Physical activity buffers fatigue only under low chronic stress

Abstract

Fatigue is one of the most commonly reported complaints in the general population. As physical activity (PA) has been shown to have beneficial effects, we hypothesized that everyday life PA improves fatigue. Thirty-three healthy students (21 women, 22.8 ± 3.3 years, 21.7 ± 2.3 kg/m²) completed two ambulatory assessment periods. During five days at the beginning of the semester (control condition) and five days during final examination preparation (examination condition), participants repeatedly reported on general fatigue (awakening, 10 am, 2 pm, 6 pm and 9 pm) by means of an electronic diary, collected saliva samples for the assessment of cortisol and α-amylase immediately after providing information on fatigue and wore a triaxial accelerometer to continuously record PA. Self-perceived chronic stress was assessed as a moderator. Using hierarchical linear modeling, including PA, condition (control vs. examination), sex and chronic stress as predictors, PA level during the 15 min prior to data entry did not predict momentary fatigue level. Furthermore, there was no effect of condition. However, a significant cross-level interaction of perceived chronic stress with PA was observed. In fact, the (negative) relationship between PA and fatigue was stronger in those participants with less chronic stress. Neither cortisol nor α-amylase was significantly related to physical activity or fatigue. Our study showed an immediate short-term buffering effect of everyday life PA on general fatigue, but only when experiencing lower chronic stress. There seems to be no short-term benefit of PA in the face of higher chronic stress. These findings highlight the importance of considering chronic stress when evaluating the effectiveness of PA interventions in different target populations, in particular among chronically stressed and fatigued subjects.

Keywords:

• Accelerometry
• chronic stress
• electronic momentary assessment;fatigue
• physical activity
• student sample
• salivary cortisol
• salivary α-amylase

Introduction

Fatigue is one of the most commonly reported complaints in the general population (Lewis & Wessely, 1992). About 10% of young and middle-aged adults in different settings report severe levels of fatigue (Akerstedt et al., 2004; Wessely et al., 1997) with detrimental effects on function and ability to work, cognitive ability, and emotional well-being. However, there is still no generally accepted definition of ‘fatigue’ but most researchers describe fatigue as a multidimensional construct that is largely subjective in nature (for an overview of assessment strategies see Dittner et al., 2004). Accordingly, we define fatigue as a subjective state that goes beyond feeling tired and sleepy, cannot be predicted by effort exerted, and interferes with carrying out duties and responsibilities. Stress, especially in the case of chronic stress, seems to play a major role in the development and persistence of this debilitating symptom in various populations (Kurokawa et al., 2011; Pawlikowska et al., 1994). The high prevalence of fatigue and related consequences such as increased sick leave or health care use (Akerstedt et al., 2007; McCrone et al., 2003) highlights the need to intensify research into intervention strategies that may help to ease the burden of fatigue and thus promote health and well-being.

Results from epidemiological and experimental studies point to a mood-lifting and stress-relieving effect of physical activity (PA; for a review Penedo & Dahn, 2005). PA is defined as any bodily movement that results in energy expenditure – as opposed to exercise, which is defined as planned and structured movements that are conducted to increase physical fitness (Caspersen et al., 1985). Research on the effects of PA in everyday life, that is, under ecologically valid conditions, on affective states is still in its infancy (Bossmann et al., 2013; Dunton et al., 2011; Kanning et al., 2012; Schwerdtfeger et al., 2008), and there is even less knowledge about the potential effects of PA on bodily complaints such as fatigue. In patients with breast cancer, momentary fatigue ratings correlated negatively, but not significantly, with objectively assessed PA during a 24-h period (Grossman et al., 2008). In line with this finding, healthy subjects reported feeling more energized and awake after self-reported (Kanning & Schlicht, 2010) as well as objectively assessed (Kanning et al., 2012) PA. However, possible moderators of this relationship have barely been investigated. In this regard, the “stress-buffering hypothesis of physical activity” suggests that PA may function as a buffer against the detrimental effects of stress on subjective well-being (Gerber & Puhse, 2009). Prospective studies showed that higher self-reported exercise levels buffered the negative effects of stressful life events and chronic stress experiences on mental health (Brown & Siegel, 1988; Lindwall et al., 2014). In particular, this stress-buffering effect was explained by moderate exercise, as compared to vigorous exercise (Gerber et al., 2010). Exercise intervention studies confirm this notion of physical fitness buffering individuals from stress responses (Klaperski et al., 2014). Whether this assumption might also be applied to everyday life PA remains to be shown.

We therefore conducted an ecological momentary assessment (EMA) study to explore the association between current PA in everyday life and general fatigue and the moderating effect of chronic stress experiences in the last three months in a sample of healthy university students. As there is first evidence for a buffering effect of PA on positive affect only during periods of high demand (Giacobbi et al., 2007), data collection was conducted during two periods, characterized by different stress levels: five days of a less stressful regular week during the semester, and five days during a stressful period in which students were preparing for their final examinations (Weekes et al., 2006). Moreover, we also took into account the moderating effect of chronic (last three months) stress. To ensure compliance, fatigue was measured by means of an electronic diary. Since there are substantial discrepancies between subjective reports of PA and objective assessment strategies (Adamo et al., 2009; Prince et al., 2008), we measured activity objectively by means of accelerometry in order to avoid memory bias.

Current PA was assumed to have an immediate positive effect on general fatigue. We hypothesized that this effect is moderated by perceived chronic stress, with those participants reporting higher chronic stress benefitting more from short-term PA than those with lower chronic stress. Furthermore, salivary cortisol and α-amylase (sAA) have been assessed as markers of the hypothalamic–pituitary–adrenal (HPA) axis and the autonomic nervous system (ANS), respectively. Both markers have been linked to fatigue (Powell et al., 2013; Yamaguchi et al., 2006) and physical activity (Filaire et al., 2013) and will therefore be considered as likely mediators of an immediate short-term PA effect on momentary general fatigue (Silverman & Deuster, 2014).

Methods

The present analysis is based on data from a larger study conducted at the University of Marburg, Germany. For a detailed description of recruitment procedures and exclusion criteria, Doerr et al. (2015) and Linnemann et al. (2015). The current report is based on a subset of 33 randomly chosen participants (21 females, 22.8 ± 3.3 years, 21.7 ± 2.3 kg/m²) from whom accelerometry data were obtained. Participants received either 50 Euros or course credit for participation. The study protocol was approved by the local Ethics Committee of the Faculty of Psychology at the University of Marburg, Germany, and all participants provided written informed consent.

Data collection took place during two weeks of the summer semester 2012, the first period at the beginning of the semester (control condition) and the second period covering students’ preparation for their final examinations (examination condition). During both ambulatory assessment periods, students were examined on five consecutive days. This design enabled us to study effects of PA on fatigue during two conditions of differing stressfulness (Doerr et al., 2015).

In the introductory session, participants were trained in handling a preprogramed electronic diary (iPod^® touch; iDialogPad, G. Mutz, University of Cologne, Germany) to report subjective general fatigue levels five times throughout the day (awakening, 10 am, 2 pm, 6 pm and 9 pm). A sixth data entry 30 min after awakening comprised only control items (not reported). Momentary general fatigue was assessed using the item “At the moment, … I feel fatigued” on a five-point Likert scale from not at all to very (Stone et al., 1997). During the trial run, all participants have been given the opportunity to ask questions, and open issues were clarified to make sure that everything was understood correctly.

At the end of the introductory session, participants were equipped with a triaxial accelerometer attached to their nondominant wrist (Somnowatch, Somnomedics, Randersacker, Germany). This validated device (Dick et al., 2010) continuously recorded movement counts every one-second interval for the whole duration of each condition, that is, 5 days, with the signal sampled at 32 Hz with 12-bit analog-to-digital converter. In cases of taking off the watch (e.g. when bathing, swimming), subjects were asked to indicate such events by pressing the event marker on the watch. As a measure of current activity level, we calculated the number of activity counts per minute and summed up activity counts from the 15 min prior to data entries (time stamps indicated by iDialogPad software). This was in accordance with previously reported analytical strategies showing averaged 15 min everyday-life activity to be the best approach for the prediction of mood during ambulatory monitoring (Schwerdtfeger et al., 2008). The first iPod® entry directly after awakening was not taken into consideration, leaving four daily measurement time points and 1320 possible “physical activity” data points in total.

Chronic stress was examined by means of the 12-item Screening Scale of the Trier Inventory for the Assessment of Chronic Stress (SSCS; Schulz et al., 2004), which assesses chronic stress experiences in the past three months on a 5-point Likert scale ranging from never to very often. This scale was completed during the control condition to capture students’ general chronic stress.

In addition to accelerometer-derived PA, participants answered two items that differentiate between those who are more or less active. The first item asked for physical activity in daily life (walking, cycling to work, etc.) over the past 12 months with the response categories “none” – “a few times” – “several times/week” – “every day/almost every day.” This was complemented by asking for exercise beyond physical activity in daily life over the past 12 months with the response categories “hardly anything” – “light activity at least once a week” – “moderate activity at least once a week” – “vigorous activity on a regular basis.” Following Leijon et al. (Leijon et al., 2010), a four-level physical activity index was created (low active – somewhat active – moderately active – physically active).

In 16 out of the 33 participants (11 women), each of the six iPod^® entries was completed by a saliva sample (SaliCap^® system; IBL, Hamburg, Germany) for the assessment of cortisol and sAA (awakening, +30 min, 10 am, 2 pm, 6 pm and 9 pm). Until analyses, saliva was kept frozen at −20 °C. Cortisol levels were measured using a commercially available enzyme-linked immunosorbent assay (ELISA) according to the manufacturer’s instructions (IBL, Hamburg, Germany). Salivary α-amylase activity was measured using a kinetic colorimetric test with reagents obtained from Roche (Mannheim, Germany) and a sample dilution of 1:400. In short, α-amylase cleaves oligosaccharides (here 4,6-ethylidene-(G₇) p-nitrophenyl-(G₁)-α,D-maltoheptaoside) into fragments that are further hydrolyzed by an α-glucosidase to yield p-nitrophenol. The absorbance of p-nitrophenol was measured at 405 nm (reference wavelength 570 nm) using a spectrometer (Spectrostar nano, BMG Labtech, Ortenberg, Germany). The rate of p-nitrophenol liberation is directly proportional to the samples’ α-amylase activity.

To account for the nested structure in the data, two-level hierarchical linear modeling (HLM 7, Scientific Software International Inc., Lincolnwood, IL) was applied, with time points at level 1 (all models include the variable time since awakening) nested within persons at level 2 (all models include the factor sex). To test for a difference between conditions (control vs. examination), “condition” and the “condition × physical activity” interaction were included in the models. The summary score of the SSCS was included as a level-2 moderator of the intercept as well as the slope physical activity. As effect size, “Pseudo-R² = (σ²reference model – σ²final model)/σ²reference model,” where the reference model is the unconditional model (without any predictors) or the final model excluding the predictor in question (Singer & Willett, 2003), will be reported. Our hypothesized mediations model with preceding PA as the putative causal variable, momentary cortisol and sAA as mediators and momentary general fatigue as outcome was analyzed according to the step by step procedure estimating lower-level mediation in random-effects multilevel models by Kenny et al. (2003). Further, we calculated the intraclass correlation using the null model (ICC = τ_00/(τ₀₀ + σ²) (Woltman et al., 2012) to determine which percentage of the variance in fatigue is attributable to the person level (level 2) and which to the within-person level (i.e. time points). Missing data points were automatically excluded (listwise per measurement time point per person) by the HLM program. Group allocation and group differences were analyzed using Chi-squared test and univariate analysis of variance (ANOVA), respectively (IBM SPSS Statistics for Windows, Version 22.0. Armonk, NY).

Results

An ICC of 0.26 suggested that 26% of the variance in fatigue is at the person level (level 2) and 74% is at the within-person level (level 1). Activity groups (n = 5 somewhat active, n = 14 moderately active, n = 14 physically active) did not differ on SSCS scores (F[2,30] = 1.088, p = 0.350) and summary scores of PA counts (F_{control condition}[2,29] = 2.332, p = 0.115; F_{exam condition}[2,30] = 0.469, p = 0.630); consequently, this factor was not considered further.

Hierarchical multilevel analyses showed that fatigue levels were significantly higher in the examination condition than in the control condition. However, there was no evidence that preceding PA was a significant predictor of momentary fatigue levels (Table 1, Pseudo-R²= 0.17 for this model). There was a cross-level interaction of chronic stress (SSCS) with PA (Table 1), suggesting a moderating effect of perceived chronic stress on the association between PA and fatigue. In fact, the lower the SSCS value the stronger the relationship between PA and fatigue. Furthermore, there was no difference between conditions (control vs. examination) in the association between PA and fatigue.

Table 1. Hierarchical linear model predicting momentary fatigue by physical activity, condition, sex and SSCS (n = 33) using restricted maximum likelihood.

Fixed effects	UC	SE	t-ratio
Intercept	1.99	0.13	14.79^a
level 2
Sex	0.05	0.11	0.67
SSCS	0.06	0.01	6.85^a
level 1
Condition	0.48	0.14	3.40^a
Physical activity	<−0.01	<0.01	−0.02
Physical activity × condition	<−0.01	<0.01	−0.14
Cross-level interaction
Physical activity × SSCS	<0.01	<0.01	2.85^b
Random effects	VC	SD	χ²
Intercept	0.13	0.37	136.85^a
Physical activity	<0.01	<0.01	50.90^c
Physical activity × condition	<0.01	<0.01	89.53^a

UC: unstandardized coefficients; SE: standard error; d.f.: degrees of freedom; SD: standard deviation; VC: variance component; SSCS: screening scale for the assessment of chronic stress; condition: 0 – control condition, 1 – examination condition; model controlled for time since awakening,.

^a p < 0.001.

^b p < 0.01.

^c p < 0.05.

To understand the nature and direction of this moderation, Figure 1 depicts two groups of SSCS scores, that is, those with an SSCS value of 24 or higher (n = 20, SSCS 32.9 ± 6.9), which is the threshold of above-average chronic stress (corresponds to a T value of 61 in the norm sample; Schulz et al., 2004) and those with SSCS scores below this threshold (n = 13, SSCS 20.6 ± 4.0). Descriptively,¹ the association between preceding PA and fatigue seems to be negative in the latter group, while in the former group PA seems to be unrelated to fatigue. Of note, activity group distribution did not differ between low and high SSCS groups (X²[2] = 1.527, p = 0.466).

Figure 1. Spaghetti plot of average (thick) and subject-specific (dotted) regression lines for momentary fatigue as a function of physical activity counts from the 15 minutes prior to data entry according to chronic stress group (< vs. ≥ a SSCS score of 24, i.e., the threshold of above-average chronic stress); SSCS: Screening Scale for the Assessment of Chronic Stress.

With regard to cortisol and sAA (for descriptive purposes see Figure 2 depicting daily slopes depending on condition), neither marker was significantly related to physical activity or fatigue (analyzes available upon request). Since the first criterion of mediation was not met (Kenny et al., 2003), cortisol as well as sAA activity had to be ruled out as mediators of an immediate short-term PA effect on fatigue.

Figure 2. Averaged daily slopes of salivary cortisol (A) and alpha-amylase (B) comparing the control (blank dots) and the exam (filled dots) condition.

Discussion

In contrast to our expectation, current everyday-life PA did not significantly influence momentary general fatigue levels in a sample of healthy university students. Testing the hypothesized moderating role of chronic stress yielded mixed findings. While fatigue levels were shown to be higher during the stressful examination period, condition-related changes in physical activity levels did not account for changes in fatigue scores. On the other hand, perceived chronic stress during the three months prior to study participation moderated the PA–fatigue relationship. PA was more strongly related to attenuated fatigue scores in those participants who experienced less chronic stress. This finding is in contrast to a previous study showing that PA was associated with subjective well-being only on days of high stress and demand (Giacobbi et al., 2007). Our results would challenge a “stress-buffering hypothesis of physical activity” insofar as current PA in everyday life had no buffering effect on momentary general fatigue in the face of chronic stress (Gerber & Puhse, 2009). It might be assumed that under chronic stress, there is no short-term benefit of an increase in PA. Chronic stress might diminish fatigue-buffering effects of daily physical activity and even reverse it,that is everyday activity episodes can be understood as an indicator of a busy, stressful day. Hence, the question arises whether the kind of PA has a meaning in this context. Participants reported more spare time- and less work-related activities in the control condition compared to the examination condition, but there was no difference between low and high chronic stress groups in respect to what they were doing at the time of prompting (data not shown). Furthermore, engagement in regular physical activity needs to be considered. As previously shown, only subjects with regular physical activity as compared to those with irregular or no activity increased their activity under stress (Lutz et al., 2010). In our study, chronic stress scores and total activity counts did not differ with respect to habitual PA, making it unlikely that this factor is a major confounder of our findings. Given our small sample size and unequal group distribution, further research is warranted to examine habitual PA as a moderator of the activity–fatigue association.

The missing short-term benefit of PA corresponds well with observations in patients suffering from the most extreme form of persistent fatigue, that is, chronic fatigue syndrome, who often report a worsening of symptoms after physical and mental exertion (Jammes et al., 2005). This so-called postexertional malaise is an important part of current diagnostic strategies in this disorder and can also be quantified via objective tests (Twisk, 2015). Although increasing PA levels is a major goal in rehabilitation, it is difficult to implement activity programs in these patients. Our findings suggest that the lack of short-term benefit from PA could provide an explanation for why particularly patients under chronic stress hesitate to engage in PA. In line with this, engaging in PA has been considered an additional stressor demanding time and energy (Lutz et al., 2007).

Besides the high ecological validity of our microlongitudinal approach and the repeated measurement of both fatigue and PA using electronic assessment methods, some limitations need to be acknowledged. When investigating subjects during their normal daily living, assessments may be biased due to increased reactivity to external cues. Participants might have changed their routines to make data entries more convenient. If this assumption holds, a random sampling method might have been preferable. Furthermore, random effects have been shown to be underestimated when investigating only 30 or fewer level-2 units, and a sample size of at least N = 50 has been recommended (Maas & Hox, 2005). Despite this disadvantage, we showed that healthy young adults benefit from current PA in the short term, at least under lower perceived chronic stress. One important question is whether chronic stress indeed increased during the examination condition. Unfortunately, this remains unanswered as we assessed chronic stress only during the control condition. However, given our previous report on only slight changes in momentary stress levels (control: 2.39 vs. examination: 2.75 on a Likert-scale from 1 to 5; Doerr et al., 2015), huge increases in chronic stress are not to be expected. Moreover, we focused on immediate short-term effects of PA. Future studies should acknowledge the possibility of delayed effects of everyday-life PA, covering longer time periods than in our current study (5 days per condition). As PA of higher intensity has been shown to have even stronger effects on arousal and affect (Schwerdtfeger et al., 2008), it is also important to consider the intensity of PA and its effects on subjective well-being. In this regard, future research should also consider the multidimensionality of fatigue by incorporating its mental and physical aspects.

With regard to biological mechanisms, our study found no significant evidence for an involvement of the stress axes (salivary cortisol and sAA) in the PA-fatigue relationship. As the HPA axis and the ANS have been implicated in fatigue (Newton et al., 2011; Powell et al., 2013) as well as in the health-protecting effects of regular physical activity (Klaperski et al., 2014), and given our small sample size (n = 16 provided saliva samples), in the future, it would be well worth investigating HPA and ANS markers as assumed biological mechanisms using sufficiently large sample sizes. Furthermore, an adaptation in study design might be advisable to acknowledge different response trajectories of cortisol and sAA. Saliva sampling immediately after physical activity assessment appears appropriate for ANS activity, while samples for HPA activity should be collected at a delayed time point (+15 min). Moreover, examining chronic stress and regular physical activities’ effects on the diurnal trajectories of cortisol and sAA as well as fatigue will be the task of future studies. Ideally, these studies will consider longer data collection times than in the current study.

Conclusion

While, at first glance, our study provided little evidence that everyday-life PA has an immediate short-term impact on general fatigue, perceived chronic stress was shown to be a highly relevant moderator of this relationship. PA resulted in reduced fatigue only under lower chronic stress. Under higher chronic stress, current PA did not have an effect on fatigue. Whether these findings in a healthy student population are transferrable to the observation of postexertional malaise in chronically stressed and fatigued populations will be the next step in determining the health benefits of physical activity. In addition, it needs to be investigated how chronic stress impacts the effectiveness of PA interventions. Future studies will have to show whether and how PA (and its intensity) needs to be adapted to reflect the target population’s needs. Moreover, future studies should expand the focus to other dimensions of fatigue as physical activity might be of particular interest in buffering physical and/or mental fatigue.

Acknowledgements

The authors thank Luisa Donath and Jannis Ziemek for their help with data collection.

Disclosure statement

Jana Strahler, Johanna M. Doerr, Nadine Skoluda and Urs M. Nater acknowledge the support of the Volkswagen Foundation (Germany; AZ.: II/84 905). Funding of participant reimbursements was partly provided by the University of Marburg, and funding for the bio-chemical analyses was partly provided by the Universitaetsstiftung of the University of Marburg (AZ.: VB1.2-5.45.26.04). The funding sources had no role in the design of the study, data collection and analysis, drafting of the manuscript or in the decision to submit this paper for publication. All authors declare that they have no conflict of interest.

Notes

1 Due to the small number of subjects within SSCS groups, a meaningful analysis of moderation is not possible. With this limitation, correlational analyses (Pearson) showed no association between preceding PA and fatigue in either group (SSCS <24: r= −0.062, p = 0.237; SSCS ≥24: r= −0.014, p = 0.715). Separate HLM analyses of PA-predicting fatigue confirmed this finding (SSCS <24: UC≤ −0.01, t-ratio= −1.36, ns; SSCS ≥24: UC ≤0.01, t-ratio =0.251, ns).