separate adjustment for
Smoking	105/3921	1.15 (0.94–1.41)	31/3921	1.99 (1.39–2.85)
Obese	102/3962	1.21 (0.98–1.49)	31/3962	2.01 (1.39–2.90)
Fasting glucose	98/3895	1.21 (0.98–1.49)	29/2895	2.15 (1.49–3.12)
Stress	95/3611	1.19 (0.96–1.47)	31/3611	1.91 (1.34–2.72)
Financial insecurity	103/3924	1.14 (0.93–1.40)	32/3924	1.97 (1.39–2.81)
Hours slept	100/3755	1.19 (0.96–1.45)	29/3755	2.07 (1.42–2.99)
Fatigue	102/3943	1.15 (0.93–1.41)	32/3943	1.93 (1.35–2.75)

Obese is classified as a BMI of 30 and over. Stress is the perceived stress, scored as follows: not at all = 1; somewhat = 2; moderate = 3; very = 4; and most stressed ever felt = 5. Financial insecurity is classified from a response of insecure or very insecure to a question on future perceived financial insecurity. Fatigue is defined as a score of 50 or less on the vitality subscale of the SF-36.

^aAdjusted for age, sex, employment grade, and time of sample collection since waking and waking time at phase 7 (2002–2004).

Discussion

To our knowledge, this is the first study to show that diurnal patterns in cortisol release across the day are predictive of subsequent cardiovascular-related mortality in men and women. Our findings suggest that slope in cortisol is related to cardiovascular-related mortality, comprising a raised late evening level of cortisol rather than a depressed morning level. Our data additionally suggest no significant association with non-cardiovascular-related mortality. No association was apparent for the CAR.

The difficulties in assessing HPA function have meant that longitudinal studies of the association of diurnal cortisol patterning with disease or mortality are largely absent in large-scale studies (2). Our findings are in agreement with a recent report from the Longitudinal Aging Study Amsterdam that evening levels of salivary cortisol are predictive of subsequent mortality in women (7) and with Sephton et al. (10) who described the association of flat patterns of cortisol with mortality in a female patient population. Our data suggest that these associations are independent of covariates examined, including smoking, acute stress on day of sampling, chronic stress associated with social position, and perceived financial insecurity, and cardiovascular medications, although we did not examine the combined effect of these factors in detail due to loss in numbers. Our findings are novel in that they suggest a cause-specific association with cardiovascular mortality in a nonclinical population. The pathway by which this association occurs remains unclear, but further follow-up of our participants with cumulated numbers of deaths will allow examination of these issues in greater detail.

Late-night salivary cortisol levels have recently been suggested for the diagnosis of Cushing's syndrome, a rare condition characterized by raised cortisol levels (22). In our analyses, we removed those with very high cortisol levels (3 sd from the mean that equated to values above 56 nmol/liter for samples 1, 3, 4, 5, and 6 and values above 75 nmol/liter for the CAR) from the analyses that would serve to remove those with Cushing's syndrome, which is associated with an increased risk of mortality (23). We observe a linear association of mortality events with cortisol levels. Furthermore, we do not observe an association of cortisol with hypertension in our study (14), and our findings were independent of hypertension, obesity, lipids, and fasting glucose, all of which are associated with Cushing's syndrome. Studies suggest an association of cortisol with the metabolic syndrome (24). We observe an association of cortisol patterns with fasting glucose (data not shown); however, this does not explain the association of the decline in cortisol across the day with cardiovascular-related deaths.

The failure of waking cortisol to predict deaths argues against a role for the HPA axis in mediating the association of fatigue with mortality. Previous studies suggest a role of vital exhaustion, a concept that overlaps with depression, which is associated with both cortisol (25) and cardiovascular events (26). We have previously reported an association of lower cortisol on waking with fatigue (16), and here we see that whereas fatigue is independently predictive of cardiovascular- and non-cardiovascular-related deaths, measures of cortisol patterns do not explain these associations that remain independent of fatigue.

The causes of flat slopes in cortisol or raised evening levels are unknown. It is unclear whether a flatter slope in cortisol is due to stress-related elevations, resulting from a stressful day, long-term changes in circadian regulation as a result of chronic stress, or impaired central negative feedback sensitivity of the HPA axis (particularly to melanocorticoid receptors) as recently described in obesity (27). Findings from our study are consistent with contributions from both of these mechanisms (14). More specifically, here we see that a flattened slope in cortisol secretion is a risk factor for cardiovascular mortality independent of the indicators of acute and chronic stress. Slopes in cortisol across the day are not significantly associated with depressive symptoms at phase 7 of our study, but we do see an association with short sleep and financial insecurity (13). However, these factors were not significantly associated with cardiovascular events in these analyses, and the effect of diurnal slope on mortality was independent of these factors.

Interestingly, we found that participants' report of events considered very stressful on the day of sampling was predictive of cardiovascular death. The mechanism by which this occurs is unclear, and we will examine this further with longer follow-up of the cohort.

The main strength of this study is that we have assessed cortisol levels repeatedly across the day in a large group of participants, who appeared to understand the study protocol and instructions and took the samples correctly. However, despite a follow-up of over 6 yr, we have few cardiovascular-related mortality events, which currently precludes us from examining a wider variety of covariates without biasing regression estimates (28). The Whitehall II study is an occupational cohort of civil servants and is therefore subject to the healthy worker effect. However, although not representative of the general United Kingdom population, it is likely that the bias in the association between mortality and cortisol patterns is limited. Because our findings use a relatively short follow-up, we cannot at this point discount reverse causation by removing deaths close to cortisol measurements from the analysis (29). Further research is therefore needed to examine whether the observed association is due to flat slopes in cortisol being a consequence of the early stages of cardiovascular disease or of occult disease rather than representing a risk factor. Cortisol measures were made on a single day only, which may have obscured the CAR to situational rather than chronic correlates (30) and may therefore contribute to the failure of the CAR to predict future events. Furthermore, cortisol patterns are measured only in the day, making it impossible to evaluate total 24-h cortisol exposure. However, our data do accord with a recent study that urinary cortisol output is associated with mortality in a very similar way to that reported here (9).

Because this is an observational study, it is also important to consider confounding as a potential explanation for our result. For example, it has been posited that low BMI reflects poor subclinical health, or health behaviors (31), which we have shown to be associated with flatter slopes in cortisol patterns (17). However, there are no mortality events in participants with low BMI, suggesting that BMI-associated occult disease is not predictive of events in this cohort. Furthermore, the association between diurnal cortisol patterns and mortality was seen after adjustment for smoking and exclusion of smokers. These findings suggest that confounding by low BMI or smoking is an unlikely explanation for our results. We cannot discount residual confounding in our analyses, although adjustment for any confounder failed to alter our HR to any great extent.

In conclusion, these novel findings from a nonclinical population suggest that flat slopes in salivary cortisol, particularly raised evening levels of cortisol, are a robust predictor of cardiovascular mortality in middle-aged adults. These associations are specific for cardiovascular rather than noncardiovascular mortality, providing evidence of a plausible pathway by which social and stressful environments are related to cardiovascular disease.