The present study uses data from the Stroke Cohort Augsburg (SCHANA), an observational single center cohort study, carried out as collaboration between the Chair of Epidemiology and the Department of Neurology and Clinical Neurophysiology at the University Hospital Augsburg. The cohort includes 945 patients aged 18 years and older with a confirmed ischemic or hemorrhagic stroke who were treated at the University Hospital Augsburg between September 2018 and November 2019.

After inclusion in the study, a standardized questionnaire was completed by the participants themselves or during a personal interview by a trained study nurse after the acute stroke event during their hospital stay. If the participant was not able to provide a self-report, proxy interviews were performed. Clinical data were extracted from the participants' medical record. Three and 12 months after the stroke event, the study participants received postal questionnaires for follow-up. Additional information about the SCHANA study can be found in the published study protocol (19).

The study was performed in accordance to the Declaration of Helsinki. Ethical approval was obtained from the ethics committee of the Ludwig-Maximilians-Universität München (Reference Number: 18-196). Written informed consent was obtained from all participants or legal caregivers.

The baseline questionnaire included information on socio-demographics, disease history, and smoking behavior. Standardized questionnaires were used to assess physical activity, general health status and depressive symptoms. The follow-up questionnaires 3 and 12 months after the stroke event additionally contained questionnaires to assess fatigue and stroke-specific health-related quality of life (see Table 1).

Information on prior diseases and stroke severity collected by chart review were included in the present data analysis. A modification of the Charlson Comorbidity Index (mCCI) was built to assess multimorbidity (37). If more than one additional disease contained in a list of relevant diseases (e.g., hypertension, coronary heart disease, diabetes mellitus) was diagnosed, multimorbidity was ascertained (38, 39).

The degree of functional impairment post-stroke was assessed by the modified Rankin Scale (mRS), which ranges from zero (no symptoms) to six (dead). The mRS is deemed a valid and reliable instrument for stroke patients (40).

The National Institute of Health Stroke Scale (NIHSS) assesses stroke severity by the number of clinical symptoms after the stroke event. Higher total scores indicate more severe neurological symptoms (41, 42).

A descriptive analysis of the total sample characteristics stratified by the different fatigue levels was performed. Multiple linear regression models were used to determine associations between fatigue and a number of possible predictors and covariables. The covariables were selected according to the directed acyclic graph (DAG) method using the “DIA” diagram editor (Version 0.97.2). The DAG is a method to identify relevant covariables from literature and prior studies. The resulting DAG showed associations between fatigue and a number of covariables such as gender, age, prior stroke, multimorbidity, cardiovascular comorbidities, physical activity, general health status, depression, stroke severity and HRQOL.

The assumptions for multivariable linear regression analyses were tested using scatterplots to confirm linearity of associations, normal distribution of the residuals and homoscedasticity. Cook's distances and leverage diagnostic plots were used to check for leveraging outliers. Variance inflation factors were used to identify multicollinearity and Durbin Watson tests were calculated to describe autocorrelation. Interaction effects of age, gender, NIHSS, mRS and multimorbidity were tested, but were not statistically significant. For statistical tests an alpha level of 0.05 was defined. Statistical analyses were conducted using SAS® Version 9.4.

From the 945 participants of the SCHANA study, 585 (61.9%) responded to the survey 3 months post-stroke and 533 (56.4%) responded to the survey at 12 months. Participants with data available from both surveys (n = 505, 53.4%) were included in the present analysis.

Baseline information of participants who were excluded from this analysis due to missing follow-up surveys was compared with the included participants in order estimate a potential selection bias. Excluded participants were significantly older, less likely to be married and more likely to have a prior stroke or a prior diagnosis of depression than included participants. Significantly higher NHISS and mRS scores indicate that they had a more severe stroke. Moreover, excluded participants had higher PHQ-9 scores (indicating more depressive symptoms) and lower EQ5D-VAS scores (indicating a worse general health status) in comparison with those included in the present analysis.

Characteristics of the entire cohort can be found in Supplementary Table 1. Characteristics of the analysis sample are shown in Table 2. The study population consisted of 293 (58.0%) men and 212 (42.0%) women with a mean age of 68.7 ± 12.3 years. Most of them (n = 489, 96.8%) had an ischemic stroke and 404 (79.6%) had a first-ever-stroke. At admission the median NIHSS score was 1.0 and the median mRS score was 2.0. Multimorbidity was common in 78.8% of the participants, with hypertension being the most frequent concomitant disease in 81.4%. A prior diagnosis of a depressive disorder was documented in the medical chart in 50 (9.1%) of the participants.

Participants with more severe fatigue were significantly more likely to have had a prior stroke, a prior diagnosis of a depressive disorder, higher NIHSS and mRS scores at admission and discharge, more depressive symptoms (PHQ-9) and a worse general health status (EQ-5D VAS) than participants with less severe fatigue (see Table 2).

Three months post-stroke, 31.1% of the participants reported to have moderate (25.6%) or severe (5.5%) fatigue (see Table 3). After 12 months the proportion of participants with fatigue was 29.1%, with 22.4% reporting moderate fatigue and 6.7% with a severe level of fatigue. Considering the temporal course, in most of the participants with fatigue at 3 months (n = 157) it persisted from 3 to 12 months (n = 96), whereas n = 61 recovered. Late onset fatigue 12 months after stroke was present in 51 participants (10.1% of all participants).

The multivariable linear regression model of the FAS score 3 months post-stroke is shown in Table 4. Independent variables were gender, age, prior stroke, NIHSS and mRS scores at admission, multimorbidity, prior depressive disorder, PHQ-9, EQ-5D VAS and IPAQ total MET-minutes/week assessed at baseline. The variables explained 17% (adjusted R²) of the FAS variance. Prior stroke was significantly (ß = 2.37, p = 0.0076) related with more fatigue, as well as higher NIHSS scores at admission (ß = 0.25, p = 0.0360). The absence of a prior diagnosis of a depressive disorder (ß = −5.04, p = 0.0001) and lacking information on a prior diagnosis of a depressive disorder in the medical chart (ß = −3.83, p = 0.0027) were both significantly associated lower FAS scores indicating lower fatigue levels. In addition, higher PHQ-9 scores indicating more symptoms of depression, were significantly (= 0.55, p < 0.0001) associated with a higher level of fatigue.

The regression model of the FAS score 12 months post-stroke showed that the independent baseline variables explained 23% (adjusted R²) of the FAS variance. Similar to the regression model for 3 months post-stroke, prior stroke, NIHSS at admission, prior depressive disorder and PHQ-9 were significantly associated with the FAS score (see Table 4). In addition, younger age (ß = −0.07, p = 0.0219), a worse rating of general health at baseline (ß = −0.04, p = 0.0287) and low pre-stroke physical activity (ß = −0.0004, p = 0.0089) were significantly associated with higher fatigue levels.

When adding the FAS score at 3 months as an additional independent variable to the model, the explained variance increased to 38% (adjusted R²). The FAS score at 3 months was significantly associated with the FAS at 12 months (ß = 0.46, p < 0.0001) (see Supplementary Table 2).

In the regression models NIHSS and mRS at admission were replaced by the same variables at discharge in order to check for any differences. However, the results did not change. NIHSS remained significant, whereas mRS did not. Thus, the information from admission was included in the final models.

Since the PHQ-9 includes one item requesting tiredness, the regression models were additionally calculated with a PHQ score which excludes this item. Both scores were strongly correlated (Pearson r = 0.98, p < 0.0001). The results of the regression models were similar (see Supplementary Table 3). The association of the modified PHQ with fatigue was also highly significant (3 months: ß = 0.59, p < 0.0001; 12 months: ß = 0.72, p < 0.0001).

Table 5 shows the significant association between dichtomized FAS and PHQ scores (fatigue yes/no, depression yes/no). A total of 14.8% (3 months) and 16.3% (12 months) of the participants have both depression and fatigue. Three months post-stroke, participants with depression had a 30.91-fold odds [95% confidence interval (CI) 15.31–62.41] to also have moderate to severe levels of fatigue, compared to participants without depression. Twelve months post-stroke, the odds were still high with an odds ratio of 21.32 (95% CI 12.22–37.18). However, among the participants with fatigue, 44.2% (3 months) and 52.2% (12 months) did not reach PHQ-9 scores indicating a depressive disorder.

In the cross-sectional regression models, SIS, PHQ-9 and EQ-5D VAS measured at 3 or 12 months, respectively, were included as independent variables. The regression model of fatigue at 3 months post-stroke showed that higher PHQ-9 scores (ß = 0.68, p < 0.0001) indicating more depressive symptoms and lower EQ-5D VAS scores (ß = −0.06, p = 0.0203) indicating a worse health status were significantly and independently associated with higher fatigue levels (see Table 6). Similarly, these variables were significant in the model calculated for fatigue at 12 months post-stroke. Additionally, in the 12 months-model a significant association between communication problems (SIS) and fatigue was found (ß = −0.14, p = 0.007).

Exclusion of the PHQ from the models changed the results. In the regression model for fatigue at 3 months post-stroke, the SIS domains communication (ß = −0.12, p = 0.0036), emotion (ß = −0.08, p = 0.0033) and participation (ß = −0.05, p = 0.0350) were significantly associated with the level of fatigue. Similarly, in the regression model at 12 months post-stroke the SIS domains communication (ß = −0.16, p = 0.0001) and emotion (ß = −0.11, p < 0.0001) emerged as significantly related variables.