The Effects of Perceived Stress, Traits, Mood States

The Effects of Perceived Stress, Traits, Mood States, and Stressful Daily Events on Salivary Cortisol MARLEEN VAN ECK, MS, HANS BERKHOF, MS, NANCY NICOLSON, PHD, AND JOSE SULON, PHD

This study examined the effects of perceived stress and related individual characteristics, mood states, and stressful daily events on salivary cortisol levels. Forty-one "high stress" and 46 "low stress" subjects were selected on the basis of Perceived Stress Scale scores from a sample of male, white collar workers. Subjects completed Experience Sampling self-reports and collected saliva samples 10 times a day over 5 consecutive days. Multilevel analysis revealed that trait anxiety and depression, but not perceived stress, were associated with small but statistically significant cortisol elevation. No effects on cortisol were found for recent life events, chronic difficulties, trait anger, or psychosomatic symptoms. Distress, as reflected by the mood states Negative Affect and Agitation, was associated with higher cortisol levels, whereas Positive Affect had no statistically significant effect. Stressful daily events were associated with increased cortisol secretion, the magnitude of the effect depending on whether the event was still ongoing and on how frequently a similar kind of event had occurred previously. Although perceived stress, anxiety, and depression did not increase cortisol reactivity to daily events, we found evidence for reduced habituation to recurrent events in subjects scoring high on these traits. Mood appeared to play a mediating role in the relationship between stressful events and cortisol secretion. These results suggest that negative affectivity is not just a confounder but is related to elevated cortisol secretion during normal daily activities. The finding that even minor events and fluctuations in mood states were associated with increased adrenocortical activity points to a possible mechanism linking subjective experience to health outcomes. Key words: perceived stress, trait anxiety, cortisol, experience sampling, mood, daily events.

INTRODUCTION

The neuroendocrine system has long been thought to play an important role in the causal pathway linking stress and ill health (1-3]. The hypothalamic-pituitary-adrenal (HPA) axis, which is involved in the regulation of a wide range of physiological and behavioral responses to stress, has been implicated in numerous illness processes (4-6), including the etiology of psychiatric disorders (7-9). Over the past decade, it has become clear that not only major life events (10, 11) but also minor daily stressors or "hassles" can have negative effects on health and well being (12-14). In contrast to the wealth of information concerning the neuroendocrine effects of major real life stressors, there is relatively little

From the Department of Psychiatry and Neuropsychology, Social Psychiatry and Psychiatric Epidemiology section, University of Limburg, The Netherlands (M.V.E., N.N.); Department of Statistics and Measurement Theory, University of Groningen, The Netherlands (H.B.); and Faculty of Veterinary Medicine, University of Liege, Belgium (J.S.). Address reprint requests to: M. M. van Eck, Social Psychiatry and Psychiatric Epidemiology Section, Department of Psychiatry and Neuropsychology, University of Limburg, P.O. Box 616, 6200 MD Maastricht, The Netherlands. Received for publication November 29, 1994; revision received December 1, 1995.

Psychosomatic Medicine 58:447-458 (1996) 0033-31 74/96/5805-0447$03.0Q Copyright © 1996 by the Americ

known about the effects of chronic stress and even less about the effects of minor daily events. The goal of the current study was to increase our understanding of the impact of daily life stress on the HPA axis. Although a growing number of studies have examined the relationship between daily events and mood (15-22) or physical symptoms (23-25), few have investigated whether stressful daily experiences have an effect on cortisol excretion. Findings have been inconsistent, which is not surprising in light of the large differences in cortisol measurement procedures and definitions of daily stress or distress. Cortisol levels have often been based on a single measurement per subject or per day. To illustrate the diversity of results, greater work demands were associated with lower cortisol levels in subjects sampled in the morning (but not in the afternoon) in one study (26); another study found that feelings of irritation, tenseness, and tiredness in assembly line workers were associated with elevated cortisol levels on workdays and that cortisol levels were absolutely higher on "bad" compared with "normal" or "good" workdays (15). Examining within-subject associations over several days, one study found elevated afternoon urinary cortisol on high stress compared with low stress days (27), and in another, no relationship between the number of undesirable events reported at the end of the day and cortisol levels

447

M. VAN ECK et al. measured in evening urine could be demonstrated (28). With more frequently measured responses to work stress in air traffic controllers, Rose and colleagues (29) found that a subgroup of subjects responded to an increase in the number of planes they had to manage with large increases in cortisol. New approaches are needed for investigating cortisol responses to daily events. Repeated measurement of cortisol not only increases reliability but provides a clearer picture of the temporal relationship between stressful events and neuroendocrine responses. This is especially important if the events of interest occur unpredictably (as is usually the case in the natural environment) and therefore cannot be directly monitored by the researcher. Additional insights into the stress process can be gained by including measures of the context in which events take place and measures of subjective responses, including mood and cognitive appraisals (30). The current study therefore combines Experience Sampling (31, 32) methodology (ESM) and frequent salivary cortisol sampling to address two main questions about daily life stress and cortisol:

Are High Perceived Stress and Distress Associated with Elevated Cortisol Levels ? Despite the lack of consistent results in the literature on the effects of chronic stress on the HPA axis, we reasoned that healthy individuals experiencing persistent but not overwhelming levels of distress and difficulty in coping with daily demands would have higher overall cortisol than persons who experience fewer problems. In addition to the effects of perceived stress level, we examined the contributions of conceptually related measures of stress exposure and chronic distress. These included recent life events, chronic difficulties, trait anxiety, depression, hostility, and psychosomatic complaints. Several of these variables have previously been linked to cortisol levels (33-35). In addition to trait-like measures of stress and distress, we assessed the effects of negative and positive mood states on cortisol. Although there is abundant evidence that cortisol increases in response to negative states (36-39), the effects of positive mood are less clear. Positive affect has been associated with lower (40, 41) as well as higher (42) cortisol levels. 448

Do Stressful Daily Events Influence Cortisol Levels? In general, we hypothesized that even minor stressors could result in increased cortisol secretion. We further hypothesized that such effects might be dependent on an individual's chronic level of perceived stress. Based on results of previous studies showing greater psychological impact of work-related events (20) and negative social interactions (17), we expected that these categories of events would have the largest impact on cortisol. Similarly, events that subjects rated as more unpleasant, more important, less predictable, and less controllable were expected to have larger effects, and ongoing events were expected to have more effect on cortisol than recently terminated events. Finally, we hypothesized that events reported to occur relatively infrequently in daily life would, due to their novelty, have a greater effect on cortisol levels than recurrent events. The study design compares two groups of male, white collar workers, with high vs. low levels of perceived stress. A total of 87 subjects completed ESM self-reports and collected saliva for cortisol determination at frequent intervals over a period of 5 days. A previous analysis of a subset of these data, based on subject mean cortisol levels, provided some support for the hypothesis of elevated cortisol levels in the high stress group, at least on workdays (43). The current analysis adds to these preliminary findings by assessing the separate contributions of "trait" (eg, perceived stress level, trait anxiety, depression) and "state" (eg, mood, appraisal) variables, at the same time controlling for diurnal and possible confounding influences on cortisol secretion. Most importantly, the application of hierarchical linear modeling, or multilevel analysis (44), enables us to investigate the neuroendocrine effects of the wide variety of stressful events experienced in daily life.

METHODS Subjects Local industries and government agencies were approached via their personnel departments to participate in the study. A decision as to which categories of employees fell under the definition of "white-collar" was made by each personnel department on the basis of standard job function descriptions. Questionnaires were distributed among these employees, accompanied by a letter explaining the goals of the study. Participation was voluntary, and care was taken to insure anonymity; 316 male employees from six different industries or agencies completed the screening questionnaire. The mean score on the Perceived Stress Scale (PSS) for this

Psychosomatic Medicine 58:447-458 (1996)

STRESS AND CORTISOL IN DAILY LIFE sample was 12.7 (SD 6.0), similar to U.S. norms (mean 13.02, SD 6.45) (45). Ninety-two subjects with scores in the upper or lower tertiles of the screening sample distribution (PSS-10 score £10 or S16) were recruited, excluding individuals who reported a history of serious chronic illness, endocrine disorder, medications known to affect cortisol levels, or treatment (past or current) for mental health problems. Exclusion criteria were reassessed during an initial interview, study aims and procedures were explained, and informed consent was obtained. During subject intake, each "high stress" (HS) subject was matched for age group, marital status, and household composition with a "low stress" (LS) subject to insure that the two groups did not differ on demographic characteristics that might affect exposure to certain classes of daily stressors. Five subjects were later excluded from analysis: four due to failure to meet ESM compliance criteria (see Daily Experience section, below) and one because he became so acutely stressed that he was unable to work during the sampling period. Of the 87 remaining subjects, 41 subjects comprised the HS group, and 46 subjects comprised the LS group. Mean age was 42.1 years (range 27-57 years); 89% were married, and 81% had children living at home.

Questionnaires The following measures were used in the current analysis: Perceived Stress. The 10-item version of the PSS was translated into Dutch. The items are rated on a 5-point frequency scale, ranging from "never" (0) to "very often" (4) in the last month. Life Events. Life events were recorded with the questionnaire form of the List of Threatening Experiences (LTE-Q) (46). Subjects were asked about the occurrence of 12 categories of events (eg death of a partner, child, parent, divorce) during the last year. Long Term Difficulties. Chronic stress was assessed with the Long Term Difficulties Questionnaire (LLM) (47). This inventory focuses on problems in relation to work/study, housing, physical environment, leisure, finance, and social relationships (partner, family, friends, neighbors). Subjects rate each of the 16 items on a 4-point intensity scale with the anchors (l) none, (2) some, (3) quite, (4) serious (difficulties). A total score is obtained by summing across all items. Psychosomatic Symptoms. The revised version of the Psychosomatic Symptom Checklist (PSC) (48) includes 17 common psychosomatic complaints (eg, headaches, backaches, nausea). Subjects rate each complaint on 5-point scales for frequency (0, "never or rarely occurs" to 4, "occurs daily") and intensity (0, "not bothersome" to 4, "extremely bothersome"). A total score is obtained by summing the cross-products of each item's frequency by intensity. Depression. Depressive symptomatology was assessed with the Dutch translation (49) of the Zung Self-Rating Depression Scale. Anxiety. Trait anxiety was measured with the Dutch version (50) of the State-Trait Anxiety Inventory (STAI). Anger. Trait anger was measured with the Dutch version (51) of the Spielberger Trait Anger Scale. The scale has two subscales: "anger-temperament" and "anger-reaction."

Daily Experience The Experience Sampling Method (31, 32) was used to collect data from subjects during their normal daily activities. Subjects received auditory signals ("beeps"), after which they filled in a questionnaire and collected a saliva sample. After receiving


detailed instructions, subjects were sampled for a period of 5 consecutive days (Thursday through Monday). A Seiko wristwatch was programmed to emit "beeps" 10 times each day, at semi-random intervals of approximately 90 minutes, between the hours of 8 AM and 10 PM. Beeps were clustered around the midpoint of each time block (8:15 AM, 9:45, 11:15, and so on), with the exact time sequence of "beeps" varied each day to decrease predictability. In a final "debriefing" session, subjects were asked to clarify reasons for missing data. The criteria we set for subject inclusion in the analysis (±20 ESM reports completed within 20 minutes after being signalled and no missing data for entire days) were met by all but four subjects (two from each group). The remaining 88 subjects completed an average of 83% of all possible responses within the time limit, for an average of 41 responses per subject. HS and LS groups did not differ in compliance rates (40.1 vs. 42.3 responses per subject, Mann-Whitney U test, p > .05). Compliance was lowest for the first "beep" (at approximately 8:15 AM), with an average of 73% valid reports. On Saturdays, response rate for first "beeps" was 61%, and on Sundays it was 59%. On weekends, 74% of all missing and invalid responses could be attributed to the fact that subjects were still asleep when signalled. The ESM form contained open-ended questions concerning thought content, the physical and social context, and what the individual was doing when signalled. The forms also included Likert scales (from 1, "not at all" to 7, "very much") for rating aspects of thoughts, mood, physical well being, individually defined (psycho)somatic complaints, present activity, and stressful events. Subjects were asked to describe any stressful events or situations that may have taken place in the interval since the last ESM report and to rate these events on a number of dimensions: unpleasantness, importance, predictability, controllability, and frequency of prior occurrence. Subjects were also asked to indicate at what time the event had started and if and when it had ended at the moment they were "beeped." Information concerning maximum level of physical exertion, smoking, food, coffee, and alcohol intake since the last "beep" was also obtained. The 17 ESM mood items were reduced to three mood measures, identified by means of principal component analysis with varimax rotation, which accounted for 78% of the total variance when subject mean scores were used. Ratings on the items "cheerful," "satisfied," "relaxed," "energetic," "self-assured," "concentrated," and "enthusiastic" were summed to form a "Positive Affect" scale (Cronbach's a = .95). Two separate components of negative affect were identified: "Negative Affect," including the items "depressed," "anxious," "worried," "lonely," "tired," and "miserable" (a = .87), and "Agitation," with the items "restless," "irritated," "hurried," and "nervous" [a = .93). The sums of the scale items were divided by the number of items so that all mood measures had ranges from 1 to 7. Subjects' descriptions of stressful daily events were first coded according to content, with categories work, network (events concerning family, friends and acquaintances), household/financial, leisure, personal health-somatic, personal health-psychological, transport, and other. Twelve events coded in the two personal health categories were excluded from analysis because of possible confounding with psychological and somatic state measures. In the current analysis, the remaining events were collapsed into the categories work (48.0% of events) and nonwork (50.5%). Some examples of reported work events were: "unclear/ vague assignment at work," "too much to organize, not enough time," "difficult conversation with boss about job performance," and "leading a big meeting." Reported nonwork events included: "having a fight with my wife about household duties," "conflict

449

M. VAN ECK et al. with spouse about how to raise our son," "child who will not listen," and "making dinner, child crying, other child taking a bath, and this all at the same time." In addition to content, events were classified according to whether or not they involved a social interaction and/or a task demand.1

Salivary Cortisol Salivary cortisol is a reliable indicator of the free cortisol in plasma, which is considered to be the biologically active hormone (53). Salivary cortisol increases within minutes in response to acute stressors and has a half-life of approximately 1 hour (54). We have found no differences in cortisol levels in uncentrifuged samples frozen immediately or kept at room temperature for 2 days (55); others report no differences for up to 2 to 4 weeks (56, 57). At the same time ESM forms were being completed, subjects collected saliva by holding a cotton dental roll in the moudi for approximately 1 minute. The saturated roll was placed in a capped plastic vial ("Salivette," Sarstedt, Rommelsdorf, Germany), which subjects stored in their home freezers each evening. At the end of the sampling period, uncentrifuged samples were stored at — 20°C until analysis. Compliance with saliva sampling was good in both high and low stress groups, with the same mean response rate (83%) as for the ESM reports. Five extreme cortisol values (>1200 ng/dl) were deleted from the data set before analysis. Salivary cortisol levels were determined in duplicate by direct radioimmunoassay (58) using 125I-cortisol (Farmos Diagnostica, Finland) and antiserum made against the 3-CMO-BSA conjugate by Dr. J. Sulon, University of Liege, Belgium. The lower detection limit of the assay was 12 ng/dl, with a mean intra-assay coefficient of variation of 4.8% (range: 2.2-7.5% for four assays). All samples from an individual subject were analyzed in the same assay to reduce sources of variability.

Statistical Analysis: Multilevel or Hierarchical Linear Model The multilevel model or the hierarchical linear model (44, 59) is a variant of the multiple linear regression model, applicable for data with a hierarchical nesting structure. In the present data set, the measurements made at the "beeps" were nested within persons. The two nesting levels were called "measurement level" and "person level." At each level of the hierarchy, explanatory variables can be added to the model. The variables that are added at the measurement level (eg, mood states, events) vary with time, and the added variables at the person level represent characteristics of individual respondents (eg, high or low perceived stress, trait anxiety). An advantage of the hierarchical linear model is that it allows for missing observations. In addition, the observations do not need to be evenly spaced over the time interval.

1 The reliability of the coding system was assessed by comparing the classifications of 345 events by two independent coders. Interrater agreement was determined by means of Cohen's K. On the whole, the qualitative information could be classified with a high degree of agreement (52), especially for the content categories. The overall K for content was .90, with intra-category KS ranging from .60 to .96; KS for social interaction and for task demand were .73 and .65, respectively.

450

At the measurement level, the relationship between cortisol and time of day can be modeled explicitly. If, for example, this relation is linear, the model would be of the following form: (CORT),, = bu, + 6,, (TIME),, + e,,, where (CORT)it is the level of cortisol at the f-th measurement of person i, (TIME)it represents the time of the day at which this measurement is taken, boi and b,, are the intercept and slope of person i, and the eits are normally distributed error terms with mean 0 and variance a2. The resulting equation can be extended with time-varying explanatory variables. For example, to study the effect of stressful event occurrence on cortisol, a dummycoded variable (EVENT)lC can be added to the above equation as follows: (CORT),, = £>„,- + bu (TIME),, + o2, (EVENT),, + e,,, where 62| is the effect of a stressful event for person i. The regression coefficients b^ [j = 0, 1, 2 ) are allowed to vary across individuals. Therefore, we can split by, into two components: a fixed component doj that is constant across persons (fixed effect) and a random component u,, that varies across persons (random effect). This gives the following person-level model for ): \ (/ = 0, 1, 2 bj, = dn/ + u,,,

Instead of estimating u-ti for each person, we postulate that the u,|S (i = 1, 2, 3 ) are random draws from a normal distribution (explaining the term "random effect"). The mean of Uyt is 0, and the variance of u,, is T r . The value of r^. indicates how much the value of u,, differs across persons: The higher the value of T^., the more the values of Uy, differ. The covariance between uti and uki (/ not equal to k) is Tjk. A positive value of r|k implies that a person with a high value offa,tends to have a high value of £>k as well. Suppose that part of the variation in b,, (/ = 0, 1, 2 ) across persons can be explained by the person variable (GROUP), that indicates whether person i belongs to the high or low stress group. The regression coefficients b)t (/ = 0, 1, 2) can now be modeled as bt, = d0, + dy (GROUP), + u,,,

where dy) {j = 0,1, 2 ) denotes the fixed effect of (GROUP), on b . The Uji (/= 0, 1, 2 ) is now a random effect of person i after controlling for GROUP,. Estimates of the fixed effects d0| and d,j (; = 0, 1, 2), as well as of the (co)variances a2 and T)k (/, k=Q, 1, 2), were obtained with the program ML3e (60). For model selection, we started with an empty model and added main and interaction effects of the theoretically important variables. Significance of fixed effects was tested by dividing the estimated effect by its SE. This ratio is approximately normally distributed. For testing the significance of the (co)variances r|k (/, k = 0,1, 2), we applied the likelihood ratio test (44). This test compares a model with and a model without a random effect. Hence, the (co)variances that are added to the model because of the specification of an extra random effect are tested simultaneously. Nonsignificant effects were excluded from the models. The postulated normality of the random effects was checked by inspecting normal probability plots of the individual estimates of b]{ (j= 0, 1, 2), which were obtained by the empirical Bayes approach (44, 59). The normality of the measurement level empirical Bayes residuals was also checked by inspection of the normal probability plot. The observed cortisol values for each person were plotted by time of day to check for the presence of outlying cortisol curves (see Results). The estimation of effects on cortisol entailed four steps, pre-


STRESS AND CORTISOL IN DAILY LIFE sented here as separate models. The analysis of repeated assessments of salivary cortisol is complicated by the hormone's strong diurnal rhythm and secretory peaks (which lead to a decline in variance from morning to evening), so the first step involved accurate fitting of the diurnal curve to allow comparisons of cortisol values across the day. Next, possible confounding factors, such as smoking (61), exercise (62), and coffee (63) and food intake (64), were included as explanatory variables in the same model. All variables with significant fixed effects identified in this model were included in all subsequent models. The remaining three models test our main hypotheses. First, the effects of level of perceived stress, mood states, and individual trait characteristics on overall cortisol level were estimated. Workday vs. weekend differences and interactions between stress level and diurnal variability were also examined. The next model estimated effects of stressful events and event characteristics on cortisol, excluding the contribution of mood variables. In the final model, mood state variables were re-entered to evaluate the relative contributions of events, mood, and trait characteristics to cortisol levels. Instead of a model with two levels, we could have formulated a three-level model, in which measurements are nested within days, which in turn are nested within persons. However, when a threelevel model was evaluated, the p values were approximately the same as the p values of the two-level model. Because the more complex model did not change the conclusions, we decided to present the simpler model. The similarity of the results of the two models can be explained by the observation that the variance at the day level, although present, was small compared with the variance at the person and the measurement levels. Again, for the sake of simplicity, we did not extend the model with an autocorrelation term because its inclusion did not change the model results.

RESULTS

Characteristics of High and Low Stress Groups High and low stress groups differed on almost all measures of stress and distress, as assessed with questionnaires and ESM reports (Table 1). Only the

number of life events experienced in the past year did not differentiate the two groups.

Controlling for Diurnal and External Influences on Cortisol Secretion Because the measurement level residuals in a model fitted with raw cortisol data were highly skewed to the right, with decreasing variance from morning to evening, data transformation was necessary to meet standard model assumptions. Log transformation, the usual remedy for cortisol skewness, resulted in increasing variance over the day. We therefore used a fifth root transformation (cortisol0'2], which gave normally distributed residuals with a homoscedastic pattern throughout the day. The observed cortisol curve was best described by a two-piece, third-degree polynomial (spline function (65)), with the node at 12:25 PM. The spline was created by adding a truncated term (defined as (TIME - 12:25)3 after 12:25 PM and as 0 before 12:25 PM) to a third-degree polynomial, which improved fit in the morning hours when cortisol levels dropped most sharply. Random terms were added to allow each subject to have his own intercept and slope. Random terms for TIME2, TIME3, and the truncated term did not change the magnitude of the fixed effects and were therefore excluded from the model for the sake of simplicity. All time effects (fixed and random) included in the model were highly significant. As shown in Figure 1, the estimated cortisol time curve closely approximates the observed mean values at the "beeps" and is clearly superior to a simple linear curve. Plots of the ob-

TABLE 1. Characteristics of High and Low Stress Groups

Total n Age Questionnaires Perceived stress Trait anxiety Trait anger Depression Psychosomatic symptoms Life events Chronic difficulties

1_S (mean (SD))

p Value HS (mean (SD)) (two-tailed)

46 42.7

(7.7)

42 41.5

(5.9)

ns

7.2 28.3 18.8 36.5 6.1 .5 19.5

(2.2) (4.4) (4.6) (5.5) (5.6) (.7) (2 3)

18.1 39.8 23.0 48.4 27.5 .8 23.3

(3.4) (7.6) (5.2) (7.7) (23.2) (.9) (4.0)