Sleep, as an underlying mechanism for regeneration and regulation processes in the human body, is considered an important marker of mental health and wellbeing [31]. Sleep disorders, especially insomnia (which is characterized in the International Classification of Diseases 11th Revision [ICD-11] by difficulties in initiating or maintaining sleep [39]) and its impact on daytime wellbeing [26], play an important role as predictors and as comorbidities of various mental disorders [23]. In addition, sleep disorders are associated with a plethora of health-related outcomes such as cardiovascular diseases or diabetes [11] and with functional impairments regarding work-related performance [12]. It is well known that sleep disorders are common in the general global population [21], which highlights the relevance of investigating potential risk factors to improve preventive strategies.

One potential risk factor for insomnia or reduced sleep quality is work-related stress. Work-related stress can be characterized by theoretical work models such as the job demand–control model [16] and the effort–reward imbalance model [32]. Karasek [16] assumed that high demands and low control at work can lead to subjective stress perception, contributing to impaired health of employees. Within the framework of the ERI model, Siegrist [32] postulated the cause for work-related stress to lie in a gratification crisis, meaning an imbalance between effort at work (e.g., demands) and reward (e.g., money, esteem, and status control) [32, 33]. In the current study, the focus is on the ERI model.

A positive association between work-related stress, especially the ERI, and reduced sleep quality or insomnia symptoms is well investigated. In a meta-analysis it was shown that ERI and insomnia symptoms were positively associated in cross-sectional analyses [40]. In addition to the included studies, Fahlén et al. [10] reported that individuals with higher ERI scores were associated with higher prevalence rates of sleep disturbances and fatigue compared to individuals with lower ERI scores. The authors reported that the strongest associations were found between low reward and fatigue, high ERI and sleep disturbances, and high effort and sleep disturbances, especially in women [10]. A recent study concurs with the positive association between ERI and impaired sleep quality [29]. Besides cross-sectional studies, Yang et al. [40] summarized in their meta-analysis results of longitudinal studies as well. In one of the two included studies, there was a positive association between ERI at a baseline assessment point and insomnia symptoms, albeit only in men [8]. This was supported by findings of Rugulies et al. [30], who reported the prediction of the onset of sleep disturbances through an increase in the ERI at the first assessment point among men but not among women. In the second included study, within the framework of the meta-analysis, it was shown that there was only a positive association between ERI at baseline and insomnia symptoms at follow-up if participants reported insomnia at baseline [22]. In a longitudinal study, ERI was associated with higher sleep disturbances [9]. In sum, more longitudinal studies investigating the relationship between ERI and reduced sleep quality are needed. Furthermore, underlying mechanisms, such as rumination, should be considered in future studies to derive therapeutic mechanisms.

To derive psychotherapeutic interventions, it is necessary to investigate which underlying mechanisms lead to impaired sleep quality. Therefore, an important aim of this study was to clarify whether individuals suffering from work-related stress ruminate, which, in turn, results in impaired sleep quality. Within the framework of the cognitive activation theory of stress, it is argued that health is only affected by sustained activation as cognitive perseveration [20] related to focusing or ruminating on the stressor [3]. Rumination itself is defined as a “class of conscious thoughts that revolve around a common instrumental theme and that recur in the absence of immediate environmental demands requiring the thoughts” [18, p. 7]. Rumination in a work-related context is specified as “conscious thoughts during off-job time […] that revolve around work and that recur in the absence of immediate environmental demands requiring such thoughts” [38, p. 2]. Therefore, the question arises of whether rumination mediates the association between work-related stress and sleep quality. Initial studies have been carried out to investigate rumination as a mediator in this relationship. It was shown that affective rumination works as partial mediator between work-related interruptions (one single item of the ERI questionnaire) and psychosomatic symptoms [41]. Furthermore, rumination mediates the association between ERI and reduced sleep quality [2]. However, these studies conducted their analyses in a cross-sectional study design, which underlines the importance of longitudinal investigations to clarify whether long-term rumination due to work-related stress leads to impaired sleep quality later. If so, psychotherapeutic interventions could be used to address rumination and consider it as a possible maintaining factor.

The present study focuses on the relationship between ERI, rumination, and sleep quality. Due to the cognitive activation theory of stress underlying the hypothesis of a mediating role of rumination, it is assumed that rumination could be a mediator between ERI and reduced sleep quality in a long-term process. The current study addresses the following hypotheses: first, that ERI is positively associated with reduced sleep quality in cross-sectional and longitudinal investigations; second, that ERI is predictive for rumination; third, that rumination is predictive for reduced sleep quality; and fourth, that rumination mediates the association between ERI and reduced sleep quality.

Methods

Participants and study procedure

In the present study, data from the Dresden Burnout Study (DBS) were used. Study design, recruitment strategy, and data collection have been described in more detail elsewhere [25]. In brief, the DBS was a prospective study investigating biopsychosocial determinants of burnout. The study was approved by the local ethics committee and conducted in accordance with the Declaration of Helsinki. Registration and data collection started in January 2015. In addition to self-evaluated questionnaires, information on sociodemographic factors was obtained. One essential questionnaire for our study was included only at the third follow-up assessment point (FU3). Therefore, for our longitudinal analysis, three consecutive assessment timepoints of the participants were included in the data analysis. Thus, for the present paper, data collected from August 2019 (starting point of FU3) to August 2023 were included. For participants who entered the study from the third timepoint (FU3), timepoints FU3, FU4, and FU5 were included. For participants who only entered the study from the fourth timepoint (FU4), timepoints FU4, FU5, and FU6 were included. This procedure ensures that a consistent number of three timepoints is available for all participants at the end of the analysis. For simplicity, the terms time 1 (T1), time 2 (T2), and time 3 (T3) are used instead in the following manuscript. Participants of the DBS were recruited via the Internet (www.dresdner-burnout-studie.de) and via the population registry mainly of the Dresden area. Initial inclusion criteria for this substudy were age between 18 and 68 years and full- or parttime employment. A sample of N = 846 for T1, N = 857 for T2, and N = 606 for T3 was obtained according to these inclusion criteria. Only individuals for whom the time intervals between T1 and T2 (mean [Mdays] = 428.52; standard deviation [SDdays] = 48.88: M + 1.5 × SD = 501.84 days ≙ 1.38 years) and between T2 and T3 (Mdays = 380.07; SDdays = 13.57: M + 1.5 × SD = 400.42 days ≙ 1.10 years) were at least 1 year and at most 3 years were included. In total, the sample consisted of 1381 individuals.

Out of the total of 1381 individuals, 368 were participants at all three timepoints. In total, 73.35% dropped out or did not participate at all three timepoints. After excluding 8 individuals due to incomplete questionnaire responses, we included 360 individuals (nfemale = 227) in the analysis of reduced sleep quality (SQ) and ERI. A detailed description of the exclusion process is presented in Fig. 1, and a descriptive summary of the sample is presented in Table 1.

Fig. 1
figure 1

Participation rate, including study inclusion criteria, illustrated with a flowchart

Full size image
Table 1 Sociodemographic and descriptive statistics of the sample including rumination, sleep quality, and effort–reward imbalance variables at time 1, time 2, and time 3
Full size table

Self-reported measures

Effort–reward imbalance

For evaluation of work-related stress, the standardized 16-item short version of the effort–reward imbalance questionnaire was used [34], assessing the three dimensions “effort” (three items), “reward” (seven items), and “over-commitment” (six items). Items of all dimensions are answered on a four-point Likert scale (1 = strongly agree, 2 = agree, 3 = disagree, 4 = strongly disagree). Sum scores can be created for each dimension , whereby the higher the sum score, the higher the deviation in the respective dimension. Furthermore, to address the research questions of this study, an effort–reward ratio can be used, representing the imbalance between the two constructs [33]. Internal consistency (ERI: α = 0.82; effort: α = 0.74; reward: α = 0.81) was acceptable.

Sleep quality

The German version of the Pittsburgh Sleep Quality Index (PSQI; [4]) was used to assess SQ at the three assessment timepoints. The PSQI questionnaire consists of a self-evaluating part (19 items) and a part evaluated by the partner or roommate (5 items). Seven different dimensions of sleep (subjective sleep quality, sleep latency, sleep duration, sleep efficiency, sleep disturbances, sleep medication, and daytime dysfunction) are examined, which can be converted into an overall score between 0 and 21. A high overall score (values >5) indicates low SQ, so the higher the value, the worse the SQ [4]. Test–retest reliability reported by a previous study was good, r = 0.89 [1]. The test–retest reliability (r = 0.70) in our study was acceptable.

Work-related rumination questionnaire

For evaluation of work-related rumination at the three assessment timepoints, the work-related rumination questionnaire (WRRQ) was used [6]. This self-evaluating questionnaire evaluates three dimensions of rumination: affective rumination (five items; e.g., “Do you become tense when you think about work-related issues in your free time?”), problem-solving pondering (five items; e.g. “In my free time I find myself re-evaluating something I have done at work”), and detachment (five items; e.g. “Do you find it easy to unwind after work?”). Items are answered on a five-point Likert scale (1 = very seldom/never, 2 = seldom, 3 = sometimes, 4 = often, 5 = very often/always). Item 6 (detachment scale) was reverse coded. The higher the rumination dimension scores, the more the person ruminates. The reliability investigated by a previous study was α = 0.84 for affective rumination, α = 0.84 for detachment, and α = 0.83 für problem-solving pondering [24].

Internal consistency in the current study was acceptable (WRRQ: α = 0.91; affective rumination: α = 0.85; detachment: α = 0.89; and problem solving: α = 0.76).

Confounders

In addition to the described psychometric measures—PSQI and ERI—several variables were used as potential confounders in adjusted analyses. Because of their potential systematic influence on either SQ or ERI, age, sex, body mass index (BMI), and depressive symptoms (PHQ-9) were included as covariates in adjusted analyses. Longitudinal and cross-sectional analyses were additionally adjusted for baseline levels of the independent variable and differences in days between T1 and T2.

Data analysis

All analyses were conducted using R Studio version 4.3.2 [28]. Sociodemographic variables were examined using descriptive statistics. Means, standard deviations (SD), ranges, and Pearson correlations were calculated for continuous variables. Correlation coefficients were interpreted according to Cohen’s [5] guidelines. Based on an outlier criterion of 2.5 times the interquartile range, 42 participants were excluded from the mediation analysis. The PROCESS macro for R Studio was used for mediation analysis [14, 15]. Inference on the statistics is based on bootstrapping methods, as many of these statistics have irregular sampling distributions, making inference problematic using conventional methods [14].

In a first step for cross-sectional mediation analysis, the model included a predictor (ERI at T1), potential mediators (work-related rumination [affective rumination, problem-solving rumination, and detachment at T1]), and an outcome variable (SQ at T1). In a second step, age, sex, and BMI (at T1) were controlled, and in a third step, depressive symptoms (at T1) were also controlled. The a‑path shows the relationship between ERI (predictor assessed at T1) and work-related rumination subscales (potential mediators assessed at T1). The b‑path shows the relationship between each work-related rumination subscale and SQ (Fig. 2a, c, e).

Fig. 2
figure 2

Proposed cross-sectional and longitudinal mediation models. a Cross-sectional mediation model. b Longitudinal mediation model. a independent variable to mediator path, b mediator to dependent variable path, c total effects of independent variable on dependent variable, c’ direct effects of independent variable on dependent variable. T1 time 1, T2 time 2, T3 time 3

Full size image

In the longitudinal analysis, all three steps are conducted analogously to the cross-sectional analysis. The predictor (ERI at T1), mediator (affective rumination, problem solving, and detachment at T2), and outcome variable (SQ at T3) remain the same, but they describe the data at different timepoints (Fig. 2b, d, f). The covariates age, sex, BMI, and depressive symptoms also describe the data from T2.

Bootstrap tests (10,000 replicates and 95% confidence interval) were used to compute confidence intervals and inferential statistics and to test the significance of the mediating effect where the 95% CI does not include zero, indicating significant effects. The total effects of the mediation models were interpreted to answer the first three hypotheses. For the fourth hypothesis, the entire mediation model was used.

Results

Descriptive statistics

Significant correlations were found between ERI, SQ, affective rumination, and detachment at each timepoint but not for problem solving across all timepoints. According to Cohen [5], almost all correlations can be classified as small to moderate (Table 2). All statistically significant correlations between covariates and ERI, SQ, affective rumination, detachment, and problem solving were below r = 0.70 at T2 (Table 3).

Table 2 Intercorrelations between the variables of effort–reward imbalance, affective rumination, detachment, problem solving, and sleep quality for all three timepoints (T)
Full size table
Table 3 Intercorrelations between the covariables for timepoint 2 (T2)
Full size table

Effect of work-related stress on sleep quality

In formulating hypothesis 1, a positive relationship between ERI and reduced SQ was proposed. Cross-sectionally, the results showed that ERI T1 was a significant positive predictor of reduced SQ T1 (total effect: step 1: b = 0.38, p < 0.001; step 2: b = 0.38, p < 0.001). The total effect remained even after depressive symptoms were added (total effect: step 3: b = 0.18, p < 0.001). Longitudinally, our results also showed that ERI T1 was a significant positive predictor of reduced SQ T3 (total effect: step 1: b = 0.21, p < 0.001; step 2: b = 0.21, p < 0.001). Overall, an effect of ERI in reducing SQ was found, thus supporting hypothesis 1.

Effect of work-related stress on rumination

Regarding hypothesis 2, it was assumed that ERI would be a significant predictor of rumination. Cross-sectionally, the results showed that ERI T1 was a significant predictor of affective rumination T1 (step 1: b = 0.38, p < 0.001; step 2: b = 0.40, p < 0.001). The effect remained even after depressive symptoms were added (step 3: b = 0.26, p < 0.001). ERI T1 was a significant predictor of detachment T1 (step 1: b = −0.23, p < 0.001; step 2: b = −0.23, p < 0.001) but not of problem solving T1 (step 1: b = 0.03, p = 0.605). Longitudinally, ERI was a significant predictor of affective rumination T2 (step 1: b = 0.29, p < 0.001; step 2: b = 0.30, p < 0.001). The effect remained even after depressive symptoms were added (step 3: b = 0.18, p < 0.001). ERI was a significant predictor of detachment T2 (step 1: b = −0.19, p < 0.001; step 2: b = −0.19, p < 0.001) but not of problem solving T2 (step 1: b = 0.03, p = 0.523). Supporting hypothesis 2, an effect of ERI on rumination was found.

Effect of rumination on sleep quality

Testing hypothesis 3, rumination was expected to significantly predict reduced SQ. Cross-sectionally, the results showed that affective rumination T1 (step 1: b = 0.34, p < 0.001; step 2: b = 0.38, p < 0.001; step 3: b = 0.15, p < 0.01), detachment T1 (step 1: b = −0.32, p < 0.001; step 2: b = −0.33, p < 0.001; step 3: b = −0.13, p < 0.01), and problem solving T1 (step 1: b = 0.12, p < 0.05; step 2: b = −0.14, p < 0.01) were significant predictors of reduced SQ T1. Longitudinally, affective rumination T2 (step 1: b = 0.30, p < 0.001; step 2: b = 0.29, p < 0.001), detachment T2 (step 1: b = −0.32, p < 0.001; step 2: b = −0.33, p < 0.001; step 3: b = −0.13, p < 0.01), and problem solving T2 (step 2: b = 0.12, p < 0.05) were significant predictors of reduced SQ T3. In support of hypothesis 3, an effect of rumination in reducing SQ was found.

Rumination as a mediator between ERI and SQ

To test hypothesis 4, a model was specified in which it was examined whether the relationship between ERI T1 and SQ T1 was mediated by rumination T1. Table 4 shows a summary of the results of this cross-sectional mediation model. Including the mediator affective rumination T1, the direct effect of ERI T1 on SQ T1 remained significant in all three steps (step 1: b = 25, p < 0.001; step 2: b = 23, p < 0.001; step 3: b = 14, p = 0.002). The model showed a significant indirect effect of ERI T1 on SQ T1 via affective rumination T1 in all three steps (step 1: 0.13, 95% BootCI (bootstrap confidence intervals) [0.077, 0.192]; step 2: 0.15, 95% BootCI [0.093, 0.214]; step 3: 0.03, 95% BootCI [0.009, 0.080]). For the mediator detachment T1, the direct effect of ERI T1 on SQ T1 remained significant in all three steps (step 1: b = 31, p < 0.001; step 2: b = 31, p < 0.001; step 3: b = 17, p < 0.001). The model showed a significant indirect effect of ERI T1 on SQ T1 via detachment T1 (step 1: 0.07, 95% BootCI [0.034, 0.122]; step 2: 0.08, 95% BootCI [0.037, 0.129]) before adding depressive symptoms in the third step. Regarding the mediator problem solving T1, the direct effect of ERI T1 on SQ T1 remained significant in all three steps (step 1: b = 38, p < 0.001; step 2: b = 38, p < 0.001; step 3: b = 18, p < 0.001). The model did not show a significant indirect effect of ERI T1 on SQ T1 via problem solving T1 in any of the steps. Indirect effect analysis showed that affective rumination and detachment cross-sectionally mediated the relationship between ERI and SQ when depressive symptoms were not considered.

Table 4 Hierarchical regression analysis for the cross-sectional mediation analysis at T1
Full size table

Longitudinally, a model was also specified that examined the relationship between ERI T1 and SQ T3 and rumination T2 (Table 5). When entering the mediator affective rumination T2, the direct effect of ERI T1 on SQ T3 remained significant for the first two steps (step 1: b = 0.12, p = 0.022; step 2: b = 0.12, p = 0.021). After adding depressive symptoms in the third step, no significant prediction was observed for the direct effect but was for the a‑path and b‑path. The model showed a significant indirect effect of ERI T1 on SQ T3 via affective rumination T2 (step 1: 0.09, 95% BootCI [0.048, 0.130]; step 2: 0.09, 95% BootCI [0.048, 0.131]) before adding depressive symptoms in the third step. Looking at the mediator detachment T2, the direct effect of ERI T1 on SQ T3 remained significant for the first two steps (step 1: b = 0.15, p = 0.004; step 2: b = 0.15, p = 0.004). After the addition of depressive symptoms in the third step, no significant prediction was observed. The model showed a significant indirect effect of ERI T1 on SQ T3 via detachment T2 (step 1: 0.06, 95% BootCI [0.025, 0.104]; step 2: 0.06, 95% BootCI [0.025, 0.108]) before adding depressive symptoms in the third step. Considering the mediator problem solving T2, the direct effect of ERI T1 on SQ T3 remained significant in the first two steps (step 1: b = 21, p < 0.001 and step 2: b = 20, p < 0.001). After adding depressive symptoms in the third step, no significant prediction was observed. The model did not show a significant indirect effect of ERI T1 on SQ T3 via problem solving T2 in any of the steps. Indirect effect analysis showed that affective rumination and detachment longitudinally mediated the relationship between ERI and SQ when depressive symptoms were not considered.

Table 5 Hierarchical regression analysis for the longitudinal mediation analysis at T1, T2, and T3
Full size table

Discussion

Summary of results

The present study aimed to examine the extent to which rumination may mediate the relationship between work-related stress and sleep quality over time. The findings confirm the first hypothesis that increased work-related stress predicts reduced sleep quality both cross-sectionally and longitudinally. It was observed that work-related stress can affect sleep quality over longer periods of time, as measurements were taken 1 year apart. In addition, the second hypothesis that work-related stress leads to more rumination was tested, which was evident in both cross-sectional and longitudinal analyses for the affective rumination and detachment subscales. The third hypothesis linking rumination to a decrease in sleep quality over time was also supported. In addition, evidence was found for the final hypothesis that rumination mediates the relationship between work-related stress and sleep quality, specifically for affective rumination and detachment. We confirmed the direct and indirect effects of this mediation, both cross-sectionally and longitudinally, even after adjusting for covariates, with the exception of depressive symptoms. Thus, we suggest that work-related stress is likely to negatively influence sleep quality over time via affective rumination as a mediator. Similarly, we confirmed that lower detachment due to work-related stress predicted lower sleep quality. However, no effect on problem solving was observed. Contrary to our prediction that work-related stress would impair problem solving, thereby leading to poorer sleep quality, we found no significant effect of work-related stress on problem solving. This suggests that problem solving is not affected by previous work stress. These findings are now discussed alongside the existing literature.

Rumination as a mediator between work-related stress and reduced sleep quality

The results of this study suggest that affective rumination and detachment function as partial mediators between work-related stress and sleep quality in cross-sectional and longitudinal analyses. While previous studies have examined the role of rumination as a mediator between work-related stress and sleep quality using a cross-sectional design [2, 41], to the authors’ knowledge, no study has examined this relationship using a longitudinal design. The role of rumination as a partial mediator in the current cross-sectional findings is consistent with previous findings from cross-sectional studies. For example, Zoupanou and Rydstedt [41] demonstrated that affective rumination serves as a partial mediator between work-related interruptions and psychosomatic symptoms. Similarly, rumination has been shown to act as a mediator between work-related stress and reduced sleep quality [2].

Specifically, the study’s longitudinal analyses show that individuals who experience elevated levels of work-related stress are more likely to engage in rumination about their work-related experiences. This can be explained by studies which suggest that affective rumination is characterized by high psychophysiological arousal [7, 13] and is associated with negative health outcomes, as shown by various research findings [13, 27]. There is also a significant negative association between self-reported affective rumination and sleep quality [27, 37]. This mental preoccupation with work can hinder the ability to mentally disengage from work-related issues, which, in turn, leads to reduced sleep quality. These findings are consistent with the stressor–detachment model [35], which postulates that work-related stressors can strain an individual’s resources and thus lead to increased stress. Employees are more negatively activated when they experience high levels of work stress. This makes it more difficult for them to mentally detach from work during non-working hours. This negative activation can ultimately lead to affective rumination about the stressor [35]. In addition, studies have shown that rumination can cause anxiety and arousal and, therefore, sleep disturbances [17, 36].

Furthermore, the mediation of the negative effects of work demands on sleep quality through detachment is also consistent with the stressor–detachment model [35]. Detachment from work has a positive effect on mood and exhaustion. However, high time pressure and workload may make this detachment more difficult [7, 35]. When employees are able to detach from work in their leisure time, stress responses such as persistent sleep problems are reduced. Our findings are consistent with research showing that high job demands predict a lack of detachment and that a lack of detachment in turn predicts poor sleep quality [19].

As opposed to affective rumination, problem solving exists without psychological and physiological arousal [27]. It is a form of thinking that involves prolonged mental engagement with a problem or evaluation of past work to identify opportunities for improvement but without the emotional process of affective rumination [7]. Another study indicates that affective rumination predicts increased fatigue, while problem solving and detachment predict decreased fatigue. This suggests a distinct functionality of the rumination styles. Additionally, this would falsify the statement that rumination generally hinders recovery [24]. This difference in the style of rumination should be taken into account when considering the current findings. In contrast to affective rumination, problem solving may have a protective effect by interrupting the rumination process by finding a solution, whereas affective rumination does not provide such relief [27]. In the present study, no comparable effects of problem solving on reduced sleep quality were found. Whether there is a protective effect of problem solving should be investigated in future studies.

Limitations and strengths

There are several limitations to the current study. Firstly, all the data collected were self-reported, which may introduce subjectivity and social desirability bias. To increase the validity of future research, it would be advisable to include objective measures such as actigraphy or polysomnography to obtain more accurate assessments of sleep quality. Secondly, the chosen time interval of 1 year between measurement timepoints is relatively long. While this allows one to identify long-term trends, it may miss more immediate changes in stress and their impact on sleep quality. To capture these dynamics more accurately, future studies may benefit from shorter measurement intervals. Finally, it should be noted that some of the data collection occurred during the pandemic period of COVID-19, which introduced changes in working conditions for many individuals, such as increased remote working and reliance on telecommunications, which could potentially influence the results or introduce unique stressors not accounted for in the prepandemic period.

Conclusion

The current study sheds light on the mediating role of rumination in the relationship between chronic work-related stress and a later decline in sleep quality over time. The established association between prolonged exposure to work-related stress and impaired sleep quality underscores the need for health professionals to consider these factors when treating patients with sleep disorders. Given that poor sleep quality can have far-reaching consequences for an individual’s overall health and wellbeing, understanding how work-related stress contributes to such outcomes is critical to the development of effective interventions aimed at improving both the mental health and sleep hygiene of the working population. Hopefully, these findings will inform future research and clinical practice to minimize the negative effects of stress on sleep quality.