1.1 Theoretical models on job stress and burnout

Burnout has been proposed as one possible outcome of prolonged job stress since the 1980’s [26]. Job stress and burnout are mostly measured with self-report questionnaires. The two most often used models to assess job stress are the demand-control model [27] (and the more recent demand-control-support model) [28] and the effort-reward imbalance model [4,29]. The first model distinguishes between demands and control on the job. Demands are measured in terms of time, quantity, and mental variables on the job whereas control is measured as the amount of decision latitude, growth possibilities, and the amount of creativity one is able to put in one’s work [30]. There is a reciprocal relationship between demands and control, where an imbalance towards high demands/low control is used to describe job stress. The second theoretical model distinguishes effort and reward on the job. Efforts are measured with variables such as demands, workload, and work pressure whereas reward is measured in terms of monetary incentives, self-esteem, and career opportunities [4]. An imbalance between the two is referred to as high cost and low gain. Siegrist et al. [4] put it as follows: “in the long run, the imbalance between high effort and low reward at work increases illness susceptibility as a result of continued strain reactions” (p. 1485).

Next to job stress, burnout is also defined in various theoretical models. A well-known and influential model of Maslach [31] characterizes burnout as a feeling of exhaustion and depersonalization, with low levels of personal accomplishment. Exhaustion includes feelings of being used up or emotionally drained by one’s work [32]. Depersonalization is characterized by feelings of callousness towards other people, while low personal accomplishment is described in terms of the perceived impact of one’s work. Considering that burnout is a possible severe reaction to (prolonged) job stress, we hypothesize that if burnout is the result of job stress, the effect of burnout on the autonomic nervous system might have an even stronger influence than the effect of job stress alone.

In the following paragraphs we will first discuss some common heart rate measures followed by the skin conductance measures that were analyzed in this systematic review.

1.2 Physiological measures of job stress and burnout

HR can be analyzed both in terms of beats per minute and in terms of the inter-beat interval (IBI). The mathematical analysis of HRV is based on the variation of the IBI interval [33], and can be divided in the amount of parasympathetic or mixed (both parasympathetic and sympathetic) activity that is reflected [9,13,34–36]. HRV can be calculated in both time domains and frequency domains [37] (and nonlinear analysis, but this was not included in the current review). Three time domain measures are used in the current review and are based on the variation in peak to peak interval. The standard deviation of these peak-to-peak intervals is also referred to as the standard deviation of normal-to-normal intervals (SDNN), which is a measure of overall HRV [37]. In addition, the percentage of adjacent cycles greater than 50ms apart (PNN50) and the root mean square of successive differences (RMSSD) are used. A Frequency domain measure is also included. These measures are based on the analysis of peak-to-peak (RR) interval sequences [33] as well. The frequency components can be calculated as the distribution of power (i.e. variance of the peak-to-peak interval) as a function of frequency. The high frequency power is thought to primarily reflect parasympathetic activity while the other measures reflect mixed activity [35]. Only the HF component is included in the current review (0.15–0.4 Hz).

The measures in the current review are predominantly parasympathetic (PNN50, RMSDD, HF) or sympathetic (EDA measures) in nature. Besides HR, a second mixed measure was included (SDNN) that is traditionally viewed as HRV [37]. SDNN is sometimes considered as the total measure of autonomic nervous system activity (see for an overview, p. 1813) [35].

There is considerable heterogeneity in methodology and measurement that might influence the recordings of HR. For instance, studies report rest measures [38,39], 24 hour recordings [40,41], HR measures during controlled breathing [39] or during a specific task [42]. In addition, ECG (electrocardiogram) devices [30,43,44], blood pressure devices [45,46] or ambulatory PPG (photoplethysmography) sensors [47] are used. Studies report on measures during work [48], leisure time [21], rest [49] or at night [46]. Studies report on untransformed values [45,50], linear transformations [41,51] or both [49]. Bivariate measures are sometimes reported [13,39] or studies report on adjusted models [6,52]. These choices might seriously influence the results that are reported. Moreover, there is significant diurnal variation in both HR [53,54] and EDA [55] which makes the comparability between studies that use different lengths of recording or different time intervals challenging.

EDA is relevant with respect to skin conductance or skin potential [9]. It is one of the most sensitive markers of arousal [12,56], and solely the result of the sympathetic activation of the autonomic nervous system [12]. Although EDA has been studied extensively in experimental research on (among others) anxiety, stress, depression, and personality, it has not often been reported as a marker specifically in association with job stress. This might be due in part to the equipment that was needed to measure EDA in the workplace (e.g., multiple sensors on the fingers and/or hand palm). Recent technological advances make it easier and less intrusive to measure EDA [57,58]. EDA has both tonic (level) and phasic (responses) components. The typical form of a response is well described [12], and several parameters can be extracted such as the height, rising time, area under the curve, or decay time of a response. For this review we will focus on the skin conductance level (SCL), the number of non-specific skin conductance responses per minute (ns.SCR) and the height (amplitude) of the non-specific responses (SCR.amp), as these have been associated most with emotional load in job-related EDA research (pp. 460–462) [12]. Boucsein [12] reported results from a few studies on the association between EDA and job stress. There is a tendency of increased SCL, ns.SCR, and SCR.amp with increased strain and stress. Therefore, these markers will also be addressed in this systematic review.

A recent systematic review on job stress and HRV concluded that “stress at work is generally associated with increased risk of disease and worsened health profiles as indicated by decreases in vagally-mediated HRV.” (p. 1814) [35]. Where these authors looked at both mixed and vagally (parasympathetic) mediated HRV measures in association with job stress, we will focus on these measures (RMSSD, PNN50, HF, HR, SDNN), as well as on EDA in association with both job stress and burnout. Moreover, we will also assess the direction of the associations, even if the effect is non-significant [35]. In addition, the former review analyzed both total as well as subscales of job stress, while this review will solely rely on total validated scales for both job stress and burnout. Only the full scales of the questionnaires were used as the current review focused on the association between both EDA and HR(V) with job stress and burnout. This served as a means to compare the results on ‘job stress’, and is an important distinction with the Jarczok [35] review that investigated the more general ‘workplace stressors’ as both full and subscales were considered [35]. For instance, Jarczok et al., [35] reported that need for control significantly decreased HF in the Hanson [59] study. Although need for control is a subscale of the job stress questionnaire, it is not necessarily considered to be job stress. People can experience heightened demands, but if control is not decreased there is not a ‘pure’ association with job stress as a full scale. The results are thus only applicable to workplace stressors, but not job stress. As the current review compared job stress to burnout, only full scales were considered to compare the job stress-burnout association instead of making comparisons between depersonalization and demands for instance.

This systematic review could therefore be considered, at least in part, as both a replication, update and an extension of their previous work. In sum, we want to know what the association is between job stress/burnout and HR(V)/EDA, which parameters might prove useful, and which recording periods are favorable over others to analyze.

Based on the outcomes of the previous review, three specific hypotheses are formulated. First, it is expected that there are positive associations between job stress/burnout and HR and EDA. Second, negative associations between job stress/burnout and HRV are expected. Third, the association between burnout and HR(V)/EDA is stronger than the association between job stress and HR(V)/EDA as burnout is a possible result of severe, enduring and prolonged job stress. Participants will include the employed adult population comparing validated job stress and burnout questionnaires examining heart rate and electrodermal activity. In addition, the timeframes that are used to assess the physiological measures vary considerably, therefore this review also aims to provide some guidelines of measurement and reporting.

2.1 Literature search and screening criteria

The literature search focused on the relationship between job stress and burnout on the one hand, and EDA, and HR(V) on the other. The review protocol was registered in the international prospective register of systematic reviews (PROSPERO) with ID number CRD42016035918. We used the Preferred Reporting Items for Systematic reviews and meta-analysis (PRISMA) to guide the reporting on the systematic review [60]. The databases that were searched for this systematic review were PsychINFO, Medline, Embase, and Web of Science. For this, we used the subscription of the Radboud University in Nijmegen, the Netherlands. The search engines, and accordingly, the search terms of the databases differ; for this reason we included the search terms in S1 Appendix which included search terms on HR, HRV, EDA, burnout and job stress. The search terms were peer-reviewed by three librarians using the Peer Review of Electronic Search Strategies (PRESS) checklist [61], which resulted in some additional suggestions, the narrowing of search terms and addition of relevant keywords. As the task force guidelines for HR measurement were published in 1996 [37] and needed some time to be adopted, it was decided to include articles from 2000 upwards. Zotero (v. 5.0.52) and Refworks were used to process the references.

The final search was conducted on December 23^rd 2016. The final searches yielded a total number of 1,814 studies. Besides citation snowballing, every included study author (only the corresponding authors) was contacted to ask if they were aware of any further or so-called ‘grey’ literature, which yielded an additional 13 studies. In the end, we included 38 articles (see Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. PRISMA flow chart of the systematic review process.

https://doi.org/10.1371/journal.pone.0205741.g001

2.2. Study selection

The following inclusion criteria were applied:

1. The studies focused on the employed adult population and should concern job stress or burnout related measures, as predictor variables and HR(V) or EDA as outcome variables.
2. The studies focused on baseline, workdays, leisure time or combinations of those.
3. The studies included a validated subjective measure of job stress or burnout.
4. The included studies were English articles published between 2000 and 2016.

We had no further requirements as far as the study design or participants were concerned. Comparisons were made based on validated questionnaires. The HR(V) and EDA measures were divided in separate outcomes for rest, task, workday, leisure, sleep time or combinations of the entire period. We performed an initial screening with three (PCdL, HN, PE) reviewers to establish if the inclusion and exclusion criteria were transparent. For this purpose, two sets of 50 randomly sampled articles were used to establish interrater agreement. This initial screening resulted in 85% interrater agreement on the title and abstract screening, and consensus on transparency of the inclusion and exclusion criteria.

We set out to perform a meta-analysis as this would allow for explaining heterogeneity in effect sizes through moderator analysis [62, pp. 198–228]. However, most of the articles did not report on bivariate correlations or means. A meta-analysis therefore would have resulted in a limited number of processed articles. We could have opted for the use of partial correlations, but there is debate on the use of them [63], even if the results of the bivariate and partial analyses are reported separately. Because we were unable to perform moderator analyses in a meta-analysis we decided to do an exploratory analysis on sample size, sex differences and age as these are expected to moderate the association between psychophysiology and job stress/ burnout.

2.3 Coding and reporting

Two authors double coded the effect directions and significance levels of the included articles (performed by PCdL and LJMC), and the interrater agreement was 88%. After a consensus meeting, both authors agreed on 100% of directions and significant associations. In addition, the articles were coded on the HR(V) or EDA outcomes, and the period of analysis, the time of the HR(V) and EDA measurement, the applied stress model (job stress, burnout), the time of stress measures, and the cut-off used for making subgroups of the participants. Furthermore, several other measures were coded (21 of 38 articles were double coded by (PE, HN, RD)). The risk of bias assessment for all articles are presented in S3 Appendix. In order to avoid simply looking at p-values (vote counting), we also looked at the direction of the comparisons. A positive direction means that higher levels of job stress and burnout were associated with higher levels of HR(V) and EDA. In case the articles reported mean differences or correlation coefficients these were used to describe the found effects. If no tables or information was available the wording of the authors was coded. For instance, Uusitalo [51] only reported all significant correlations in a table and concluded in the text that “the pure vagal time-domain index RMSSD was the only HRV measure which correlated with ER-imbalance” (p. 835). Since that was the only significant effect, it was assumed that the other investigated measures in the study, that is both HF and SDNN, were non-significant. These measures were therefore set to “no effect” and significance level as “not reported” because we also did not have information on the direction of the association. If tables were available we extracted the direction of the association from the tables, if no significance test was applied to the subgroups in those tables we reported that the significance was n.r. (not reported). All significance levels were set at p < .05. All available data is included with the article to both enhance reproducibility and compliance with the PLOS policy.

3.1 Study characteristics

We included 38 articles which reported on 119 outcome measures (Table 1). All reported outcomes in these 38 articles turned out to be HR(V) measures; No EDA studies met the inclusion criteria. In the initial full text review, we did identify 4 EDA studies that were possibly eligible, but they were excluded because of the use of a non-validated job stress questionnaire (2 studies) or because it was a simulation study (1 study), or a real-life stress exposure (1 study).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive characteristics of the articles.

https://doi.org/10.1371/journal.pone.0205741.t001

Nineteen of 38 study authors responded to requests of grey literature or additional information. For four authors we were unable to retrieve a valid email address.

3.2 Association between both burnout and job stress with (para)sympathetic measures

The first hypothesis states that there is a positive association between HR and job stress/burnout. The associations between HR/SDNN and job stress/burnout are summarized by recording period in S2 Appendix. There were thirty-five reported outcomes on HR, of which 32 outcomes are reported in the Appendix. As can be seen in S2 Appendix, the majority of articles on HR and job stress/burnout found positive associations. The three non-reported outcomes were difficult to categorize (see Table 2). First, Poorabdian et al. [64] reported on an omnibus Chi-square test of which it was unclear at which time point the measures were taken. However, the direction of the association was significantly positive. Second, only one reported outcome for the Moya-Albiol et al. [48] study was significantly negative while one outcome was mixed. The study reported a significant negative association caused by measures in the middle of the workday. Third, Borchini et al. [40] reported on a non-significant positive relationship during a period of leisure and night. Therefore, in sum, twelve reported outcomes were significantly positive effects. The ratio of significant effects is in favor of a positive association (12 positive effects vs 2 negative effects; 12:2). Finally, seventeen outcomes reported on SDNN, of which three reported outcomes were significantly negative, and one was significantly positive.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Codesheet from the included articles.

https://doi.org/10.1371/journal.pone.0205741.t002

The second hypothesis states that there is a negative association between the reported parasympathetic outcomes and job stress/burnout. Reported outcomes on HF, RMSSD and PNN50 were included in this systematic review. The association between parasympathetic measures and job stress/burnout are also summarized in S2 Appendix. Thirty-four outcomes were reported on HF. Only 33 outcomes are reported in S2 Appendix as Hanson et al. [59] reported on a non-significant negative association for work leisure period, which was not a defined category in our study. Seven reported outcomes were significantly negative, two were significantly positive. Twenty-seven outcomes were reported on RMSSD, five of the reported outcomes were significantly negative, one was significantly positive. One study outcome was mixed on the conclusion; on the first day of the assessment no effect was found, whereas on the second day a significant and negative effect was found [51]. Six outcomes were reported on PNN50. Only one of the reported outcomes was significantly negative. One further remark has to be made for S2 Appendix. For PNN50, two of the outcomes were measured in rest. One of the outcomes was reported on a significant negative effect for the female sample [39] while the other reported outcome was positive, but non-significant for the male sample.

To summarize, we found 13 significantly negative reported outcomes for the parasympathetic outcomes, compared to 3 significantly positive reported outcomes. The relatively high number of negative effects seems to be in support of the second hypothesis, at least as far as the HF and RMSSD outcomes are concerned. However, most of the 67 reported outcomes were non-significant.

3.3 Burnout and job stress

The third hypothesis states that the association between burnout and HR(V)/EDA is stronger than the association between job stress and HR(V)/EDA. Six outcomes were reported on the HR-burnout association, however, only one was significantly positive [70], with the caution that the burnout group was entered as a dummy variable and part of a multiple regression analysis. Twenty-nine outcomes were reported on the HR-job stress association. Eleven outcomes were significantly positive. These results indicate the opposite of the hypothesis, the association between job stress and HR seems to be found more often than the association between burnout and HR.

As for the parasympathetic outcomes, seven outcomes were reported on the HF-burnout association. Only one reported outcome had a significant negative effect. Twenty-seven outcomes were reported on the HF-job stress association. Five reported outcomes were significantly negative. Five outcomes reported on the RMSSD-burnout association. Only one reported outcome had a significant negative effect. Twenty-two outcomes were reported on the RMSSD-job stress association. Four reported outcomes were significantly negative, one of the effects was mixed [51]. Note that there are two articles and 11 reported outcomes using both JDC and ERI to divide subjects in high/low strain. In sum, these results do not indicate that the parasympathetic association between burnout and HRV is found more often than the association between job stress and HRV.

3.4 Exploratory analysis

Because we were unable to perform a meta-analysis and for purposes of generating hypotheses, we explored whether the effects changed as a result of sample size, sex, or age (Table 3). In order to get a good contrast between samples, the results were analyzed by median splits based on these three variables. We expected HR to have a positive association with job stress and burnout. For the parasympathetic measures, a negative direction was expected. Interestingly, articles with higher sample sizes, and thus presumably providing more power to find a true effect, indeed found twice as much of the hypothesized effects than articles with lower sample sizes. The interpretation of the sex split is less clear as the samples are mixed, but the samples with a higher percentage of women seem to have more negative parasympathetic effects (11 vs 1). No age effects were found.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Amount of positive (HR) and negative (parasympathetic) significant reported effects.

https://doi.org/10.1371/journal.pone.0205741.t003

4.1 Main findings for the hypotheses

The first hypothesis stated that there is a positive association between HR measures and job stress/burnout. First, support for this hypothesis came from both the high number of positive associations and the ratio of positive and negative effects (12 vs. 2), which clearly showed that the likelihood of a positive association between HR and job stress/burnout is higher. In other words, the results of this review support that high levels of job stress and burnout are associated with an increased HR. Second, one-third of all reported outcomes (12/35) showed a significant positive association between HR and job stress/burnout. However, if we leave burnout out of the analysis, 11 of 29 of the effects for job stress were positive. The number of articles on burnout included in this systematic review was too small to draw any firm conclusions. As van Doornen [69] also pointed out, the daily hassles of job stress may be incomparable with scales of exhaustion, depersonalization and personal accomplishment as a result of enduring job stress, but further research is necessary to investigate this claim.

Hypothesis 2 stated that there is a negative association between parasympathetic outcomes, indicated by HRV parameters, and job stress/burnout. Support for this hypothesis came from both the number of negative directions and the ratio of negative and positive effects (13 vs. 3), which showed that the likelihood of a negative association is higher. In addition to the ratio, one fifth (13/67) of all reported outcomes showed a significant negative association. If we leave burnout out of the analysis, one sixth (7/42) of the effects were negative. Twelve of the 19 articles included by Jarczok et al. [35] were also included in this systematic review. The twelve articles included in both systematic reviews showed comparable results, although Jarczok et al. [35] included fewer reported outcomes for the parasympathetic measures than in our study (38 articles). In Jarczoks’ sample, half of the parasympathetic outcomes were significantly negative (17/33) on the (sub)scales. Our analysis resulted in a significant negative effect in only one fifth (6/29) of the outcomes on the full scales, which explains the difference as the unit of analysis of the scales is different. For example, Jarczok et al. [35] also included effects of the separate scales for demands, control, effort or reward of the JDC and ERI (workplace stressors vs job stress) models while we only considered articles that reported results on the entire scale. However, the direction of the effect in both studies (i.e. the study by Jarczok et al. [35] and the current study) was overwhelmingly negative, that is, higher levels of job stress are associated with lower parasympathetic activation. It is worth mentioning that the correlation between HF and RMSSD is usually high (>.90) [43]. Therefore, it seems intuitive to expect a significant association of RMSSD if HF also has an effect, and vice versa, which results in an overestimation of the effects found in our study, due to multicollinearity. Finally, although SDNN is not a primarily sympathetic or parasympathetic outcome measure, there was a tendency towards a negative association between SDNN and job stress/burnout in our study.

The third hypothesis stated that there is a stronger association between psychophysiological measures and burnout than between psychophysiological measures and job stress. However, this hypothesized relationship could not be confirmed. One explanation for this might be the small number of included articles on burnout. Another potential reason might be that people with burnout symptoms do not experience job stress symptoms anymore as they are on sick leave which might have a calming effect on the body, and thus the (para)sympathetic measures.

In conclusion, it is important to state that most of the reported outcomes were non- significant, and for some articles that reported that there was no effect, we could not determine the direction of association. We did find partial support for the two hypotheses regarding a positive direction for the HR measures and a negative direction for the associations of parasympathetic measures with job stress and burnout. The ratio for the HR measures (12 vs. 2) was slightly higher than the ratio for the parasympathetic measures (13 vs. 3). Also, some evidence was found that articles with a larger sample size more often found a significant association. However, the third hypothesis regarding a stronger association between psychophysiological measures and burnout as opposed to job stress was not supported.

4.2 Methodological considerations

Related to the findings for the three hypotheses, four observations can be made regarding methodology and measurement. First, most reported HR outcomes came from rest (n = 11) measures. The number of reported parasympathetic outcomes was almost twice as high, most measures were taken at rest (n = 16), workday (n = 13), for 24 hours (n = 14), and at night (n = 13). It is remarkable that only few articles used rest measures since these seem to be most easily obtained, although some researchers may disagree. However, a rest protocol does assure that there is no movement or psychosocial demand on the participants, which might result in an artifact free signal. Also, there is some heterogeneity within these rest measures. Borchini et al. [40] pointed out that a strict standardized ECG protocol is necessary to obtain precise results. We recommend to include rest as a baseline measure in future studies. The baseline measure could also be used to adjust for between-person variation in psychophysiological recordings. It is worth noting that both HR and EDA show diurnal variation which suggests that this has to be taken into account as well when obtaining rest measures [53–55].

Second, regarding the prolonged measurements of HRV there are two points worth noting. First, Uusitalo et al. [51] suggest that nonlinear measures of HRV are less movement prone. In comparison with frequency domain measures, they argue that time domain and nonlinear HRV might prove to be more stable and suitable for situations in which a person may show a lot of movement resembling the real-life situations. Only few included articles report on controlling for physical activity, which is expected to have an influence. It is recommended to adjust for these movement artifacts if possible. In addition, Kamath et al. [81] mentioned the influence of respiration on the HF component of frequency domain measures; as breathing decreases, HF decreases as well. The authors state that HF measures can only be obtained in case one controls for breathing which might not be possible for all mobile devices that are currently on the market. We recommend careful consideration controlling for movement and breathing if possible.

Third, Clays et al. [43] point at the fact that one cannot compare all parameters obtained in different time intervals as they are dependent on time of analysis. For instance, SDNN is highly dependent on the length of the recording, and there are apparent differences in the duration of measurement intervals between studies. SDNN is typically used for 24-h recordings only. This is in line with remarks made by Kamath et al. [81] that long term measurements are preferably analyzed by time domain methods, and short term measures are preferably analyzed by frequency domain methods. With time domain measures it is difficult to discriminate between sympathetic and parasympathetic measures. In the current review, five of the outcomes on rest measures, which are usually 3–15 minutes long, were time domain measures as opposed to 11 outcomes on the frequency domain. There is a need for guidelines on the use of time domain or frequency domain methods and the duration of assessment from different laboratories [81]. Although a recommendation on duration of measurement is beyond the implications of this review, we would recommend reporting exact timeframes, methods of analysis, and transformations and filters applied to compare data more easily, even if mobile devices are not used in laboratories, but in real life situations.

Fourth, a final notion on real-life measurement is made by Rau et al., [46]. The authors suggest that the assessment of leisure time differs between studies, which may consequently lead to differences in findings. Some studies operationalize leisure time as the time between work and sleep while others consider resting days as leisure time. We recommend the use of after work leisure time as a separate terminology from a day that is completely without work.

Many of the reported outcomes are not independent. Therefore, we evaluated how many of the significant associations came from the same study. For the HR outcomes, the 12 significant positive outcomes came from 11 articles, whereas the 19 non-significant reported outcomes came from 14 articles (no significance test was reported for 3 outcomes; and 1 study reported mixed effects.) Thus, there appeared to be more independency among the significant associations than among the non-significant associations, which increases the reliability of the significant relationships. For the parasympathetic reported outcomes, the 13 significant negative outcomes came from 10 articles, whereas the 49 non-significant outcomes came from 22 articles (no significance test was reported for 2 outcomes; and the association for 3 outcomes was in the opposite direction of our hypothesis–positive instead of negative.) Again, there was more independency among the significant associations than among the non-significant associations.

Regarding the exploratory analyses it is worth noting that for sample size both sympathetic and parasympathetic measures show a tendency to obtain more associations in the hypothesized directions if the sample size is larger. A larger sample size often means a higher power, which implies a higher chance of obtaining a true effect, in this case a positive relationship between HR and a negative relationship between parasympathetic functioning and job stress/burnout which increases our trust in the relationships that we found. We do not want to extrapolate conclusions based on sex as we had less female samples, which is an indication that more female samples are needed in future studies to increase our confidence.

4.3 Directions for further research

A particular strength of the current study is that the results reported by Jarczok et al. [35] were replicated. This is especially relevant as psychological research is currently dealing with a replication crisis [82]. Second, this review extended the former one by only considering the full scales of job stress, adding the concept of burnout, and including EDA as a purely sympathetic marker. A third strength of this review is the additional search that was performed for gray literature by emailing all included authors. In spite of this strength, this study has a limitation on ‘vote counting’. If bivariate correlations or adjusted models with the same covariates are unavailable, it is difficult to summarize the true effect. The problem with vote counting is that it does not consider the magnitude of an effect, and control for heterogeneity or moderation is impossible. A second limitation is that we were unable to perform a meta-analysis which makes the results less compelling than they could have been.

We also have some additional recommendations. First, there is high variability in the number of covariates and bivariate or partial correlations reported. We strongly recommend to report both bivariate relations and adjusted models as this can seriously alter the effects that are found. This becomes even more evident as we consider that, as mentioned earlier, there is dependency between the reported outcome measures as they are all based on (variation in) HR, and some outcome measures have high correlations (i.e. the correlation between HF and RMSSD is usually high (>.90) [83]. We recommend that studies use the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology; [84] statement to guide reporting for future studies.

Second, the use of validated questionnaires is highly recommended, there was an abundance of studies using only one job stress question making it unclear what the construct of the measure was (none of those non validated studies were included in this review). As the subjective meaning of job stress differs between jobs and people within these jobs, for comparability a thorough investigation is important.

Third, in burnout research there is often a predisposed clinical cut-off. In the job stress samples this is not always the case as some report on the quadrants while others use a median split or a 20/80 division. It is recommended to use the same validated cut-offs. A problem with comparability between articles arises as some articles included relatively healthy employees while others included a relatively large number of stressed individuals. In addition, providing the data along with the publication would allow for a meta-analysis of individual data.

Fourth, it is essential to report on both sex and age as there are few articles that report differences between men and women [38,39]. The concept of job stress might not be applicable to men and women alike as Riese [78] already mentioned. The original constructs might hold true more to men than to women. The psychophysiological profile of stress and burnout in women might be different.

Of course the question remains to what extent self-report questionnaire measures correspond to physiological measures of job stress. HRV is often used as a measure of stress while it does not always match the subjective stress that people experience. It would therefore be worth constructing a psychophysiological profile in which the within-subject baseline is taken into account, which would be possible with a baseline measurement. People differ from each other in baseline measures. It is unclear if their baseline is the same in all circumstances, let alone if the baseline can change in situations where they get stressed. In other words, people might have a moving baseline. However, the vast majority of research until now has been cross-sectional in nature, and comparisons were made between subjects or between groups. The baseline of the subjects is only considered in some of the more recent research [80].

The fact that there is a small difference in the number of reported parasympathetic associations between our study and the study by Jarczok et al. [35] stresses the importance of reporting all bivariate correlations or means on all scales, for this will enable a meta-analysis, as was also suggested in more recent research [85]. Moderation analysis can then be performed on different sets of confounders and covariates. The current systematic review includes articles that use a variety of confounding adjustments and covariates, which makes it difficult to compare them.

Lastly, no EDA studies were found that met our criteria. A few studies were considered for inclusion. For instance, Cendales-Ayala et al. [86] did a simulation study in which it was shown that high demands in bus drivers resulted in significantly increased EDA. Considering the sympathetic nature of the EDA, we expected to find more studies. However, only recently it is possible to obtain EDA measures via mobile devices, such as wristbands, in real life during prolonged periods of time. Based on this development we expect to find more studies on EDA and job stress in the near future.

4.4 Conclusion

In conclusion, this study examined whether HR was positively associated with job stress and burnout. In addition, it examined whether there was a negative association between parasympathetic markers, and job stress and burnout. Support was found for both hypotheses. No support was found for the hypothesis that the association with burnout was stronger than it was for job stress. There is a need for more extensive reporting of effect directions, and on female samples, which were underrepresented in the current review. In this sample job stress was mostly related to increases in HR and decreases in RMSSD, HF and SDNN. Maybe these measures can be used as indicators and warning signals of increases in job stress, whereas the relationship with burnout is less clear.