Association of post-stroke fatigue with physical activity and physical fitness: A systematic review and meta-analysis

Background: It has been hypothesized that post-stroke fatigue (PSF) is associated with reduced physical activity (PA) and impaired physical fitness (fitness). Understanding associations between PSF and PA, and/or fitness could help guide the development of targeted exercise interventions to treat PSF. Aims: Our systematic review and meta-analysis aimed to investigate PSF’s associations with PA and fitness. Summary of review: Following a registered protocol, we included studies with cross-sectional or prospective observational designs, published in English or a Scandinavian language, which reported an association of PSF with PA and/or fitness in adult stroke survivors. We searched MEDLINE, Embase, AMED, CINAHL, PsycINFO, ClinicalTrials.gov, and World Health Organization’s International Clinical Trials Registry Platform from inception to November 30, 2022. Risk of bias was assessed using Quality in Prognosis Studies. Thirty-two unique studies (total n = 4721 participants, 55% male), and three study protocols were included. We used random-effects meta-analysis to pool data for PA and fitness outcomes, and vote-counting of direction of association to synthesize data that could not be meta-analyzed. We found moderate-certainty evidence of a weak association between higher PSF and impaired fitness (meta r = –0.24; 95% confidence interval (CI) = –0.33, –0.15; n = 905, 7 studies), and very low-certainty evidence of no association between PSF and PA (meta r = –0.09; 95% CI = –0.34, 0.161; n = 430, 3 studies). Vote-counting showed a higher proportion of studies with associations between higher PSF and impaired fitness (pˆ = 0.83; 95% CI = 0.44, 0.97; p = 0.22, n = 298, 6 studies), and with associations between higher PSF and lower PA (pˆ = 0.75; 95% CI = 0.51, 0.90; p = 0.08, n = 2566, 16 studies). Very low- to moderate-certainty evidence reflects small study sample sizes, high risk of bias, and inconsistent results. Conclusions: The meta-analysis showed moderate-certainty evidence of an association between higher PSF and impaired fitness. These results indicate that fitness might protect against PSF. Larger prospective studies and randomized controlled trials evaluating the effect of exercise on PSF are needed to confirm these findings.


Introduction
Post-stroke fatigue (PSF) is a debilitating condition that affects rehabilitation outcomes, 1 quality of life, 2 and mortality. 1The reported prevalence of PSF ranges from 25 to 85%. 3 PSF's etiology is unknown, but is believed to be multifactorial and different types may exist. 4he literature suggests that PSF could be associated with physical activity (PA) and physical fitness (fitness), 5 which are two distinct, yet related outcomes.PA is defined as "any bodily movement produced by skeletal muscles that results in energy expenditure," 6 while fitness is "a set of attributes that people have or achieve that relates to the ability to perform PA." 6 Components of fitness include, but are not restricted to, aerobic fitness, muscle endurance, muscle strength, and body composition. 6PSF may lead to reduced PA and consequently to impaired fitness, but it is also possible that impaired fitness following a stroke may trigger PSF, leading to reduced PA and further impairments in fitness. 5PA and fitness are both modifiable factors, so that, although causal relationships are difficult to establish, a better understanding of their associations with PSF could guide development of targeted exercise interventions to reduce PSF. 5 This would be an important step, as there are currently no established interventions for treating PSF, although exercise training interventions have been successful in decreasing fatigue in other conditions. 7,8A systematic review on associations between PSF and impaired fitness concluded that the three eligible studies yielded insufficient evidence about any such association. 5As additional studies have since been published, we have updated and expanded this review using broader inclusion criteria.
Our primary aim was to conduct a systematic review and meta-analysis to determine whether PSF is associated with PA and/or fitness in patients with stroke.Our secondary aim was to explore PSF's associations across different subdomains of PA and fitness.

Methods
The study protocol was registered in the Prospective Register of Systematic Reviews (PROSPERO; CRD4202 1216435) and reported in accordance with the Preferred Reporting Items for Systematic Review and Meta-analysis (PRISMA) statement. 9

Eligibility criteria
Eligible studies had cross-sectional or prospective observational designs, were published in English or a Scandinavian language, included stroke survivors aged 18 years or older, and examined associations between PSF and PA and/or fitness.Studies that only measured balance, activities of daily living, or functional limitations were excluded.We included published study protocols/clinical trial registry records (TRRs)) if they met our inclusion criteria.

Search strategies and selection criteria
We performed a systematic search in AMED, CINAHL, Embase, MEDLINE, PsychINFO, ClinicalTrials.gov,and World Health Organization's (WHO) International Clinical Trials Registry Platform (ICTRP) from inception to November 30, 2022.The search strategies are described in Figure S1.
We de-duplicated records in EndNote X9 10 and imported them into DistillerSR 11 or Rayyan. 12Two authors (PL and EE) independently screened titles and abstracts, and fulltext articles using a pre-piloted eligibility-criteria checklist.TRRs were screened in Rayyan or manually by one author (PL).

Data extraction
Two authors (PL and EE) independently extracted data from eligible studies using a customized data extraction form in DistillerSR.For the updated search, PL extracted the data and another author (MU or AL) cross-checked the extraction.If disagreements could not be resolved, a third author (JB) was consulted.Data included publication details, study and patient characteristics, PSF outcome, PA/ fitness outcome, and measure of association (Table S1).We contacted study authors if additional information was needed.Data from multiple reports of the same study were linked together and treated as one.For each study, only one association per outcome (i.e.PA/fitness) was extracted.If a study reported more than one association per outcome, we chose the most relevant (Table S2).

Data synthesis
We performed a meta-analysis when two or more studies reported a correlational value between PSF and PA, or PSF and fitness.We used Comprehensive Meta-analysis software 13 to calculate meta correlation (meta r), 95% confidence intervals (CIs), and prediction intervals.We used randomeffects models, which assume that study population or design differences may influence the data. 14Heterogeneity was assessed by prediction intervals 15 and by visually inspecting the Forest plot.We could not perform sub-group and sensitivity analyses predefined in the PROSPERO protocol, due to an insufficient number of studies.Data that could not be pooled were synthesized by vote-counting of direction of association.We calculated the proportion of studies reporting a negative association (higher PSF and lower PA/fitness) along with 95% CIs (Wilson intervals) using an online calculator. 16A two-sided p-value was obtained from the binomial probability test using Stata software. 17Vote-counting does not consider statistical significance or magnitude of association. 14

Risk of bias assessment
Two authors (EE and PL) independently assessed risk of bias.If consensus was not reached, a third author (JB) was International Journal of Stroke, 18 (9)   consulted.We used the Quality In Prognosis Studies (QUIPS) checklist, 18 modified for our study.QUIPS has six domains (Table S3), each rated as having low, moderate, or high risk of bias.
To detect outcome-reporting biases, we searched for existing protocols for the included studies.We compared the planned outcomes reported in the protocols with reported outcomes in the final publications.

Certainty of evidence
One author (PL) rated certainty of evidence and a second author (JB) cross-checked the rating using the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) framework. 19Quality of evidence was rated as very low, low, moderate, or high.We summarized the evidence in GRADEpro. 20

Search results
The search identified 4125 records.One additional record was obtained from the reference list of a systematic review. 5fter duplicates were removed, 2298 records remained for screening.We excluded 2146 records based on titles and abstracts.We assessed 285 TRRs for eligibility in Rayyan, 40 TRRs manually, and 153 full-text documents in DistillerSR/Rayyan.Thirty-four published papers and two TRRs met the inclusion criteria (Figure S2).

Study characteristics
We included 32 unique studies (total n = 4721 participants, 55% males).Michael and colleagues published two articles 21,22 on the same sample, both referred to as Michael 2007. 21ne published protocol and one TRR described included studies. 23,24The second TRR was ongoing. 25Most studies were of cross-sectional design (n = 22), the others were longitudinal prospective studies (n = 7) or randomized controlled trials (RCTs; n = 2) (Table S4).All studies were published in English between 2006 and 2022.Table S5 describes the characteristics of each included study.

Associations between PSF and PA
The three studies in the PA meta-analysis included 430 participants. 21,38,41PSF was measured with the FSS-9 21,38 or a single question about the impact of fatigue on daily activities. 41PA was measured as steps-per-day 21,38 and with a question about weekly exercise habits. 41The pooled data (Figure 1) showed no correlation between PSF and PA (meta r = -0.09;95% CI = -0.34,0.16; p = 0.47; I2 = 79%; very low-certainty evidence).Prediction interval was not calculated because of few included studies.
Combined, PA meta-analysis and vote-counting showed no association between PSF and reduced PA; very lowcertainty evidence.
Combined, the fitness meta-analysis and vote-counting showed a weak association between higher PSF and impaired fitness; low-certainty evidence.

Risk of bias
For the 30 included studies, the QUIPS domains most frequently assessed as having low risk of bias were "Statistical analysis" (all studies) and "Measurement of prognostic factor, fatigue" (all but three studies). 29,35,41The domain most frequently assessed as high risk of bias was "Study confounding" (11 studies) 21,32,33,38,40,41,44,45,49,50,54 (Table S7).Our GRADE judgments are shown in Tables S8 and S9.The most common reason for downgrading certainty of evidence was risk of bias, followed by inconsistency and imprecision.International Journal of Stroke, 18 (9)

Discussion
To our knowledge, this is the first systematic review including meta-analyses of PSF's associations with PA and fitness.In contrast to the 2012 review, 5 there were sufficient studies to pool data and assess certainty of evidence.
Fitness vote-counting showed statistically non-significant, very low-evidence of a higher proportion of studies reporting a negative association between PSF and fitness.The analysis was underpowered 14 and should be interpreted with care.However, the meta-analysis showed moderate-certainty evidence of a weak association between PSF and impaired fitness.This finding could support the hypothesis of fitness as a protective factor against PSF.Since fitness is a modifiable factor that can be improved by exercise, 6 exercise training could be a viable intervention for PSF.Zedlitz et al. 56 reported that cognitive therapy combined with exercise training was better at reducing PSF than cognitive therapy alone and suggested that improving physical endurance may help reduce PSF.Their finding is consistent with studies showing that exercise can reduce fatigue associated with other diagnoses. 7,8he potential benefits of exercise training for preventing and treating PSF have been discussed for at least a decade, 57 but few advances have been made toward this goal.We found no studies designed to investigate the effect of exercise training on PSF.Nonetheless, there is now sufficient evidence of a connection between PSF and fitness to warrant RCTs on exercise interventions to relieve PSF.
We estimated no correlation between PSF and PA in the meta-analysis; very low-certainty evidence.Vote-counting showed very low-certainty evidence of a higher proportion of studies reporting a negative association between PSF and PA, but the test was not statistically significant.Our results are supported by a previous meta-analysis that reported no association between PSF and PA. 58In theory, PSF may cause individuals to be less physically active, 5 but increased PA could also exacerbate PSF. 5 Causality is difficult to establish and may even vary between individuals.Our findings may reflect such variability.Intrapersonal factors affecting patients' participation in PA may also play a role: some individuals may choose to avoid PA because of PSF, while others may choose to participate in PA despite PSF. 59Because PSF and PA are complex concepts, with many dimensions and associated factors, 4,58,60 a straightforward causal relationship between the two may not exist.Larger prospective studies, designed to investigate specific hypotheses about PSF's relationship to PA, are warranted.
In prior meta-analyses, the most frequently-reported risk factors for PSF included female sex, depression, and disability. 60,61PA and fitness were not among the factors analyzed, but PA was highlighted as a candidate that could explain more of the variance in PSF. 60Based on our metaanalyses, we propose fitness as a more likely candidate.However, the weak association of our estimate suggests there are additional factors involved.Previous research has found a non-linear relationship between PSF and age, with higher PSF among both younger and older individuals. 60,62 similar relationship could exist between PSF and fitness, but our study only evaluated linear relationships.Perhaps more likely, age and higher demands on younger individuals 60 may confound PSF's relationship with fitness.In our clinical experience, younger patients often suffer from PSF despite being relatively fit.Thus, age-adjusted associations may be stronger, but some included studies did not adjust for age.Well-designed studies are needed to further update the evidence-base on both PSF risk factors and interventions to improve care for stroke survivors of all ages. 63

Limitations and strengths
There are several limitations of the included studies.The majority of the evidence came from studies with small sample sizes, using different outcome measures, measured at different time-points.Fatigue instruments are known to measure different dimensions (e.g.intensity, physical fatigue, mental fatigue, impact on activities), and there is a lack of content overlap between instruments. 64This means that the fatigue instruments themselves may have varying degrees of associations with PA and fitness.The same applies to the different PA and fitness outcome measures.The included studies also measured associations at different time-points poststroke, and associations between PSF, PA, and fitness may differ during the acute phase, rehabilitation, or a more stable phase.The extracted associations were reported as a secondary aim or finding in 15 studies, 26,27,30,32,33,38,41,43,44,46,47,49,[51][52][53] suggesting that these studies were designed and powered for other primary objectives.
As for limitations of our review, since most of the included studies did not have published protocols, we did not use protocols in conjunction with risk of bias assessment as we had planned.This decision did not affect the overall risk of bias in the two studies 23,24 concerned, nor did it affect the GRADE confidence in certainty of evidence.The PA meta-analysis estimate should be interpreted with caution due to high heterogeneity.A very low certainty of evidence means that the actual estimate of association may differ in either direction.Finally, selection bias may exist, as we only included studies published in English or a Scandinavian language.Our study's strengths include the multidisciplinary research team, adherence to systematic review guidelines, the extensive search (including TRRs), ability to pool data, and use of GRADE to appraise the certainty of evidence.

Clinical implications
Given the multiple benefits of PA and exercise for poststroke recovery, 65 it is arguable that PA and exercise training should be recommended regardless of whether patients have PSF or not.As PSF can be a barrier to PA, 66 proper International Journal of Stroke, 18 (9)   screening, assessment and individually-tailored management of PSF is essential.

Conclusion
The fitness meta-analysis showed moderate-certainty evidence of an association between PSF and fitness, indicating that being physically fit might protect against PSF.This result needs to be confirmed by RCTs evaluating exercise interventions, with PSF as the primary outcome.The PA meta-analysis showed very low-certainty evidence of no association between PSF and PA, highlighting the lack of large prospective observational studies in this field of research.

Figure 1 .
Figure 1.Standardized association between post-stroke fatigue and physical activity.Results are shown for individual studies reporting Pearson's correlation.

Figure 2 .
Figure 2. Standardized association between post-stroke fatigue and physical fitness.Results are shown for individual studies reporting Pearson's correlation.