Cognitive performance in patients with burnout, in relation to diurnal salivary cortisol

Abstract

This study investigated cognitive performance in patients with burnout, in relation to the flexibility of the hypothalamic–pituitary–adrenal (HPA) axis. Clinical cases with work stress-induced burnout (n = 65), and demographically matched, healthy reference subjects (n = 65), were given six neuropsychological tests and a self-rating scale for cognitive problems. Diurnal salivary cortisol was measured among burnout cases and an external reference group (n = 174), including a dexamethasone suppression test (DST) among burnout cases. Compared with referents, the burnout group under-performed in a cognitive speed test (Wechsler Adult Intelligence Scale-Revised (WAIS-R) Digit Symbol), but not in any other test of sustained attention, episodic memory, or vocabulary. Burnout cases had considerably more subjective cognitive problems, but ratings were unrelated to test performance. Compared with referents, burnout cases had similar morning salivary cortisol levels and similar awakening response, but lower evening cortisol. Among burnout cases, lower diurnal cortisol variability was related to slower performance in several tests. The DST response showed no consistent relationship with any cognitive parameter. Hence, despite considerable subjective cognitive problems, the burnout group showed only a partial, mild deviation in cognitive performance. A flatter diurnal cortisol profile was related to lower cognitive processing speed, but diurnal cortisol pattern and DST response were normal, suggesting a maintained HPA axis flexibility.

• Attention
• cortisol
• exhaustion
• memory
• mood
• neuropsychological tests

Introduction

Work stress has been proposed to be a major cause of health problems and sick leave in European Union (EU) countries in recent years (Parent-Thirion et al. 2007). The major contributing factor suggested is the rapidly changing structure of work, with a shift from manual to mental demands (Kompier and Taris 2005). One disorder, burnout, has been suggested to be specifically related to long-term work stress. Burnout is characterized by exhaustion, cynicism, reduced professional efficacy (Maslach et al. 2001), and neuroendocrine disturbances, in particular a dysregulation of the hypothalamic–pituitary–adrenal (HPA) axis, with elevated concentrations of glucocorticoids, sleep disturbances, and cognitive dysfunction (McEwen and Sapolsky 1995; Ekstedt et al. 2006).

Despite extensive animal and human experimental research, the precise relationship between cognitive problems and HPA axis dysregulation has remained unclear. Whereas many early studies focused on the vulnerability of the hippocampus to long-term elevated cortisol level, recent data suggests that other limbic structures (e.g., the amygdala) and the prefrontal cortical areas may be more vulnerable (Lupien and Lepage 2001; McEwen 2005; Radley et al. 2006). A widespread effect of cortisol on brain functions also seems to fit the observations that reduced performance is not restricted to hippocampus-related cognitive processes [i.e., primarily memory consolidation and retrieval (Roozendaal 2000)], but extends to executive functions, such as sustained attention and working memory (Horner and Hamner 2002). Only a few studies of cognitive performance in work stress-induced burnout have been published (Sandström et al. 2005; Van der Linden et al. 2005; Rydmark et al. 2006; Öhman et al. 2007). These studies have shown slight impairment in complex attention and memory tests, but normal results in more basic tests of memory and intellectual ability.

Long-term increased cortisol secretion may also change the density and sensitivity of the cortisol (glucocorticoid and mineralocorticoid) receptors in the brain, altering the set levels of the feedback system for maintaining HPA homeostasis. Transient effects on cognition from increased cortisol level, induced either pharmacologically or by stress provocation, have also been convincingly demonstrated (Het et al. 2005; Sauro et al. 2003). However, cognitive function can be compromised by too low as well as too high levels of free cortisol (De Kloet et al. 1999). This means that a transient change in cortisol level may either hamper or facilitate cognitive functioning, depending on the long-term cortisol levels of the individual (Lupien et al. 2002) and the point on the diurnal cortisol profile at which the increase occurs (Maheu et al. 2005). This complex interaction of long- and short-term effects of cortisol makes it hard to relate a given cortisol level to cognitive functioning, not least among burnout cases with a probable, but undocumented, history of long-term increased cortisol secretion. Studies of the diurnal cortisol rhythm in burnout have shown conflicting results, with the awakening response increased (De Vente et al. 2003; Grossi et al. 2004), decreased (Pruessner et al. 1999; Sonnenschein et al. 2007), or normal (Mommersteeg et al. 2006). It is unclear which of these patterns is more theoretically plausible since elevated cortisol levels have been found in major depression and low cortisol levels in post-traumatic stress disorder (PTSD) and chronic fatigue syndrome, disorders with symptoms that to some extent resemble those of burnout although they are regarded as separate entities (Iacovides et al. 2003). One possibility is that the earlier stages of burnout are characterized by increased levels of free cortisol, while the later stages are associated with lower than normal cortisol levels, representing a breakdown of the endocrine feedback mechanisms (McEwen 1998; Fries et al. 2005). Altered feedback sensitivity of the HPA axis is commonly assessed using the dexamethasone suppression test (DST). Signs of DST hypersuppression have been found among high scorers on a burnout scale (Pruessner et al. 1999), but this could not be replicated in a recent study (Mommersteeg et al. 2006).

The purpose of the present study was to assess the character and scope of cognitive problems in burnout in relation to cortisol, by testing: (i) whether burnout is associated with lower performance in neuropsychological tests, (ii) whether the degree of subjective cognitive problems in burnout is associated with the objective cognitive performance level, and (iii) whether subjective and objective performance are associated with diurnal cortisol pattern and the DST response.

Materials and methods

Participants

The subjects were primarily recruited to a broad study of the characteristics of burnout, focused on a workplace intervention to facilitate return to work (details to be published elsewhere). Participants were identified in collaboration with social insurance offices in southern Sweden. Over a 3-year period the offices notified the research group of cases on recent sick leave, with work strain as a probable cause, having any of the International Classification of Diseases (ICD-10) diagnoses within the category F43 (reaction to severe stress, and adjustment disorders), with the exception of PTSD (F43.1). Potential participants (randomized selection if exceeding the capacity) were informed by letter that the intent was to offer employees with short-term sick leave due to work stress an examination and an intervention at their workplace. During the course of the study, a neuropsychological examination was added to the examination program. During this period 729 persons were mailed the information letter. A subsequent telephone call, intended to recheck the relation to work strain, showed that 13% could not be reached, 9% had non-work related exhaustion or other disease, 12% had no motive for the workplace intervention (e.g., had quit work, were about to retire, could not assimilate the intervention), and 53% were uninterested (30% reported improved health, 3% too poor health, and 20% other factors). The remaining 101 persons underwent an initial clinical interview, after which 68 persons were considered eligible, of which 65 agreed to participate in the neuropsychological examination.

The main clinical examination comprised assessments by a senior physician and a psychologist at the Department of Occupational and Environmental Medicine, Lund University Hospital, Lund, Sweden. The medical workup showed that participants had generally good physical health, but 26% had minor somatic disorders (Table I). Current state of burnout was validated with the Maslach Burnout Inventory-General Survey (MBI-GS) (Maslach et al. 1996), which showed a mean exhaustion score of 4.6 (SD 1.2) compared with mean scores in the 1.2–1.9 range found in Scandinavian populations (Richardsen and Martinussen 2005; Ahola et al. 2006). In all, 89% of the participants met the criteria for exhaustion disorder suggested by the Swedish National Board of Health and Welfare (Socialstyrelsen 2003), which requires the presence of one or several clearly identifiable strain factors during at least six months. Our impression was that most participants had experienced a severe workload for years, accompanied by increasing symptoms. Screening with the Primary Care Evaluation of Mental Disorders (Prime-MD) (Spitzer et al. 1994) showed that 69% of the subjects met the criteria for a Prime-MD diagnosis, most commonly depression and/or anxiety disorder (Table I). Self-rating scales for depressive symptoms, Symptom Checklist-90 (SCL-90) (Derogatis 1992) and Beck Depression Inventory (BDI) (Beck et al. 1988), also showed elevated levels (Table I). Antidepressants, anxiolytic drugs, or sleeping pills were used by 38% (including in combination), and 42% had entered psychotherapy. In general, treatment had begun only weeks before the examination, so major treatment effects were unlikely. At the time of examination, 60% were on full-time sick leave, 32% were on part-time sick leave (working 50–75% of their regular hours), and 7% had returned to full-time work. At the time of examination, 90% had been on sick leave for more than 85 days.

Table I.  Characteristics of the burnout group (n=65)

Characteristic	n (%)	Mean (SD)
Somatic disorders, total	17 (26)	–
Hypertension (treated)	6 (9)	–
Type II diabetes	2 (3)	–
Whiplash	2 (3)	–
Other^*	7 (11)	–
Medication
Antidepressants; SSRI, SNRI, or NaSSA^†	19 (29)	–
Anxiolytics or sleeping pills	10 (15)	–
Medication for somatic disorders^‡	13 (20)	–
No medication	36 (55)	–
Psychotherapy	27 (42)	–
Alcohol consumption
Low to moderate	63 (97)	–
High	2 (3)	–
Tobacco use, regular	10 (15)	–
Prime-MD
Major depression	13 (20)^¶	–
Slight depression	7 (11)^§	–
Anxiety disorder	7 (11)	–
Anxiety disorder + major depression	14 (21)^‖	–
Anxiety disorder + mild depression	3 (5)	–
Social phobia	1 (2)	–
MBI-GS, exhaustion score	–	4.6 (1.2)
SCL-90 subscales
Anxiety	–	1.6 (0.9)
Somatisation	–	1.3 (0.8)
Depression	–	1.9 (0.8)
BDI mean raw score	–	19.3 (8.7)
Severe depression	8 (12)	–
Moderate depression	31 (48)	–
Slight depression	15 (23)	–
Minimal depression	11 (17)	–

Prime-MD, Primary Care Evaluation of Mental Disorders (Spitzer et al. 1994); MBI-GS, Maslach Burnout Inventory-General Survey (Maslach et al. 1996); SCL-90, Symptom Checklist 90 (Derogatis 1992); BDI, Beck Depression Inventory (BDI) (Beck et al. 1988)

^* These subjects each had one of: thyroid disorder (treated), gluten intolerance, allergy, chronic bronchitis, recurrent diverticulitis, atrial fibrillation, slight B12 deficiency, and mineral deficiency (treated)

^† Selective serotonin reuptake inhibitor (SSRI), serotonin–norepinephrine reuptake inhibitor (SNRI), or noradrenergic and specific serotonergic antidepressant (NaSSA)

^‡ Mostly antihypertensive, antihyperlipidemic, and antidiabetic drugs, beta blockers, vitamin or mineral supplements, or weight control drugs

^¶ Two subjects also fulfilled the criteria for somatoform syndrome NUD

^§ One subject also fulfilled the criteria for somatoform syndrome NUD

^‖ One subject also fulfilled the criteria for obsessive–compulsive disorder.

The reference group comprised healthy persons examined with the same neuropsychological tests and medical screening in the context of previous studies (Österberg et al. 2000, 2002). From a total pool of 98 men and 102 women, 65 referents were successfully matched with the members of the burnout group for gender (71% women), age (burnout: mean 48.2; SD 9.5 years vs. referents: mean 47.9; SD 8.3 years), and education (54% had studied at university, 28% at upper secondary school, and 15% had 9-year compulsory schooling). Cortisol data were derived from an external reference group, comprising blue- and white-collar daytime workers in various professions and companies (n = 174), with similar distributions of age (mean 47.5; SD 10.2 years) and gender (70% women).

Neuropsychological tests

The tests were chosen from a larger set with available reference data, and the final selection was based on a demonstrated high sensitivity for subtle brain dysfunction (Österberg et al. 2000; Link et al. 2006), while covering basic cognitive domains and limiting the test session to approximately 1 h (due to other examinations). Each subject was tested individually in the afternoon by one of three experienced psychologists, two of whom had also examined the reference subjects in the previous studies (Österberg et al. 2000, 2002). Manual scoring was cross-checked by a colleague. The tests were given in the following sequence:

1. the Cronholm–Molander verbal memory test (Cronholm and Molander 1957), an associative learning task comprising immediate and delayed recall sections;

2. the Digit Symbol test from the Wechsler Adult Intelligence Scale-Revised (WAIS-R) (Wechsler 1981), a test of perceptual and fine motor speed;

3. the WAIS-R NI incidental learning (Kaplan et al. 1991), a test of memory for the digit–symbol combinations used in the preceding Digit Symbol test. The modified version used showed only the stimulus key row with the symbols blanked out (scoring range 0–9);

4. the k-test from the Automated Psychological Test System (APT) (Levander and Elithorn 1987), a test of sustained attention. The k-test requires visual scanning of computer screen images, containing a scatter of 10 letters, for the presence/absence of a critical stimulus (k). Around 100 stimulus sets were shown during 5 min. The task is to respond correctly and quickly to each stimulus set by pressing the corresponding keys (“k” vs. “no k”). Results were expressed: (a) according to signal detection theory, as d' (accuracy) and β (balance between missed hits to targets and false hits to non-targets; a negative β indicating that missed hits were more common than false hits); (b) as reaction time (RT) level in relation to stimulus laterality (the mean of single RT responses for correct hits, left-center-right); (c) as error ratio (proportion of missed and false hits) and (d) as search strategy (the RT ratio for correct yes/no responses with “k” present/absent on a scale from sequential (obsessive) to global (impulsive), a higher value indicating a more global strategy);

5. the Austin Maze test with the Milner pathway (Milner 1965; Walsh 1985), a test of spatial learning, strategy, and speed in a modified computer version (Österberg et al. 2000). Ten trials were used, and error and performance time refer to the nine memory trials;

6. the SRB:1 vocabulary test from the DS-battery (Dureman et al. 1971), a verbal knowledge or intelligence test containing 30 items of multiple-choice type.

The test session ended with the delayed recall section of the Cronholm–Molander test.

Self-ratings of cognitive problems

The memory and concentration scale of the Euroquest (Chouanière et al. 1997) was used to assess subjective cognitive problems. Six items address memory problems, and three items problems with attention and concentration. The 4-point (1–4) response range has the labels “seldom/never”, “sometimes”, “often”, and “very often”. The scale has been validated as a sensitive measure of slight cognitive impairment (Karlson et al. 2000; Carter et al. 2002).

Salivary cortisol

The participants collected saliva in sampling tubes with cotton swabs (burnout group) or polyester swabs (the external reference group) (Salivette^®; Sarstedt Ltd., Leicester, UK) on weekdays at awakening, 30 min later, and around 9 p.m. in the evening. The burnout group also carried out a DST, by ingesting 0.5 mg dexamethasone (Decadron^®; MSD, Whitehouse Station, NJ, USA) at bedtime and taking two further saliva samples the following morning. Oral and written instructions emphasized that swabs should be kept in the mouth until thoroughly hydrated ( ≤ 5 min), avoiding smoking and eating heavy meals 1 h prior to sampling. To avoid contamination due to microbleeding from the gum, tooth brushing and eating were not allowed until after the morning samples. Samples were stored in refrigerator until they were returned by mail, and then frozen until analysis. Determination of cortisol in saliva was carried out with a competitive radioimmunoassay (RIA), designed for quantitative in vitro measurement of cortisol in serum, plasma, urine, and saliva (Spectria Cortisol Coated Tube RIA, Orion Diagnostica, Espoo, Finland), according to the manufacturer's specifications. The sample volume was 150 μl, the range of the standard solutions prepared was 1.0–100.0 nmol/L, and the incubation time was 30 min at 37°C. The specifications given by the manufacturer were a sensitivity of twice the SD of the zero binding value in saliva (0.8 nmol/L), a bias of 110% (103–115%), an intra-assay variation of 5.4%, and an inter-assay variation of 7.3%. The cross-reactivity to cortisone was < 0.2%. A 1470 Wizard gamma counter (Wallac, Turku, Finland) was used for measurement of radioactivity. A method evaluation of certified reference material in water, performed by our laboratory, showed no bias of the method, recovery being 97% (95% confidence interval: 94.0–100.9). Limit of detection (LOD) was 1.59 nmol/L. Between-run coefficients of variation (CVs) were 19% at 11.5 nmol/L and 16% at 49.2 nmol/L (Garde et al. 2003). To show equivalence between different runs, natural saliva samples at two concentrations (5.9 and 18.5 nmol/L for cotton swabs, and 5.9 and 49.2 nmol/L for polyester swabs) were used as control materials and analyzed together with the samples. Westgard control charts were used to document that the analytical method remained under analytical and statistical control, in other words, that the trueness and precision of the analytical methods remained stable (Westgard et al. 1981). The performance of the methods has been validated by inter-laboratory comparison schemes (Garde et al. 2003; Hansen et al. 2003). Due to an empirically verified variation in cortisol values depending on the swab material (cotton vs. polyester) and the version of the Spectria RIA kit (new vs. old), the raw cortisol values of the external reference group (Cr) was corrected to correspond with the values of the burnout group (C) by the formula:

In addition to the raw cortisol values, the following indices were used:

1. the cortisol awakening response (CAR), defined as the increase in cortisol concentration from the awakening sample to the +30 min sample, expressed in absolute (nmol/L) and proportional (%) changes;

2. the cortisol morning peak (CMP), defined as the highest concentration of cortisol in saliva found in either of the two morning samples;

3. the cortisol range (CR), defined as the decrease in cortisol concentration from the CMP to the 9 p.m. evening sample; and

4. the DST response, defined as the proportional (%) decrease in cortisol concentration after ingestion of dexamethasone, in comparison with the baseline weekday morning values.

Data analysis

The statistical analysis was performed using SPSS, version 15.0 (SPSS, Inc., Chicago, IL, USA). Visual inspection showed that neuropsychological test scores and Euroquest scores were approximately normally distributed, except for a positively skewed distribution for the APT k-test error ratio, necessitating ranking prior to statistical analysis. The cortisol raw values from the sampling points, and the CMP variable, had positively skewed distributions, so they were log-transformed (ln). The CR, CAR, and DST response variables were ranked.

To provide a useful summary measure of cognitive performance (see Link et al. 2006), an overall outcome variable for all tests (Z_tot) was computed. Each subject's raw score on each test was transformed to a standardized score, based on the z-distribution in the reference group (i.e., each test variable in the reference group had a mean z-score of 0 (zero) and a SD of 1.0). To give each test an equal impact on the Z_tot, the scores on tests with several outcome variables (e.g., the APT k-test) were computed as the mean of the z-score of the separate variables within the test. The APT k-test results were truncated to the mean z-score of the two core variables d-prime and total RT (i.e., the mean z-score of the three RT variables). WAIS-R NI incidental learning was excluded since male reference data were unavailable. Each subject's z-scores across the final five test variables (one per test) were summarized in a total mean z-score (Z_tot), expressing the overall cognitive performance.

Between-group comparisons of cortisol, test, and questionnaire data were carried out using univariate analysis of variance (ANOVA) in the general linear model of SPSS, corrected for the covariates age, gender, medication with psychopharmacological drugs, SCL-90 depression score, and education level (only for neuropsychological test scores). Initially all covariates were entered, and then removed one at a time (based on highest p-value), until only covariates with p < 0.20 remained (Mickey and Greenland 1989). In case gender remained a covariate in the final model, the interaction Group by Gender was explored in order to detect a differential effect of gender across groups. In case of statistical significance for the factor Group, the effect size is given as partial eta square (η_p²). The within-group comparison of DST cortisol concentrations, compared with the normal weekday values, was carried out using the general linear model repeated measures analysis in a two-factorial model (day; time point) and the same set of covariates. Associations between variables were determined with Pearson's partial correlation, controlling for age and gender as a baseline, and, when justified, for remaining covariates. p-values < 0.05 were considered statistically significant.

Ethics

All subjects gave written informed consent to participate in the study. The study protocol was approved by the Ethics Committee of Lund University (LU 784-03).

Results

Neuropsychological tests

In tests of sustained attention and processing speed, the burnout group performed significantly lower on the WAIS-R Digit Symbol (F[1,124] = 9.6, p = 0.002, η_p² = 0.072), while the APT k-test scores were similar across groups (Table II). Group differences in favor of the burnout group emerged in the Cronholm–Molander verbal memory test (immediate: F[1,125] = 6.71, p = 0.011, η_p² = 0.051; delayed: F[1,125] = 5.50, p = 0.021, η_p² = 0.042) and in Austin Maze performance time (F[1,123] = 4.52, p = 0.036, η_p² = 0.035), after correction for depression score (higher depression score was related to better performance in the burnout group). Group comparisons of raw scores uncorrected for depression score did, however, not show any group difference in the latter variables (Cronholm–Molander, immediate/delayed: p = 0.22/0.44; Austin Maze performance time: p = 0.75). Remaining memory variables, Austin Maze errors and WAIS-R NI incidental learning score, did not show any group difference, nor did SRB:1 vocabulary or Z_tot (burnout: mean − 0.11 (SD 0.57); referents: mean 0 (SD 0.56); F[1,124] = 1.36, p = 0.25; corrected for age, education and gender). The interaction Group by Gender, explored for Austin Maze, WAIS-R Digit Symbol, APT k-test β and RT, did not reveal any differential effect of gender across groups.

Table II.  Neuropsychological test raw scores for patients with burnout disorder and healthy referents.

Cognitive domain and test	Patients (n = 65)		Referents (n = 65)		ANOVA, Group
	Mean	SD	Mean	SD	F^*	p^†	ηp^2
Verbal intellectual ability
SRB:1 vocabulary	24.2	3.7	24.2	3.8	1.96	0.16^D,E	–
Memory
Cronholm-Molander verbal memory, immediate recall	20.9	4.4	21.9	4.8	6.71	0.011^D,E	0.051
Cronholm-Molander verbal memory, delayed recall	17.2	5.1	17.9	5.4	5.50	0.021^D,E	0.042
Austin maze, total errors^‡	38.1	22.7	39.3	19.9	0.18	0.67^A,G	–
Austin maze, total performance time (s)^‡	308	114	316	128	4.52	0.036^A,D,E,G	0.035
Incidental learning WAIS-R NI, number correct^¶	6.8	2.1	6.6	1.6	0.10	0.75^A	–
Sustained attention and processing speed
WAIS-R Digit Symbol	53.1	11.1	58.4	11.1	9.61	0.002^A,E,G	0.072
APT k-test,^§ d'	4.10	0.45	4.01	0.52	1.07	0.30^A	–
APT k-test, β^‖	− 0.62	0.70	− 0.72	0.65	0.82	0.37^G	–
APT k-test, RT level for correct responses, left part of screen (ms)^‡	1138	275	1111	236	0.04	0.84^A,G,P	–
APT k-test, RT level for correct responses, center part of screen (ms)^‡	993	268	961	193	0.03	0.85^G,P	–
APT k-test, RT level for correct responses, right part of screen (ms)^‡	1242	308	1266	282	0.27	0.61^A,E,G	–
APT k-test, search strategy^#	0.56	0.10	0.59	0.09	3.21	0.075	–
APT k-test, error ratio (%)^‡	2.2	2.6	2.8	3.1	1.33	0.25	–

^* df ranged from [1,123] to [1,127], except for Incidental learning WAIS-R NI [1,89]

^† Univariate ANOVA p-values for between-group comparisons, corrected for age (A), Symptom Checklist-90 depression score (D), level of education (E), gender (G), and medication with psychopharmacological drugs (P) in case p < 0.20 for each covariate

^‡ A lower score signifies better performance

^¶ Reference data were available only for women; consequently, comparison data is given for women only (n = 46 per group, the covariate gender excluded)

^§ n = 64 in the patient group; for one case, the k-test data was excluded due to the patient misunderstanding the instructions

^‖ Qualitative variable; a negative β indicates that false rejection (missed ‘k's) was more common than false alarms (alarms when k not present)

^# Qualitative variable; a lower value indicates a more sequential search strategy, while a higher value indicates a more global search strategy.

Self-ratings of cognitive problems

The Euroquest scores were substantially elevated in the burnout group, showing a mean total score of 2.91 (SD 0.62) vs. 1.78 (SD 0.49) among referents (F[1,124] = 28.6, p < 0.0001, η_p² = 0.19). The concentration and attention problems subscale showed a more conspicuously elevated score compared with referents (2.93 (SD 0.69) vs. 1.55 (SD 0.44), F[1,125] = 55.9, p < 0.0001, η_p² = 0.31) than did the scale for memory problems (2.90 (SD 0.65) vs. 1.89 (SD 0.58); F[1,125] = 15.7, p = 0.0001, η_p² = 0.11). The interaction Group by Gender was significant for total score (F[1,123] = 4.8, p = 0.030, η_p² = 0.04) and memory problems (F[1,124] = 4.2, p = 0.041, η_p² = 0.03), indicating larger differences between burnout patients and referents among men than among women (1.32 vs. 1.05). However, within the burnout group, none of the Euroquest subscales, or total score, showed a statistically significant relationship with any of the cognitive test parameters shown in Table II, nor with Z_tot. However, SCL-90 depression score correlated with Euroquest total score (r = 0.30, p = 0.02), but not with Z_tot (r = 0.09, p = 0.50).

Associations with salivary cortisol concentration

Diurnal cortisol concentration changes

Analyzable cortisol samples across all three time points were obtained from 46 burnout subjects, while nine and two subjects had missing data for one and two time points, respectively, due to insufficient amounts of saliva (dry swabs). No data were obtained from eight subjects, of whom seven had not returned any samples and one had invariably dry swabs. The raw values in the burnout and reference groups were similar at the awakening and +30 min measurements, but the evening values were lower in the burnout group (F[1,225] = 28.05, p < 0.001, η_p² = 0.111) (Figure 1, Table III). The CAR, CMP, and CR did not differ between groups. For the CAR, the interaction Group by Gender did not reveal any differential effect of gender across groups. Within the burnout group, neither absolute nor proportional CAR was statistically significantly related to any of the neuropsychological measures (Table II) or to Z_tot. However, the CMP and CR values were related to the APT k-test reaction times for left, central, and right stimuli (CMP: range − 0.37 to − 0.32, p = 0.007–0.02, n = 55; CR: range 0.30–0.33, p = 0.02–0.04, n = 53), which relationships mostly remained after controlling for depression score and medication with psychopharmacological drugs (CMP: range − 0.35 to − 0.29, p = 0.012–0.037; CR: 0.27, 0.32, and 0.30 for left, center and right visual field, respectively, p = 0.064, p = 0.026, and p = 0.038). The CR was also related to the WAIS-R Digit Symbol score (r = 0.29, p = 0.04, n = 53) and to Austin Maze total performance time (r = 0.29, p = 0.03, n = 53). After controlling for depression score and psychopharmacological drug use the relationship remained for WAIS-R Digit Symbol (r = 0.30, p = 0.034), but only bordered on significance for Austin Maze total performance time (r = 0.27, p = 0.059). Thus, a smaller decline in cortisol from morning to evening was associated with slower performance in these three tests. To illustrate this relationship, the burnout group was divided into halves based on a median split of the mean z-score of the three APT k-test RT variables (Figure 2). This indicated that the primary cause of the relationship was a higher cortisol level at +30 min in the high-performing subgroup (F[1,46] = 12.9, p = 0.0008, η_p² = 0.22, corrected for gender). None of the cortisol measures was related to the Euroquest scores of subjective cognitive problems.

Figure 1 Salivary cortisol concentrations (medians and quartiles) among patients with burnout (broken line with triangles; n = 50–55) and reference persons (solid line with squares; n = 50–55), at awakening, 30 min after awakening, and at 9 p.m., on ordinary weekdays, and among burnout patients after dexamethasone ingestion (broken line with circles; n = 41). Statistical significance for group differences (ordinary weekdays) in univariate ANOVA: *p < 0.001; n.s. p>0.05.

Table III.  Diurnal salivary cortisol concentrations and computed index values (medians and quartiles) on a normal weekday among patients with burnout disorder and healthy referents, and after dexamethasone treatment among patients with burnout disorder.

Cortisol measure (values in nmol/L unless otherwise stated; n refers to number of participants in the burnout group with valid data)	Patients		Referents (n = 174)		ANOVA, Group
	Md	Q1/Q3	Md	Q1/Q3	F^*	p^*	ηp^2
Raw values
At awakening (n = 53)^‡	9.2	6.1/13.2	9.8	7.0/13.1	0.21	0.65	–
30 min after awakening (n = 50)^¶	14.4	9.7/21.6	14.1	10.1/18.3	0.09	0.77	–
At 9 p.m. (n = 55)^‖	1.3	1.0/1.7	2.1	1.6/3.1	28.05	< 0.001^D	0.111
After dexamethasone treatment (n = 41)
• at awakening^‖	0.8	0.4/1.1	–	–	–	–	–
• 30 min after awakening^∥	0.7	0.5/1.2	–	–	–	–	–
Indexes
Cortisol awakening response (CAR; n = 47)^§,^#
• absolute increase	4.2	− 0.9/8.7	4.4	− 0.3/8.4	0.09	0.77^A,G	–
• proportional increase (%)	53.5	− 6.1/112.6	49.2	− 2.3/111.5	0.04	0.84^G	–
Cortisol morning peak value (CMP; n = 56)	14.2	8.8/21.7	14.8	11.6/19.0	1.27	0.26	–
Cortisol decrease from morning (peak value) to evening (CR; n = 54)	− 12.6	− 19.0/ − 6.8	− 12.1	− 16.8/ − 9.0	2.22	0.14^D	–
Proportional cortisol decrease after dexamethasone treatment (n = 41) compared with the corresponding time points on a normal weekday
• at awakening (%)	− 90.9	− 95.1/ − 86.2	–	–	–	–	–
• 30 min after awakening (%)	− 93.9	− 96.9/ − 90.0	–	–	–	–	–

^* df ranged from [1,285] to [1,295]

^† Univariate ANOVA p-values for between-group comparisons, corrected for age (A), Symptom Checklist-90 depression score (D), gender (G), and medication with psychopharmacological drugs (P) in case p < 0.20 for each covariate

^‡ The median and Q1/Q3 values for the subsample of 41 subjects included in the DST analysis were identical to the values for the total burnout group

^¶ The values for the subsample of 41 burnout subjects included in the DST analysis were Md = 14.7; Q1/Q3 = 6.2/21.8

^§ Computed as the increase from the awakening value to the value 30 min after awakening

^# A negative CAR was found among 28% (n = 13/47) of the burnout subjects and among 26% (n = 46/174) of the referents

^‖ Most cortisol concentrations in the burnout group were below the detection limit (1.59 nmol/L) and are thus approximate.

Figure 2 Salivary cortisol concentrations (medians and quartiles) among patients with burnout, divided into group halves based on a median split of the z-score reaction time performance in the APT k-test (the subgroup performing above median shown as broken line with triangles, and below median as solid line with squares), at awakening (n = 27/25), 30 min after awakening (n = 23/26), and at 9 p.m. (n = 26/28). Statistical significance for group differences in univariate ANOVA: *p < 0.001; n.s. p>0.05.

DST response

Analyzable morning cortisol samples across both days were obtained from 41 burnout subjects. Besides the seven subjects that returned no samples, all missing data was due to insufficient amounts of saliva (dry swabs). Repeated measures analysis of raw values showed a strong main effect of day (F[1,40] = 481, p < 0.0001, η_p² = 0.92), indicating considerably lower DST morning cortisol levels (Figure 1 and Table III). All burnout subjects (n = 41) showed a substantial cortisol concentration decrease (≧67%) in response to the DST. A significant main effect of time point (F[1,40] = 12.1, p = 0.001, η_p² = 0.23) was observed, indicating a higher concentration at the +30 min time point, but the significant interaction day * time point (F[1,40] = 4.1, p = 0.047, η_p² = 0.09) indicated that this awakening response occurred only on the baseline weekday (F[1,39] = 9.5, p = 0.004, η_p² = 0.20), not on the DST day (F[1,40] = 0.8, p = 0.37). In none of the above analyses were the covariates depression score and psychopharmacological drug use related (p>0.20) to the outcome, and were thus not included in the final model. The sole relationship found between the DST decrease and the neuropsychological test variables was that a stronger DST response at the awakening measurement correlated with a higher WAIS-R-NI incidental learning score (r = − 0.38, p = 0.02, n = 41, controlling for all five covariates). None of the neuropsychological test variables, including Z_tot, showed any relationship with the decrease at +30 min. Nor were the Euroquest scores related to any of the DST response variables.

Discussion

This study showed that persons with work-related burnout had mostly normal neuropsychological test scores on tests tapping cognitive domains vulnerable to mild cognitive impairment, such as short-term verbal and spatial memory, sustained attention, and cognitive speed. The only exception was a slightly lower speed in one perceptual and motor test, the WAIS-R Digit Symbol, corresponding to − 0.5 Z. However, in the APT k-test, posing similar demands on attention and processing speed, no group difference was found. Moreover, the summary index, Z_tot, was equal in the burnout and referent groups. After correction for self-rated depressive symptoms (SCL-90), the burnout group actually showed a statistically significantly higher performance in the verbal memory test and the speed variable of the spatial learning test, due to a trend for higher performance among burnout cases with higher depression score. The latter observation is probably best regarded as a statistical artifact, since scores uncorrected for depression did not show a group difference on memory tests or any other test variable, besides the WAIS-R Digit Symbol score. Nor were any differential effects of gender across groups seen on the outcome of the cognitive tests.

The Euroquest ratings of memory and attention/concentration problems were substantially higher in the burnout group than among referents, and men showed a more pronounced elevation than women did. Within the burnout group, the extent of subjective cognitive complaints was, however, unrelated to all scores and the Z_tot. This may suggest that a depressive and negative self-perception is a major determinant in the reporting of subjective cognitive problems, a notion supported by the observed correlation between SCL-90 depression score and the Euroquest score. Another possible explanation, based on our clinical impressions, is that being on sick leave due to burnout for the first time gives rise to worries about future health and career, which leads to distraction in everyday situations, perceived as cognitive impairment.

Concerning cortisol, the overall diurnal profile of the burnout group was essentially normal, as reflected by the similar values of the morning samples and the CAR, CMP, and CR indexes among burnout subjects and referents. However, the burnout group showed a significantly lower evening value. It might reflect that the burnout cases, of whom the vast majority was on sick leave to some extent, were more relaxed in the evening due to lower everyday demands. Among possible relationships between cortisol and cognitive performance, only the maximum morning concentration and the highly interrelated CR variables showed small, but solid, relationships with the speed measures of the APT k-test, the Austin Maze test, and the WAIS-R Digit symbol test (Figure 2). This suggests that a more flexible HPA axis may be related to speedier test performance. It might also indicate that the burnout subgroup with the highest CR are in a more active or early state of burnout, in which elevated levels of cortisol and catecholamines still uphold vigilance and speed, while the subgroup with lower CR have entered into a more dysfunctional state, characterized by a blunted diurnal rhythm and a less adaptive response to everyday stressors such as challenging cognitive tasks (McEwen 1998). The relationship between CR and cognitive speed may however reflect a general phenomenon not restricted to burnout; an issue that we were unable to address because reference data for cortisol and test scores were not available from the same reference group. The substantial decrease in morning concentrations in response to the DST found in all cases of burnout, ranging from 67% to nearly 100%, does however indicate that the basic HPA axis flexibility was normal (Mommersteeg et al. 2006). The DST response was also studied in relation to test performance, which did not reveal any convincing relationship; the weak correlation found between the WAIS-R NI incidental learning score and the DST decrease at awakening was not corroborated by data from the second morning measurement, which suggests a spurious finding due to the large number of correlations carried out. Moreover, the extent of subjective cognitive problems was unrelated to all cortisol measures.

The observed difference in psychological test scores between burnout cases and healthy referents seems small in relation to the findings in previous comparable studies of work-related burnout. Sandström et al. (2005) compared 67 female patients on sick leave with 15 matched referents, and found a generally lower performance in a complex attention test, the integrated visual and auditory continuous performance test (IVA), and in the Rey Complex Figure spatial memory test, but no difference in the full-scale WAIS-R. Van der Linden et al. (2005) studied 13 burnout patients on sick leave, 16 subjects in active work scoring high in a burnout questionnaire, and 14 referents. The burnout subjects showed unstable performance and impulsive errors in both attention tests used, and a strong inverse correlation between attention and subjective cognitive problems was found. Rydmark et al. (2006) compared 29 female patients on sick leave for job-induced depression or maladaptive stress reaction with 28 matched referents. The patients showed poorer attention in a complex RT task and lower performance in a working memory task (digit span). However, they showed normal performance across all three tests of declarative memory. Öhman et al. (2007) compared 19 patients on sick leave with 19 matched referents. The patients showed a slower rate of verbal learning and had signs of impaired executive functioning, such as poorer attention and processing speed (WAIS-R Digit Symbol), lower verbal fluency, slower performance in a working memory task, poorer prospective memory, and a disproportional reduction in verbal memory during distraction. The general trend across these studies support the notion that problems with executive functioning, such as reductions in sustained attention, working memory, and flexibility, and increased distractibility, are discernible in burnout. In contrast, the results of the present study only give partial and unconvincing support to this hypothesis, through the somewhat lower performance in the WAIS-R Digit Symbol test, which seemed more sensitive to the effects of burnout than the tests of declarative memory (e.g., the Cronholm–Molander and Austin Maze tests). Still, the modest cognitive impairment found in the previous studies of work-related burnout is hardly surprising in view of the general picture of mild impairment in attention and immediate memory observed even in cases of PTSD after prolonged and severely traumatizing events, such as war combat or prisoner of war internment (Horner and Hamner 2002).

There are several limitations to this study. First, many potential participants declined participating in the study when initially contacted by phone, often on the grounds that their condition had substantially improved during a short sick leave. Only a few claimed to be too ill to participate, but on the other hand we only invited cases with recent sick leave, without recurrent burnout episodes. Although we find it unlikely that that the participants were healthier and in better spirits than the total population on sick leave for work-related burnout, it can hardly be excluded as a possible reason for the mostly normal test scores.

Second, the neuropsychological test battery we used was not as extensive as that used by Sandström et al. (2005) and Öhman et al. (2007), though it was in fact larger than the ones used by Van der Linden et al. (2005) and Rydmark et al. (2006). A more extensive neuropsychological examination might have revealed reduced performance in further tests besides the WAIS-R Digit Symbol test. It could also be argued that the tests used in the present study may have been insufficiently complex to enable detection of slight impairment in executive functioning, related to stress-induced dysfunction of prefrontal brain areas (Koenen et al. 2001). However, there is reason to assume that several of the present tests have a fair sensitivity to prefrontal dysfunction; the APT k-test requires fast decision making, continuous self-monitoring and cautiousness, the Austin Maze test involves a strategy component, the WAIS-R NI incidental learning test is largely dependent on working memory, and the WAIS-R Digit Symbol test requires speed, flexible perception, and fine motor control, all of which are capabilities reflecting aspects of executive functioning.

The baseline comparability of the burnout and reference groups may also be questioned, in view of the fact that many patients described a very high job performance for several years prior to the sick leave. However, the vocabulary test SRB:1, intended as a proxy measure of premorbid intellectual level, gave very similar scores in both groups, indicating that the matching for age, gender, and educational level was indeed successful. Öhman et al. (2007) and Rydmark et al. (2006) used the SRB:1 with similar results, and Sandström et al. (2005) showed the same using the WAIS-R vocabulary. It is reasonable to assume that the extraordinary job performance preceding burnout merely reflects a short-term ability to perform exceedingly well, which, however, is unsustainable in the long run.

One may also criticize the fact that cases were included who, in addition to burnout, met the formal criteria for depression or anxiety disorder of the Prime-MD (Spitzer et al. 1994). The inclusion of these cases was motivated by the finding that symptoms of depression and anxiety are considered intrinsic parts of the burnout syndrome, and that burnout can hardly be reduced to only these psychiatric disorders (Iacovides et al. 2003). Depressive symptoms and use of psychopharmacological drugs were explored as covariates in the within- and between-group comparisons of test scores and cortisol, which showed that they were mostly unrelated to outcome and did not reduce any group difference.

Concerning cortisol, it would have been preferable to collect data not only on ordinary weekdays, but also, before and during the neurobehavioral testing. This might have shed further light on the possible impact of transient changes in cortisol on test performance, for example, whether a blunted stress response to the test situation could be related to lower cognitive performance (Fries et al. 2005). In order to increase reliability, cortisol measurements on several weekdays would also have been preferable, but this was considered likely to jeopardize participation. Moreover, morning measurements seem to be reasonably stable for individuals across days (Clow et al. 2004).

To conclude, despite considerable subjective cognitive problems, the burnout patients performed as well as referents did across a number of neuropsychological tests, apart from a slightly slower performance in a single speed test. Moreover, their diurnal cortisol profile and DST response was essentially normal, suggesting a preserved basic flexibility in the HPA axis. The degree of subjective cognitive problems was unrelated to the actual performance in any test, but a weak relation was found between CMP and speed in a few tests of attention and learning, suggesting that a flatter diurnal cortisol profile is associated with lower cognitive speed. It is tempting to speculate that the diurnal cortisol profile found in the present group of patients represents normalization from sustained higher levels during the years preceding sick leave. If so, the neuroendocrine feedback may have been recalibrated to operate efficiently at high cortisol levels, i.e. during allostatic overload Type 2 (McEwen and Wingfield 2003), which might explain why the return to more normal diurnal cortisol levels is perceived in terms of listlessness, fatigue, and cognitive problems. Could this be one of the reasons for the marked discrepancy between subjective cognitive problems and neurobehavioral test performance?

Acknowledgments

This study was supported by grants from the Swedish Research Council (2003-5724) and the Swedish Council for Working Life and Social Research (2003-0765). Palle Ørbæk, M.D., Ph.D., is acknowledged for financial support through the grant from the Swedish Research Council. Thanks are due to Birgitta Pålsson, B.Sc., Gunnel Åbjörnsson, B.Sc., Lars Havewald, M.Sc., Birgitta Malmberg, M.D., and Britt Larsson, M.D., Ph.D., for medical assessments and data collection, and to Lisbeth Prahl, for administrative work.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.