Impact of stressors in a natural context on release of cortisol in healthy adult humans: A meta-analysis

Abstract

Increased hypothalamic–pituitary–adrenal (HPA) activation, culminating in elevated circulating cortisol levels is a fundamental response to stressors. In animals, this neuroendocrine change is highly reliable and marked (∼5–10-fold elevations), whereas in humans, the increase of cortisol release is less pronounced, and even some potent life-threatening events (anticipation of surgery) only elicit modest cortisol increases. Meta-analysis of factors that influenced the increase of cortisol release in a laboratory context pointed to the importance of social evaluative threats and stressor controllability in accounting for the cortisol rise. The present meta-analysis, covering the period from 1978 through March 2007, was undertaken to identify the factors most closely aligned with cortisol increases in natural settings. It appeared that stressor chronicity was fundamental in predicting cortisol changes; however, this variable is often confounded by the stressor type, the stressor's controllability, as well as contextual factors, making it difficult to disentangle their relative contributions to the cortisol response. Moreover, several experiential factors (e.g. previous stressor experiences) may influence the cortisol response to ongoing stressors, but these are not readily deduced through a meta-analysis. Nevertheless, there are ample data suggesting that stressful events, through their actions on cortisol levels and reactivity, may influence psychological and physical pathology.

• Cortisol
• corticosterone
• stress
• naturalistic
• ACTH

Introduction

At one time or another, most individuals encounter stressful events that challenge their psychological and/or physiological well-being. Obviously, not all individuals who encounter stressors respond in a uniform fashion nor do all stressors elicit comparable outcomes. Ordinarily, stressors will give rise to several behavioral, cognitive and emotional changes, some of which are aimed at contending with the challenge. Concurrently, a concatenation of brain neurochemical and hormonal changes are provoked, whose function is one of enhancing the effects of other neuroendocrine processes, preparing the organism to cope with the insult, and blunting the physiological and psychological impact of the stressor (de Kloet et al. 1999; McEwen 2000a,b; Sapolsky et al. 2000). Although stressor-induced neurochemical alterations may be adaptive, it has been suggested that when the stressor is sufficiently protracted, the strain on endogenous systems might become excessive (allostatic overload), thereby increasing vulnerability to disturbances, such as depression (McEwen 2003; McEwen and Wingfield 2003).

Of the many physiological changes that emerge in response to stressors, one of the most fundamental, and most frequently studied, is the increase of cortisol concentrations in blood, saliva or urine that accompanies stressful experiences. Increased secretion of cortisol by the adrenal cortex is thought to be a fundamental feature of the stress response with multiple beneficial effects. However, in humans, this hormone is not elicited under all stressor conditions, and even some fairly strong stressors elicit only moderate cortisol variations (Biondi and Picardi 1999). The goal of the present meta-analysis was to define the contribution of several factors in determining cortisol release in humans in response to naturalistic stressors. In this regard, it was of particular interest to establish the contribution of variables such as stressor controllability, predictability and chronicity, as well as social evaluative threats in promoting these cortisol changes. These factors are known to affect stress responses in animals (in the case of the former three variables), and social evaluative threat has been implicated as being particularly cogent in promoting cortisol changes in a laboratory context (Dickerson and Kemeny 2004). Thus, it was hypothesized that such factors might also be pertinent in moderating the cortisol response in humans in real world settings.

Physiological responses to stressors

In response to stressful experiences, certain behaviors, notably those associated with arousal, vigilance and coping processes predominate, whereas behaviors that are not productive in a defensive capacity (e.g. sexual and feeding behaviors) ought to be suppressed, although in humans, there are instances where eating, especially carbohydrates, may be a response to stressors (Levine and Marcus 1997; Dallman et al. 2003; Peters et al. 2007). Concurrently, neuronal functioning is increased within several stressor-sensitive brain regions. In this regard, studies in animals indicated that neuropeptide and monoamine changes are evident across multiple extrahypothalamic regions (including various amygdala nuclei, medial prefrontal cortex, locus coeruleus) and in hypothalamic nuclei, presumably to facilitate the ability to deal effectively with the ongoing challenge (Munck and Naray-Fejes-Toth 1994; de Kloet et al. 1999; McEwen 2000; Sapolsky et al. 2000).

To a considerable extent, organismic variables (age, sex, genetics) moderate the neurochemical changes elicited by stressors, as do experiential variables (previous stressful experiences, maternal factors) (Anisman and Matheson 2005). As well, some neurochemical stress responses (e.g. monoamine turnover in limbic regions) are influenced by factors such as stressor controllability (Weiss et al. 1981; Petty and Sherman 1982; Heinsbroek et al. 1989; Anisman et al. 1991; Bolanos-Jimenez et al. 1995; Nankai et al. 1995), chronicity, ambiguity and predictability (Matthews et al. 1980; Osuna 1985; Baker and Stephenson 2000) and it seems that different types of stressors may be selective in activating particular neural circuits (Anisman and Merali 1999). Whether this holds true in humans is uncertain, although some types of stressors (e.g. anticipation of adverse events) are particularly effective in promoting anxiety (Paykel 1982; Reno and Halaris 1990), whereas others (e.g. loss, social conflict) are more aligned with depression (Roy 1983, 1985; Brown et al. 1987; Monroe and Depue 1991). Moreover, these particular effects may vary with gender (Harris 2001; Kendler et al. 2001; Mazure and Maciejewski 2003).

Stressor provoked variations of hypothalamic–pituitary–adrenal (HPA) activity

Psychological (psychogenic) or physical (neurogenic) stressors (both classed as “processive” stressors as they involve appraisal of the stimulus and the context in which this stimulus is presented), as well as systemic insults (e.g. immune activation associated with infection), stimulate hypothalamic–pituitary–adrenal (HPA) functioning, although they may do so through different neural circuits (Herman and Cullinan 1997; Yokoyama and Sasaki 1999; Sawchenko et al. 2000). By example, animal studies revealed that neurogenic stressors, innate psychogenic insults and learned psychogenic stressors may differentially influence neuropeptide processes (Merali et al. 2004). It is equally possible that, in humans, different types of stressors might not engage all of the same processes, and hence certain stressors may be more likely to influence HPA functioning.

Ordinarily, when a processive stressor is encountered, various brain regions may be activated. Some regions may be involved in the development or elicitation of fear and/or anxiety (e.g. central and medial amygdala and bed nucleus of the stria terminalis) (Lee and Davis 1997; LeDoux 2000), whereas others may be more important in the appraisal of the stressor or in executive functioning (e.g. medial and orbital prefrontal cortex) (Fuster 1989, 1995) or the learning/memory of fear/anxiety (Nader et al. 2000). After emotional meaning is assigned to the sensory information, the amygdala guides emotional behavior, likely through projections to the hypothalamus, hippocampus and prefrontal cortex (LeDoux 1986; Fellous 1999; Vertes 2006). Ultimately, the paraventricular nucleus (PVN) of the hypothalamus is activated, giving rise to the release of corticotropin-releasing hormone (CRH) from terminals located at the median eminence, thus promoting the release of adrenocorticotropic hormone (ACTH) from the anterior pituitary, which then stimulates cortisol release (or corticosterone in rodents) from the adrenal cortex.

As prolonged cortisol increases may increase vulnerability to immunosuppression, and to autoimmune-related and metabolic disorders, it is essential that an individual be able to terminate cortisol release appropriately (Munck et al. 1984; Sapolsky et al. 2000). Once released into the circulation, cortisol activates corticoid receptors in the hippocampus (Jacobson and Sapolsky 1991; Herman 1993; McEwen 2000), or may directly impact the hypothalamus to attenuate further neuroendocrine activity. In the latter instance, cortisol action on the PVN down-regulates CRH activity, and attenuates ACTH secretion (Uht et al. 1988; Whitnall 1993). In addition, given that stressors activate several limbic sites, it is not surprising that brain regions involved in emotional responses or appraisal processes also contribute to HPA regulation (Viau and Meaney 1991; Buijs et al. 1993; Diorio et al. 1993; Larsen et al. 1994; Suemaru et al. 1995; Herman and Cullinan 1997). Interestingly, unlike the suppression of hypothalamic CRH release elicited by cortisol (stemming from negative feedback), in limbic regions, particularly the central amygdala and bed nucleus of the stria terminalis, cortisol acts to increase CRH release, as well as GABA_A-mediated neuronal functioning, which may be fundamental in the promotion of various anxiety states (Schulkin et al. 1998, 2006; Duvarci and Pare 2007).

Glucocorticoid functions

Glucocorticoids (GC) have multiple actions that facilitate the ability to deal with stressors; GCs influence glucose metabolism, lipolysis, alterations in regional blood flow and may prevent overshoot of immune reactions (Sapolsky et al. 2000). Moreover, their function in dealing with adverse events can take several forms; they may act in a permissive (exert an effect prior to stressor and prime defence mechanisms), suppressive (arise from stress-activated defence reactions and prevent them from overshooting), stimulating (enhance the effects of the first wave of hormonal responses to stress) and finally preparatory capacity (do not affect the immediate response to a stressor but modulate the organism's response to a subsequent stressor (Munck et al. 1984; Sapolsky et al. 2000). Despite the adaptive and highly beneficial effects associated with cortisol release, as indicated earlier, sustained activation of these processes might culminate in excessive wear and tear on biological systems (allostatic overload), and hence might increase vulnerability to various pathological outcomes (McEwen 2000). Indeed, protracted increases of cortisol levels have been associated with depression and cognitive impairments (Lupien et al. 2005) as well as hippocampal cell loss (McEwen 2000).

Corticosterone (or cortisol, the primary GC in humans) has been taken to be the prototypical stress hormone, and it has frequently been assumed (incorrectly so) that GC changes provide an index of distress (witness for instance that cortisol levels may be reduced among those with post-traumatic stress disorder, PTSD; Yehuda 2002). In studies conducted in humans, at least within a laboratory context (e.g. using the Trier Social Stress Test, TSST), cortisol elevations as high as 2–4-fold (much lower than the ∼10-fold increase seen in rodents) have been reported, but changes of this magnitude are infrequent (Biondi and Picardi 1999). The factors governing the magnitude of cortisol changes elicited by stressors within a laboratory situation remain to be fully elucidated, although in their recent meta-analysis of laboratory-based studies, Dickerson and Kemeny (2004) indicated that cortisol variations are most pronounced in tests that involve uncontrollable social-evaluative threat (i.e. assessment by others).

The analysis of endocrine reactivity to psychological stress within a laboratory setting has been exceptionally valuable as it is permitted standardization of the stressor and control of potential confounding factors. Yet, little is known about the generalizability of these studies to actual life circumstances. However, field studies have been conducted to determine those conditions that favor cortisol changes that occur in response to day-to-day stressors or those of a more severe or chronic nature. These studies have included analyses of cortisol changes associated with daily hassles, events such as academic and occupational stressors, athletic stressors and various social stressors. Some naturalistic stressors (e.g. surgery) reliably increased HPA activity, but cortisol fluctuations in response to other stressful events have been less consistent, typically increasing from 25 to 100%, or have not been detected at all (Biondi and Picardi 1999). Of course, meaningfulness of these variations is difficult to compare across studies, as the stressors were qualitatively different from one another (e.g. differing in their relative severity/emotional impact, controllability, predictability, chronicity), as were the fluids in which cortisol was determined (i.e. blood, saliva, urine). Moreover, as already indicated, several organismic (genetic, age, gender) and experiential variables may also contribute in this regard.

HPA diurnal rhythm associated with stressful experiences

Ordinarily, cortisol release follows a well-defined diurnal rhythm, and is secreted from the adrenal cortex in a pulsatile manner (Deuschle et al. 1987; Mershon et al. 1992; Windle et al. 1998). In humans, cortisol release, already high at awakening, increases to reach a morning peak during the ensuing 30–60 min, and declines precipitously thereafter, reaching an evening nadir at about midnight (Linkowski et al. 1993; Schmidt-Reinwald et al. 1999). Although, it has been suggested that the increase of free cortisol (∼50–60%) within the first 30 min after awakening is largely independent of the time of awakening, sleep duration, sleep quality, physical activity or morning routines (Spath-Schwalbe et al. 1992; Wust et al. 2000; Wilhem et al. 2007), there have been several reports indicating that time of awakening may, indeed, influence the course of the morning cortisol rise (Edwards et al. 2001; Kudielka and Kirschbaum 2003; Kudielka et al. 2006; Federenko et al. 2004). Nevertheless, it does appear that the magnitude of the increase is related to general levels of distress (Pruessner et al. 1999; Schmidt-Reinwald et al. 1999) and may also be predictive of the response to other challenges, including CRH challenge and the TSST (Schmidt-Reinwald et al. 1999).

It has been suggested that the factors that moderate the morning cortisol rise have not been sufficiently examined, and it is unclear whether this increase is functionally significant (Clow et al. 2004). Indeed, although increasing stressful experiences have been associated with elevated morning cortisol release, relatively severe trauma has been associated with a blunting of the morning cortisol response (e.g. in patients affected by PTSD; Yehuda 2002). This may be coupled with an increase of the afternoon cortisol levels, so that the diurnal profile of cortisol release is flattened (Caplan et al. 1979; Hart et al. 1996; Adam and Gunnar 2001). Furthermore, there is evidence suggesting that chronic stressful experiences may be associated with diminished cortisol levels (Pruessner et al. 1999), although the case for such an outcome is not unequivocal (Melamed et al. 1999; Steptoe et al. 2000; Grossi et al. 2001), and defining what constitutes a chronic stressor has not been satisfactorily addressed. This should not be taken to imply that stressors do not favor elevated morning cortisol secretion, but that such an outcome is subject to the moderating influence of other factors, including a backdrop of traumatic experiences that provoke PTSD. In their recent meta-analysis regarding the influence of chronic stressors on HPA functioning, Miller et al. (2007) indicated that much of the variability that has been reported concerning cortisol changes elicited by chronic stressors could be attributed to features of the stressor, individual factors and the time since the stressor was experienced. It was suggested that those stressors that involve trauma, are uncontrollable, and threaten personal integrity tend to produce a high, flat diurnal profile, although the degree of morning cortisol release tends to be somewhat lower than that associated with stressors that do not promote PTSD.

The present analysis

There clearly exists an array of variables that could account for the individual differences of cortisol reactivity that typically occur in response to stressful experiences. Factors such as stressor predictability, controllability (with respect to stressor onset, stressor termination, and the consequences of stressor experiences) and chronicity have long been thought to play a pivotal role in behavioral disturbances and central neurochemical changes (reviewed in Anisman and Matheson 2005). Likewise, as already indicated, in their meta-analysis, Dickerson and Kemeny (2004) indicated that within a laboratory context stressor controllability and social evaluative threats were fundamental in determining stressor-provoked cortisol increases. However, a comparative analysis is unavailable concerning the influence of different types of stressors on the cortisol response under naturalistic settings, or whether common denominators exist that moderate this stress response. In the present analysis, it was of interest to establish whether evaluative stressors (e.g. examinations, oral presentation, athletic competition), as well as stressor controllability, predictability and chronicity are fundamental in determining cortisol increases under naturalistic conditions. These variables were selected largely on the basis of animal studies that have identified them to be particularly important in determining behavioral and neurochemical changes associated with stressors (Anisman and Matheson 2005) as well as the meta-analysis reported by Dickerson and Kemeny (2004) concerning the factors that influence cortisol output in a laboratory context.

Methods

Literature search

Articles for consideration were identified through a computer-based search using Pubmed and PsycInfo databases from 1978 through March 2007. The key words were cortisol, HPA, neuroendocrine, hydrocortisone, psychoneuroimmunology, psychoimmunology, psychoneuroendocrinology and psychoendocrinology with naturalistic terms such as stressor, natural stress, psychological stress, as well as more specific terms such as caregiving, academic examination, competition and surgery. Additional articles were also obtained from the reference lists of these articles and several reviews (Kirschbaum and Hellhammer 1994; Biondi and Picardi 1999; Kiecolt-Glaser et al. 2002; Miller and O'Callaghan 2002; de Kloet 2003; Burke et al. 2005).

Study inclusion criteria

Articles were included only if they met the following criteria: (1) used naturalistic stressors, defined as stressful events or situations likely to be encountered by many or most individuals in the context of their everyday lives (e.g. academic examination, job strain, marital conflict, work-related noise exposure, anticipation of a competition or medical procedure) as well as stressful events of a chronic nature (e.g. caregiving). Inasmuch as severe stressors that might promote PTSD (e.g. rape, war) may involve processes distinct from the more common stressors, these were not included in the present analysis, but were recently reviewed by Miller et al. (2007); (2) the sample involved healthy adult participants. This criterion allowed control for the effect of confounding variables with cortisol levels, namely age and the presence of physical and psychological pathologies that may also influence cortisol reactivity. Thus, studies involving severe illness or those maladies that require drug treatments that directly or indirectly affect HPA functioning (e.g. cancer, heart disease, diabetes, autoimmune disorders, depression and so forth), were excluded, as were those that examined stress reactivity in children and/or adolescents (i.e. where the mean age was under 18 years or the age range included participants under 18 years old); (3) reported or provided data from which an effect size was provided in the text or could be extracted or extrapolated by inferential statistics. Studies involving the prediction of cortisol levels from various personality characteristics were excluded given that it was impossible to determine from these data whether the stressor itself evoked significant changes of cortisol levels.

Coding

Demographic characteristics of participants as well as several features of the methodology and the nature of the stressor itself were coded. Reliabilities of the coding schema were calculated with the intraclass correlation (r1) for the continuous variables and kappa (κ) for categorical variables (Orwin 1994) by randomly selecting 15% of the studies to be coded by a second independent, trained judge. As will be seen in ensuing sections, the interclass correlations were relatively high, attesting to the inter-rater reliability with respect to the evaluation of the characteristics of the stressors.

Participant characteristics

Participant characteristics included (a) sample size (r1 = 1.00), (b) mean age of participants (r1 = 1.00), (c) gender composition (coded as percent female, r1 = 1.00) and whether exclusion criteria (e.g. depression, inflammatory diseases, medications used, κ = 1.00) were absent (coded 0) or present and reported in the studies (coded 1).

Methodological characteristics

Various aspects of the methodology of each study were coded. In this regard, studies were coded for the time of day that cortisol was collected (κ = 1.00). As described earlier, cortisol levels follow a diurnal rhythm, wherein an increase in level of the hormone is evident within the first 30–60 min following awakening, and then declines over the course of the day. Furthermore, in a laboratory context, stressors tend to have greater effects in the afternoon (when basal cortisol secretion is relatively low), thus it was deemed important to evaluate studies on the basis of the time of day in which they were conducted. Even though there is variability over the course of the morning and over the course of the afternoon, like in Dickerson and Kemeny (2004), those studies in which the cortisol samples were collected before 12 p.m. were coded as morning studies (AM coded 1), whereas studies in which samples (saliva or plasma) for cortisol determination were collected after 12 p.m. were coded as afternoon studies (PM coded 2). Studies in which samples were collected at several prescribed times of the day were coded as diurnal (coded 3). Those studies in which saliva or plasma sample collection occurred at varied times of day, and were too broad to fit into a circumscribed code (e.g. between 9 a.m. and 4 p.m.) coded as AM/PM (coded 4). Finally, when the study did not report the time of day at which samples for cortisol measurement were collected, it was coded as non-specific (coded 5).

Given that the method by which samples for cortisol measurement were collected may also differentially influence cortisol values, this variable was also coded. Specifically, cortisol can be measured either in saliva (coded 1), blood (coded 2) or urine (coded 3). Not only do these methods vary in their degree of intrusion (e.g. venepuncture is a more intrusive method of collection that may, itself, constitute a distressing factor thereby influencing cortisol values) but they also assess different cortisol fractions. Plasma samples reflect levels of cortisol bound to protein as well as biologically active free cortisol (unbound), whereas urine and salivary samples reflect only the levels of free cortisol (Peters et al. 1982; Kaye and Crapo 1990; Bonnin et al. 1993).

Characteristics of the stressor

A stressor classification was used (κ = 1.00) so that each stressor fell into one of four types: Occupation-based stressors (coded 1; including job strain, burnout, work related noise exposure, unemployment or academic stressors); social stressors (coded 2; including events such as marital conflict, loneliness, interpersonal stress and caregiving); medical procedures (coded 3; including surgery and dental procedures); sports (coded 4). It should be underscored that this does not mean that caregiving is being equated with, say, marital conflict. For the purpose of the present investigation caregiving was considered a chronic stressor, whereas marital conflict was considered to be acute (or subchronic) because it was conducted within an experimental setting.

Additionally, whether or not the stressor was: (1) acute/chronic [κ = 1.00 (acute was coded as 0 and chronic was coded as 1)]; (2) predictable [κ = 0.78 (no was coded as 0, and yes was coded as 1)]; (3) had an uncontrollable or unpredictable outcome [κ = 0.77 (no was coded as 0, and yes was coded as 1) and included and (4) an evaluative component [κ = 0.77 (defined as ‘being judged or evaluated’ with no being coded as 0, and yes coded as 1)] was also coded.

Statistical analysis

Effects sizes: Meta-analytical technique

For cortisol analyses, we calculated the effect size of the difference between conditions (e.g. cortisol concentrations at baseline compared to those following a stressor) or between groups (e.g. cortisol concentrations of those individuals not suffering from burnout compared to those of participants suffering from burnout) using the standardized-mean change statistic, Cohen's d, which is appropriate for repeated measures effect size estimates (Becker 1988; Dunlap et al. 1996; Morris 2000). This statistic can be interpreted as the magnitude of the difference between pre- and post-stressor cortisol values in standard deviation units [d = (Mean post stressor level − Mean pre stressor level)/pooled SD] and were further corrected for sample size bias (Hedges and Olkin 1985). The direction of the effect size was positive if the study reported an increase of cortisol levels from pre- to post-stressor. Cohen classifies an effect size (d) of 0.20 as small, 0.50 as moderate, and 0.80 as large (Cohen 1977).

To the extent that they could be, effect sizes were calculated using the means and standard deviations provided. However, for studies in which these statistics could not be found in the article or directly obtained from the author, inferential statistics were used (Hedges and Olkin 1985; Rosenthal 1991). Specifically, for the between-subject designs, the effect size r was first computed from t or F statistics with 1 df in the numerator using the formulae provided by Rosenthal and DiMatteo (2001) and then transformed into Cohen's d, [r = (t²/t² + df)^1/2 or r = (F/F + df_error)^1/2; d = (4r²/(1 − r²)^1/2. For within-subjects designs, the formula d = tc{2/[n(1 − r²)]}^1/2 or d = (F)^1/2{2/[n(1 − r²)]}^1/2 was used in order to control for possible effect size overestimation (Dunlap et al. 1996). F and tc correspond to the values reported in the text comparing the pre- and post stressor conditions. Frequently, the value of r was not provided in the text. Therefore, a value of 0.40 was use to calculate the effect sizes. Additional analyses indicated that r values from 0.20 to 0.60 did not alter the results of the analyses.

For those studies that only reported significant results, we assumed p < 0.05 and this value was used to calculate the effect size, whereas those studies reporting “no significant effect” were given an assumed effect size d = 0.00 (Rosenthal 1991). Such inferences were made for 40 studies; importantly, omitting these studies did not alter the results of the analyses. Even if diurnal studies comprised several cortisol samples, separate ds were calculated for each stressor-related cortisol assessment relative to a control group not exposed to the naturalistic stressor in question. In order to maximize the condition of independence in the meta-analysis, when saliva samples were collected on multiple days from the same participants (e.g. job strain among bus drivers; Aronsson and Rissler 1998), only effect sizes from the first day of the study were calculated and used in the analyses.

The hypotheses were tested through regression analyses, comparisons of means (t-tests to determine that effect sizes differed from 0), analyses of variance (ANOVAs) to compare stressor types and characteristics. Effect sizes served as the dependent variable in all the analyses, whereas methodological factors and participant characteristics served as predictor variables in the regression analyses.

Results

Sample of studies selected

A total of 140 studies met the inclusion criteria for the present analysis resulting in 181 effects sizes. In total, 10,976 participants (mean age = 38.3 years old, SD = 14.0) of whom 47% were females (range = 0–100%, SD = 0.37), contributed to this study. Saliva collection was the most frequent method of cortisol assessment (n = 112; 61.9%), followed by plasma measurements (n = 47; 26.5%) and urine measurements (n = 19; 10.5%). The majority of studies were conducted in the morning (AM; n = 80; 44.2%), whereas some were conducted in the afternoon (PM; n = 35; 19.3%), throughout the day (Diurnal; n = 19; 10.5%), at varied times throughout the day (AM/PM; n = 16; 8.8%) or at unspecified times (n = 31; 17.1%).

Methodological factors: Time of day and method of cortisol collection

As alluded to earlier, given that the time of day that samples were collected may influence the observed fluctuations, coupled with the fact that the method of collection (saliva, blood, urine) influences different cortisol fractions, the effect sizes were first regressed separately onto dummy coded variables representing the four time periods (i.e. time of day was dummy coded using three variables and type of cortisol collection dummy coded using two variables).

Time of day did not influence the effect size associated with the stressor experience, R² = 0.001, F(3,143) < 1, ns, even though this variable explained 5.0% of the between-study variance. Furthermore, the method of sampling for cortisol measurement was not a significant predictor of the effect size, R² = 0.012, F(2,176) = 1.05, ns.

Participant characteristics

A regression of the effect sizes for stressor-induced cortisol variations as a function of participant characteristics (number of participants, mean age and gender composition) was also conducted. This regression analysis was not significant, R² = 0.035, F(3,128) = 1.57, ns, indicating that neither the number of participants, the gender composition, nor the mean age of the sample were significant predictors of the effects sizes. Given that none of the methodological factors (type of sample for cortisol collection, time of day) or participant characteristics qualified as significant predictors of the effect sizes, these variables were not considered in subsequent analyses.

Types of stressors used

The majority of the naturalistic studies involved occupational stressors (n = 112; 61.9%), whereas the remaining studies comprised social stressors (n = 33; 18.2%), medical procedures (n = 13; 7.2%), and sports-related stressors (n = 23; 12.7%). Table I provides a summary of the studies used in the present analysis as a function of the type of stressor used, the characteristics of the stressor, and the time of day at which the study was conducted. The average effect size and the primary coded dimensions for each study are provided in Table II.

Table I.  Characteristics of the studies.

			Methodological characteristics		Stressor characteristics
Type of stressor task	No. of effect sizes	Subjects	Time of day	Collection	Acute	Unpredictable	Uncontrollable	Evaluative
Occupational	112	7744	47 a.m.	65 saliva	30	22	58	29
			17 p.m.	28 blood
			16 a.m./p.m.	17 urine
			13 diurnal	2 ns
			19 ns
Social	33	2304	15 a.m.	24 saliva	3	30	32	31
			11 p.m.	8 blood
			6 diurnal	1 urine
			1 ns
Medical procedures	13	574	7 a.m.	5 saliva	13	10	13	0
			2 p.m.	7 blood	21	0	0	21
			4 ns	1 urine
Sports	23	354	11 a.m.	18 saliva
			5 p.m.	4 blood
			7 ns	1 ns
Total	181	10,976

Table II.  Methodological characteristics and average effect sizes for the studies included in the meta-analysis.

	Author/date	N	Gender*	Age	Time^†	Collection^‡	d^¶
Occupational
Academic Examination	Frankenhaeuser et al. (1978)	30	0.61	19	a.m.	Urine	1.06
	Johansson et al. (1983)	2	0.50	31	Diurnal	Urine	0.27
	Kirkeby et al. (1984)	9	1.00	26	–	Blood	0.17
	Hellhammer et al. (1985)	9	0.00	–	–	Saliva	2.14
	Lovallo et al. (1986)	28	0.00	23	a.m.	Blood	1.12
	Meyerhoff et al. (1988)	11	0.00	25	a.m.	Blood	2.33
	Herbert et al. (1986)	38	0.00	–	a.m.	Blood	1.97
	Semple et al. (1988)	9	–	21	p.m.	Blood	0.69
	Johansson et al. (1989)	27	1.00	27	a.m.	Blood	0.05
	Evans et al. (1994)	7	–	–	Diurnal	Saliva	0.87
	Glaser et al. (1994)	45	0.00	–	p.m.	Blood	0.00
	Malarkey et al. (1995)	55	0.00	22	Diurnal	Blood	0.00
	Guidi et al. (1999)	28	0.29	–	a.m.	Blood	1.19
	Song et al. (1999)	38	0.66	–	p.m.	Saliva	0.06
	Lacey et al. (2000)	15	0.50	27	a.m./p.m.	Blood	2.38
	Vedhara et al. (2000)	60	0.40	22	Diurnal	Saliva	0.20
	Ennis et al. (2001)	36	0.76	24	a.m.	Urine	0.86
	Ng et al. (2003a)	11	0.18	33	a.m.	Saliva	0.94
	Ng et al. (2003b)	31	0.32	23	a.m.	Saliva	− 4.52
	Al-Ayadhi et al. (2005)	48	1.00	20	a.m.	Blood	1.78
	Droogleever Fortuyn et al. (2004)	15	0.20	35	p.m.	Blood	0.19
	Gaab et al. (2006)	15	0.40	24	a.m.	Saliva	1.39
	Weekes et al. (2006)	67	0.51	20	p.m.	Saliva	0.28
Driving examination	Dugue et al. (2001)	7	0.00	18	–	Blood	0.21
Job Strain	Caplan et al. (1979)	85	0.00	40	Diurnal	Urine	0.49
	Rose et al. (1982)	195	0.00	–	–	Blood	0.22
	Harenstam and Theorell (1990)	1034	–	–	a.m.	Blood	0.00
	Schreinicke et al. (1990)	77	0.00	43	a.m.	Saliva	0.50
	Zeier (1994)	22	0.03	–	a.m.	Saliva	0.57
Job Strain (continued)	Zeier et al. (1996)	16	0.00	42	a.m.	Saliva	0.61
	Burton et al. (1996)	115	0.57	–	p.m.	Saliva	0.00
	Fujigaki and Mori (1997)	10	0.00	28	p.m.	Saliva	0.00
	Aronsson and Rissler (1998)	10	0.50	30	a.m.	Urine	1.10
	Schulz et al. (1998)	12	0.51	26	a.m./p.m.	Saliva	0.00 – 0.61
	Sluiter et al. (1998)	10	0.00	47	a.m.	Urine	0.18 – 1.15
	Steptoe et al. (1998)	61	0.62	35	–	Saliva	− 0.55
	Fischer et al. (2000)	139	0.81	–	Diurnal	Saliva	0.27
	Hanson et al. (2000)	77	0.44	40	a.m.	Saliva	0.00
	Rissen et al. (2000)	31	1.00	44	–	Saliva	0.15
	Sluiter et al. (2000)	115	0.00	37	–	Urine	0.00
	Steptoe et al. (2000)	105	0.61	39	a.m.	Saliva	0.50
	Ganster et al. (2001)	198	1.00	–	–	Saliva	0.95
	Sluiter et al. (2001)	55	0.00	40	Diurnal	Urine	0.41
	Ohlson et al. (2001)	103	0.92	–	a.m.	Blood	0.00
	Yang et al. (2001)	73	–	30	a.m./p.m.	Saliva	0.75
	Lundberg and Hellstrom (2002)	82	–	–	a.m.	Urine	0.69
	Lundberg and Frankenhaeuser (1999)	60	0.50	44	a.m./p.m.	Urine	0.05 – 0.51
	Hansen et al. (2003)	101	1.00	41	a.m.	Blood	3.43
	Sluiter et al. (2003)	20	0.00	42	–	Saliva	0.19
	Weibel et al. (2003)	8	0.88	43	Diurnal	Saliva	0.50
	Fujiwara et al. (2004)	8	1.00	28	a.m./p.m.	Saliva	− 0.13
	Kunz-Ebrecht et al. (2004a)	124	0.46	52	a.m.	Saliva	0.54
	Kunz-Ebrecht et al. (2004b)	75	0.49	–	Diurnal	Saliva	0.67
	Otte et al. (2005)	76	0.13	28	–	Saliva	0.59
	Schlotz et al. (2004)	214	0.53	49	a.m.	Saliva	0.04
	Steptoe et al. (2004)	91	0.47	–	–	Saliva	0.59
	Dahlgren et al. (2005)	25	0.63	47	Diurnal	Saliva	1.68
	Ritvanen et al. (2006)
	Low vs. high stress (young)	14	1.00	31	a.m.	Blood	0.19
	Low vs. high stress (old)	14	1.00	54	a.m.	Blood	− 0.05
	Smith et al. (2006)	32			a.m.	Blood	4.00
Commuting	Evans and Wener (2006)	208	0.48		a.m.	Saliva	0.38
Burnout	Melamed et al. (1999)	89	0.04	43	a.m./p.m.	Saliva	0.81
	Pruessner et al. (1999)	66	0.64	44	a.m.	Saliva	− 0.95
	de Vente et al. (2003)	46	0.36	42	a.m.	Saliva	0.13 - 0.59
	Moch et al. (2003)	16	1.00	38	a.m.	Blood	− 2.30
	Grossi et al. (2003)	63	–	48	a.m.	Blood	0.00
	Grossi et al. (2005)	42	0.52	40	a.m.	Saliva	0.27
	Mommersteeg et al. (2006a)
	Control vs. burnout (awakening)	94	0.55	43	a.m.	Saliva	− 0.05
	Control vs. burnout (15 min after	94	0.55	43	a.m.	Saliva	0.14
	Awakening
	Mommersteeg et al. (2006b)	33	0.68	45	a.m.	Saliva	1.20
	Mommersteeg et al. (2006c)	109	0.28	43.9	a.m.	Saliva	0.07
	Langelaan et al. (2006)	88	0.00	45	a.m.	Saliva	0.00
Unemployment/socioeconomic status	Kapuku et al. (2002)	24	0.00	19	–	Blood	− 1.00
	Brenner and Levi (1987)	110	1.00	–	–	–	0.24
	Arnetz et al. (1991)	227	0.75	35	–	–	0.00
	Ockenfels et al. (1995)	60	–	–	am,pm	Saliva	0.25-0.50
	Toivanen et al. (1996)	162	–	–	a.m.	Blood	0.27
	Grossi et al. (1998)	11	0.37	42	a.m.	Saliva	0.27
	Decker (2000)	31	0.00	–	a.m./p.m.	Saliva	0.00
	Steptoe et al. (2005)	160	0.45	55	am,pm	Saliva	− 0.44–0.07
	Wright and Steptoe (2005)	139	0.49	71	a.m.	Saliva	0.07
	Maier et al. (2006)	71	0.40	42	p.m.	Blood	0.38
	Steptoe et al. (2003)
	Low vs. high grade (men)	73	0	53	Diurnal	Saliva	0.51
	Low vs. high grade (female)	63	1	52	Diurnal	Saliva	− 0.59
Noise chronic	Melamed and Bruhis (1996)	35	–	–	–	Saliva	0.27
	Evans and Johnson (2000)	38	1.00	37	–	Saliva	0.02
Public lecture	Basset et al. (1985)	29	0.24	29	Diurnal	Urine	0.19
	Bolm-Audorff et al. (1986)	10	–	–	a.m./p.m.	Blood	1.42
	Lucas et al. (2006)	11	0.73	38	–	Blood	− 0.18
Medical procedures
Surgery	Manyande et al. (1992)	40	0.45	42	p.m.	Blood	1.47
	Manyande et al. (1995)	48	0.41	45	p.m.	Blood	1.06
	Dahanukar et al. (1996)	23	–	–	–	Blood	0.65
	Smith-Hanrahan (1996)	35	1.00	–	a.m.	Saliva	0.65
	Augustin et al. (1999)	45	0.56	34	a.m.	Saliva	0.27
	Kain et al. (1999)	56	0.50	62	a.m.	Blood	4.89
	Ozarda-Ilcol et al. (2002)	16	–	–	–	Blood	9.58
	Ilcol et al. (2004)	26	1.00	49	–	Blood	1.21
	Pearson et al. (2005)	39	0.31	71	a.m.	Urine	3.15
	Nicholson et al. (2005)	–	0.25	31	–	Blood	− 3.23
Dental procedures	Benjamins et al. (1992)	13	0.46	35	a.m.	Saliva	1.87
	Miller et al. (1995)	196	0.00	35	a.m.	Saliva	2.75
	Brand (1999)	31	0.00	42	a.m.	Saliva	0.46
Severe physical injury	Offner et al. (2002)	22	0.38	35	–	Blood	4.15
Social stressors
Marital conflict	Miller et al. (1999)	–	0.50	31	a.m.	Blood	0.27
	Kiecolt-Glaser et al. (2003)	138	0.50	25	Diurnal	Saliva	0.24
Marital Conflict	Heffner et al. (2004)	85	0.50	26	a.m.	Blood	− 0.27
	Kiecolt-Glaser et al. (1996)	90	0.50	26	a.m.	Blood	0.27
Abuse	Pico-Alfonso et al. (2004)	162	1.00	46	a.m./p.m.	Saliva	0.00 - 0.61
Loneliness	Pressman et al. (2005)
	Low vs. high loneliness (morning)	83	0.55	18	a.m.	Saliva	0.54
	Low vs. high loneliness (afternoon)	“	“	“	p.m.	“	0.53
Interpersonal Stress	Hansen et al. (2006)	393	0.64	44	Diurnal	Saliva	− 0.20
Caregiving	Irwin et al. (1997)	133	0.58	70	a.m.	Blood	− 0.32
	Bauer et al. (2000)	116	0.49	72	Diurnal	Saliva	0.49
	Cacioppo et al. (2000)	27	1.00	67	a.m.	Blood	− 2.82
	Da Roza and Cowen (2001)	58	0.60	69	am,pm	Saliva	0.38 – 2.47
	Davis et al. (2004)	30	1.00	71.4	Diurnal	Saliva	1.31
	Provinciali et al. (2004)	75	0.79	54	–	Blood	0.00
	de Vugt et al. (2005)	112	0.65	60	Diurnal	Saliva	0.03
	Wilcox et al. (2005)	35	1.00	65	a.m.	Saliva	–
Caregiving (continued)
	Gallagher-Thompson et al. (2006a)	22	1.00	51	a.m.	Saliva	0.67 - 1.93
	Vedhara et al. (2002)	103	0.34	43	Diurnal	Saliva	0.92
Bereavement	Jacobs et al. (1987)	56	0.50	62	Diurnal	Urine	3.22
	Gerra et al. (2003)	14	0.57	38	a.m.	Blood	− 0.42
Sports	Urhausen and Kindermann (1987)	8	0.00	25	a.m.	Blood	4.04
	Booth et al. (1989)	6	–	–	–	Saliva	0.00
	Passelergue et al. (1995)	6	0.00	26	p.m.	Saliva	1.47
	Filaire et al. (1996)	31	1.00	19	a.m.	Saliva	0.27
	Kugler et al. (1996)	17	0.00	40	–	Saliva	3.29
	Eubank et al. (1997)	5	0.00	24	p.m.	Saliva	− 0.48–2.41
	Gonzalez-Bono et al. (1999)	16	0.00	23	p.m.	Saliva	0.00
	Salvador et al. (1999)	28	0.00	18	a.m.	Blood	0.00
	Carre et al. (2006)	17	0.00	18	p.m.	Saliva	1.03
	Suay et al. (1999)	26	0.00	18	a.m.	Blood	3.41
	Bonifazi et al. (2000)	8	0.00	–	a.m.	Blood	2.29
	Filaire et al. (2001)	12	0.00	22	–	Saliva	4.13
	Wagner et al. (2002)	8	0.00	–	–	Saliva	0.57
	Bateup et al. (2002)	17	1.00	20	–	Saliva	0.33
	Iellamo et al. (2003)	7	0.43	27	–	Saliva	1.51
	Filaire et al. (2003)	20	0.00	25	a.m.	Saliva	0.00
	Salvador et al. (2003)	17	0.00	–	a.m.	Saliva	3.11
	Kivlighan et al. (2005)	12	0.50	–	–	Saliva	1.30
	Edwards et al. (2006)	40	0.45	20	p.m.	Saliva	0.00

* Denotes % female

^† a.m. = morning; p.m. = afternoon; a.m./p.m. = during the day

^‡ Medium of cortisol collection

^¶ Effect size estimate.

Effect sizes and stressor types

A summary of the effect sizes for all stressor types is provided in Table III. As indicated by t-tests and the 95% confidence intervals involving the various stressor categories, it appeared that average effect size across studies differed from 0 (i.e. the stressors had a significant effect), and this was the case for each of the stressor categories.

Table III.  Effect sizes of different stressor types.

Type of stressor task	No. of effect sizes	Subjects	d	Confidence interval (95%)
Occupational	112	7744	0.33**	0.12–0.56
Social Stressors	33	2304	0.63**	0.20–1.06
Medical Procedures	13	574	1.42*	0.16–2.68
Sports	23	354	1.29***	0.72–1.86
Total	181	10,976	0.58***	0.39–0.77

t-Tests indicated that effect sizes were significantly different from 0 at *p < 0.05; **p < 0.01; ***p < 0.001.

Furthermore, univariate analyses of effect sizes indicated a significant effect of type of stressor, F(3,177) = 6.00, p < 0.001, η² = 0.092. Follow-up comparisons indicated that there were significant differences between medical and occupational stressors (Mean Diff = 1.09, p < 0.05), with medical stressors demonstrating the strongest effects on cortisol levels. As well, sports stressors promoted stronger effects on cortisol levels than did occupational stressors (Mean Diff = 0.96, p < 0.01). No further differences were found between stressor types.

Stressor characteristics

In order to determine which stressor characteristics explained additional variance of effect sizes, a 2 (chronicity) × 2 (predictability) × 2 (controllability) × 2 (evaluative) between subjects ANOVA was conducted. A main effect of chronicity was found, F(1,167) = 8.21, p < 0.01, η² = 0.047, in that acute stressors had greater effect sizes than did chronic stressors (Table IV). Levene's test of homogeneity of variances indicated that error variance across stressor categories were not equal, F(8,167) = 3.49, p < 0.001. Specifically, in order to determine whether the inclusion of medical and sports stressors in the sample biased the results (and indeed, these were shown to have greater effects than occupational stressors), a second analysis was conducted in which these two categories of stressors were excluded [i.e. the analysis only included occupational stressors (i.e. academic stressors, job strain) and social stressors]. A similar 2 (chronicity) × 2 (predictability) × 2 (controllability) × 2 (evaluative) ANOVA was conducted and no significant main effects of stressor characteristics or interactions were observed (Table V). Levene's test of homogeneity of variances indicated that the error variance across stressor categories were not equal, F(7,134) = 2.51, p < 0.05. However, in order to keep the theoretical meaningfulness of the effect sizes, these data were not transformed.

Table IV.  Effect sizes of stressor characteristics across stressor types.

Stressor characteristic	Mean d ± SD
Chronicity**
Acute	0.76 ± 0.40
Chronic	0.22 ± 0.35
Predictability
No	0.42 ± 0.47
Yes	0.60 ± 0.32
Controllability
No	0.57 ± 0.30
Yes	0.36 ± 0.65
Evaluative threat
No	0.68 ± 0.27
Yes	0.21 ± 0.61

**p < 0.01.

Table V.  Effect sizes of stressor characteristics without medical and sports stressors.

Stressor characteristic	Mean d ± SD
Chronicity
Acute	0.53 ± 0.46
Chronic	0.22 ± 0.32
Predictability
No	0.41 ± 0.46
Yes	0.33 ± 0.32
Controllability
No	0.38 ± 0.32
Yes	0.36 ± 0.59
Evaluative threat
No	0.53 ± 0.30
Yes	0.11 ± 0.55

In the analyses there were no other significant interactions involving the different stressor characteristics (i.e. stressor controllability, predictability, chronicity and evaluative component). From this perspective, it seems that these variables did not promote synergistic actions, although this does not imply that they did not have additive effects with respect to cortisol variations.

Discussion

The objective of the present meta-analysis was to identify those factors that were fundamental in determining increased cortisol levels under naturalistic stressor conditions. It is understood, of course, that unlike discrete stressors employed in a laboratory context, stressors experienced in situ typically comprise compound adverse events and stimuli (e.g. medically related stressors may have implications for employment, and stressor experiences are often accompanied by rumination that may exacerbate the adverse experience). As these stressors share a common source, as well as several fundamental characteristics, they might be highly correlated with one another. Not surprisingly, the contribution of different elements of compound stressors often cannot readily be disentangled from one another, making it difficult to conclude whether observed neuroendocrine alterations are specific to a particular element of the stressor.

Factors such as time of day, gender, age and the medium used for cortisol analyses (blood, saliva, urine) have been found to influence hormone levels in human studies that involved laboratory stressors (Seeman and Robbins 1994; Dickerson and Kemeny 2004; Kudielka et al. 2004; Otte et al. 2005a,b; Burke et al. 2005). Although such factors are known to influence basal cortisol levels, they did not appear to influence the effect sizes of cortisol changes elicited by stressors in natural settings.

A variety of different naturalistic stressors influenced cortisol levels (d = 0.61), but the overall effect size was moderate (Cohen 1988). Indeed, the magnitude of the cortisol increase was generally much smaller than that evident in stressed rodents (where 8–10-fold elevations are common), or in response to a laboratory challenge such as TSST, where ∼2–4-fold increases have been reported (Kirschbaum et al. 1992, 1993, 1995, 1996; Biondi and Picardi 1999; Al'Absi et al. 2000; Gerra et al. 2001; Wolf et al. 2001; Dickerson and Kemeny 2004). Indeed, across studies it appeared that in response to naturalistic stressors, cortisol increase ranged from 0 to 180%.

Medical and sports stressors were particularly potent in stimulating HPA activity relative to occupational and social stressors. Given their nature, threat and potential severity, it is not surprising that medical stressors were associated with greater increases of cortisol compared to other type of stressors (Ellis and Humphrey 1982; Kehlet 1984; Douglas and Shaw 1989; Kincey and Satmore 1990). In addition to constituting a source of distress, anticipation and anxiety (Johnston 1986, 1988; Doering et al. 2000; Pearson et al. 2005), medical stressors may involve a pre-existing disease that acts to sensitize reactivity to stressful aspects of medical procedures, particularly surgery. Similarly, as indicated earlier, medical procedures might be associated with other experiential or circumstantial factors that place an additional burden on the individual (e.g. possible financial and personal burden; hospitalization may represent a marked change of environment, sleep cycle, eating habits, and so forth). Of course, it is also necessary to consider individual differences related to surgical history, the nature of the surgery (risk, success rate), and the individual's age (as a risk factor).

The relatively large cortisol increase in association with sports-related events might, at first, be considered surprising. However, the distress associated with such events may involve multiple factors. Among other things, these include concerns regarding the impact of a mediocre performance on an individual's career or status on the team, and the fear of embarrassment or of being condemned following a poor performance. Furthermore, sports-related events often entail a social-evaluative threat (performing in front of an audience) and anticipatory arousal, and thus might have been particularly potent in eliciting cortisol secretion (Dickerson and Kemeny 2004). Moreover, training leading up to the athletic competition itself, may place a strain on some physiological systems (e.g. cardiovascular system, pulmonary system), and hence cortisol changes may be secondary to such factors rather than just the psychogenic aspects of the stressor. The difficulties of assessing the impact of sports stressors on cortisol variations are compounded by the fact that some studies involved well-trained and experienced athletes, whereas others used novices. Thus, differences in experience performing in public, self-confidence and self-esteem may all have contributed to variations of the cortisol response. Finally, it will be recognized that sustained exercise (especially among athletes) may result in hypothalamic–pituitary–gonadal neuroendocrine variations (e.g. in association with amenorrhea in females) that might not only influence basal cortisol levels (Ding et al. 1988; Brundu et al. 2006), but might also affect the response to stressors (McComb et al. 2006).

Although stimulated cortisol secretion for limited periods can be beneficial, as indicated earlier, prolonged elevations can promote several adverse physiological and psychological disturbances (e.g. hyperlipidemia, hypertension, chronic immunosuppresion, dysphoria, affective disorders, cognitive disturbances and sleep disorders) (McEwen 2000; Sapolsky et al. 2000; McEwen and Wingfield 2003). Thus, when faced with a chronic stressor, it would be adaptive for cortisol functioning to be down-regulated. Indeed, the present meta-analysis revealed that across stressor types, those of an acute nature elicited greater cortisol changes than did chronic stressors. Furthermore, in studies that assssed the impact of acute and chronic stressors, a similar outcome was observed. By example, the cortisol rise associated with parachuting diminished over successive jumps (Deinzer et al. 1997), as did the response to chronic exercise (Wittert et al. 1996). Further, it seems that with chronic illness, biphasic changes of circulating cortisol level occur. The distress initially leads to high levels of cortisol, coupled with a reduction of the cortisol binding protein, corticosteroid-binding globulin, hence resulting in greater free cortisol levels. With continued distress hypocortisolaemia may ensue (Beishuizen and Thijs 2004; Johnson and Rn 2006). These findings are in keeping with animal studies that indicated that the initially high levels of cortisol elicited by a stressor are abated with chronic exposure (Weiss et al. 1975; Nankova et al. 1993; Haleem and Parveen 1994; Armario et al. 2004). What is less clear, however, is whether the adaptation, such as that evident in chronic illness, represents an adaptation that reflects diminished distress and has beneficial value, or is actually a reflection of exhaustion of those processes governing HPA functioning (Beishuizen and Thijs 2004).

In humans, including the studies reviewed here, certain types of stressors tended to be more chronic than others (e.g. caregiving or job strain vs. academic examination or sports event), and frequently stressor chronicity and stressor type were confounded. Further to this same issue, certain stressors have a lengthy anticipatory period and/or a ruminative period following the actual experience. Thus, it is difficult in some instances to define when a stress experience began and when it ended. Finally, determining what constitutes a chronic stressor is dependent on a host of individual difference factors as well as those related to the characteristics of the stressor itself (i.e. severity, controllability, predictability, threat). Thus, although it generally appeared that the cortisol response was diminished with chronic stressors, it needs to be re-emphasized that this may depend on other factors, including the severity of the stressor as well as other ongoing stressors being encountered.

In most studies that assessed cortisol levels associated with stressful experiences, these were done at specific time(s) within a day, but relatively few studies assessed diurnal cortisol variations, including the early morning increase of cortisol secretion. Yet, it will be recalled that the rise of morning cortisol (i.e. over the first 30 min following awakening) appears to be particularly sensitive to the influence of ongoing life stressors (Melamed et al. 1999; Pruessner et al. 1999; Schmidt-Reinwald et al. 1999). Just as chronic stressors have been associated with a dampening of the morning cortisol rise, it was also reported that morning cortisol release was diminished among individuals experiencing PTSD (Yehuda et al. 1995a,b, 1996; Boscarino 1996; Goenjian et al. 1996; Anisman et al. 2001; Abercrombie et al. 2004; Lauc et al. 2004; Pico-Alfonso et al. 2004; Griffin et al. 2005; Wessa et al. 2006), although contradictory data have been reported in this regard (DeBellis et al. 1994; Lemieux and Coe 1995; Young and Breslau 2004; Young et al. 2004; Inslicht et al. 2006). Among the studies showing attenuated morning cortisol increase (or a flattened diurnal cortisol profile), it did not seem that this effect was unique to any given stressor that involved threats to the person, having been reported among war veterans with PTSD (Lauc et al. 2004), severely abused women (Pico-Alfonso 2005), and women undergoing the chronic distress of metastatic breast cancer (Abercrombie et al. 2004).

Based on a meta-analytic review regarding the cortisol changes associated with chronic stressors, Miller et al. (2007) concluded that a variety of factors related to characteristics of the stressor, as well as person variables, determined the cortisol profile that emerged. Specifically, it was observed that, in general, chronic stressors were associated with reduced morning cortisol release, coupled with greater afternoon/evening secretion (and hence a flatter diurnal cortisol curve), resulting in a higher overall daily cortisol output. These outcomes were most pronounced for traumatic stressors and those that entailed physical threats. In contrast, stressors that influenced the social self (including those that elicited shame) tended to promote elevated morning and afternoon/evening cortisol levels. These cortisol variations tended to diminish with time following the stressful experience.

Given the apparently protracted effects of chronic stressors, even if these are diminished relative to acute insults, the possibility ought to be considered that such events may give rise to a variety of pathological conditions associated with cortisol alterations, and these may be evident well after the stressor experience has ended. Indeed, these data raise the possibility that the processes leading to allostatic overload, and hence vulnerability to pathological outcomes, may be less related to the magnitude of the cortisol changes induced by the stressor than to the chronicity of these variations (McEwen 1998; McEwen and Seeman 1999). However, as far as we are aware, the possibility that chronic stressors, rather than stressor severity, may be most closely aligned with allostatic overload has not been reported in animals or in humans.

The present meta-analysis, as well as the discussion to this point, focused on the immediate effects of acute and chronic stressors on cortisol release. However, in assessing the impact of ongoing life stressors on cortisol reactivity, it might be considered that beyond their immediate effects, stressors may proactively augment neurochemical responses to subsequently encountered challenges (sensitization) (Anisman et al. 2003). This is particularly pertinent as such experiences may influence the recurrence of depressive symptoms, as well as development of PTSD (Post 1992; Breslau et al. 1999; Kendler et al. 2000; Heim and Nemeroff 2001; Penza et al. 2003). Essentially, stressors might influence vulnerability to later affective changes as the neurochemical substrates of the illness may evolve over time and repeated illness episodes. Studies in animals have indicated that although neuroendocrine adaptation occurs in association with a chronic stressor (although this effect may vary with the nature of the chronic stressor paradigm employed), when chronically stressed animals are subsequently exposed to a novel (heterotypic) stressor, an exaggerated cortisol response may be engendered (Armario et al. 2004; Dallman et al. 2004). The processes responsible for such an outcome have not been fully deduced and may involve either altered HPA feedback mechanisms or increased co-expression of ACTH secretagogues (CRH and arginine vasopressin (AVP)) at the external zone of the median eminence, that synergistically stimulate HPA activity (Tilders and Schmidt 1999; Anisman et al. 2003). For the present purposes the essential point is that stressors have ramifications on cortisol functioning that may persist long after the initial stressor experience has ended, provided that individuals again encounter a stressor (even if it differs from the initial traumatic or chronic event). Thus, although cortisol variations were reported to decline over time following a chronic stressor experience (Miller et al. 2007), this should not necessarily be interpreted as HPA reactivity being normalized. To the contrary, among some individuals, increased HPA reactivity may persist for extended periods.

Intuitively, one might expect that uncontrollable stressors would be more aversive and threatening than controllable stressors, and hence would lead to greater neuroendocrine variations. Ordinarily, when faced with an acute stressor, it would be adaptive for certain neuronal responses to be elicited rapidly, irrespective of the attributes of the stressors (e.g. controllable vs. uncontrollable; predictable vs. unpredictable). After all, when a stressor is initially encountered, especially if it is one that has not previously been experienced (making appraisal of the stressor difficult), a robust neuroendocrine response would assure adequate resources to deal with the stressor. Once appropriate appraisals of the situation have been made, and the threat is deemed to be modest or controllable, then the biological response might taper off. Indeed, those systems that are necessary for immediate responses to stressors (i.e. those that threaten the well-being of the organism), including activation of the sympathetic nervous system and even fundamental immune responses that act against pathogenic stimuli, should react rapidly and comparably to both controllable and uncontrollable stressors (Sapolsky et al. 2000). However, in actuality, it is often difficult to dissociate controllability or uncontrollability of a stressor from the chronicity of the stressor. In most instances, controllable stressors might be deemed to be acute (after all, if a stressor is perceived as being controllable, by definition it ought to be possible for the individual to terminate it, and hence its duration should be brief), whereas uncontrollable stressors can be either acute or chronic.

Summary and limitations

It appears that stressors experienced in natural settings were associated with increased cortisol release, but the magnitude of the effects were generally smaller than those observed in certain laboratory contexts. Not surprisingly, stressors related to medical procedures were most apt to promote elevated cortisol levels. Interestingly, the effect size associated with chronic stressors was smaller than that associated with acute stressors. Yet, given the possibility that stressors may engender sensitization of neuronal processes leading to changes in cortisol regulation, it should be considered that subsequent stressor experiences might elicit pronounced neuroendocrine responses.

The influence of social-evaluative threat found in the meta-analysis of laboratory stress studies by Dickerson and Kemeny (2004) might also manifest itself in the individual's self-evaluation, cognition, or beliefs about the self and might thus influence behavioral and neurochemical changes (Dickerson et al. 2004; Gruenewald et al. 2006). Such factors have not been intensively examined within a naturalistic setting, and hence their contribution to variations in cortisol secretion is uncertain. Nevertheless, it may be significant that internalized racism (i.e. where the individual accepts the racist views of others as being correct) was associated with increased cortisol levels (Tull et al. 2005), which likely reflects low self esteem of these individuals. It was likewise reported that threats to one's group was associated with elevated levels of cortisol, particularly among individuals who tended to express higher levels of anger (Matheson and Cole 2004).

Despite the relatively large number of studies that have evaluated stressor-provoked changes in cortisol secretion, the influence of a great number of variables remains to be determined. In this regard, as already indicated, most life stressors are complex, typically comprising multiple strains. Thus, it is difficult to identify the unique contributions of each element of the stressor mosaic, although based on the present analysis it did not seem that experiencing stressors with multiple components (i.e. those stressors with several attributes that might augment cortisol levels, including stressors that are chronic, uncontrollable and unpredictable) synergistically enhanced cortisol levels. As well, limited information is available concerning the influence of individual difference factors (personality factors), stressor appraisal and coping strategies, as well as previous stressor experiences (including trauma) in determining HPA functioning. It is conceivable that the use of oral contraceptives influences cortisol levels in natural settings just as they were reported to do so in laboratory tests (Kirschbaum et al. 1999). However, as most studies did not report this information, nor was there typically any mention of phase of the menstrual cycle, their influence could not be assessed. These limitations notwithstanding, the present findings reinforce the view that changes in HPA activity represent dynamic processes to accommodate challenges, and may contribute to the provocation or exacerbation of various stress-related illnesses.

Acknowledgements

Supported by the Canadian Institutes of Health Research and by the Natural Sciences and Engineering Research Council of Canada. HA holds a Canada Research Chair in Behavioral Neuroscience.